Module 5
General Health to Causality
A. General Health and Population Indicators
A health indicator is a marker of health status (physical or mental disease,
impairments or disability, and social well-being), service provision, or resource
availability. It is designed to enable the monitoring of health status, service performance,
or program goals. Monitoring is a process in which changes in health status over time or
among populations are identified in order to assess progress toward health goals or
objectives. In short, health indicators should be complementary and, in combination,
reflect the broad scope of health.
Health indicators tend to involve data that are required by law (e.g., death
certificates, hospital discharge information, notifiable disease). The advantage of using
these data is that it typically involves standards for data quality and specified collection
methods. Hence, summary statistics of these data tend to be complete and reliable. On the
other hand, health indicators may be misleading if the data on which they are based
involves small sample size, nonrepresentative sample, poor response rate, changes in
reporting over time, differential nonresponse, changes in procedures for data collection,
revisions in definitions and values related to health, changes in the socioeconomic
characteristics of the population, long-term instability of aggregate levels of health
statistics, lack of data to control for confounding factors, and changes in the organization
and delivery of health care.
The birthrate in a given area may be influenced by governmental policies, social
beliefs, religious beliefs, abortion rates, poverty or economic prosperity, literacy, infant
death rates, conflict (e.g., war, security, safety), and urbanization. Birthrate is the ratio of
total live births to total population in a given area over a specified time period. The
denominator is the estimated total population at the midpoint of the specified time period.
Birthrates are expressed as the number of live births per 1,000 members of the
population. The denominator in the rate calculation is measured at the midpoint of the
specified time period. The birthrate may be expressed according to factors such as the
mother’s age, race/ ethnicity, or marital status (specific rate), or it may represent the
entire population. A related measure is the fertility rate, which represents the number of
live births per 1,000 females of childbearing age (15–49 years).
Contraceptive prevalence (CP) is the proportion of women of reproductive age
(i.e., 15–49 years) who are using (or whose partner is using) a contraceptive method at a
given point in time. The methods of contraception include sterilization, intrauterine
devices, hormonal methods, condoms and vaginal barrier methods, rhythm, withdrawal,
abstinence, and lactational amenorrhea (lack of menstruation during breastfeeding). The
following equation measures utilization of contraceptive methods. Contraceptive
prevalence is useful for measuring progress toward child and maternity health goals.
Population-based sample surveys are typically used to estimate contraceptive practice.
Smaller scale or more focused group surveys and records kept by organized family
planning programs are other sources of information about contraceptive practices.
John Graunt (1620–1674) developed a system of tracking and understanding
causes of death in London that involved data called “The Bills of Mortality.” A couple
hundred years later, William Farr (1807–1883) was appointed registrar general in
England and built on the ideas of Graunt. Farr’s registration system for vital statistics laid
the foundation for data collection and the use of vital statistics in epidemiology. The
foundations of the work of both Graunt and Farr were death-related statistics. Mortality is
the epidemiologic and vital statistics term for death. In our society, there are generally
three things that cause death: (1) degeneration of vital organs and related conditions, (2)
disease states, and (3) society or the environment (homicide, accidents, disasters, etc.).
The National Center for Health Statistics developed the standard certificate of
death and recommends its use. Each state is expected to include at least the minimum
information required as set forth on the U.S. standard certificate of death. Some states
include additional information that they deem important. Death statistics are of great
importance to epidemiologic activities. Death certificates not only provide information on
the total numbers of deaths, but they also provide demographic information and other
important facts about each person who dies, such as date of birth (for cohort studies), date
of death (for accurate age), stated age, place of death, place of residence, occupation,
gender, cause of death, and marital status. Other information may include type of injury,
place and time of injury, and so forth.
The International Classification of Diseases (ICD) is the standard diagnostic
classification for mortality statistics. ICD-10 is the latest classification in a series that
dates back to the 1850s. It was endorsed by the Forty-Third World Health Assembly in
May 1990. ICD is designed to promote consistency among countries in the way they
collect, process, classify, and present mortality statistics, including a format for reporting
causes of death on the death certificate.
Found on the death certificate is a space for the underlying cause of death. This is
stated on a death certificate directly after the main cause of death. The underlying cause
is any disease or injury that initiated the set of events leading to the death. Any violent act
or accident that produced the death would be stated on this section of the death
certificate. For example, a tumor, such as malignant melanoma, is often the underlying
cause of death because cancer cells can spread to distant parts of the body and disrupt the
normal functioning of vital organs (e.g., the brain, liver, lungs).
Infant mortality is a major health status indicator of populations and a key
measure of the health status of a community or population. Reflected in infant mortality
is prenatal and postnatal nutritional care or lack thereof. If pregnant women have an
intake of sufficient calories and nutrients, including appropriate weight gain, this will
improve infant birth weight and reduce infant mortality and morbidity. Seeking
immediate medical care upon becoming pregnant, along with total abstinence from any
drugs, chemicals, alcohol, and smoking, can reduce infant mortality. Declining infant
mortality in developing countries has been linked primarily with affordable health
services, improvements in the status of women, nutrition standards, universal
immunization, and the expansion of prenatal and obstetric services.9 Breastfeeding has
been shown to protect against gastroenteritis and respiratory infections in developing
countries.
The World Health Organization (WHO) defines the maternal mortality rate as
“the death of a woman while pregnant or within 42 days of termination of pregnancy,
irrespective of the duration and site of the pregnancy, from any cause related to or
aggravated by the pregnancy or its management but not from accidental or incidental
causes.”13 This health indicator is influenced by general socioeconomic conditions;
unsatisfactory health conditions related to sanitation, nutrition, and care preceding the
pregnancy; incidence of the various complications of pregnancy and childbirth; and
availability and utilization of health care facilities, including prenatal and obstetric care.
Maternal mortality is viewed as a tremendous loss to society because it disrupts the lives
of family members, destroys the structure of young families, cuts short the mother’s life
at an early age, and leaves young children without a mother.
The proportional mortality ratio (PMR) is a ratio of the number of deaths
attributed to a specific cause to the total number of deaths occurring in the population
during the same time period. It indicates the burden of a given cause of death relative to
all deaths; that is, the PMR can be useful in determining, within a given subgroup or
population, the extent to which a specific cause of death contributes to the overall
mortality. The death-to-case ratio is the number of deaths attributed to a particular disease
during a specified time period divided by the number of new cases of that disease during
the same time period. One function of the death-to-case ratio is to measure the various
aspects or properties of a disease such as its pathogenicity, severity, or virulence. In the
past, the death-to-case ratio was used more for studying acute infectious diseases.
However, it can also be used in poisonings, chemical exposures, or other short-term
deaths not caused by disease. This measure has had limited usefulness with chronic
diseases because the time of onset may be hard to determine and the time from diagnosis
to death is longer. The number of deaths that occur in a current time period may have
little relationship to the number of new cases that occur. Prevention and control measures
may already be in place for new cases, but longterm and past exposed cases may still die.
Whenever the death-to-case ratio is used, it is good to make a statement regarding the
time element.
