Presentation

Correlational (Ecol()gical) Studies

A 6.drreliitiari.a] (et;.t>logicrtt}. study,· pdf2y{atioti-"leVfl1 data tb e;c4itt,ine the: r.e:' (ation,ship b'ett>lRNJJ! rt!_tes· tJ[$eqse

O!Jjective

Population

Requirement

Whafld watth'Olit ftir9

leYvels:oi exposNre·.ancl d.iseasRin-seve.raJ :popula,tl,of.JS' . .

no ij $)("po;sure have: a ntgber h:ite otaisease?·

+ -· - .': •• ...... ,!' ...

!fOpul&tion-leveLdata are :parti'fipatlt$. .

'Hieaim

Ttii.Er topic has,notbee.n pnzvlouslyexpiored:uslng<ii1diiliam11· : 1evef<d:ata: ·· ·

of·c:fqtf! · ':2. int;lu.d$ i.n

tanac:yJ: ! .• , .

'.·.

.•

FIGURE B -'1 Key Qf Cbrtei'!:XtionaJ Sfudi.es

4.3

• 8.1 Overview • Does the percentage of adults with multiple sclerosis tend to be higher in countries

farther from the equator? • Does the prevalence of diabetes tend to be higher in provinces with a higher preva-

lence of obesity? • Does the rate of asthma tend to be higher in cities with higher levels of air pollution?

Each of these questions can be -explored with a correlational study. Also called an study" or "aggregate study," a correlational study uses population-level

data to look for associations between two or more group characteristics (FIGURE 8-1 ). Because existing data sources are almost always used for correlational studies, the

key to success is identifying a data source that contains comparable information about the variables of interest. Information about all the variables of interest must be avail- able for a suitable number of populations, which can be grouped by place or time. For example, place-based populations could consist of all member nations of the United Nations, all 50 states from the United States, the largest 20 metropolitan areas in the United Kingdom, all the counties in the state of Michigan, or a random sample of cen- sus tracts in New York City. Time-based studies could use historical data for the past several decades from one or more place-based populations.

B 8.2 Data for Correlational Studies For most studies, at least one characteristic of the populations being examined is des- ignated as an exposure, and at least one is designa ted as an outcome or disease. Most exposures and outcomes used in correlational studies are in the form of aggregate

such as the proportion of each population with a particular characteristic or-the average value of the variable in the population. For example, the exposure may be the percentage of adults age 30 and older who have not completed at least 12 years of ed- ucation, the mean income in the population, or the median age. Alternatively, an ex- posure variable may represent an environmental measure that is likely to be fairly consistent across an entire population, .such as the number of rainy days over a given year or the average ultraviolet radiation index during midday in the hottest month of the year. The disease m ay be measured as, for example, the prevalence of obesity among adults or the annual mortality rate from asthma. .

Before conducting a statistical analysis of ecological data, the data must be en- tered into a spreadsheet. Each population should be assigned to its own row in the spreadsheet. Each exposure and outcome should be assigned to its own column. The data should be filled into the cells in each column so that they line up with the correct population. (See FIGURE 8-2 for a sample data table.) ·

44 CHAPTER 8 Correlational (Ecological) Studies

A s, c

t.xposyre'l 48.2 6.5.1 3l.'8

FIGURE 8-2 Sample Data Table

Qutcome ·l 14.1 lZO 14.9

The analy$is, will be valid only' if the data point$ are tom parable, If mu1tiple sources of data are used br il the data were coUected over a lengthy period of time, then the definition of exposure or disease may differ from one population to another. In some populatio.ns, exposures and diseases· may be routinely undercounted or routinely over- diagnosed) compared to other populations. Because of the bility, researchers should interpret e·cologic associations conservatively. .

• 8.3 Analysis: Correlation On :a used to cprrelation, each polnt represents or1,e populatioi1 in the study; The exposure is plotted on the and the outcome or disease is pl<;Jt- ted on they-axis (FIGURE 8-3).

•· When all the points fall neatly, in a line, then the correlation is strong. e: When the points are not exactly linear but a lilfe fortrenq cw be then the

correlation is mild or moderate. • When the p·oints to be randomly placed and no qui be drawn

through th:em, then the correlation is or nonexistent. • If higher levels of ex:pos1,1re: are linked to higher rates of disease:,. then the :slope is

positive. • If higher levels of exposure are linked to 1owet rate.s. of disease, then the is

negative.

For continuous variables and other variables with responses that be plotted on a number line;. a Pe;:trson correlation coefficient (t) ,should he used ·to calculate the correlation. For ·variables that assign a rank to responses or that have ordered cate- gories, us·e the Spear man rank-order correlg:tion (designated by tP:e letter r or the: Greek letter p [rho] in most statistical programs). Eor both tests, the value of r ranges from -1, when lie ·perfectly on a lin·e with ·(,1 negative s1ope,.to 1, when all points lie perfectly on a line with a positive slope. When ·r = 0,. there is no association be- tween the exposure· and Qutcome. ( Ch?pter. 2 7 the difference between para- metric tests like lhe Pearson correlation and nonparametric tests the Spearman

:8:3 Analysis: Correlation 45

Positive slope

•• •

• • •

FIGURE 8-3 Correlation

• • • • • • • • •

• • •

rank-order correlation and Kendall's tank correlation, which is often designated with letter "C [tau]. )

The association between two or more yariables can also be reported as r2, which shows how strong a correlation is without indicating the direction of the association. The value of r2 ranges from 0 for no correlation to 1 for perfect corrdation.

Sometimes more than two variables are being compared or the goal is to under- stand the relationship between two variables while controlling or adjusting for the ef- fects of other variables. In such cases, linear regression models are used to assess the associations (see Chapter 28).

Note that a measure of correlation is .different from a test for intercorrelation. A test for inter correlation exa,mines whether two or mor.e related variables that are part of a survey instrument measure various aspect s of the same thing. For example, Cronbach's alpha and the Kuder-Richardson Formula 20 (KR-20) are measures of in- ternal consistency among items on a questionnaire. Tests of intercortelation examine the reliability of survey instruments and are not the same as tests of correlation that "

pare two. or more independent variables. For ecological studies,, correlation, not intercorre1ation, should be assessed.

• 8,.4 Age Adjustment Sometimes the populations bejng compared have very differ.ent age structures. For ex- ample, one or more populations might skew considerably younger or older than the others. If so, age adjustment may be necessary to make a fair comparisQn among pop- ulations. Direct age adjustment requires knowing the exposure and/or disease rates by age group in ea.ch population. These are then applied to-a starrdatdized population, and

46 CHAPTER 8 Correlational (Ecological) Studies

a summary age-adjusted population rate is calculated for each population being com- pared. Indirect age adjustmen:t methods can spmetimes be used to compare populations for which age distributions are known but age-specific rates of exposure and/or disease are not known.

• 8.5 Avoiding the Ecological Fallacy Correlational studies compare groups rather, than individuals. No individual-level data are included in the only populatioh-level data. The incorrect attribu- tion of pDpulation-level associations to individuals is called the ecological fallacy, and this error should be avoided. Even though a popu"lation with a higher rate of exposure to something has a higher rate of disease than populations with lowe_r exposure rates, individuals in that population who have a high level of exposure do not necessarily have the disease. The experience of an individual in a population may vary significantly from the :pQp,ulation For example, it w,ould be incorrect to that any one individual from a country with a high average body mass index (BMI) will be obese or that an individual froin a country with a low average BMI will not be obese. However, it is appropriate to identify trends in populations and to use those observa- tions to generate hypotheses for individual-leveLstudies that will test for relationships between the characteristics of interest in individuals. Correlationalstudies are a use- ful starting point for genera,tjng hypotheses ;:tb_out associations, bu( they are not the final word on risk factors for disease.

H.S Avoi,ding the Fallacy 4 7

Case Seri'es . .

ca$e strie$ two ar ifi,i;iht patients .who hl;lf;ri :tb.e of· fiJhcxhave .

Q.bjective ! .. Describe a group of [ndiVIduals witfi.a · /:. '

V\'b<Wfire trt $tl;ldy · pdtf:41.\a\iQtl:? '· . .

same-disease or.,tle same . .

- .. , .

;VYnen A t1!ld. grq,UJ-1 ts- . ;.gr e ·

·Requirement .l.. ,$pecitY'What'new and imp.ortant fnfofmationtne. analy$ts;W.!l1

- . o , -·

s , .. .- -'

. '

4. DeclGie of the stucly popula-flon that will :- . -

FIGURE 9-1 Key Characteristics ot .a, Case Series.

4 9

• 9.1 Overview A case series describes a group of individuals with a particular disease (FIGURE 9·1 ). A case series is po.ssible only when a researcher has access to an appropriat.e source of cases and when there is a compelling reason to write about those cases. This stud y ap- proach can be useful for:

• Describing the characteristics of and similarities among a group of individuals with the same signs and/or symptoms of disease

• Identifying new syndromes and refining case definitions • Clarifying typical disease progression • Developing hypotheses for future research

Some case series for rare. conditions may require only a handful of participants. Others may include several htmdred individuals.

• 9.2 Case Definitions A rese archer conducting a case series must select one disease of interest, .determine what will be new and interesting about the study, and identify an appropriate and

category Disease/procedure Person

Place·

Time

Exqmple 1 Whooping cough (ICD-10 code A37) Any person with a confinned case of whooping cough, defined. as an acute cough of any duration with

otBordatella pertussjs from a or a cough lasting 2

or more weeks with p.aro?(ysl}ls of coughing inspiratory "whOOJJ/ or post- tussive vomiting and contact with a laboratory-confirmed case. of pertussis Residents of.: Big City whose diagnoses were reported to the Big City Health Department (which requires notification of all .diagnoses of pertussis)

First soiJght:cliriical care t?,·etween January 1 and March 31, 2011

FIGURE 9-2 Case Definittons

50 CHAPTER 9 Case Series

Example 2 Liver Adult patients (ages 18 and older at the time of transplant), excluding those who were not receiving their first liver transplant and those who received multi-organ transplants

Patients whcrhad transplant . surgery at the Oakville Regional University Medical Center

Recipients Qf Ilver transplants between January 1, 2000, and December 3t 2008, who were followed tor a minimum of 2 years posHransplant

:

.available source of cases. The next step is to establish a clear case definition that spells out inclusion and exclusion Pq.rtlcipants may be selected from clinical loca- tions that use ICD (i.e., diagnoses based on the International Classification of Diseases, known more formally as the lntern.atiqnal Statistical Classification ofDiseaseso ·and Related Hea'lth Problems). If so, the ICD number can be part of the case defini- tioi)., but a code aJone is' rarely sufficient to cover atl inclusion and ex.clu,sion criteria .. A more comprehensive case definition will indude a disease description plus any rel- evant person, place, ·and time ( FIGURE 9-2). Case definitions are also es.'" sential for any outbreak investigation, no matter which study approac'h is used to investigate the epidemic.

• 9.3 Special Considerations A case series might involve primary data a·cquired by interviewing cases about their experiences using a qU,estionrtaire. and/or qualitative The· might be supplemented or confirmed with a review of the participants' medical records .. Alternatively, a case series can be (4nd oft_en is) based solely on secondary data, usu- ally acquired frorn a review of patient charts. .

When medical records will he consulted as part of the dat<l .<;:olledion process, it is often helpful' to ;create a questionnaire that guides the· extraction of information from medical records. One, of the limitations of relying on patient chart:s. is that they usually contain only irrfor·mation deemed at the time 'of ·examination to be .clinically relevant. The medical intormation in patient files is not recorded for pur .. poses; so records are. unlikely to contain all the ihfotmation that researchers would like to know .. Less relevant signs and patient comments;, and clinician observa- tions are usually not recorded . As a result, the a.gsence of a note about a symp- tom history does-noJ necessarily mean that the exposure was not present, just that itwas not recerded. A data extraction tool should include'·space to indicate the absence of a desired piece, of inform_a tion in the record. During the analysis ,and interpretation stage of the re·sea,.rch project, the researcher should carefully consider tlw amount and type, of missing information. . Case· series come with special requirements: All case series studies approval

by a research ethics committee;-, as well as informed consent from participants and/ or the careful use of ex;istihg records. Case series :researchers mtJstpay special attention to pro- tecting the identities of participants.,'This is especially important when the disease or pro- cedure is relatively rate and/or when the place time chqta:cteristics so nattowthat individuals familiar with the source community might be able to recognize the partici- pants. In most all potenti,ally information m:ust be removed prior to publication .. For example, an image should not contain any identifying marks that could reveal the participant's identity. (In some situations,, patients may be allowed to give permission for-potentially identifiable information or images to 'be published.)

9.3 Special c_onsiderations 51

• 9.4 Analysis Most case studies do not require any numbers beyond simple counts and frequencies, but· some may henefit ftofil use of well-defined measures of motbidity and mor- tality. For example, the case fatality rate is the proportion of persons with a particu- lar disease who die as a result of that condition. (This is different from the crude mortality rate, which is the proportion of members of a general population who die of any condition during a time period. It .is also different from the propor- tionate mortality ,rate, which is the proportion of deceased members of a popula-tion whose death was attributable to a particular cause.) lh some situations, comparative statistical tests may be possible when comparing subpopulations within the popula- tion of cases or when comparing before:-and-after measures for the same individual participants.

Although many case setie:s do nothav¢ any time dimension, somefollow patients for days, months, or years.ln this type of study approach, the case series becomes, functionally, a cohort study i11 which all participants are defined by their disease status. Chapter 12 discusses cohort study approaches.

52 CHAPTER 9 case Series

:

Cross·Sedional Surveys

A cross-sectional survey provides a snapshot of the health status of a population at one point in time. Cross-sectional surveys, sometimes called "prevalen-ce stud- ies," are among the most popular study approaches in the health sciences because they allow for the relatively rapid collection of new data.

Objective Describe the exposure and/or disease status in a population Primary study question What is the prevalence of the exposure and/or disease in the

population? Population The study participants must be representative of the population

from which they were drawn. When to use this approach Time is limited and/or the budget is small. Requirement The exposures and outcomes are relative{y common, and the

likelihood of being able to recruit several hundred participants is strong.

1. a population. First steps· pevelbp"a strategy for recruiting sample.

3. Deci<;Je on the methods to be used for data. collection. What to watch out for Nonrepresentativeness of the study population. Key statistical measure Prevalence

FIGURE 10-1 Key Characteristics of Cross-Sectional Surveys

53

.·

• 10.1 Overview The goal of a crQss-sectionalsurvey, also called a prevalence study, is to measure the proportion of a population with a particular exposure or disease. This determination should be made at one point in time based on a representative sample of a popula- tion. (See FIGURE 10-1 .) Cross-sectional surveys are used to:

• Describe communities • Assess pop11lation needs • Evaluate programs • Establish baseline data prior to the initiation of longitudinal studies

• 10.2 Representative Populations In some ways, cross-sectional studies use the simplest study design. The researcher just asks a few hundred people to complete a short questionnaire and then analyzes the. data. However, there is one very impurtant requirement: the participants must be reasonably representative of $orne larger population. The researchers cannot simply ask friend s, the fans attending a youth football game, or individuals attending a chi- ropractic clinic to complete a-survey and then assume that the results of the survey will be generalizable to all town residents. If the results are intended to reflect the pro- file, of an entire town or other population group, then the st,udy's sampling strategy must recruit a population that is as diverse as the town. .

Chapter 16 has more detailed informatipn about populations for a cross-sectional survey, and Chapter 17 explains how to estimate sample size requirements.

• 10.3 Analysis: Prevalence Cross-sectional surveys measure the prevalence of various demographic characteris- tics, histories, and disease states in one well-defined population at one point in time. The most common way to report results for a cross-sectional survey is simply to report the prevalence rate, which is the proportion of the population with a given trait at the time of the survey.

Comparative measures can also be used. For example, ratios compare the: prevalence of a characteristic in two population subgroups by taking a rati.Q of their prevalence_ rates. Because a cross-sectional survey has no time dimension, it can- not be used to assess causality. An exposure can be said to be "associated" or "re- lated" to a disease, but a cross-sectional survey cannot show that an exposure caused a disease.

54 CHAPTER 10 Cross-Sectional Surveys

Case-Control Studies

A ca$e-control study compares the exp·osure h'istories of people with and with- out a particular disease in order to identify likely risk factors for- the disease.

Objective Compare exposure histories in people with disease (cases) and

people without dise(:lses (controls) Primary study question Do cases and controls have different exposure histories?

Populatiu.n Cases and controls rn:ust be similar except for their disease

status.

When to use this approach The disease is relatively. uncommon, but a sQJJrce of cases is available. Requirement A source of cases is available. .

1. Identify a source of cases. 2. Assign a case definition.

First steps 3. :oecic;le. what tyj'>,e population yvilfbe appropriate for tile

4. Decide whether c:ases and controls will be matched. What to watch out for Recall bias

Key statistical measure Odds ratio (OR)

FIGURE 11 -1 Key Characteristics of Case-Control Studies

ss

FIGURE 11-2 Framework for a Case-Control Study (The letters a, b, <;, ano o <;:orrE:spond to the equation shown in Fig!Jre 11-4.)

• 11.1 Overview Individual participants in a case-.control s_tudy are selected for inclusion in the study based on their disease Participants with the of interest are cla.ssifie.d as cases. Those without the dise.ase are classified as contro-ls. Both cases and controls are asked the same set of questions a bout past exposures ( FIGURES 11-1 and 11 -2). A .case- cohtrol study is often the best study approach for identifying risk factors for a dis- ease. This is especially so when the disease is relatively uncommon and a study of the gerteral population is unlikely to yield more than a few cases. A special type of statis- tic-an odds ratio-is used to identify likely risk factors:.

• , 11.2 Finding Cases and Controls Because case-control studies require a fairly sizable number of cases, the first step is to identify an appropriate and accessible source of individuals with the disease of in- terest. Hospitals, specialty clinics, physicians' offices, public health agencies, disease registries, and disease support groups may be able to assist researchers in identifying individuals who :are likely to .meet the st,ttdy's case definition.

(Regardless of the source, in most situations these_organizations will not release any information about individuals until after a research project has received approval from an appropriate ethics oversight committee. When the information is

researcher must exercise extreme care to protect the privacy of potential partici- pants .and the confidentiality of their personal information.)

56 CHAPTER 11 Case-control Studies

:

Chapter 16 provides additional about the selection of All cases must have the -same disease, disability,. or other health-related condition.

So a ca:se-cbhtrolstuqyis td a working ca;se definition that spec- ifies exoactly what characteristics must be present or absent for a person to. be deemed a Clipis:;al manuals ·and publicitions previous studies oJ the dis- ease can he helpful references for-drafting and refining the inclusion and exclusion criteria. The case: definition should include person, place, and time (see Figure 9-2). ·

Next, an -appropriate source of controls rm,1St be §elected. Depending on the goals of the study,_ controls may be recruited from, among other sources: .

·• Friends and relatives of cases • Hospital or clinic patients without the disease of interest • The general population,

CohtrQlsmust similar to cases except fortheir disease sraws. SQ the inclusion and exclusion criteria for :cases that -do not specifically relate to the disease should als_o apply to controls. For cas<;s he between ?'Sand 39 years ,of age, contrDls must alsCl 'he men in this .age group. · .16 provides additional a:bout-rhe selett!qn o£ controls for control studies ..

11.3 Matching Early in the design process, a decision must be mad.e about whether to match <3ase·s and controls; There are three basic options for matching: no matchi'rig, {group) matching, ·and matched-pairs (individual) matching,

·Some studies use no matching. They s.imp.ly that similar inclusion a,np ex- dusion criteria for cases -and controls will result in case and control populations that hav:e:similar distributions according to. sex;, age,graup; status, ;tnd other eharacteristics.

Some studies use frequenl:y (group) matchitig for a few V!ilfi<itbles to com- para ble case and c0ntrol populations. For example" suppose .a study is using hospita l- ized controls. The researcher tnay select,for ea,ch case, one ftom the hospital registration file·s who was admitted the -same week as t he case, who is the

,st;:xas' .th-e case, and who 1s ± 3 yeats of <)ge of the case. IIi. addition to identifying .only· one s·imila-r case, frequency matching can be used to· identify two, three, q:t rnpre, times as controls as .cases,., (Estimating the S.i.lm_ple size for different ratios of cases to controls is described in Chapter 17.) For group match- ing like this, the goal is to tecruJ:t a cob.trpl ·populatio.rl that is similar tb the case: pop- ulation. Individual cases are not tied to individual controls durin$ analysis. So fhe analysis uses the same .·approa.che'$ as useCl for unmatc;he,d studies.

11 .3 Matching 57

Some studies use matched-pairs (individual) matching. Each case is personally linked to a particular individual control. This approach is fairly common in genetic stud- ies, in which a case is linked to a genetic sibling or other close genetic relative for analy- sis. This kind of matched-pairs approach requires a special type of analysis that is discussed at the end of this chapter.

For both frequency matching and matched-pairs matching, it is important not to overmatch. The variables used for matching criteria cannot be considered as expo- sures during analysis. For example, suppose cases and controls are frequency matched based on the date of hospital aqmission, sex, and age. The case and control popula- tions will likely end up, perhaps artificially, having the same proportion of admissions in April, the same percentage of males, and about the same mean age. As a result of this forced similarity, the study will not be able to examine whether cases are more or less likely than controls to require hospitalization in a certain month, to be males, or to be octogenarians. Additionally, when there are more matching characteristics, it can be difficult to find controls who meet all of the matching criteria.

• 11.4 Special Considerations Once the decisions about study design are made, planning for data collection may begin, as described in the third section of this book. Researchers must keep two special points in mind when designing the survey instrument for case-control studies.

First, 'in case-control studi_es, all participants must be asked. questions that con- firm whether each is a case, a control, or neither. The questions must ensure that only confirmed cases and controls are included in the analysis. Adhering to strict defini- tions for what constitutes a case and what constitutes a control minimizes the risk of misclassification bias.

Second, researchers must be aware of the risk of recall bias, which occurs when cases and controls systematically have different memories of the past. This type of risk is particularly important in studies. Participants are asked to recall distant events from the past that cannot be confirmed by documents from when the exposure would have occurred. Cases may be searching for answers to questions about why they have become ill. As a result, they may have more vivid memories of partic- ipation or lack of participation in activities perceived to be risky or beneficial. For ex- ample, adult cases in a study of night blindness may report that they rarely ate carrots as children. They may say this· not because they never ate carrots but because. they as- sume that, if they had eaten lots of vegetables high in vitamin A when they were kids, they would have had good vision as adults. Alternatively, cases may overestimate child- hood carrot intake. They may wonder why they developed night blindoess when they have such fond memories of happily munching on carrot sticks every day at lunch in grade school. The reality may be that they ate carrots only once a month. Controls,

58 CHAPTER 11 Case-Control Studies

..

on the other hand, are unlikely to think much about risk factors for poor eyesight. They may recall eating carrots sometimes rather than rarely or often.

Because of recall bias, this study might find a significant difference in the reported childhood consumption of carrots between cases and controls. This may be the result even if in reality there was no difference in the average diet of the two groups. Alternatively, the survey may fail to capture a true difference in dietary history. Although there is no way to prove that recall bias is occurring, the results of case-control stud- ies must be interpreted cautiously in light of the possibility that differential recall may have influenced the findings.

• 11.5 Analysis: Odds Ratios Researchers considering using a case-control study approach must become familiar and comfortable with the concepts of odds and odds ratios. The odds ratio is the mea- sure of association that readers will expect to be reported for a case-control study. Odds are the most familiar from their connection with betting. A horse with an equal chance of winning a race (50% likely to win ) or of losing a race (50o/o likely to lose) is said to have "even odds/' or odds of 1 (50%/50%). Similarly, a case-control study com- pares the chance of having had a particular exposure to not having had it ( FIGURE 11-3) . If 50% of the participants in a study report a history of exposure and 50% report no exposure history, then the odds of exposure are 50 %/5 0 o/o , or 1. If 25% report having the. exposure and 75% do not, then the odds .are 25 % /75o/o,'Or 0.33. If 2 % re- port being exposed in the past and 98% report not being then the odds are 2%/98%, or 0.02. ·

Case-control studies use, as their main measqre of association, the ratio of the odds of exposure in cases to the odds of exposure in controls. This is called an odds ratio (OR). FIGURE 11-4 shows a 2x2 table for a case-control study. Two-by-two tables compare two dichotomous (i.e., yes/no) variables. In the 2x2 table for an unmatched case-control study, the columns are for disease status (case = yes and control = no) and, the rows are for ex- posure status (exposed = yes not exposed = no). All of the participants in the study are assigned to one of the four resulting boxes: (a) cases with an exposure history, (b) con- trols with an exposure history, (c) cases with no exposure history, and (d) controls with no exposure history. As a check, the total number of cases in the study should be a + c, and the total number of controls in the study should be b + d.

The odds of exposure in cases are the numbe_r of cases with exposure (a) divided by the number of cases without the exposure (c). The odds of exposure in controls are the number of controls with the exposure (b) divided by the number of controls with- o ut the exposure (d) . Simple algebra shows that the· equation for the odds ratio of (a+ c)l(b +d) can be simplified to

OR=ad be-

11.5 Analysis: Opds Ratios 59

100%

d%

FIGURE 11-3 Odds

FIGURE 11 Odd? Ratio (Point Estimate)

60 CHAPTER11 Case-Control Studies

'' If the odds of exposure are the same for cases and controls, then OR = 1. • If the ()R is greater than ·1, then cases. have odds of exposure than

implying that·the exposure was risky. · • I£ the OR is less than 1, then cases have lQwer odds ofexposure than controls, im-

plying that the exposute was. protective. .

The 95% confidence interval shows whether an OR is statistically significant (FIGURE 11-5).

• Suppos·e the95% confidence-interval (95.% CI) overlaps OR= LThisoccurs when the lower-end ofthe coniiden.ce interval less thqn 1, while the higher end of the confidence interval is greater than l, suggesting risk. In that situ.ation, the OR is :said to be not statistically significant., and the exposure and disease are deerrted to have no

• If the entire 95% interval is less than l, then the ·OR is statistically $ig- nificanr,. and the exp·0sure is; deemed protective.

• If the entire o/6 corifidence.interval is greater than 1, then the' OR is statistically significant, a,nd the. exposure· is deemed risky.

Computer- and Internet-based statistical pr.ograms are,available that cakulafe the point estitnaJe for the OR (.the value .()f adlbc'}, .:,1long with its correspon<ling 95 o/o con- ·fidenceinterval, when the valuesJor·a, b, c, and d. are entere.d. Sample is shown. in FIGURE 11-6 .. One .example hM an odds ratio ol1.58S' an'd a 95% confidence inrer- val:o£'(1.027., 2.453 ), implying that the. exposure was risky. The other example has an odds t;:rtiQ of 1.158 and a 95% confidence interval of (0.650, 2.Q66), 5l.nce the· 95% CI overlaps 1, the association 'is not statistically significant. The CO;rrect conclusioni'n this e:Xaniple iS: that,.therejs no ,associatiov .and the disease.

FIGURE 11-5 Interpretation ohhe Odds Ratio Based onlts 95°/ci confidence Interval

'11.5. Rafio,s 61

Lower Cl = 0.650 Upper Cl =2.066

FIGURE 11-6 Examples of Odds Ratio calculations

For a case-control study, it is incorrect to say·that "the exposed had a higher (or lower) rate of disease than the unexposed" because the rates of disease in exposed and unexposed. participants a:re not known. Case-control studies recruit participants be.- cause they have or do not have a disease. Usually·about 50% of participants in a case- control study are cases even if cases make up Jess than 1% of the community from which the study population was drawn. As a result, the prevalence of disease among exposed persons in the study population could be 70% even when the prevalence of disease among exposed persons in the community from which participants were drawn is less than 1%. Because th e study population is usually not representative of the com- munity as a whole, case-control studies are unable to calculate rates of disease among the exposed and not exposed.

Case-control studies are, however, able to examine odds of exposure among the diseased and the not diseased. For case-control studies, the orientatio.n should always be from disease status to exposure history, and from odds rather than risks or rates. So the phrasing of results should always be that '4 cases had greater {or lesser) odds of exposure than controls."

• 11.6 Matched Case-Control Studies . Individually matched case-control studies require the calculation of a matched-pairs odds ratio that uses a special kind of 2x2 table that shows how often pairs of cases and controls had the same or differ-ent exposure histories ( FIGURE 11-7) . When both the

62 CHAPTER 11 Case-Control Studies

PR=btc

FIGURE 11-7 Ratio

case and controUn a m<rtched·pair his'tory of exposure or no exposure, their c;.onq;Jrdanv. (cells a .anp C,]),_ th{!y do not ptovld¢' much use-

Jul information about the ·potential r elationship the exposure and 'the dis- ·eas·e,. However, w:lten histories for a p:9.it dt'Sco.rqcl/rtt (cells .b -and q), they provide .an indication about wheth:e:r the exposure is: likely t:o he. risky or protef;- tiye. A Qt nuil{ber ofting:s the was eX'pos:ed and the. G,<5il'tto1 w'a.:S: not (h,) to the number ofti:mes the control was,.exposed·arrd the cas.e 'was not (c) provides l}n for a spe¢ial type ofoJlds tatiq.

If bl c is greater than · 1 and the 9 confidence-irrterval ( ca.lculated using all four categories in the. ngurl;!, indudi11;g the pairs) not oveda,p thert ca-ses are more than controls:ro· have been exposed. This impliesthatthe;ex:- ,. posure is risky.

• Ifblc is less;th'art 1, and'the 9'5% confidence interval does notovetlap 1, then·'cases- were less likely .than controls to have had the: exposure,. This implies the ex:- pos.ure is

• If .the 9So/o confidence interval indudes 1 ,. then there, is no assodati:on between the diseas·e a·nd the exposure.

For additional information :about how to analyze individually matched (matched- ·paits} _cas,e"'conttol studi¢s., ,,¢on!iult. a that 'addtes;$es pairs me.thods,·and: analysis,

11 :.6: Case-Control Studies li3:

Coh,ort Studies

A eoh:ort study-follows participants through time trJ' cafculate ·th:e ·'rate at which new an,i:L

Approach

Objective

Ptlmatystudy question

. Populati.ofl.:

When tq·pse thfs approach

Pr<J>spectfve or Retrospective conofl Compare rates :of new-(lm:ident): dis¢.ase tm.;p'eople.-,wtth different e,xpofiu'r'e histortes; Is exposure assotiated with an fncr:e:a,sed .i ndden€e of

must be similar

Becau-se tHe goaT is to :iook for incident diseas.e; no ,one..,can:have the disease of inferestaMhe oft he.·

Arf:e'*JYosur:e. ts .r:ela;tivety .. uncommon, but.a soutce:of exf;)osed indiViduals isavaila.vle;. A source ofindLvJdO<\II$WltFi the. · exposure is: available .

. , FIGURE 12-1 Key of CuhorfStudies;

' ;,.

Follow'a ,p:opuJatron 'forward..in tim€Ho.I.ook tor:new nnci.oent)

fs exposure assdtiated with an . lncreased.inci.dencEhOf disease? PartiCipants must be avaiTabJe for foJiow-up·.rnt;>nth§.·or .aft¢r. entailment. · The study participants must n·e · r=easomibly representative .of the populatiofl trom tneywere. drawn_. rhe gpal is to examine ml:llt,iple axposure:s:and multipleoutcomesj

-andfirne is HOt a .concern. $dequ:ate time and

m<?ney for the study:

.'

65.

First steps . l. ldentify·a source of individuals 1. Select a population. with the exposure. 2. Select the exposures and

2. Decide what type of unexposed outcomes that will be assessed. individuals_will be an appropriate 3. Decide how often data will be comparison group. collected.

4. Develop a strategy for minimizing the bu.rden of participation and maximizing . benefits and incentives .

What t(;) watch Participant drop"buf:> out for (prospective studies) or missing

records (retrospective studies) Information bias, in which the Potential data management exposed participants are more challenges if lots of information thoroughly examined for disease is collected at many points in than unexposed participants time.

Key statisti!ial Relative. RR) Relative risk (rate .ratio, RR) measure

FIGURE 12-1 (continued)

• 12.1 Overview A cohort is a group of similar people followed through time together. (See FIGURE 12-1 . ) Health research makes use of several types of cohort-based study approaches. All co- hort studies have at least two measurement times.:

• An initial survey that determines the baseline exposure and disease status of all participants

• One ot more follow-up assessments that determine how many participants have developed a new (incident) disease since the initial examination (FIGURE 12-2)

Because information is collected from individuals at multiple points in time, researchers can know with certainty which exposures were present in individual participants be- fore the O:nset of new disease. This information allows for the identification of poten- tially causal exposures. ·

• 12.2 Types of Cohort Studies Cohort studies can come in many forms. For simplicity, this chapter will group co- hort study designs into three categories: retrospective, prospective, and longitudinal

66 CHAPTER 12 Cohort Studies

FIGURE 12-2 Framework for a Cohor:t Study (Thl:! a, b, ,, and d co(respondcto. the· equati0n StloWI'i Jf:l Flgt:Jre l2·8.)

(Figure 12.-1 )" Both cohort studies :and prospedtiue cohort "Stu_dies. recruit partici pants. based o.u their exposure status. One group ofpa,_tticipc;tnts is.tectuited be- ca qse they aJ?e known :tQ. have had :a parti,ufar ·exposure, A s-ecorrd group is· recruited because they are known :rr:ot to have been exposed'. Retrujting hasedn:rt es:posut·e stat- us-makes retrospective and prospective cQhort studies :the·nptimalstudy :approache:S for u_rtcom.mbn exposures. · ··

The-members of the two comparison groups for hotht}"pes of studies :should he similar except fot their :F6r e·xample:

• A cqhqrt study might co·mpa.re i:tl'QUstrjal workers expqs:ed ro to workers:in a plant·that doesnot us:ethe. c:hemicaL lt would not be helpful, ever, to compq.te factory woikets to office

• A cohort st udy might compare, health outconres in children with. high blood lead levels and lqw :blood lead levels whq -atten.d thesarh_e It: wou1d not be_.as however, to exami11e the · impact of'bJoud 1ea.d lev.els if the ex- posed st4dent ·ftotn one and, the were from another scho;ol. Any differences, in health observed might he -due to differenc:es in socio- economic rati-rer lead expqsur.e-.

12.2 Types ,ofCohoriStudies 67

FIGURE 12"3 Times of Baseline and Follow-Up Data Collection for Cohort Studies

The key difference between retrospective and prospective studies is when the baseline measurements are established ( FIGURE 12-3).

Retrospective cohort studies use baseline information collected at some point in the past and follow the cohort to another point in the p-ast or to the present. Retrospective studies establish baseline information from birth school records, medical files, occupational records, or other sources that may be decades old. Then the researcher matches the baseline records to later files or to information so- licited directly from the same individuals in the present. For example, a retrospective cohort study might track down two groups of young adults in equal numbers: those born at a particular hospital in-a particular year who had low birthweights and those born in the ·same hospital in the same year who had normal birth weights. The aim could be to see how birth weight influenced adult health status_. Similarly, a retrospec- tive study might track down the causes of death after retirement from the armed ser- vices for soldiers whose military records indicate whether they served or did not serve in a particular deployment zone.

Prospective cohort studies have a different time orientation. Prospective studies collect baseline data about exposures and outcomes in the present and follow the co- hort to some point in the future.

Bec:ause a1l cohort studies examine incident retrospective and prospective studies must be able to demonstrate that the outcome of interest was not present in any members of the cohort at baseline. A retrosgective cohort study that looks at the causes of death after the baseline assessment will have no trouble proving that the out- come-death-was not present at the time of the initial assessment. It is more chal- lenging, to conduct a retrospective study when the outcome of interest is a condition that may have been present.at baseline but not documented.

L ongitudinal cohott studies follow a group of individuals forward in time but do not recruit them based on exposure status. Instead, participants are recruited based on membership in a well-defined source population. Longitudinal cohorts may follow all the residents of one town, a representative sample of members of one professional or- ganization, or a cohort of students. recruited frorh the same university.

68 12 Cohort Studies

:

FIGURE 12-4 longltudfnal Studies

Individual participants are assessed at baseline for several exposures and diseases. Then they followe.d forward. in time. to determine the. incidence 'fate for one or more outcomes ol interest. Longitudinal s tudies may use· a pqpulation irt whi,ch all participants star:t the .study at the same time and no one is allowed to later. Alternatively; they may use a d,ynamic population with rolling admission artd

dtbJ:>olits ( FIGURE 12-4}. For drnamic-populations,_ the time to follow up is usually base-d on individual participants' dates of enroJlment than on a fi:Kecl calendar date.

Several variants -of longitudinal stu'di®s, such as time s·eries stp.dies und p(;'(nel stu<i- ies, rneasqrethe same individuals repeatedly over time, as cohort studies do. Surveillance systems c an he used tO ·monitor qvet .an extended period us-ing continuous data -rather than discrete. time ·points. Alternatively., some types· CJf studies measure individuals from the s.ame p'dJ>l)lations at. dif- ferent points in time. They do not neces_sarily capture the same individuals in each tound of questlonjn.g, These tyfJes of studies use a; s_eries of .. surveys rather than a longi, tudinal cohort s-tudy approach.: ·

• 12.3 Special Considerations For studies? the ,first step· is identify u source· of existing records that can ptovi:de the b}:tseline data .. Irt some: cases" existing records may .also be able to videaiLre:quired follow'" UP dat-a, an·d no contact with-the individucds will be required. :Pot thatrequire contact withindividuals,.·a method for c;;:ontact- ing those i-dentifietl in historic records will need to be tested,_ and shown to resl.llt ii;l areasoriable pa:rtic:ipatlo.nrate.

12.3 lipedal Considerations. 69

For a prospective study that will recruit participants based on exposure status, the first step is to 'identify two available sources of individuals: one for those with the ex- posure of interest and one for those without the exposure.

For longitudinal studies, the first step is to select a source population . . Alternatively, if the goal is to conduct secondary analysis of existing data, the first

step is to identify an existing source of data. The secondary analysis of existing data is the most cost-effective way to examine study questions on two conditions:

• A completed or ongoing prospective or longitudinal cohort study has assessed the exposures and outcomes of interest. ·

• Electronic data files are available to outside researchers for analysis.

For prospective and longitudinal studies, decisions must be made about how often data collection will take place and how long the study (or at least the first wave of the study) will continue. Because loss of participants to follow-up before the end of the study period is a major concern of studies that follow participants.forward in time, re- searchers must develop strategies that minimize the burden of participation and that maximize interest in continuing to participate. Some studies may increase retention rates by offering participants free medical tests or other incentives. Sufficient motivation may also be provided by reminders of the significant impact of the disease on affected persons and their family members or by notifications of the important discoveries be- ing made as a result of their continued participation.

source populations haye been identified, it is time to initiate collection, as described in the next secti'on of the book. When developing the survey instruments for cohort studies, remember the importance of establishing exposure and disease sta- tus for all participants at baseline and at follow-up. All participants must complete the same assessments to prevent the information bias that might result when exposed participants are more thoroughly examined for disease than unexposed participants. A strong data management system must be prepared to link baseline and follow-up data while maintaining the confidentiality of the information provided by participants. Data management is discusse'd in Chapter 25.

• 12.4 Analysis: Incidence and Risk Ratios The goal of cohort studies is to examine the incidence of new disease. The incidence rate is the number of new cases of disease in a population during a specified period of time divided by the total number of persons in the population who were at risk dur- ing that period. Individuals who already have the disease of interest at the start of the study period are not at risk of getting new disease, so they are removed from the de- nominator (FIGURE 12-5). For example, suppose a cohort study examined the incidence of disease over 1 year in a population with 50 members and that 7 of those 50 already had the disease at the start of the year. In that situation, the denominator should be

70 CHAPTER 12 Cohort Studies

FIGURE 12-5 Incidence

43 rather than 50. If 4 of those 43 are diagnosed with the.disease during the year,: then the incidence rate is 4/43, or 9,3' per 1000 per year. (Incidertoe converted to units of "per 1000,_" "per 1Cl,OOO," or theJike_sothat they can be more:easilrcom- pared.)

Som,e.-cohort studies, especially those withdynam1cpopulations (Figure 12-4) and those that n,1il formany use-person-time as a dynQtniflatq,r. is a wa,y of accounting for different individuals in the study population being observed for dif-

lengths- of time. a study recruits iO in.9ividuals at baseHne (FIGURE 12-6). After 4 years, 6 of the 10 participants are still active in the study :and have not been diagnosed with the disease ol interest. these .6 individt.Jafs haye con- tributed 24 person-y-eans of observation _after 4 years. Suppose that 2 :ofthe-· 10 origi- nal partic'ipants .ate dia:gri<Jstd ·with the: .disease of inter:est a:t their ,study examinat-ions .. One person is diagnosed 2 years into the study, and the other 4 years Ji1tq rhe study: Together, these 2 individuals contributed G per:s·on-years .of tion to the study; However, cmcethey :are diagnosed and no lenger at risk of getting thedise·?s·e, they a.re noJonger ahle to c:.ontribute p¢rs.on-y·ears. to the deno;mihator for

12.4 Analysi£:. 1ncidence and Risk Ratios 71

10 person- years of -

· :, observation .. /: -1 year

, incidence: ,-· p_erson- .

_: ·years ·' .

FIGURE 12-6 Person-Year Analysis

the of incidence. Two other participants also leave the study. One drops out of the study after the second year but before the third year;_ this participant is con- sidered to have contributed 2 person-years of observation. Another dies after the first year and contributes only that 1 person-year of observation. In total, over 4 years, the 10 original participants experience 2 incident cases of disease over 33 person-years of observation. For the calculation of rate ratios a.nd other measures that rely on the comparison of incidence rates, it does not matter whether the incidence rates are meas- ured per 1000 persons 12-5) or per 1000 person-years (Fi.gur!;, 12-6), as long as all incidence rates in the equation use the same units.

An initial and simple way to compare incidence rates is to compare the rates of new disease in the exposed and unexposed members of the cohort. Excess risk, or at- tributable risk, is the absolute difference in the incidence rate. (See FIGURE 12-7. ) For example, if 10 % of the unexposed and 15% of the exposed became ill during the study period, then the excess risk in the exposed was 15 % - 10%. = 5%. This numb- er represents the additional risk of disease in the-exposed that can he attributed to the exposure. It assumes that the exposed would have had the same rate as the unex- posed if they had not had the exposure. This assumption is one of the reasons why the exposed and unexposed populations must be similar except for their expos ure status.

72 CHAPTER 12 Cohort Studies

FIGURE 12-7 ,Attrit?,utable .(Excess) Ris.k ..

The attributable risk per£ent (AR%) is the proportion of incideht among the yxposed that, ate due. to the ex;pos:ure. The percentage is by the excess, risk to the incidence rate in the exposed. F ot the preceding the· AR% is 5% + 1.5% = 3.3%. In, other words, one-third of the cases· of disease could have been prevented if the exposure was removed.

The most common mea oi assoCiation for cohort studies is the r:ate ratio (RR), also known as the relative rate, ·risk ratio, or relative risk. The Rlt compares the: inci- dence rat¢. among the expQsed t .o. the. rate in the unexposed (FIGURE 12-8.). The RRjs easy to "interP'ret;

• If the RR is ·equal to 1 or dose to 1, then the irtc:ideiit'e tate was, the sam.e in. tlte: exposed c,1nd.in the une:xposed. The e;xposure is not asss:>ciated·with the disease.

• If the RR is greater than 1; ihen the incidence tate wa:s highet in the exp.osed than in the unexposed. exposure is conside·red -risky. ''·

• If the RR i s: less than 1, then the irtcidente n1te'W<}$ lowe:r in the exposed than irt the unexpose.d. The :exposure is protective ..

the 9,5% confidence interval indicates·whe:ther the RR is staristkally signi1i¢artt (FIGURE 12-9).

-12.4 Analysis: Jntidence and Risk_Ratios 73

FIGURE 12-8 Rate Ratio (Point

• If the 95% confidence interval (95% CI) overlaps RR = 1, the lower end is in the protective range, and the higher end is in the risky range. The association between the exposure and the is not statistically significant. The appropriate con- clusion is that there is·noevidence for an association between the exposure and the disease.

• If the entire 9·5% confidence interval is less than 1, then the RR is statistically sig- nificant. The exposure is protective in the study population . .

• If the entire 95% confidence interval is greater than 1, then the RR is statistically significant. The exposure is a risk factor for the disease in the study population.

RR

Exposure is PROTECTIVE

FIGURE 12-9 Interpretation of the Rate .Ratio Based on Its 95% Confidence Interval

7 4 CHAPTER 12 Cohort Studies

For the protective exantple it). 12-9, it would p'e accurate to ·re_port that ''par- ticipants with the exposure-were half as likely to develop the disease. as· those. without the exposute.'·'' For t,he risky example in 12.'-;9 ,the repott :could.state ''par- tid pants wi'ththe exposure were twice as likely to develop the disease-as. participants without the exposl1i'e. ''

Computer- and Internet-based statistical programs are available for the calculation ofstatistics that cart be derived from a 2><.2 t;1bt:e (FIGURE 12-10),

• Incidence in the ·exposed • Trwidence in the unexposed • Attributable, risk

A- UO/ • JX /0

FIGURE 12-10 Examplrs of Rat.e

le: (20.2-o(c;, 30.9'%) lu: 34.8% (29.2%,. 40.9%)

AR = -9-;6% (-17.6°k:,, -1.6.%) -____ . .ARo/o

RH= 0.724 (0.551 . 0.951). ' -.

Chi-square·= 5.486 2-sided p·value = .0.019

le: 21.2% (16:6%, 26.7%) lu: 14.2%.(11.4%; H.5%)

AR = 7.0%. (1.1%, 12.§%) '"R-0 1 - 3'3 0-0 " (7 '.6"Ql 51 5' 01 ') M. . _10- - . ( 0 . • __ 10:, . ... f O

Chi-s·quare = 5.918 p-value = Q .. Olo

l2.4 Ami-lysls: tnddence and Risk Ratios 75

• RR • CI for the RR

Both examples in Figure 12-ro have statistically ,Significant rate ratios. One example has a rate ratio of 0.724 and a 95 o/o confidence interval of (0.551, '0.951). The expo- sure is considered protective because the entire 95 % CI is less than 1. The other ex- ample has a, rate ratio of 1.493 and a 95% confidence interval of (1.082, 2.060). The expdsurei's risky b,ecause the entire 95% CI is greater than 1.

76 CHAPTER 12 Studies

Experimental Studies

A'tz tp ct,·w4. -c;<1ntr¢l' _- cattse$ilnt1i.t'e.tt(led_:.oJ;tttcome: - ... ,._.: --·· . - -. .- - . . . ... -

-

):>opulatioll

use.this -

First steps.

. Whafto-watch . . · · ·., ... ' . . •,. "' '

Compare;outcomes ih partlclpants<a?stgn'ecl to an Intervention · · ormntmtgroup

-- - - ' . . .

· · tStf11llpt,pattidpaf1t$ ·?re::ra:nckitriiY:·a$slgned wan or group. · . .. ' -- .. ' ,., .

· ·. l: 01J ·:2. ccmsfltute<a faYntable"outcome, . . . .3: Dec-idewhah€0ntroJ is-an a.ppropfiate .cmnpari5on fer-the

fnterveti:tion. · 4:. oHndingWtll to.'prevent partidRarttS · a-ncubttt+e. f(Qhl · ·

9'$.stgneg tE> Hiterventibn ce>flti:o1 grgu;J1 ... · .

s,l, seiect t.fl'e ·rnethod .for ranaoniizrrig partiEip.ants to an group.. . -

. <' '

FIGURE 13c 1 Key Ch(;ltacteristits,of E:xpetimetnal Studies

..

77

;

• 13.1 Overview Experimental studies assign participants to receive a particular exposute (FIGURE 13-1 ). This is the primary distinction between an experimental study and other study de- signs. Observational designs (such as cross-sectional, case-control, and cohort studies) do not "do" anything to participants; they simply ask for a report on their experiences. An observational study may ask whether participants eat or do not eat an apple a day, run or do not run on a treadmill for at least 30 minutes three times each week, take or do not take a particular medicine twice a day, othave.or have not seen an ad for a health promotion campaign. In contrast, an experimental study may assign som.e or all study participants to eat one red delicious apple daily, run on a treadmill every other day, take a pill every 12 hours, or read a health brochure.

Experimental studies are the gold standard fot assessing causality. Because there- searcher assigns participants to receive a particular exposure, the exact dose, dura- tion, and frequency of the exposure are known. The researcher knows when the exposure occurred and so can compare the health status of each p"articipant before and after the exposure. The researcher can therefore assess whether the exposure may have caused a particular outcome.

FIGURE 13-2 Framework for an Experimental Study (The letters a, b; c, and d correspond to the equations shown in Figure 13-8.)

