Epi 200B 2010

Practice Midterm

Question 1

You are given the task to examine if a high parity protects against malignant melanoma (MM, a skin cancer) that is expected to be caused by genetic factors and environmental exposures such as sunshine. By using data from a regional cancer registry all incident cases of MM were identified over a period of one year (400 cases). A similar number of controls were identified from the region that gave rise to the cases by using density sampling. All of these cases and controls were invited to take part in case-control study that would include genetic factors as well as other exposures such as parity (see E). Among the identified cases 300 accepted the invitation and among the 400 invited controls 200 took part in the study. For cases 200 had no children 75 had 1 child, 25 had 2 or more children. In participating controls these numbers were 100, 75, and 25 respectively.

A)Estimate the association between MM and number of children using nulliparous as reference.

B)Under which circumstances will the non-responders not cause selection bias?

C)Is selection bias likely for a study on parity, explain.

D)Is selection bias likely for a study on genetic factors, explain.

E)Estimate the odds ratio for parous versus nulliparous women.

Please note that Parity means the number of live born children a woman has delivered. Nulliparous means women had not given birth to any children and parous means that they gave birth to at least one child.

F)Assume some of the cases and controls who did not want to provide a blood sample and take part in a lengthy interview are willing to respond to a few questions. Assume that you are allowed to ask 5 questions, which questions will you ask?

G)Assume that the variable on parity mentioned in question “E” [nulliparous versus parous] has a similar distribution in (non-responding) controls as in responding controls, how must the distribution be in non-responding cases to get an OR that indicates no association between MM and the dichotomized value for parity?

H)If you want to adjust your analysis for sun exposure and number of sunburns, asking questions on this exposure will provide inaccurate data. Do you expect differential misclassification? Explain and explain what is meant by differential and non-differential misclassification.

I)If you assume that there is association between parity and the risk of MM but sunburns increases the risk of MM two fold. Could that bias your results on the parity-MM association, when you adjust this association for sunburns.

  1. When sunburns data are recorded with no misclassification
  1. When they are recorded with misclassification?

QUESTION 2

A case control study was conducted to assess the association between prenatal exposure to organic solvents (a frequent exposure among shoemakers, painters, cabinet makers and others) with neural tube defect (NTB) in the offspring. All cases were included within a certain period and controls were selected from newborns without NTD. Exposure was measured by asking the mother if they worked with solvents during the pregnancy. Among 200 cases 146 were reported un-exposed and among 200 controls 173 were reported un-exposed.

1)Which study design did they use and what are the main method problems in using this design?

2)What is the Odds Ratio (OR) based on the reported data and what is its interpretation?

3)Suppose the sensitivity for estimating exposure using the questions that provide the exposure data was 90% among cases and 80% among controls; specificity of measuring exposure were both 100% among cases and controls.

A)What is this kind of misclassification called? Explain.

B)Estimate the corrected Odds Ratio (given no misclassification of the disease, no selection bias and no confounding)?

C) Exposure assessment was based on job titles and use of a so-called job exposure matrix (a list of job titles with a high frequency of exposure) developed by trained occupational hygienists. Comment on the likelihood of exposure misclassification as a result of using a job exposure matrix.

D) The investigators could have selected newborns with another congenital malformation than NTD as controls. Please mention main advantages and disadvantages of using this strategy.

Question 3:

TABLE 1: Association between consumption of the cumulative average of cured meats and newly diagnosed chronic obstructive pulmonary disease (COPD) in former and current smokers, Health Professionals Follow-up Study, United States, 1986–1998 (n = 27,755 men)

1 Multivariate relative risks have been adjusted for age and energy intake

2 Multivariate relative risks have been adjusted for age, energy intake, smoking status (former or current smokers), pack-years, and pack-years squared.

3 Multivariate relative risks have been adjusted for age, energy intake, smoking status (former or current smokers), pack-years, pack-years squared, race/ethnicity, US region, body mass index, and physical activity.

a). Table 1 above was published in a recent AJE article.

Draw a DAG that includes high vegetable/fruit intake when estimating COPD rates for cured consumption; explain why you would or would not adjust for this variable (high vegetable/fruit diet) when estimating a relative risk in this study.

b). Please calculate crude relative risks for cured meat consumption of <1 serving per week and for >=1 serving per day of cured meat compared to never/almost never eating cured meat. Compare the crude estimates to the ones shown in table 1 and explain what you conclude with regard to confounding in this data.

c). Explain why the authors adjusted for US region in their third adjustment in table 1; i.e. argue why you would (or would NOT) consider US region to be a confounder.

d). The authors conducted the analysis in table 1 after excludingnever smokers. Another analysis was also performed including never smokers of cigarettes: this time, the cohort included 111 cases and a total of 246,822 person-years of observation. Comparing the rate for highest cured meat consumption (>=1 serving per day) to the rate among never/almost never consumers in this expanded cohort they reported a RR of 2.64 (95%CI 1.39, 5.00). Explain, why you think they performed these two different analyses (with and without never smokers), and what would you take away from this additional information.

1