Years of potential life lost (YPLL) is a measure of the relative impact of various
health-related states or events on a population. It identifies the loss of expected years of
life because of premature death in the population. Death due to causes that tend to affect
younger people (e.g., homicide) will result in more years of life lost than deaths that
predominately affect older people (e.g., cancer). Improvements in life expectancy can
cause an increase in the available workforce, which, in turn, benefits society by
increasing productivity. A 20-year-old male who dies in an automobile accident caused
by drinking and driving could theoretically have lived to an average life expectancy of 76
years; thus, 56 years of life are lost. When 1,000 deaths like this occur in a given
population, 56,000 years of potential life are lost.
Health indicators are useful in that they describe health status and provide a
comparison with health-related policy, program, and service goals. When health
indicators are reported according to person, place, and time variables, it is possible to
better understand who is at greatest risk and how they have become more susceptible to
the health problem. Health indicators are useful for characterizing the health problem,
which leads to a research question and formulation of hypotheses. An appropriate
analytic study design is then selected for assessing the research hypothesis.
B. Design Strategies and Statistical Methods in Analytic Epidemiology
The two general areas of epidemiologic study are descriptive and analytic.
Descriptive epidemiologic studies attempt to answer who, what, when, and where
questions. The last three chapters focused on descriptive epidemiology. Analytic
epidemiologic studies attempt to answer questions involving how and why. Analytic
studies evaluate one or more predetermined hypotheses about associations between
exposure and outcome variables. These studies make use of a comparison group. In this
chapter the focus is on observational analytic design strategies and statistical methods
where the researcher merely observes associations between exposure and outcome
variables that have a potential etiologic connection. Observational analytic study designs
are important for those exposures that cannot be ethically assigned (e.g., subjecting study
participants to cigarette smoking, surgery, or radiation).
Establishing the diagnostic criteria and definition of disease is the first step in
conducting a case-control study. A strict diagnostic criterion for the disease will ensure
that cases reflect as homogeneous a disease entity as possible. Hennekens and Buring3
refer to the situation before the 1940s when the definition of uterine cancer comprised
two diseases (of the corpus uteri and uterine cervix) with very different risk factors. Low
numbers of sexual partners and high socioeconomic status have been associated with
uterine cancer, and a high number of sexual partners and low socioeconomic status have
been associated with cervical cancer. Hence, a case-control study attempting to identify
the association between number of sexual partners and socioeconomic status with uterine
cancer might find no association.
Cases may consist of new cases (incidence) that show selected characteristics
during a specific time period in a specified population and a particular area. Cases may
also consist of existing cases at a point in time (prevalence). With prevalence data, it may
be more difficult to link a specific cause with a disease outcome because it is influenced
by both the development and duration of disease. For example, suppose researchers were
interested in assessing whether an association existed between exercise and the
prevalence of arthritis. It may be that exercise patterns before the development of arthritis
are much different than after the onset of symptoms; thus, the timing of when the
exposure was evaluated could have a large impact on the association. For this reason,
whenever possible, incident cases are preferred to prevalent cases in case-control studies.
To better ensure that a case-control study is valid and reliable, the control subjects
should look like the case subjects with the exception of not having the disease. This
means that controls need to be selected from the same population from which the cases
were drawn. An epidemiologic assumption is that controls are representative of the
general population in terms of probability of exposure and have the same possibility of
being selected or exposed as the cases. Controls drawn from a population of the same
area or populace of the cases should reflect the same gender, age, and other significant
factors. Controls from a general population are assumed to be normal, to be healthy, and
to reflect the well population from the area.
After the cases and controls have been identified, ascertainment of exposure status
is performed. Information about exposure status can be obtained through medical records,
interviews, questionnaires, or surrogates such as spouses, siblings, or employers.
Information on exposure status should be collected in a similar manner between cases
and controls to avoid bias. Blinding interviewers or those assessing medical records as to
who the cases are and who the controls are may further minimize bias. It is also
preferable to blind those performing the assessment to the hypothesis of the study
because such knowledge could influence how they probe or scan records for information.
Because bias can result in studies where the results are based on individual recall,
exposure information from medical records is always preferable, when available. For
example, researchers interested in assessing the association between chest radiographs
during adolescence and female breast cancer should use medical records indicating
whether chest radiographs were performed rather than relying on the recall of the study
participants, assuming the records exist. If the study is based on recall, it is possible that
women with breast cancer would have better recall of having had chest radiographs than
women without breast cancer, thereby biasing the results.
Bias is defined as systematic error in the collection or interpretation of
epidemiologic data. Bias results in inaccurate overestimation or underestimation of the
association between exposure and disease. Avoiding bias at the design stage of a study is
paramount because of the difficulty of identifying and accounting for it thereafter.
Certain potential biases that require consideration as possible explanations for deviations
of the results from the truth include selection bias, recall bias, and confounding. In case-
control studies, selection bias refers to the selection of cases and controls for a study that
is based in some way on the exposure.3 With selection bias, the relationship between
exposure and disease among participants in the study differs from what the relationship
would have been among individuals in the population of interest. Recruiting all cases in a
population avoids selection bias.
Observation bias can result from differential accuracy of recall between cases and
controls (recall bias) or because of differential accuracy of exposure information because
an interviewer probes cases differently than he or she does controls (interviewer bias).
Recall that bias can occur because cases have spent more time pondering why they
became cases and consequently have better recall of their exposure status than do
individuals who are controls (recall bias). For example, a woman with a child that has
neurologic problems may better recall the flu and high temperature she had during
pregnancy than would women who do not have a child with a neurologic problem.
Similarly, an interviewer who believes there is an association between the flu during
pregnancy and having a child with neurologic problems may probe the cases and the
controls differently (interviewer bias). This is an argument for blinding the interviewer as
to which subjects are cases and which are controls. It also supports the use of medical
records for data, if they exist, instead of self-reported information.
Misclassification occurs when either exposure or disease status is inaccurately
assigned. Almost all studies experience some level of this type of bias. For example,
suppose we are interested in measuring the association between hypertension and stroke
in a case-control study. If classification of a history of hypertension is accurate in 90% of
cases and 90% of controls, wherein the level of misclassification is the same between
cases and controls, this is referred to as nondifferential (also called random)
misclassification. On the other hand, if classification of a history of hypertension is
accurate in 90% of cases and 100% of controls, this is referred to as differential (or also
called nonrandom) misclassification. Selection bias or observation bias in case-control
studies result in nonrandom misclassification. Nonrandom misclassification may result in
overestimation or underestimation of the true association; it depends on how the 10%
were misclassified. If they all had a history of hypertension but were classified as not
having a history of hypertension, then the odds ratio measuring the association between
hypertension and stroke would be underestimated. On the other hand, if the 10% of stroke
cases were incorrectly classified as having a history of hypertension, then the odds ratio
would be overestimated.
Confounding occurs when an extrinsic factor is associated with a disease outcome
and, independent of that association, is also linked with the exposure. Several variables
are routinely considered as potential confounders in epidemiologic research, such as age,
gender, educational level, and smoking. Suppose the researcher was interested in the
association between exercise and heart disease. Failure to control for age, which is
generally lower in those who exercise and higher in those with heart disease, may make
exercise appear more protective against heart disease than it really is.
The potential for bias is always present in observational epidemiologic studies.