78 CHAPTER 13 Experime'ntal Studies

A typical experimental study desigp in the health sciences is. a randomized con- trolledtriql(RCT), in ·

• Some participants are,randomlyassigned to an active Intervention group. • The 'rerrtaip.ing pat:titipants-are· assigned to, a control group. • · Then all participants from both groups are foltowe:d forward in time to. see who

ha.s a favo.tabre out.c;oine anq who does not ( FIGURE 13-2 ) .. -

Randomized controlfecl trials require careful definitions of:

• The intervention • · How p?rtitipants will be· randomized to one of the e,xpo$ute groups • ·What Wpe of control is appropriate -

What constitutes· a. favorable ·outcome for the trial

T:hey also call for a c0nsiderati:on of the 'ethical challenges of ·assigning partici- ro ail eve:rt ifthat exposure iS, exp;e.ct;ed to improve health These

issues are discussed .in the follow'ing sections.. ·

• 13.2 Describing the lnterventio·n The first step in an experimental ,study ist:o· carefully define the intervention that p;;tr'- ticipants assigned to the_ active intervention group will r.eceive. The description should state exactly:

• What the intervention will be • Where and hqw p;:rrticipartts will teceive the.il)terve.ntion • When, how often, anrl for- what duration participants. will receivethe intervention • The elig1bili_cy qtiteii<:! fot paxtic!pahts

For example, a new· drug trial will d-gdare vety strict requirements for tion of the pill to, be 'ingeste.d, how :often it will be ta.ken, for how many we.eks it. will be taken, and who will meet .the case definition ft>r· eligibility to partiCipate. A ·new strength- building intervention will provlde detailed descriptions of' the procedures; and how thc;;y will change in intensity over the how participants will be coached ot supetvjsed? whete participants will engage/in the exercises, how long, the study period will and wha.t inclusion and exClllsion criteria will apply to potential volunteers.

• 13.3 D.efining Outcomes M<Jst experimental s:(qdj¢s ate sllperiodty that aim to, tha:t a new intervention is better than same type nJ controL( FIGURE 13-3). S:omestudies·aim instead

13.1 DE!f!ning 79

Goal Superiority trial NoninferiQrity trial Equivalence trial

FIGURE 13-3 Types of Success

Intended Intervention Outcome New'diet"" and Significant . weight loss weight-loss program

New drug , Improvement: therapy of the quality

of life for those with a particular disease o;mdition

New preventive The prevention vaccine of infection

'

Success The intervention is b'etter than the controL The intervention is not worse than the control. The intervention is equal to the control.

Favorable Unfavorable Outcome for Outcome for Favorable Outcome for an Individual an Individual the Study Population

-

The loss. of The loss of The proportion of those body <ib<:i!o'body who lose at least 10% of

weight and weight or failure their body weight and maintenance to maintain maintain that loss for at of lower weight loss of least 6 months is higher weight for or more in the intervention group :?:6 months for ::::: 6 months than in the control

group.

Improvement Failure to The rate of Improvement in quality of life demonstrate in the drug therapy

improvement in (intervention) group is quality of life higher than the

improvement rate in the placebo (control) group, according to a carefully defined and validated set of criteria for what constitutes Improvement

Incident Incident infection The incidence of infection infection does occurs. in the vaccinated not occur. (intervention) group is

lower than the incidence of in the unvaccinated (control) group; as confirmed by laboratory testing .

. FIGURE 13-4 Examples of Favorable

80 CHAPTER 13

..

to show that the new treatment is as good or no worse than existing interventions. "Better" could mean that an intervention is better than a current therapy at curing existing disease, or it could mean that a new intervention is better than a placebo at preventing new disease from occurring. Because the term "better'' can be defined in so many ways, the researcher must carefully define what constitutes a favorable outcome for the experiment. Because the data collected during the study are intended to show whether a better outcome was achieved, the measure of success must be stipulated prior to the initiation of the study.

For example, an individual participant's success in a weight-loss program could be defined as the loss of at least 10o/o body weight and the maintenance of the lower weight for at least 6 months (FIGURE 13-4). Alternatively, success could be defined as the loss of at least 15 pounds over a 2-month intervention period or as achieving a body mass index (BMI) of less than 30 by the end of the study period.

If the goal of a study is to test wheth er a new drug is better than a placebo at im- proving the quality of life of those with a particular disease condition, the definition of "improved quality of life" should be carefully considered and validated. This can be done by means of clinical examinations and/or survey instruments designed to as- sess various aspects of quality of life. If the goal is to evaluate whether a new vaccine prevents infection, then laboratory tests should be used to confirm the presence or ab- sence of past, recent, and/or current infection.

• 13.4 Selecting Controls Experimental studies usually assign some participants to the active intervention and the remainder to a control group (FIGURE 13-5). The most typical control is a placebo, an inactive comparison that is similar to the therapy being tested. Examples of place- bos are a sugar pill used as a control for a pill with an active medication, a saline in- jection used as a control for an injection of an active substance, and a sham procedure that is designed to look and feel like a real clinical procedure used as a control for that active procedure.

The mere act of taking a pill or receiving some other form of therapy, even if it is inert or inactive, is often enough to make recipients feel better and/or behave better. Placebo-controlled studies allow the effect of the active therapy to be examined sep- arately from the boost in health status that people may experience simply by partici- pating in a research project or receiving some sort of intervention.

It is sometimes unethical to use a placebo when .an effective. the.r;;:tpy is already available. However, if the goal of the experiment is to see whether a new therapy is bet- ter than a current one, theri it is appropriate to compare the new therapy to some ex- isting "standard of care/' w hether that is the best therapy currently or the standard local therapy. Sometimes the new therapy may be given in addition to the ex- isting therapy.

13.4 Selecting Controls 8 1

Jype of Control Active Intervention Comparison Placebo/inactive Active pill pill comparison

lnjectien otan Injection of saline solution ; active substance

Acupuncture needles Acupuncture needles inserted at inserted at acupuncture locations in the body that are not points acupuncture points (sham acupuncture)

Some other active An jnattive substance that is ingretljent indistinguishable from the active

intervention in terms of appearance, odor, taste, texture, and dellvery mechanism

Active comparison/ New therapy Current best therapy for the condition standard of care being studied

Newth€rapy Current standard therapy (' ' Some other existing therapy New tfterapy

Current therapy plus Current therapy alone new therapy

Some dose of a Alternate doses of tile rnec;:lication medi-cation

Some duration of a Alternate durations of the therapy therapy

Nd intervention New fnte'rvention Participants assigned to the control group are asked to maintain their normal · · routines.

Self New intervention Each participant's status before the intervention Is compared to his or her

.

own status after the ·1ntervention,. :

New intervention Each participant receives the new intervention for some duration and the compari?on for some duration, preferably in a random otder.

FIGURE 13-5 Examples of Typesof Controls

82 CHAPTER 13 Experimental Studies

Sometimes the goal is to determine how much of an intervention is required. For example, should the dose of<). substance be changed? (Is 100 mg of a medication as effective as 200 mg?) Or should the duration of therapy be reconsidered? (Do 4 weeks of physical therapy work as well as 8 weeks?) In such cases, varying dose$ and dura- tions may be tested and compared to one a nother. Sometimes different interventions are compared in various combinations in one study using a factorial study design.

Although experimental studies sometimes involve a control group of participants who are rap.domly assigned to maintain their normal routines, this method is usually not preferred. The approach raises ethical concerns about discouraging the adoption of healthier lifestyles during the course of the study. It also raises concerns about a type of bias called the Hawthorne effect t hat can occur when participants in a study change their behavior for the better. For example, suppose a researcher i s initiating a study of a new weight-loss program and plans to randomly assign participants either to the new therapy or to a normal routine group. In this situation, simply informing the controls that they will be weighed at the start and end of the study period will be enough to spur a sizable proportion of the control group to initiate an exercise pro- gram, start eating a healthier diet, or take other steps to lose weight. These changes may interfere with the accurate measurement of the impact of the new intervention.

When there are ethical concerns about the appropriateness of assigning some par- ticipants to a potentially risky intervention or about not assigning all participants to a potentially life-saving intervention, sometimes participants can serve as their own con- trols. Sometimes each participant's status before the intervention is compared to his or her own status after the intervention. The results of this experimental design are not as clear-cut as using a placebo because time alone can lead to significant improvements or declines in health status, especially among those who are severely ill. When possible, it is advisable to use what is called a crossover design. In this approach, the researcher as- signs some participants first to the active intervention and then the control and assigns other participants first to the .control and then to the active intervention.

• 13.5 Blinding Blinding, sometimes called masking, occurs when participants in an experimental study and perhaps some r esearch team members do not know whether a participant is in the active intervention group or the control group. In a single-blind study, par- ticipants are unaware of their exposure status. In a double-blind study, neither the participants nor the persons assessing the participants' health status· know which par- ticipants are in the active a,nd control groups.

Blinding minimizes the information bias that can ·occur if participants or asses- sors are able to assess outcomes differently based on the results they ex.Pect for an ex- posure. For example, blinding prevents participants in the active intervention group from reporting more favorable resq.lts because they expect a positive outcome. Blinding

13,.5 Blinding 83

also prevents assessors from recording more favorable results, either intentionally or unintentionally, for participants in the active intervention group.

Blinding is usually possible only when all participants are assigned to similar ex- posures. Ifparticipants in both the active intervention group and the control group are taking pills (of the same color, shape, size, and taste), if both are getting injections, or if both are participating in similar exercise programs, a blinded study may be pos- sible. In contrast, if the active intervention is a special diet and the controls eat their usual diets, if the active group will participate in exercise classes and the controls will be 'oh their OWl\, or if the active intervent_ion will include both diet and exercise ponents and the control only a diet plan, then a blinded study may not be possible. To minimize the likelihood of bias in studies that are not blinded, it is helpful to identify objective outcome measures (such as laboratory tests) rather than subjective outcome measures (such as participants' self-repo.rted feelings).

• 13.6 Randomizing A variety of approaches can be used to randomly allocate participants to an active in- tervention group or a ·control group (FIGURE 13-6).

• Simple randomization uses a coin toss1 a random number generator, or some other simple mechanism to assign each individual to one of the groups.

• Block randomization randomly assigns groups of people, such <:'\S whole commu- nities or whole schools, to -an intervention or control group.

FIGURE 13-6 Examples of Types of RanJdoniizati,on

84 CHAPTER 13 Experimental Studies

•. ra:ndomization t<!tr4.omly assigns individu:flh w!thip_ c¢rt.ai:n (such as males and females or various agegroups) to a particular exposure. This typ.e q{ randomiz.ation is useftd when simple randqm_ization may rtqt result in, e-nough members of certain subgroups being randomized to each of the exposure groups.

Reference· books that focus specifically on experimental studies provide additional de'- ta'ils. about methods for randomizati·on.

II 13.7 Ethical Considerations All research with human participants or their personal data raises ethical concerns that. must: bP:.t stuP.it+s a hig}l level Df ethical risk. In experimental studies, the researcher assigns participants to ex- posures that tlie do .hot choose an<;l rn:a_y have: be·ert unlikely to in norma1life had they nO:t volunteered to: participate in a ,research project. This means that a m,ttrlber of issves he considereci before, aQ. study (FIGURE 13-7}. For example:

e; The,p,rin(:iple of e,quipoise that should be conducted only ·when there · is genuine un-certainty about which tre:atm'ent 'witl work better;

• The prin<;:iple of disJributive ju:stiCe,implle:s thatthe.source:popuJation_mu-st bean ap- propriate ;one and that, the research must not exploit low-resou-rce individuals who are tJ.nlikely t.o ac:cess.to. the therapy ifit is fQund to be.suc¢essful.

• The principle ofrespectforpersons requires.tWothings of participants; (1) that they v:olunteer .fo.r a study without being unduly by the ·prospeCt: of being compensated for their participation and (2;) that they are able to understand what it means to. be. a research subject, inc;_luding the, possibility oJ being assigned to a :control group instead of the newinterV:enti'on. ·. . .

. . · St!)dy really

.· necessary? . ·· .

· Is the population an

appropriatE! and justifiable

one.?

Is induqement

to participate :appropriate?

D.a_ partipipp.ntp 'trt)ly ·understand tharthey may not (ec.eive the active

intervention?

it.ap:propriate to use· a ptacebo?: Is it appropriate·

to use ,some other

outcomes be monitored and

addressed?.

When might an; experiment need

to ' be discontinued early?

yYhat a, . . partrcipant experlei)ces ·:

harm after the "conelusloh

.of the study? ·

Will have .

access to the therapy . . it it is shown to be ·

su¢cessfu!?

FICU RE 13-7 Exatnptes of to Q)nsidetWh.eh Ptcl'rining an Ex·peotim€r'llal Stutly

13.7 Ethical Considerations · 85

• The principles of beneficence and nonmaleficence require that researchers balance the likely benefits and risks of the study. For example, researchers must consider the use of a placebo or another control, put in place a monitoring system for ad- verse reactions, and identify the conditions under which an experiment would be discontinued early. The study could be discontinued either because one of the ex- posures proves to be risky or because the new intervention appears to be so ben- eficial that keeping it from the control group would be unethical.

Chapter 21 discusses additional ethical principles that must be considered when plan- ning and conducting research with human subjects. Research ethics committee review is required for all experimental studies, as explained in Chapter 22.

• 13.8 Analysis Experimental studies use many of the same measures of association that cohort stud- ies do, including relative rates, attributable risks (excess risk or risk reduction) , attrib- utable risk percentages, and measures of survival. Cohort studies use these measures to examine the impact of an una ssigned exposure on the incidence of disease. Experimental studies use them to examine the impact of an assigned exposure on the likelihood of having either a favorable or unfavorable outcome.

Also, several measures are specific to experimental studies.

• Efficacy is the proportion of individuals in the control group who experience an unfavorable outcome who could have been expected to have· a favorable outcome had they been in the active group instead ( FfGURE 13-8). A high efficacy is an indi- cator that an intervention is successful.

• The number needed to treat (NNT) is the expected number of people who would have to receive a treatment to prevent an unfavorable outcome in one person {or, alternately stated, to achieve a favorable outcome in one person). A small NNT indicates a more effective intervention. If a drug is intended to prevent stroke and has an NNT of 5, then 5 p·eople have to take the drug for one year (or some other .. specified time period) to prevent one of the 5 from having a stroke. If the drug has an NNT of 102, it means that 102 people have to take the drug to prevent one of the 102 from having a stroke.

• A related concept is the number needed to harm (NNH), which is the number of people who would need to receive a particular treatment ih order to expect that one of them would have a particular advetse outcome. A large NNH indicates a more effective intervention. NNT and NNH are often used for cost-effectiveness analysis.

Another special consideration is whether to use a treatment-received approach, which lim- its analysis to the participants who were fully compliant with their assigned intervention.

86 CHAPTER 13 Experimental Studies

Rate of unf&vQrabl,e outcome in intervention

'group ..

FIGURE i3-8 Efficacy and Number to Tre.at (NNT)

Efficiency= (rc- rj)lrc

NNT = 1i(r¢- r1)

. Excess, rate of .

outcome ih c-ontrol ·g_roup

Rate of

o.utcome expected in t.l;le control :group if they haq, been·

In the i nterve)l'tion

group

FJCU RE 13-9 FtoW (jf Par{icipantS: ih (In St!lPY

13.8, Analysis H7

:

These are the participants who never missed taking a pill at the prescribed time and never missed a scheduled clinical exam. An alternative is a treatment-assigned ap- proach (or intention-to-treat approach), which includes all participants even if they were not fully compliant with their assigned intervention. ''Treatment-received analysis is better for testing the ideal-world efficacy of the intervention; treatment-assigned analy- sis is better at measuring real-world effectiveness.

No matter which analytic approach is used, the research protocol should include specific plans for promoting compliance and minimizihg dropouts. The flow of partic- ipants through the study, from the recruitment and enrollment stage through the analy- sis stage, should be included in the report for any experimental study (FIGURE 13-9).

• 13.9 Screening and Diagnostic Tests The goal of some studies is to compare two tests that are supposed to measure the same thing. In most situations, this goal involves comparing a new test to an existing one. Perhaps the new test is cheaper, quicker, and/or less invasive than the current test, which may be reasonably reliable and may therefore serve as a reference for the par- ticipant's "actua:l" status. For example, a new blood antigen test for a type of cancer may be compared to biopsy results. Most trials of laboratory-based tests can be con- sidered observational because they do not require the researchers to do anything to the

True positive

(TP)

False negative

(FN)

Fl G U RE 13-1 0 Test .Resu Its

88 CHAPTER 13 Experimental Studies

False positive

(FP) TP

· TP+FP

TN

partiaipants, other than collect a biological specimen. Other tests involve experimen- tal procedures and .. are appropriately classifie.d as experimental.

Studies of new screening and diagnostic have a <dea:r s·et of eligibility Gri- teria. They may· call for intentionally seeking out some individu;1ls known 'to have the di:sease of interest :a,nd some known tb be disea.se fre·e. An appropriate reference ·standard must be identified, and a· rationale for any cuto_ff points fo.cthe. new test .and referen!Z;e t¢st be determined. 'for exqmple, the protocol should specify the concentration of antigens.in the blood thatwilrindicate a positive versus a negative test result. A sys- tem should be put in place to ensure that the examiners-. the ·clinicians or lilborato.ry· scientists conducting the assessments-are blinded to the status of the partici- vavts; as indicgted by test.

FIGURE 13-10 shows how to calculate the sensitivity, specificity,. posidve predictive, va;lue, and negatiVce value qfnew screening o:r diagnostic in comparison to a reference standard. The sensitivity is the proportion of people who actually have a

( a<:<:ording to ilie standard) who te§t positive (using the tWw test).. 'Ihe .specificity is the proportion of people who do not have the· disease who tes,t negative,. The positive p:redictive (PPV) is th;e proporriop 'of those who test positive who acrually have· the disease. The ·negative predicitive: value (NPV) is the proportion of those who test :negative who q,ctmrlly do not. have the A good at· will have high·values for each of these measures.

SimHar §,tatistics can be us.ed to determine the e·:;l,{::tenr .of between rwo a·s- sessors ·who are eva1uating the same study participants . ..For :example,, a measunement: kpown as 'the kc;lpp,a.sJqtistie: tan indicate whether two 'radiologists examining the -set. of X-rays reach the same conclusion :about the presence or absence· of a fracture mof.e or kss oft;Cn than expected by cli:-ance. Other l;Ileasgferp:ents; ofinter-obseruer· agreemeut ( a1so ca'lled carrcordance) can also be· used to assess the validityan(Lconsis- tenc'y' of tuols. and procedureS.. More qehtiled infQttn,atiqn ab,out quahty- control teGhniquesis available-in a varietynf reference books.

13.9 S.cre.ening and Tests 89

.·

Qualitative Studies

.A qualitative study looks for the themes a.nd that emer:gefromthe ob- . set,-'vcl.'ffo'ft i;md ei!q,;luatio11i of tt tonte?ct. · ·

• · 14.1 Qualitative Study Methods Qualitative :data collection is not a detached, structured process based on a random sample of irtdividu(lJS:. Instead,, researchers, have interJs.e with (!. seJected :group of informants (often identified by means of purposive· sampling)·. The researchers are allowed. to express empa.thyaJJd, when appropriate, to be Pal'ticipc;nt who gairr·.access to and under>Standingof a community by immersing in {ts practices .. Because qualitative .researchers are so do.sely·.engaged with participants, they n eed to reflect qn how their backgrounds might bias· their· observations, and they need to be transpar- ent about. these potential limitations when reporting findings.

A carefully considered approach is used to gather and int:erpret data. For example!

• Phenomenology to understand how participants understand, interpret, and find m¢ani'ng in their own p.niqu,e life and feelings. - ,

• Grounded :theory is. an inductive reasoning process that uses, observations to de- velop ge'neral theories thst.t hJ1man behavior.

91

• Ethnography aims to develop an insider's view (anemic perspective), rather than an outsider's view (an etic perspective), of how members of a particular cultural group see their world.

A set of somewhat flexible techniques is used to ensure the comprehensiveness of the collected information. This approach may involve a combination of listening and watching, with field notes being taken to record both verbal and nonverbal cues.

Two of the most common methods used to collect data are in-depth interviews and focus groqps.

• In-depth and semi-structured interviews of individuals use open-ended questions to explore viewpoints. The interviewer is allowed to probe for more details about any response in order to gain fuller understanding of the participant's experiences and perspectives. ..

• Focus groups of about 4 to 12 people are moderated discussions led by a facilita- tor from the research team. The facilitator encourages participants to interact with one another and to clarify their individual and shared perspectives.

Interviews and focus groups are usually audio- or video-recorded, then transcribed so that the exact words (and sometimes also the nonverbal expressions) can be coded and interpreted. Interviews are often supplemented by other methods, such as partic- ipant diaries or journals. The analysis of qualitative data usually involves coding and classifying observations {sometimes using software designed for this purpose) and de- riving major and minor themes from the ·groups of observations. Reports of the find- ings of qualitative studies often incorporate quotations that express participants' perspectives and experiences in their own words.

Qualitative research can stand alone, or it can be used in conjunction with quan- titative studies. Before initiating any qualitative study, researchers should consult spe- cialty references or experts in qualitative methods.

• 14.2 Consensus Methods The goal of some studies is to identify areas of consensus and areas of contention among individuals who are experts on a particular topic and/or a particular community or or- ganization. The results of the deliberations are then used, for example, to select research priorities, to identify best practices, and to agree on plans of action.

Several techniques have. been developed for shaping these conversations and the resulting conch1sions. For example, the Delphi method_is a structured decision-making and forecasting process in which participants engage in several rounds of:

• Completing individual questionnaires • A facilitator summarizing ang sharing the responses

92 CHAPTER 14 Qualitative Studies

• Panelists reconsidering their perspectives after reflecting on the opinions expressed by others

The goal is for each iteration to move the panel of experts closer to agreement.

• 14.3 Program Evaluation Program evaluation includes a variety of approaches for examining processes, and/or outcomes of projects (specific, time.-limited activities)., programs (ongoing groups of and/or policies. The goal of these assessments is usually not to identify what is being done correctly or incorrectly, but to provide feedback about what is working well and what can and should be improved.

In the health sciences, a typical evaluation begins with a meeting at which stake- holders describe the purposes of the program, how it was intended to function, how it is actually functioning, and what they themselves hope to learn from the assessment. Based on these conversations, an evaluation approach is selected. Evidence is gath- ered from a variety of sources, possibly including a review of existing program documents, surveys of stakeholders, interviews with key informants, and observations at program sites. All the evidence is then reviewed and categorized., perhaps using a framework like SWOT. SWOT identifies strengths (internal organizational strengths), weaknesses (internal organizational limitations), opportunities (external-strengths), and threats (external limitations, which might be political, economic, sociocultural, technological, environmental, or legal). Finally, practical suggestions .ateinade based on the conclusions of the assessment. · ·

A similar process can be used as a component of other form-s of evaluative re- search, such as:

• Needs assessment • Cost-effectiveness analysis • Health services research, which examines factors related to the types of health

services and providers available to a population, the organization and financing of those health services, and the impact of governments and policies on popula- tion health

14.3 Program Evaluation 93

Designing _the Study and Collecting Data

Identify study

question

Select st udy

approach Analyze

data Report

findings

The third step in the research process is to develop and implement a detailed study plan. This section describes how to create a protocol for primary, secondary, and tertiary stud.ies. o Overview: developing a proposal and protocol

Primary studies: collecting new data • Selecting a sample population

Estimatjng s-ample size o Developing a questionnaire o Surveys and interviews o Additional assessments • Ethical considerations • Ethical review and approval

o Secondary studies: existing data sets o Tertiary studies: sy$tematic reviews ar:rd meta-analyses

Overview: Developing a Prop·osal and Protocol

A proposal isu$uaJZ'Y lor afipr<jval. Apto"" ·tqc9l de.tt!filed tke a,ottbns {/:J,qttk!ilti?t?=til.ke:n ,r!,U.tlng thtk imp leme:nttt'#on cr,f :S,t,i/t_·f$y.

• 15.1 Overview of Research Plans by Study Approach a study questio.n and a :Stqdy approach have heen the next .. step is to

create a detailed research plan. The components of this plan will vary hat ac- cording to the study approach (FIGURE 15-1 ). For. the collection oi new data froJn indi- viduals, the researcher n eeds to::

• Develop a questionnaire and ofher data. collection tools • Identify an_ app:ropriate w·ay to t¢.cruit • Select methods for collecting. and recordingre$ponses from participants

Prepare an for a r ¢view comnlit.tee

1£ existing data, will be analyzed-; an appropriate data file must he,identified and the data, set and sQpporting materials If a. literatu.te review will conducted, search strategy must he de-fined,, :eligible· articles identified, :and reie.va nt information £r,ptn each extracted into .a :For a'il study it is helpful to. create: a protocol that will guide each step ofthe-data collea.ti0n and management process ..

97

' Tertiary (review e)Cisting • -. I · -

,.-----'-___:.._., ! ·,..-------'; 1 Define

Define. study

question

. I I -search

strategy & eligibility

criteria articles for eligibility

FIGURE 15-1 Research Plans for Primary, Secondary, and Tertiary Data Collection

• 15.2 Resources for Research

Write& report

A first step in creating a research plan is to assess a ll the resources available for there- search project and all the resources that are expected to be required. The aim is to en- sure that the anticipated resources are adequate for the intended study design. Many research projects require little in the way of material resources. A secondary analysis or review article may require only access to a computer, a statistical software program, and a decent collection of electtoruc journals. Some primary studies incur only relatively mi- nor expens,es, such as the cost of photocopying a limited number of questionnaires. Other primary studies may become quite expensive. They may, for example, involve travel to a distant field site, laboratory testing or other clinical assessments, lengthy du- rations of d ata collection, and/or the hiring of interviewers and data entry personnel.

Money and materials are not the only resources to consider. For many studies, the most important resources are the individuals who are available to their time, expertise, and/or connections to the project. (See ChapterS for information about working with collaborators.) Other resources may include things like access to:

• Potential study participants or data sets (perhaps through a personal·contact, a pro- fessional organization, a community organization, or a patient or client database)

98 CHAPTER 15 Overview: Developing a Proposal and Protocol

• Laboratory space, office space, or a meeting room • Equipment, suth _as COIJ1puters· and copying inGtchines

• 15.3 Funding Sources and Budgets Although not all research projects require financial support, sometimes projects need outside sources of-money, or, at ·a minimum, they would be significantly enhanced by them. Common sources of funding are in_terna!,grant$ from a schoe<)l or employer and external grants from private foundation-s,.- government agencies, o'r other sources.

A proposal must align with the goals of the sponsoring agency and its typical fund- irig level. Some organizations support only research focused on a very specific disease or population, and others are much more general in s-cope. Some -student..:focused awards consist of only a tew hundred dollars, and some government agencies distrib- ute 'millions of dollars to established researchers.

Granting agencies prefer to fund research projects that will answer well-defined and significant study questions and that ha,ve ·a budget appropri::I;te for the work to done. The budget should cover all the-essentiaLcosts-of the researchprqject without being excessive in its total ·amount ot in any A student or tt_a,inee applying for a small grant may want to request funding for only basic direct expenses, such as travel and A larger grant may request:

• Salary support for core members of the research team • Stipends for consultants • Funds for the purchase of equipment and supplies • Funds tot administrative costs (such as p<:lynients for facilities usage, utilities,, com-

munications., and support staff) • Junds forcompensatirtg or reimbvtsing the ofstqdy p,articipants

It is not unusual for the funding cap from a source such as a competitive student resean;:h award to be lower than the actual amount requited for aproject, In this sit- uation; the r-esearcher should shcJw in the gra.ht application which .expenses will be paid by the new grant, if funded, and which will be by other sources. When. submitting a proposal, the researcher .should also ascertain the absolute minimum amount ot support require.d .and, if adequate funds cannot be secured, be prepared to abandon a project (and forgo possible offet<S 0£ financial support for specific. project) . . For example"_ if transportation and photocopying costs will be $800 and funding is secured Jor 'only $250, the; researcher may .decide to use the time, e:t1ergy; and money elsewhere;

Funding rates are often extre-mely and prQcessing time vades. Waiting for funding can stall projects for lengthy periods of time.

-15.3 Funding Sources .and Budgets 99

• 15.4 Research Timelines and Responsibilities Most research proposals and protocols include a fairly detailed schedule for the planned research project. It is therefore helpful to:

• Create a list of all the steps from planning the study through the dissemination of results

• Create a calendar that shows-when each of these steps is expected to begin and end, • Set deadlines along the way that will help ensure that the project stays on track to-

ward timely completion

The schedule will need to be somewhat flexible because predicting how long some steps will take can be difficult. For example, waiting for ethics approval or for the dis- bursement of funds from a granting agency might take several months instead of sev- eral weeks. Data collectionmight be completed far more slowly (or more quickly) than expected. Data entry might take much more time than originally anticipated. Additionally, relying on collaborators to complete some aspects of the work may re- sult in delays. This is especially likely when the lead researcher is not in a position of authority. For example, a student researcher might not be able to push for a faster re- sponse when a supervisor is slow to provide feedback. Sometimes these holdups are not a major concern. However, missed deadlines may be a serious problem when some collaborators have inflexible schedules or are completing degree requirements.

Research projects proceed most smoothly when all research team members have; the same understanding about each person's roles and responsibilities. The protocol should include dates for the completion of all tasks and the incentives or reminders to be used to encourage careful and on-time completion. It may also be helpful to iden- tify a process for resolving conflicts. Sometimes one person, often a senior researcher, is designated to adjudicate delays in the submissions of agreed-upon deliverables, dis- agreements about the interpretation of the protocol or the nature of an assigned task, and other differences of opinion or awkward situations.

Universities, hospitals, and other institutions typically require one researcher to act as the primary investigator (PI) and to accept responsibility for guaranteeing that:

• The protocol is followed. • Any adverse outcomes are immediately reported to the institution's research ethics

committee. • The budget is properly managed.

In some situations the PI is the person doing the, greatest amount of work on the proj- ect, but many institutions require a senior employee to be designated as the Pl. For example, some universities require a professor to be listed as the primacy investigator on any research project that involves human subjects, even if a student is taking the lead role in the conduct of the project.

1 00 CHAPTER 15 Overview: Developing a Proposal and Protocol

Backgrountl · ''- • B.rlef sumrnaryof whadsAlready.kn0wn about the tqpic.

. • -Uterafure review, With dta'flons of the previous w.o'ikoJ other· • 'St:nnmazy o.rme owii 9riq c:mSr. preliminary results

(it.aJ:!fJJica'I:JJe} - • Purpose' of • Signitltance and tmpartarrce:ohhe new. proje-Ct • Definitlon ot key terms

• StU4Y.:.tf8$i.gn,_ • pop4Iation (fQl' new tte1ta t:pllecti.o.nl dfdata 'soprte (ftir' analysis Qf,oexlsting • sampJrpg .. Recruiting procedures·{for new data collectian) • neti.nition and measuremenh'Jf k'€y variables · • tlafq prQtedu(es • (ff!:aPpJita:btg)

Ana·lysis plan • p,Jgn ' plait · -

• (s-atattes . . • tile. {sgJ:h as artq

and paper)<and _ · ·· · • phones; and lnternetan;ess). • Trave.I ori·cars;-park'ing, and p-ossibly:· airfare, hotels,

fo<;>Cl) · · .-.. ·' . ' · • <:!3,?\S 9hg tp the-Jnstitution by

funding:agendes)

Re_seArcl:i'er: il)fO:mla.tion.JsOf;h as a otosketcl;l, cv; or: resume) .. " ' ..

• rq$t!'tJtn.ent .• review.('-1RP! i(;atrotJ ·a.nq

---

FIGURE 15-;2 Typical PrQposal .(:o_ntent

1-5.4 Resea rc:h nmeline$ an.d Respprisibilitres , l a 1

• 15.5 Writing a Research Proposal Most granting agencies require the submission of a formal research prQposal, and stu- dents and employees conducting research are often expected to submit a proposal for review by supervisors. Most research proposals (FIGURE 15-2 ) contain:

• A brief background that explains the importance of the proposed project • A goals statement • A des<;:ription of the methods that will be used • An analysis plan • A plan for the dissemination of findings • A timeline • A budget • Basic information about the researchers

The application guidelines usual\y provide specifications about what content should be included in a proposal; how the-proposal should be organized, and how long each section should be.

• 15.6 Writing a Research Protocol The research proposal serves as the backbone for developing a very detailed research protocc>l: The protocol explains the exact procedures that will be used for every step of the research process. For a primary study, the protocol will explain, among other things:

• The exact processes that will be used for contacting and recruiting participants. • The desired sample size and the steps that will be taken to acquire an adequate

number of participants. • The procedures that will be used to obtain and document informed consent

(or to document refusal to participate). • The exact questions that will be asked (and, if interviews will be used to gather in-

formation from participants, the exact instructions for how those questions will be asked).

• The exact codes for the entry of various responses to survey questions (including missing responses) into the computer database.

• The exact steps that will be. taken to maintain the confidentiality of any personal information that might be contained in that data set. ,

• If applicable, the research protocol must also describe in detail any laboratory procedure that might be used.

For a systematic review, the protocol is quite different but equally detailed. It de- fines the exact criteria for an.article's eligibility for inclusion in the review. It spells out

1 02 CHAPTER 15 Overview: Developing a Proposal and. Protocol

exactly how articles that :cannot easily be classified according to the eligibility criteria Will be handled. }?or ail $fudy the protocol shoufd anticipate lik¢ly dilemmas_ and areas -of confusion and address them ,as c0mpletely as possible.

Ideally, a protocol

• Fully describe all the procedures that will he used for data collection <;).iid • Provide,deta:ils about the, responsibilities ofeach member of the research team. • List the deadlines 'for corhpktion of all the in the· process; • Describe the mechanism for updating, any part of the researc.h plan, should the,

.need ;;t:rise after approval of the iP.ltial ptbtocol. (Significant. to the., protocol may need to be approved by.all relevant researoh ethics committees he- fore they Cqn be implemented.)

A strong protocol provides enough detail that another researcher could easilyrepli- the study, be, det:ailed enough th.aJ theyntire methods section oJ arty

paper that will result from the project could he written before data :collection begins.

• 15.7 Preparing for Data Collection Before initiating data "coUection, ma-ke sur-e all preparations have finalized.

• Are all supplies and equipment ready for use? ·• Have all eolla,borator.s approved ot their roles, and

deadlines? • ]-:lave ·all cbll?bot.at_ors cQmplete,d ethics trail).i.ng? • Have 'the final versions ofall study documents (such as the informed consent state-

ment ;;tnd the questionnaire) approved 'by ·e-ll relev:;trtt ethics .tqhlrnit:- tees, if required?'

• Bas th¢ dat.amanage,ment system peen te'sted <:J.nd found tq. b·e tel:i.abJe? • Ifapplicable, ar·e all participating lahoratories ready to' begin processing samples?

The following provide irtf.ormati:on about some of the steps in ·preparing for primary data coilection: identifying a sample developing a qll.esti,onnaite and. ot.her fo(ms of and ensuring that ethical standards are, followed:

l5.7 Prep,aring-forOata toneoiciil 1 03

Primary Studies: Seleding a Sample Population

A. -source of' study partkip.ants for primar,'y_ studies" should be idrmtified ear{y in- t/ie · ·

• 16.1 Types ofResearch Populations At le-ast four _ditfetertt types_ of populations must he when ptepadng_ to .col- lect data (FIGURE 16-1). (Several.different names are used to desc::tibe,thes:efour entities, but the concepts are the f()t all health fldds.)

• The broadest group is the target population to which the:results·qf the·study should be applicable.,

• The source population-is a weH-define:d subset of :indi_vidu9-ls from the tqtget population.

• The sa:Jttplepopulati:on consists ofthe [JTdividuals 'from popula#on who. are asked to participate.

• The 'Study 1JOP'ttlation c-onsist,s_ of the Plembers of sal]lp;le; pqpulatioh whQ consent. to participate in the study.

105

. po.puJC!tion::that the , to ': ' · ·

. .. .. ,•

FIGURE 16-1 The Study Population Is a Subset of Individuals Sampled from a Source Population

• 16.2 Target and Source Populations A well-defined study question identifies a target population to which the results of the study should apply. A target population might b(i! quite narrow. The goal might be to identify the cause of an outbreak of a drug-resistant bacterial. str.ainin one wing of a long-term acute care hospital. Or the goal might be to measure the prevalence of binge drinking on one college campus. Alternatively, the target population might be rela- tively large: all adult males, a whole country, or all people with type 2 diabetes. Unless the target population is very small, measuring the entire target population or even randomly sampling from it may be impossible.

Instead, a more specific source population (sometimes called a sampling fram e) should be identified. Ideally; the source population consists of an enumerated list of population members. For example:

• All women with a breast cancer diagnosis in the past 2 years who are indexed in a particular cancer registry

• All members of a professional sports league . • All households within 2 miles of a particular nuclear power plap.t

In each ot these examples, it would be possible for a researcher to acquire or generate a list of all members of the source population. (The list does not have to include the names of the members of the source population. The list might instead include registry identification numbers or street addresses.)

1 06 CHAPTER 16 Primary Studies: Selecting a Sample Population

• 16.3 Sample Populations Sometimes every person who is listed as a,_ member of the ,soqrce population will be ?sked, to participate-in a study, especially when the source population is smalL In this situa- tion, the soun;e populationis the sart1e as the s<:trnple population._ Howeve,r, -a. source population is often much larger than the sample size required for a study, In this situa- tion, o_nly portion of the_ source populatiQn is selected to serve as a sample population.

A variety of methods can be_ used to select the sample population. If a list of every individual in the sou.rce population is available, a computer program can be used to s.elect at random the individuals who will be asked to participate. If the sample will be from an entire city, then duster sampling: can be used to identify at random whole city blocks for inclusion in the sample :population. Alternatively, the sample pqpulation might consist of all residents in the city living on every tenth street that runs north to south. Examples of these types of probability-based samples are in FIGURE 16-2.

Sometimes a non-probability-based convenience pbpula-tion CaJ) be .selected based on the ease of access to those individuals, schools, or communities. Convenience sam- piing must always be used with caution. Convenient populatitms <;1t-e often sys- tematically different than the communities they ,are intended to represent.

No matte:r which sampling method is used, the goal is to ,end up with a sample population that is representative of the source population and, ideally, of the target pop- ulation: too. If random is the need.s to a,void the nbrt- random-samplin,g bia,s that could. occur if each individual in the source population does not have an equal tha,nce :of selected for the sarpple population. If non- probability .sampling, (convenience sampling) is the researcher must avoid the ascertainment bias that can: o¢tlir if the --coh venience pie i$ npt representative ott he source or target population as a whole. These and many other potential :sources of bias· can be eliminated or tninimize<.f with caref ul pla nning.

FIGuRE 16-2 Examples of Types of J?robabi11ty Sampling

16.3 Sample Populattons 1 07

• 16.4 Study Populations The individuals identified as the sample population will later be asked to participate in the study. The study population will consist of the members of the sample popula- tion who can be located, who consent to participation, and who meet all eligibility criteria. A 100% participation rate is extremely rare. At least some of the individuals in the source population will ignore an invitation to participate. Some who respond to the invitation will choose not ro participate. Others will turn out to be ineligible be- cause they do not meet the inclusion criteria. A low response rate may result in non- response bias if the members of the sample population who agree to be in the study are systematically different from nonparticipants. However, a less than 1 OOo/o partic- ipation rate is usually not a problem as long as the researcher:

• Uses suitable and carefully explained sampling methods. • Takes appropriate steps to maximize the participation rate. • Recruits an adequately large sample size. (Chapter 17 explains how to estimate the

sample size required.) • Reports the number of potential participants at each stage.

• 16.5 Populations for Cross-Sectional Surveys In a cross-sectional survey; the source population must be representative of the target population, and the sarhple population must be representative of the source popula- tion. The goal of most cross-sectional surveys is to describe a specific target popula- tion accurately. The results of these surveys are often used to make important resource and policy decisions.

Convenience samples rarely result in a study population that is representative of the target population. For example, suppose the goal is to quantify the prevalence of tobacco use in all high school students in a county. Designating only one high school as the source population is not sufficient ( FIGURE 16-3) . Working intensely with one school might maximize participation rates. However, the school may enroll students who are different from county students as a whole-more rural or urban, more or less diverse, or tnore or less wealthy. In such a situation, the results would not be an accu- rate reflection of health across the county. Similarly, recruiting participants for a gen- eral population survey from among the spectators at a football game, shoppers in a grocery store, or donors fll a v()Iunteer blood drive would likely r-esult in a sample population that did not represent the target population. ·

Ideally, the researcher needs some way to confirm-that the -source population is similar to the target population and that the sample population is similar to the source population. For example, the sample population for the survey in Figure 16-3 can be checked to see whether it yields a proportion of students by grade and sex that is sim-

1 08 CHAPTER 16 Primary Studies: Selecting a Sample Population

approach ·

·stt.Jdy question ..

:study method

Target population

.Sot,Jrce populatron

Source population list

·sample population

Study population

Confidentiality

survey What proportiqp ofhigh.scb0ol students in North County smoke

cigarettes? · · Participants will complete their own paper-based questionnaires. Students ingrades 9- 12 in North County

All students enrolled ih qny of 14 high schools in North County

A list of the number of students in each homeroom provideQ by each high s.chool

Based on estimated sample size requirements, 20% of hGmerooOi'$ will be .selected fr0111 the lists provrded, and all-students in these homerooms will &easked to · participate.

Eligible individuals from the sample populatton who agree to participate

·No studeril names will everbe provided to researchers; will be anonymous.

FJGURE 16-3 Population Example for a Cross-Sectional Survey

ilar to the proportions in the source county as a whole. Similarly, the sample popula- tion for a survey of the public should reflect the demographics of the target pop- : ulation in the most recent population census.

• 16.6 Populations for Case-Control In identifying pbssible particip ants for a case-control study, the first step is to find an appropriate and available source of cases. All cases must have the same disease, dis- a bility, or other health-related condition. T he study's case definition should be very clear about the characteristics, signs, and symptoms t hat must be present E>r absent for an individual to be categorized as a case. For example, a researcher may want to sdect

16.6 Populations for Case-control Studies 1 09

u.-s ;,:;i,lp_men ages 70:...79 years

Study approach

Study question

Study method Target population

Source population

Source population Jist

population

Study population

Confidentiality

Case-control study What are the risk factors for hip fractures in adult women in the

United States? Participants will be intelViewed in person or by telephone. Women ages 70-79 living in the United States

All women ages 70-79 who were admitted to St. Luke's Hospital System·in Center City with an incident hip fracture in the past 12 months

A list of the hospital registration numbers for each inpatient female age 70-79 Whose computer files indicate a diagnosis . of a hip fracture (ICD9 code= 820 or I COlO code= 572.0)

All members of the wlll to participate as cases; and each case wifl t?e .aske<;t to provide the names of three female friends in. the' same age range who might be able to serve as controls.

Eligible individuals from the sample population who agree to participate

Names, addresses, and phone numbers of potential participants will oe provided to the researcher so that potential participants can be contacted. identifying information, such as names, addresses; telephone numbers, and so cia I secutity numbers, will not be included in the file that contains questionnaire responses.

FIGURE 16-4 Population Example for a Case-Control Study

only candidates with advanced disease or, alternatively, may prefer to study only cases with relatively recent onsets of symptoms. The case definition should specify both inclusion and exclusion criteria.

Hospitals, specialty clinics, public health offices, disease support and ad- vocacy organizations may be helpful resources for locating individuals or groups of in-

11 0 CHAPTER 16 Primary Studies: Selecting a Sample Population

.·

dividuais who are likely to meet the study's case definition. However, care must be taken to ensure that the sample population is not sicker, or more or le.$·S so- cially connected than the average person who meets the case definition. (Another op- tion tn<;W' to \l.se tbe participants of 4 lar:ge longitudinq.l cohort as tpe population for both cases and controls. This kind of nested case-control study design minimizes recall bic:;ts because information abo1,1t past exposures was, collected :at the, time of the exposure and is not based on participants' memories.)

Once a source of ca·ses is ·a, valid control grovp must be selected. This is a critical decision. The controls musrbe similar to the cases in every way except for their disease status. For example, comparing older adtdt women to teenage boys, cbm- paring_people with chronic heart disease to marathon runners, or comparing big-city businessmen to men who are subsistence would not be valid. All cases and all controls must meet the same eligibility criteria except for the ones relating to disease status .. (Having both a case definition and a control definition is helpful so that.bor- derline cases·are excluded from servring as either cases ·or contrtils.) Thus, a study that targets ·septu,agenarian women should require both Qq_ses and to be women in their 70s (FIGURE 16-4). A study examining chronic disease should ch:oose a control population r.epresentative of the general public, not a population thatls unusually phys- ically active. And cases and controls 'tot any one study should be,drawn from similar geographic and demographic populations.

Controls cart. he drawn from many different types of source populations. For .some hospital-based studies, it may be appropriate to use as controls individuals hospital- ized with a condition other than the one being, studied. For some. population-based studies, random-digit t elephone dialing may yield a representative population. Or, because many people will refuse to an:swet personal questiqns over the telephone, it ·may result in a very unrepresentative In some situations., friends or family m:embers ohhecases may be the because they ar:e'likely to have similar hack- grounds .. When making this important decision, the researcher should consult a ref- :erence specifically addresses the selec::tibn .of ;ipprqprrate for case-control studies, including the possibilities for matching cases to controls.

• 16.7 Populations for Cohort Studies Identifying source and' sample flOpulations for a longitudinal cohort study is fairly similar to identifying these populations for a su.rvey (FIGURE 16-5). There. are some added conce·rns about needing tG a stable study population 'in order to retain as many participants as possible for the duration of the stvdy. For cohort studiesthatseek to exposed and unexposed populations, identifying exposed and unexposed participants is s-imilar to the steps for identifying cases controls for. ·a case-control study. ·

16.7 Populations for Cohm1 Studies 111

;

Target population -·

All children with "'. cystic fibrosis

in Canada

Study approach

Study question

Study method

Target population

Source population

Source population list

Sample population

Study population

Confidentiality

. CphoH: study What is. the incidence rate for lung lnfe(tions in children with ·cystic fibrosis?

Rarticipants' parents will be asked to log all irifedions throughout the 2-year prospective study period, and these will be checked against the patients' medical records.

All children with cystic fibrosis in Canada All children ages 2-12 years who were patients of the cystic

fibrosis clinic of University Children's Hospital (UCH) in the past 12 months

A list of all children ages 2-12 who were examined .at the UCH cystic fibrosis clinic in the past 12 months

The parents of all individuals in the source popufation will be asked if they will allow their .children to partiCipate in the'stuc;ly. ,·

:Eiigiole indivl{jUals from the sample popu:laijonwh.ose parents agtee to let them participate ' .

All guideiines and regulations for the protection· of patient information will be strictly adhered to, and only essential personnel will have access to patient records.

FIGURE 16-5 Population Example for a Cohort Study

• 16.8 Populations for Experimental Studies As is true for cross-sectional surveys, experimental studies require a source population that is reasonably representative ofthe target population. For exa:tnpley suppose the goal of an experimental s-tudy is to see whether nutritional counseling during the first semester at a residential college. prevents weight gain during the first ye9-r of college. The researcher has to recruit a reasonable cross-section of the first-year student pop- ulation (FIGURE 16-6). Some sample selection approaches would likely in a sam- ple population that was much more concerned about weight than the average first-year

11 2 CHAPTER 16 Primary Studies: Selecting a Sample Population

Target population

First-year students attending primarily residential colleges

Study approach

Study question

Study method

Target population Source population

Source population list

Sample population

Study population

Confidentiality

Experimental study Does nutritional counseling during the first semester of college

prevent weight gain? Half of the participants will be assJgned to meet weekly with a

nutritionist during their first semester, and half will have no intervention. All participants will complete nutritional assessments during the first and last weeks of the fall and spring semesters of their first year at college.

Rrst-year students at primarily residential colleges All first-year students at East State College A list of all students enrolled in the mandatory first-year seminar

class at East State College A randomly selected sample of students from the source

population Eligible individuals from the sampl'e populatiorrwho agree to

participate Nutritional counseling and assesstne.nt sessions will be

conducted in a private setting, and only essential personnel will have access to participants' records.

FIGURE 16-6 Population Example for an Experimental Study

student: asking for volunteers, recruiting students enrolled in nutrition classes, or sam- pling from among student athletes. As a result of any of these methods, both the inter- vention group, who received nutritional counseling, and the control group, who did not, would be unlikely to gain much weight during the study period. Because there would be no significant difference in weight gain in the intervention and control groups, the intervention would be deem-ed unsuccessful. Had a more representative study pop- ulation been recruited, the results might have shown the intervention to be a success.

Some experimental studies require participants to· be exposed to potentially risky substances or activities. In such studies, the risk of harm can be reduced by selecting an appropriate source population and defining strict inclusion and criteria. For

16.8 Populations for Experimental Studies 11 3

studies that involve exercise must target potential participants likely to be healthy enough to engage in physical activity. Studies of new drugs for advanced forms of cancer are often open only to extremely ill patients for whom standard therapies h ave not been effective.

In general, safety is always the top priority in designing an experimental study. All necessary steps must be ta ken to protect participants. For example, suppose all sam- pled individuals for an experiment will require the injection of a solution or the inges- tion of a pill. The researcher must ensure that participants have no known allergies to any of the ingredients in either the experimental substance or the placebo. An allergy to any ingredient must be listed as one of the exclusion criteria.

