statistics for public health and epidemiology

Question 1

A cross-sectional study was conducted to examine the effect of gestational age on systolic blood pressure (SBP) of low birth weight babies who weigh less than 1500 gms. Data was collected on 60 such babies and posted on Moodle in the Excel file Assign2Q1.xls. The dataset contains the following variables.

ID: Participant ID number

sbp = Systolic blood pressure (mmHg)

gestage = gestational age in weeks

a) What are the study factor and the outcome factor?

b) To explore the association, calculate the correlation coefficient and interpret it?

c) Conduct a simple linear regression using Stata and report the Stata output. What are the assumptions for a linear regression? Examine the assumptions with the support of relevant graphs and statistics.

d) Write down the regression equation and interpret the regression coefficients and their 95% confidence interval from part c.

e) What is the expected systolic blood pressure of a newborn whose gestational age is 24 weeks? Show your workings.

Page 2 of 9

PHCM9498 Epidemiology and Statistics –

Question 2

A case-control study was planned to investigate whether there was an association between a mother being diagnosed with toxaemia (A condition in pregnancy, also known as pre-eclampsia characterized by abrupt hypertension, albuminuria and oedema) and the baby being born with low birth weight. The research team wished to recruit the cases and controls from antenatal clinics. Based on a pilot study, the team expected that the odds ratio of the association in question would be 2.5 using a two-sided significance test and the prevalence of toxaemia among women giving birth to a normal weight baby is 6%.

a) If equal number of cases and controls could be recruited in this study, how many in each group would be required to achieve 90% power at 5% level of significance? Include a screenshot of your Stata command and output with your response.

b) One of the researchers thought that prevalence of toxaemia among the controls would be 4%.

i. What effect will this have on the required sample size to detect an OR of 2.5 with the same power and level of significance as in part a)?

ii. If the prevalence of toxaemia in the control group is uncertain, would it be preferable to assume that 4% or 6% of the control mothers have the condition? Describe your reason.

c) A similar study on the same source population found that approximately 80% of the mothers approached for the study would agree to participate. From this information, how many mothers of newborn will need to be approached to achieve the sample size that you estimated in part a)? Show your workings.

d) The research team found that they do not have sufficient number of cases as per the calculation in part a) and decided to recruit 2 controls per case. How many cases and controls would be required if the power, effect size, prevalence of exposure among the controls and level of significance are the same as in part a)? Include a screenshot of your Stata command and output with your response.

e) For 28 of the newborns their apgar score at 5 minutes (ranging from 0 to 10) was available. A description of apgar score can be found here. The research team also wanted to examine if the apgar score differs by the history of toxaemia among the mothers. Apgar score is highly negatively skewed in both groups and for some of the newborns it was zero. What statistical test is appropriate for this study? Explain the reason.

Page 3 of 9

Question 3

Read the Extract, which is a highly edited version of a published paper. Use the information provided in the Extract to answer the following questions.

a) Using PICO format, write the research question for the study.

b) What type of study has been used to answer the research question? What would be the best study design to answer this research question? Provide reasons for your answers.

c) i. In your own words, briefly describe the source population and study participants.

ii. Are there any issues of concern about selection bias in this study? Provide reasons for your answer.

d) i. What is the main study factor? Are there any problems about measurement error of the study factor?

ii. What is the outcome factor? Are there any concerns about measurement error of the outcome factor?

e) i. How have the researchers dealt with potential confounding in this study?

ii. Are there any problems with confounding? Provide a reason for your answer.

f) In your own words, briefly summarise the main results of the study shown in Table 2 of the extract. Please focus on 3 factors that are associated with LHC and one that is not associated with LHC. [Note: You do not need to provide the formal explanation for each set of values.]

g) What is your assessment of the internal and external validity of the study findings?

Page 4 of 9

PHCM9498 Epidemiology and Statistics –

Question 3 – edited extract

Background

The causality of the various types of lympho-haematopoietic cancers (LHC) is multifactorial and includes physical, chemical and biological agents, for example radiation, solvents, viruses etc. In addition, there are a number of agents that are suspected causes ofLHC.

Several studies have suggested that occupational exposure is associated with LHC. These include chemical and woodworking industries and farming, particularly animal breeding and the application of pesticides.

Pesticides, which are widely used in agriculture, comprise a wide variety of chemicals mainly used to control damage to plants from insects (insecticides), mould (fungicides) and weeds (herbicides).