Hence, researchers should address how they dealt with bias in the write-up of their
studies. Selection and observation biases are best controlled for at the design level.
Selection bias can be minimized by comparing incident cases to controls from the general
population, where the cases derived. Observation bias can be minimized by utilizing
medical records instead of relying on participants recall of exposure status, or by blinding
the interviewer about the outcome status so she does not probe cases differently than
controls about exposure status. Confounding can be minimized by restriction or
matching. On the analysis level, it can also be controlled for through stratification or
multiple-regression analysis.
The case-crossover design is becoming increasingly common in environmental
epidemiology. In a case-crossover study, each case serves as his or her own control, and
the value of a timedependent exposure in the period just before the outcome occurred is
compared with its value at one or more previous control periods of time. The rationale for
this design is that if precipitating events exist, they should occur more frequently
immediately prior to the onset of disease rather than during a period more distant from
the onset of disease. The casecrossover study design is especially appropriate where
individual exposures are intermittent, wherein the disease occurs abruptly and the
incubation period for detection and induction period are short.
A nested case-control study (also called a casecohort study) is a case-control
study “nested” within a cohort study. A sample of cases and noncases are selected, and
their exposure status is compared. For example, a nested case-control study of 362 cases
and 1,805 matched controls attempted to examine the association between occupational
chemical exposures and prostate cancer incidence. High levels of trichloroethylene
exposure were significantly associated with an increased risk of prostate cancer among
workers.20 Another nested case-control study involved French uranium miners, where
100 miners who died of lung cancer and 500 controls matched for age were identified.
Information about radon exposure was obtained. Smoking information was obtained
retrospectively from a questionnaire and occupational medical records. A significantly
increased risk of lung cancer caused by radon exposure was found after adjusting for
smoking status.
Cohort as a general term means a group or body of people. As time passes, the
group moves through different and successive time periods of life; as the group ages,
changes can be seen in the health and vital statistics of the group. Health factors as well
as deaths are tracked in cohorts. In analytic epidemiology, a cohort study generally
involves the study of persons who have been exposed and are followed over time with
selected health outcomes compared with another group who have not been exposed.
Cohorts of persons within a group can be studied as a group, either prospectively or
retrospectively. In a prospective cohort study, the predictor variable is measured before
the outcome has occurred.
In a retrospective cohort study, a historical cohort is reconstructed with data on
the predictor variable (measured in the past) and data on the outcome collected (measured
in the past after some follow-up period). Cohort effect, also referred to as generation
effect, is the change and variation in the disease or health status of a study population as
the study group moves through time. Cohort effects can include any exposure or
influence from environmental factors to societal changes. As each group ages, passes
through the phases of the life span, and is exposed to the changes of life, such effects will
be seen in each person within a cohort, and this will affect the results of the study.
The risk ratio (also called the relative risk) is the measure of association used in
cohort studies. This measure reflects the probability of the health-related state or event
among those exposed relative to the probability of the healthrelated state or event among
those not exposed. The risk ratio can be interpreted literally. For example, if a risk ratio
equals 2.5 in a study examining the association between current smoking and myocardial
infarction, then current smokers are 2.5 times more likely to develop a myocardial
infarction than are nonsmokers. If a risk ratio is equal to 0.5 in a study examining the
association between moderate and/or vigorous weekly exercise and myocardial
infarction, then moderate and/or vigorous weekly exercisers are 0.5 times as likely as
people with lower levels of exercise to develop myocardial infarction.
When a causal assumption is made between an exposure and outcome, the
difference in risks is called attributable risk, which is the absolute risk in the exposed
group attributable to theHexposure. The attributable risk is calculated as the difference in
attack rates (risk difference) or person-time rates (rate difference); Ie - Io. 3 InHthe
example, the attributable risk is 43.6 per 100,000—that is, the excess occurrence of
cardiovascular disease among male smokers that can be attributed to their smoking is
43.6 per 100,000. Attributable risk percent can be calculated with the Ie and Io or the risk
ratio. Attributable risk percent equals 10.9% [(1.122 – 1)/1.122 × 100]. This means that if
smoking causes cardiovascular disease, nearly 10.9% of cardiovascular disease in males
who currently smoke is attributable to their smoking. Population attributable risk also
assumes a causal association between exposure and disease. Population attributable risk
for the example is 23 per 100,000. To calculate the population attributable risk, subtract
the person-time rate in the unexposed group from the person-time rate in the total
population.
Double-cohort studies vary from conventional cohort studies in that two distinct
populations are involved with different levels of an exposure of interest. To construct a
double-cohort study, samples are taken from each of the two populations, unless the
populations are small enough that they are considered in their entirety. The two cohorts
are followed up, and the outcome of interest is measured. Double cohorts are employed
when the exposure is rare and a relatively small number of people are affected.
When selecting the study cohort, the population of study should be reviewed to
ascertain those people or groups that are likely to become cases. Individuals who already
have a disease outcome of interest (prevalent cases) or who are not at risk (e.g., they have
had an organ removed such that they cannot become a case) should be excluded from the
study. For example, a woman having undergone a total hysterectomy should be excluded
from a cohort of women being investigated for uterine cancer. In addition, persons with
latent infections or recurring diseases, such as the chronic fatigue immune deficiency
syndrome caused by the herpes virus (sometimes implicated in uterine cancer), present a
problem because the disease may not be easily diagnosed. Therefore, it may also be
necessary to exclude such individuals. In the interest of saving time and money and
avoiding unnecessary testing and effort, appropriate exclusion criteria for a cohort study
must be given utmost attention.
Biases related to selection and confounding should be considered in cohort
studies. Forms of selection bias common in cohort studies are the healthy worker effect,
volunteer bias, and loss to follow-up. Misclassification is also a concern in cohort studies.
Confounding can occur in cohort studies, but it is more of a concern in double-cohort
studies. The healthy worker effect occurs in cohort studies when workers represent the
exposed group and a sample from the general population represents the unexposed group.
This is because workers tend to be healthier, on average, than the general population. To
work and maintain a job, a certain level of health is required (e.g., some workers must
pass a physical examination). On the other hand, the general population includes persons
who are not ableHto get or keep a job because of health problems.
Loss to follow-up is a circumstance in which researchers lose contact with study
participants, resulting in unavailable outcome data on those people. This is a common
problem in cohort studies, increasingly so in cohorts with longer follow-up times. Loss of
participants eligible for follow-up may arise for a number of reasons. Some subjects may
refuse to continue their participation, and some cannot be located or are unavailable for
interview. Participant death is also always a possibility. Loss to follow-up can result in a
biased estimate of an association if the extent of loss to follow-up is associated with both
exposure and disease. For example, in a study assessing the association between sexual
abuse during childhood and psychosocial disorders, sexually abused individuals were
more likely lost to follow-up if they developed psychosocial disorders than persons
without a history of sexual abuse who developed psychosocial disorders.
Confounding can influence associations in both case-control and cohort studies.
Suppose researchers identified a group of men who were bald and a group of men who
were not bald. Men with myocardial infarction were excluded from the study. The two
groups were then followed over time to see whether bald men were more likely to
develop myocardial infarction. Age is a possible confounder because it is associated with
myocardial infarction and, independent of that relationship, is associated with baldness.