• 16.9 Vulnerable Populations Vulnerable populations in health research include some people with poor health, some people with limited decision-making capC!city, and some members of socially marginalized groups, among others. Despite the potential risks of including members of these populations in research studies, including them is the only way to study health issues in these groups. The critical health concerns of prisoners, for example, can be identified only by conducting research in prisons. Individuals with severe mental health disorders must be included in studies of psychiatric diseases. The impact of inter- personal violence on health can be understood only by asking survivors about their experiences. Pregnant women and children must be included in tests of the safety of pharmaceutical agents before they can be approved for wider New therapies for severe chronic diseases must be tested in people with advanced illnesses.

Research conducted with members of vulnerable populations requires extra consid- eration of the potential risks of research to participants. The study must be sufficiently important to justify gathering new data from members of a vulnerable population. The ab ility of every participant to provide informed consent free from coercion must be as- sured. (For young children and those with significantly diminished mental capacity, a legally recognized representative must provide consent.) Concerns about the increased risks of adverse effects from study participation must be addressed. For example, people with fragile health may have an elevated risk of injury from physical tests. Those with histories of abuse or mental illness may have a heightened risk of psychological damage from answering questions about sensitive topics. Chapter 21 provides additional infor- mation a bout the requirements of research involvin& vulnerable populations.

• 16.10 Community Involvement Some studies benefit from or require the participation and/or support 'of whole geo- graphic, cultural, or social communities and their leaders. A cross-sectional survey that

11 4 CHAPTER 16 Primary Studies: Selecting a Sample Popula1i!Jn

;

will coUett W(o,tmati6n trQm ihe ofs<.;hqo1 des·1 in addition to the consent of parents and the app-roval ofan ethics· committee. A longitudinal,s.tuch that to reciuir and fpllo:w 'whok"villages will npt a: c:ess ifformctl and infutmal tomtnunity leaders and other lnc;al representatives are not .actively in·plaiining, :recruitment; _an.cl: Communlty-b.,ised studies: Qftert.workhest'When they use research methods such as for Community'"

Paf#.cipaJory Research., .or CBPR . .A clinical study'that to pattki- panrs with ·an unusual dis·easemiay benefit greatly by; partner:ing with an active·disease ·support and advo,cacy ti:¢twork, Thes.e cohnections·should early ip't.he: res:eq"tch pla,Ming ·and maintained throt:rg}totu-the data':collection perio(L Also,, the resul:ts olthe:.study should be shared with partn_ers as soon they are availabfe,

l6c.io Community Involvement 115

.·

Primary Studles: Estimating Sample Size

An number o(stu,dy _par:tix;i'pants is -required to achieve valid and -sig- if1if!;rt¥rtt ·· · · . ,

• 17.1 Importance of Sample Size In a _popular children's story, Goldilocks explores the home of three bears. She finds that some of are to.b:smafl, some ate. too big, and some are the right size; Similarly, when determining how ma:nyparticipants .are neededfbr a study to be mean- ingful, the gail is to recruit jus:t the right number of participants. The right number is based on statistical estim·ations alYout how many p'eople are required to answer the study question with a specified level ofcertainty. If more recruited than are $fa.- tistically required, resources are wasteq, :includipg-the time of both the researcher and . the participants. If t9o few participants ;;u;e recruited1 the whole study wilLbe plmost wortltless because the sample will not have en:ough power to answer the 'Stqdy question. Few researchers ,evet have.. the luxury of worrying a,botit a,.surplus ofpartici- pants, but many to recruit a sufficient study populatioh. A shortage can make getting statistically significant results ,almost

117

• 17.2 Bigger Samples Are Usually Better There is value in having a large sample size. Large samples from a population are usu- ally betterthan small ones at yielding a sample mean close to the true population value. For example, suppose that the average (mean) age of 20 people in a total population is 39 years (FIGURE 17-1 ). If a sample of only 3 people is taken (15% of the total pop- ulation), there is some possibility that the sample mean will be close to the population mean of 39 years. But there is also a possibility that the sample mean will be distant from the population mean. Il a larger sample of 8 people ( 40% of the total popula- tion) is selected from the population, then the sample mean is likely to be fairly close to the population mean.

FIGURE 17-2 shows an alterilative display of the sample means that combines the mean age with its 9 5% confidence interval. A confidence interval is a statistical estimate of how close to the population value (in this case, the population mean age) a sample of a particular size is expected to be. When the sample size is small, the sample mean may be quite far from the mean in the total populatjon. This is represented by a wide confidence interval that reaches far from the sample mean. When the sample size is large, the sample mean is expected to be close to the population mean, and the confidence interval will be narrower.

The black dots in the center of each of the five lines in Figure 17-2 represent the five sample means from Figure 17-1. The 95% confidence interval for each sample population is represented by the lines extending from the sample means. The 95% confidenceinterval is calculated for each sample_ based 'On the sample size, the mean

FIGURE 17-1 Sample Size and Means

11 8 CHAPTER 17 Primary Studies: Estimating Sample Size

:

FIGURE 17-2 from 9 Population NCJ:rrowet 9S%.Confi:dence lntervarThan smaller

;age, of the sample, and the standard ;deviation of the sample., which is -a mea.sure of how,:: far 'i:tpa.rt t}1¢ ages t)fthe irtdivid_uals, in the sa'mple a·re,. for. the top line hi 17-2, the sample m·ean is .34 years-, and the confidence interval -stretches fr.om 18 to 50 years. Based on this: a can he 95% the mean ·in·the total population i's somewhere betw.een 18 and 50 years. lnde_ed1 the population mean 9f 3'9 year·s 'is in thit.t If h:t.milred$ of:ra'rt_dow .of 3 5n.dividtfals :are drawn ·from the total population of20,-about 95% of those samptes' will have: a 9:5% ctmfi<;l¢nc¢ that overlaps with. population mean ot 39 yeats . . About 5% of the time, ·the random sample o£':3 individuals· will by chance an unusually"young or oJd. set df inclividuals, :and the con:fiden.c.e interv:ai ·w11J not overlap with the population mean •.

If q.ll 2Q people in the·t_dtaJ p:opula,tion are induded ih the a:nalysis, no confi,dence iilterval is require:d; the population m.ean age Will be known exactly. If a sample of 18 o£ 20 members, of the populadort is dr;:rwn, the sampl-e mean will be very to 39 years. The confidence interval will be so narrow that it ·w'ill extend y'ond'the.dot representing the $'ample meari. Figp:re shows, that la:z:gePsamplesizes. ;generally result·in samp1e means that ate closet to the pup·u1atit:m mean. sam- ple sizes also. have conliden¢e intervals that are.;narro:wer.than·the con:ildenceintervals: generated by smaller More !generally,. htrger sample sizes-ma-ke it tno:t'e' likely Jhat a study will signilkanr resuhs,,.

1Z2 Bigger Sampfes.Ate Usuafly Better 119

• 17.3 Sample Size Estimation A sample size calculator is more accurately called a sample size estimator because the range of suggested sample sizes is based on a series of guesses about the expected char- acteristics of the sample population. This type of tool should be used to identify an ap- propriate sample size goal. Sample size calculators are available online, often at no cost, and are bundled with a number of statistical software programs.

Case-Control ,.

Characteristic Survey Study Cohort Study Experimental Study Are cases Are exposed Are-exposed people

What proportion more likely people more likely :til ore likely than of the population than controls than unexposed Lingxposed people has the exposure to have the people to develop to have a favorable

Study question or disease? exposure? the outcome? _outcome?

Population size 5000 - - -.. Anticipated percentage 15% - - - wi1h exposure or disease Confidence for ±3% - - - anticipated exposure percentage

RatiQ of controls - 2 ,,

- - to cases

Ratio of unexposed to - - 1 l exposed

Anticipated percentage - 25% - ·: - of comrols exposed

Anticipated percentage - - 10% 10% of unexposed with disease or outcome

.. - OR worth detecting - 1.5 - -_. RR worth detecting - - l3 -, 1.25 ·· .

. Gmfiden<;:e level 95°/6 ;95°/o 95% . 95°lo . '. "

; Power ('1 - '· .. - ,- i ·gQ'% 8o.o1o - . 80%

-: ., . ' . 1 •. EstimClte.d - -aso·cases and -18-50 exposed and :-9JJ: and

700 controls 1850 unexposed 90 unexposed

FIGURE 17-3 Examples of Sample Size Calculation

120 CHAPTER 17 Primary Studies: Estimating Sample Size

FIGURE 17-3 shows examples of t he kinds of inputs that must be provided for var- ious study approaches in order to get a rough estimate of the required number of par- ticipants. A best guess must be used for most of these inputs because accurate information will not be available until after the study has been completed. Trying a variety of val- ues for these variables will show that even slight changes in inputs may result in a con- siderable difference in the sample size estimate. When the level of certainty about inputs is low, erring on the side of a larger sample size is wise.

Also important is that the sample size estimates refer to the study population. Because the participation rate, is unlikely to be 100%, the sample population needs to be larger than the number suggested by sample size calculations in order to yield a study population of adequate size.

• 17.4 Power Estimation Another way to check for sample size requirements is to work backward from the number of participants likely to be recruited to see whether that sample size provides adequate statistical power for the study design. Power is related to the ability of a statistical test to detect significant differences in a population w hen differences really do exist. For ex- ample, suppose that there really is a substantial difference in the mean weight of cases and controls in a case-control study or that an exposure in a cohort study is truly a risk factor for the disease outcome of interest. In such situations, a study needs adequate power to detect those meaningful differences . .or relationships. When differences between means· are close to one another or when relative risks and odds ratios have a point estimate close to 1, sample sizes must be especially large to increase power.

Type 2 error (p)

FIGURE 17-4 .Power and Errors

difference

Type 1 error (a)

17.4 Power Estimation 1 21

FIGURE 17-4 ilh1strate$,·the definition of statistical powet. Population-based studies. aim to have study populations that reflect their source populations. Sometimes, how- ever, either because q£ chance or because of a study design flaw (such as too small a sample .size), the sample does not capture the true of the population. Type 1 errors occur when a study population yields a, significant statistical test result when one dnes not exist i:n the source p0pulation. The'probahility of a type 1 error is often nqted by the Greek letter alpha (d). Most studie.s aim to have a= 5%,, which corre- sponds to statistical tests using a·95'% confidence intervaL Type 2 errors occur when a statistical testoJ data from the stvdy population finds no significant result when one actually exists in the source population. The probability of a type 2 erroris often re-

to :qsing the Greek letter beta (B). Power is define.d .as 1- S'o = 20%. corre:- sponds to power= 80%.

Examples of power estimation for various study approaches are shown in FIG- URE 17-5. Like sample size estimates, power require best guesses aboutthe ex- pected findings of the The standard expectation is that a study should ha'Ve a power of 80o/o or greater. If the power is desired, the way toimprove power is usually to increase the sample size.

N,umber of unexpos.ed

of:cases· Number of controls

·crQss-S.ectional Survey tQO

of expos-eo. 400/6 with G:lisease or outcome

.o'f 2'61J/o ·wtth ·oisea$e· or outcome ·

casgs witb expo _sure

Of controls ' with €{{pbs.ure

:9:5% {1- q)

FIGURE 17-5 Examples of Power Galcul51tion

Study

250

49.0

95<1/o

122 CHAPTER 17 Primary Studies: Estimating Sample Size

250'0. ]0 l5od 70

l30Zo .,85.70/o

• · 17.5 Refin.ing the Study Approach Be prepared to rethink the study approach if the power for the estimated number of participants is not sufficient. For example, if a researcher expects to be able to recruit about 300 participants but the sample size estim-ates suggest that 870 participants will be required, then the intended study design will not work. In this situation, the study question, study approach, and/or target and s0urce populations must be refined. A new plan must be crafted that is suitable fo_r tl)..e sample that the can rea- sonably -expect to. recruit.

17.5 Refining the Study Approach 1 231

Prim_ary Studies,: Developing a Questionnaire

A qa¢stt<i m:t4ife;;-i5 ,a"tooff6t gath.r£ring ·pm;tit;lpants. ·· Questionnqires-cqn, be designed fat self:.i ep:o:rting --or a'S sc'fij!Jts, for

' '

;• 18.1 Questionnaire Design Overview Questionnaires can be ,developed fo.r almo.st a11y health topic A good questionnaire is carefully crafted fo:r a specific purpose .. Questionnaire. desigrt usually· wbrk:s best ·when it starts with the identification of the_ gene ral and specific content to be o_ove):ed py the .surv.ey instrument and thert progress.es to choosing the ty pes of questions, and answers for each topic to be assessed. The wording of e:ach question should be checked carefully, The qtJestions -wjrhin and the s:ectio-rts themselves should be in .logical order;· The formatting of the document should be easy tq follow. Prior to its u.se, the s.urvey instn11nertt h;as to be and r evis·ed as n:ec'" essary (FIGURE 18-1 ). This: chapt er pr o.V:ides details abo,ut each step. T:he.researcher: $hould·also specialty reference m}lrtliR.lS for about de- :signt·ng a valid and useful questionnaire:£ or ·a 11articular .study question.

125

FIGURE 18-1 Questionnaire Design Plan

• 18.;2 Questionnaire Content The first step in designing a questionnaire is to list the topics that the survey instru- ment must cover. This list must include the expos:ure, disease, and population (demo- graphic) areas that are the focus of the study question (FIGURE 18-2). The questionnaire may also include questions about factors that might influence the relationship between exposures and outcomes; these factors are often called potential confounders. For ex- ample, adults who smoke tobacco products may be more likely than other adults to consume large volumes of alcohol. In a study of the relationship between smoking and liver .disease, the use of alcohol could be a potential confounder. Because smok- ers are more likely than nonsmokers to drink, tobacco users may appear to be at a greater risk of liver disease than nonsmokers, even if the rate of liver disease is the same in smokers who drink as it is in nonsmokers who drink. Asking questions about both tobacco use and alcohol use enables the researcher to statistically adjust for dif- ferent levels of alcohol use by smokers and nonsmokers and thus to more accurately examine the possible relationship between smoking and liver disease. A thorough search of the literature for studies on similar topics will help in iden.tifying the range of question areas that should be included in the questionnaire. ·

It is often helpful to start with a list of the main categories of questions to be asked, and then to add detail about the specific topics to be covered. For example, a survey of physical fitness might include sections on demographics, cardiorespiratory fitness, muscle strength, muscle endurance, flexibility, and body composition. (These cate- gories would work equally well for a self-assessment tool and for a laboratory-based assessment :tool.) A survey about risk factors for breast cancer might have sections on:

• Sociodemographics (such as age, ethnicity, education level, and income) • Family health history • Personal health history (such as previous diagnoses of benign breast diseases and

the date of the last screening mammogram) • Reproductive history (including questions about gravidity, characteristics of men-

strualcycles., and use of hormones) • Lifestyle factors (such as alcohol use, exercise history, and working the night shift)

The questionnaire must include questions confirming that participants meet the el- igibility criteria for the study. For example, if only currently registered students are

126 CHAPTER 18 Primary Developing a Questionnaire

FIGURE 18--2 Question Areas

supposed ·to particip-ate in a c:toss-,settional survey, one: of the first questions should he aoout enrollment status.

The questionnaire must alSo' be ·abl¢ into key cat- egories .. For ,e:x:ample:, in case-control studies, researchers need to ask questions that qllow them to copfirm that all cases meet the, cas.e definition. Cohort series of questions about exposures 'and about disease status. The answers to these questiort,s ptov!de evidence; that participants did not have the disease olitcotpe of interest at the start of the observation period.

A fihal cojlsideratiori is the length of the A survey tha tis too sho;it will miss potentially crucial information. cA s.urvey that is too long may yield a low response (ate.

• 18.3 Types of Questions After the bro:a:d c}ftegories of questions and the specific topics to be ad- dressed in each section, the next: step is tO' d-e:Cide which type'S of questions ·ate appro- priate. E:adi question topiC $hould be assigned a, spedficiype of qu,estion,; such as a date question or a-yes/ho· question. Examples of these and othet question types are slwwn in FIGURE 18-3. types knd themselves to ·parti<:ular types It m:ay b:e"helpful to refer to Figure26-2,'wh,ich has additional in:£orrttatfon about var-iable

·Close-ended ques#ons, which allow a limitecl number of possible answers, are usually easier to statistically than apen-ehded_, or free'-resfJQnse, qpestiorrs .. The main limitation, of dose-ended questions is that they may force respondents to select answers that· do n_ot truly express their sta,tus ot ppini'qns. Open-ended tions ail ow participants to explain their selections and qualify their responses·; to give multiple af1d tcr·prqvide.responses·no.t anticipate-d by the However, open --ended questions take longer to ask a nd answer, and they may result in irrelevant

Recopirig free·-_respon$e. into obj:e¢tive-q:nd meaningful categories for statistical analysis is often time-consuming and · difficult., Open-ended .questions are often must useful wheri they _to g¢t i:mp,res:sions;or to clarify re$ponses to close-ended questions. .

q\les.tiohs come in a ·variety of date and time vail- abies . (which can be-used to calculate the length of time between :e-ventsh numeric vari-

ttnd yarial5les.

1 gt Questitm_s 127

Response Options for the Type Sampt_e Question Sample Question Date What is your birth date? - -- - -- ----

m m- d d - y y y y Numeric What is your height without shoes --·- inches

(rounded to the nearest half inch)?

Yes/no Ouring your J1fetime, have you smoked 0 Yes 0 No - more: than 10Q

Categorical! What is your sex? 0 Female 0 Male multiple-choice:

What is your favorite type of film? 0 Action/drama nominal (no rank) o Comedy/musrcal 0 Documentary 0 Other:

Categorical/ What is fhe highest level of education 0 Less than high school multiple-choice: you have completed? 0 High school ordinal (ranked) 0 Some college but no degree

D College degree or more advanced

How much do you agree with this D Strongly disagree statement:. D Disagree "No matte(how much 1 exercise, 1 will 0 Neutral not be a,bleto lose weight." 0 Agree . .

0 Strongly agree

On a scale of 1 to 5, with 1 meaning Poor----------Excellent poor and 5 meaning excellent, how 01 02 03 04 05 would you rate your hearing (without the use of a hearing aid)?

Paired-comparisons Do you prefer to drink coffee or tea? 0 I prefer coffee 0 I prefer tea 0 I like coffee and tea equally 0 I do not drink coffee or tea

Rank-ordering List the following four political issues in Number from 1 (most important) order from most important to you (1) to to 4 (least important): least to you (4): crime/safety, _ Crime/safety

foreign policy/ _ defense; _ Foreign pq!ity/defense

- Taxes/ revenue

Open-ended/ What is your biggest personal health free-response concern atpresent? ...

FIGURE 18-3 Examples of Types of Questions

128 CHAPTER 18 Primary Studies: Developing a Questionnaire

Categorical variables can have .as 'few as tw_o options (called diehotomou,_s vari- ables), like yes/no or malel'female, or they can have dozens of possib1e answer·s. Categorical variables can also be ranked ( ordina.l) ox unordered (nominal). Ordinal re- spons·es· have an order, and nominal responses do nothave any built-in or- der. For example, a question about educational level is ordinal because somelevels of edutaiion involve more ye(lrs of school than other levels do. A question about occu- pational category is no_min.al because· there is-no obvious way to rank occupations as divers·e as plumbing, farmi't1g, teaching, nursing, sales, -and taw.

Less used question types include paired-comparisons and rank-orderingquestions.

• 18.4 Anonymity Many questions .can be asked in more than one,valid way. The researcher must·de.cide which question type is most appropriate arid will best protect participants'· anonymity. For example, _participants' ages can he ascertained in a number of ways. One is by· :asking for the date o'f birth and ca,l:culating the n_umber of years between the birth date and the interview or survey date. 'However, asking for such specific-personal informa- tipn may rais·e q.bout q,nonyntlty because. birth dates tQu1d be pe_rsonal identifiers in a small population. Additionally.; the fear of providing identifying infor- mation lna •fmean that many participants will skip the qp.e$tiOn may even drop out of the study rather than provide a birthdate. To protec't participants and re- dqc·e 'feats about privacy, the researche.r could ask for e:ach. participq.nt's current age. ,in years rather than for a .da:te of birth. Even then, age in years could reveal the iden- tity ofsome individuals 'in 'the study populq.tiort. In SQtveys of Stl}dents,.tor in- stance·,- most participants will fall .into a fairly narrow age range; much younger or older student$ will stand out. In such C:a.ses, it might be. best simply to ask participants to indicate which age range they belongto (such as s;20, .21-29,

Similar decisi<;ins must be. made -about component of the_ qlJ_estionnaite that could link participants to their answers. If a name, an adaress, a hirthdate, ur other information could link a pq.rticipant to the stupy, then there must be a sqlid plan in place forprotecting the privacy of and the confidentiality of the informa- tio.ti they share (see Chapter2l).

• 18.5 Types ·of Responses Once the: types of have be.en sdected, a decision must be made. about kinds -of responses that are appropriate for ·the.question.

• For numeric the question should state exactly how specific the answers should be. Shoulq height be reported to the neare-?t inch, to nearest half inch, to the nearest quarter inch, or to the nearest centimeter?

18.5 Types of Responses 129

• For categorical questions, the response categories should be listed. Sometimes an "other'? category should be included s.b that respondents can fill in their own an- swers if none of the listed responses is applicable. Consider all possible responses for each question, and include as many as needed. .

• For ranked questions, decisions must be made about how many entries to include on the scale and whether there will be a neutral option. Most scales with a neu- tral option list 5 to 7 categories; most scales without a neutral option list 4 or 6 categories. Sample response scales are shown in FIGURE 18-4 .

For self-report surveys, a decision must also be made about whether to add a cat- egory for "not applicable" or "do not know." For questionnaires to be used as scripts, a "refused to answer" category is needed. Other examples of these types of alternate responses are "no opinion," "not sure," "hard to say/' " no answer," "I prefer not to answer," "1 do not understand," and " I forget."

Some variables require a definitive answer, so the option for skipping a response might be removed. For example, for many surveys, knowing whether the participant is female. or male is important. Most people can answer that question very easily, so an "I do not know'' response is not required. However, for a question like "When was the last time you had your blood sugar levels tested?" it may be important to know whether a person is uncertain about the answer. That uncertainty is a valid and inter- esting response. Neglecting to list " I do not know" as a possible response would force many people to choose an answer they were not sure about. This may hide important information. It may also lead to systematic inaccuracies in the data. For example, in- formation bias may occur if participants. who de, not know the to a question systematically default to providing the answer they assume the researcher wants to hear.

Strongly disqgree Disagree Neutral Agree Strongly agree ' .

Dissatisfied Somewhat Neutral Somewhat satisfied Satisfied dissatisfied

Very negative Somewhat Neither negative Somewhat positive Very posrtive negative nor positive

Poor Fair Good Very good Excellent

None: Few Some Many Very many. .

Not at all Not too important Somewhat Very important Extremely important important important .

FIGURE 18-4 Examples of Responses for Ranked Questions

1 30 CHAPTER 18 Primary .Studies: Developing a Questionnaire

• 18.6 Wording of Questions After drafting the check each question for clarity.

• each ask what it. is intended to ask? • Is' the language of each question clear and neutral? • Will members; o£ the :s.tudy population the. language? • Is: the question sensitive to potential cultur:al issues telate<::t'to language?

Also check to he sur.e that the responses are carefully worded.

• Is the: choic:e oftesp_onse de:at? • For scaled ·questions, is the rank order· dear'? (For example" is it dear that 1 is:

disagree.'' and 5 i$ "'strbrlgly agr¢e·;'? Or, alternatively,_ that 1 is 4 exs:;el.., lene? and 7 is '

• Fbr que.stiorts w·ith uhrariked tategoties.,. is the order ;;tlpha- betical or ·otherwise neutral?· ··

FIGURE 18-S lists exam pies of potential problems wtth questions:,. including problemsre,- lated to language, content, and responses. ··

undeffhed_ abhreviaUons

Ambiguous, meari-ings

Vagu.eness

'

•.

.tt myocan.flal infarction?

; ' ' .

.. .

Hq:\(e -you.ever thatyou have.::rfPH?

-::·'o: ,. .... ... .· What kintt of t)ouse tio:

··

Do, regtilarJy? ·'· ·

. ... '·"

FIGURE 18-5 Problems to Avoid

iiarti:cipants nqtkllOW'thafa a 'fancy narnefor- .

a heart · · · pot:!<n<?w that spA is . ··

short>fbir ·'fuen'ign ·prostatlt2 hyrpertrophy or that BP.H f9rostate.

- . - ,___

wHhout seei11g/9 llstof HRpropriate.: . · .. sbQu!cf

be.Uan apartment/ aa rerital;7>. ta duplex,"'·ot: ':a sfqgle-fatiilly h.o.me.:" "Regularlf.isnoi-clear: 1\Jlersonwhe:

.. _mosldays ·wegek·rtftght-, ttaily''l(n<:f,

s9y_ "-tlft '' Wh9 ex:ertrse,? may tt.Ia1fregtJJ<n: ·Jt

a typiEal·week,,' howmany;day.s :ooy.ou exerciaefor-at leasla-o fri:Jmifest · · ... ' . . . . ' .... .

18.6 Wording of Questions 1 31

Problem Double: negatives

Faulty·ps$Ur:Jiptions

Two-in-ohe

lmpossibl'e tb recall accurately·

Too much detail

1- '

Sensitive. questions

questions

Leading questions

.

FIGURE 18-5 (continued)

Example I dl.d find this visit with my doctor to be unpleasant. 0 Disagree 0 Neutral o Agree Do your gums. bleed

cleqnings? 0 Yes 0 No

Do you exercise at least 3 tlriies a week and eat a diet? OYesONo

HowmP'ny servings of carrots:did you eat most weeks when you were a child?

List any prescription medications you have taken for 1 month or longef.hiJhe past

Have you ever hit, scratched, bruised, or otherwis.e physically injured an intimate partner?

Have you ever thought that you would like to lose io or more pounds?

What is your impression of the quality of work done'b¥ the dedicated pubiR: sE!tvarits who work at the county heafth department?

Pr-oblem with the Exampie 'The' wQrding of ftijsquestibn makes it hard ·to·figure out whether a fierson who was . satisfied with a visit shorild agree or disagree.

The question assumes that everyone has routine dental cleanings .. l.f do not visit the dentist" is not an an5Wer option, a person who does not have dental cleanings is forced to answer no.

Combtnes two separate questions: one for exercise and one for diet

Adults will not be able to remember this level of detail about theirchJldhood diets.

Unless the respondent has had very few pt'escriptions, ansWering this question is iinpossible without tanking up medical records.

\

This question is unlikely to be answered truthfully if the response should be yes, (;lnd it may raise concerns about confidentiality and potential legal . requirements for reporting abuse.

. Anyone could have felt this at some point in Ume; but the question does not clarify · whether this is a long-term tpnging or a thought that crossed the respondent's mind for the first time upon reading the question.

This question clearly intends to lead respondents toward a positive answer (and may unintentionally have the opposite ·effect).

1 32 CHAPTER 18 Primary Studies: Developing a Questionnaire

Answer? Witft pd,()f':SCple

Mi$:Sing_answer options·

Over(cllJPlng ah,sW.ef 9Pt!i?P$ .

FIGURE 18-5 (continued)

is your imptessiqh · CIDOL!fthe Of

· b9 center CHy HospJta(? tJ Fair 0 Cood .0 Great 0 Excellent

' HPW many hbtJf:> Week; do_ television? 0 0 04...,7 DB. orrnote

Wh9l q;?lor 9te yblll eye?Y D Brown D Biue

ln'.a typi,(al week, now- ')ilany Y9P

d no 1-3· D 3-SD S-7

• 18.7 Order of Questions

Problem with the This questioii's response opfiqns ctea'riY 9re q - ·respcmse; is no "f:>oOf" option;·

Evell though n)95t pE;bple IJY:atch 1 hQ,l!r ofte:leVisie>n qauy,·whJch woufd,,put them in the "8 onnon? respons£H:ategOJy;, they may not-want.to:choosean, "extreme" answer. Their fmil<:cunite-respons_es Will -lea.d tQ a faLse tt)ese, resp,onse pptions may

the qpestiop as. how many hours a day they watch television.

not clear tq¢&ftlir)gs; hour; week,

it refers to, pte" post-tax income,

Paftitip'qnfs 3 d9ys a .s a. week Wjli' r:fbt kfibw response to select. ·

Ma,ny start \Yith easy o.r at lea.st general questions before m.oving to· more difficult or sensitive questions. The questions shcmld be in an order that flows na:turally· ftom topic; tq anot.her, and sirttilar questions should grouped. Sometirrles simibr questions with similar response types: can be best asked consecutively. Other times, it is better to _mix up ·such questi(ms prevel)t babitua[.ion. This occurs when respondents have given the same answer to .so many questions 1n a x ow ( . .. agree .. . . agree, .. ';) th:;tt they contim,1e to reply with ·the, same respunse has becom€ Think carefully about. how previous questions could taint the an- swer$_ to later one.s. For ex:ainple, once q participant has yatiety of options,

18.7 Order of Questions 1 33

he or she can no longer provide an unbiased first impression. Thus, the researcher may want to order questions about impressions this way:

• First, an open-ended question to garner a first impression from participants: "What do you do most often when ? " ·

• Second, a series of yes/no questions to clarify beliefs and practices: "Do you ever ?" --

• Last, a concluding, open-:ended question to allow participants to express final im- pressions: "Now that you have considered ·the possibilities, what would you say you do most often when ? "

• 18.8 Layout and Formatting The next step in questionnaire design is formatting the document so that it is orga- nized,. easy to read, and easy to record answers on. Use a readable and large font. The answer sheet should clearly indicate where and how responses should be marked. White space (blank areas) on the page is helpful. It clearly separates sections and makes the page visually appealing. If necessary, very clear instructions for skips should direct interviewers or respondents to jump over sets of nonapplicable questions.

The layout of the survey instrument will vary depending on the mechanism of data collection used. A written survey (FIGURE 18-6) that respondents complete on their own on paper may require a cover letter and instructions about how to indicate answers, such as:

Basic Information 1. What is today's date? - --- - - ----

mm d d y y y y

2. What is your date of birth? - -- - - - - --- mm d d y y y y

3. What is your sex? 0 Female DMale '

Health History (Check one answer box for each question.)

4. Have yo,u ever been DYes D No NoJ then skip to diagnosed with breast cancer? question 7,

5. Have you had a mastectomy DYes D No (either partial or complete)? .

FIGURE 18-6 Example of a Self-Reported Questionnaire

134 CHAPTER 18 Primary Studies: Developing a Questionnaire

- -____;,;_ nt h1 d_ d y ·y y y

Thai1k,.YQu 5t!-{dy rm gqing tQ a§kfng_ygu _$6me .bask qqestions:

_2. What Is yqur m m d d y Y 'Y y

' D.Feruale

Read :. Nawrm going "to ask you a fow questions about your m.edicaf a/story. . · J ,-Have ybu-ever been diagm:lsed

with breastltamcer?

4. l:::iaveyou had aJr-mstectomy (.eitheF pariia'l or-complete}?

DYes

DYes

QNo

DN.o

No, . questiem fi. ·

D ·Refused to --answer

FIGURE 18-7 Example. of -a Telephone Interview Scrrpt (for the, same questions as in Figure rs-6}

• "Select the. one answer that best describes you. " • "Fill in the oval in front _of,your answer completely using blue, or black • "Circle all options that app1y to you." • your answer in block capiralletters; as shown in the. example below.1'

An Internet:-based s_urvey has 'the benefit of aHowing the researcher to build skips inro the program so that irrel¢vant questions de> not even on the· screen; Computet- based surveys can also force a person to give an answer to one or more questions before

of a problem can result from making some fields requ'ired. Required fields £orce ,a person who does not want to provide an .answer to quit the -survey "4! that question.

In an oral survey-(FIGURE 18-7), the interviewer will read the questions to the respon- artd note the given. In additioh to the it will requite a script

that·has an opening statement,-transitions betweense-dions closing sentences, The ;nust clearly in(;liq ue th_e sections to be read al.OlJ,d) spa.c-e.s Jar recording responses, and other instructions.

After formatting, the docurnent _should be carefully' for grammatical er- rb'rs, m:isspellings,, rni'ssing questions, gaps in logic, undear instructions, formatting errors, and othe:r organiz_a.tional If the questionrta irt:; to be too long, _it may be helpful to cut some questions.

18.8 layout-and Fonnatting 135

• 18.9 Validation A valid questionnaire (or other assessment tool) measures what it was intended to measure in the population being assessed. One way to seek validity is to include sur- vey questions or modules that are identical to the ones used in previous research proj- ects. However, access to survey questions is often not possible in the -health sciences. Copies of questionnaires are almost never included in published papers and are only rarely posted on researchers' websites. As a result, new research projects usually re- quire the development and testing of a completely new survey instrument. Pilot test- ing of the new questionnaire is essential for the development of a valid and useful tool.

• 18.10 Commercial Research Tools Several widely used and validated tests are available to researchers, such as:

• For psychological status: the Beck Depression Inventory and the General Health Questionnaire (GHQ)

• For cognition: the Mini-Mental State Examination (MMSE) • For health-related quality of life: the SF-36 and SF-12

The Buras Institute's Mental Measurements Yearbook provides reviews of thousands of available tools used in psychological and educational assessment. Some of these tools are free of charge, but most are commercial products that require payment for use. For some instruments ·ptovided at no charge, researchers ha:ve to pay to have the results scored and validated against previous users of the survey instrument.

• 18.11 Translation Translation of the survey instrument into orte or more additional languages may be nec- essary if the population contains speakers of more than one language. (Translation may also· be required when an ethics review committee may require materials to be pre- sented in the committee's preferred language as well as in the language of the sample population.) Check to be sure that the translated version expresses the same meaning as the original survey. Accuracy may require the rephrasing of whole sentences, not just direct word-for-word translations.

One. way to ensure that the correct meaning is being conveyed is. to use back trans- lation, or double translation. One person translates the questionnaire from the origi- nal language to a new language; a second person then translates the survey instrument in the new language back into the original language. A comparison .of the original version of the survey with the back-translated version will reveal where the second-

1 36 CHAPTER 18 Primary Studies: Developing a Questionnaire

,langu'age ttanslatioh does' not match the intended meaning of cite original version. A 'Second approach is.t0 independently translatethe surv.ey ins1ru- ment fto!n 'the· or-iglngl to dre new1anguage. Then the:tWQ <rre 'CQ.rnpated to see which words and phrases best corwey the precise meaning lind complexity of h ·. . .1· . ··- .. t _ e :q'i;l-e:str_OJ.1'Q<}:tJ:'e•

:• 18.12 PilotTesting · A P.ilqtttJsl") or pr'¢f:i3_$t, q£ .qu¢s'tiO:o.naite help'fui fpr c)leckip_g, ·a_mon-g

• The and clarity of the questiorts. • The order of the_ questions. • The ability andwillingtress ofp:articipantS: to answer the questions, • :rrhe re:spo:_tlses and wh.ether the re:sp:onse's! match the. intended typ-es of

tespe>ns-es·. • The arnount oltime-Jt takes to .c_omple.te the-

'The, sh9'uld. to with p_ilottest. They "$hould b:e from the target _population,_ meetthe ehgibihty criteria fof' the, stud-y '(in tenns of age,.

statvs., <Jt otb;e:r- but· not pe in .. T}r¢y s_h;ould be_ asked to ·complete the. preliminary; survey :and then proYi:de feedbaek ahout -cnn-

riming, :anti 'othet' They or in..a focus group. ·

The tShould he· tevi:s'ci<l; art S'everal rounds of pilot testing may be ·re<:Juired·t.o develop · a/souRei ·survey instrument.

:

Primary Studies: Surveys and Interviews

Most primary studies collect data from individual participants using an inter- view method or a self-administered questionnqiYe. Self-reported surveys are usu- ally the least costly and least time-consuming way to gather information. interviews may allow for more detailed information to be gathered and can be accompanied by laboratory and other tests.

• 19.1 Interviews Versus Self-Administered Surveys The first decision to make about data collection is whether to have a member of there- search team interview participants or to have participants record their own answers ( FI GURE 19-1 ). Interviews may he conducted in person or via the telephone. The primary advantage of an interview is that trained interviewers record the responses, and they can ensure the accuracy and completeness of each questionnaire. Self-administered surveys : can be completed at a specific research site, such as a workplace or school or hospital, or they can be delivered by mail or the Internet. The key benefits of self-administered ques- tionnaires are that data collection from a large number of participants is cost effective and they may be the best way to get honest answers to sensitive questions.

The most important considerations when deciding which approach to use are the goals of the study and the expectations of the sample population members. Secondary considerations are cost, time, and potential barriers to participation. For example, in terms of cost:

• For interviews, the highest cost is usually personnel. • For mail surveys, the highest costs are typically photocopying, postage, and data entry.

139

· Interview

·y A member ot ll!ie. research team

.. · ·· records , ·· . . .

In-person (face-to-face)

interview

Telephone interview

survey

PartlBipants are provit:Jed

. · ret't;lr:'d;.tneir own·an$wers

Completion in presence of researchers

Mail (postal) survey

E-mail/ internet-based

survey

FIGURE 19-1 of Methods for Collecting Data

• For Internet-based surveys, the costs may be relatively low if a free or low-cost survey-hosting Web site is used.

In estimati11g the cost per participant, consider the likely participation rate. Mailing. out 10 surveys may be necessary to receive one completed questionnaire, and the budget and sample size estimates should reflect this expectation.

Time is another consideration. Asking participants to complete their self-administered questionnaires at the same place and time£an generate a lot of data quickly. One-on-one interviews may take a considerable amount of time per participant, but all interviews may be able to be scheduled within a relatively short period of time. Mail surveys may trickle in over an extended period of time and make it challenging for a researcher to know when to stop waiting for additional responses to arrive.

The barriers to participation also vary according to the data collection method. Transportation to the inter.view site may be difficult for some interviewees. Discomfort with the telephone or computer may be a challenge for others.

• 19.2 Recruiting Methods Once a data collection method has been selected, the next step is to decide on an ef- fective method for recruiting members of the sample population to be participants in the study. The goal of recruiting is twofold: ( 1) to recruit as many members of the sample population as possible and (2) to yield a study population that is reasonably representative of the sample population. With regard to: the second goal, the researcher ,should try to find a way to compare the characteristics of participants to those of the source population. For example, the age distribution of the source population can be compared to that of the study population. A statistical test can be used to determine whether the study population skewed o.ld or young or wa.s spot on.

140 CHAPTER 19 Primary Studies: Surveys and Interviews

The besunethod for initiating conta<,::t with potential pati\dpa;nts is.- re1at.ed to the-intended data tdlleotion method (FIGURE 19-2J. .

• Jf the plants: to interview-peopJe'in person, th:e bestrei;.ruitingmethod ·may .be to potential :recruits at "wotk, :!:it schci<Jl;: ·at home;, a pubiic venqe, or at an-

other <appropriate location. interviews coul.d be: Scet up by ·calling :sampled irtd;ividuals: or sen.d,ing or .e-rnaiL· lO£ these methods· of :contac-t require the'indiyidu:<il's contaet information to he known. This would he tht;· patiellts of a. parti'cula,t clink ot of a particufar organization.}

• 1f the plan :i·s. tb 'interview by it nfitY to ipants-with cold caUs. However, the participation ,rate will likely be higher if a let- 'tt::t of invitation first. (Se:odirtg a wilJ also callpvy tqr the atquis#ion nf ·signed inf@rmed eons.ent foFms prior to the interview, if the>,r are 'Feq

FJGURE 19-2 Examples. of M¢thq:qs torCon'tactiri,g Of the Population

i9.2 Recruiting,Metbods- '1-41

• If the plan is to collect data via the Internet, then contacting potential participants via e-mail or a website may be the most effective method.

Participation rates will likely be higher if recruits understand the importance and value of the research project. For example, suppose that the plan is to interview mem- bers of a particular organization by phone. The response rate is likely to be highest if interviewers start each phone conversation with potential participants by explaining why the participants are being contacted, how their contact information was acquired, and how completing an interview will assist the organization. The participation rate may be quite high even for unscheduled telephone calls because the importance and relevance of the study is addressed at the start of the call. Support for the study may be even higher still if the study plans are shared ahead of time in an organizational newsletter or via an e-mail to all members.

In contrast, only a few out of every hundred calls made by a random-digit dialing- calls to a computer-generated list of unscreened telephone numbers-may yield one person willing to complete a survey. Even then, many willing participants may turn out to be ineligible for the study. Additionally, a growing problem with using random- digit dialing is that mobile phone numbers are often unlisted and are not necessarily indicative of the user's geographic location. These issues may further reduce the rep- resentativeness of study populations recruited by random-digit dialing. Nevertheless, using the first minute of a phone call to explain why a particular study will make a dif- ference in the world or to a particular community may: increase the willingness of ran- domly dialed individuals to participate.

Another way to increase the participation rate is to pro-yide multiple invitations and opportunities to participate and to make participation as easy as possible. Mail survey packets should include a concise cover letter that explains the purpose and im- portance of the survey. The cover letter should also disclose any necessary informa- tion such as financial sponsorship and contact information for the research team. The mailed packet should also include the survey instrument and a preaddressed stamped envelope so that the completed survey can be easily returned to the researcher. A few weeks after the initial mailing, a reminder postcard or a second copy of the question- naire should be sent to those who have not yet responded. The follow-up mailing should reaffirm the study's importance and express gratitude to those who have al- ready returned a completed survey as well as to those who intend to do so. Similarly, multiple phone calls on different days of the week and at different times of the day may have to be made to reach potential participants by phone. Multiple e-mail invitations tQ complete a computer survey may be required to get recruits to fill out an online questionnaire. Including a step-by-step guide for using the survey Web site may make participation more accessible to those who are uncomfortable with new technologies.

Finally, incentives-such as small gifts or the opportunity to be in a drawing to win an award-may be an effective means of encouraging participation among those in- vited to be in the study.

142 CHAPTER 19 Primary Studies: Surveys and Interviews

B 19.3 Data Recording Methods A decision must also he made about how; response_s will be recorded and when they will be entered ·into JI computet databc;se·. There ·are twq basic options (FIGURE 19-3). One is to record the responses on paper and to enter them into a computer database later. The other is to have or participants en,tet responses intb a; database.

Paper have several benefits. I_n some; they are quired for the collection of data from a large number of participants at :one time. An example is when alLstt:).dents attenQ.ing a ;;cho91 ·need to compLete-a questionnaire dqt- ing the same 20-min ute period. Paper instruments allow for the -easy .collection of sig.:

on into.tmed consent ;and sQine reseatchets v'fllue ,having paper re,cords as a backup. Hut pa;per-based surveys have a serious disadvantage: unless some- what expetisi ve opticai ?tan forms are used, all te$J}Qnse-s have tO manually ·en tere.cl into a computer at a l'ater time. Data entry is often a very time.:consuming process.

The J:najo.r advantag,e of.conipl;lter-a·s.s.isted su·tveys is that they eliniinate the.rteed for later data entry;. They may also simplify the questionnaire hy automatically remov- ing any questions not relevant to a parhculat Stl!dy participant. For· example, they may skip questions specific to females 'for: participants who identify themselves as be- ing male. The main limitation 0fcomputer-a..ssisted surveys is:that some populations

uhtomfortable with cotnputertechn:ology.The' members of these population gr0ups will thexefore systematically choose, rtot top.att;icipate in 4.11 Internet:-ba:se.d Discomfort with technology may take other forms .. Some interviewees will be dis-

by an who is-entering responses into a computet they give tneir responsest These individuals might not be similarly bothered by an interviewer with

E-mail/

FIGURE 19-3 Methods for Collecting and_ Recording survey Dater

19.3 Data Recordtng MethOds 143

a clipboard. Additionally, for some studies having a limited number of computer ter- minals or portable electronic devices so severely limits the number of participants who can complete a survey a t any one time that computer use becomes a barrier to proj- ect completion.

• 19.4 Training Interviewers The interview process should be the same for all participants in a Stt!dy, whether they are being interviewed in-person or by telephone interview. Uniformity is easiest to ac- complish w hen all interviewers are provided with the tools they need to follow a stan- dardized set of procedures. All interviewers should undergo role-specific training and have an opportunity to practice their interview skills. Each interviewer should be given a interviewe.r handbook that provides information about the purpose of the study, details about interview logistics, an annotated script for the interview, and annotated copies of all study forms . The training and handbook should:

• Explain the interview process step-by-step. • Specify exactly how to ask questions and record responses. • Identify any prompts or follow-up questions that the interviewer needs to use. • Provide preset checklists for managing problems during an interview.

All of this information should be contained in the study protocol. Interviewers usually feel more prepared for their role after attending one or more

training sessioqs. Facilitators often begin training sessions by: ·.

• Explaining the purpose of the research project. • Emphasizing the importance of strictly following the procedures spelled out in the

interviewer' handbook. • Making clear the absolute necessity of maintaining the confidentiality of all infor-

mation that study participants share with them.

Interviewers may also need to complete additional institution-mandated research ethics : training sessions.

The bulk of the training session is then usually dedicated to understanding and practicing the interview process. The questionnaire should be examined in detail so that all interviewers understand what each question is asking, how to pronounce all the words in each question, how to phrase the reading of each question, and how to present the possible answers for quest'i:ons that are not open-ended. All response form s and/or computer-assisted data entry programs should be closely examined so th at every interviewer understands exactly how to record participant responses. Each in- terviewer should have the opportunity to participate in several mock interviews from start to including the. informed consent Clear guidelines ·and lots of practice will help to create skilled- and confident interviewers (FIGURE 19-4).

144 CHAPTER 19 Primary Studies: Surveys and Interviews

:cnaracterfstk

Respecttui

:G:onsiderate

Impartial

Actions-That nernonstraH.rlh:e Characteristic • Communkates'pleasantly and afi study parti;cipants and

mer:nbers ofthe research team • Has pt.adkeefinteiviewJng eno.L!gh to be tomfortablew.itb both -thescrlpt and

th-g>tnterYH:!w. -proce$S,: ,. As!(s ·sl:JpeTV!sorsfgt

•- Begins scheduled lnteNiewse-ssion on t.tme • Has all necessar§ materials-on handpriorto the -startofeach·tnterview·session • Maintains,meticulous records :and eo:0mpletesall fitesand,paperwork promptly • Dresses anct ·grooms-ap!Jropriatel-y-for .interviews ,. Is alert coodltiohsthat may make inteT:l'Jewees oncotn{ortable.;

as loVd b<J:tkground Pim Hgntihg • Allows fldeqyatetirpe fqr to respohd tq • 9\'ltf .! .' cleC!tiY

Uses;;an appropriate tone ofvdice {and, for in-pe-rson intef'Vtews, faCial ;expressions and gestures) · ·

• Jlstof responses·when·a partidpant· ti.Oe$Xiot

., Reacl}}fffe scriht c,tsJt'iS. writt¢ri "' Probes.fw.r;rn;SWer:s only'wnen the ·ss:YiPf in{.lica_tes i.$ appro\!ed •· ones (luestit;m :unless qn e'JP!snatiotriS:

.provided ln'·the:seritJt or approved in theiinterviewer handbocik Av.oids veibal and' nonverba'l :Of disapproval

• Does ncit ·express_, personal qpinions · · •· interviewees toward a,-paftlcutar (for exarnpfe, ;by

c!h patti<;\J:Iarworqs i_n or by pr,opthg until teceivJng:a :response} ·' · ·

··' l:}oe$ ttqt ttlbrtcate orfillsify if' ,R;e-cords responses to:ope:n-e:ncted verbatin); Without-

-'i-nterpreting them

• Compi-etes:ailsteps of the interview process lnthetorred:order1 as prescribed by.·the,fnterviewerhanctb.aak:

• DoeuritentS'ioforrtJe{:( conse.i:lt{Jfior to conducflng an ihtervJew .r any·-:cah1ponent;ot the lntervi,ew ·• tQr,ros

FlGURE 19-4 Characteristics of Well-Trained Interviewers

19.4 Training lntetviewers 1 45

:

Primary Studies: Additional Assessments

Surt;ey$ are th? most Qoinmoti of but•other measurements, a-re-often important supplements. ta: in-formation.

• · _20.1 Supplementing Self-Reported Information Self-reports, .such as those made during interviews and the completion o£ question- Paires, ,are da.ta so1.1rces, but they have significant limitations. Respondents rnay n-otte11 the truth, either because they do not accuratdy·remembet the (;lrtswers pr

they want to provide "'c;orrect_" answers. Also,, they ma-y not know :some of their health measures, such as theircur:tent weight of blood pressure, Laboratory tests : and other objective measures-can be used to su-pplement and validate -self-reported data ahd to explore factors require independent assessment. This i:s usually coJfected in pe;rson -by a member of the research team. This-chapter presents SDme of these ·additional types of data.

• 20.2 Anthropometric Measures- Anthropoinf!try is the_ of the human body, and anthrop,ome.tric mea- surements are often e-specially in studies of nutritionaLstatus,., Some of the; rno'st coinmon body rheaslit¢ments are:

147

·• Height (stature) • Weight • Waist circumference • Hip circumference • Mid-upper arm circumference • Skinfold measurements that estimate the body fat percentage

Standardized methods should be used to these measurement$. Any tools used for the measurements should be ·carefully calibrated to ensure accuracy and reliabil- ity. The individuals taking the measurements should be trained to :use all equipment properly and to record results to the appropriate level of precision. They should also ensure privacy for participants while the measurements are taken.

