A descriptive analysis of updated information and an evaluation of the effects of non-response in a case-control study of breast cancer and maternal serum alpha-fetoprotein

Abstract

The participants or their proxies of a case-control study of the relationship between maternal serum alpha-fetoprotein (AFP) and breast cancer (target population) were invited to complete a mail questionnaire to gather updated information about health and reproductive events. The response rate among the target population was low (40%) but was similar for cases and controls. The internal validity of inferences about the target population using only questionnaire respondents was assessed in three ways. The first compared exposure odds ratios (OR) for selected characteristics collected at baseline for the target population to the same measure of association calculated for questionnaire respondents only. The second calculated the odds of being in 'exposure' categories of selected characteristics among respondents compared to non-respondents. The third compared questionnaire responses about selected pregnancy and reproductive details to information about the same details collected at baseline. Respondents were considered to be in almost perfect agreement with baseline data when giving details about year of termination of the first full-term and the index pregnancies (k=0.93 and k=0.98), but coding inconsistencies between baseline versus questionnaire data affected the level of agreement found for most other variables. Comparing the respondents to non-respondents, respondents drank more alcohol (2.253︢.59 glasses per week vs. 1.89 3︢.20 glasses per week), were more likely to be white(82.89% vs. 62.03%) and have attended college (60.09% vs. 34.78%). Respondents compared to non-respondents were also more likely to have experienced a first full-term pregnancy at 21 years of age or older (83.70% vs. 69.86%). Among respondents being a case was also strongly associated with having a family history of breast cancer (X2,=19.25, p<0.001). The distribution of women in the high AFP category among respondents was Selection bias due to non-or differential response was evident when baseline data for respondents was compared to that of the target population. Misclassification bias was not adequately addressed due to theoretical and technical problems associated with each variable selected for evaluation. These biases should not affect the utility of the updated information obtained on the questionnaire.