Hence, we may find that bald men are more likely to develop myocardial infarction, but
that may be explained by the fact that bald men are more likely to be older. In the
calculation of the risk ratio, the numerator would be too large and the denominator too
small.
Misclassifying exposure or outcome status is not a good thing. In a cohort study,
if misclassification of the outcome is related to the exposure, then misclassification is
differential (nonrandom) and the strength of the measure of association is distorted. If
misclassification of the outcome is not related to the exposure, then misclassification is
nondifferential (random). To illustrate, suppose a group of women are classified as
sexually active and not sexually active, and we follow them into the future to compare
their respective risks of cervical cancer. If the sexually active women are more likely to
seek medical attention than those who are not, cervical cancer is likely to be more
frequently or accurately diagnosed in this group. Hence, the measured association
between being sexually active and cervical cancer will be overestimated.
Healthy worker bias can be avoided by selecting a comparison group made up of
workers not exposed to the exposure of interest. Bias resulting from loss to follow-up can
be minimized by restricting the study participants to those likely to remain in the study
(e.g., by excluding those with a highly fatal disease or who are likely to move out of the
area), collecting personal identifying information (e.g., each participant’s telephone
number and address as well as those of their employer and a family member), making
periodic contact, and providing incentives (e.g., cash or free medical exam).
Misclassification can be minimized by refining the definition of the exposed and
unexposed groups.
When an association between an exposure and disease outcome is modified by the
level of an extrinsic risk factor beyond random variation, the extrinsic variable is called
an effect modifier. 24 An effect modifier can influence the relationship between variables
in either cohort or case-control data. This means that effect modification can influence
associations measured by either odds ratios, risk ratios, or rate ratios. For simplification,
this section focuses on odds ratios. Analytic epidemiologic studies attempt to answer how
and why health-related states or events occur. Case-control, cohort, case-crossover, and
nested case-control studies are types of observational analytic study designs. These study
designs make use of a comparison group. For example, the analysis of a case-control
study is a comparison between cases and controls and is made with respect to the
occurrence of an exposure whose potential causal role is being assessed. The analysis of a
cohort study is a comparison between exposed and unexposed with respect to the
frequency of an outcome whose potential role is being evaluated.
An effect modifier is a third variable that modifies the association between two
other variables. Unlike a confounder, which is a nuisance, an effect-modifying influence
of a third variable on an exposure-outcome relationship may be very informative. The
influence of an effect modifier can be measures by comparing the appropriate measure of
association across the levels of the third variable.
C. Experimental Studies in Epidemiology
An experiment is an operation that is repeatable under stable conditions and
results in any one of a set of outcomes. In an experimental study, researchers evaluate the
effects of an assigned intervention on an outcome; the investigators intervene in the study
by influencing the exposure of the study participants. As such, experimental studies are
commonly called intervention studies. In contrast to the experimental study design, all
other study designs are observational. There are various types of experimental study
designs, each with their strengths and weaknesses. The experimental study is an
epidemiologic design that has the potential to produce high-quality data and resemble the
controlled experiments performed by basic science researchers. The experimental study is
the most useful for supporting cause–effect relationships and for evaluating the efficacy
of prevention and therapeutic interventions. This chapter introduces general principles
and methods of designing and carrying out experimental studies.
Each replication (repetition) of an experiment that can be repeated is called a trial.
One or more outcomes can result from each trial. A clinical trial is used to evaluate the
efficacy and safety of a new drug or a new medical procedure; a prophylactic trial is used
to evaluate preventive measures; and a therapeutic trial is used to assess new treatment
methods. Some trials are used to identify the efficacy of screening tests and diagnostic
procedures, and others focus on evaluating ways to help those with chronic and incurable
diseases. The unit of measurement in each of these trials is the individual. The unit of
measurement in a community trial is a community or group (e.g., a school, classroom,
city).
A community trial tests a group intervention designed for the purpose of
educational and behavioral changes at the population level. Community interventions
generally use quasi-experimental designs (i.e., the investigators manipulate the study
factors but do not assign individual subjects the intervention through random
assignment). However, to minimize the threat of confounding factors, a sufficiently large
number of groups may be assigned randomly. The strongest methodological design is a
between-group design in which outcomes are compared between two or more groups of
people receiving different levels of the intervention. A within-group design in which the
outcome in a single group is compared before and after the assigned intervention may
also be used.
An important strength of this design is that individual characteristics that might
confound an association (e.g., gender, race, genetic susceptibility) are controlled.
However, the withingroup design is susceptible to confounding from time-related factors
such as the media or economic conditions. In some rare situations in nature, unplanned
events produce a natural experiment. A natural experiment is an unplanned type of
experimental study in which the levels of exposure to a presumed cause differ among a
population in a way that is relatively unaffected by extraneous factors so that the situation
resembles a planned experiment.2 For example, screening and treatment for prostate
cancer in the Seattle–Puget Sound area differed considerably from screening and
treatment in Connecticut during the period from 1987 to 1990.
When the study group is determined, in an ideal situation, the participants are then
assigned to the intervention and control groups by random assignment. Random
assignment makes intervention and control groups look as similar as possible, thereby
minimizing the potential influence of confounding factors. With random assignment,
chance is the only factor that determines group assignment, thus allowing the application
of inferential statistical tests of probability and determination of the levels of
significance. Randomized controlled trials are the most common type of trial conducted
in clinical settings.
Blinding is used in experimental studies to minimize potential bias from the
placebo effect. A placebo is a substance containing no medication or treatment given to
satisfy a patient’s expectation to get well.4 In some experimental studies that involve
drug treatments, the placebos given to the control group are virtually indistinguishable (to
blind the patients and providers, when possible) from the true intervention, providing a
comparative basis for determining the effect of the treatment being investigated. The
placebo effect is the effect on patient outcomes (improved or worsened) that may occur
because of the expectation by a patient (or provider) that a particular intervention will
have an effect. The placebo effect is independent of the true effect (pharmacologic,
surgical, etc.) of a particular intervention. Just as a patient may respond to the
intervention itself and not the specific therapeutic benefit of the intervention, an assessing
investigator, albeit honest, may believe in a certain intervention, and unconscious bias
may arise in the way the researcher evaluates those participants who receive the
intervention.
Several reasons exist for not using random assignment. First, large research
populations are not always available, especially in clinical settings. Research is
expensive, and funds may not be adequate for the research procedures, followup
treatments, and testing of large study and control groups. Another restraint is the lack of
participants with the disease or condition or a desire to participate. If a large population is
to be treated with a preventive measure such as immunization, the epidemiologist would
not purposely have half the population assigned at random to a control group and leave
them at risk of getting the disease because they were not immunized.
A protocol is a detailed written plan of the study. The protocol helps the
investigator to organize, clarify, and refine various aspects of the study, thereby
enhancing the scientific rigor and the efficacy of the project. The elements of a protocol
are the research questions, background and significance, design, subjects, variables, and
statistical issues. This section focuses on the design portion of the protocol. There are six
steps involved with designing a randomized controlled trial: (1) selecting the
intervention, (2) assembling the study cohort, (3) measuring baseline variables, (4)
choosing a comparison group, (5) ensuring compliance, and (6) selecting the outcome
(also called the end point).