• 20.3 Vital Signs Basic vital signs are physiological measurements that can be accurately taken after minimal instruction. These include:

• Temperature • Blood pressure • Pulse (heart rate) • Res,pitatory rate (breathing frequency)

A thermometer is used to measure body temperature. A map.ual or electronic sphyg- momanometer (a blood pressure cuff) is used to measure systolic and diastolic blood pressure. Resting pulse and respiratory rate do not require any instruments other than timekeeping devices. All assessors should be trained to use the same techniques, and tests of inter-rater reliability should be used to confirm that all assess_ors get similar or identical. results when they measure the same person.

• 20.4 Clinical Examination A well-trained clinician can m_ake accurate and reliable assessments of many health states that machines are unable to assess well. A clinician can examine:

• Heart s.ounds • Breath sounds and other respiratory functions • Bowel sounds and the condition of the abdomen • The range of motion (ROM)' and the condition of the joints • The condition of the skin, hair, and nails • The health of the eyes, ears, nose, and mouth • Mental status

148 CHAPTER 20 Primary Studies:: Additional Assessments

:

• The ability to conduct activities o"fdaily'living Other signs of health or disease · ·

Sometimes a clinical examination will he part of the data collection process. It so, an a;sses$mep_t form sho1,1ld carefully deS,c:ribe each C:()mportertt of the examination, ing the exact procedures to he used and the specific. diagnostic criteria for each item on the form, as orQer rn which lhese should be examined.

• 20.5 Tests of Physiological Fundi on Tests of physiological function can provide helpful informati0n about health status. For example, spirometry measures lung function, elec::troc;ardibgr<;rphy (ECG) measures heart function, electroencephalography·{EEG) measures audiom- etry measures hearing acuity;

• : 20.6 Laboratory Analysis of Biola-gical Specimens Tests ,o£ serum., urine,, stool, saliva1. and/or other biological :specimens may be helpful fbt iderttify'ing: ·· ·

·• The risk factors for a disease :e, The presence: of a disease or matkers fo·r a disease • The characteristics a:ssocia ted with having a disease.

Sort):e <;;lne.w spe.<;imehs, eithe): as part of routine clin- ical practice or specifically tor the purposes of,the reseanch project. If so, a resea:rch ethies (;ontrtritteernq>?tbe a·s&U:red that the potetiti:aJ p}tyskal Tisk:s. to caqS,ed by the collection of the. sample will be minimized and that.:the·privacy af.partieipants Will be prdte¢tecl. ·

Some, studies. may be able to make use oJ existing -specimen banks) wheJh:er the: samples ·are anonymous, qt linked to other jnforntation .-a,bout the donor. Tile use of :existing samples also .requires ethics comnlittee.review and approvaL Participants. ma¥ · have· a riglit to tlle.tes\llttH:>fthe lahor4tory .te.sts -qn their imens, and the protocol .should.discuss. how n'G:itifkaticm will occur ..

:• 20.7 Medical Imaging MedicaL imaging techniques .are .sometimes used to visualize parts of .the human body . . ate. radiqgraphy (X:-rays), te$onance. (MRI), coniputec,l tomography ( CT) ancl.ultrasoun:d. The resulting images. may be useful to researchers for purpose$; of diagnosis-artdlbr.fort}ie resp.onses. to. therapies.

20.7 Medical Imaging 149

• 20.8 Tests of Physical Fitness A number of different tests can be used to measure physical fitness levels.

• Cardiorespiratory fitness can be assessed using a 1-mile walking test, ·a 1.5-mile run test, or some other test of aerobic fitness.

• Measures of muscle strength and endurance include timed curl-ups, push-ups, pull-ups, flexed arm hangs, bench presses, leg presses, and grip tests (using a hand- grip dynamometer) .

• Flexibility can be measured using a sit-and-reach test (often measured with a flex- ometer) and other activities that stretch the lower back, hamstrings, or other muscle groups.

• Additional tests of fitness may assess agility, balance, coordination, speed, power, and reaction time.

• 20.9 Environmental Assessment Both the natural and built environment can have an impact on human health. Exposure to pollute.d air or water or to toxic substances (such as asbestos, lead, mercury, radon, or pesticides) can compromise health. Access to outdoor recreational areas and t he installation of safety equipment, like grab bars in the bathrooms of older adults, can contribute to increased health. Some research projects may benefit, from measuring the levels of environmental contaminants or the presence of hazards in the home and/or work environments of research participants. For example, the ·arriount of radon in the lowest floor of a participant's home, the presence of lead in paint chips from the home, and/or the amount of daily precipitation falling at a participant's home location could be linked to a health questionnaire.

• 20.10 GIS (Geographic Information Systems) Sometimes a map and/or spatial analysis of important features in the study area would help answer the study question. If so, a GPS (global positioning system) receiver can be used to acquire the geographic coordinates (in latitude, longitude, and altitude) for relevant locations, such as the homes of participants, hospitals and other healthcare facilities, roads, schools, houses of worship, grocery stores, recreation facilities, wa- ter sources, and industrial sites. The coordinates for public locations ¢an be collected by anyone, but permission from the owners or residents. of private land may be needed before entering their property to take a GPS reading. The GPS coordinates for the homes of participants is identifying information; extra care must be taken to protect this data (see Chapter 2 1 ).

150 CHAPTER 20 Primary Studies: Additional Assessments

Pr-imary .Studies: Ethical Considerations

hdve_an. ethical obligf!lt.iqtL til irzirr;imite .the ris_ks may pose ,to _participa-nts.

• 21.1 Beneficenc·e, Autonomy, and Justice The thr-ee main prineiples in biomedical research ethics are beneficence, autonomy (.sometimes called "respect for persojls" ), and distribptive justice. Each ptQ:tb.toJ shopld be carefully inspected to ensure its compliance with these principles .. This ·SOft of scrutiny is··required by research ethics cotriinittees, and,. mo.reimpo:rtantly, need to adhere to the. highest stanaatds.of profe·ssiona1ism. ·

Benefic;enc.e means that the study should -c'do good" and is 0ften paired with non- which mea·ns.th?t the $tildy should ('do no harm/' To meet the. require-

ment of beneficence, a research proposal must have a highJikeiihood benefitting. individual participants a,ng/orthe communities from they atedrayvn. For most studies, the opportunity to contribute to scientific knowledge· is .considered an ade- q'Q,ate benefit to patticipants, "although in $9me" mo.['e specific individu4l artg ·community benefits :can be offered. _

NQ:ntnaleficence requires Jhat steps taken to "minimize potential physical, psy., chologieal, financial, social, or ether harms to _participants, as well as to ensure an :,tcceptable·balance hetween risks. arid benefits. For example,·theptinciple or doing no harm means that experimental stadies must identify ahead of time what e;vents would

lS1

category Examples· o.fQuestions to Ask

Contribution • Why is the proposed project important? • How will individuals and/or communiUes benefit from this study?

• Will individuals or communitie.s that participate in the study be offered any form of inducement, reimbursement, or compensation? If so, what will be offered, and is it appropriate? Is the offer so high that it could be seen as

Compensation coercive or so low that the study could be seen as exploitative? • Are the risks ofparticipation mihima.l? • How will st.Ody-related injuries be hgndled? • Are the risks and benefits balanced?

• How will potential participants be informed about the study? • How will consent to participate be documented?

Consent • Will a test of comprehension be required? • If applicabJe, how will consent (and possibly assent) be acquired for children

and other members of potentially vulnerable populations? • If applicable) win community meetiFigs be held prior to beginning the study?

Confidentiality •· How will the privacy and confidentiality of participants and their personal

information be maintained?

• Why is research in the selected population important? • Is the source population appropriate for the goals of the research study? • Will the selection process be fair?

Community • Will the saJ11ple size be adequate? • Are potentjally vulnerable particip.9tfts adequately protected? • Has th:e protocol been adapted ·to -aadress the cultural, expettatibns of the

source population? .. • If applicable, has the community agreed to participate in this project?

Conflicts of • Who is contributing to the project's finances and/or logistics? • Might potential conflicts of interest inhibit the ability of a researcher to

interest conductethical and unbiased research? -

• Are afl members of the research team adequqtely trainedto conduct ethicat research?

Collaborators • What be taken· during -data collec:tion and anafysis to ensure that the protocol and all ethical standards are adhered to by all members of the researth team?

• Which research ethics committee(s} needs to review the project? Committees • If applicable, what community organizations have been consulted about tf:le

propos€to project?

FIG URE 21-1 Eight Central Considerations (8 Cs) in Research Ethics

1 52 CHAPTER 21 Primary Studies: Ethical Considerations

lead to.early termination of the study. Discontinuation.mightbe appropriate when the intervention appears to be dangerous or when it appea,rs to be so bene:ficial that wou1d be unethical not to imrnedi'ately offer the, intervention to those in the placebo group. Another way to minimize harm is eo provide participants in studies that might c.ause emotiDnal distress with information abo_ut local counseling services.

Autonomy means that participation in research should be tompletely voluhtary. For almost all research projects that involve interaction with individual participants and/or their personally identifiable records, each potential participant must he fully in- formed about the benefits and burdens of the study, the procedures involved, and the plans for use ofthe data collect_ed. They rnqst also be given a free choice to participate or not.

Respect for persons is: a broader concept that includes, for example;

• Justifying· the necessity for :and the, importance of the research project -•- C hoosing an appropriate· sautee: popqla'tion • Using a fair process to sample and recruit participants • Ensuring an adequate sample size so that the _study a,dequate sta,tistical p_awer

to assess the study objectives - - • Making research procedures as ·minitnalfyinvasive as • Maintaining. the ·confidentiality of all info'tmation

Distributive fus,tice seeks to ensure, that the_ benefits and burdens of research are equitable. For example, if an expetimental pharmaceutical therapy proves to be eJ- fective and safe, participants in the dinicaltrial should be guaranteed ac- cess to the drug_ after the tri-al is _over. ·

'FIGURE 21 -1 highlights some of the many questions that researchers should ask about their own prior to a farinal ethics committee review. International health re- search guidelines, such as those developed by the Council for International Organizations of Medical (CIOMS), and national research guidelines may identify.-(ldditional areas o'f concern that need.to be considered as protocols are .. develope.d.

• -21.2 Incentives Incentives are sometimes offered to research recruits and participants. Researchers nted to consider the t!thical implications of oHering an indU;cernertt to poten,tia1 ticipants to encourage them to enroll in ·a study, aJ offering reimbursement for the co.sts of participa.t._ion, or of cnmpertsa,ting participants for their efforts.

Research intentives do not have -to be monetary; For ne·ariy every study; a benefit to pardclpants Is contribution to sc:ienJific;:,knowledge and to the increa,sed under- standing of their own health risks. and disease conditions-as well as their communi- ties'-that ca.h result from s.ciehtific J:i!search. For many res.earch in the health sciences, this is su-fficient :reward for participation. At; the same time, the principle of

21.2 Incentives 1 53

:

nonmaleficence requires that participation in a research project should not be an un- due burden to participants. So in some situations Teimbursing participants for their travel expenses or compensating them for their time may be appropriate.

To increase the participation rate, researchers may reasonably offer a small gift to all participants or enter all respondents to a questionnaire into a drawing for a more substantial gift that one randomly selected participant will receive. It may also be ap- propriate to provide free treatment for certain conditions examined by the study, such as iron pills for participants fourid to have anemia or de-worming medication for par- ticipants with intestinal parasites (so long as the relevant health education materials are also provided to participants receiving these treatments). In the case of clinical tri- als, covering all medical expenses directly related to participation in the study is some- times appropriate.

However, the desire to reward participants must be balanced with the need for participation in any research project to be voluntary. When an individual feels coerced into participation, the principle of voluntariness is violated. Coercion could include social pressure or requests from authority figures that make it difficult for an individ- ual not to agree to enroll in a study. For example:

• Employees asked by their supervisors to entoll in an occupational health study may fear losing their jobs if they do not agree to participate.

• Patients asked by their own physicians to register for a research study may fear that their medical care will suffer if they do not comply with the request.

• Prisoners may believe that participation in a research study is mandated and/or will yield unspecified rewards.

Coercion can also include generous incentives, such as free medical care and monetary compensation, that could significantly impair the ability of an individual to make an informed decision about the risks as well as the benefits of participation. To minimize the risk of coercion, researchers have to be very transparent about what participants will gain from participation in a research study and what they will not gain.

• 21.3 Informed Consent Statements Informed consent statements provide essential information about research projects to potential research participants so that they can make a reasoned decision about whether to enroll ·in a study. The key components of an informed consent statement are sum- marized in FIGURE 21-2. The statement must use clear, simple language that the reader is able to understand to describe the study aims, the procedures and expectations of participants, and the benefits and the possible risks of participation. It should empha- size that participation is voluntary and that any participant can withdraw from the study at any time. Informed consent statements should be tailored to the source population, and they may need to address cultural attitudes, beliefs, and traditions.

154 CHAPTER 21 Primary Studies: Ethical Considerations

:

conte_nt Area

Pufpose.

Procectures

Confitlenttqlity

Voluntariness

Contact infmmation

Signature·

Descrtpfi¢r1 A aefinrtibn of ''reseqrc;.n· and a that the stqoy An explamitior'l '·of the p'urpbse a no aims pf the res.earch . prot.ess in

the.f are .in which ,that With the tesearch"gpals) A description,of hovv anq why certain

invited to pa·rticipate in the fese9fth prqje.ctand an .estimate of th'e o( inqi·vtguals'Who will J:tE:

A d.esctiption oHhe study procedures (Including any physic: a I exanis, cone,Jlpn samples" r.anaGrnizat1dn ·orqlindin.g p,ro.cesses, interventions; ot othe( ptcltedures t.hat:are part of the stl,IQY proto con and, tlie .expe(ied duration·of the· ·the5t\JdY

A descrtpfion dfuenefits· to paitt¢ipants,ahd/or to rnduoing a. clear bf neJ;lffe_req or a. clear ·statement that

the. participant will direct benefits

A descf-iptipn df the possible risks,. discomforts( and. costs associateq with particJpattoh, a sta'ternMtthafinvolvement in the·projeq involve

· unforeseeable a description of h·o_w study-rel<.n:ed injurtes .WiH behancfled

A descripfid'n Qf tlie tHat will oe. tQkeJi to maintain confiqentlality A statement i's voluntary and thaHhe participant may,

:Withdraw from th_e study at ·any time with no penalty., along with ,an · explana.tion for the- prqcess: of trorh

Contact.information forthe researchers

Space for the participant's signature

Fl G u RE 21 "2 Content for the. hi formed Consent Statetnerit

• 21.4 Informed Consent Process Altho. ugh researchers often more concerned about·acqulri.ng a signat"qre from pa r-

than explaining the research process to them, informed consent is intende"d to be a process, not merely a. piece of The principle. of autonomy dictates that potential participants in a research study have the right to make their own decisions ahout whether to participate and thad::hey must be provided with:information that will allqw them to make infqrmed cl}.oices. ·

Thejnformed consent process consists of the following steps:

' " Reading the infotmec;l consent statement aloqd to a potential participant ·and/or allowing the individual to rea:.d a -copy of the· statement

21.4 Informed Consent Process- 155

• Allowing adequate time for the potential participant to con$ider whether he or she wants to participate

• Answering any questions • Only then asking whether the individual wants to participate in the study and is

willing to sign an informed consent form

Acquiring a signature is .not the end of the process. The lines of communication between r:esearchers and participants must remain open during and after the data col- lection process. All participants must be given a copy of the informed consent state- ment that includes contact information so that they can contact the researchers if they have concerns about the study or desire to withdraw.

The researcher should ensure that participants understand the research process and the consent document. A brief test of comprehension may be _helpful. For ex- ample, recruits for an intervention study may be asked to say in their own words w hat "randomization" means. A correct answer will demonstrate an understanding that each participant may be assigned to a control group rather than to the active interven- tion group and that participants do not have a choice in the matter, An incorrect or incomplete answer may require additional explanation of the research process prior to acquisition of a signa ture on a consent document.

• 21.5 Informed Consent Documentation For most studies, the expectation is that each study participant will sign a printed copy of the informed consent statement. This written record provides ·legal protection for the institution sponsoring the research project because it shows that participants agreed to the terms of the study. For telephone interviews, informed consent documents may be mailed to potential participants, signed, and mailed back to researchers prior to the interview. For computer-based surveys, an electronic signature can be provided.

In some situations, written consent is not appropriate, such as when participants could be harmed by being linked to the study or when the source population has a low literacy :rate. When few potential participants are able to read and write, partici- pants might provide a thumbprint or some other mark to indicate consent. Alternatively, when it is inappropriate to ask people who cannot read a document to sign it, oral consent may be preferable. Oral (or verbal) consent must usually be witnessed by an independent third person (someone other than the researcher or the participant), and in some cases a statement of consent is also audio-recorded.

In a limited number of studies, the individual informed consent process may not be required. Some anonymous questionnaires do not require an intense informed con- sent process if they:

• Cannot be linked to individuals • Do not ask sensitive

156 CHAPTER 21 Primary Studies: Ethical Considerations

• Do not physically examine indivi'duals or collect biological specimens • Are SO' short that describing the study would take longer than .completing the:

questionnaire

In these situations, when there ·are no fores.eeable. risks to participantS, the ·completion of the survey can sometimes be considered adequate proof of willingness to participate.

bthet stqdres might alsq not require consent. For _if researchers will ·s.erve groups of individuals in public places,· where participants have no reasonable:ex- pectadonof will not interact-with the researchers, consent .-is not require_d.

Different research ethics committees will have -different levels of comfort with waiving consent or :allowing alternativ-e methods for documenting_c;ortsent._ The rele- vant committees should be consulted about what they will consider

• 21.6 Confidentiality and Privacy Privacy is-the individuals have the right to choosewhat'information they reveal about themselves. The right to privacy means that:

• Individuals have the _right to refuse tg allow their personal in(ormation to be shared. with researchers .

. -Inqividuals who agree to. participC;Jte in a study inv'olving face-to-fa<;;e, interviews. should. have the option of meeting with researchers. in a place where no one out:.. sl.de the res.e.arcb te-a.m wilt be: able to obsexve .. or gverhea,_r the interview.

Confidentiality is· the. protection of per.sona:l i-nformation provided to researchers , One way to guarantee confidentiality is.not to colleet any personally iclentifiable in- fermation? Such as names, addTes·ses·,; identification numbers, Or. other data that can be linked to an individual. This is often .an option for cro-ss-sectional surveys, but it is not possible for prospective or longitudinal studies in which baseline data a'bovt irt- .dividuals mus.t be linked to their own tollow-up data. When individually identifying. 'information must be corlected;,, many thtoughout the research proces's can be taken to protect the information.

• All paper records should be :stqred in a locked tile box in a locked :roa.tn; and all ·computerized data files should be .

• Names a.rtd otheJ; personal identifiers ·should nolhe in B.le.s that ,con- tain sensitive personal information. Instead_, two separate files should be created,, one for info.tina,tion and ope f:or 'lll other data .. should be linked only by a unique study identificatinn -number.

• Onfy essential research shoplp have: the file containing persop- ally-.identifying information.

21.() Confidentiality arid Privacy 157

• At some point after the end of the study (and in compliance with the rules of the relevant research ethics coinmittees about how long documentation of informed consent must be stored), individually identifying records should be destroyed.

Researchers asking questions about sensitive issues must decide ahead of time how to handle disclosures. In some situations, guaranteeing confidentiality m:ay not be pos- sible if doing so would violate the law. For example, legal mandates may oblige re- searchers to alert the police aboat child abuse, intimate partner violence:;, or suicidal ideat.ion, or to inform public health authorities about diagnoses ofsome infections. The decision about how to collect data related to these issues may requite consultation with a legal expert and local authorities. Participants in studies of serious genetic dis- eases should be offered genetic counseling and given the opportunity to decide whether they want to know the results of tests.

A research ethics committee may waive the usual requirement that consent forms include the names of participants if a study will collect information that could cause harm to the participants if their names could be linked to the study. This may include studies of: ·

• Drug or alcohol abuse • Sexual practices and preferences • Psychiatric illnesses • Immigration status • Participation in illegal • Genetic disorders

If the study will examine any of these potentially sensitive areas, the ethics com- mittee should be consulted early on in the planning process so that appropriate pro- tocols for handling the inform:;ttion can be developed.

• 21.7 Cultural Considerations A protocol must be appropriate to the culture or cultures of the expected study par- ticipants. Culturally appropriate recruiting may take different forms. For example, in some cultures, a small gift may be expected as a token of goodwill before an individ- ual is asked to participate ·in a .study. In othe,r parts of the world, this would be. con- sidered coercive because it, would create a "debt'' owed to the tesearcher .. In some cultures, participants may a small gift upon completion of the study. In other cultures, such a gesture of appreciation might make volunteers feel that the gift some- how devalues their donation 'to science. Participants from some culques expect to share tea or a light meal with researchers before any questions are asked. People from other cultures may expect that all health research will be conducted in a clinical

158 CHAPTER 21 Primary Studies: Ethical Considerations

setting. In .some parts of the world, prospective participants might need to know that community sqch as goverpment otficial$, oreligiom; leaders, or tribal leaders, have approved of the project and are monitoring it. In other cultures, the assuciation of with a research project ,may ra.ise concerns about volunt4riness, confi- dentiality, and/or the potential misuse of data.

The informed consen.t process may also to be ada·pted to local custom .. Although individual participants are always required to provide consent for their own participation,. potential participants may ,need time to consult with their spouses', ents, or other family ·members: prior to giving conse:nt. For some community-based Stvdies, a rneet.Lng ofthe community should be held so that e\reryone is .confi- dent that they all hearing the same story from the researchers. It may be helpful to have a local advisory bo'ard se.rve as intermedia;ries between the commurtityand the research team. The informed consent statement and study materials may need to be available in multiple languages.

The survey instruments· and data collection process must also be culturally appro- priate, and researchers must be trained in ¢ulturally respectful interv.iew.technique_s. Topics that are openly discussed in one culture may be sensitive ."in another. Tests that are o'.rJJy mildly uncomfortable in one,Gulture,may be extremely distressing in another. For although peopfe in some are sensitive about the measurement "of bqdy other cultures may not c:are about wei'ght but may be uncomfortable, with the.measurement of height. There may be formal or informal restrictions on who tan conduct ah interview or a physical Female participants may be un- willing to be' examined by a male, or older participants tnay be uncomfortable beipg interviewed by a much young.er person. Participants may expect to be. alone with,just a researcher, but in other cases participants will expect to have a family· rhember'pres- ent fqr ,the entire proce,ss.

H the research team does not include members of the target populption, it is iih- po.rtaht to work with of the source community when developing, and revising the protocol. Additionally\ some research ethics conl.thittees req11ire a cultutal expert to examine the protocol as part of the review process.

• 21.8 Vulnerable Populations Vulnerable populations are :discussed in 16. In addition. to defending why a particular research project must fo.cus· on a potentially vulnerable extra care :must taken t.o ensure that:

• The selection process is fair. • Pot¢ntial participants ar.e assured that participation is voluntary .. • Participants (and/or their legal representatives) are fully ihformed about the pos-

sible benefits and risks of the study as wdl as about the expectations of participants·.

21.8 Vulnerable Populations 1 5,9

Although most members of vulnerable populations can make their OWn choices about whether to participate in a res.earch project, children and some adults with cognitive impairments may not be <::onsidered competent to make an informed decision. In this situation, a legally approved guardian is allowed to grant consent on behalf of the study participant. Whenever possible, in addition to having the legal representative's consent, potential participants should assent to their own participation.

• 21.9 Ethics Training and Certification Research ethics committees usually require everyone who will be 'in direct contact with research participants and/or their personal data to complete formal research ethics training. Many institutions offer their own courses, either in...:petson or online, and several funding agencies and nonprofit organizations also offer training that is avail- able to anyone. After completing modules on various aspects of research ethics and pass- ing an exam, a certificate of completion (usually valid for one to three-years) is issued, affirming that the investigator has been trained in ethics. Copies. of these certificates should be saved because research ethics committees often require proof of ethics train- ing for all members of the team.

160 CHAPTER 21 Primary Studies: Ethical Considerations

Ethical Review and Approval

Research ethics committees protect study participants, researchers, and host in- stitutions by carefully reviewing research protocols prior to their implementation.

• 22.1 Ethics Committee Responsibilities The three primary goals of research ethics committees (RECs), often called Institutional Review Boards (IRBs ) in the United States, are to:

• Protect the " human subjects" who will participate in o bservational or experimen- tal studies or whose personal information will be examined by researchers. (Separate animal care and use committees oversee research with animals.)

• Legally protect the researcher's institution from t he liability that could occur as a result of research activities.

• Protect researchers by preventing them from engaging in activities that could cause harm.

The major functions of ethics review boards are to:

• Review new and revised research protocols • Approve or disapprove of those protocols • Ensure that informed consent is documented (if required) • Conduct continuing review of long-term research

161

• 22.2 Warning: Ethics Review Takes Time Research ethics committees are usually composed of at least five members, prefera bly from diverse backgrounds, including scientists and nonscientists. Each member re- views the proposal and then meets with the others to discuss it and to decide about whether it meets the ethical requirements of the institution. An outside scientific ex- pert and/or community representative may also be consulted about the research plan. As a result, even the most efficient ethics review committees, when reviewing even the simplest proposals, may need a month or longer to issue an exemption or an approval or to make a request for a revision to be made to the protocol, which must then be re- considered by the committee before final approval. For complicated proposals, the re- view may take several months. Examples are studies involving:

• Invasive procedures • Sensitive questions • Potenti·ally harmful interventions • Deception about the study aims • Waiver of written informed consent • Multiple sites • International research teams

A research time line should assume a lengthy review period. The application should therefore he submitted to the ethics committee as soon as possible in the planning process.

• 22.3 Application Materials Some research ethics committees ask applicants to provide a narrative research state- ment that addresses a list of possible ethical concerns. Others require the completion of dozens, of pages of forms that require answers to a long series of questions about the project (even though most questions require an answer of only ''no'' or "not ap- plicable"). FIGURE 22-1 summarizes the questions that research ethics committees com- monly examine during the review process. A research protocol or narrative statement about a planned project should address each of these points and any others required by the committees evaluating the proposal.

Proposals for the analysis of existing data may be significantly shorter than pro- posals for new data collec:tioh, but both primary and. secondary analysis propo$als need to: ·

• Describe the expected study participants • Discuss the risks and benefits of the study • Explain how confidentiality wiJl be ensured

1 62 CHAPTER 22 Ethical Review and Approval

Category considerations • What is the anticipated composition and size of the study population? • How will participants be recruited? Does the recruitment method

Participants r-aise any concerns about-coercion?

• What are the inclusion and exclusion criteria? Are they reasonable? • Is the source population appropriate for the study question? • Are potentially vulnerable subjects protected, if applicable?

• Why' is the study important and necessary? How will the proposed study benefit partici'pants .and/or their communfties;t

• How will data be collected? wm existing data, documents, records, or specimens be used? Wiil -individuals or groups be examined using surveys, interviews, focus groups, oral histories, program evaluations, or other methods? Will data be audio or video recorded? Will noninvasive clinical measures be taken? Will participants be asked to

Risks and benefits engage in exercise or tests of endurance, strength, or flexibility? What machines will be used to coJiect data, and will collection jnvolve Will blood, hair, rlail clippings, sweat saliva, sputum, skin cefls, or other biological sped mens be collected noninvasively? Will drugs_ or devices be tested?

• What are the potential physical, psychological, financial, or other risks to participants?

• Are the risks minimal (or at least minimized)? • Are the risks reasonable compared to the anticipated benefits?

• Does the informed consent st,atement aqhere to institutional guidelines?

• Howwill informed consent be sought? • How will informed consent be documented? • Is any modification to the usual methods of documenting informed

Informed consent consent being requested? Is the request reasonable? (For example, are parents being asked to provide consent for their children, and are the children being asked t() assent to participation? Or is a waiver of a signed consent form being requested because the source population has a low literacy rate? Or is a request being made t<;> have no documentation of consent because the .existence of a form linking an individual to the study could harm the-participant?)

• How will privacy and confideniiality be maintained? Privacy and confidentiality • What are the plans for the protection of computerized and

noncomputerized data?

• 00es the informed consent statement clearly state how research

Safety monitoring participants can cont-acfthe research team and/or the ethics review board if they have concerns?

• What constitutes an adverse eveAt? HoW will such events be handled?

FIGURE 22-1 Examples of Information Requested and Examined by Ethics Committees

22.3 Application Materials 1 63

Conffltts .Ofihtere.st

Resear¢her tNin:tng ,.

Dncumentation

FIGURE 22-1 (continued)

-"·

• Hqw i_S"tbe · • nEled be\

afsclosed and/or Bddress.e:d? ·-

-. tO: cohtll.,ltitelhical res.earcft? • .. (ifaf!¥l,attached?-

- • .Are . .otherassessment t1;1.ols .. attache'ct·? ·

• -,s a ((my. qt tfom··studysites-and!o_r·utJ1er

efh ks-revjew commlft.ees attacheq1 ilapp:fica'bfe?' • ls. a..j:Qpy_.Gfthe ·§tant pr.Qposal Gltta€hed; if:applieabfe? • oJ tre\iriirtg for a'! I memb,erS,, of

te-q):h ·

• Disclose_ potential conflict.s of interest •, Provide, proof of training.

Supply all relevant documentation

For ptirn}lfy st1,1dies, the Q.ocumerttaticm may include :a copy of t he informed consent statementt the. questionna'ire, :and recruiting materials. For secondaryanalyses, the ap:- phca_tiq_n mpst inchtde that the. data afe in the p.uhlic ciqmairi or that appro- priate individuals or organizations havegranted the researcher permissron to analyze -the' d'aq .. ,

•22.4 Review Process Once, all applic,ation materials have heen submitted to a research ethics committee, there are three possible. next Steps: (1) exemption., (2.) expedited review, and (3) view. The"·ethics' review boar.d decides whlch acfion -is appropriate.

Exemption fto'm review rna y'lie granted-but not have to gtant¢ti-whert the research involves the analysis, of .existing r,eco.rds or biological specimens thatt:anpot beJi:rtk:edto individua,ls. - · ·

An exemption can also he granted for data colle(;tedas part of normal practice tha,t isrto·( Intended tQ contribute tQ ,gerreralizable knowledge'. It to make a distinction between routine practice activities .and .i!-1-tentioaal research activities ..

include as:sessing their students' knowledge o£ J)la:te- rial, clinicians examining their patients,, community 'health organizations initiating monitotin,g and .evaiu·,atj'Qn projects, and colJecJing, sprvelllaJJCe data and conducting outbreak investigations. None:oftllese activities requires review

164 CHAPTER 22 Ethical Review arid Approval

by a research ethics committee; all are considered to be within the accepted scope of practice. However, ethics review is required if these practitioners or organizations choose to engage in research activities.

• An educator plans to have students take special pre- and post-tests to assess a new pedagogical approach and hopes to publish the results in a teaching journal.

• A clinician reviews patient records so that they can be presented as a case series at a professional conference. -

• The results of a survey of clients of a community organization might later be pub- lished in a professional journal.

In such situations, an exemption might be appropriate, but the decision is up to the committee, not the researcher.

An expedited review may be possible when a minor change to a previously ap- proved protocol is requested. Sometime expedited review is also possible for new stud- ies in which the risk to participants is. no greater than what is encountered in ordinary daily life (or, in the case of clinical work, during routine examinations). Expedited re- view may allow the chair of the ethics committee to approve the proposal without a full meeting of the committee. However, all members must be notified of the decision and given an opportunity to express concerns.

Full review of the research proposal is usually required when data will be collected through interaction with individuals, an intervention will be tested in individuals or a community, or identifiable private information will be collected. The ethics review committee has the right to approve the proposal or to deny approval. If a protocol is not satisfactory at initial review, the committee usually inform·s the investigators of changes to the protocol that the committee feels will make the proposal acceptable. Some requests may be easy to accommodate, and researchers should simply comply with them. At other times, the requested changes would significantly alter the nature of the project or would be unfeasible given the intended study population. In this sit- uation, the researchers need to present their concerns to the ethics review committee and to try to work with them to find an acceptable resolution. However, the commit- tee does not have to acquiesce to the desires of the researchers. The ethics review board has the right (and sometimes the duty) to deny approval to any protocol that does not meet its standards. Furthermore, the board can demand proof that certain standards (for example, standards for data storage or investigator training) are met prior to ap- proving the protocol.

• 22.5 Review by Multiple Committees Multiple research ethics committees may be required to review studies that involve researchers from multiple institutions and/or participants from multiple countries or multiple study sites. Additionally; funding agencies may require review by their own

22.5 Review by Multiple Committees 165

ethics boards. For .example, a stu"dent planning on conducting thesis research in an- other country mt1st have the rest!arch protocol reviewed by, at minimum, two ethicS boards: one from his or her own university and one from an ethlcs committee in the study country (often a local1Jniversity or a teaching hospital).

At least three issues must be resolved prior to submission of a research proposal to multiple committe'es: the applictltioli documents that ·will be required, the wording of the informed consent and the order of review.

First, ea.ch review board mqsfbe £:Onsulted 4bout the app.lic:ation materials it wants to receive. Sometimes submitting the same paperwork to all committees is possible. More likely, .each board will require its own ilhique application materials, perhaps in addi- ti'on to copies of all documents submitted to other ethics committees. The. researchers have the to ensure thaf each application pa<;ket clescribes the study ob- jectives.and protocol in the same way. · ·

S.ecortd" m.any insiitt+tions have their own.prefen'ed ·wording for in·£ormed con- sent statements. The informed consent statement is seen as a legal document, and in- stitutions want to be sure that the wording protects them. Howeve.r, the preferred wording may differ for each participating institution. A resolution about how to merge!. cqnsent state.inenJ requirements-while making sure that the sxudy participants will understand the language-. must be reached.

Third, the order of revie:w must be established .. SQmetim.es all the committees in- dependently review the proposal at the same time'. Other times the reviews con- ducted "domino" style, with the proposal being independently reviewerl and approved by one committee; then passed to the next committee, and so on. Committees com- .monly stipulate that approval by their institution will be contingent on approval from aU -other participating institutions, even when they r·eview concurrently. If a modifica- tion of the protocol. or informed consent document is mandated by one committee, then all other committees must re-review the proposal. A t;ignificant amount of exrta time for ethics revie:w should be built into the project time line when multiple r-esearch ethics committees will be involved.

• 22.6 Ongoing Review Studies that can be completed within one. year may not require further contact with the res·earch ethics C011J.mittee initial ·approvaL however, the res.earcher is required to provide updates to the committee for the du:ration of the project. Routine reports ab<:>.utthe numbef' of particip;ints recruited may be ·requited. All adverse events. must be immediately reported to the ethics review boar4. Any changes to the informed tc>nseht statement, the questionnaire, recruiting materials, or other StQdy do<;tJ.ments' must receive prior approvaL In addition, all ongoing research :protocols must be re'-

(or more at the discretion. of ethics committee) until the completion of data collection or; in ·some cases, until the completion of data

1 66 CHAPTER 22 Ethical Review Approval

analysis. The progress report for re-review may need to include (depending on insti- tutional requirements):

• Current versions of the protocol, informed consent statement, questionnaire, and other study documents

• A report on the number of participants enrolled in t he study, including basic de- mographic information about them

• A report of any adverse events or complaints • A list of any amendments to the protocol or study materials that a re being

requested • A summary of findings (which are especially important for experimental studies

that might need to be stopped early if the intervention appears to be harmful or very beneficial)

• 22.7 Conflicts of Interest Most ethics review committees and an increasing number of journals require researchers to disclose potential conflicts of interest related to the study. A potential conflict of in- terest is mostly likely to occur when:

• A new product is being tested, such as a new medication or medical device, and one or more members of the research team earns a salary (or a c·onsulting fee or an honorarium) from or holds equity interests (like stocks. or ownership ) in the company that produced, developed, or will market the product.

• Intellectual property rights (such as the ownership of a patent or copyright) may result in earnings for one or more researchers.

When a financial or other interest could bias the design, conduct, or reporting of the study-or could merely appear to have the possibility of biasing"the study-the po- tential conflict of interest must be disclosed. The disclosure of a potential conflict of interest is not a confession that bias has occurred or will occur. It is, however, an im- portant assurance of transparency. Most universities and other institutions involved in scientific research have policies about what constitutes a conflict of interest and about when and how potential conflicts need to be disclosed.

• 22.8 Is Ethics Review Required? Ethics review is required for almost every proposal that will involve human subjects, whether those people will be directly contacted by the research team (in person, by tele- phone, by mail, by Internet, or via any other method) or their existing personal infor- mation will be analyzed. A small su bset of projects might be exempted from review,

22.8 Is Ethics Review Required? 1 67

but the :decision to exempt. ct ptoj,ect from tl:tview G3:rt he ma:rle ohly by the relevant tohimittees:._ Researchers .ilr.e notallowed simply to ·de<:date that their projects :do. not need ro he reviewed. · -.

Many int:entivts. are in to J:ncoura;ge f)artic'ipati'on in the farmai review process: institutional approval pr.ovides-a clt1gree o£ legal protection to" the reseq'tthet An approval letter 1s· eviden<:e that the r¢s.earoh ph1,n was ear.efully considered. and deemed reascmahly- s·afe by a of prior to i{le initiation of d:;rs, col- lection and:analysis. Another-incentive is: that some; granting agencies will notrelease funds urttil a research plan has approved by a re$earch -ethics finally, an incre_asing nt+mber oJ jo-urna1s ·arEt requiring that ·authors provide details about w:hic:h re·search ethics .the p_r_dj:ect (,eve11 if it subseqqently

from review). Some are even re_quiring. ,cotpies. of '.the, official apprGvallet- ter$. t.eseat:cbers rnust: tak;e time to un9ergp Jorm:a1 review priQt tQ c-o.llecting any data, or analyzing any data files. Research pr0tocols 'Cannot he,retroae- t1vely ffpprqv;ed. · ·

lf:i8, CHAPTER 22 Elhical ApprovaJ

Secondary Studies: Existing Data Sets

Some health research studies analyze existing clinical records> survey data> or population data rather than collecting new information from study participants.

• 23.1 Overview of Secondary Analysis In some situations, data collection means acquiring existing data sets for secondary analysis. These data files may be publicly available individual-level or population-level data, privately held survey data, or filed clinical records. Whatever the data source, what makes a project a secondary analysis is that the researcher conducting the statistical analysis has not had (and does not have) any contact with the individuals whose data are being examined.

A researcher conducting a secondary analysis contributes to scientific knowledge by analyzing and interpreting accumulated data that might otherwise remain untapped. Sometimes a researcher can download an entire data set from an Internet website or have it sent by e-mail. Such files often contain already cleaned-data ·that are ready to analyze within minutes of receipt. At other times, the data are availq. ble only as paper or electronic records from which the relevant information must be extracted and en- tered into a computer database ptior to analysis.

169

• 23.2 Publicly Available Data Sets A growing number ofgov·e.rnmental agencies (and sometimes research teams sup- potted federal funds and private organizations) make their data sets available to the public or routinely m(;lke data available to re:searchers upon Th,ese organiza:- tionso are experts at collecting data but often do not have the resources to conduct a thorough statistical analysis of an entire clata set before it betomes relatively obso- lete. Sharing data is therefore a cost-efficient way to extract as much information as possible o4t of qata sets, when the data we.r;e expensive to collect. For ex- ample, the U.S. Centers for Disease Control and Prevention: (CDC) provides, on its

data from several nationwide. tross-s.ectional studies, including the National Health and Nutrition Examination Survey (NHANES), the NationalHealth Interview Survey (NHIS), and the Behaviotal Risk Factor Surveillance System (BRFS$), Statistics Canada provides access to data. sets such as the Canadian Community Health Survey (CCHS} via the Research Data Program. The United Medical Research Council has a Data Support Service to link researchers to available popula- tion: data. Additional datq. sets are available ftom United Nations agencies like the World Health Organization and from various national governments.

Researchers may be able to download an entire data set immediately and at no cost directly frotn the website of the sponsoring orgariization. Sometimes there is a

proc.ess. The researcher is required to submit a req1,1est form, which must be approved by-an oversight body before the requestor can be provided with a copy of the <:lata hy e-niail or via a link to a download site. Although these data files-are ofte·n provided at no cost to the sometimes they tnust be purchased, and access to some data files. is limited to citizens or residents of the country in which the data were collected.

A conducting secondary analysis needs to understand all the methods ·that were used for data collection artd the contents of the data file. So, in, addition to downloading the data files, the researcher should download ·and read all supporting documents, such as the project overview,. protocol, or handbook; the questionnaire; the codebook; and any published articles that describe the origins of the data set.

Investigators who make their data .avftila'ble to the public 'qften dq not e;xpe·ct to he coauthors on papers written by independent analysts. However, they may .expect ·their contributions to be recognized and/or the soufces of funqing anti technical sup- port to be-acknowledged. The supporting documents should stat.e the expectations; if they do not;, the' reseaq:her should ask a t ontact person fot clarific·ation. Aha, some- tim.es, the analysis reqtiires assistance from the individuals involved in designing the study and/or colletting and processing the data. If so, those individuals may qu;:tlify for coauthorship even if the supporting documentation does not say that this' is necessary, individuals. s.hould a.sked about their expectations.

There are some major limitations to using already ava:ilable data. One is that the analyst is limited to exploring orily.the topics included in the original survey . .A r:dated

170 CHAPTER 23 Secondary Studies; Existing Data Sets

conceui is that the analyst has, to trust thai the data were collected using valid and stan- dardized methods and that the supporting documenta;tion accurately describes the 'ac- tual procedures, used for data collection. Another challenge is that finding someone who can answer questibns about the procedures used might be difficult. Some down- load websites do not list the name of a contact person, and some of the listed contacts may not have' been integrally involved in the study design -and data collec:tion process. A final issue is the risk of duplicating the analysis that-someone else has done or is do- ing.-A literature search may uncover related works that have q"een published or are in press, but it will not identify analysis in progress or papers under review by a journal. The contact ,petson fo:r the data set may not know whether other researchers are con- ducting an analysis of the data or what topics other researchers are focusing on.

Despite its secondary analysis is often excelknt option for researchers with strong statistical skills but limited;time and/or data collection resources.

• 23.3 Private Data Sets Individual researchers ·and small research teams may have data available that have not yet been analyze.d. Sometimes-the researcher-shave computerized data files that have not yet-heen fully explored. Sometimes paper records have been set because they 'are not a current priority of the resea.reh te?m. Or the original researcher or research team may- have published the r.esults of somep·ortion of the data set1 but left unanalyzed some of the other potentially s:ignificant, interesting, and novel aspects of the data. In this sit- uation, the original researcher may be open to a new taking the lead on ana- lyzing tha:t portion Qf the data set and writing up the results for possible publication.

A request for access to a private data set is most likely 'to be, granted when the new researcher has some connection to the original Tesearcher.' Students are most likely to have success asking,their professors for data sets to analyze. Alternatively, if student_s are interested in the work of a research group a.t another university or hospital, they may ask their professors to reach out to colleagues at the other institqtioh.

When privately held data are shared with a new investigator,_ the original researchers usually expect to he coauthors on any tesulting publication. The roles and res ponsi- bilitie.s of each party should be agreed on as early as possible.

• 23.4 Clinical Records Clinical re:cords are a common source of data for case series. Individuals working in clinical settings commonly have access to p_atjent for research purposes, pro- vided that the research project receives all requir-ed approvals and wilLnot violate any

_la,w or such as the Health Insura,nce Portability f}nd Accountability Act (HIPAA) Privacy Rule that must be adhered to in the United States.

23.4 Clinical Records 1 71

Sometimes the relevant information can be extracted from an electronic database. When electronic records are not available, a data extraction form can be created and used to extract the relevant information from each file. The extracted information can be entered directly into a computer database or recorded on paper for. later data entry. When possible, the data files should not contain any individually identifying information.

A major limitation of using existing clinical records is that patient records are of- ten incomplete. Researchers camiot assume that missing informatiop. is the same as a no. For example, they cannot <).ssume that the absence of information about a symp- tom means that the patient did not experience the symptom. The patient might have had the symptom but failed to mention it to the clinician. Perhaps the clinician did not specifically ask whether the symptom was occurring. Maybe the patient did men- tion the symptom but the clinician did not record it, perhaps because the symptom did not seem especially relevant. Similarly, even if a patient's records at one clinical site do not mention that he or she is taking a particular drug, the patient might have been prescribed the medication by a clinician at some other site. When data must be col- lected in a particular way to be useful to the researcher, a primary study design may be necessary.

• 23.5 Ethics CommiHee Review Additional approval by an ethics conunittee at the institqtion where the secondary analysis will be co·nducted is usually not requited if several co.nditions are met:

• The data to be analyzed are publicly available. • The data set contains no individually identifying information. • The data were collected following approval by a federal government or some other

widely recognized and reasonably trusted entity.

If the data come from a private source, then, prior to even looking at the data set, the analyst must obtain clearance from his or her own institution and perhaps also from the institution that houses the data. Use of hospital records for research pur- poses always requires review by a research ethics committee.

Researchers with questions about whether their project requires review should consult the appropriate committees. It is better to err on the side of submitting a perhaps unnecessary proposal than to erroneously presume that a project is exempt from review without confirmin.g the validity of this assumption. Chapter 22 provides information about the ethics review process.

172 CHAPTER 23 Secondary Studies: Existing Data Sets

Tertiary Studies: Systematic. Reviews and Meta-Analyses

A; s;y.s_tcpn<JJfli¢ is·. thf! careful c,oitfp:i14t.1ort.,fi.nd s.un{mdf.y flf:allFP,u.l:diQtitiali:§ ·. A syt!J'fl{i'tlryst(fti,s'r ·

tic for tie.s1ft,lt8c of . .

• · 24.1 Overview of the Systematic Review Process As noted in the intro.ductiort to r.eviews of liter-ature in Chapter 7; the systematic review process requires th·e

• Identification otan appropriately narrow study question. •· Se1ecti·on of .·q: well se·arch ·s,xratt;;gy; • Screening ofaJl potenballycrelevant articles to determine whether they meet the . . .

predefined 'eligfbi1ity ctiteria. • Extraction oJ re1evant 1n£ormati'rm from all eligible .. ·artide:s. • ot the flndirtgs or i:trtides,!\ ..

FIGURE 24-1 .summarizes the systematic review process. In some situations it is qppro,. pti;:Jte c,o create ·<;t swflJll<lry by pooling data; frvrn the included ,studie.s,. a process Galled but this is not . ..

173

.·

FIGURE Systematic Review Process

• 24.2 Search Strategy After identifying a well-defined study question, the next critical step in a systematic re- view or meta-analysis is to select appropriate search terms and search limiters. For example, a systematic review might require included articles to:

• Be indexed in MEDLIN£ with certain specified MeSH terms • Be written in English • Be in or after 1:995 • Use a case-control or cohort study design • Have a minimum sample size of 20 humans

Or a systematic review .might involve searching two or more databases with the same set of keywords, allowing publications in any language (assuming that coau- thors and :friends can assist with translation) and in any publication year, but restrict- ing eligible articles to randomized controlled trials.

Or a systematic review might involve looking up every article cited in an included article to try to fully capture the entire published literature on th.e topic (a process some- times called "snowballing"). The goal is a complete, unbiased list of related articles.

To check the appropriateness of search terms, identify a handful of articles known to be relevant to the study question. Then determine whether the search terms capture all of these articles. If the search misses one or more of the reference pieces, then the search strategy needs to be· modified. However, this process must not be used to ex- clude disliked articles, which would cause the indusio·n bias that systematic reviews seek to minimize.

Once a system for identifying eligible articles is in place, abstract databases are systematic.ally searched for articles that meet all the inclusion criteria. If the topic is appropriately narrow, then keyword searches and limiting factors can often reduce the number of abstracts and/or articles that must be screened for eligibility to a rea- sonable number, often less than 100 articles. Most systematic.reviews end up with about 10 to 25 included articles after screening, although some have many more than that. The full text of each of these screened artides must be read to determine final el- igibility. Ideally, each article should be assessed by tw·o independent reviewers. FIG- URE 24-2 summarizes this process. The count of articles at each step-identification, screening, checks of eligibility, and inclusion in the manuscript-should be included in the research report.

1 7 4 CHAPTER 24 Tertiary Studies: Systematic Reviews and Meta-Analyses

..

# of abstracts - . Identified

from each database. or

strategy

FIGURE 24-2 Systematic s,earch St.rategy and Counts to Report

• 24.3 Data Extraction Once all eligible articles are identified, the content of these articles is extracted into data extraction tables that list descriptive characteristics like:

• The study location • The study years • The study design • The study population and sample size • The key findings of interest • The strengths and of the study

A data extraction table allows for easy compilatiqn and comparison of-observa- tions relevant to the study question.