The aim of this study was to investigate the relationship between exposure to pesticide and LHC in the region of central Greece, which has a substantial agricultural labour force. The study investigated application procedures as well as the use of personal protective equipment (PPE) in farming and quantified exposure using detailed questionnaires administered to patients and controls from two hospitals in a central Greece region.

Methods

Study setting

A study of LHC was designed and conducted in two regional hospitals from October 2003 to October 2006. The region is one of the thirteen in the central eastern part

of continental Greece. The total area is

14,036 square kilometres (km2), which represents 10.6% of the area of the entire country. According to the last census conducted in 2001 by the General Secretariat of the National Statistical Service of Greece (ESYE), the population of the region was approximately 740,000 inhabitants, which represents 7% of the total population of the country. The population breakdown is 44% urban, 16% semi-urban and 40% rural. The region ranks second among the agricultural producing regions of Greece and contributes 14.2% of the total primary sector produce of the country.

Cases

Five hundred and ninety-seven (597) histologically confirmed cases of LHC were collected from the medical records of the two hospitals within a two-year period. The FAB (French American British) and REAL (Revised European-American Lymphoma) classifications were used, since both were used by the attending physicians during the study period.

During the data collection period, 35 cases died before entering the study. There were also new cases enrolled (16), and a number (102) of cases could not be identified due to data missing from the medical records (incorrect address or phone number etc). Out of the 476 available cases, 354 agreed to participate in the study and completed the questionnaire (response rate 74.36%).

Controls

The control group comprised subjects from the same hospitals who were admitted for acute conditions such as orthopaedic, eye, ear, nose, throat or surgical diagnoses and

acute cardiac conditions. Some subjects

Page 5 of 9

PHCM9498 Epidemiology and Statistics –

were also recruited from the internal medicine and blood donor departments. All of them were self-declared cancer-free patients. Cases were divided into groups with age ranges of 5 years (e.g. 60-64, 65-69 etc) and were matched according to age and sex.

Almost all controls who were asked to participate completed the questionnaire (response rate 93.2%).

Questionnaire

To ascertain information from both cases and controls a structured questionnaire was developed and pre-tested. It includes the following sections: 1) demographics, 2) residence, 3) occupation, 4) exposure and agricultural practices, 5) family history and 6) habits and medical history.

The demographic section included the hospital where treatment was obtained, level of education and type of insurance. The residence section included full address, chronological residential changes with duration of stay and residential level (urban, suburban or rural). The assignment of the residential level and the coding of the residence were performed using lists from the Ministry of Internal Affairs of Greece. In the occupational section, detailed information on previous employment and other activities were reported.

Subjects who reported at least one year of work in farming were asked to complete the part of the questionnaire on pesticide exposure. The information collected included type of crop, duration of farming, surface area of the farm, characteristics of the adjacent farm and crop infestations.

The participants were asked to provide commercial names of pesticide products, number of applications per year, total years

of use and type of application (spraying,

during planting and usage of treated seeds). To provide help to the participants in recalling the pesticides used, a list of the most common brand names was provided. Moreover, questions were included regarding adherence to application guidelines and familiarity with caution signs/first aid help. The type and purpose of the use of PPE (e.g. uniforms, masks, boots etc), the use of equipment machinery, the frequency of consulting experts, aspects of product handling (e.g. location of storage) and the existence of an integrated farm system were also investigated. In addition, information on certain habits such as changing clothes before entering the house, smoking, eating and accidents (e.g. accidental poisoning), was also collected. Finally, details of work with animals and pet ownership were considered. In the family history section, all types of cancers, LHC and immunological disorders as well as the family proximity (first, second- and third-degree relatives) were recorded.

Categories for smoking habit and alcohol consumption according to the Centers for Diseases Control and Prevention (CDC) 2002 classifications were adopted. Information regarding the subject’s medical history included other cancers, cardiovascular disorders, respiratory diseases, immunological disorders, inflammatory diseases, endocrine disorders, diabetes mellitus and other metabolic disorders.

The questionnaire was pilot tested among both cases and controls aged 27 to 73 years and modifications were made according to the results obtained. The time needed to fill the questionnaire was reasonable with the only delay at the section regarding the specific pesticides used. The interviews were

conducted by a trained occupational

Page 6 of 9

PHCM9498 Epidemiology and Statistics –

physician and the date of diagnosis was used as a reference date for both cases and controls.