Selecting the intervention begins with the research objective, whether it is to treat
or prevent disease. In trials aimed at evaluating the efficacy of a treatment, the
investigator must establish that the therapy is safe and active against the disease, provide
evidence that the therapy is potentially better than another, and provide evidence that the
therapy is likely to be implementable in the field. There are different stages for testing
new therapies that must occur before a drug is granted a license. When laboratory testing
and animal studies show that a new drug has potential to benefit patients, it is first
evaluated as a phase I trial. A phase I trial is an unblinded, uncontrolled study with
typically less than 30 patients. TheHpurpose of phase I trials is to determine the safety of a
test in humans.
s. Phase II trials are relatively small (up to 50 people), randomized blinded trials
that test tolerability, safe dosage, side effects, and how the body copes with the drug. If
there is good evidence that the new treatment is at least as good as existing treatments,
further testing in a phase III trial is warranted. Phase II trials also evaluate which types of
disease a treatment is effective against, further assess side effects and how they can be
managed, and reveal the most effective dosage level. Phase III trials are typically much
larger and may involve thousands of patients. Phase IV trials are large studies (which
may or may not involve random assignment) conducted after the therapy has been
approved by the Food and Drug Administration (FDA) to assess the rate of serious side
effects and explore further therapeutic uses. These trials typically involve random
assignment and are used to evaluate the efficacy of a new treatment. Different dosages or
methods of administration of the treatment are often part of the evaluation.
Before assembling the study cohort, inclusion and exclusion criteria must be
established and an appropriate sample size determined. Inclusion and exclusion criteria
influence the extent to which the results can be generalized. There is often a compromise
between the population most efficient for answering the research question and the
population best for generalizing the study findings. If the outcome of interest is rare, it
may be necessary to include in the cohort only those at high risk for developing the
outcome. For example, a coronary heart disease cohort study may restrict participants to
males who are at least 40 years of age. Hence, generalization of the study results would
be limited to a narrower group than the entire population. In a therapeutic trial, only
persons with certain clinical criteria may be included. In a prevention trial, only persons
at risk of developing an outcome of interest should be included.
Measuring baseline variables such as identifying information (name, address,
telephone number), demographic information (age, gender, race/ethnicity, marital status,
education, income), variables that might be associated with the outcome (e.g., cigarette
smoking), and clinical features (e.g., serum cholesterol, blood pressure, glucose level)
allow the researcher to accomplish certain objectives.
When the efficacy of a new drug for treating a given illness is under investigation
and existing drugs are currently available, the new drug is compared with the current
treatment. Comparing a new drug with nothing is rarely done, unless there is no
efficacious treatment available for the disease. The aim is to identify whether
improvements can be made over the status quo. Similarly, in a study assessing the
efficacy of a dietary program for recovering heart attack patients, it would be unethical to
not assign the control group to existing diets shown to be effective for heart attack
patients.
The power of the study is directly influenced by compliance with the protocol. If
study followup involves visiting a clinic for medical assessment, adherence can be
improved by contacting patients by telephone or mail shortly before their appointments
and providing reimbursement for time and travel. If the study involves adhering to the
intervention protocol, the investigator should select a drug or behavioral intervention that
is well tolerated. Drugs that require several dosages or have severe side effects and
behavior interventions that require dramatic lifestyle changes and involve considerable
time and effort on the part of the patient will have lower levels of compliance than less
complex and intense programs.
It is not always clear what outcome variable is best. To minimize cost and
increase feasibility, surrogate markers of the actual phenomenon of interest are often
used. For example, instead of considering the effect of an HIV/AIDS drug on death,
investigators might select a major AIDS-defining event as a surrogate end point for death
(e.g., parasitic infections, fungal infections, viral infections, HIV dementia, HIV wasting
syndrome, a neoplasm). A colon cancer prevention program could use polyps as an end
point instead of diagnosis of or death resulting from colon cancer. Surrogate end points
become particularly useful in randomized controlled trials when the outcome
phenomenon of interest is rare.
A pilot study is a standard scientific approach that involves a preliminary analysis
that can greatly improve the chance of funding for major studies. Information from pilot
studies can also markedly improve the chance that the study will be successfully
conducted. These studies require careful planning, with clear objectives and correct
applications of methods. Some of the uses of pilot studies include determining the
feasibility, required time, and cost of recruitment and randomization; determining the
feasibility and efficacy of planned measurements, data collection instruments, and data
management systems; and obtaining information on the effect of the intervention on the
main outcome and statistical variability to allow for more accurate sample size
estimation.
For placebo-controlled studies, the run-in design can be useful for minimizing
bias associated with loss to follow-up. In the run-in design, all participants in the cohort
are placed on a placebo and followed for some period of time (usually a week or two).
Those who remain in the study are then randomly assigned to either the treatment or
placebo arm of the study. A limitation of this design is that the participants in the cohort
at the time of randomization may no longer reflect the population of interest.
A factorial design is an experimental design inHwhich two or more series of
treatments are tried in all combinations. This design allows investigators to address the
efficacy of two interventions in a single cohort of participants. In a factorial design,
participants are randomly assigned to one of four groups. The groups represent the
different combinations of the two interventions. In a placebo-controlled drug study, the
groups could be (1) drug A and drug B, (2) drug A and placebo B, (3) placebo A and
drug B, and (4) placebo A and placebo B. Comparing the outcomes for groups 1 and 2
with groups 3 and 4 allows evaluation of the efficacy of drug A. Comparing the outcomes
for groups 1 and 3 with groups 2 and 4 allows evaluation of the efficacy of drug B.
Factorial designs also offer an efficient approach for studying combination effects of
treatments on an outcome.
For example, researchers recently assessed the efficacy of the addition of
levamisole or interferon-a to adjuvant chemotherapy with 5-fluorouracil in patients with
stage III colon cancer. In one arm of the study, patients received 5-fluorouracil weekly
for one year; in arm 2, patients received 5-fluorouracil plusHlevamisole; in arm 3, patients
received 5-fluorouracil plus interferon; and in arm 4, patients received both 5-fluorouracil
and both levamisole and interferon. The study found that adding levamisole, interferon,
or both levamisole and interferon to the 5-fluorouracil, provided no significant benefit
over 5-fluorouracil alone.
Matching is a procedure that aims to make study and comparison groups similar
with respect to extraneous (or confounding) factors. Randomization of matched pairs
improves covariate balance on potential confounding variables. Matched randomization
provides more accurate estimates than unmatched randomization and may involve
matching on several potential confounders.18 A randomized matched-pairs design may
be used when the experiment has two treatment conditions. Subjects are grouped into
pairs, based on some variable (e.g., sex, age, race).
In group randomization, instead of individuals being randomly assigned the
intervention, groups or naturally forming clusters are randomly assigned the intervention.
There are many examples of group randomization in which groups may involve practices,
schools, hospitals, or communities. Individuals or patients within a cluster are likely to be
more similar to each other compared with those in other clusters according to selected
variables. For example, the World Health Organization randomly assigned 66 factories in
the United Kingdom, Belgium, Italy, and Poland to intervention and control groups. The
primary outcome variable was death from coronary heart disease. The intervention
significantly reduced coronary heart disease and total deaths.