When interpreting the results of a systematic review, studies that find no statistically significant results for an item of interest are just as valuable as those that find a signifi- cant association. The researcher should record and report both statistically significant

24.3 Data Extraction 1 7 5

findings (p < 0.05) and statistically insignificant findings (p;:::: 0.05). A report may state, for example, "Five of 40 published studies of the association between exposure A and disease B found an increased rate of disease B among those exposed to A, and the re- maining 35 studies found no association." That is a more accurate depiction of the lit- erature than if the report merely says, "Five studies found an increased risk of disease Bin those exposed to A." The latter statement incorrectly implies a consensus that ex- posure A is significantly associated with disease B. One of the primary contributions of systematic reviews to the health science literature is the ability to identify areas of con- sensus and areas of disagreement that need to be further examined.

Systematic review reports also need to address the possible influence of publica- tion bias on the findings. Publication bias occurs w hen articles with statistically sig- nificant results are more likely to be published that those with null results. If 10 studies look at the association between the same exposure and the same disease, the one study that finds the exposure to be risky is much more likely to be published than the nine null result studies. Even if the other nine studies are published, they are likely to high- light some other statistically significant aspect of their research and to downplay the lack of a positive or negative association between that exposure and disease. Proving that publication bias has occurred may not be possible, but the presence of consensus should be conservatively interpreted when only a limited number of studies have been published on a topic or the results are mixed.

• 24.4 Meta-Analysis A meta-analysis pools the results of several studies identified during a systematic re- view to create one summar y statistic. Only similar statistics from similar studies can be pooled. For example, a summary estimate of efficacy can be estimated from several high-quality randomized controlled trials with the same active intervention, the same type of control, and similar population groups. However, the results from studies that use different study desigiJ.s br dissimilar population groups should not be pooled, Pooling several unadjusted (crude) odds ratios may be appropriate, but pooling a mix of crude odds ratios and age-adjusted odds ratios is usually not.

Before pooling the data, the researcher must show that the results of the studies are comparable. Homogeneous (similar) studies can be combined into a summary sta- tistic, but a great deal of caution should be used if the studies are heterogeneous (dis- similar). The amount of variability in the measure between studies can be examined using a Q-statistic for homogeneity or another appropriate statistic ..

If a summary (pooled) statistic appears to be appropriate given the variability among the studies, the next step is to select a model that will be used for creating a pooled estimate of the effect size, which is the estimate of a measure like a sull}mary odds r a- tio, rate ratio, efficacy, correlation coefficient, or difference in means. There are two main choices: a fixed effects modd or a random effects model.

176 CHAPTER 24 Tertiary Studles.: Systematic Reviews (!nd Meta-Analyses

• A fixed effects model can be used to create a pooled estimate (such as a Mantd- I!aens·zel odds ratio) when .the ate fairlybomogenous:•

• A-random. effects model is required when the tests of heterogeneity show that the studies .ate .di$sitnilar.

The point estimate ·for the sumtnar:y measure will be similar for both model types. Bowevet;, -a random ·effects model wilLresult in a wider 95% confidence interval for the summary statistie because the tahdom effects model will adjust for the variabUity' between the included studies.

o ·nce a model is:selected,, a specialized computer software program can be used to estimate the value of the po,oled statistic .and its confid:ence interval. The contribution of each study·to the pooled estimate is :usuallyweighted on ·the sample, size of the included studies, although other ,approaches to weighting can be used .. Step-by- -step guides to meta-analysis techrtiques'are a,vailable from The Cochtane Collaboration :and oJher reosource

The contributing ,srqdi.es th¢ sutnrrurry tn,easute often dl·splayed tJS:irtg @ forest plot ( FIGURE 24-3). A forest plot usually has:

• ?J"is ,showing ef£eGt size. • A ·vertical line showtngthe effect;size that indicates :no ·effe:cr (such as .an -odds ra-

tio qJ 1). • A .row for information from each inducled study· that uses a square or othet marker

tQ indi,Ca.te point for the effect si;t:e' au,d us·e,s a horizontal line to show the 95"o/o confidence,interval.

• for the. point' in varying ·siZ:es that show 'how ·each study was weig;hted in the meta-,analy.sis:. Small markers usually indicate Studies:With,sma11 sample $izes,. and usu"q,lly indicate S:tudies with large sample sizes: ..

• A. representation ofthe.summary; often sh()wn usinK a diamohd shape.

Point .estimates and corrti'dence intervaJs

induqed stl)qies

Pooled: point esti'mate:and confidence ir'l,terval

' ' I { ·• '

1

•

• q;

FIGURE 24-3 Example,.ofa. ForestPiot

• •

2 2.5

··· · ,, is

·•

• - . : . ",,

24.4 177

There are two main threats to the. validity of a 'meta-analysis: poor quality of in- cluded studies and publication bie1s·. The selection criteria used during the systematic review process can eliminate any studies of questionable validity. . .

The possibility of publication bias-the preferential publication of studies that re- port a statistically significant and/or favorable outcome--can be examined using a funnel plot. A point for each included study is graphed on a funnel plot that shows ef- fect size on the x-axis and sample size on the )'-axis (FIGURE 24-4 ), If no publication bias has ·occurred, the poii1ts for· the included studies will form a oone shape. If pub- lication bias has reduced the number of publications with statistically insignificant re- sults, a part of the cone will be missing. In that situation, the pooled estimate is likely to have overestimated the true effect size.

, ; ':·: PunneLwhen there is · '• ;n0 effect and and ''' a.pcrsitive effect · no p4blioation bias ndjitiblicatioh bias · pu.Biicatlon bias

Effect > 0

Each point repre-sents one study the

meta-analysis · ·

FIGURE 24-4 Example of a Funnel Plot

178 CHAPTER 24 Tertiary Systematic Reviews ;;tnd Meta·Analyses

;

Analyzing-Data

Identify stuqy

'Question

Select study

approq,ch

Qesign :stutly .&

collect oata Report

findings;

the fourth. step l n ·tne rese.qrch Is 1@ 'compife <lftd we.re col during step 3.-Most research only fheuse·,of descriptiWHirtd perhaps sante t om- paraflve. statistics. • Data ;management

Descriptive st'ItiStics , eom'J:>atati9e:$tat lsti<;s.: • Aovanted:healtb.statfstics ,,_ ·( . ' . . .. . . ,. .. .

179

Data Management

Data -·n1:anagement refers 'tO the e;ntit'e process of reco_ril keepifl:_g,;whether t-rack- tug eligibilitJ'i1ui systematic re-vieW, B:;qtracting.data frorh p-ertientcharts for:ad:ase e'(tteringthe resp.·onses to a cross..:sectional or case- denttol or reciordi.n:g:. a:U the"'result$ of.cli'l:tz'<Jal asse-ssment conducted dur- ing 4 'QY ¢xperitf.Jental stuqy. After data are the files·

be c}etJPJ-ed andi(lerhctPste·cod-ed.before lie,gi11:hin:& s:tcitisticaLanalysis.

• 25.1' Codebooks Prior to beginning data entry, it is. useful t0H:r,eate a codebaok-tha:t de.scrihes each vari- able-a.rtdspedfies how the collected information will be en.tere,d tprtiputer data.- base (FIGURE 25-1}. For quantitative surve;ys, numeric Dr alphabetical codes can he assigned tQ th:e for dose:..ended answers provided on the For {)pen- ended questions and qualitative surveys, a codebo.okis even more essential hecause it provides de,4f in$Jfuctions how· to

In addition to providing specificinformation piece of informat:ion $hoqld b,e: ip.tp the cori1ppt¢r 'file,the'co4e.boQk sh<;mld specify:

• The name. of each variabLe (whkh usually employs only capital letters or a: com- hination o'fcapitalletters· and and. avoids starting with a symbol, such as an urrder.score.)

ti' The wording. of the ques.tiort that,was asked e, The vatiable type . • The ·options listed on the survey as possible answers to the. question

l81

:

Question Variable Variable Variable Number Name Question Type Length Codes 1 INTDATE {Oate of 'date 8 • Enter-as

interview} DD·MM-YYYY

2 AGE What is your numeric 3 • Enter·number age in years? • Misstng = 999

3 · SEX What is text 1 • your sex? • Female = F

• Missing= {leave blank}

4 WORK Which of the text 10 • Working full time= following FULLTIME

• Working part time = trest describes PARTTIME your work • Unemployed but want status? to wqrk = UNEMP

• Retired= RETIRED • Student= STUDENT • Homemaker = HOME: • Other= OTHER

If OTHER go to 4bJ otherwise skip to 5

4b WORK_ OTHER Other text 50 • {Enter text as reported by respondent; only

description enter for those for whom WORK= OTHER.}

5 STUDENT Are you text 1 • Yes= Y currently • No=N enrolled in • Don't know !Missing! 5chool? Rejused =D

FIGURE 25-1 Example of Codebook Entries

• The way answers should be entered into the computer database • What to do with missing answers

The codebook is also the place to describe how anticipated problems will be handled. For example, what should be done if a respondent selects two answers from a multiple choice list when the instructions said to select .orily one? What should be done if the same person accidentally turns in two copies of the completed survey form? What if the handwriting on a form is illegible or the data entry person is not absolutely certain about what the words say or which box was checked? If unanticipated quan- daries arise, the code book should be amended to state how the situation was addressed.

182 CHAPTER 25 Data Management

Question Variable. Variable Variable Numb¢t Qu.esfion Type Length <:odes ·5 ALC How often do numeric 1 • Never-=0

you drink • Less than r time. a alcofi()l? m.orith = 1

• Ab.ot.Jt 1 timea month=2

. • About 2 times a month =3

• About 1 time a week= 4· AIJQut 2-3 times a week= 5

• AbDut 4-5 times a week=6

• Every day or.. almost e:vei)f day = 7

• Don't khdw = 8 • Refused/missing = 9

7' Has a ddttO'r numeric. 1 • Yes= ·1 ever told-you . No-= '0 tbat,you had a • know= 7 sex:uaUy . Rr;fused to an>wer = iJ · . Missing =9 disease?

FIGURE 25-1 (continued)

The codehook will also specify for each variaJjle whether missirtg answers should be left .blank in the data bas:e, indica ted with a n umexic code (such as entering a 9 if the expected entry code is 0 or 1 for a drchotomous variable), or marke·d ·with the word "MISSING." The statistical analysis pr:ocess may needadjustmentba·sed on what the codehook how missing data were handled. For example; if missing infor- mation about a,ge is-entered as those entries will need to be remo_ved prior to analysis, or else the mean age will end up artificially high and the standard deviation will be very large.

• 25.2 Data Entry Data are usually entered into a database program (like Microsoft Access). One of the benefits of these programs is that they can be. designed to be. vis:ually aJ3pealing and tQ indude preapproved responses to questions . .and automatic skips between questions .. This ensures cqnsistehcy oferrt:ries an_d the completeness ot the. file.

25.2 Data Entry 183

An alternative option is to enter the data directly into a spreadsheet program (like Microsoft Excel). Variable names should be entered in the first row, with one variable per column. Each individual's data should be in a new row. The advantage of this data entry approach is that it does not require creating a data entry form., defining fields and variable names, and doing other coding and testing of the data entry system. The disadvantage is that it is easy to input inconsistent codes, which makes cleaning the data much more difficult, or to accidentally enter new data over an existing row of data.

Both database and spreadsheet files can be uploaded into standard statistical soft- ware programs for analysis.

It may be worth doing double-entry of at least some of the completed survey forms (often a minimum of 10% of them) to check the accuracy of data entry. Double- entry consists of two individuals entering the same data (or the same person entering the data twice) into two different computer files, then comparing the records in the two files for agreement. Special software programs (such as the Data Compare utility that is part of the U.S. CDC's free Epi Info program) allow the individual records stored in two files to be linked by a unique ID number or other variable and compared. These programs usually provide statistics about the agreement level. If the agreement is not extremely high, then it means that the double-entry and comparison of all records is probably required to ensure the accuracy of the final data file.

The file comparison programs usually facilitate the creation of a clean final data file. They identify disputed entries and allow the researcher to select the best response for the final clean data file after consulting the original survey forms. For example, sup- pose one of the two data files indicates that a participant was 32 'years old, and the other says that the participant was 42 years old. The origina1 form completed by the participant may indicate that the true age is 42, and 42 can be selected as the cor- rect entry for the cleaned file.

• 25.3 Data Cleaning Data cleaning is the proces's of correcting any typographical or other errors in data files. FIGURE 25-2 shows how errors such as extra spaces, typos, and the use of lower-case instead of capital letters can be fixed so that the responses all adhere to the codebook. When paper-based data collection methods are used, fixing incorrect en- tries sometimes requires looking up the original survey forms. For example, while an "N'' for SEX might reasonably be assumed to be a mistyped "M," it is not clear whether an ''R" for STUDENT refers to a "Y" or an "N." In such situations, there- spondent's file must be consulted. Missing values .in a computer database may also re- quire reference back to the Driginal survey forms. Sometimes information missing in the computer file may have been on the survey forms but overlooked by the data en- try person.

184 CHAPTER 25 Data Management

·Variable F M m N N

N R y y [lVlt5sJng]

' .

FIGURE 25-2 ExampJe of Data Cleaning

;2

1 87 ., 4

N ·v ffvliss-iogr -, ·:: ' _ '·

9'03 89.

2"

This is. als.o tqet:iine t¢ dean up extrefi+ely unre.a·sonable answers. Far sup- pos_e: a participant's age in years is listed as 192. This number canreas:anably· be. as-

to be a typo, and tht; origirt;;tl s:urvey fotm shquld be: consulted' to';r th¢ true age. If the s urvey form lists th:e age ·as· 192' or if the survey was computer-assisted and there is no trail, then this, va'lue ·should be ex<:lud.ed from an:alys1·s bes:ause it is :an impossible age .. However, a study:ofadults ·could:reasonably include a-n individual with an age of 105 yeats. So this value would hot· be tea$,d}rable to ot although it would be worth checking the original survey -form for agr.eement with the ehtty in the

Data cleaning should also ensure that duplicate entries are-removed from the data- base 'a.nd with .4Il. data froh} all entered into the .datab-ase.

• 25.4 Datil Recoding The recoding into rtew categories can be done either prior to ot during dat<(l a,nalysis .. Rec;oding prior to intense analysis. is often the easiest approach when the in- tended new categqries .. are known. For exa:mple1 the variable AGE could be used to cre- ate a new ADULT that:is:coded as 0 (no).for :any participant younge:r than 18 years old and 1 (yes) fot any a_ge 18 or ·older. The variable WORK could he used. to create a new ;varial:>le .caUe.d FULLTIME that is coded as l.(yes) if the an-·

to W01U\. was FULLT!ME and 0' (no) £or any O'ther respt>nse. ·

25.4 D,ata Re.coding 18.5

A few basic practices will help protect a cleaned data file. Never do any recoding until an original version of the cleaned data file is safely backed up elsewhere. A saved file allows a researcher to go back to the original and start anew if a file is damaged during recoding. Also, never recode into the same variable; that is, do not.replace the original values with the new recoded values. Instead, always recode into a different (new) variable. Having the original variable and the new variable in the file enables there- searcher to compare the original and recoded values, thus confirming that the recod- ing was done correctly.

• 255 Maintaining Confidentiality Chapter 23 emphasized the importance of maintaining the confidentiality of any in- formation participants disclosed to researchers. One way to maintaJn confidentiality is to safely store paper records, including signed informed consent statements, in a locked and secure room. Another is to destroy individually identifying information once the records are no longer needed (such as after the data have been entered into a computer file and the files have been thoroughly cleaned) and a research ethics com- mittee has approved the secure disposal of consent statements and other documents.

Another way to protect confidentiality is to create secure computerized data files. In general, no individually identifying information (such as a name or national iden- tity card number) should be included in an electronic file containing other informa- tion about participants (such as. responses to surveys or the results of laboratory tests). If there is .a need to link records to individuals-and there is no need to do this- then the records should be linked to identifying information by a unique study iden- tification number. The file containing individual names should be stored in a separate and secure place, not on the same computer as the other participant data. Files con- taining identifiable information should be password-protected, and access to them should be limited to essential research personnel. Consult with an information tech- nology expert or a research ethics committee if que&tions or concetn,s about securing participant information arise.

186 CHAPTER 25 Data Management

Descriptive. Statistics

Wnen appropriately and ·statistics provide essential and :useful1nforination for making sense of health research da(a. Descriptive s_tatis- ti,cs (flre t-tsed ttY desmdb,e the basic c;/har,+tcteristi'(s of study populatiorts' and other ·ddt a sources.

• 26.1 Analytic Plan by Study Approach Statistics can he used 'to tell a complete .and compelling story about the data collected during. a research study. For mo_st papers_, and especially those written by researchers. ·with iitilited experience in adva.nced statistics, the -goal of analysis should be to use the simplest statistics possible to make the results of the study clear. Most rese·ar<::h -studie$ du not requite the use of complex statistics like regression, and usin.g advanced statistical tests incorrectly is, of course, never helpful.

The type of-analytic plan thc;rt is .. commorrly used with each of the m4j0r study' a,p- proaches is shown in FJGURE 26-1 . Each starts with a description of the study popula- tion. with no comparison group, like case :Series and <;ross -sectional .surveys,. may need only univariate analysis. Simple statistics, like counts (:frequencies), propor- ti'on.s,. and, avs;rages, .are likely to provide adequate description of the study popu- lation . . For studies that compare two o.r more ·populatiorrs-includiqg case-control, cohort, and expedme·ntal studies-Jhe description of the study ·population must be completed before moving on to bivariate analysis, such as the calculation of rate ra- 'tips, odds ratius, and other comparative s,tatisticaJ tests. Cdes<::ribed in Chapter 11).

"187

,•. --

Describe Regression and the study Compare other advanced population .. groups - analysis

r ..-

(Univariate (Bivariate (Multivariate analysis) analysis) analysis) i '

.. _ .. .series Cross-sf3Gtienal.survey

case:.control study -----------:-------···············-···························· Cohort study __ __._.;.;__ ___ --=---------··········································

Experimental study ----------......;.;..-----·········-·······························

FIGURE 26-1 Analytic Plan

Advanced statistical analysis that examines three or more variables at one time is rarely required (and is briefly described in Chapter 28 ).

• 26.2 Types of Variables A variable is a characteristic that can be assigned more than orie value. Examples of variables that could be examined during a population health study are age, sex, an- nual income, languages spoken at home, frequency of alcohol ingestion, history of chicken pox, and use of contact lenses. The value of a variable for an individual does not have to vary (change) over time, but the response among individuals within a pop- ulation ·should be something that might differ.

In many statistical and database programs, responses from individuq.l participants are displayed in rows with each column representing one variable-. For example, one column of data may represent sex. One value for sex-an F for females or an M for males-will be listed in each row. Another column may represent age in years, a nd one value for age-usually a whole number-will be listed in each row.

There are several ways to classify variables ( FIGURE 26-2).

• Ratio uariables. have numeric r esponses plottep on a scale on .a value of zer-o stands for "nothing.'' .For example, if height is in feet, a measurement of 0 feet tall means that there was no height. As a result, the ratio of heights is meaningful. A person who is 6 feet tall is twice as tall as a person who is 3 feet tall, yielding a ratio of 2 to 1.

188 CHAPTER 26 Descriptive

Ratio

lnt_ervar

Norriinal/ categorical

8Jnomial

Definittbn Numl)e(s 6ii a thq_t has· a

zero

Numbers on a scale that does· not Mave:a meaningful zero

An· series ranK !b responses (front firsno last in the series) but for which the nurri'oers assigned to the \ta'lues are. not meaningtu.l Categorieswith no -- inhererit-r:aBk or oroer Nomhial tor which only, ±wb

FIGURE 26-2 Types. of Vari(lb_les

neTght weight (If theweighqncreases from 10 kg to' 20kg weight has doubJed; soUle' 'ratio of 20 .k\gto H) kg is meaningfuL)·

Temperature ("F or oq (Tbe temperature does not do,uble if it increases frqm zoo to o0 .does not Qf all heilt.) HigtiesJ de-gree t;arped1 scales fornever (1) t.o;always (5), scales: for strongly tHsagree (TJ tg·strongly agree (5)

EmpJoyment cat¢gory; blood type.

• Interval variables are also numeric, but they are plotted on a . scale on which zero do_es not St;1nd tor '''nothing."' .(\n outside-temperature of 0°C dcJes not mean that there is no heat; tfthe weatherturns colder,; the temperature ,may fall to -l0°C or lower. A day with a high is not as hot ;3.8 a day wrth 'a ,maximum tem'{)erature of 20°C.

-• Qr(/inal vci,riqble.s, qr ranked va,ri'ables, order responses from first to last or &om best to worst or from most favorable to least favorable. The rank order can he assigped a numbe.t:; t,e·sponse.s, to a 'SUtvey to indiea_te their level of agreement with a statement:can be coded with agree, as "3, '' neutral as "2," and as "1.," Alternately, responses could he coded. with as ''1-';. and 9is- agree 3·." Or neutral could he set as agre.e;:as: ''1," and -disagree as "- No matter what the scale is, the orde;r of the responses is h1dicated by th.eir numeric: val- ues. (Figure 18-4 prnvides,- exa:mples -ofother types of ranked responses.)

• ·or categotical ,vatictbles, ,categorical responses with no inhe-rent rahk or order. For example, there is nt> obyious.way to numerically rank participants' -favorite recr_eational sports activities or bloo.d types. Bin:O.mia} v{f.,ri- ables a,,te a subtype· of c-ategorical variable's with only two usu- ,ally :yes and no.

26.2 Types of Variables 189

Ratio and interval variables can be further classified as either continuous variables or discrete variables.

• Continuous variables can take on any value within a range. For example, although height is often rounded to the nearest inch when it is measured, a person's height could actually be 64lh inches or 73% inches or 58.1528 inches.

• Discrete variables typically result from counting something, so there are gaps be- tween acceptable values. Fot example, a family can own 2 egg-laying chickens or 17 chickens, but cannot own 2lh chickens or 51.4 chickens.

• 26.3 Measures of Central Tendency Descriptive statistics are often used to describe the "average" response to a variable in a population. (For numeric variables, the average is often referred to as the central tendency.) There are several ways to report the average (FIGURE 26-3 ) .

• The sample mean is calculated by adding up the values of all responses provided to a question and dividing that sum by the total number of individuals who answered the question.

• The median is the middle number when all responses are put in order from least to greatest. Half of the responses in a data set will be greater than the median, _and half will be less.

• The mode is the most common answer given by respondents.

For ratio and interval variables, the central tendency can be described using means, medians, and modes. For ordinal variables, a median or mode can be reported. A mode can be reported for categorical variables.

Measure Values Reported of Central by Participants Tendency Value Calculation

25 Mean 39.5 (25 + 30 + 30 + 40 + 50 + 62) -;- 6 40 =237 + 6 = 39.5

30 Median 35 The two middle values from are 50 30 and 40; 35 is halfway between 30 and 40. 30

Mode Two participants provided a response of 30; no other 30 62 responses were listed more than once.

FIGURE 26-3 Example of a Mean, Med1an, and Mode

190 CHAPTER 26 Descriptive Statistics

.·

:• 26.4 Measures of Spread Means and medians, provide informatiorvahout the center :of a. data set, but they do, not providejp:tormation about how much v·ariabil'tqr in the,dq.ta, ·sci., for ·exarn- :p:le, the partki19ants in a :study of adults witn a mean age of 50 years::may all be SO years. old, or theY cctqJd range from 1 s· t.o io4 yeats; old .. Th?1tinfot111:ation is v·eJ:y iPJ.p6r.tant to have wben interpreting the meaning of the results. Measures of spread;· also called ·dtspe.:rsiQ.n, are, l;lse,d t.Q the: variability .a.Q.d range ot re:spon$es ..

The range for a variable is the difference between· the responses .. with the greatest and least. num¢i:ic Fot it the youttge.st participant in a stll.d)f is }_8: years old iandtheoldestis 104 years old, the rangeis ,l04 -18 = 86 years.. ·

The median int;trks the value that _divides into two h:llv¢s·with equal numbers of observations. Quartifes mark the:three valueS'tha:t divide a data set into four equal parts' . .Similarly, tert,iles dJvide a data set into three equal par:tJ:." ;quintile$ divkle :a data, set into five e-qual patts., and deciles divide a data set into 10 ·equat'parts. The interquar-

(IQR) is the rau,ge for the 25th to 75th petce:nt1lese which captures middle 50% of A /;Jp:vplbt (also called,a.'bo:xc-and-wfiisker plot) c'.ln be.us-ed to -display this:info.rmation (FIGURE 26.-4). Bo;xplots. can b.e esp.edally helpful for.displayiqg the

more, than 1----

1.5 tQHs trom the median)

. {iriter- 75th J IQR,

L::·+---- Median. quartile . ,.___ __ 261h percentile ra:n.ge)

11---- Outlier

FIGURE 26·4 Sample Boxpiot

the "whisl<ers'' (inn.er fen:ces)

:.show the highest :and lowest values .or (in case') 3 times the JQH,

rs narrewer

26:4 Measures of Spread 1 91

8

7

6

5 .... c ::::::1 4 0 (.)

'3

2

1

0

Sample size= 37 Mean=30.4 Median=30 Mode=30 Range=37- 22=15 Standard deviation=2.5

20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 '37 38 39 40 Response

FIGURE 26-5 Sample Histogram For the same data shown In Figure 26-4.

distribution of responses when the responses are skewed. Skewing occurs when the "whiskers" on the boxplot extend much farther on one side of the median than on the other side.

A histogram is an alternate way to display the responses to a numeric variable like a ratio variable or an interval variable (FIGURE 26-5 ). On a histogram, the x-axis shows the values of responses, and they-axis shows the count of the number of times each re;sponse was given. For a graph to be considered a. histogram, each bar must be the same width. Importantly, there should be no gaps betWeen the bars in the middle of the distribution, where responses are clumped together. (There can be gaps to indi- cate values of the variable with a count of 0 responses.)

A histogram showing a normal distribution (Gaussian distribution) or approximately normal distribution of responses will have a bell-shaped curve with one peak in the middle (FIGURE 26-6). However, not all numeric variables have a normal distribution. The distribution may be skewed, with responses that extend farther from the peak on either the left (left-skewed) or the right (right-skewed) side of the histogram. The dis- tribution may have a bimodal (two-peaked) distribution instead of being unimodal (one peak). Or it may be uniform, with about equal numbers of people ·providing each response.

For variables with a relatively normal distribution-a reasonably bell-shaped curve-the standard deviation describes the narrowness or wideness ·of the range of responses. When the responses are normal:

192 CHAPTER 26 Descriptive Statistics

-:a: -2 Sta,ndard SD d.evia:tiorts

(SO) I .

I , . . <

: .68%of individuals: • .... II!•

' ' I I ·: of ·individu.i;ils : I I I . ·t

:>99% df

·-1 SD'

z-scor.e: ·Z = -3: z·= ,_2 .Z = - 1 z=1

+2 SE>

,Z ==2

. I

+.3 SD

i=.3

FIGURE. 26-6 Exarnple ohhe Distribution ofRespo:nses ·tor a Normally distributed NumericVariabte

• 6"8% of :respan<Ses £all within Dtie standa:rd deviation above o;r the hiea,n_. • :9S% ofrespunses are within tWo standard deviationsabove .. or below the: mean. • More than 99Uo ofresponses are within three .. standarddeviations above o,r below

the mean (Figure

A\S·mall standard deviation indicates that most responses were f:airly close to tne mean. thatthe (esp·onses:WC;l.S wi-qe.

A z-score indicates hcr.w many standard deviations: away from the sample mean is. Fot

• An indiv:iduaJ whose age is :exactly the :mean in the population will hacv e ·a z- 0-.

·• A pers·on whose. age is one' standard ·deviation ahove;the mean in the population will. have a z-s'Core, o.f 1 ..

• A. persc>n ·whose.· age is two· standard devi:atit1ns below the populatron mean will have A histogram otbo.xplot cannqt be·us,edt() displ;ty th& responses vari-

ables. The distribution of resp.onse;S must instead be displayed in a bar chart less,of-- feJi, a pie C.hart.}.Like·,q, histog:ra,m, of'q..bat tJ1.i;i.rt7shows of

26:4 Measures nfSpread 1;g3

-c 25

20

5 15 (.)

10

5

0 A

FIGURE 26-7 Sample Bar Chart

8 c Response

D E

and they-axis shows the count of the times each response was given. However, for a bar chart the x-axis can display either a number or a word. And while histograms re- quire numbered bars to be evenly spaced along a number line, responses on bar charts may appear in any order (FIGURE 26-7 ). The bars in bar charts can be displayed verti- cally or horizontally, and there are usually spaces between the bars.

The goal of descriptive statistics is to describe accurately all the responses to a , variable (FIGURE 26-8).

• For ratio and interval variables, the mean and standard deviation are typically reported.

• For ordinal variables {and for ratio and continuous variables without a normal distribution like the bell-shaped curve shown in Figure 26-6), the median and in- terquartile range are often reported.

Common Measure Common Measure Typical Means of variable Type of Central Tendency of Spread Display Ratio Mean Standard deviation Histogram Interval Mean Standard deviation ·Histogram ordinal/ranked Median' lnterquartile range Boxplot Nominal/categorical Mode - Bar chart pie chart Binomial Mode - Bar chart

FIGURE 26-8 Common Descriptive Stattstlcs by variable Type

194 CHAPTER 26 Descriptive StatistiCs

• For .categorical variables, the proportion of participants who provided a p.articu- lar is us11ally used to describe the population.

• 26.5 Statistical Honesty .Researchers are obligated to describe their data accurately andto correctly report the results ofstatis.tical tests. To do otherwise is a torm of research misconduct. Three of the most serious forms of research misconduct are:

• Falsification-the misrepresentation of results. • Fabtitation-. the creation of fake data. ·• Plagiarism-the use of other peoples ideas or words without proper· attribution.

Statistic;il honesty requires more thah merely avoiding o,ptright falsification,. fabrica- tion, and plagiarism. It also requires adherence to accepted statistical practices .. For ex- ample, it tnay be tempting. to look tor statistical tests that will yield the the researcher desires, such as ones that are considered statistically significant. However, scientifkintegiity requires researchers to foil.ow established statistical Examples of unaoceptable practices include:

• Running a. dozen different types of statistical tests, on a data set1 hoping that one o£ them will happe·n to. yield a statistically significant result to feature in a report. lQ.st ead,. the researcher must select the cor:re.ct test £,or the question being :asked and the variables being; examined. -

• Recoding ratio variables into categorical variables by preferentially ·selecting the cutoff points·that yield statistically significant results for tests of the new c_ategor- icaL variable,

• Ignoring,outliers.-unusual responses to a question-w'ithout a. valid and standard reason. For example1 a recorded birth weigpt of 80 pounds may be reasonably as- sume·d to· be an error in the data file, it. can be removed from But it. is not. reasonable to remove an 8Q""pound adult from the data file because :an adult could weigh pounds.

Statistical analysis is about discovering the true story in a data set, not about cre- atively ma,nipul.atihg data towar d ca preferr-ed result.

• 26.6 Consultation and Collaborati-on the researcher shou]dconsult with a statistician duriqg_ the .study design process:

to ehsure that: ··

• The sampling methods and sample size are appropriate.

26,6 Consultation arid. Collaboration 195

• The questionnaire will yield usable data.- • The analytic strate·gy is a reasonable one.

Checking with an expert for the first time later in the study increases the risk of un- fixable flaws in the study da_ta. If answering the study question adequately requires the use of elaborate analytic techniques, invite an expert in that technique to serve as a col- laborator and as a coauthor on the resulting paper. An invitation to collaborate should be made as early as possible in the project and in .cons1Jltation with. other coal,lthors (see Chapter 5).

'196 CHAPTER 26 Descriplh/e StatistiCs

Comparative Statistics

Compa:rative.statistics compare groups of participants by sex or age, by exposure or or by 'qthe:r characteristics. Examp{es of cbmpatadve statistj .. , cal tests include rate odds and Chi-square tests. This chap- ter provides_ a: brief oifervrew a( p-values, cfJnfidefzce {nte'rl!t?l'$, aftd sem.e of the most' com.mon comparative statistics used in the health sciences.

• 27.1 Comparative Analysis by Study Approach Some tJpes of stt,tdies require use of comparative statistical tests. These tests egprize study partkipants into two or more groups and compare the characteristic:s, of:

study' Approach First Step Key Analysis Case:-control study Show that cases ·and controls use o·dds,ratibs' (ORs) to see

are simiiar,ex-Geptfar disease whether cases and controls have status different exposwre histories-

Cohort study Show that the exposed>and, Use rate ratios· (RRs} t0. see_ whether unexposed aresim'ilar the exposed and· unexposed nave for expdsute stt';ltus different of

.Expenmental_gtudy . Show that the inc::Hyidl!al$ rate ratios oth¢( the intery¢iitibM measures-to s.¢e·rt the interveJiti,Qn

control a'nd control have-different except for exposure· status outcome-s

FIGURE 27-1 Analytic Plan for C<;Jmparing Groups

197

the groups. For example, the analysis of a case-c·ontrol study req'uires using compar- ative tests to show that the cases (people with the disease} and controls (people with- out the disease) in the study were similar in terms of age distribution and other demographic characteristics. Then additional comparative tests are applied to deter- mine whether the exposure histories of cases and controls were different. Comparative tests can also be used to compare before and after characteristics of participants in longitudinal and experimental FIGURE 27-1 summarizes the lJ-Se of compara- tive statistics for several common study approaches.

• 27.2 Hypotheses for Statistical Tests Comparative statistical tests usually are designed to test for difference rather than for sameness. Accordingly, statistical test questions are usually phrased in terms of dif- ferences: Are the means different? Are the proportions different? Are the distributions different? Each question about statistical difference has two possible answers: The values are either different or not different.

Statistical Null Hypothesis Alternative Goat Question (Ho) Hypothesis (Ha) Test whether the average ages of Are the means The means are The means are cases and controls in a case-control different? not different. different. study' are s_imilar enough to be considered equal

Test whether the mean age of Was the mean The mean is not The mean is participants drawn from a populatjon age in the study different from 40. different from 40. with a mean age of 40 years is cto:se . population enough toAD-years that the study different from popt.llatlon can be considered · 40 years?

of the source population

Test Whether the proportion o_t Are the The distributions The distributions responses to ,a categorical question distributions of are not different. are different. about the-frequency of flossing was responses similar for-male and female different? particlpi.ilf;lts in a cohort·study Test whetper participants, on average, Are the before The scores are The scores are had a :Ciiange in their scores on a scores of not-different. different. pretest-acfrninistered prior to an participants intervention and a post-test different from . administered after the intervention the after scores?

FIGURE 27-2 Examples of Hypotheses for Statistical Tests

198 CHAPTER 27 Comparative Statistics

The term ·null hypothesis (Ha) describes the expected result ol a statistical test if there is no difference between the two values being.compare_d_. (N1:-tll means nothing or zero. A null result means that there was no statistically significant difference.) The alternative hypothesis (Ha) describes the expected result if there is a difference ( FIG- URE 27-2). For example, for a test to compare the mean ages of two groups of study par- ticipantsJ the hypotheses could be. as follows:

• 'H 0 : There is no sign'ificant difference between the two • 'Ha: There. is a significant difference between the two means.

A test to pare the distribution of responses to a question in two groups would have:

• Ho: There is no significant difference in the distribution of responses in the two populations. .. ·

• Ha: Ther.e is a significant difference in the distribution of responses in the two populations.

• 27:3 Rejecting the Null Hypothesis Because st.atistic.al tests do not ask questions the answers provided by statistical tests do not allow a researcher to say conclusively whether two values are the same. Instead, a researcher must make a conclusion about whethe,r the results of a statistical test inqicate that values are different ot not differen.t. The language used to describe this decision is that the researcher will either ''re)eGt the null hypothesis" or ¢'fail to reject the null hypothesis." · ·

• Rejecting the. null hypothesis means concluding that the,values are different by re- ject'ingthe that the values a;r-e not different.

• Eailing to reject the null hypothesis means concluding that there is no evidence that, the val\leS are differen't. Functipn,ally, this is like saying' that the val\leS are. dose enough to he considered similar? but failing to reject the null hypothesis. shbuld never'be taken ·as evidence that the. values are the; same.

The decision to reject or fail to reject the null ·hypothesis is based on .the likelihood that the result of a test was d.ue to chance, One way to understand the concept of chance is to consider the·variability in sample populations. When a sample population is drawn from a .source population, the mean age in the sample population is usually not exactly the mean age of the source: populaticm. ($ee Figure 17-1 for t:tn iHustra,.. ticm of the variety oJ sample means that can oc.cur in different samples-drawn from one source The' range, of expected.va:lu:es .fot the mean age of ·sample· pc>pula- tions drawn from a source population can be estimated using statistics ( FIGURE Sc>rne samplei populations. will have meah ages that are very dose tQ the mean in the

27.3 Rejecting the· Null Hypothesis 1

Mean in the source population

Sample means that are "extremely" far from the

mean In the source population

(the 5% of most extreme sample means - the 2.5% of means that have the lowest

values and the 2.5% of means that have the highest values)

Mean

FIGURE 27-3 Example of the Distribution of Mean Ages for Sample Populations Drawn from a Larger Source Population

source population; other sample populations will have mean ages that are quite far from the mean in the source population. No set cutoff defines what will be considered ex- tremely far from the mean agein the source population, but the standard is to say that the 5% o£ sample means farthest from the true 1Jlean ar.e extreme. Thus, by chance, 5% of the samples drawn from a source population will be expected to have an ex- treme mean.

Similarly, if two sample populations are drawn from the same source population, their mean ages will not be identical even though they are drawn from the same pool of individuals. Comparative statistical tests accommodate this expected difference when testing whether two groups in a study population are different. For example, a test that compares the mean ages of cases and controls in a case-control study adjusts for the fact that there will be some difference between the mean 'ages of cases and con- trols, even if the cases and controls are sampled from source populations with identi- cal mean ages. The test will also examine whether the mean ages are so far apart that, if the cases and controls were drawn from source populations with the same mean age, the difference between the mean ages of the cases and the controls would fall in the 5% of most extreme differences expected by chance.

When the difference in rnean ages is great, the statistical test will show that it is highly unlikely that the group means are not significantly different. The researcher will therefore reject the null hypothesis and conclude that the mean ages of the cases and the controls are different. The difference between the mean ages of cases and con- trols will be taken as evidence that the mean age of individuals in the source popula- tion for cases and the mean age of individuals in the source population for controls

200 CHAPTER 27 Comparative Stt\tistics

are different. This conclusion assume_s that the difference between the source popula- tions is reflected in the sample of cases a:nd contro-ls that happened to be drawn from their respeqtive source,_populations.

If the statistical test shows that the inean ages of cases anq controls are fairly dose, the researcher will fail to reje€t the null hypothesis and will conclude that the means are not diffetent. .

B 27.4 lnterpretingp-Values A p-value, or probabtlity for a !)tatistical te,st used to decide whether tlie re- sults observed are likely to reflect real differences between groups. The interpretation is tor all statistical tests: the p-vaJue the stugy determines whether the null hypothesis.(Ho) will be r·e}ected .. The standard is. to use a significance levetof a= Oo.OS, or 5%. Any statistical test with a result that is in the 5% of most extreme responses expected by chance will Tesult in the rejection of the null hypothesis ( FIGURE 27-4 ). Alternatively, studies use a= 0.01, which makes it harder for a test to find a sta- tisticaJly significant E,esult that would .cause the rej·ection of the null hypothesis-. Others

0.10, which makes it more likely that a tes·t willyield a. statistically significant

Some p-values are reported as beihgoile-side:d or two-sided, based on the alterna- tive hypothesis for the statistical test. While mnst statistical re·srs use an alternative hy- P-Othesis thatslmply expresses difference (such as "themeans are diffetent"), some tests allow for an altern:ative hypothesis that states the, direction oft}).e difference.· (like ''males have,a higher mean age 'than females") (FIGURE 27-5). If a direction is-specified in.'the al- ternative hypothesis, then all of the exttem'e values (all of the shaded area shown 'in Figure'2 7 -J) will be on side of the distribution {either ali on the left of the distri- bution or :all on the right). When this direction is specifie·d" a one-sided p-valu·e can b'e

Ho =reject H0 p z 0.:05"' =Jan to reject R0; The are notdilferent:. . . . The means are different The means. are not different.

The,,propqrtions,.are The: prqporti€HJScare not different. pifferent.

dfsVibuliorts .are QOt .different. The are . The distr:iblitiohs:ate riot diff-etent. ct iffer'e,:nt. _

·*'Assuming a= 0.05. FIGURE 27-4 Interpreting p-VaJues,

27.4 Interpreting p;Values 201

1\vo•Sided Alternative Example of a Null Hypothesis (Ho) Hypothesis (Ha) Alternative Hypothesis (Ha) The means are The means are different. The mean of cases is higher than not different. the mean of controls.

The proportions are The proportions are different. The proportion oflhe intervention not different. group is lower than the proportion

. of the control group . The scores are The scores are different. The after scores were, on average, not different. higher than the before scores.

FIGURE 27-5 Examples of One-Sided and Two-Sided Alternative Hypotheses

used. In all other situations a two-sided p-value should be used to make the decision about rejecting or failing to reject the null hypothesis.

• 27.5 Interpreting Confidence Intervals Confidence intervals (Cis) provide information about the expected value of a measure in a source population based on the value of that measure in a study population ( FIG- URE 27-6). For example, if the mean age in a study population of 1000 people sampled from a large city is 30 years, a researcher should not assume that the mean age in the whole city is exactly 30 years. The 95% confidence interval states how close to 30 years the mean age in the source population (the whole city) is expected to be. If the 95% confidence in- terval for the mean age in the study population extends from 28 to 32, a researcher can be 95% confident that the mean age in the city is between 28 and 32 years.

The width of the interval is related to the sample size of the study. A larger sam- ple size will yield a narrower confidence interval. If every member of the source pop- ulation is included in the study population, then a confidence interval is not needed because the exact value for the source population will be known.

A 95% confidence interval is usually reported, and that corresponds to a signifi- cance level of a= 0.05 for a statistical test. This means that 5% of the time a 95% con- fidence interval is expected to miss capturing the tr ue value of a measure in the source population. Using a 99% confidence interval (a= 0.01) would make the confidence interval wider and make it more likely that the value in the source p_opulation would be captured within the confidence interval. But it would also make it mqre difficult to classify a result as statistically significant because fewer results would be classified as extreme. Alternatively, a 90% confidence interval (a= 0.10) could be used. A 90% con- fidence interval would be narrower and make it easier for a result to be deemed sta- tistically significant because more results would be classified as extreme. However, a 90% confidence interval woulct be less likely than a 9 5 % confidence interval to

202 CHAPTER 27 Comparative Statistics

Statistic Mean_age of aT( participants(yearsy

Proportion of <:lll partiCi'parit? with a

(D/o)

Otlcls rat1 o ·

Rt?lative ri?k (RR)

Result with g_sofo Cl ·rnterprefation 30 (28, 32)

9:.o (7:3.,. 10,9)

'1.7 (0.6, 53)

1.6 (1.1, 2A)

Basetron the.mean age in the stud,y pgpulat!nn. (3.0 years)1 we· are 9soto conftdent'ttiafthe,mean age in the source popul1:1tion is between '28 and 3'2 years.

13as,ed' .on the proportion of individual's. in. the study 'Nho'had the dis¢ase w¢ ate

9SW9 confident that th'e preval.en·ce ;of dJ.sea_se irt . the sourcEr popu·lation is between 7.36/o and 10 .. 96/a. Base:d on the OR in the stud,y popula!ibrt = · 1.7t we are:9 S'Qfo confident that the: OR in the source pGpulation .is somewhere b·etween 0.6 and s.·l. Because this overlaps.with :OR= l, we .conclude that there is no 9nd. g Based'on the RR inthe studypopulatiqn'(RR= 1.6)., .are 959/o confitlent that the RR in the. sciur<;e pop.ulation' . is between lJ and.2.4 .• Because thrs-range does not · overlap wlth RR = 1, we conclude ,thai the e)(posure i's

'with an iri.cteaseq fisk o.f dJ!;iease. ·

FIGURE 27-6 Interpreting Confidence Intervals (Cis)

capture the true value in the source population .. For example, a 90·% confidence in- terval for an odds {OR) i$less: likely to overlap with 1 than a_ 99% confi- dence interval. So, although the 90% confidence interval is less likely to the trlJ.e odds rado,,tt is also more likely that the OR ,will be deemed to show a statistically significant asS'otiatio.n between the expo·s11:re ·and the outcome {FIGURE 27·7).

;·· .. ·>

.. ,·;

FJGURE 27-7 90°/o, 95%; and Collfidence Intervals (Cis) fort he Same Odds Ratio (OR)

27.5 lnterpretin_g Confidence Intervals 203

• 27.6 Measures of Association Some of the most common types of comparative analysis are the measures of associ- ation explained in the chapters on the various study approaches, such as the correla- tion used for ecological studies (Chapter 8}, the odds ratio (OR } used for case-control studies (Chapter 11 ), and the rate ratio (RR} used for cohort studies (Chapter 12).

The OR and RR compare responses to two variables that have each been divided into two levels using what is sometimes called 2x2 analysis . Prior to using a computer to calculate an OR or RR, variables that are not already divided into two categories must be recoded into binomial variables (often coded numerically as yes = 1 and no = 0 ). In some situations, the cutoff points for the categories are o bvious, such as t hose that divide an ordinal variable into categories for disagreement (strongly dis- agree or disagree) and agr eement (agree or strongly agree). Sometimes the population can be divided into groups of r elatively equal sizes using the median, quartiles, or other sample-based cutoff points. Alternatively, biologically or socially meaningful cutoff points can be defined, such as using the 18th birthday to divide a study popu- lation into children and adults. The choice of cutoff point will influence whether the

Percentage of of Controls

cases(AMI) (no AMI} Exposure (n = 150) · (n = 250) OR(95%CI) Interpretation

Female 39.3% 4 2.4% Reference cases and controls in

Sex group the study did not have

significantly different Male 60.7% 57.6% 1.14 proportions of males.

(0 .75, 1.72)

no 37.3°7o 51.2% Reference Cases had greater odds

Waist group than controls of having

a waist circumference circumference yes 62.79/o 4 8.8% 1.76 greater than 35 inches. >35 inches (1.16, 2.67t

. Never 68.00/o 73.6% Reference Cases and controls in smoked group study population Former 1.1'.36/o 10.8% 1.14

did nofhave · Tobaccoose · smoker '(058, 2.18)

signJficanUy ,different smoking histories.

Current 20.7% 15.6% 1.43 smoker (0.84, 2.44) .

· statistically significant at 0.05 level. FIGURE 27-8 Example of Odds Ratios fOr a Case-Control Study of Acute Myocardial Infarction

204 CHAPTER 27 Comparative Statistics

.·

exposure and outcome have a statistically significant association. Accordingly, the de- cision about how to define categories should be justifiable.

The results of 2x2 analysis are often presented using tables like the one shown in FIGURE 27-8. The reference group for an odds ratio or rate ratio should be well defined. In the example, males are compared to females, those with a waist circumference greater than 35 inches to those with Smqller girths, former smokers to never smokers, and cur- rent smokers to never smokers. (Two separate odds ratios were calculated for tobacco use because the variable for toba_cco use had three possible responses insteqd of just two.)

The 95% confidence interval provides information about the statistical signifi- cance of the tests. For exa,mple, the 95% confidence interval for the odds ratio com- paring the sex distribution of cases and controls contains OR:::: 1. This means that it is not clear from the test whether cases are more likely or less likely than controls to be male. The conclusion is therefore that there is no statistically significant difference in the proportion of cases and controls by sex.

• 27.7 Selecting an Appropriate Test For statistical comparisons more complex than 2x2 analysts must select a test that is appropriate to the goal of the analysis and the types of variables being analyzed. The steps for identifying and using a statistical test are summarized in FIGURE 27-9.

First, the variables be compared should be selected and the goal of the test clearly stated. The goal could be.:

• To compare the mean age:? ofmales and females (variables: 'age, sex). • To see whether the proportion of cases and controls with various blood types is

similar (variables: blood type, disease status). • To determine whether individuals starting an exercise program had, on average,

lower heart rates one month after starting the program than they did when they enrolled (variables: initial heart rate, heart rate after one month).

Then select a test that is appropriate for the types of variables being examined. Some tests require the variables being examined to have particular distributions or other characteristics. The researcher must confirm that the variables meet these as- sumptions of the test prior to running it and interpreting the output.

FIGURE 27-9 Plan for Hypothesis Testiflg

2Z7 Selecting an Appropriate Test 205

Statistical tests are often classified as being either parametric or nonparametric. The basic difference between these two types of tests is that parametric tests make more assumptions about the variables being examined than nonparametric tests.

• Parametric tests assume that the variables being examined have particular distri- butions, often requiring the variables to have normal or approximately normal distributions. These tests may also require that the variance for the variable of in- terest-the spread of observations around the mean-be equal or at least similar in the population groups being compared.

• Nonparametric tests do not make assumptions about the distributions of responses.

Parametric tests are typically used for ratio and interval variables with relatively normal (bell-shaped ) distributions of responses. Most parametric tests are more sta- tistically powerful than nonparametric tests. So the preference is to use a parametric test whenever the variable being examined fits reasonably well with the assumptions the test makes about sample size, distribution, and the equality of variances.