Due to changes in pesticide exposure over time, an expert phytopathologist evaluated each questionnaire and the data collected regarding pesticide exposure and handling practices, as well as commercial names of pesticides and types of application. All active ingredients were grouped according to the target pest into three major categories: 1) insecticides 2) fungicides and 3) herbicides.

A new variable was developed for total pesticide exposure for all subjects, in order to classify them in the “Low/No” (referent), “Medium/High” category. Participants who had never worked in the agricultural sector as their main or secondary occupation and declared urban or suburban residences were classified in the “Low/No” exposure category.

In order to avoid reporting exposure that occurred after LHC diagnosis and to allow a minimum latency period, the period preceding one year before the reference date for both cases and controls was used for recorded exposure.

Statistical analysis

Standard statistical procedures were carried out using the epidemiological software Epi-info (version 3.4.3) and the Statistical Package for Social Sciences (SPSS) (version 15.0). Data were validated for completeness before statistical analysis was performed. Descriptive analysis was conducted for each variable including frequencies, ranges, means ± SD, median values and interquartile ranges (IQR) for both cases and controls. The chi-square or Fisher exact test was used to analyze qualitative data, whereas the student t-test

or Mann-Whitney test was used for

quantitative data. Total pesticide exposure and smoking habit as well as smoking and eating during farming were made categorical (yes/no) variables. Agricultural occupations were grouped together (farmer, pesticide applier, seasonal farm worker, animal breeder) and compared to the group of other occupations (e.g. teacher, trader, public servant, unemployed etc.).

Logistic regression analysis was used to control for confounders and to identify independent risk factors. Exposure to pesticides was an independent categorical variable. Age, sex, smoking and family history (all types of cancer, LHC and immunological disorders) were included as confounders. Differences were considered statistically significant when the two-sided p value was ≤ 0.05.

Ethics

The study was approved by the University of Thessaly Scientific Committee which is responsible for ethics. The Institutional Scientific Committee of both hospitals approved access to patients’ files and allowed interviews. The participants provided informed verbal consent and no monetary incentives for participation were offered.

Results

Overall, 354 cases and 455 controls (81 orthopaedics, 84 otolaryngology, 64 ophthalmology, 93 surgery, 42 cardiology, 59 internal medicine and 32 blood donors) were collected. In Table 1 the characteristics of the study population, including sex, agricultural occupation, pesticide exposure and smoking habit, are presented. The median age of both cases and controls was 70 years (IQR 62-

76).

Page 7 of 9

PHCM9498 Epidemiology and Statistics –

The results of univariate data analysis for LHC and different exposures are shown in Table 2 including specific protective behaviours. No association was found between the types of pesticides and LHC (data not shown).

Using logistic regression analysis, pesticide exposure was found to be independently associated with a LHC (OR 1.41, 95% CI 1.00-2.00) after controlling for the following confounders: age, sex, smoking and first-degree family history of all types of cancer, family history of LHC and family history of immunologicaldisorders.

Page 8 of 9

PHCM9498 Epidemiology and Statistics

Table 1: Description of study participants

Cases Controls

Sex Male 188 53.1% 238 52.3% Female 166 46.9% 217 47.7% Agricultural occupation Yes 185 52.2% 212 46.6% No 169 47.7% 243 53.4% Pesticide exposure Yes 282 79.6% 331 72.7% Low/No 72 20.3% 124 27.2% Smoking Yes 148 41.8% 205 45.0% No 206 58.2% 250 54.9% Total 354 455

Table 2: Univariate analysis of exposures among LHC cases and controls

Exposure OR 95% CI

Agricultural occupation 1.25 0.95-1.65 Pesticides 1.46 1.05-2.04 Smoking during pesticide application 3.29 1.81-5.98 Eating during pesticide application 18.18 2.38-381.17 Not changing clothes after work 1.51 1.00-2.29 Handling pesticide-treated seeds 3.01 1.19-7.77 No proper ventilation when using pesticides 3.60 1.77-7.31 Self-application of pesticides 1.14 0.82-1.60

Protective practices OR 95% CI

Compliance with product labels 0.85 0.56-1.28 Use of PPE during application 0.86 0.53-1.38 Familiarity with caution signs 1.05 0.66-1.65

Page 9 of 9