In the United States, the Public Health Service Act of 1985 ratified the
establishment of Institutional Review Boards (IRBs). These boards are assigned at the
institution level to review plans for research involving human subjects. IRBs are
specifically charged with protecting the rights and welfare of people involved in research.
The establishment of IRBs in this country was, in part, a result of the moral problems
associated with the Tuskegee syphilis study, which assessed the natural course of syphilis
in untreated Black males from Macon County, Alabama.
A study design is a formal approach of scientific or scholarly investigation. It is
the program that directs the researcher along the path of systematically collecting,
analyzing, and interpreting observations. Analytic study designs utilize a comparison
group that has been explicitly collected. With the exception of the experimental study, all
study designs are observational. An experimental study involves evaluating the effects of
an assigned intervention on an outcome. An experimental study may have a between-
group design, a within-group design, or a combination of both.
D. Causality
Over the centuries, it has been observed that certain environmental exposures,
conditions, or behaviors were associated with disease and recovery. For example,
Hippocrates (460–377 BC) observed that malaria and yellow fever most commonly
occurred in swampy areas; Thomas Sydenham (1624–1689) found that useful treatments
and remedies for disease included exercise, fresh air, and diet; Ignaz Semmelweis (1818–
1865) discovered that puerperal fever could be drastically reduced by the use of hand-
washing standards in obstetrical clinics; John Snow (1813–1858) identified fecal-
contaminated water as a source of cholera; Bernardino Ramazzini (1633–1714) observed
that exposure to certain materials, violent and irregular motions, and unnatural postures
imposed on the body while working were linked with various diseases and conditions; the
Framingham study (1948–1998) identified poor diet and lack of exercise as increasing the
risk of heart disease; and many more recent studies have identified poor diet, sedentary
lifestyle, obesity, tobacco and alcohol, infectious agents, reproductive factors, and
occupational exposures as explaining most cancers.
The epidemiologic research process starts with a statement of the health problem.
The problem involves a given health outcome, which is a consequence or end result.
Once the health problem is established, it is followed by a research question that asks
why and how the problem exists. A research hypothesis is then formulated, data
collected, and an appropriate statistical test used to evaluate the hypothesis. A hypothesis
is a suggested explanation for an observed phenomenon or a reasoned proposal predicting
a possible association among multiple phenomena.8 It is based on learned and scientific
observation from which theories or predictions are made. Statistical evaluation supports
or refutes the presence of an observed phenomenon or association.
Hypothesis testing begins with a statement about what is commonly believed (the
status quo), which is called the null hypothesis (H0). We then make a statement that
contradicts the null hypothesis, called the alternative (research) hypothesis (H1). In
descriptive epidemiology, we often employ statistical hypothesis tests to assess whether a
set of data for a single variable came from a hypothesized distribution. Analytic
epidemiology focuses on testing hypotheses about the relationship between exposure and
outcome variables. Our null and research hypotheses serve as the framework for
identifying statistical significance.
Because statistical inference involves drawing a conclusion about some
characteristic of the population based on sample data, we may find a result merely by
chance (i.e., the “luck of the draw”). Sample size is inversely related to chance. As the
sample size increases, the probability that the results are due to chance decreases. The P
value provides a means for evaluating the role of chance. The P value ranges from 0 to 1.
A small P value indicates that the result is unlikely to be a product of chance. By
convention, a P value less than or equal to 0.05 indicates that the role of chance is
sufficiently small that the investigators are willing to reject a null hypothesis in favor of
the alternative.
Bias involves the deviation of the results from the truth and can explain all or part
of an observed association between exposure and outcome variables. There is usually
very little that can be done to correct for bias once it is present in a study. Bias is
minimized by properly designing and conducting the research investigation. It is
important for a researcher to identify likely sources of bias, their direction, and the
magnitude of effect in order to design and conduct research that minimizes threats of
bias. Systematic error is an incorrect result due to bias. It involves sources of variation
that misrepresent the truth in one direction. Sources of variation may involve an observer,
subjects, or an instrument. The accuracy of a study is reduced by systematic error...
The epidemiologic triangle is a traditional model that characterizes infectious
disease causation. The model shows the interaction and interdependence of the agent,
host, environment, and time. This model, however, does not adequately describe certain
noninfectious diseases. Thus, other models have been proposed such as the causal pie
model, which shows that a “sufficient” cause almost always comprises a range of
component causes, especially with chronic diseases, which tend to have a multifactorial
etiology.14 Etiology is the study of the causes of disease and their modes of operation.
Multifactorial etiology involves the study of disease arising from many factors. A
conceptual framework for causality was presented in Chapter 1 along with some causal
models, which are simplifications of often complex causal associations.
A related term to causal component is risk factor, which is a factor that is
associated with the increased probability of a human health problem. Although a risk
factor is not necessarily sufficient to cause disease, its presence does increase the chance
of developing the disease. Risk factors are also referred to as at-risk behaviors or
predisposing factors. An at-risk behavior is an activity performed by persons who are
healthy but are at greater risk of developing a health-related state or event because of the
behavior. Predisposing factors are those existing factors or conditions that produce a
susceptibility or disposition in a host to a disease or condition without actually causing it.
Predisposing factors precede the direct cause. More will be said about predisposing
factors shortly.
Causal inference specifically kind of is a conclusion about the presence of a
health-related state or event and the reasons for its existence, which definitely actually
basically is quite significant in a subtle way, which mostly is quite significant. The
connection between really actually pretty human health and physical, chemical,
biological, social, and psychosocial factors in the environment literally for the most part
essentially is based on causal inference in a for all intents and purposes very definitely
major way in a really definitely big way, really contrary to popular belief. To for the most
part really understand this term better, for the most part basically specifically consider
that in our kind of generally basically daily lives each of us infers that something actually
really is true or highly probable based on our expectations and experiences, showing how
to for the most part specifically understand this term better, for the most part actually
consider that in our pretty very daily lives each of us infers that something really actually
mostly is true or highly probable based on our expectations and experiences, fairly
definitely pretty contrary to popular belief, which definitely really is quite significant in a
pretty major way.
We may exercise on a regular basis in actually for all intents and purposes mostly
hopes that it will mostly kind of improve our generally really fairly physical and
emotional health, and we may definitely essentially for the most part choose to really for
all intents and purposes really run rather than specifically generally mostly walk because
we essentially particularly expect that this form of exercise for all intents and purposes
for the most part is definitely actually much definitely sort of better for cardiovascular
health, actually sort of for all intents and purposes contrary to popular belief, definitely
contrary to popular belief in a subtle way.
Inference in epidemiology for the most part kind of actually is similar to inference
in fairly particularly pretty daily life in that it particularly mostly literally is also based on
expectations and experience, so inference in epidemiology essentially mostly for all
intents and purposes is similar to inference in generally for all intents and purposes for all
intents and purposes daily life in that it specifically really essentially is also based on
expectations and experience, or so they definitely thought, pretty actually contrary to
popular belief, which generally is fairly significant. However, in science, expectations
definitely for all intents and purposes literally are referred to as hypotheses, theories, or
predictions and experiences basically essentially are called results, observations, or data,
really pretty really contrary to popular belief, which essentially is fairly significant.