Nonparametric tests are o_ften used for ranked variables, such as the responses to surveys that ask participants to indicate preferences using scales from 1 {strongly dis- agree) to 5 (strongly agree). They are also used when the distribution of a ratio or in- terval variable is non-normal. Additionally, non parametric tests are used for categorical variables, including variables with just two groups (such as cases and controls, males and females, children and adults).

• 27.8 Comparing a Population to a Set Value The goal of some statistical tests is to compare the value of a statistic in a study pop- ulation to some set value. For example, suppose that participants in an experimental study are students at a university at which the mean age of undergraduate students is 21 years. The researcher wants to confirm that the mean age of the study participants is reasonably close to 21 years. If the distribution of ages in the study population looks like the distribution in Box A of FIGURE 27-10, then 21 years is captured within two

years years

(c)

21 years

FIGURE 27-10 Comparing the Sample Mean to Some Other Value (One-Sample t-Test)

206 CHAPTER 27 Comparative Statistics

standard deviations of the study's mean age. The conclusion would be that the sam- ple mean is not so far from 21 that the means would be considered different. In other words, the sample shown in Box A fails to reject the null hypothesis that the means are not different. The conclusion is that the means in the study population and the university student population as a whole are not significantly different. Box B also captures 21 years within the 95% confidence interval, even though the mean age of study participants is farther from 21 than it was in Box A. In Box C, however, the study participants were several years older than the average student at the university, and 21 years does not fall within the 95o/o confidence interval. This indicates that the study population may not be adequately representative of the university's undergrad- uate student population. In this situation, the null hypothesis is rejected, and the con- clusion is that the means are different.

• 27.9 Comparing Independent Populations Sometimes study participants are grouped into independent populations, which are pop- ulations in which each individual can be a member of only one of the population groups being compared. For example, if the populations being compared are divided by age, the population of adults ages 18 to 49 will not overlap with the population of adults ages 50 to 99. Each individual participant in the study population can be assigned to, at most, one of these groups. So the populations are considered independent.

Type of Variable Being Examined Ratio/Interval Ordinal/Rank Nominal (Parametric Tests) (Nonparametric Tests) Binomial Categories

Statistic being Mean Median Proportion Proportions evaluated Test for whether the One-sample One-sample Binomial Chi-square statistic in one popula- t-test median test test (x2) goodness- tion is different from a of-fit test hypothetical value Test for whether the Independent- Mann-Whitney U test Fisher:s .. st(}tistic differs in two S§mples (two- (Wil,coxon rank sum exac1 test (X?) test populations sampfe) Hest test, Wilcoxon-Mann-

Whitney test) Test for whether the One-way ANOVA Kruskai-Wallis test Chi-square statistic differs in two '(F-test) (x2) test . (x2) test or more populations

FIGURE 27-11 Tests for Comparing Two or More Groups

27.9 Comparing Independent Populations 207

..

Males Females variable Test of p-value Variable Report (n = 200) (n =200) Type Comparison for Test Interpretation

Mean Ratio Independent- The means are Age (SO) 43.7 (Z8) 41.1 (8.1) (normal) samples t-test 0.001 djfferent Current % 12.0% 9.5% Binomial Asher's exact 0.519 The proportions smokers (yes / no) test are not different Home n (%) . Nominal Chi-square 0.864 The proportions district: (x2) test are not different.

North 90 (4S.O%) 87 (43.5%) Central so (2S.O%) 48 (24.0%) South 60 (30.0%) 6S (32.5%)

FIGURE 27-12 Examples of Tests for Comparing Males and Females in a Study Population

A variety of statistical tests can be used to compare independent populations. The appropriate test to use depends on the type of variable being examined (FIGURE 27-11 ) . For example, a two-sample (independent-samples) t-test could be used to compare the mean ages of cases and controls participating in a case-control study. A Fisher's exact test could be used to examine whether the proportions of males in the exposed and un- exposed groups of a cohort study are similar. A Chi-square test could be used to de- termine whether the distributions of participants by race or ethnicity are similar for the intervention and control groups of an experimental study.

When running statistical tests, it is often beneficial to create a table of basic infor- mation about the variables of interest for each of the comparison groups as well as the result of the statistical tests used to compare those populations. FIGURE 27-12 shows sample output for tests of whether responses differed for the male and female partic- ipants of a cohort study. In this example, the males have a significantly greater aver- age age than the females because the p-value for the independent-samples t-test is less than 0.05. However, the proportion of males and females who smoke is not significantly different because the p-value for Fisher's exact test was greater than 0.05.

Characteristic Males (n = 200) Females ( n = 200) p-value Age Mean (SO) 43.7 (Z8) 41.1 (8.1) 0.001* Current smol<ers % 12.0% 9.5% 0.519 Horne district: n(%)

North 90 (4S.O%) .87 (43.5%) 0.864 Central so (2S.O%) 48 (24.0%) South 60 (30.0%} 65 (32.5%) .

•statistically significant at a= o.os level. FIGURE 27-13 Simplified Version of Figl1re 27-12

208 CHAPTER 27 Comparative Stafistics

table shown in Figure,2 7-12 includes mote infoxmation than is usually in- cluded in published manuscripts. Howev'et, it allows the researcher to double-check that the correct tests were used and that the correct interpretations were made. It also facilitates the writing of the statistrc-al ,methods portion of the methods section of a research report. A more succinct eomparison table is usually prepared tor the final re- port. A satriple results table is shown in FIGURE 27-13.

• 27.10 Comparing Paired Data A set -6£ is used when the goal is to compare bdore-and-after results in the same individuals ( FIGURE 27-14). If the goal is to see whether, on average., a partici- p::tnt in a cohort study gained weight between the baseline exam and the 1-year follow- up exam, a matched-pairs t-test can be used. If the: goal is to see whether a safe driving course the pass rates fot a driving licensure exam, McNemar's test can be used t0 examine how many participants ,switched fro-m failing a ,pretest to passing a post- test, how many switched from passing a pretest to failing a post-test, and how many had no change in status. McNemar's test tan also determine whether the differences inditqte that the:cours,e had a sig,nificartt impact on exam pass rates:

FIGURE 27-15 shows:samp1e output for paired tests. In this example, participants in a 3-tnonth e.x·eftise program lost weight during 'the-study period because the p-value for the matched..:pairs t-test was less than 0.05, but the participants did not increase their ab!lity to run 1 mile in less than 1 Q _minutes'. The p -value for McNemar's test was greater than 0.05, which indicates that there was no ·differenfei n this variable during the study period.

. Test for wnefh¢,r the value of the variable is

' ,. ' . . . - . di,fferenfin Olie J:)Qpula" tion measured twice '(such as before-and-after ln or il} twqpa)red grOUp$ Test for whether :the. vaiue of Variable is different, in two or more

groups

Type· of varia_ble. Befng R9tio/lhte.rval OrdinaVRalik· Nominal :

(NQtiparametritTests) Bfhpmial Categorie-s _ (paired)

Wilcoxon pairs) t¢_srorsJgn test'fqr matched pairs

Qne: 'v'Vay Friedman te?t . . measures-.AN()VA

McNemar's McNetnar:s· fe$t

(:o:chrah"s Q Q test tesf

FIGURE 27-14 Tests for Comparing Matched Populations

27.1 0 Comparing Pa'ired Data 209

N .....

0 ('" \

:I : > m ;:o

IV

"- A

"0

C J

V l S"

0 : C!

" '.I

. "'

(;

,·

·' I

? :.:;,

Va tia

bi e

pr cl

es t

Po st-

Te st

W Pe

· T

es t o

f C om

pa ris

on

fo rT

es t-

.

sa m

pie Si

te

n, ·

. -,

"

40 :

40

0 C

b'! ltl

t -

- -

W el

gh t

M ea

n 17

7. 9

17 1.

5 -6

.5

.. R at

to

M at

ch ed

-p ai

rs t-

te st

<

0. 00

1 pt

e- :an

ct po

st -te

st w

ei gh

ts

(p ou

nd s)

.;(s

o)

(1 8.

8)

(1 8.

6)

(5 .7

) (h

or m

aO

·fo r in

di vi

cl ua

ls w

er e,

o n

av er

- ag

e, _s

ig nl

fic an

tty d

iff er

en t.

Ab le

to ru

n n

(% )

12

16

6 no

-to -y

es ,

B in

om ia

l M

cN em

ar 's

te st

0.

28 9

Th e

pr e-

an d

po st-

-te st

ab ili

ty fo

r 1

m ile

in

(3 0%

) (4

0% )

2 ye

s- to

-n o,

an

in dM

du al

p ar

tid pa

nt to

ru n

les s t

ha n

32 n

o ch

an !!€

1

in le

ss th

an 1

0 m

in ut

es

10 m

in ut

es

w as

, o n

av er

ag e,

n ot

d iff

er en

t.

FI G

UR E

2 7-

15 E

xa m

pl es

o f T

es ts

fo r C

om pa

rin g

Pr et

es t a

nd P

os t-T

es t

fo r P

ar tic

ip an

ts in

a 3

-M on

th E

xe rc

ise P

ro gr

am

A Brief Guide to Advanced Health Statistics

Only a very limited number of studies require regression analysis or any of the other advanced statistics that are described in this chapter. User-friendly statis- tical software programs. hcwe made it possible for nearly everyone to run ad- vanced statistical analyses, bu.t these programs still require the user to select appropriate tests and decipher what the output means. Researchers should not use these tests without first knowing when to use them, what conditions have to be met to make their use appropriate, how to run them, and how ·to interpret them. This chapter provides a quick reference to some of the most commonly used advanced statistical techniques. ·

• 28.1 Confounding and Effect Modification One of the main reasons researchers use multivariate statistical models-that is, analy- ses of three or more variables at one time-. is to examine the inter actions that may occur among variables. This, can be especially helpful when a third variable (also called an extraneous variable or lurking variable) may be concealing or distorting the true relationship between two other variables. Several different types of third variable ef- fects might occur, including confounding and effect modification.

A confounder may make the association between an exposure variable and an out- come variable appear more or less significant than it truly is. For example, the crude (unadjusted} odds ratio for the relationship between physical inactivity and a first heart attack may show that inactive adults are four times more likely than active adults to have, a heart attack. Yet age may confound this association because older adults are more likely than younger adults to be inactive and also more likely to have a heart attack. Age- specific analysis may show that inactive young adults are twice as likely as active young adults to have a heart attack. It might also show that inactive older adults are twice as likely as active older adults to have a heart attack. These results for analysis stratified

211

.·

by age would indicate that age was confounding the association between exercise and heart attacks. When a third variable is shown to be a confounder, an adjusted measure of association, such as an age-adjusted odds ratio, should be reported for the associa- tion between the exposure and the outcome. In the example, instead of reporting a crude odds ratio of OR= 4, it would be more accurate to report an age-adjusted odds ratio of OR = 2. ·

An effect modifier (sometimes called an "interaction term") is a third variable that often represents biologically distinct groups of individuals who might experience dif- ferent biological responses to. various exposures. For example, menopausal status may be an effect modifier for some studies related to women's reproductive health issues. A particular exposure may be associated with a decreased risk of breast cancer in pre- menopausal women but be associated with an increased risk of breast cancer in post- menopausal women. If a third variable is shown to be an effect modifier, it is usually best to report separate stratum-specific measures of association for each level of the effect modifier (such as separate results for premenopausal and postmenopausal women). Pooling the results for the biologically different groups may hide meaning- ful differences, so an adjusted or crude measure of association should not be reported when effect modification is occurring.

FIGURE 28-1 summarizes the steps required to identify confounders and effect mod- ifiers. To be a confounder or effect modifier, the third variable must be independently associated with both an exposure (or predictor) variable and an outcome variable.

3rd variable (sex, in this example)

y (3)

Exposure---+ Outcome

How to identify confounding:

• Confirm that (1) is statistically significant • Confirm that (2) is statistically significant • Calculate three measures of association

(ORs or RRs) fo r (3): • Crude OR between the exposure. and outcome

• OR for 1 (female, In this example) • OR for stratum 2 (male, in this example)

When a third variable is associated with both (1) the

exposure and (2) the outcome of interest, unadjusted analysis may hide the true association between (3) the exposure and the outcome

ORtemale = OR male * ORcrude Confounding

I'·' ·"··C• .' ,_.;.:,,<:

.

ORtemale * OR male * OR crude I ' Effect modification

ORtemale = OR male = ORcrude Neither

• Interpret results

FIGURE 28-1 Confounding and Effect Modification

212 CHAPTER 28 A Brief Guide to Advanced Health Statistics

These two relationships should be confirmed. Then a crude odds ratio (or othermea- sure of association) for the relationship between the exposure and the outcome should be calculated, along with a separate measure of association for each level of the third variable, such as separate odds ratios for males and females. The crude and stratum- specific measures are compared using a Breslow-Day test for homogeneity or interac- tion, a -2 log likelihood test, or another appropriate statistical test. After running a suitable test, the interpretation is as follows:

• If the crude and stratum-specific odds ratios are all similar, then neither confound- ing nor effect modification is occurring. Report a crude measure ..

• If the stratum-specific measures of association are equivalent to one another but different from the crude measure of association, the third variable is a confounder. Report an adjusted measure.

• If the stratum-specific measures of association are different from one another and different from the crude measure of association, the third variable is an effect mod- ifier. Report stratum-specific measures.

• 28.2 Regression Regression is often the easiest way to adjust for one or more confounding variables or in- teraction terms during analysis. Regression models seek to understand the relationship be- tween one or more predictor (£ndependent) variables and one outcome (dependent) variable. The models allow the effect of one predictor variable on the outcome to be ex- amined while controlling for other predictor variables (that is, while keeping their values constant). The two most common types of regression are linear regression and logistic regression, which are discussed in the following sections. The steps for model fitting are similar for both types of models and are summarized in FIGURE 28-2 .

Some statistical software programs require the analyst to:

• Select a variety of specifications for the model, such as the particular estimation technique (often an ordinary least squares, generalized least squares, or maximum likelihood estimation model).

• Choose the method the computer will use to select variables for inclusion in the model. For example, an "'enter" method will include all predictor variables in the model. A "forward stepwise" method adds the best predictor variables to the model one at a time until adding .an additional variable does not significantly improve the fit of the modeL A "backward stepwise" method deletes vari;ables from the model until deleting a variable significantly reduces the of the model.

• Check the fit of the model by examining its residual terms, which measure how well real data match the values predicted by the model, and the r·esults of statistical tests of the goodness-of-fit for the model.

28.2 Regression 21 3

Step 1 Select outcome variable.

.,

2 Identify the appropriate type of regression (such as a linear or logistic model) for the out- come variable.

3 Select one or more predictor (independent) variables. 4 Check to make sure that any assumptions required for the model (such as the variable

types or of outcome and predictor variables) are met. 5 Choose a selection helping the computer to sef vari-

ables will produce the "best-fit" model (the model that the com·puterdetem'Hhes is the best at explaining the relationship between the predictor variables and the outcome variable).

6 Examine the model for potential problems. For example, examine residuals for possible au- tocorrelation, check for possible interaction between predictor variables (such as the multi- co !linearity that might occur when two predictor variables are highly correlated), and look for other potential problems that might need to be addressed.

7 Interpret the results of the regression model, and consider whether they are logical (for ex- ample, that all necessary covariates are included and all illogical ones are excluded).

FIGURE 28-2 Steps in Fitting a Regression Model

A statistics reference or a ,statistician should be consulted for detailed information about these and other advanced analytic techniques.

• 28.3 Linear Regression A linear regression model is used when the outcome variable is a ratio or interval variable.

Simple linear regression models examine whether there is a linear relationship be- tween one predictor variable and the outcome variable. FIGURE 28-3 provides an ex- ample of how to interpret the reshlts of a simple linear regression. The relationship between the predictor and outcome variables can be visually displayed using a scatterplot, and the regression model finds the best-fit line for those points. The slope of the line is the coefficient for the predictor variable (often designated as in the output of statistical software programs). The y-ihtt:;rcept for the line is the coefficient for the constant in the regression model.

These values can be used to write an equation for the best-fit lipe, and that equa- tion can be used to predict the expected value of the outcome variable for various val- ues of the predictor variable. The r1 for the model, which is the square of the correlation

214 CHAPTER 28 A Brief Guide to Advanced Health Statistics

The output for ttie regression model is: The .r2 for.the regression

13 SE line.is r 2 = 0.79, whieh (¢oeff.) (stpndard

p-value. means thatthe predfctor error)

Predictor..:_ 1 3.1 0.4 :<:0.01 variable 7Q<Vo of

the variation in the values Constant 0.9 6.4 o,sg of the outcome vada,ble.

The equation for the regre'ssion line is: 100, • • OUTCOME = 3.1 *PREDICTOR _1 t . ao .. •

1j) 60 •• . E • .. 1 0 .. B

value outcome value "' 40 . . 0 • • • 10 -20 •

0. • 15 47:4 Q 5 10 15 20

·2B ·87.7 Predictor_ 1

FIGURE 28-3 Example of a Simpl e Linear Regres·sion Model

The output for the. regression int>del is:

SE p-va,lue Predictor_ 1 0.5 0..1 <0.01 Predictqr _:2 0.6 02 0.01 Constant - 6.2 6.2 0.33

The. equation for the regression line is:

T he. ,'2 for is r2 = 0.87\ whl9h

means thatthe predictor variables e:J(plain 87% of

the Vatiation in the VC;ll ue.s of the outcome variable.

OUT.CO_ME = 0.5*PREQICTOR_1 + 0.6*PREDICT OR_2 - 6.2

If Predictor_2 is held constant, a 1-unil inq eas& in Predlctoc1 isassociated with a 0.!;;-unit. in the :e)\pecjed. value. of the·

out'Come \rarfable: Preclictor_1 value

fo 11

11

Predictoc2 value

30 -30 3 1

16.8

17.3 17.9

If Predictor_1 Is heJd. a 1-unit

in PredJctor_2 i:S,associater:l with a 0.6-unit increase. in the ex,pecte9 value: o.fthe,

outcome variable.

30

FIGURE.28-4 Example of a Multiple Linear Regression Model with Two Continuous variables

28.3 LinegrRegresslon 21 5

provides information about how well the regression model predicts· the variation in the of the otttcome variable. The vah;1e of r2 ra,nges frorri 0 to 1,, with larger values_ indicating a better model fit.

Multiple, lineqf regressi<;>n mori(:ls examine the, effects of several predictor vari- ables on ,the. value of the outc:ome. variable. FIGURE 28-4 provides a'n example of how to "interpret the output for a m't.1ltiplelinear regression with two continuous predictor variables. The for the predictor-variables and the constant,can be used to wtite an for a best-fit line. That equation be used to examine the ef- fect of each predictor variable on the outcome variable while contr.olling the other predictors by hoJdit;lg their values constant.

Multiple linea-r regression models can have both .continuous and categorical pre- dic:'tor. vatiable:s, as long a,s the responses to -c_ategorical variabks are expres.sed by numbers. F1GURE 28-5 shows how,to interpret models with multiple types of predictor variable$ that do not interact. In the example, a 1-unit increase in the va,lue of the

variable is associated with a 2·-unit increase in .the. value of the outcome e. the, coefficient for "Predictor_2" is :::::; 2 .. 0). This relationship between

"Predicto:r_2" ,and the outcome is the ·same for both males ,and £em2!-les, even 'though males have an 1.8. 7- unit higher value far the outcome than females (since the coeffi- cient for ,sex 18;7).

Th!3owtput for the regr_es:sion mod,el i.s: 100

13 SE p-value o·Females •Male$

Sex l8'.7 <0.01 Predictor_;'2 ;:?;p .0.1 0: co·nstant - 17.9 2.7 <0.01 0 10 20 30 50

The equation for the regression line is· OUTC"OME "" + o:2*PREDJCTOR_2 -17.2 IT Pred'lct?L2 is h.els:J constant, a male·has an

ta<7-u:ni(higher expected Pre!;lictor_2 Expected outco-m,a- v<\hJe than a

value value outcome value / female. 0 29 2.2.i y / For males, a 1-unit

increase in Predici,tor_2 1 20 40.8-Jl / wf!h a 2:.0-uoit

42.§ _r increa:s.e-in thtfexpe:cted 21 Ql,ltcome. vati\ilile, FIGURE 28-5 ExampJe of a Multiple Liile<u Regression Model wilh One Continuous·and One Categorical Variable with No Interaction (As lndi<::ated by the Parallel Lines for Fema.!es and Males on the Graph)

216 CHAPTER 28 A Brief to .Advanced Health Statistics

The output for the regression ·model is: 100

SE p-vah,1e ,80 ,C) Femal,es · (]) •M;;IIes. :$ex 4.1 < 0.01

c60 $:l 40 :::l

Predictor_Q 2.4. 0.1 :< 0.01 0 20 Sex* Predictor _2 - 1.2 ·o. 1 < 0.01 0

' o 10 20 $0· 40 50 Con:stant -20.3 2.8 <0.01 __:2

The equation for the regression line is· .. '· .OUT;COME = + 2.4*PREDICTOR.,..:2 For females, a 1.7yolt incre.a:se in Predic'tor_2-is

- 1.2*$EX .. PREDICTOR_2- associated wit.h ·a 2A-unit increase. in the

Sex Predictor_2 Expected / value of'the outcome value value outcome value variable. 0 (female) 20 27.7v · ,. Fo(males,.a H,mit 0 (female) 21 30.1_1 L incre.ase in Pr.e_dictor_2 is .associated with a 1.2-unit

1 {male) 20 31.7v increase in the 32.9.J

·expected yalu·e of the 1 (male) 21 ·outcq,me variable,

FIGURE 28-6 Example of a Multfple Lih·ear Regression Model with One Continuous and One Categorical Variable With (As Indicated by-the Non-parallel Lines for Males and. Females on the Graph)

The predictor variables in multiple linear regression models :may interact .. For ex- ample, interaction may be occurring w.heq :the best"'6.t tegr.ession for irrales a nd females have considerably different.slopes. FIGURE 28-6 shows how to interpret mod- els when ipteractiqn i§ oct;,utring :between sotne of the pt;.edktor variables. In the ex- ample, a 1 -unit increase in the" va1ue of the Predictor_2 variable. is associated with a. '2.4-unlt. ihcteqse i'n the of the Qlitcame. for females-o:ut only a1.2-unit for males. The equation f0r th.e -regression model expresses this interaction through the use· of a sp·ecial term,

• 28.4 Logistic Regression Logistic m.odels (.sometimes .called logit.regression models:) are use.d when the outcome variable is a dichotomqusvati'able.- i's in ease:-control studies, fpr which the outcome variable is usually case status, with ·case= 1 and .control = o: Forouttome variables that ate oth!er types yes/no vari-ables-, it· is typical to. let yes = 1 and-no = 0 . Predictor variables for a logistic regression ca n be or c'o:ntinuous.

28.4 Logisttc Re&ression 21 7

The output for the regression model predicting, beinQ a case (not a eontrol) is:

13 SE OR (95% Cl) p-value sex 0.59 0.49 1.8 (0.7, 4. 7) ._0.23

ate_food 1.44 0.52 4.2 (1.5, 11.7) 0.01 constant 0.37 0.03

OR for sex= exp(0.59} = 1.8 OR for ate_food = exp(J3) = exp(1 .44) -= 4.2 lower bound of 95%CI = exp(l3- = lower.bound of 95o/q,GI = exp(J3 -1.96*SE) = exp(0.59 -1.96. 0.49) = 0.7 1 .96*0.52) = 1.5 upper bound of 95%CI = Jixp(J3 + 1.9.6*SE) = upper bound of 95%CI = exp_(J3 + 1.96*SE) = exp(0.59 + 1.96*0:'49) = 4.7 exp(1 .44 + 1.96*0.52) = '11.7

Controlling for Controlling for sex, those The r2 for the model is (_a yesfno .variable for whether who ate the suspected food. r.2 = 0 .14, participants ate a certain item had significantly which means that the food),_ there is no difference odds of being a case than predictor variable explains by sex in the odds of being a those who c;li«;; not eat the 14%61 the vari'afion in the case: item: vatue:.sbf thE:foutcome' OR= 1.8 (0.7_, .4.7) OR=4.2(1.5, 11.7} variable.

1,.96 is multiplier for a 95% confidence interval, 2.576'for a 1.645 for a 90% Cl. FIGURE 28-7 E.xample of a MUltiple Logisttc Regression Model

FIGURE 28-7 prdvides a!). example ot the output Jor a iogistic regression and ex- plains-how to interpret it. The coefficient for a predictor variable in a logistic regres- sion model is the natural log of the odds ratio:> ln(OR). So the odds raJiQ, for the association between that predictor variable and the-outcome variable -can be found by taking the exponential of the coefficient, exp(p). The odds ratio for e?.ch predictor variable represents the change in the odds of the outcome-typically the odds of be- ing a case or being classified as a yes-fot a 1-unit in the predictor variable. The confidence-interval for the odds ratio can he calculated using the ;value of the co- effident. and its standard ,error, as shown in tht\ figure.

• 28.5 Dummy Variables The predictor variables in regression models 0an take a variety of forms but must have numeric responses. Nominal categorical variables responses that cannot be or- dered and assigned a rank; but a series of dummy variables that ,convert :categorical respons·es to a series of dichotomous ( 0/l) variables can be created. Additignally, when fitting a; logistic-regression model, it might be helpful to convert ratio ·and interval vari-

218 CHAPTER 28 A Brief Guide to Advanced Health Statistics

R$'punse tO: thE:!"Origihal Qu¢stion was .,,. A

a· t 0

Tf)en the V;;tlues of the: Dummy Varia.li>les. Are . . . ·

'B_Dummy (Was .Bthe

tne o (hol ·

1 (yes) .

0

Q

·(Was :c fl1e: response to, t.he otigin.al

?)

0

1' Q

D_Dummy (Was D the

question?) 0

d 0

l

Cprt_clus:ior'i a'a$ea ·on_ the Dummy vari_ables ·The was' not B; C, or D, so it was A i The response was Et The response was C.

28-8 Dummy Variable,s

abies to dummy variables so that ·a series of odds. ratios for the levels of the variable can be estimated.

FIGURE 28-8 provides an e:xample ofhow this recoding works·. If the; original cate,. gori<::al vaiiable has n. pos_si:ble responses, th en n- 1 dummy variabJes are to capture all the -re;spons·es to the nrigina l questiqn. AlT n- 1 variables shoulq be in - cluded in a regr-ession model (even if some -ma y be elimirrated d uring a .stepwis_e selec- tion ptote:ss ). · ·

• 28.6 Survival Analysis Survival analysis examj'ves the distribution. ofthe durations of time that individuals in a study-population .experience· from an initial time point {such as· the time of enrollment in

o r the tirne"of diagnosis ofa parri'cula:r well-defined event, which can be death or some other ®utcome. Measures o f surviv:aHnclude:

.•. Median survival time. • Cumulative sur·vival at set ti'me·s· a:·fter •' Life tables t hat record conditional and e;umulative probabiliti'es of survival. ' Kaplan-Meier plots that display cumuiative- sutvi'val FIGURE 28-9) .

Log.,.r a nktests can he used to det ermine whether s urvival is shorter in one: popula- · don than iu another. ·Cox haz}trds regression, which a hazard ratio that compares durations to an event (such as dearh):in 'two-populatibns; can also be; used !ot Sl}tvivalanalysi$.

2.8.6 Survival Analysis. 2·19

. ----------l ------------------- 1 : I ' I : About 85% of l About 55% of i the study · · l the study- : participants : participants l survived 1 year ! survived 3 ye_ars 1 2:' 3 5 _,

_,

FIGURE 28-9 Example of a Kaplan-Meier Plot

• 28.7 GIS/Spatiai Analysis If GPS (global positioning system) coordinates or other geographic data have been collected, then spatial software programs may be useful for conducting the geographic portion of the analysis. The geographic data should he incorporated into a GIS (geo- graphicinformation sy$t¢ifl,) ._ The GIS allows for spatial analysis such as:

• The identification of sp'J:tial disease clusters (using a statistic like· Moran's coeffi- cient or Geary's coefficient). · ·

• The determination of associations, if any, between the social or physical environ- ment and disease.

• The estimation of distances between locations. • The a_scertainment of the ge0graphic factors that are related to access to health

·serv1ces.-

A medical geography or health geography reference should be consulted for assistance with spatial analysis.

220 GHAPTER 28 A Brief Guide to Advanced Health Statistit:s

ld.entify ·study

questian

Select study

approach

Design study &

collect data

Analyze data

The.f!fth lti te'search. is;'tO vyrite. :a report an:q pre iUor :presentation .and puolication. This s:ecti.e.m provides tips torwrlting, fevrsing and disseminating .findings. ... Article structure • Citing • Writ!ng.strategies ! '' Critically reVising • Posters and p(esenlafJcJ'Qs. • · SeJecti:ngtargetj"Ournals

the subroissien,.review; and pubflcattonprocess • Wh,Ypubnsh?

Article· Strudure

Rese.dt'rch af.t.icles·almost always,hcuJe t/tJ:e scime structure: metfJodsi arrd diicussibnf -- ' . . ·' .... . ... ;:. ·' -.- -.. -

• 29.1 Abstract Th:t abstra ct is a summary of the article. Its-·most important function is to ·serve as an a<dvttttisement for the ·martuscript because computer cfatabases· and -search en-

have,:access onlytoabstracts:: Ev,en ·when researchers have a.copy 0£ the full text of an article, they willlJot pe likelyto readpa:srthe abstt;il.Cl if does not draw their at-,

Therefore,, an abstract ought to be accurate? reasonably complete, and com- pelling .. Writing. the can be :a when most jou:rnaJs limi1 abstrq.cts to a max:i'mum of 150 to 250 words .. lt is usually .easi'est to write'the.abstract after the rest qf paper· has 'alt·eady been the ·fo.cus, key tes11.lts, a;vd conclusions <H¢ dear.

A stta.ctured 4bst.r(lct use$ 'subheading§·, like objective, methods, and !:;on- elusion, ·to highlight content.

An,,unstrirctt,tred abstta.<;t us:t.ta:lly follows th:e $ame outline but ,doe$ rtot list the section titles.

223

• 29.2 Introduction The introduction provides the background information that a reader must know to un- derstand the methods and results of the article. This section often includes informa- tion about the study population, the study site, and the study years. Person, place, and time characteristics are usually included somewhere in the text of the article in addi- tion to appearing in the abstract.

The length of the introduction section compared to the discussion section varies according to the target publication venu e. For some journals, a typical introduction might consist of only one or two paragraphs, but a lengthy discussion section is ex- pected. For other journals, the introduction might be several pages long, but the dis- cussion section is relatively short.

The introduction section might include a comparison to previous studies and a discussion of what is novel about the new study, but that content might appear in the discussion .section instead. Some introductions provide a list of key definitions, but these might be placed in the methods section. No matter how much information is provided in the opening paragraphs of the paper, most introduction sections conclude with a statement about the importance or significance of the study and the specific aims, objectives, or hypotheses that the paper will address.

• 29.3 Methods The methods section should begin by clearly identifying the design used. If per- son, place, and time characteristics were not provided in the introduction, they should be listed in this section. Definitions should be provided for the key exposures, out- comes, and other variables. For example, for a case-control study, the case definition should be spelled out; for an experimental study, the intervention and control should both be described in detail. For some studies, supplying the exact phrasing and order of questionnaire items, along with the steps taktin to validate the survey instrument, might be important.

For primary studies, the methods used to identify, sample, and recruit participants should be described and the inclusion and exclusion criteria listed. The methods for collecting data should also be described, including interview techniques and (if rele- vant) laboratory methods and physical examination checklists and measurement meth- ods. For secondary analyses, the report should specify who collected the data originally, how they were collected, how they were acquired-for secondary analysis, and the role, if any, that the authors of the new paper had in data collection.

The methods section should provide information-about ethical considerations, such as whether inducements were offered, how informed consent was documented, whether community groups were consulted, and which research ethics committees

224 CHAPTER 29 Article Structure

reviewed the proiect. Ethical issue_$ can also be included in the _on the preference of the journal.

This section 'should end with a destr'iption of the statistical methods used. The methods s.ection can often be written even before data collection begins be-

cause most .of the fire finalized before data_ co1lef:tion starts.

,• 29.4 Results The re,sults section should start with a description of the study population that <;leady identifies the sample, size and the demographics of the participants. Additional results of statistical 11nalysis shol;lld theri be provided7 using_ tabks and figures when posS'ible. Most ,studies do. not require fanoy ·multivariate statistics. The results of a Statistical test shouid not be reported unless the authors fuily u.nderstand when that,test can be used and hew it,should he interpreted.

-• 29.5 Discussion The discuss'i'bn. settion'Qsuall-y' with of the-l<'ey findings of the: ne:w study. Ideally, the key findings should match ·the. aims, objectives:, or hypt:>t heses spelled out in the last paragr;tph of the iptto:duetion The pqtagraph:s shoul-d cQmpa.re the new study to previous studies and include a. thorough discussion of the. rdev;ant existing literature and art adequate number- of

Every,paper needs to indude.at1east one. paragraph on the limitations of the study. The limitations paragraph identify pro,blems. (such :as type.s of b_ias ). that could make the study results invalid ·or inaccurate. It should also explain the steps taken to problems -and why 'it is 'unlikely ·serio1,1s problem'S, occutred.

The final paragr-aph of the discuss,ion should state the conclusions of the. study. The appropfiate cqndusioils; vary by discipline and journal, but they might include new theories that emerge from. the analysis,. the policy implications of the study, or di- rectforts £or future research.

•• 29.6 EndmaHer Some journals list information between the conClusion and reference list. The so- cafled e:ndm:atter rn:a y in

• The affiliations of the-authors :and their contact information (if this is .not listed on t;itle page)

• The contributions of each author to thef>a_per

29.6 Endmatter 225

• Acknowledgments of people who assisted with the study but who did not meet authorship criteria

• Information abo ut some ethical aspects of research (such as a declaration that each participant gave informed consent or the names and locations of the commit- tees that reviewed the project)

• A list of all funding sources • Disclosures of the presence or absence of possible conflicts of interest (both per-

sonal financi al conflicts of interest and potential conflicts related to being em- ployed by an organization with a financial interest in the study)

Some journals provide this information in the final published version of the paper but request that it be removed from the submitted manuscript. The reason is that some journals use a double-blind review process, and this information could reveal the iden- tities or affiliations of the authors. Author guidelines of each journal will indicate what information should be provided in the endmatter.

• 29.7 Tables and Figures Many health journals limit the number of tables and figures allowed for each article, often to a maximum total of four (tables and figures combined). This limit means that the content for tables and figures must be carefully selected to highlight the most im- portant .q.,spects of the study. Tables should be used to organize and present statistical results that cannot easily be listed in the text in a sentence or two. Graphs and other figures should be used when a visual presentation of the material is more effective than words at conveying a result. Any images used should be meaningful, not merely decorative. There is no need to repeat information in the text that is provided in a table or figure, but be sure to have a callout for each table and figure that indicates when the reader should refer to the table or figure.

A table should provide ep:ough information ·so that it can be independently inter- preted and understood even in the absence.of the text (FIGURE 29·1 ).

• The title of the table should provide a brief but clear description of the content. • The rows and columns should each have a descriptive label and, when applicable,

provide units and/or sample sizes (which are often designated by n for the num- ber of participants).

• For each statistic, provide a confidence interval, p-value, and/or other measure of uncertainty, such as a standard deviation or standard error for a mean or an in- terquartile range for a median.

• A note just below the table (or in the title bar) should explain the meaning of as- and other symbols (such as t, :t:, and§ ) commonly used·to denote sta-

tistical significance and other items of interest.

226 CHAPTER 29 Article Structure

sex .,. ,., Ma,J,e Fernt:ue

,. blatrhea

' .' . V0miting.

· FIGURE 29-1 Example Frequency Table for a Case Serles

47(S4Wg) 40

· · .so (92Pto) .56.(6:40/o)

9 (ro%)

• Consistent fonts, spacing; and number of decimal points should ee used £or all ta- bles ih m:anusq:ipt. ·

A :grqph should provide enough information in the title, figure, and/or legend or key for a reade·rto be able to interpret the graph eyen witho1,1tteading the.re.latep por- tion of the text. FIGURE 29-2 highlights some n£ the features that may make. a g_raph

-or mote-difficult to interpret cortec:tly:. High-resolution photographs, maps, and other imag€s provided bythe .

. authors can. also he used as figures. Note that photographs ·of study participants are usually not :allowed without the written :permission ofthe-subj<wt or subjects.

FIGURE 29-2 Examples of Correctand Problematlc Graphs

29.7 227

• 29.8 Writing Checklists The most common information included in each section of an article-is shown in FIG- URE 29-3.

Section Content Abstract/summary the article. using key words.

Provide essential background infC>rmation. lntrpductionlbackground

State the objectives of the study (or, for experimental stuc)ies, the hy- potheses tested).

Identify the study design for experimental studies, the ran- domization method).

Describe the source population (including selection methods and eli- gi{jility criteria and, if re.cruiting methods), the setting, and the dates of the study .

. Methods

Define key exposures, key outcomes, and other variables.

Explain how data were collected.

Descrrbe how the required study size was estimated. Discuss ethical considerations (such as which researcsh ethics commit-

the project, whether an Inducement was offered, and hbW infoJmed consent was,gotumented).

Describe the statistical mefhbds u-sed for analysis.

Describe the study population, including the sample size (using a flow diagram to show the number of individual participants at each stage

Results of the study if that will be

Report relevant results (using tables and figures possible).

Sutnmarize key findings:·al)d bOW they relate to thB, ObjectiVes (orhypbtheses).

Discuss the limitations of the study. Discussion Provide a conservative and well-supported interpretation of the re-

suits, state how the new studylits with other relevant evidence (such as previous studies), and discUss the generalizablllty of the study (the populations to which findings.might reasonably apply).

Acknowiedge the contributions of each author, the by people whodld not meet authorship criter'ia (if any), the sources of fund

End matter and potential conflicts of interest {if any), if requested by the journal. -· References.

FIGURE 29-3 Key Content for Articles Reporting on Analysis of Individual-Level Data

228 CHAPTER 29 Article Structure

A number of checklists have been .developed for the content. to-indude· in reports pri :va·ri_l)ps·typc:s ·of stu.di(}$• Sqm:e q.£ checklists- list.e,d in FIG- URE 29-4.

(heddist ..

PreferredReportihg Jt.emsfor systemati¢ Reviews ana Q.f: in_t __ erven_ ti_.ons_) ·

PRISMA 1

Meta.;analy5Js: et ·observational :Studies in· EJiictemiolow

ntro __ l s .. _tu_·ay· ·· STR._ ... o· B'_E-c·a.-"'.·.e· _50n·t.r"'_-.· __ 1 D:trengtl'lenlng tne R¢portJng r-------· -----''-'----L--.,_"""' __ "'_ ${l,.ldies in Eptgemiolqgy= ·

:Cphqrt_ -STROH E'--H:)hort

.,_ .: .

FIGURE 29-4 Common Reporting Guidelines

consoUdated.stanctar.ds @l Reporl:iqg· Trfals{for randomized ,.controlled trials) TransparentRHJ:YQrti'rrg of Eva I uatiCJns-witb

t onsolleate<;f'Ctiteria · '

29Jl 229

Citing

Research-reports must previde ·accurate reference' information for ever.y publica-: tio'n tbl!t iS; t(s?dto. supp,on ihe m£tthods:, aitd ota new sJudy.

• 30.1 Referring to the Scientific Literature Authors. of every new scientific paper need to explain how theiFnew r.esearch.fitswith previo!).s ·stugies:! The introduction ot .a milnuscript l,rsually provides 'the back- ground neces,Sary to understand the importance of the. new work .. The discussion sec- tion provides an extensive cotnp'arison of resLJlts o'f the new stu.dy to the results of previously published works. A typital article in the health sciences refers to ;;tbout 20 or 30 other -articles published in reviewed althpugh cite only a .handfu:Land some (especially review articles} may cite hundreds.

Pertinent articles cart be found by -s.earching electronic databases-and by looking at the reference: lists nf articles akeaay identified and-'determined to he helpful; since the$e sources ate likely to also he relatc:d. (See Chapter 3 Jor a revi,ew of how to find relevant articles.) References-should selected to sup·port the itnpt)rtance, validity, and condtisions ofthe study. cart abo be used to.ack11owledge the alternative methodological approaches that could have been "Qsed, to identify both 'ar- eas in which the new findings' agree with the existing literature and a-teas where the findings, contradict studies, and to provide .varying persp·et:tives on the pol- icy and practice of the, study.

231

Citing an article is a way of endorsing the work of its authors (except in the rare instances when specific flaws need to be pointed out). So it is important to read the full text of every cited article and make sure that the methods and conclusions are sound. (Reading the full article carefully is even more important when criticizing the work.} Do not trust abstracts to be reliable. Abstracts may incorrectly or incompletely summarize the methods and results of a study. For example, they may leave out crit- ical information, like a very small sample size, a very low participation rate, or the use of a data set that is many decades old. Or they may report only the statistics that ate most: shocking or the most congruent with previous studies. Additionally, abstracts often state conclusions that the study's data do not support. Before citing any article, read and understand the full article.

Journal articles are the preferred source of evidentiary support for scientific ar- ticles, although books, book chapters, and formal reports (such as those published by governmental agencies and international organiz-ations) are also acceptable. FIG- URE 30-1 sQmmarizes the characteristics of formal reports, like those typically found in peer-reviewed journals. Fact sheets, websites, and other materials that have not been published in a formal online or print venue by a trusted organization should be cited only when a more reliable and permanent source of information is not available (FIGURE 30-2 ).

Formal Sdentific Reports ... Are published in a peer-reviewed journal (or sometimes a peer-reviewed report or book), not on a website, in a newspaper, or in a popular magazine. Describe the study design and why it was appropriate for the objectjv.es of the study. Explaifl hbW the study population was selected and demonstrate that the sample size was sufficiently large; Explain how exposures and outcomes were defined and assessed. Describe the analytic approaches-used and present results using easily interpreted tables and graphs. Draw conth:Jsions that are reasonable and based on the study's data. Discuss the_ llrilitations.of the study. Com·pa'r'e the new study to previous studies.

Follow a standard outline and other conventions for scientific writing. . FIGURE 30-1 Characteristics of Formal Scientific Reports

232 CHAPTER 30 Citing

.;.':

Ne.wspaper or 'popular - . magpzine:

' ,.

Yes.

Official repo11 · ·•·

formal

FIGURE 30-2 C:itablesources

Remarks ue

Jt.llstartlog·· p.laces for tnforma'J bi:l:t;'sftbu(t:I at¢d in a· formi;il

ttbttt ? trusted ,apt! liP, fOJTI1()1 a_rticle qr

repm;tprovktesJhe same lhformation .. shoui:d 6e re,.

terredto.on,ly;when ···n,o- . Jifk. or f.epqtt the

S:arne . . :r· :··. - ... . -·· fCjfestatlsticat·databases ,and re'poits• .only the· 'oata; n.ow it wa-s.·collec.ted, and:wheh. -i! ··· · · · ···

dted only when ·lhey p.u(bncanons ·{wfttr as'

y.ears:.anct/br'other ,'QjofJograt?hit rntotrhaJic;Jn).from :ttll.:sfe&'

·- ·

· '9fe sourcesJor forrnal

general tetbriical

to c:if¢; .·. · -

Jie;'sure.

Jpu.tr:m Is :ate fpr- mal maffi:Jscrtpt'S.

II 30.2 Writing in One's Own Words. Few scientific articles quote-directly from ·another source word for word.Thereare many reasons tp avold q:upting ftotn anotbet p:ublieatioh. 0I).e. of the tnqst in1port(!i11t is borrowing phrases and .sentences from other writers can make the writing in a. docu- m ent choppy: Some p-eople wh9 11se ql!ote& do so beqal.J.se they feehhat'the original works were so perfectly ·written that they could not say- the sa·me thin& equally well using

'J0.2 in One's Own Words 2"33

different words. This is not true. Saying the "same thing" in one's own writing style usu- ally is better for the new work use it means that the entire article has the same voice.

Another benefit of paraphrasing is that it helps ensure that the article being cited has been understood. Paraphrasing accurately requires a level of comprehension that direct quoting does not. Using a quote that is not fully understood is never a good idea.

Paraphrasing does not remove the requirement to cite an original source; it just means that quo,tation marks do not have to be used. When a direct quote is lifted from a pa- per and reu:sed, the entire quote must be in quotation marks (or indented from the left margin, depending on the length of the quote and journal formatting preferences). Additionally, an in-text citation must be provided. When the ideas or findings of other scholars are paraphrased, quotation marks are not used (because the words are not

Quotation Reference (Almost Never Used in Journal Paraphrase (Always Requrred tor Either a Articles) .(Often Used) -- Quotation or a Paraphrase) A case-control study exam in- A case-control study of 1. Risch HAMarrett LD, Jain ing risk factors for ovarian Canadian women found no as- M, Howe GR: Differences in cancer in Canadian women sociation between ovarian can- risk factors for epithelial found that Rage at first full- cer and the ages of participants ian cancer by histologic type: terni pregnancy was not asso- at the time of theirfirst full-term results of a ;case-control study.

risk of ovarian . ' . 1 .Am J Epiaemiol pregnancies. cancer.''1 1 g96; 72.

The authors acknowledged The authors of the study · 2. Zang El\ Wynder EL. that usince we did not adjust pointed out that it was possible Differences in lung cancer risk for depth of inhalation and that they might have underesti- between men and women: age at smoking onset, the RR mated the magnitude of the in- examination of the evidence. for women, compared with creased risk of lung cancer in J Nat/ Cancer Jnst that for men, due to .smoking female smokers compared to 1996;88: 183-92. was likely to haye been.un- male becquse they derestimated by our results."2 had not statistically adjiJsted for

·smoking behaviors, such as the depth of inhalation}

The investigators noted that The investigators concluded 3. Sutton R(;. An outbreak of Hcholera is usually considered that the most likely cause of the cholera in Australia due to to be a w(lter-borne disease, cholera outbreak was food food served in fHght on an in- but, in this outbreak, the avail- served to passengers on atr- ternational aircraft. J Hyg able evidence that a (London) 19.74;72:441-51. food item served as part of a meal was the most likely vehicle of infection.H3 . .

FIGURE 30-3 Examples of Quoting and Paraphrasing

234 CHAPTER 30 Citing

being copied), but an in-text citation for the' source of the original information must still be pro.vided. FIGURE 30-3 illustrates the difference between a quotation and a paraphrase.

• 30.3 What Is Common Knowledge? Anyspecific knowledge,. such as a statistic or the results of a particular field or labo - nitory study, mu,st be.cited when it is referred to in a scientific paper. However, some areas of general knowledge, or common knowledgel do not require a citation. Common knowledge refers to what a typical pers,o_n in the discipline would know; it does notre:- fer to what a randomly selected persbn at the grocery store would know . .For example, it would be common knowledge that influenza is caused by a virus and that Germany is located in Europe. Both of these facts ·<lte well established, and 'a quick search fot pa- pers on influenza or about studies conducted in Germany would show that this infor- mation is not usually accom,panied by ·a, citation. lrt contrast, a cotnment about the results of a particular epidemiological study of flu ·in, Germany or a statistic. about the pr<J'po_rtion, of Germans·; a#ecte'd by flu in a typicaJ year is specifk know ledge; it would need to be cited. When in doubt about whether a. bit of information is common knowl- edge, err on the side of a citati'on, Also, .any disputed fact should be well sup- ported by one or more reliable sources.

• 30.4 Avoiding Plagiarism P.lagiari?m occurs when w·ording, thinking, or <:rea'tive output is repeated in a· new document without attribution. Copying the exact wor ds- of another person without using q\mt;;ttjon marks ·arid providing a full citation, paraphrasing a ·u.riique theory or observation without. providing a citation, and using an image:withoutper- mission and an ·acknowledgment are q.ll forms of plagiarism. Failing to acl<nowledge the source of the original work deprives the author' or creator of the material the recog- nition that the. person and it may r:eslJlt in the plagiarist getting. cre:,dit for work that he or 'She did not do. . . · .

Piagiarismis a major violation of scholarly integrity, and it can have a damaging lo·ng-terni impact on a professional C<freer. For ;example, a publishech1ttide with ex- tensive plagiarism must be retracted, which requires the publi2 acknowledgment of guilt and results in a permanent .open record of wrongdoing. (The consequences of plagiarism and other forms of rese.arch misconduct, as redundant publication or the fabrication or of data:, a:re. discussed in detail on the website of the Com·mittee on Publication Ethics.) For students, plagiarism can result in expu.lsion from schonl.

30.4 Avoiding Plagiarism 235

Several habits can be adopted to ensure that plagiarism does not, occur. One help- ful practice is never to cut and paste information from a website, article, or any other source into a document file that contains any draft material for an article. It is far too easy for those words, phrases, or even whole sentences to be unintentionally incorpo- rated into the text of a manuscript. When browsing websites for background material, take the time to paraphrase the information rather than cutting and pasting the con- tent for later review. .

Another good habit is always to include a reference in research notes about any ob- servation that will later require a citation. For example, if an article presents a theory that explains the findings of the new project, do not just make a note about the theory. Jot it down and put a bracket with the author and year next to it, as in a journal manuscript. Also write out the full bibliographic information for the article so that the source of the theory can be easily identified later on when writing is under way.

