Public Health Epidemiology STATS HW (Confounding and Effect Modification)

    

Assignment: Problem Set: Confounding and Effect Modification

Public Health Epidemiology: Decoding the Science

DUE 1/31/18 8 P.M EST BE ON TIME..

This Assignment will allow you to practice individually and increase your skills assessing confounding and effect modification. Your goal for this Assignment is to compute and interpret the effects of confounding and effect modification.

Instead, copy and paste the problem set into a Word (or equivalent) document and complete all problems, showing your work. . Be sure that your completed problem set includes your responses to all questions and shows the calculations and/or reasoning to justify all answers.

NOTE:PREVIOUS WORK WAS DONE BY ANOTHER TUTOR WITH MOSTLY WRONG ANSWERS. I’VE ATTACHED REMARKS FROM PROFESSOR HELPING WITH ANSWERS. PLEASE READ CAREFULLY

1) A study followed 900,000 US adults from 1992 to 2008. At baseline, all participants were screened and determined to be cancer free and their body mass index (BMI) was calculated. Body mass index is a measure of obesity that is calculated using a person’s height and weight. Subjects were separated into the following groups according to their BMI: (a) normal weight, (b) slightly overweight, (c) moderately overweight and (d) greatly overweight. 57,145 deaths from cancer occurred in the population during the follow-up period. . The following results were seen for men and women when the heaviest members of the cohort (greatly overweight) were compared to those with normal weight:

Men: Risk ratio of cancer death = 1.5, 95% confidence interval = 1.1–2.1

Women: Risk ratio of cancer death = 1.6, 95% confidence interval = 1.4–1.9

State in words your interpretation of the risk ratio given for the men.State in words your interpretation of the risk ratio given for the women.Are these results confounded by gender?The authors stated that they controlled for confounding      many risk factors using a multivariate analysis. State an alternative      method that the authors could have used to control for confounding in the design or analysis. In addition, name two confounding variables that you think      should be controlled using this method.

2) A cohort study was undertaken to examine the association between high lipid level and coronary heart disease (CHD). Participants were classified as having either a high lipid level (exposed) or a low or normal lipid level (unexposed). Because age is associated with both lipid level and risk of heart disease, age was considered a potential confounder or effect modifier and the age of each subject was recorded. The following data describes the study participants: Overall, there were 11,000 young participants and 9,000 old participants. Of the 4,000 young participants with high lipid levels, 20 of them developed CHD. Of the 6,000 old participants with high lipid levels, 200 of them developed CHD. In the unexposed, 18 young and 65 old participants developed CHD.

Construct the appropriate two by two tables using the      data given above. Be sure to label the cells and margins.Calculate the appropriate crude ratio measure of association combining the data for young and old individuals.Now, perform a stratified analysis and calculate the appropriate stratum-specific ratio measures of association. What are they?Do the data provide evidence of effect measure      modification on the ratio scale? Justify your answer.

3) A study used self-administered mail questionnaires to gather data on height and weight in order to calculate a measure of obesity. Which of the following types of problems were likely avoided by this method of data collection, and why?Interviewer bias,Exposure misclassification, Confounding,Selection bias,Loss-to-follow-up

1 PREVIOUS WORK THAT WAS SUBMITTED THAT WAS WRONG,ADDED PROFESSOR REMARKS HELPING WITH ANSWERS

(PROFESSOR’S REMARKS)  For the first Question, I don’t know where it came from. If you could, copy it and send it to me. For Q

2

, the tables are in error and the calculations are incorrect. The answer is as follows: a. Construct the appropriate two by two tables using the data given above. Be sure to label the cells and margins. b. Calculate the appropriate crude ratio measure of association combining the data for young and old individuals. c. Now, perform a stratified analysis and calculate the appropriate stratum-specific ratio measures of association. What are they? d. Do the data provide evidence of effect measure modification on the ratio scale? Justify your answer. a. Total: CHD No CHD Total High Yes 220 9780 10000 Lipid Level No 8

3

9917 10000 Total 303 19697 20000 Young Old CHD No CHD CHD No CHD High Yes 20 3980 4000 Yes 200 5800 6000 Lipid Level No 18 6982 7000 No 65 2935 3000 38 10962 11000 265 8735 9000 b. Risk Ratio = 2.7 c. Risk Ratios= 2.0 for young people and 1.5 for old people For Question 3, Interviewer bias is the only on avoided.

A. Temporality: This states that the effect must follow the cause, as well if there is delay amongst the cause and effect, then this delay also must show up, i.e. the effect must occur after that delay.

Among the statements mentioned above, the first one provides evidence for temporality: “the main strength in this study involve that the overweight condition was established prior to development of cancer.”

B. Biological Gradient or dose-response: This guideline is based on the exposure and its effect. It implies that greater exposure will lead to greater effect. However in some case inverse relation may develop too, i.e. greater exposure leads to lower degree of effect.

The second statement provides the evidence for this guideline, stating that increased BMI (increased exposure) led to increased cancer mortality (increased degree of effect). “The study discovered that morality related to cancer increased at the same rate as BMI increased. Moreover, comparing to normal weight men, the risks for slightly overweight men as well as moderately overweight men and also greatly overweight men were 1.2, 1.3, and 1.5.

C. Consistency: This would be supported by the consistent findings (results) by different studies conducted by different individuals in different place with various samples. This increases the possibility of the effect.

The thirst statement provides evidence for consistency, while asserting that fact that many epidemiological studies (different studies) conducted in US and abroad (different place)

have found similar results (consistent results). “Many epidemiological researches carried out in the United States as well abroad have also discovered that the risk of cancer mortality is mainly increase by high body mass index.”

2

High Lipid Level

Developed

Not Developed

Young

4000

7000

Old

6000

3000

CHD

Developed

Not Developed

Young

20

3980

Old

200

5800

Crude ratio for High Lipid level is 0.55 and crude ratio for CHD is 0.15

Hence the High Lipid Level has higher risk within its group and CHD has lower risk within its group.

3

Interviewer bias : No. Since questioners are being mailed. Not an interview.

Exposure misclassification or Information bias is the error arising from measurement error. Yes, it can happen. In fact for different groups of people this error can be different .

Confounding : No . Confounding is a state whereby the effect between an outcome as well as exposure is biased by the occurrence of additional variable. But here , no other variable can affect it.
Selection bias : YES. Exposure group to study should be selected randomly. And from every age, sex, financial group , ethnicity the chance of a sample drawn should be equal.

Loss-to-follow-up . Not really. If it’s just the 1st time gathering data only , then ‘no’.