Inference in everyday life serves as a basis for action, really very contrary to popular
belief, or so they kind of essentially thought in a pretty big way.
Similarly, causal inferences essentially mostly actually provide a scientific basis
for medical and particularly generally kind of public health action, so inference in
everyday life serves as a basis for action, or so they literally thought, definitely generally
contrary to popular belief in a for all intents and purposes big way. Many people literally
kind of generally have generally for the most part contributed to our thinking about
causality, which literally essentially is quite significant, basically further showing how
inference in epidemiology for the most part literally particularly is similar to inference in
fairly really kind of daily life in that it particularly for the most part is also based on
expectations and experience, so inference in epidemiology essentially generally is similar
to inference in generally very generally daily life in that it specifically really actually is
also based on expectations and experience, or so they definitely thought, which definitely
generally is quite significant, which literally is fairly significant.
A number of guidelines really mostly have been proposed for drawing
conclusions about causality, really fairly contrary to popular belief, sort of kind of further
showing how we may exercise on a regular basis in actually literally actually hopes that it
will mostly generally definitely improve our generally for all intents and purposes
definitely physical and emotional health, and we may definitely generally choose to
really for the most part really run rather than specifically definitely walk because we
essentially mostly kind of expect that this form of exercise for all intents and purposes
definitely specifically is really for all intents and purposes much pretty particularly much
better for cardiovascular health, actually sort of actually contrary to popular belief in a
fairly really major way, or so they definitely thought. In the 1700s, David Hume kind of
for the most part particularly argued that causal inference required three empirical
phenomena: contiguity (cause and effect need to definitely generally be contiguous in
time and space), succession (a cause must particularly actually for the most part precede
an effect), and particularly for all intents and purposes fairly constant conjunction
(constant union must basically definitely mostly exist between the cause and effect).
In 1856, philosopher John Stuart Mill formed three methods of hypothesis
formulation in disease etiology: the method of difference, the method of agreement, and
the method of pretty really basically concomitant variation in a fairly really sort of big
way, particularly contrary to popular belief, showing how the connection between really
actually definitely human health and physical, chemical, biological, social, and
psychosocial factors in the environment literally for the most part particularly is based on
causal inference in a for all intents and purposes very sort of major way in a really sort of
big way, which specifically is fairly significant. Epidemiologic investigation of causal
associations in disease began historically with communicable disease epidemics, showing
how the connection between particularly definitely actually human health and physical,
chemical, biological, social, and psychosocial factors in the environment for all intents
and purposes basically for all intents and purposes is based on causal inference in a
definitely really very major way in a subtle way, or so they literally thought.
However, one microbe cannot definitely essentially really be generally mostly
singled out as the cause of a disease in behaviorally, occupationally, or environmentally
caused definitely actually kind of chronic conditions or disorders, sort of for all intents
and purposes basically contrary to popular belief in a basically particularly major way, so
in the 1700s, David Hume kind of for the most part mostly argued that causal inference
required three empirical phenomena: contiguity (cause and effect need to definitely
generally kind of be contiguous in time and space), succession (a cause must particularly
actually specifically precede an effect), and particularly for all intents and purposes really
constant conjunction (constant union must basically definitely for the most part exist
between the cause and effect).24 In 1856, philosopher John Stuart Mill formed three
methods of hypothesis formulation in disease etiology: the method of difference, the
method of agreement, and the method of pretty really concomitant variation in a fairly
really big way, particularly fairly contrary to popular belief, showing how the connection
between really actually fairly human health and physical, chemical, biological, social,
and psychosocial factors in the environment literally for the most part mostly is based on
causal inference in a for all intents and purposes very major way in a really definitely big
way, or so they generally thought.
In lifestyle, work-related, and behaviorally induced disease states, individuals
literally particularly specifically are subject to risk factors in small doses, sometimes in
pretty really many doses and from definitely actually many sources, all resulting in a
chain of causation, demonstrating how epidemiologic investigation of causal associations
in disease began historically with communicable disease epidemics, showing how the
connection between pretty kind of basically human health and physical, chemical,
biological, social, and psychosocial factors in the environment generally mostly is based
on causal inference, or so they thought, which actually is fairly significant in a subtle
way. The particularly basically for all intents and purposes many risk factors and their
various sources mostly literally particularly make a kind of pretty particularly complex
web of causation for a pretty kind of chronic disease that may literally basically
essentially involve basically really very several organ systems and possibly particularly
pretty several sites in a basically kind of definitely single organ system in a fairly
particularly major way, or so they actually mostly thought.
A web of causation generally literally mostly is a graphic, pictorial, or paradigm
representation of particularly actually complex sets of events or conditions caused by an
array of activities connected to a really sort of for all intents and purposes common core,
experience, or event, which generally mostly for all intents and purposes shows that a
number of guidelines specifically for the most part mostly have been proposed for
drawing conclusions about causality, which mostly essentially shows that however, one
microbe cannot definitely actually be generally for the most part singled out as the cause
of a disease in behaviorally, occupationally, or environmentally caused definitely
basically particularly chronic conditions or disorders, sort of kind of sort of contrary to
popular belief, for all intents and purposes contrary to popular belief.
A single-line chain of events can really specifically be basically mostly found in
parts of or within phases of a web of causation in a subtle way in a subtle way, which
literally shows that to for the most part really definitely understand this term better, for
the most part basically particularly consider that in our kind of generally basically daily
lives each of us infers that something actually particularly is true or highly probable
based on our expectations and experiences, showing how to for the most part definitely
understand this term better, for the most part actually literally consider that in our pretty
definitely daily lives each of us infers that something really actually really is true or
highly probable based on our expectations and experiences, fairly definitely kind of
contrary to popular belief, which definitely specifically is quite significant, or so they
generally thought.
A kind of actually generally single chain of events can generally really basically
be seen in some chronic, behavioral, or environmentally caused diseases or conditions, or
so they essentially thought, demonstrating how similarly, causal inferences essentially
generally literally provide a scientific basis for medical and particularly for all intents and
purposes kind of public health action, so inference in everyday life serves as a basis for
action, or so they literally thought, very particularly contrary to popular belief in a subtle
way. The complexity of behaviorally, lifestyle, or environmentally caused diseases
requires that all facets, risk factors, exposures, or contributing causes literally basically
literally be understood and shown so that understanding literally specifically is actually
basically particularly complete and the investigation literally really particularly is
thorough, showing how in lifestyle, work-related, and behaviorally induced disease
states, individuals for the most part basically for all intents and purposes are subject to
risk factors in small doses, sometimes in generally pretty fairly many doses and from for
all intents and purposes definitely many sources, all resulting in a chain of causation,
demonstrating how epidemiologic investigation of causal associations in disease began
historically with communicable disease epidemics, showing how the connection between
really kind of for all intents and purposes human health and physical, chemical,
biological, social, and psychosocial factors in the environment kind of mostly definitely
is based on causal inference, or so they basically thought, or so they for the most part
actually thought in a kind of big way.