• 30.5 Citation Styles Citations typically appear in two formats:

• As in-text citations where the sources of information are briefly identified in the text.

• In a reference list at the= end of the document that provides full bibliographic in- formation for each source.

No one citation style is used in the health sciences. Most medical and public health journals use some version of a style alternately called ICMJE (International Committee of Medical Journal Editors) style or Vancouver style, or they use the very similar NLM (National Library of Medicine) style or AMA (American Medical Association) style. Alternately, a journal may adopt the APA (American Psychological Association) style or some other style. (APA style is commonly used for social science journals as well as for nursing journals.) Howeyer, although some journals strictly adhere to one widely used reference style, many have their own customized styles. Reference manuals and style guides are available for all of the widely used styles, and most journals provide an author's guide on their websites that spells out the journal's own style. Articles re- cently published in the target journal provide additional examples of the journal's pre- ferred style. Whether preparing a manuscript for publication or writing a less formal report, the goal should be to use a consistent citation and reference style throughout the document.

In-text citations are abbreviated bits of information-about the cited work that allow the full reference to be located ill the reference list at the end of the article. Examples of formats for in-text citations are shown in FIGURE 30-4. (Some journals wiTI convert brack- eted citations to superscript numbers during the editing and layout process. The author guidelines must be carefully examined to see which submission style is preferred.)

236 CHAPTER 30 Citing

..

·tot .. first last name publitatiun

. :year 9rtd

pubffcaHon year

Numl'>er·inbrackets · bracketS).

'· •i

Ntmiber in parehthe- ses,(round fur:ackets}

.... 2004).

:•' ·

. . . flj.

..... Dl

... .(1).

.... (1)

FIGURE 30-4 In-Text Citati.bn Styles

. .. (R!,Ji:z; 2004; Yamt:u:nuto, 2001). . . . . •..•. (RUiz &8artthez1 2004; Yamamoto et aL1 2001J .·.

... ,lt, 2] .

. .. ·.[\2]

... f1,2).

.... {1,?) 1,2

.: , '(lv'Cl(IOV, 2l'J08; Ru.iz;2004; Yamamoto( 2001 ) •

. ... (Ivanov, Ruiz a San.dlez,,ioo4i Yamamoto etaJ.,:2CJ0'1 ... tT-3]: .... Jl-3] ... (1-3);

.... :(1-3) 1-3.

The, retere,nce. hst at end pfthe article presents works either alph<ttbeticaJJ.y by the .first author's last name orin the. order of firs.t appearance of the cited work in the text i:rrticle. Wher,i pteparinga manu,script fot submission to a journal, ¢4ec.k the do<wment :carefully for compliance-with the journai?s house style. Journals that useJCMJE style or a typically Hst autht;>ts by .name, a'no first irtit:ials,then the title (with eapitalletter.s only for. proper nouns),. an abhr.eviated fournal name; the publication year, yolume,:and page numbers! The components, may bt :Se}rar:at:ed by pe- riods (full stops·) :or by-se-micolons ,or·'commas. Howeve.t;,the journals may make minor ,adjustJ;nents to these components. Sorrte jov,rnals 'e4p,ed ail authors ttl be listed no matter how many there an:,;-Bome journals use an a bhreviated version f()r six or more

such as three authors, tQllowed hy: "et aL') Some use 'ab- breviations for journa1s;; others use the full journal na.m<=t .. Some list issue number-s;, ni,ost do hot .. Sotn,e list the fufl page numhe.rs (such as 202-209), other:s· Us,e a slightly shorter version (such as 202-9). Some use italics. or bold type for some :parts ofthe bibliographic entry, The key is-'to, be consistent, no ma.tterwhkh style is adopted.

Writing Strategies

This chapter provides tips for moving through the writing process successfully.

• 31.1 The Writing Process By the time a researcher is ready to write a final report about a project, the vast ma- jority of the work on the project has been completed: A study question has been iden- tified and refined, a study approach has been selected and a protocol developed, and data have been collected and analyzed. The end of the project is in sight, but the prospect of creating a report that is intended to be disseminated beyond those imme- .· diately involved in the project can be intimidating. Putting off the writing process is easy. The writing can drag on, and in some cases it is never completed.

Few writers have the ability to sit down and crank out a complete manuscript in one burst of productivity. Most writers experience cycles of high motivation and pro- ductivity and then of absolutely no interest in their work. So most to find strategies to get the first words em paper and then to see the manuscript.through to com- pletion. FIGURE 31-1 illustrates a typical writer's productivity levels during the writing process. The durations of each stage will vary among writers and for different papers, but writers usually need to address their motivation at three key times:

239

. ' -· .-. ... . I I I

I I • I I "

I

- :roelays in : Initial

"·" I I I I I I I

l getting 1 burst of , ·: ":;>tarted : praqqctivity :

Waxin_g and wa:fling

1 Exnillistion sets in 1:-FJrja:l -burst ! 1 - L . !

·:'stalls _

FIGURE 3i -1 Typical Variations irr Productivity during the Writing Process

• First, writers· must overcome the barriers to getting started. • Second, writers must find ways to prolong the period of high productivity that of-

ten occurs at the start of a writing proj:ect. • Finally, most writers become fatigued during the writing process apd at some point

lose all desire even to think about their projects. At such points, i:hey must find the motivation to persevere and to complete the manuscript.

• 31.2 Getting Started Scientific papers follow a standard outline (see Chapter 29). There is no need to think creatively about the structure of a paper. By the time a researcher has defined a study question, designed a study; collected data, and analyzed data, there should be sufficient information to answer the key study question and explain the findings. At that point, the only way to get started on a writing project is to start writing.

If a researcher does not know how to begin writing, an easy way to start filling pages is to:

• Put a working title for the paper at the beginning of the file, along with the names of all the coauthors.

·• Add in the headers for the Abstract, Introduction, Methods, Re.sults, Discussion, Acknowledgments,, and References.

• Fill in the names of the people to thank in the acknowledgments section. • Add page numbers. _ • Paste in a table or figure that was created during the analysis process and will be

included in the. final report. • Paste in some relevant lines about methods from_ the protocol.

240 CHAPTER 31 Writing Strategies

Then start filling in the gaps. Per-haps find a model article and use it as" a template to create a detailed outrine that specifies exactly what paragraph in the paper will cover. For example, the headers for paragraphs. on .statistical methods and ethi- cal consideri:!tio11s can be at the end of the methods section. A brief list of what to cover In each of those paragraphs can then be added bas.ed on what was re- ported in the model a,rtide .. (Be, careful nqt to plagiarite any ideas .or phrases from the model article.) Then write a sentence .or two for each of those key points"! a sentence about informed ·consent, a sentence about_ ethics committee review, _a s_entence about the significance level used for statistical tests.

The content of the rpanusctipt does not need to he added in any particular order. Many authors of scientific papers 'find it easiest to start with the methuds, then to write the resJ.Jlts, then the introduction and dist:u:Ssion, and finally the. but that order is not required. Many authors skip around in the paper, adding a few sen-

aJ a time here and there. Soine authors tind it 'helphtl to write throughout research process. They may draft the introduction ns sonn as the study question has. been identified, the metho,ds once the .study is designed, and the results after data hgye been collected and Then they draft the discussion section and edit the ear- lier d.r:atts of the manuscript to ensure that the paper tells a fa,cused story. When get- 'ting started, a good plan is to first write-wh-atever part of the paper is ready t<J be put into words. Then just keep on writing.

• 31.3 Staying Motivated Most writers expedence time.s when they have no desire at all to write. A number of steps can help a writer to regain motivation. Sometimes changing habits or scenery helps-writing in a new pla<;:e or at a n:ew time of d;ty' or rej;noving distractions ftom the writing area. Sometimes setting a time <line for small _parts of the pa- per is helpful. A titne_lin,e,m,akes it ro tasks. being Build in re- wards that .celebrate those. intermediate successes on the way to.a, completed paper.

Many man l:t'scri pt$ for health science j ourrra,ls are limited tq a, maxim uin ..of abou 3000 words. So writing only 10CY words .a day-j_ust a few sentences-will lead to a completed: first draft o.f a manuscript in less than two months. Revising the paper hi'tly take month or two of daily writing (see Chapter 32 ). Time lines and goals .can be. for that ptocesso p_S well.

Setting up a weekly meeting with an advisor or a writing __ group may a1so help with stayin,g on tratk. If the mental block is truly about the content Q'f the,manl).script, talk- ing through the" project with a friend can be help'ful. Another approach is to speak the c.ontent. aloqd or write 'it very informally ,or in sentence fragments . .The ideas 'that emerge from these exercises can be formalized-tater. Alternatively, ,move on to a ferent sedio.n while waiting for inspiration about the trickier_ parts' of the paper;

3.1.3 Staying 241

• 31.4 Conquering Writer's Block Sometimes a case of block gets serious and can last for weeks or months. FIG URE 31-2lists various types of writer's block. In each case, a negative thought cycle develops that has to be broken. Acknowledging the underlying issues-fear of being judged and fear of failure are common-is one step toward getting back to writing. But writers also need to initiate new behaviors to facilitate success.

One is to set asjde something like, 30 ,minutes at the same time each day for Then stay at a desk with a writing tablet or a word pro'cessor for the full 30 minutes with no distractions: no videos, no music, no computer games, no e-mail. Although it may take several days or even weeks for this dedicated writing time tore- sult in meaningful output, writer's block can be overcome only when an aspiring writer makes writing time a priority. Additionally, those with writer's block benefit from con- sultations. with senior coauthors, a writing support group, friends, and others who can help motivate them.

The 'completed manuscript will not be perfect .. No paper is perfect. By the time a report is written, there are likely to be a lot of imperfections that cannot be fixed. For example, the methods used to select participants cannot be changed. Participants can- not be a question that was not included in the questionnaire. These flaws are normal and expected. What authors can do is:

• Fully explain the actual methods used • Runall the appropriate analyses • Include a helpful set of references that support the results • Polish the prose · • Honestly identify the limitations of the study and explain what was done to ad-

dress them

242 CHAPTER 31 Writing Strategies

:

Reason to Ay0id'Writihg "This studyJs .boring ana unimportc:mt •

M0ststudies,-make only minor contributions fo tnovihga fie.JdJ(;)rwarq,t!ufth¢-o.hJywaytoJnake ;,my cqntribution is tq p:u,DHsh. ·

'111[5 research project son:lfo} Every;study flaws1;out few oHs flaws: The premise w.ru; bad, the not allow major mistakes to be. rnade. , study design was poor, and the data _ colledjon didnlg;o.

"r don't kiiowliowtowrite a a·rty

Tm stuck on this one se-Ction; and 1-can!t wmk·onanythingelse:untif tfinishthis part" · ·

"I don't ·know w.hatto do nexU

''I:don'twant to dlsappufnt/be by my protessor/supe!Vis.o'r(

Tm nota gdoo writer:'' "1'111 n9'fgo9d .at writing in "I don't think this project was :esti ng enougfJ to pub.ilsh :anything about it:" ' .

it.ls te- jetteo, 1 will .oe

Pff t11lspgp{3(;1splHJli5ht!9, someope '·njigpt a flaW '11:1 It, :and that

· wo.ukl ;h€r:embarrassl11g\" . ,;:1 have time to wrrtet

way to learn write_.is _bywritrrfg, A writing. 5t:Jpp.ott grQUR, ·ang/or advisors h'elp With tflis:p·rotess; Wtiiting .and, rewtiti'1gthe same section nver again or .sp.end- ing weeks searching th·e lilerature for more SUJ)pOrting ey)- deQ<;e ooth The besfway to:move fQ;Ward.is to IAforls, on .an:othet ot tne Pi=liler'or tq asR a or a'dvlsorfbr assistancEt

-.. , , -- .. ' . . .. .- . ' _.- . -

C::oauthets;and/or advisors will be happy to offer advite aboat hoW"to mcwe .. forward. 5uperv1sorswant·a paperto-be,asgood as itcan o..e, and

to make ;;IbQuf I revi" siqiis iftli¢y cpl!:uthQfs,: A wrftitlg5lip)J.brt g(o\lp'aricjjot

encotJragtn& ;andhone·stfrientl c.n1dtor colleague can reviewl:lrafts be.fore the manuscrfpt is shared with·a s.u- pervisor. ProcrastJnatlon-WIII of!ly in<:rcease,an)(ie.ty·abo,ut bein'g, · · · ,

i;l:dvis.9t?,.aqg help polish the but,only after it has b'een drafted. · check with· a coauthor anGI/or advisor aboufhow to, appro- priately disseminate the findings,

a The otlly who will of a manu>ctiP!. p,re·those· \<\fhorrrtl:le authors tefl it Also1 · win know that manyipatJers·are-sul!>mittecHo severafjour- · nals·before they ,ar;eaccepted for publication. Ptonastination will only delay-the start of tt1e revtew proc<:ess a no ·the pos.Si-

Gofi!Jtho(:s, editbrs witl .flOtlet a!') .. dbviously, 1tawed or badly wiitten to publication,·

Almost,evecyone oatttirid 15 or3:0 Do 'nOt as .writfri.g:

FIGURE 31-2 FOrms bf Writer's Block; 4

3.1.4 C.onquering Writer's Block 243

.·

Critically Revising,

:CJn:ce a , 'complete has 1ieen it ·tre.eds t.a &e revised and pol- Coauthprs'4t(d other aolleugues may help wii/:7 this ptoaess:; but the lead is responsible for 'oheckittg th-e tnanus;bripct very catefu.lly. Thre.e .check-

li:sts q,re ·prJ),vif4e.4 tq: fi;J,i;ilitat.e th¢ · ·

• 32.1 Does the Paper Have a i'Piot"? Every paper should tell a "'story" that has:

• A beginning-the introduction: sets the stage. :• A middle-the methods and results say what happened. • An end-the discussi.on provides a cohdusion that ties .all the parts of'the story

the story line ·shotJ.ld be able to he in a sentence or tW.o,. Indeed, s·brne journals require. a ·preeis that is 35 words or less·. The ahstra'Ct for the report should iell Jlie: who!¢ $'to!y in one compelling And the; wh.iole tn.(;lJiuSctipt must also work to :convey a cuhesiv:e ·message (FIGURE 32-1 ). The first step in editing is to make pic:t:ure is be.ipg

245

Does the paper have a clear "'story line*? Can the "'plot* be summarized in one sentence?

Does the title of the paper reflect the key aspects ofthe study? ., '

Ooes the abstract;telrthe key parts of the story? ., Do the opening paragraphs draw the reader into the story?

- Is the goal of the study clearly stated in the introduction section?

Does the methocts make' it clear howthe_methods WE:re:heJP.ful in answering the study question? Do the results and discussion sections provide the answer to the study question?

Is the story missing any parts that need to be added so that it is complete and compelling? Do any in logic need to be addressed?

Are a·ny parts of the manuscript redundant? Are any parts peripheral to the main story? Can these be-removed to tighten the storyline? Are the conclusions fully supported by the data?

Qoes each paragraph have a theme?

FIGURE 32-1 Does the Paper Tell a Compelling NStory"?

• 32.2 Structure and Content Once the pieces of the paper's story are clear, the next step is to check the structure and content of the manuscript (FIGURE 32-2) .. The paper should be well organized, com- plete but concise, and accurate about what was done and what was found .. The text, the tables and figures, and. the reference list must all meet these same requirements.

ls·the paper well organized? Is' thE) content focus:ed?

Does the introduction provide all essential background information? (For exampl.e, are the person, place, and time details listed somewhere in both the abstract and the text?) ·Does the introduction make the. research proje·ct}Ippear necessary an9 importa·nt?

· ihtrod'·udion say why:thes.tudy is novel?

Are the methods described in adequate detail? '

.Is enough statistical analysis presented? Is each .statistic included in the paper nece?sary?

Are the tables aqd fi,gures weltdesigned? FIGURE 32-2 Check1ist for the Structure and Content of the Paper

246 CHAPTER 32 Critically Revising

.·

- . . .• . ·. . . . ooes the qi5(1,f$slpn !)rdViete a- c:oncise of finding$ and t.he- findings fn·the tontext.of prevlou> research? discussion sectron avoidsimply .relterafihg the results· section:? ·

Js every· in tne d.i.S{:l,JssLdli s.e.ctjon suppqrtect oy Should Q:ade,d td IS-eY _Is listed important and neee'ssary.? Cart any.enttles,. in the HasJhe pap-er been dout:ile·che.cked to:ensurethat: no part of iUs plagiarized or paraphrased withoufproper attribution? Is_ part -oftlie.J)artet trutllfiii?"(For example; pqper report methogs tnat were .gdJJ- ai!Y -th9ti an· or them? ooes,it re)iort the r,¢S!Jits:J>qhe most ap(tt¢lpri-

tests .r;3ther than res,ults frqmle?s appropriate'teSts'th?J happened to produce-·'· statisticallY significant resultS?)

FIGURE 32-2 (continued)

• 32.3 Style and Clarity In a final check,, look at e:ach word; sentence, paragraph, and·s.ection, examining style and dadty ( FIGURE 32-3) : . ,

• Words must be used carefully. • Sentence-s must be concise and clear. • The voice must be consistent. • The grammar and spelling must be. ptopet

Are words usea precisely?:(For ex:ample, .'cUe termsJike "',associated, .. u.correlated/ .anO: "'caused'" appropri<:itecytAre anct

Ar.e--au abbteviations,.introdutted at first - - ·· ··-·- . _.,,_, . ·. . ·' - -'.;; . ' • · . . Is the tone·.ot the writing-appr::opr.iate?' ls the wri'ting,styte fact based;ratherthan emotion based?: Does the·artide consistently-use -a lhird•personvQice.or; in rare :situations; consisfemtl;y use -a petson ("'"or "we").voi(.:e? Is' the voice cornKt

FIGURE 32-3 for Style qnd Ctarity

32.3 Style and Clarity 247

DoaH sgbietts (nouns) vetb,s'? since "data" is a plural word, is "data a_re ... • used rathet Rdata is .. :?) Are f]JI oth.er grammatical followed? Are active verbs rather than passive verbs used whenever possible? Is the verb tense consistent? {For most papers the past tense is used rather than the present be-

collected..in the past.) Is eae::hsent.ence clear? Are concise as Are all words spefled' correctly:? report should 'consistently follow the sp'ellfng·tonventiohs of one cotmtry.) Is all punctuation correct? (For e)qlmple, are there extra or missing commas?) Does the paper adhere to the specifiq1tions of the target journal (if the manuscript-is being pre- pared fqr ?Ubmission to a journal)?

FIGURE 32-3 (continuep)

.248 CHAPTER 32 Critically Revising_

Posters ·and Presentation-s

Research results are ·often publicly shared for_the,first ti-me' during an oraL pre- s_enta#on,/Jt dit an a,e:aci,emic;-or

• 33.1 Purpose of Conferences The-primary outcome of most professional and academic conferences is:·networking: ·me'eting new people working in the ofin_terest, cq,tc'b,ing n,p with oLd 'ftien.ds, and making,and nurturing professional connections that may he helpfuL in the future. Conferences are q pla_ce to ideas; tO learn about what others irtthe field axe doing, to learn new methods and techniques in·a discipline,and to share--current work with receivedeedback on it. Pr.esenting new in the, fortn of a poster or an oral presentation can be a particularly useful way t·d get feedback on a pro.ject before submihing the work for review uy au journal. Sharing findings is· a ,way to whatothers >find most intere:sting about the project and to identify the weaker aspects. of the stJJ,dy and the questions that need to he in a formal written

•• 33.2 Structure of Conferences Some, ,conferences are annual events sponsored by professional organizations that :draw th9usand:s of attendees .. Others are. small gathedngs, of a few dozen sc;:holars, working in a narrow field of study; Most conferences include a mix ,of:

249

• Plenary sessions where keynote addresses are given • Business meetings run by the officers of the sponsoring organization • Concurrent sessions in which multiple panels of oral presentations are held at the

same time in different rooms • Poster sessions in which attendees can mingle while reviewing research posters

Presenters are usually assigned to deliver either an oral presentation or a poster presentation. Oral presentations require speaking in front of a potentially large audi- ence and may involve facing an open question-and-answer period in which the work can be discussed. This interaction can be helpful in improving the work prior to pub- lication-so much so that some presenters are disappointed when no one in the audi- ence raises a concern about their work. However, the process can be incredibly stressful. Oral presentations are generally considered to be more prestigious than posters, in part because there are usually more slots for poster presenters than for oral presenters.

Poster sessions are usually held in a less formal venue. Posters may be taped along the walls of a room or displayed on long rows of easels, and attendees can browse through them at their own pace and interact with presenters if they want more infor- mation about a project. Having one-on-one or small group conversations about one's work can be very helpful. However, posters often require more preparation time than oral presentations, and they may also be expensive to print and a hassle to transport. (They are, however, nice to display in the hallway of an academic department or work- place for several months after the conference.)

• 33.3 Submitting an Abstract Researchers interested in presenting at a conference are usually required to submit an abstract for consideration by the organizing committee. The conference organizing committee and other reviewers:

• Rate the submitted abstracts. • Decide which researchers will be invited to present. ·· • Select who will give an oral presentation as part of a panel and who will be assigned

to a poster session.

Abstracts selected for a conference are usually printed in a conference bulletin for easy reference by attendees, to help them decide which sessions to attend and which posters to seek out.

A good abstract includes key words and conveys one clear health message that is appropriate to the audience expected at the conference. If the conference focuses on clinical practice, the abstract's applied message should be readily translatable into im- proved patient care. If the conference focuses on research theories and methods, the abstract should emphasize the novelty of the approaches used and their applicability

250 CHAPTER 33 Posters and Presentations

to other research topics. If the. conference focuses on health policy, the abstract should have a clear p.olicy implication.

Since the submission deadline is often many months before the conference, some- what prelimiuq.ry resl}lts inay need to. be presented in the abstract. However, the ob- jectives 'and methods must be clear,_·and, very importantly,, final results must be ready ,to share by the conference ..

Applicants may be asked about their preferred presentation formats. Those who indicate a to d.o either an oral presentation or a poster mew th(:dike- lihood that·their abstracts will be accepted for the conference·.

Submitting_ an abs.tratt tor consideration infers a commitment to attend the .con- ference if selected to be a presenter. The fine-print instructions for the conference of- ten specify the of applicants. For example, the sponsoring organization may keep track of dropouts artd absentees -and not allow thetn 'to present at future conferences. Most conferences require .presenters to pay a registration fee· (often s.ev- eral .hurtdted dollars) as well as covet all of their own travel expenses, .although some s·chools and employers may reimburse some or all of these. expense-s.

• 33.4 Preparing a Poster Conference attendees are drawn to visually appealing posters. Sn when preparing a po·ster,o give equal attention to and to 1t$: design ( FIGURE 33-1 ). should be

J\eep,Jhe content focuseo on .Qn!e, core lnessage. Ch_oose a· descriptive title. Include author names; brief author affiliations, and contact information foF at least one authar. bo,not-li-st in'formation "aboutthe conference (suGI1 as name, dates, or focatiori) on the poster. Consi(!:er skipping the qbStract to space.

Content Clearly state mai,n goal, the specJfic objectives or hyp.olhese.s, and the bl the study. Use a .. sfructw:ed format. introduction/bacRground, methods, results, and conC:Iuslon/c;Jiscusslon sections a r<eference list, in small f6nt if previous stui;ljes-are Cife<;l). Be concise; Us.e ana.'bulfe:ted lists when possible. Images like graphs1 tables(, flowcharts, photographs, .an<;! maps are more effhtivethan words at conveying.information.

FIGURF Suggestions for Post er Content, Layout, and Formatting

33,4 Prepgring a 25;1

•1'. the size. and Shape (horjzontal or vertical) of'ttre. display are13 thatthe conference organizer:S will provide and create a poster to fill the

. space . Decide whether to print one .large poster (preferred) or smaller panels that can be joined together at the conference venue. Organize content into three or four columns or anottier structure with a logi<::al flow.

,·

. \-.:· -c_dlor, and/or.lines,;to group the informiJtioi:l . Leave. adequate white space, and keep the color palette simple.

Layout arid formatting Ensure adequate contrast between the background (usually light) and content (usually dark). lt·is often helpful to use borders for photographs and -other image content. Use and consistent fonts that can easily be rea.d several steps back

... from the poster. Simpfffygraphs and make sure that they can be read from a distance (which· may require adding a litle and directly labeling lines or bars

. .

rather than using a key). Use high-resolution images (and remember that enlarged photographs become fuzzy).

FIGURE 33-1 (continued)

well organized and have focused content, an eye-pleasing balance between text and im- ages and white space, and an inviting color palette.

Posters can be created using either specialized graphic design software or a simple presentation software program like Power Point. The size of a slide can be adjusted so that the dimensions match those required by the conference. A sample layout is shown in FIGURE 33-2, and the Internet has many examples of other poster·formats. Be sure to have several people check both the content and the design of the poster before it is printed. Also, inquire about: .

• Printing costs (which will vary significantly depending on the size of the poster, the amount of color, the type of paper or fabric, and any special options like lam- inating or mounting)

• The amount of time required for printing • Whether a special carrying case is needed

252 CHAPTER 33 Posters and Presentations

FIGURE 33-2 sample Pbster Layout

• 33.5 Presenting a Poster At most the poster presenter is responsible for setting up the poster at

assigped Although sume cqnfetertce· organizer.s provide all the· necessary sup- :Plies,, presenters usually find out what is required only after·they-arrive at the confer- ence site'. Pos.t.et presenters shqu1d therdore tome prepared with <;lips (for clipping a poster to a board set on an eas:el), pushpins (for pinning,a paster to a corkboard), a.nd tape (for ta.pjng poster to a Th¢ presenter is also down the poster at an appojnted time. It is considered had form to take down a poster eqrly otto leave it vp after the assigned time, 'when someone else rnay b;e waitin;g to set up a paster for the next session.

S'cime .ses:Sion times: when :ate ·expt:¢ted to stand by their posters and intera:ctwith attendees. 'These sessions are .an opportunity for one,-on-one conversaticH:lfi with interes.ted It is apptopria.te to each person who stops to view the poster, Thus;the presenter-should respond toques- tions: without beco.ming so engaged with one that all others ·with questions or comments are ignored. Some presenters prepare a handout that7is either a page-sized printout of the full poster 6r sheet with highlights .. Mos,t pr,esenteJS: have. business cards with contact information available for distribution.

33.5 Presenting a PQster" 253

• 33.6 Preparing for an Oral Presentation A typical oral presentation time slot is about 15 minutes long. Because a minute or two is required for setup at the beginning a nd questions at the end, about 10 to 12 min- utes of this time slot are available for the actual presentation. Most presenters at health science conferences prepare a set of computerized slides (typically using PowerPoint) that will guide their talks and provide information to the audience. Since· most present- ers cart cover ·1 or 2 slides a minute, about 12 to 20 slides are appropriate for a 10- to 12-minute talk (FIGURE 33-3). The slides should not attempt to reproduce a paper on the screen; they should highlight the key message of the presentation using images in place of words as often as is appropriate. FIGURE 33-4 provides a checklist for the con- tent and layout of slides for a presentation slide show.

Number Content Area of Slides Note

Title slide 1 Include the title, names of authors, and contact information for the presenter.

Research goal/importance 1 Start with the key message.,

Outline or summary ·1 Include citations (in small font) for any

' Background/specific aims 1-2 previous work mentioned and for images taken from other sources.

Methods 2-4

Figures and tables of statistical results usually Results 4-8 need to be very simple to be readable during

a presentation.

Strengths/limitations 1

Conclusions 1 End with the key message:

Acknowledgments and/or 0-1 invitation for questions

Total 12-20

FIGURE 33-3 Sample Distribution of Slides for a 10- to '12-Minute Talk

254 CHAPTER 33 Posters Presentations

;.fhe COntent bf each slide is·aGcurate. ; .. · . . -

·. Graphs, and _othertmages .i:lre irr of wor<;ls as:'Gfte_n, as !

are _instead of.fUII sentertt:e$.· ' • ' .

·There are no:'more than about six fines per slide; All words are spelled. correctly,

All bulleted phrases on one.slide use a coAs.istent {foJ example, all start vyith the· WQrd "fo"' 6r all \iYlth·an "-tn(· \v.bfd). .

qppropriate-forthe scheduled presentation duration {about 1-2 sl·ides·per minute, time set aside;for questions). - ··

. .

Every sHdeJs relevant. . • . ' . ..

Layout -and forni:CJtting. . . The background i:lild .rtotQistra.{ting; · and the text {eifher-darkJe:ttei:soi:r a (fght background or tigiltlefters;OA a dark. b.acJ<gr()l)Mct). The'.tor:i-tra$t is under

are on ·or qffJ. A eoi1sfste.rit cindfeaclahle font is thro£tghol)t. irrarge::rc:>htstie·ls.llsed throughout, induding ta;,bles and figures(which may· require .sirnpHfying them ana: enlarging the:font used for-various componentS); •. ' . . .

A :and pleasapt <;:olar is:Hsed·lhroughout..;

Al1 :tables.·:and are .Unnecessary:effects iikes0unds( animated components, and distracting. slide transitions ave avoi:ded.

FIGURE 33-4 Checklist for Presentation Slide Show

3_3.6 Preparing for an Qral Presentation 255

Preparing the slide show is only the first step in preparing to make ·an oral presen- tation. FIGURE 33-5 provides a list of content-, voiGe-, and performance-related items to practice extensively in the weeks before a presentation. Practice in front of people who will provide honest feedback. Consider video-recording a practice performance, reviewing it, and identifying areas for improvement. No one can plan for everything that might be encountered at the conference, including nerves, but practice makes a positive experience more likely.

Content Opening lines Practice the exact opening sentences; these need to capture the attention of the audience.

Message Master the content of each slide enough to describe each one without referring to notes,

Phrasing Use· relati'vely.short, ·precise sentences with actiVe verbs.

A ow Practice transitions from one slide to the next. Closing lines Practice exact closing sentences about key

conclusions. Voice Pace Speak at a moderate to slow rate.

"

Volutne ' ' .·. Sp,eak loudly.

Pitch Vary your voice inflection. ' Enunciation Speak clearly;-avoid words and phrases that are

difficult to pronounce, if possible. Pronunciation Check on the pronunciation of technical words and

names. Fillers Try to avoid, fillers (such as *urn, .. . ah, ... . like . ..

you know;;). Performcjhce Engagement Smile and in'ake eye contact with members of the

audience.

Posture Stand tall or sit straight.

Delivery Do not just read the slides or read from a script. Movement Try not to fidget, sway, pace; or m!3!5e other

distratting'gestures or movemehts. Technology Become comfort&ble with advancing slides (using a

mouse, keyboard, and/ or clicker) and with using a pointer, if applicable; face the audleoce when using these tools, if possible.

FIGURE 33-5 Things to Practice Before the Presentation

256 CHAPTER 33 Posters and Presentations

A few week} before the confirm what equipment will prov1d¢d in thtt presentation room {such as a computer and an LCD projector).

•· expect presenter's to provLde their own laptop computers. • Som·e conferene·es require presenters to upload their presentation files to a website

in advance of the conference, • Some ask presenters to e.;mail -their files to the session moderator. • S,orne expect presenters to have the file on CD, and others prefer files

to be on a flash drive .

No matter what forma tis preferred, always bring a backup copy 0fthe presentation file. -

··• 33.7 Giving an Oral Presentation FIGURE 33-6 the key orr the p(;ly ofthe pre·seilta.tion_. Ct>nferehce. or- .-ganizers often advise presenters to:

• Arrive at the presentation room at lea;St,15 minutes: before the panel begins: {not 15' minutes before an individual presentation-rime) . -

• Check .in. with the moderato_r. •- Set up the computer .and projectot or confirm that slides ar:e ready to be projected.

Presenters are also often reminded to be considerate of other presenters In their ses- by $trictly adhering to their time limits. ·

At most conferences; time is allotted for questions from the. either after eqch pres,entation or _after the pa,nelists in se,ssiort Tf;;t is not.availab.le for thoseasking questions,. the respondent should repeat the question before _answering The usually to:·

• Keep responses short. • Thank those who offer s_uggestions for improving _the work, • Acknowledg_e thelimitations :of a "study yet highlight its· Str'engths:. • Be respect_ful to :everyone.

At the -epd of the sessiop.,_ one-on -on:e or small group conversation the te.search may :continue. Presenters should have busi11ess cards available to-givelothosewho have o·verlapping interests. - ·

:33.7 Givingan Oral Presentation 257

.i• . 7, Time !asks 15 minutes Moderator Check in with the session moderator or chair, if is one. before the Q&A Ask the moderator whether the question-and-answer time will assigned take place after each presenter or after all the presenters are presentation finished. panel is scheduled to Time <:onfirm the amo.unt of time for the presentation, and ask the

· begin moqerator whether there is a .and, what sort of warning signs will b.e given when the q)lOtte<Ftitne is nearly finished. If there is no timekeeper, ask a friendly person in the front row to serve as one.

Computer If using a computer and/or projector, check to be sure that the devices are set up and that the presentation is loaded on the computer and ready to use.

Pointer If using a pointer and/or c)icker, check to be "Sure that they are working.

Microphone If using a microphone,.check to be sure that it works. Water Check to be sure that drinking water is available. Copresenters Greet other presenters in the session.

During other Listen Pay attention to the other talks; do not focus on personal notes presenters' or preparation duringthis time. talks in the Connect Listen for points of connection between the research talks session being especially if the question-and-answer period

comes at the end of the session. During the talk Relax Trust that practice wfll·result in a proficient presentation.

Control Be alert to nervous behaviors, such as adding ,fillers to speech or swaying the body.

Keep time Do not exceed the allotted time period. After the; taJk Thanks Thank the moderator}, timekeeper, technology support person,

and/or fellow presenters. Belongings Check to be sure thatpersonal items are nor-forgotten. Conversations Wait in the room for at least a few minutes in case anyone has

follow-up questions; move the discussion into the hallway as soon as the for the next session b:egin setting up their talks.

FIGURE 33-6 Checklist of Tasks on the Day of the Presentation

258 CHAPTER 33 Posters and, Presentations

Seleding Target Journals

-eonductedliealtb:r:eseanth p.roj- edjs oftert t_h_'e dissem,inatiqn.' t>'(r¢s-tlttK thto4.gh .(Jn f11!t:b/icat'ton.

• 34.1 Choosing a Target Journal

.....

Researchers who want to publish their fl'ndings; must identify one or more journals that could reasonably be expected to disseminate their reports·. Selecting a mrget jour- nal :early in the writing _pr.oc:ess makesJt :easier to hone' the paper's message for the journal's audie·nce.An of recent artides in the journal _pt:oiides about:

•· The best to follow • .How tO' divide commentary between the introduction and discussion sections: •· What snbsettions to the rti¢th'Qds $ection • The appropriate voice and writing style . , The .amow1t·of to include: • The reference and citation style

2S9

Choosing a target journal entails many considerations, including:

• The aim and scope of the journal • Its audience • Its impact factor and other characteristics • The possible costs of publication • Online access options

• 34.2 Aim, Scope, and Audience The most important considerations when considering potential target journals are the fit of the research topic with the aims, scope, and audience of the journal. Some jour- nals are very broad in focus, while others are very narrow and publish in only one subspecialty area. Some are international journals that publish research from around the world. Others have a very specific local or regional focus and publish articles per- tinent only to that geographic area.

Determining whether an article is a fit with a specialty or regional journal is often straightforward. As an example, a journal focused on liver disease in Argentina will not be interested in a paper about osteoporosis in Mongolia, but it will review a man- uscript on cirrhosis in Buenos Aires. A journal focused on nutrition in Southeast Asia would not review a manuscript on vision disorders in Sweden,. but it would consider a paper on iodine deficiency in Cambodia.

Knowing what topics fall within the scope of a general journal is a little harder. Some prestigious general journals will publish only articles expected to have a significant and nearly immediate impact on clinical practice. Some general journals in medicine, public health, nursing, and other health science fields will consider articles on just about any topic that is remotely related to the aims of the journaL

Considering the primary audience for a manuscript is also important. For ex- ample, if the article's message is targeted to people working in a focused geographic area, a journal sponsored by a regional professional society that provides a copy of each issue to all members of the organization might be the best venue. Publishing in such a journal will ensure that the paper reaches those who will most benefit from it. On the other hand, if the study has conclusions that are relevant to an international au- dience, then a journal known to have a global readership might be more appropriate. However, the expansion of the Internet is making regional and ihternational journals less distinct. Libraries and researchers nearly anywhere in the world are able to ac- quire copies of even relatively obscure publications.

One way to identify journals likely to consider a paper for publication is to exam- ine the manuscript's reference list. The jo urnals cited most often in the manuscript are likely to be suitable target journals. Abstract databases and library holdings may also provide a sense of which journals are likely to be interested.

260 CHAPTER 34 Selecting Target Journals

• 34.3 Impact Factors The target journal should not be selected ptim:lrily becaU:se of its impact f;1ctor, rank- ing, or reputation, even though these are all factors to consider. The impact factor .is base'd on the number of times a typical article in a. is cited in its first yeat or two after pu.blication. A few of the most prominent journals (like Nature, ]AMA, The Lancet, and the New England]ournalofMedicine) have an impact factor of 10 or greater, but most journals in the health sciences have -an impact factor closer to 1 or 2 , Specialty journals may have an impact (actor less:than 1, but they can still be important within the specialty area. Impact factors are often listed on Journal websites, and resources sl1:ch as the Web of Knowledge (art electronic resource often available through university libraries) compile ratings for many journals.

B 34.4 Journal Characteristics After identifying potential journals, lookat the journal requirements. For a review ar- tide., make sure that the target journal will accept reviews. Some journals specifically solicit short reports, which ·may ·allow authors a maximum of 1000 or 1500 words, one table ,or figure, and a limited numher of references" These condensed manuscripts are an appealing option ·for a case report; a small cas:e series, or an update to a p:revi-

published article. Alternatively, a comprehensive report of a large study that will exceed theusua] 3000- or 3500-word limit or the standard limit of four tables anc;l/or figures will require a journal that has more flex ible word limits.

S,o·mej ournals provide inforn:ra tion a bout their turnaround iime (the ·average time from submission to first decision) and/or their rates. Many big- name jour- nals With low acceptance rates have a tt1rnatound time of only a tew days or weeks because they send very few manuscripts out for external review. 'Specialty journals with higher rates may have ;1 turnaro11rtd time of.several months beGause three or more external referees review nearly every manuscript;

Another consideration is the method ;o£ submission. Most journals. have moved 'to online submiss-ion systems, These allow authors to upload manuscriptsto a website and track th¢:progress of their articles through the journals ask authors to.e-mail.a copy of the paper to the editor; and some still require several copies of the pa::- per to be sent by postal mail. often prefer online systems because, of the, ability to monitor the':status of their manuscripts, but this is not a priority for some researchers ..

·• 34.5 Publication Costs Aliho'ugh many journals are able to cover costs through subscriptions, advertising, and/or the' support of a professional spdety, an ins;:reasing number of them are resorting to a variety of mechanisms that comp-e:l authors to cover some of the costs of publishing,

34.5 Publiqition Costs .261

• A few journals require authors to pay a small submission fee and will not review an article until this payment is received.

• Some charge a small or large publication fee. The fee may be per article, usually called a processing fee or processing charge. Or the fee can be per ar:ticle page, usually called a page fee or page charge. The number of pages is determined by the final ready-to-be-published article, not by the number of pages in the submitted manuscript.

• Some journals that are rqn by professional societies require the corresponding au- thor of an accepted paper to become a member of the sponsoring society. In this situation, publication requires payment of a membership fee.

• Some journals require an open access fee, which allows the journal to make the article available online immediately upon publication w ith no restrictions.

• Some journals give authors the choice of whether they want to pay for open ac- cess. Researchers sponsored by funding agencies that require articles written with their support to be publicly available may thus pay for open access. Authors with- out tunding may publish at no cost.

A few journals that charge fees may allow authors to request waivers of some fees if the authors are from low-income countries and/or if the project was not supported by a contract or grant. These requests usually must be made before the paper is reviewed. Publication fees are usually disclosed in a journal's author guidelines or somewhere else on the journal's website. Look for this information. when considering publishing options.

• 34.6 Online Journals Some journals publish only print versions of their articles. However, the vast major- ity of print publications also offer online access to subscribers (usually libraries), even for articles that are not publicly availa ble through an open access option. Articles pub- lished in these journals are assigned to an issue and given page numbers, but they are also available to subscribers as electronic files.

Some recently founded journals are available only online. Although many of these journals are likely to remain available online for many years to come, some researchers are wary about publishing in new, unproven journals that do not leave a paper trail. A subset of these new online open access journals have a reputation for not having rigorous review standards and being a sort of pay-to-publish scheme. In contrast, some open access online journals have quickly becom.e well-respect ed journals that are re- garded as having strong peer-review systems.

Before submitting to an online-only journal, be sure that the journal is legitimate and is indexed in r elevant databases. For example, being indexed in a competitive database like MEDLINE, which examines t he quality and editorial rigor of all candi- date journals prior to accepting them for inclusion in the database, is validation of the journal's legitimacy. -

262 CHAPTER 34 Selecting Target Journals

,•

The Submission, Review, and Publication Process

Manuscripts submitted to,peer-reviewed journals are evaluated by editors and external reviewers, who provide feedback about how to improve a manuscript and make a decision about whether it is ready to be published.

• 35.1 From Paper to Publication Many brilliant and artfully written articles are published every year. And, every year, a lot of not-so-brilliant articles are published. A manuscript has a high likelihood of eventually being published if it is written in decent English (or written well in some other language)o, if the methods were reasonably rigorous and valid, and if the findings have a clear application or message.

Publication is a priority for many health researchers because, from the perspective of the broader scientific community, a project that has not been published is a project that never happened. Submitting to a journal as soon as a revised and polished man- uscript has been crafted is critical. Procrastination can render the study useless be- cause data in the health sciences quickly become obsolete and no longer publishable. (See Chapter 31 for writing strategies.) Submission is not the end of the writing process. Additional revisions will likely be required, even if first journal to which a manu- script is submitted accepts the paper. This is another incentive to submit as soon as is reasonably possible: revising a manuscript is never easier than when the project is fresh in mind.

.263

• 35.2 Journal Selection Once all coauthors are satisfied that the manuscript is ready to be submitted for peer review, one journal must be selected as the first journal for submission. Chapter 34 has suggestions for selecting an appropriate journal. A preliminary target journal may have been selected early in the research or writing process to serve as a guide. However, once a manuscript is completed, a variety of journals should again be considered. Only one can be selected as the first place to submit the completed manuscript.

Submitting to two or more journals at the same time is not permitted in the health sciences. Although editors of some popular magazines may compete for manuscripts from paid freelance authors, nearly all of the labor in the professional health journal system is voluntary. Editors may receive little or no compensation for their time, and reviewers and authors are unpaid volunteers. It would be a major strain on the edito- rial and review system if every manuscript was sent to several journals at the same time. Thus, most journals require a statement with each submitted manuscript affirm- ing that the manuscript is under consideration only by that one journal. This rule should be assumed to be true for all journals. Once a manuscript has been submitted to a journal, it cannot be submitted elsewhere until either the authors are notified that it has been rejected or the authors formally withdraw it from consideration. The web- site of the Committee on Publication Ethics, whose membership includes the editors of several thousand biomedical journals, provides additional information about ap- propriate conduct for authors and the repercussions for those who violate standard protocols.

• 35.3 Manuscript Formatting Each journal provides author guidelines that state how manuscripts should be for- matted. The guidelines must be carefully followed. See FIGURE 35-1 for examples of formatting preferences, which vary by journal.

Special attention should be paid to tables, figures, and other images when format- ting the manuscript. The tables in the manuscript do not need to match the typographic style of the journal. Most journals will reformat the tables of all accepted manuscripts into their house styles when they convert the text into the single-spaced, small font, multicolumn format that is popular in health science journals.

However, graphs,. maps, and other illustrations are rarely reworked by a journal's graphic designer prior to publication. So all figures should be polished prior to sub- mission. Journals may require image files in a specific electronic format, which may or may not be a standard file type. Most journals charge a fee for printing color im- ages but not for grayscale images. So use color only when it is absolutely necessary. (Alternatively, some journals charge for color in the print version but allow the online

264 CHAPTER 35 The Submission, Review, and Publication Process

Conterit Should only the title listed on the title. page? Or should authors, word counts, key words, running h,eaders (abbreviated versions of the title}, or other information also be listed?

Author information Should identifying information be blacked out in the manuscript (possibly including the citation information for references to previous works by the research team)? Or shouJd author names be listed? Should authors' degrees, job titles, lnstitutlonal affiliations, and/or information be li$te.d?

Abstract/summary Should the abstract be structured (showing subheadings for each section, such as "objective ... methods ... results ... conclusion·) or unstructured? If a structured abstract is expected, are there preferred subheadings? What is the word limit for the abstract? Should the abstract appear on its own page? Is an additional one-sentence summary or precis required? Are add-itional separate statements required, such as bwhat this paper will contribute to the literature'?

Keywords How many keywords (search terms that wilfbe 'linked to the article) should be provided? Must these be MeSH (medical subject heading) terms? Should these belisted somewhere in the manuscript?

Sections Is there a preference for how sections within the document are labeled and formatted?

Ac;kn·owleqgments/endmatter . Should acknowledgments of funding sources· or personal assistance be included at the end of the, manuscript? Is any additional endmatter to be suth as information about the role of each coauthor, details about ethics committee review, declarations of potential conflicts of interest, or other disclosures?

In-text citation style How should in-text references to works listed in the reference section be shown: as superscript numbers, as numbers within brackets, by the lqSt name of the first author and the publication year in brackets or parentheses, by the riatnes of several authors along with the publication year in brackets or parentheses, or by some other method? (Chapter 30 shows . examples of these methods.} '

Reference Jist order Should the entries in the reference list be in alphabetical order (by the first author's last name) or in of first appearance in the manuscript?

FIGURE 35-1 Manuscript Formatting Requirements Addressed by Journals;Author Guidelines

35.3 Manuscript Formatting 265

" .·

formatting

Fonts.and fontsizes

wotg Jimi15/page limitS

Tabl-es. anciffigures

FIGURE 35-1 (c;_ontinued)

format\ingJ:loes thej.qu,rnal ret)Uire f(:jr the reJetente list? 'Par :example, h0w'many authors,shouM be listed for a(tides wifh morethan 6:coauthors? Shcmld thefuiljoumal tiUe .. ble 1isjedior an ab.breviatlon·for Should the issue and voll!m¢.be:listeo, ·or jUsfthe Snoqlo "of the p<ltrtS .oJ ttrerefe-rencE! t5'e itallcs? Wl:latmarglns and lin'e spacing should oeu.s.ed? Dothe lines on-

Should page,numbers be-sh0wn at the bottbm center of eaEh pa;ge,,the.top ofeC!ch m· elseWhere? whaUoniS and font slzesshould -be used?

-; ,.

Wh2rt;.·islhe W6rq limit o(pcl!ge limit? Does the word li.tnit include·pnly main texrof the .a:rtitle; or does it,also the:abstract; ·references, and tables·? · · · · · · is the. number oftahles and/or figures limited?Should tables and 'figures appear ·-in the manuscript following the-paragraph ln. whic.h ar€ first mentioned, or should they .all be placed at the :eild ·qf the<fua.nushlpt file the referen.ces? .$b01,Jld

as q t.ile, .or b¢ left'p!q€ed. at-the .enu ofth'e tile burtigures savt:!d separcite files? -

version of the manuscript to use color .at no cost. In this situation, authors ·may sub- mit a color version 'of the image but must ma:ke sure that vets:ion has ap- propriate tones and adequate Also, .. since an image. may he ·resized prior to publication, checktha.r each ima.g¢ can be;: enlarged or red'!lceg without_ distortion.

• 35.4 Cover Letter Even though most submissions (lre made via computer in.stead of by postal delivery, most online :submission systems still expect a coverletter to be uploaded. FIGURE 35-2

of a cover letter. The should summarize the manuscript and seek to convince the editor thatthe work is impprta)1t, valid;, original,. and <l good fit with th.e ;aims of the journal. Once ·submitted, the editorial decision about <Whethe·rto consider the article for publication may be m'ade solely on the of abstract and/ ox covet leiter. So both of these items must be compelling.

2.66 CHAPTER 35 The Sut;>misston, Re:liiew, an.c! PUblic;;ltion Proc:ess

Basieinf6rmatiol'l

:Summary.

_ lmportanee

Fit

"

AdGiress th-e letter to the editor(s) ·eilh-er,generically VDear·e.dito(') or byname. ,

Provide the title otthe manuscript anq, .if the jmJrnal publlsnes diffeninte::ategortes of articles, the type'ot article (such as original rese.aKh, or shor-t repoM).

Provide a short summaFy ofthe· study design. and the cqse ·forwhy tllg mam,1stript is Important ?ignifiEant, _

and original.

Make-th.e'q;tse for why the is a good fit for tt:te Journal.

Some affirm thaf-the mtmusi:fipt is n<;Jt r_evieWel'sewhere· ah,d has 1/ot pUblished,, that all Jfsted coauthors meet authqrshlp criteria including the. approval of submitting the manuscript to the jourmil, and/or that no-'conflicts of interest Aeed .to be disclosed to the editot:s; S.ome.journals may aaditionaHy ioft:mllation about

of each':copuinor a no/or .of proje:ct

manuscript fOr ppsslble rev.iew and publjcatfon.

