“Confounding” and “Effect Modification”.

FPH 7240 – Introduction to Epidemiology

Fall 2018

Assignment 7 – Bias

Due: October 24, 2019 at noon

Readings: This homework is based on the “Bias” chapters in Aschengrau & Seague. For next week, please read chapters 11, “Confounding” and 13 “Effect Modification”.

A recent case-control study was designed to identify family, medical, and behavioral risk factors for ovarian cancer among African-American women. Cases were enrolled using cancer registries in several states, and controls matched by age, sex, race, and state of residence, were selected from the population using random-digit dialing including both land lines and cell phone numbers. A total of 391 cases and 640 controls provided survey data; however, not all answers were complete for all of the women.

1. Family history is one risk factor the investigators were interested in studying in relation to ovarian cancer. Among these women, 382 cases and 621 controls provided information on family history of ovarian and/or breast cancer. Suppose that in reality, 136 cases and 181 controls had a family member with a history of breast or ovarian cancer.

a) What is the “true” odds ratio of ovarian cancer associated with having a family member with a history of breast and/or ovarian cancer? Please calculate the OR and put your result in a sentence.

b) Unfortunately, not all of the women correctly remembered their female relatives’ history of breast or ovarian cancer, but cases were more accurate in remembering their family history than controls. Only 90% of cases and 75% of controls with a family history of one of those diseases reported that there was a history of one of those diseases in their family. Given this information, what is the odds ratio the investigators in this study observed? Please report the OR and state the observed result in a sentence.

c) What kind of bias does this represent?

d) Is the bias toward or away from the null?

e) In one or two sentences, please describe the observed OR, the true OR, and the type and direction of the bias present.

2. In the same study of risk factors described above, the investigators also were interested in whether age at menarche (first menstruation) was associated with ovarian cancer. All 391 cases and 640 controls provided information about age at menarche, and of those, 72% of cases and 413 controls were actually at least 12 years of age at menarche.

a) What is the “true” OR of being at least 12 years of age at menarche compared with being <12?

b) Among women whose actual age at menarche was younger than 12 years, 20% reported that they were 12 or older at menarche. Given this information, what would be the observed odds ratio of ovarian cancer associated with being at least 12 years old at menarche?

c) What kind of bias is present in 2b?

d) In a sentence or two, please indicate the observed and true ORs, and they type and direction of the bias present in 2a-2c.

e) What would the observed OR of age at menarche and ovarian cancer be if 90% of cases and 80% of controls who were actually younger than 12 at menarche reported their age at menarche accurately?

f) Please describe the results of 2e in a sentence including the type and direction of the bias present, and how it compares with your results in 2b.

3. In the same case-control study looking at risk factors for ovarian cancer, the investigators also wanted to examine whether parity (whether the women had children, and how many children they had) was associated with ovarian cancer. Among the 391 cases and 640 controls, 311 cases and 543 controls had at least one child.

a) What is the OR of ovarian cancer associated with having at least one child compared with having no children? Please describe the observed association in a sentence.

b) When recruiting women into this study, 203 women diagnosed with ovarian cancer who were approached and asked to participate did not respond. In the case population in this study, it was common for women diagnosed with ovarian cancer to move in with relatives, including their children, during treatment. Suppose that among those 203 women with ovarian cancer who chose not to participate, 65% were living with children who encouraged them to participate, and those women finally got tired of their children telling them what to do, and they agreed to be in the study. What would the observed OR be after these women joined the study?

c) What kind of bias would be introduced due to the circumstances described in 3b?

d) In one or two sentences, please describe the type and direction of the bias in 3b, and the biased result compared with the “true” association.