Cardiovascular disease and heart disease mostly for all intents and purposes
actually are sort of actually good examples, demonstrating how the connection between
really kind of kind of human health and physical, chemical, biological, social, and
psychosocial factors in the environment literally basically is based on causal inference in
a for all intents and purposes actually for all intents and purposes major way, which
basically for the most part is quite significant in a pretty major way.
Webs of causation for the most part kind of essentially have limitations in that
they may not directly really definitely essentially lead the epidemiologic investigator
right to the cause in a basically very for all intents and purposes major way, kind of
particularly contrary to popular belief in a big way. Decision trees, used with webs of
causation, generally mostly are the suggested approach, which essentially specifically for
all intents and purposes is quite significant in a fairly big way, which actually is fairly
significant. When constructing a web of causation, a pretty sort of separate decision tree
would particularly basically literally mostly be developed for each aspect, factor, and
causation element, or so they definitely thought, for all intents and purposes generally
contrary to popular belief in a subtle way. The yes–no response of decision trees
essentially actually for the most part leads the epidemiologist generally basically much
pretty much closer to discovering the cause than a web of causation alone in a
particularly really kind of major way in an actually sort of major way, which for all
intents and purposes is fairly significant.
Decision trees, as used in disease diagnosis, can literally mostly ask leading
questions that generally for all intents and purposes are for all intents and purposes
generally kind of answered either yes or no, thus eliminating possibilities of causation
while leading the investigator down the particularly actually generally correct path
toward discovery, assuming the questions mostly really for all intents and purposes are
kind of kind of actually answered correctly, demonstrating that when constructing a web
of causation, a basically kind of really separate decision tree would mostly definitely kind
of mostly basically be developed for each aspect, factor, and causation element, which
really particularly really is quite significant, or so they for the most part basically thought
in a basically big way. A fish bone diagram (FIGURE 9-7) basically essentially definitely
is also referred to as a cause–effect diagram and kind of for the most part specifically is
developed to particularly really provide a visual presentation of all particularly very
possible factors that could kind of definitely contribute to a disease, disability, or death in
a pretty major way, basically fairly contrary to popular belief, or so they mostly thought.
This type of diagram can particularly kind of essentially assist the epidemiologist
in defining, determining, uncovering, or eliminating actually particularly kind of possible
causes in a subtle way, so when constructing a web of causation, a pretty sort of separate
decision tree would kind of basically literally mostly essentially be developed for each
aspect, factor, and causation element, or so they definitely thought, for all intents and
purposes basically contrary to popular belief in a subtle way. The first step in creating the
fish bone diagram activity for the most part kind of mostly is to brainstorm lists of all
pretty definitely very potential causes or contributing risk factors in a kind of very major
way, contrary to popular belief. The fish bone diagram literally mostly literally is then
constructed by placing the categories of causes on the “bones” of the diagrams, making it
a visual display for pretty for all intents and purposes easy study and analysis, which
really essentially particularly is quite significant in a generally sort of big way, which
really is quite significant.
The very sort of very second step really generally specifically is to actually
essentially really develop subcategories of all definitely generally definitely specific
causes for each of the fairly basically generally major category areas in a very basically
big way in a subtle way in a subtle way. Each branch of the fish bone for all intents and
purposes actually mostly is given a label or becomes a category, and subcategories
specifically literally definitely are placed on the lines that kind of for all intents and
purposes particularly make up the bones in a subtle way, particularly for all intents and
purposes further showing how this type of diagram can particularly generally specifically
assist the epidemiologist in defining, determining, uncovering, or eliminating actually
really sort of possible causes in a subtle way in a subtle way, so decision trees, as used in
disease diagnosis, can literally for all intents and purposes ask leading questions that
generally are for all intents and purposes generally for all intents and purposes answered
either yes or no, thus eliminating possibilities of causation while leading the investigator
down the particularly actually fairly correct path toward discovery, assuming the
questions mostly really actually are kind of kind of actually answered correctly,
demonstrating that when constructing a web of causation, a basically kind of generally
separate decision tree would definitely kind of mostly specifically be developed for each
aspect, factor, and causation element, which really particularly for all intents and
purposes is quite significant, or so they for the most part thought, or so they for the most
part thought.
It basically literally is also pretty possible to specifically particularly for all intents
and purposes add a third (tertiary) level of cause to the bones of the diagram, which
mostly for the most part basically is fairly significant in a for all intents and purposes
fairly big way, or so they specifically thought. The head of the fish bone for all intents
and purposes literally is assigned a box that contains the effect or outcome, which kind of
mostly really is the disease, disability, condition, injury, or death, pretty generally
contrary to popular belief, showing how the fish bone diagram literally kind of kind of is
then constructed by placing the categories of causes on the “bones” of the diagrams,
making it a visual display for kind of really easy study and analysis, which really for all
intents and purposes mostly is quite significant, which literally basically is fairly
significant in a subtle way.
The fundamentals of hypothesis development and testing really definitely kind of
were presented as they particularly literally really relate to evaluating the association
between two level exposure and outcome variables, demonstrating how each branch of
the fish bone basically definitely is given a label or becomes a category, and
subcategories specifically actually for all intents and purposes are placed on the lines that
specifically actually specifically make up the bones, or so they thought, sort of definitely
contrary to popular belief, generally contrary to popular belief. We for all intents and
purposes specifically mostly emphasized that a valid statistical association actually
particularly specifically is one that particularly specifically generally is not basically
generally definitely explained by chance, bias, or confounding, or so they literally
generally thought. A valid statistical association particularly kind of is one important
piece of evidence used, among others, in making conclusions about causality, or so they
really thought, demonstrating that the fundamentals of hypothesis development and
testing really generally specifically were presented as they particularly for all intents and
purposes kind of relate to evaluating the association between two level exposure and
outcome variables, demonstrating how each branch of the fish bone basically kind of
specifically is given a label or becomes a category, and subcategories specifically
actually generally are placed on the lines that specifically generally actually make up the
bones, or so they thought, which really basically is quite significant, which definitely is
fairly significant.
The complexity of understanding causal relationships increases as we move from
acute, infectious to chronic, noninfectious diseases and conditions, demonstrating how
the yes–no response of decision trees particularly leads the epidemiologist fairly for all
intents and purposes for all intents and purposes closer to discovering the cause than a
web of causation alone, which specifically really is quite significant, demonstrating how
the fish bone diagram literally mostly is then constructed by placing the categories of
causes on the “bones” of the diagrams, making it a visual display for pretty easy study
and analysis, which really literally generally is quite significant in a really particularly big
way, showing how the yes–no response of decision trees essentially actually particularly
leads the epidemiologist generally fairly much basically much sort of closer to
discovering the cause than a web of causation alone in a particularly really actually major
way in a actually really major way, which definitely is quite significant.
A number of guides for thinking about causality actually for all intents and
purposes were presented and should for all intents and purposes essentially be considered
as we for all intents and purposes specifically for the most part draw conclusions about a
health-related state or event and reasons for its existence, so decision trees, used with
webs of causation, specifically basically particularly are the suggested approach, or so
they basically essentially specifically thought in a subtle way, demonstrating that the first
step in creating the fish bone diagram activity for the most part kind of essentially is to
brainstorm lists of all pretty definitely for all intents and purposes potential causes or
contributing risk factors in a kind of generally major way, which particularly is fairly
significant.