9f an a,uthbi's to appear on the tover A''Signetlletter can be scanned into a tompnt'et and up!Gaded on thejournal's ·submission website, orit ·can be-faxed to the journal off!ee.

FIGURE 35-2 Sample Cover Letter Content

• 35.5 Online Submission 'Once the manuscript files have been prepared and all the required supplemental infor-

have: the is reaqy to be subtrtitt.ed. The authnrs may need to send·paper.copies by postal mail, sometimes along with a computer disk containing the .files. Alternatively, g journal might rej,uice authors. to e-·maH the manuscript. and cover letter to. the I ournal. Most journals, however, :require online .submissiqn.

Creating an account with a journal's, ·submission website usually takes only a few 'i.Uillutes. Only the cprrespanding Czf:lthpr- the coauthor who Will commuh:icate with the journal and answer questions from readers after the paper is published--needs to register. The co.trespondirtg a.utho.r may be the first thor, the coau,thor, or the coauthor with the moststab1e address and affiliation. In addition to facilitat- ing submis$ion of the. the online a,ccount enables the

355 online su]?tniss.Jon 267

author to track the manuscript's progress through the review process. Most online systems will indicate when the editorial office is cbnsidering an article, when the arti- cle is undergoing external review, and when a decision is pending. Online submission usually takes about half an hour, although it may be faster or slower depenqing on the amount of information requested and the number of steps in uploading.

Most submission websites start by asking for basic information about the article, such as the title, abstract, and keywords. The keywords may be able to be typed or pasted in, or they might be selected from a list provided by the journal. Some journals will also ask for:

• The type of article (such as original research, review article, or letter) • The word count • The number of tables • The number of figures • Statements about ethics approval, funding, possible conflicts of interest, and au-

thor contributions • Confirmation that the article is being submitted to only one journal

A second step asks for information about all contributing authors. The correspond- ing author should check ahead of time w ith coauthors about the preferred forms of their names. Most a uthors in the health sciences choose to use a middle initial w hen publis hing, since PubMed and several other abstract databases list authors by their last names and first and middle initials. Some journ_als also request a job title, affilia- tion, and contact information for all authors.

There may be additional steps. For example, the journal may request the names and contact information for three or more potential reviewers and/or a list of people who should not be reviewers because of a known conflict of interest. Some journals require a list o f potential reviewers before a submission will be processed; some make this information optional.

The final step is uploading the manuscript ffies.

• Some journals require the title page to be uploaded separately from the rest of the manuscript, especially if they use double- blind review, in which reviewers are not told the names of authors and authors are not told the names of reviewers. (Some use single-blind review; reviewers are provided with the authors' names but authors are not provided with reviewers' names. Others use an open review process.)

• Many journals require each table and figure to appear in a separ;ue file. The file types accepta ble for figures vary among journals.

• The journal may request additional files, such as a publishing agreement signed by all authors or a checklist showing compliance with required contents and/or formats .

. All of the manuscript files are typically combined into one pdf file during the sub-

mission process. The corresponding author should carefully review this file for com-

268 CHAPTER 35 The Submission,_ Review, and Publication Process

pa;ge numbering, and the of tables and figures prior to. finalizing the submission. The author may ·also have an bppdrtunity to. review -<1I1 html version of the uploaded paper and/or check the references for accuracy. (Some systems ·auto- rnatically link 1nanuscript references to ahsttatt dat.ahases so can ily access the abstracts of the cited articles. Incorrect references may be flagged as having errors.) Onee the trtanuscrjpt ·supporting file:s are tonfinrted to be· cotrect, the suhmission is complete.

• 35.6 Initial Review Once a manuscript is submitted, the journal's editorial staff' does a preliminary review

;deCides. whether to sen,d the manuscript tO ex.terna;l peer reviewers, or tO ]:-ejeCt it withaut review (FIGURE 35-3 ). Although the organizational structures of journals vary, typically the editor-in-thief who overse.es. the journal assigns, submissions t6 assistant editors for initial review. For manuscripts deemed ·worthy .of review, the as-

editors-identify ctd hoc. reviewers .. These-are who not on the iour- nal's editorial 'board who are asked to serv€ as peer reviewers use ·of their expertise on the paper1S topic or metho.ds: Sotne j.ournals send nearly all ·manu,scripts out to re- viewers; others.select:nnly a small 'fraction of them for peer review:

Rejection without review (sometimes called a desk tt}je.ction) is· often not a com- mentary on the quality of the manuscript. It is rather a decision based on thepetceived fit of the papet with thejournal's:current interests, If an article is rejected without re- view; the authors ·should identify a different journal be a better _fit, make

· Accept

FIGURE 35-3 The Journal Review Pto!;ess

35.6 Revlew 269

any edits deemed necessary, reformat the manuscript for the new journal, and submit there.

One of the advantages of the initial review process is that it allows authors to quickly submit their work to a more suitable journal. Authors are often notified of a decision to reject without review within two weeks of submission, although in some situations notification may take three months or longer. When an article is selected for external review, notification usually takes at least two or three months, if not longer. Authors should usually not contact editorial offices to inquire about the status of their manuscript until at least four months after submission. Even then, ·a request for an update should be made only if the status of the paper has not recently been updated in the online submission management system.

• 35.7 External Review Results Decision letters sent after peer review are almost always accompanied by comments provided by one to four reviewers. Reviewers usually provide two sets of comments to the journal.

• The first set of comments is on the quality of the manuscript. These observations are intended to be shared with the authors and often include specific points that the authors should address to strengthen their manuscript.

• The second set of comments is intended only for the editor. Reviewers may be asked to rate the manuscript's novelty, importance, and fit with the. journal in ad- dition to the quality of the work.

An external peer review can lead to three possible results: rejection, an opportu- nity to revise and resubmit, or acceptance (Figure 35-3). An article determined to be methodologically sound and well written may receive low scores in the areas of inter- est or relevance to the journal. So it is possible for a manuscript to be rejected even if all the comments shared with authors are very positive. Alternatively, an article deemed to be somewhat lacking in writing quality may receive high scores for the originality of the topic and the apparent significance of the work, and may yield an invitation to revise and resubmit. Often reviews are mixed, with one or more reviewers being very critical and one or more being quite positive. Mixed reviews may lead the editor to de- cide either to reject the article or to offer the opportunity to revise and resubmit.

• 35.8 Rejection Some manuscripts are rejected because they are poorly written, incomplete, or of lim- ited interest to those not directly involved in the project. However, many rejected man- uscripts are well written, thorough,. and interesting to a wide audience. Many journals

270 CHAPTER 35 The Submission, Review, and Publication Process

have low acceptance rates and reject high-quality papers. An appeal to the €ditorto reconsider a rejected manuscript will almost never result in a different out- come .. So ·authors should simply put their energy into revising the rilq,nqsctiptfor sub- .missionto another journal. . "

Rejection does not rn_ean that the article has been rejected by all journals and will never be published anywhere. It simply means that one journal has decided that the paper is not s4itable for its a.uqie.ilte, Many authors ·findit helpful to take, a few days to be disappointed about the refection, to vent about some nf the reviewer comments, anq to complain about edito.ria1 dedsion making. But Qne rejection-or even several of them-does not mean that a paper will never be published. Each set oJ reviewer comments .. can strengthen a paper. Most papers are not fatally flawed, that is,,.so badly de:signed and conducted that they cannot be ·rescued. Most papers can be made' suit- able for publication somewhere, although gaining acceptance may require several weeks or several months of additional work As long as researchers are willing to learn from each set of reviewercornments, the manuscript will continue to become stronger with each submission.

JQegin work on revis1ons as so.on as -possible after receipt of a letter. As time elapses after the completion of data collection and rel}lernbering the original aims, methods'- and results becomes increasingly difficult. All reviewer com- ments should 'be read and carefully c_onsidered, with appropriate ed)t{; made. (The next section describes, how to interpret reviewer comments.) Never submit-to a second journal without taking advf{ntage of the inpgt provided by the first set of reviewets. For one thing, their feedback will improve the paper; For another, the manuscript may be sent to the sa·me reviewers, who will not he happy Hthe1r eva1uat1ons were rgnore,d. The Tevising process may require relativ:ely little time, or it may demand significant reworking of entire se.qion:s of 'th(;: fuantts'cript. The background an,d discusBion sec- tions: may need to. be expanded to include. more emphasis on the importance of the new pape:r and more, citations ofthe ·rdeva.nt literatute. The methods settion rnJ.y to provide ·more details about the techniq!JeS used. The results s.ection may need to §how .additional output.

Once the manuscript has been edited to the satisfaction of all coauthors, a new target journal should be selectt;d. The wtiting style .and. formatting of the paper may need to be. updated to reflect the style of the new'target journal prior to submission.'

• 35.9 Revision and Resubmission Whenp.uthors ate invited to revise and resubmit.(R&R) their manusctiptto the same j-ournal; they need to edit the ·manuscript and prepare a response to eac:h reviewer comment. Apthors may be given a deadline for resubmission. If they miss: the dead- line, the revised manuscript may be ,treated as a new submission and be sent out to· new reviewers1 which ma.y significantly decFease the likelihood of acceptance. Some, but

35.-9 Revision and Resubmfssion '271

not all, journals make a distinction between a minor revision and a major revision. A minor revision may be reviewed only by the assistant editor after resubmission, whereas a major revision may be sent back to the original reviewers for a second look. A jour- nal may allow only a very short time, perhaps three or four weeks, for a minor revi- sion to be returned. A major revision may be given a deadline of three months or longer.

If the original reviewers are asked tore-review the manuscript, they will be pro- vided with a copy of the authors'' responses to their comments. Accordingly, every re- sponse needs to be carefur1y constructed and respectful. Examples of responses to comments are shown in FIGURE 35-4 . Some reviewer suggestions-often marked as "minor"-will be easy to re.spond to, such as correcting typos, reformatting tables, or adding a few more citations. Others-often marked as "major"-may require more thought and time. Responding to comments that are complimentary or to points that the authors agree strengthen their papers is fairly easy.

Responding to critical comments is much more difficult. Authors who disagree with the suggestion of a reviewer are not obligated to change their paper to suit the reviewer, but they do need to write a thoughtful and conciliatory explanation of their point of view.

• Sometimes reviewer comments are hard to decipher or vague, such as "The entire manuscript is lacking foclJ.S and clarity." An appropriate response is to refer to ex- actly where and how the paper has been improved.

• Sometimes a reviewer's comments exhibit a lack of comprehension. Although it is tempting (and sometimes accurate) to assume that the review.er was reading care- lessly, the authors should consider how that part of the manuscript might be re- vised to promote clarity.

• Sometimes two reviewers offer conflicting advice. The responses to both of the comments should summarize both comments and explain how a decision was reached.

The response to reviewer comments should be prepared as a separate file from the manuscript. Additionally, some journals require a version of the manuscript that high- lights or tracks the changes made in the document between the first submission and the resubmission. Once the revision is complete, a new cover letter, a revised manu- script, and the responses to reviewer comments can be uploaded to the journal submission website. The cover letter should thank the editors for the opportunity to revise and resubmit, thank the reviewers for their comments and state their ad- vice has improved the paper, and affirm that each reviewer comment has been reviewed and responded to.

The time needed for second review varies widely among journals, ranging from a few weeks to several months, depending on how many parties are mvolved in the re-rev1ew.

272 CHAPTER 35 The Submission, Review, and Publication Process

The spedfk aims of this paper We have edited th:e.final paragraph of the introduction. shovld be .. dearly stateo. early iM the section· to .rtlaf<.e'it clear that the. three specific aims of the manuscfJpt. . p:aper.a·rt:f(ll to .... , (2) to ... , ahs1 (3} to . . ...

Did,yO:ursurvey inclu<Je a qJJestioh , about. ,,.?

The sample size seems tooJow·to have.:adequate power-lor this; study ctesfgn,

Table 3 .. seems incomplet-e; ltshould a I so 'results oft he ... test for eacH row.

Vol! useo the . ... test to analyzg ... , buta .. . appr,opriate.

sectionr the :ctothg:rs·'c;lairrL .. ,,but is it gossible,

,. .. , i.? llqppening l.nstef'ld,?

We this pc,1ragrapn:to improv,e ctarity and to. .. ; . ·

The we analyze,d 919 not incl4cte '!:variable for ... , even without that informatit<m, o-ur analysis· shows

that ....

This WOL(I<;J have beenafrelpfUI questioh'to ?sR, but, unfortonatelyt it wa,? not intlu;ded In O'Ur questionnaire.

"

·we did:not tflis:que:stiQn baseline survey preSeQted in this paper, but we do plan tcHlSka··question about·. , . in 0u:r study. nextvear. We agree that

. thiswill be an interesting question to explore; ·

We did askthJs .quest;nn an<ffound' .... We hqve :added- this'flndlng to the·results section.

we ¢id ask,this qpestJon":and fb.und .. : .. We used ... software. before inhiatingour study to estimate our required sample size. With eX!pederl inputs. o.f. ... and . p.ower of8.0'%, a size of ... was estimated .to be requrreq. In totarw¢ rectil1tect , .. . partitip&n:tS. Hase<J oJt.the results of our study, ;3ng our estrmates,ofpower duting-'(jata an·a·lys-is, which shOWed ; .. I our sample size 'is,estimatedto have sufficient power to yield results. We have done the' additional analysis requested and have added a new n :.ilumn to Table 3 that sh0ws the resuUs ,of th.e ... test, What we founo was_ . , . , whieh ·is cemsJstent witli the resutts of our (t)ther statisticC!I tests. The ... _ test that we used is the appropriate; ... , . The alteniate ... test is not apRrbpriijteJiecause .....

Our qsseitlon that . . .. is happening is on ..... This interpretation (s suppQrted by several J)ublitaficms, rncll):dlng ... . we have :pur rationare conclusion in the discussion ·s.ection and added additional referenees;to prevfousliterature: · · · ·

re¥ie\I'Jer rqise:s a ve'ty interestil}g p0int. We that. both ofthese:'interpretations are possible, and now discuss both perspeCtives· in the discussion

FIGURE 35-4 Sample Responses to Reviewer Comments

35,.9 Revision and Resubmission 2 73

Sample comment . Sample Response($) The conclusion about . . . is not We have removed this claim. Our primary conclusion, which supported by the data.

' is fully supported our results, is ....

You should include a discussion Thank you for raisfng this interesting point. We added of. ... commentary on ... to the discussion sectiOn.

we agree that this is an interesting topic, but since ... is - only tangentially related to our specific. Clihls, we do not have space to distt[ss it in this paper.

You need to add a paragraph on We have added a paragraph on limitationsto the discussion the limitations of the study. section.

Several recent publications have Thank you for bringing these articles to our attention. Both addressed the themes of your of the articles by .. and ... were helpful in supporting our work and should be cited, findings and are now included as references. including ... , ... , and .. ..

I am not convinced that the study-is we have added an additional paragraph t.o the introduction importanJ enough for publication in that highlights what is new and significant.about our an international journal. It may be findings. We have also added an additional paragraph to a better fit for a regional journal. the discussion section that discusses the Implications of our

findings for other settings. We believe that our paper is important because ....

There are typos in lines ... and ... Thank you for catching these typos. We have corrected both ... of them.

FIGURE 35·4 (continued)

• 35.10 After Acceptance Papers are rarely accepted for publication as is, especially for a first submission. Acceptances are often provisional acceptances with final acceptance pending until a few minor adjustments to the manuscript are submitted. If a provisional acceptance is offered, journals will often ask that the required updates be made within a short period of time, sometimes in as little as one week. After the journal receives the cor- rected manuscript, a final acceptance letter will be sent to the corresponding author, usually by e-mail.

Once a paper is formally accepted, it is usually sent to a copyeditor, who checks the paper carefully for grammar, spelling, and to the journal's style. (Some journals have a style manual for copyeditors that specifies the preferred phrases, terms, abbreviations, and spellings for articles published in that journal.) The paper is then sent to a layout specialist who formats the document to look like all the other articles published in the journal. The page proofs (or galley proofs) are then sent to the

274 CHAPTER 35 The Submission, Review, and Publication Process

:Corresponding author for review, usually as :a :pdf. file .. Authors are usually given only :one to thre.e to ·meticulously· check the :do<:J:t,tnenr, . to p.rty que6es. 'frpm the·editor, and' make . .arry·"other ·editing requests·. 'This is not the: time to rtTake any sub- stantive ·change.s; ;s-uggeste-d .correction·&: are· limited to new· p,rol:ilern-s;, like. formattinK :errors anti copyedits .. that:have the" meaning,&£ the text. read every 11ne .e:xamirte eyery darity <l:nd.'¢h<;<;k the. ·speUing.o£ names, the contact information pr:Ovided for the corresponding au- thor, :and the order ofreierences. This i,s. the last opporlunity to· oa:tch 'errors,.

After the authots return the the to puoiication de:- pend,&· on the fo.urmiL SQme: fournals wili a pdf file ·of the correet.ed page pt:oofs: on their we'bsites ;an 'advance,£ux;ess. ar:dde ot ptep·rirzt,, Othe"rs wifl rtot p't>st the ar- ticle online, untii it has be:en assigned to and p,uhhshe:d itt print form. An ar-

be in tnete; Week$ after (?r many months the pa;ge proofs: a(e approved. So.on after article is published., the·.abstradt will be,

:,added to. the' thqt'index. the The pvbli%he(t artide citeiJ f6t' tim.e in anO:thex article about a. year 0r soc after publica'tiorr. At this pornt,;the

re$.e,atch is, ·

Why Publish'?

:Rese{lrc'hers take the lime a1/iil 'f3:ffort to·:se,e a re?et:lrch the wt:zyto"pqb.,. licatiori: fot,·man"Y reasons. .· '

• 36.1 Scientific Dialogue The' peer-reviewed publicatiorr:system (and, to a lesser-extent" ·the professional confer-

network) is the primary way scientists communici;Ite. Submitting'a manuscript to a journal for review isthefust ·s'tep:in a series of conversat-ions a:bontthe,research proj- ect. The fir,st back-and-forth conversation occurs with editors and reviewers. After the ·article is published, the cohtinpes· l:JS other .tesearche.rg, read, discus;S; and apply the work. So, if the results. o't a research .study are ndt published, for all pntctital ·purpose's itis as if the was nev:et done: The findings do fi9t become· part of the conversation among scientists because there is·no formal record of the proj- :ect. Although may have learned from p:r<;>jectev:err If it is nevet -fpr- mally written up, .anunfinished report does not further scientific knowledge or improve . . practtce.

277

• 36.2 Critical Feedback The peer-review process is an important step toward producing a high-quality paper. Reviewers are usually quite adept at identifying weaknesses in an article and asking authors to carefully think through the problem areas and to respond to them. Responding to suggestions from editors and reviewers requires an author to:

• Understand and appreciate different perspectives • Balance conflicting sets of advice about what would strengthen a paper • Deal with the frustration of needing to rethink and rewrite whole portions of a

paper to make the intended meaning clear (since unclear writing is usually at least partly to blame for reviewers who completely misread a part of a manuscript)

• Recover from a harsh review and move forward

Subjecting a manuscript to criticism and possible rejection can be intimidating and unpleasant, but the peer-review process produces stronger manuscripts and better scientists.

• 36.3 Resped for Participants and Collaborators If participants donate their time to a project, then the researcher has an ethical obli- gation to make sure that their time was not wasted. One way to fulfill this responsi- bility and to show respect for their contributions is to share the results of the study widely so that others learn from it.

If a project finds a significant association between an exposure and an outcome, that finding should become part of the scientific literature. If a project finds no asso- ciation between an exposure and an outcome, the results may be even more important to publish so that other scientists do not waste their time and resources coming to the same conclusion. (Publishing a study with null results is often more challenging than publishing a study that finds an unexpected or strong association. However, a finding of no association is just as important as a finding of a statistically significant associa- tion as long as the study used valid methods.)

Seeing a project through to completion also shows respect for collaborators and mentors. Being a coauthor on an article or listed in the acknowledgments section is a reward or gesture of appreciation for everyone who assisted with the research process.

• 36.4 A Step Toward Future Research The research process does not necessarily end with a report. The research process is a cycle, in which data analysis and reporting naturally feed back into the formation of new study questions {FIGURE 36-1} and the establishment of a personal research trajec- tory. Publishing marks an import<fnt step in this cycle.

278 CHAPTER 36 Why Publish?

Design study

Collect data

Analyze data

FIGURE 36-1 Research Project B.q.ises New QUestions to Be Explored .in F!.!tyre Projects

Publishing on the topic twi<;:.e is Redundant_ or dupli- cate'publication is a violation of profe:s:sional standards,,. and may result' in the retrac-- tion of the articles. However, asp:et ts ofthe qa ta set that were not coveted in the first publication might be worth exploring •. The newly _published researcher should consider the rela,ted research th4t might pursuing. The-report p·rob- ably number of gaps in the literature that could be investigated. Furthermore, by qoqtributing to the research a published pa,pet allows other health sci- entistS- to continue the research process by examining new study questions, raisecl by the published.re.port.

• 36.5 Personal Benefits Publishing enhances the authors' re-sumes/CVs. It indicates that a person ispartof the. s¢hdlady cbmtrtlJ.nity, can s.ee :a proJect through to completiqn, and has the ability to handle constructive criticism .. A published article. becomes a p art of author's per- marterit rec(>rd; the paper will be indexed in ab stract databases for deca·des -and cer- tainly for the length of each author's career. And, although .authors ofscholarly J.;o1;1.rnal a;rtides: are not, paid for their writing (and; ip: gre often happy when they do not have to pay fees) , the payoffoften comes in terms of improved job oppor-

and prornQtions. publi$hirtg is, unlikely to brin;g a person fame and fortune, but/it does provide a tangible product after all the many hours that the au- thor -spent reading, planning, data., ru:nn'ing sta,tistical an·alys·es, and writ- ·ing. A published paper is evidence <::>f the expertise and commitment to fot individuals .a.nd

36.5 Personal Benefits :27g

Index

A ·abstracts

16--17 2.65

f:or. reading -232

16:9'? 274-27:5

rate,Z6l a\:knowledg_eme.r1ts,26, .?26., 16;5 .ad hoc· :269 ;advan4'e ac-c·ess, 2 7S advets;e event; 163', 166- 167' ;1ge aggregahi studies,44 ;alpha (txl, 1:21-tZ:t

hypothesis (Ha.l, 198,...19-9 AMA Artferlcan <3aficer SodetW 16 ANOVA '2'09 . ' . . ,,.._ .. ' . ' ' - ' iu'J:thtopometry> 147-143· AI> A :2.36

structure;, 223'-2:29 :ascerta:fnmt::nt 1?ias, '107 ass:ent\ 160

:ttttributabfe r-isk (Al\)1 7.4:..:.z3, pe.rce.nt(AR'?aJ,, 73

.a!ltho.r';guidelines,1 2'64-26·6 cauth.orship;ccrlte.ria, auihp.rship :order, autocoirelation; 2'14·

;:¢utonomy, 151\JSs, ISS

'!3 137

sec(j:on, :2;2,4

b_.ac.k_ward .!itep\'yil),e mydef,:21.3 bar chart) 193:-194 Fee!< In yentqry,.136 BehavibfaLRisk FactorSuivetllance·

()?IfFS,$)., i 7Q heha VJ:biS'; :.u

8.6 151-lS-3 ' . :;·: •, - ,, :- - •"'- .. ,_ .

beta (!3.), bias,

ascertainment bras, iQ:7 Hawthqrne effect, 83 ihfl;)rrtfatip:n bias·, 70, '8'3, 130 rnisclasSificatittn bias, 58;

bias, f67 fion-response 1Sias1 ws· pufthcaticm '1?13, recall bias, n r

b,..i,.,w, 107 bimodal distribbdorit 192 .Qji}qmi?l test,207

129_, 139 29?, 109

descrlp'fiv:e•st:atiStics; '194 logistic

biolqg,ical '149. biva.riat,c; t:$1?,::..210

¢orrelar;ion, 4.5:-46

rat.io (0R}, S-9-:...63, 203-1'04 :Z,.o'6'-2o7 ·

one,v.fay A:NOVA, 207, 209 pairecL ZJb' relative risk (R'R:lf73-76-,.204'

bJind . ,blinding; 8'3.'-'84, 8·9

84

operatQ'rs; i1 boxploc,_vn- 19.2 br(li:nst9np,ing,_ 9'-10 Breslow-Day te:st,,:213, 'bMdgets .99,; 'tdl ·

c. 'callout, '216 Ga.nadi:an Cmunall,ni+y He;tltb

. . (CCHS), i7b S'<;i .

·ca$¢ rlefiniti'Qn, 57, case 'fB:taHrs· rate, 52:

34-3.6., 49-52', 187-'188 ,case :Study ('dis.e report)l 49

.:analytic 59-6·3,:187-'19:9, -· ., . . . , 19:7-19'8.

\mtk other1ap., _34-c.. 36;

:l'bgisfk regression;-217...:21..8'

res, 6o2-63 S,T-:5:§ .

titlds ratiO.' ( 0 '203

:5:6- 57, 1(19:..111

i wdlihg checklist; 229

18:9 · statistics3 207. 20'S(

stat_isti<:Js, -.causalitY. 66, YS c;:'ent:ers.:for C&ntr0l &:"

P.rev..eiftiori (US. 1S, tid ' '

children, 114, 160 Chi-$quare (:X2 ) test, 207-208 CINAHL, 16 citing, 231-237 clinical examinations, 148-149 clinical records, 51, 171-172 dose-ended questions, 12 7 cluster sampling, 107 coauthorship, 25-29, 170, 171 Cochran's Q test, 209 Cochrane Cqllaboration, 177 codebaok, 181-183 coefficient of regression (p),

214- 218 coercion, 154 cohort, 66 cohort studies, 65-76

analytic plan, 70-76, 187-199, 197-198

comparison with other ap- ,pro;lches, 34-36

excess risk, 72-73 inciaence t:ate, 70-72 longitudinal cohort studies,

65- 70, 111-112 person-time, 71-72 prospective cohort studies, 65-68 relative risk (RR), 73- 76 ret(ospeCtive cohort studies.,

sample ·population, 69-70, 111- 112

sample size, 120-122 writing checklist, 229

collaboratio_n, 25-29, 98, 100, 195-196

commercial r.esearch tools, 136 on Publication Ethics,

23S,264 common knowledge, 235 community meetings, 159 Community-Based Participatory

Research (CBPR), 115 comparative statistics, 197-210,

2 11-220 by study approach, 197-198 confidence intervals, 202-203 confounding, 211-213 hypothesis testing, 198-201,

205-206 independent-samples t-test,

207-208 interaction, 212-213 linear regression, 214-217

282 Jnqex

logistic regression, 217- 218 9dds ratio (OR), 59-63, one-sample t-test; 206-207 one-way ANOVA, 207,209 paired data, 209- 210 p-value, 201- 202 regression, 213-219 relative risk (RR), 73-76, 204 survival analysis, 86, 219-220

compensation, 153-154 computer-assisted 143,

144 computer-based surveys, 135,

139-140 . concordant pairs, 63, 89 concurrent sessions, 250 conferences, 249-258 confidence interval, 202-203

and sample size, ilS-119, 2.02 for cohort studies; 73- 76,203 for matched

63 for unmatched case-control stud-

ies, 6 1-62,203 confidentia lity, 157-158, 186 conflict of interest (COl), 167 confounding, 126, 2'11-213 consensus methods, 92-93 COn,5_ent, 154-160 . CONSORT 22-9 cqnsultants, 25- 26. continuous variables, 190 control, 56, 81-83, 110 control definition, 11 0 convenience sampling, 107, 108 copyeditor, 274 C OREQ checklist, 229

45-46 correlational studies, 34:...:35, 4'3-4 7 corresponding 267-:268 cost-effectiveness analysis, 86, 93 <:ouncil for International

Organizations of Medical Science (CIOMS), 153

cover letter for manuscript resubmission, '272 for manuscript submission,

266-267 for surveys, 142.

Cox proportional hazards regres- sion, 2 19

C:ronbach's alpha, 46 crossover design, 83 cross-sectional surveys, 53-54

analytic plan, 54, 187- 199 comparison With. other ap-

proaches, 34-36 cross sectiortafseries, 69 prevalence, 54 sample pop_ulation, 54, 108-109 sample size, 1 20-122 writing 229

crude mortality rate, 52 cultural considerations, 114,

158-159 cumulative survival, '219 cutoff 89, 195, 204

D data cleaning, 184-185 data entry

computer-assisted surveys, 143-144

database: pn;g:rams, 183-184 184

paper-based questionnaires, 143-144

spreadsheet programs, 44-45, 184, 188

data extraction form for medical records, 51 , 1 72 for systematic reviews, 175-176

data man<t:gemeo:t, 181-186 181-183

computer-assisted s urveys, 143-144

data cleaning, 184-185 data recoding1 1 85-186 database programs, 183-184 maintaining confidentiality, 186 paper" based' questionnaires,

143-144 spreadsheet programs, 184

deception, 162 dec iles, 191 Delphi method, 92-93 dependent variables, 213 descriptive statistics, 187-196

averages, 190 bar <;harts, {93-195 boxplots, histograms; 192- 193 mean, 190 median, 190 mode, 190 quartiles, 19f range, 191 spread, 191-193

,standard deviation,.l92_:193 193-194 .

'·desk rejection, '269 diagnostic tests., 8S-89 di:chotbmou's variables, 129, 1?9

st,adstics, 20'7; 2.09 ·descriptive statistics, 194 lpgisqc, t.egr.e;;sion, 217 .. .::.219

direct age adjustment, 7 djscqmin\l::rrion, 8 _s.....,.$_6, 151-15 3 discordant pairs, :63 ;discrete vaiiabks, 190 discussioil ·section, 225

it-13,-44 191

.di'str:it?.utive 85, 153 dose':..response, 82-83

.·qouble .. translation, 1;3.6'--137

doupbentr.y, 184 .. .dropouts, 69). 70, 88

218-219 d up licate-p:ubl 23 5,,2. 79 duplicate submiss:ioh '235, 279 pynamic· .. population, -69

E ecologi€al 4 7 'ecologica-l 4)-47 :editor, 264, 266_, 269-:271

ll).oqific<u:ion, .111-:2 i 3. ,effect size,'176-177 · eff_ka<:y, 89--'S 8 eligibility criteria

Jor r,rimaw 10·8, 111 fot· system<J..tic reviews; 40:-4'1,,

1'02-103, 174 Embas-e, 16 ,ernie· perspective, 91 ·e,ndtna:tter, ?25-226 •envil:onmental 11_, 44,,

tso Epi Info, 184

.8$ equivalence 80-8 I .'(iJhics ethics training, 160

92 etk 9,2 e..xs;e.ss risk, hdusici..ri-Criteria

qse definitions\ 57 f<:lr experimenta1 studies,

113-114

exempt_ion from revieV'{, 167-1.68

exp(P).,. 218 expedited review;, 165 experimeqtalistpdies, .. 77-8:9

analytic plan, 86-88, 197::...198 '

blindir% ·'8'3-84; ·89 c<imparison with othet ap-

proar;;hes, 34-36 efficaty, 86-8 S

85-87 number needed .to treat (NNT),

' outcomes; 79-81 placebo, 114 rahdomizaJion,

controlled trials (RGTs),79

s.amp)e pop4J.:ttion, tl2-1 J:4 sample size, 12:0-t22.

e·xposure-disease-p.opuladoti and hrainstorrning;t'l-13 and i s:-.:.19 and appfoach, J6 and ·stucly goals, 22-..:.23 for corfelatiorial stuaies, 44

on;al '44 in questionnaire development;

11-l2l 44-45 99

review,.-269-2 74, 2 78 . 7.0. ,

tejecti'dn 270-471 revisiqn a.nd resupmissJof1

271-274

F' Ftest, 207 'fabrication, 195, 235 factqtial des-ign, 83

15, 2.:32--:2,33, falsifica'tion:; 195; 235 field notes, !}i' ..

226-2'2.7 first ·a\JthQr, 28 .. Fisher's exact test, '207-208 fixe,:Lelfe.c:ts mqdel, 176;_177 fixed po_pufa:tiori, 69 focus.grQ1ll?S, 9i fc'Jrest plot; 177· 'forward ste,pw-is,e nwflel"213

free-respQI}se qu:est_!ons, 127 Jrequeney matching, 5]-5,8

tesr, .209 .. fuil '.review, 165: fun,ding ,s_oun.:es, 99 funnel p'iot, 1 is:.

G galley proofs, 274-275 gaps in t he literature, 1 .18-19·, . 279 Gaussian distribution, i92 Geary's coeffibje-nt; 220 Qenerat Questionnaire;

(GHQ);l36· ·general kn9'Wledge, '235 ,geneialiiabilit-y, 2, S4; 16:4-165

least genetics; 2; 58? '15 8

ic lJ1forrna1ion.S 150, 22D

gl:tostauthqrship, 28 gift authorship; 27:-2;8

142', H4, 1,58 GIS, .150, 220 goafs,of researcl\, 1-4, 19, Goqgle Schalar,. 17 GPS (global pesitionillg re-

ceiver, 220. ·g'ran ts, '99'

phs,.22 6-227 ,grounded the0rr; 91 group rn.atching,. 57'""" 58:

H habituation; 1J3,-J34 hand"se:archiQg,,'174 Hifwthothe etfect, 83

ratio, 219' health (definition),. 2

and' Act

f71 .. health research. overvit:wt 1-S health·seryices: reseg,rch, 93 height; l48, 1£9 'histogram, ·192:....193, 194 hyp,9tbe:ses: for statistical tests,

198:....201

I . IC_MJE ICMJE relerehce:style,, 236....:23 7 impact.factQrs:, ·261

incentives, 153-154 incidence rat\'!, 70-72 inclusion ct'iter'ia

case definitions, 51, 57 for experimental studies,

113-114 for meta-analyses, 40-41 for observational studies, 108,

111 for systematic reviews, 40,

102'-103,174 independent populations, 207-209 independent variables, 213 independent-samples t-test,

207-208 indi rect age adjustment, 47 individual matching, 57-58, 62-63 inducements; 153-154 informant, 91 information bias, 70, 83, 130 informed consent, 154-157

vulnerable populations, 114- 115, 159-160

informed consent statements, 154-155

informed consent process, 155-156

informed consent documentation,

informed consent documentation, 156-157,163

informed consent process, 155-157, 159

informed consent statements, 154-155,166

Institutional Animal Care and Use Committee (IACUC), 161

Institutional Review Board (IRB), 161-168,

application materials, 162-164 exemption from review, 164-165 expedited 165 full review, 165 ongoing review, 166-167 purpose," 161 review by multiple committees,

165-16,6 review of secondary analyses,

172 time for review, 162

intellectual property rights, 167 intention-to-treat analysis, 88 interaction term, 212-213,216- 2 17

284 Index

intercorrelation, 46 internal grants, 99 Internet-based surveys, 135, 140-143 interobserver agreement, 89 interquartile range (IQR), 191, 194 interval variables, 189-190

comparative statistics, 207, 209 descriptive statistics, 194

intervention, 79 intervention studies (See experimen-

tal studies) .interviewer training, 5 interviews, 92, 135, 139-140 in-text citation, 234-235, 236-237 introduction section, 224

J,K journal aim, .scope, and auqience,

260 Kaplan-Meier plot, 219-220 kappa statistic, 89 Kendall's rank correlation (1), 46 keywords, 10-12

for conference abstracts, 250 for manuscripts, 265,268 for abstract database sear<:hes,

16-17 for systematic reviews, 174 MeSH terms, 10-11, 17

Kruskal-Wallis test, 207 Kuder-Richardson Formula 20

(KR- 20), 46

L laboratory research, 2-3 laboratory testing, 149 last author, 28-29 lead author, 28 lead researcher, 25-29 life tables, 219 limitations, 225 linear regression, 214-217 literature reviews, 15-19

abstract databases, 16-17 full-text articles, 17-18 review articles, 37-41, 173-1 78

In( OR), 218 l0gistic regression, 217- 2 1$ log-rank test, 219 fongitudinal cohort studies, 65-70,

111-112 loss to follow-up, 69, 70, 88 lurking variable, 211-213

M magazines, 15,233 mail surveys, 139-142 major revision, 272 Mann-Whitney U test, 207 Mantel-Haenszel adjusted odds ra-

tio, 177 masking 83-84, 89 matched-pairs case-control studies,

62-63 matched-pairs matching, 57- 58,

62-63 matched-pairs odds ratio (OR),

62-63 marched-pairs t-test, 209 matching, 57- 58 maximum likelihood estimation,

213 McNemar's test,)09 mean, 19Q

and sample size, 118-11 9 and standard deviation, 192-193 and z-score, 193 comparative statistics, 207

median, 190 comparative statistics, 207

median survival, 219 medical imaging, 149 medical tecotds, 511. 171-172 MEDLINE,16_:17, 262 Mental Yearbook ,

136 MeSH terms, 10-11 , 17 meta-analyses, 34-35, 37-38,

40-41, 173-178 forest plot 177 funnel plot 178 ·pooled statistic, 176-177 research process, 41 topic selection, 39 writing checklist, 229

methods section, 224-225 Mini-Mental State Examination

(MMSE), 136 minor revision, 272 misclassification bias, 58 missing data, 1'83 . mode,, 190; 207 MOOSE checklist, 229 Moran's todficient, 220 mulcicollinearitx, 214 multiple linear regression, 2 15- 217 multiple logistic regression, 217- 218

multivariate analysis, 211-220 by study approach, 197-198 confounding, 211-213 dummy variables, interaction, 212-213

N

linear regression, 214-217 regression, 217-218

regression, 2 13-219 survival analysis, .86, 219-220

narrative reviews, 37-38, 40 National Health and Nutrition

Examination Survey (NHANES), 170

National Hea lth Interview Survey (NHIS), 170

needs assessment, 54, 93 negative-predictive value (NPV),

88-89 -2log likelihood test, 213 nested case-control studies, 111 n:ewspap.ers, 15, 233 NLM reference style, 236 nominal variables, 128:.._129, 1 89

comparative statistics, 207, 209 descriptive statistics, 194-195

noninfe.riority trial, 80-81 nonmaleficence, 861 nonparametric tests, 206

comparative statistics, 207, 209 correla tion, 45-46

nonrandom-sampling bias, 1 07 non"response bias, 108 normal distribution, 192-193, 206 null hypothesis (Ho), 198-202, 207 null result, 175-176,1 99,278 number needed to harm (NNH ), 86 number needed to treat (NNT),

86-87 numeric variables, 127- 128,

188-190

0 observational studies, 78, 229 odds, ,59-60 odds ratio (OR),. 59-63, 203 one-sample median test; 207 one-sample t-test, 206-207 one-sided test, 201-202 one-way ANOVA, 207, 209 ongoing review, 166-167 online journals, 262

online submission, 261,267-269 open access, 262 open-ended questions, 92, 127, 134 open review, 268 oral consent, 156 oral presentations, 250, 254-258

concurrent sessions, 250 question and answer time,

257-258 slide shows, 2,54-255

ordinal varia bles,_ 128-129, 189 comparative statistics, 207, 209 descriptive statistics, 194

ordinary least squares, 213 originality, 18-19 outcome variables, 213 outcomes

for correlational studies, 44-45 for cross-sectional surveys, 44-45 for experimental studies, 79-81 health outcomes, 11-12

outliers, 191, 195

p page charge, 262 page fee, 262 page proofs, 274-275 paired data, 209-210 paired c-rest, 209-210 pal:red-comparison questions, 128 panel studies, 69 parametric tests, 206

comparative statistics, 207, 209 correlation, 45-46

paraphrasing, 233-23,6 participant diaries, 92 participant observation, 91 participation rate, 140-142

and estimated sample size, 121 and incentives, 154 and study population, 108

patient charts, 51, 171- 172 Pearson correlation coefficient (r),

45-46 peer review, 269-274, 278 person-place-time, 5Q.....51, 57, 224 person-time, 7 1-72 phenomenology, 91 photographs, 227 physical fimess tests, 150 physiological function tests, 149. pie chart, 193-194 pifot test, 13 7

placebo, 81-83, 114 placebo effect, 81 plagiarism, 195, 233-236 plenary sessions, 250 pooled statistic in meta-analysis,

176-177 Population Reference Burea,u, 16 population research, 2- 3 populations, 12 positive predictive value (PPV),

88-89 postal surveys, 139-140 poster presentations, 250, 25i-253

poster content, 251 poster 252-253 poster sessions, 250, 253

power (1-P), 121-122 precis, 245 predictor variables, 213 pregnant women, 114 preprints, 275 pretest, 13 7 prevalence, 54 prevalence rate ratio (PRR), 54 prevalence surveys (See cross-

sectional studies) primary investigator (PI), 100 primary studies, 21-22, 34-35

ethical considerations, 151-168 funding, 99 interviews, 139-144 proposals, 101-102 protocols, 102-103 questionnaire design, 125-135 question,naire validation, 136,

137 recruiting, 140-142 research plan, 98 research populations, 105-)08 sample size, 117-121 l self-administered surveys,

139-144 timelines, 100

PRISMA checklist, 229 prisoners, 114, 154 privacy, 157-158 processing charge, 262 processing 262 program evaluation, 54, 93 project evaluation, 54, 93 proportionate mortality rate, 52 proportions, 207 proposal, 97, 101-102

Index 285

prospective cohort studies, 65-68 protocol, 97, 102-103, 158-159 provisional acceptance, 274 publication bias, 38, 176, 178 publication costs, 261-262 publication ethics

authorship criteria, 26-28 duplicate publication, 235, 279 plagiarism, 195, 233-236

publishing as part of the research process,

5-6,278-279 costs, 261- 262 online journals, 262 page proofs, 274- 275 peer review, 269-274 preparing for submission,

264-266 provisional acceptance, 274 resubmission, 271- 274 submission, 266-269 target journals, 259-262, 264

PubMed, 16-17 PubMed Central, 17 purposive sampling, 91 p-value, 201-202 PysciNFO, 16

Q Q-statistic, 176 qualitative studies, 91-93

comparison with other ap- proaches, 34-35

writing checklist, 229 quality control, 89 quality of life, 81 quartiles, 191 questionnaire design, 125-137 quintiles, 191 quoting,233-235

R r, 45-46 ,2, 46,214-216 random effects model, 176-177 random-digit telephone dialing,

111 , 14 2 randomization, 84-85 randomized controlled trials

(RCTs), 79 range, 191-192 ranked variables, 128- 129, 189

286 Index

comparative statistics, 207, 209 descriptive statistics, 194

rank-order variables, 128- 129 rate ratio (RR), 73-76 ratio variables, 188-190

comparative statistics, 207, 209 descriptive statistics, 194

recall bias, 58- 59, 111 receding

for 2 x 2 analysis, 204 of close-ended questions,

185- 186,204 of open-ended questions 127

recruiting, 140-142 and cultural considerations, 158 and incentives, 153-154

redundant publication, 235, 279 reference group, 204-205 reference list, 236-237 reference standard, 88-89 reference styles, 236-237 regression, 213-219 reimbursement , 153-154 rejection of manuscript, 269-271 rejection without review, 269-270 relative rate (RR), 73-76 relative risk (RR), 73-76 reporting guidelines, 229 representativeness, 54, 107- 109,

140 research (definition), 2, 164-165 research ethics, 4, 151- 168

autonomy, 151, 153, 155 beneficence,86, 151-153 conflicts of interest (COls), 167 distributive justice, 85, 153 informed consent, 114, 154-157,

159, 160 Institutional Review Boards

(IRBs), 161-168, 172 nonma leficence, 86, 151-153 respect for persons, 85, 153 (See also various aspects of re-

search ethics) Research Ethics Committee (REC),

161-168, 172 (See also Institutional Review

Board) research miscond uct, 195, 235 research process, 5, 97-98 residuals, 213-214 resources for research, 98-99

respect for persons, 85, 153 responding to reviewer comments,

272- 274 response scales, 130 resubmission, 271-274 results section, 225 retention, 70 retrospective cohort studies,

65-68 review articles, 34, 37-41 revise and resubmit (R&R),

271- 274 revising, 245- 248, 263 rho (p ), 45 risk ratio (RR), 73-76

s salami publication, 279 sample population, 105- 106, 107

and confidence intervals, 202 sample size, 117-123

and confidence intervals, 202 and power estimation, 12 1-122 sample size estimation, 120-121

sample size estimation, 120-121 sampling, 107 sampling bias, 107 sampling frame, 106 scatterplot, 45,214 schools, 114-115 screening tests, 88-89 secondary studies, 21-22,34-35,

169- 172 research plan, 98 publicly available data sets,

170- 171 private data sets, 171 clinical records, 171-172

self-administered surveys, 139-140 and layout, 134 supplemental data, 147

semi-structured interviews, 92 senior author, 28-29 senior researcher, 25-29, 100 sensitive questions, 132, 139, 158,

159 sensitivity, 88-89 sign test for matched pairs, 209 significance level (a), 121- 122,

201-203 simple linear regression, 214-216 simple random sampling, 107

simple randomization, 84 study, 83 .

skew; 192 skips

iq papct-Qascd ,$urvcys, i34- 135 in surveys,

l30., us, 143 in data entry' p:rog_rams, 183

snowl:>alling, V4 . socioecohorri:ic position {SEPl, '11 soc;ipeconomic status. {:SES), 11 source

al).dconfiqence intervals, 202- 203

spatiill ang.Jy(sis, 150, 22.0 Spearman tank order coftelation

_ (r, p),_ 45-4€ specific a ims, 2.-2-23

hypotheses, specin<: knowledge, 23:5

oqjedives, 22-23 speojficity, ·88-851 spread, 1.91-195 srandatd deviation, 19,2-1 93

and coding for missing infonria- tion, i&B

and cdmparative tests, 206"""207 and s?mpl:e 119 antl z-sc;d're>- 193

2 15.-21$ starrdard·d care, 81- 82 sJatisticaJ honesty, 195 statisxica:l powei: (1- P!, 121..:122

model, '213 · sttatified·tat1dom1zatiob, 84-85

:sampli,ng, 1 d? STROBE 229 .sg-uctured 223

gonls, 22-23 study population, 105- 106, 108 :s-uhmissiorr'fee;262

of a mamrs·cript, 261, 467-269

sujni:nai:y:statisiic ih meta-analysts, 176'-1'77 .

suj:Yeridtftr trial, 7-5J-·so 6,9, 164

survival analys·is, S6.r21.9-2?6 SWOT 9} systemp.ti¢ tevle:ws, 40,

173- 178 data ex.tratrion, 175-17/J eligibility .criteria-, 17 4- 17 5 1·e$earch praces.s, 38 &earth 174

39, 173--174 writing 229

$amplmg, 107

T tables, 226-227 targetjQJ.Ifnal, 259-262,264 targetpopulatidn, 105-10€1 tau {<r), 46 telephon.e iritet\rie\VS, 139'-141, 156

21- 2:-2:, 173-179

research plan,. 98 systematic revrews, 3?-j8, 40,

173-176 57- 3'8, 40-4'1,

176-178 testQf comprehension, 1_5,6 third ·variable effects, 211- 213 tirp.e series, 69 timellriesJ 3·5- 36; 100" 162 to:pi<;' ma,ppil)g, 9- rd, transcrjprrofi, 92 traosl.ation, 159 tteattnent-assigned analysis, 86-8'8

8-6- 8.8 TREND cb:ecklist, 229 t-test, 209 2 ?< 2 table,

fqr cqhoJ;t 74-7.5 for experimental studies, .&7 for matched case.-control studies,

.62.'-63 for sereening ttists, 8 for unmatched casecc'Qnrroi srqd-

:iesj 59-60 recQdit;rg for ·anaLysis., 2.04- 205

tWo-sample tcn'ist,_ 207--'208 te:St1 2Ql-202

type 1 error, 121- 122 type 2 error, 121- 122

u .UNQP; 16 UNICEF; 16 unifoJrn distribt1tioq, 19Z Unifttrm Req ui:tementdor

Manuscripts"Sub.PJitwd t9 Biomedical Journals, 26- 27

unimod.al distr:ibutipn, J 92 univariate analysis, 187-195

abstract, 22.3

v validat:iori, 136 vglues, 4, 10 Vancouver refere-nce s}yle, 236 variability, 1'91, 214-218 vatiables, 127-129, 188-1,90 verbal vollmtarin.es·s,, vulnerabkp6pulations, 114,

159-190

W,X, Y,Z waiver of.cdfisent, 158 \Veb .of 2-P) Web .oHidence, 16

weight, 148_, 15.9 w4ite U4, 25:2 Wilcogdn ra:nksum test, X07 Wilcp;KCHl-sjgned..,rank test, 20'9

id7 wo(d limit's, 223,241,261 Wqrl\1 llank, 1 p Wqrld Health O.ri?;iiri'iz;ariq_n

. (wHO), 15, 16, 170 writer's bkKk, :242-243 writing 2A1 wr-iting strategies, 23-9-243 writing 2?3- ;434, 2.47-4.48 wi'ltten coiiseiit, 156 ,Z-'$COre, 193'

Index 287

' I