A search for recall bias in a case-control study of diet and breast cancer.

BACKGROUND In retrospective studies of dietary habits and breast cancer risk, recall bias is a concern since diet has been publicized as a cause of breast cancer. METHODS In a case-control study of diet and breast cancer risk nested within a cohort of women screened with mammography, we contrasted answers to a retrospective dietary interview with answers to a dietary questionnaire which was filled out before any diagnostic procedures for breast cancer were undertaken. The source population was all women aged 40-74 in two counties in Sweden invited to mammographic screening and asked to fill out a questionnaire before the screening. Cases and controls were subsequently defined -- matched on age, county of residence, and time of mammography -- and approached for an interview. RESULTS In all, 265 cases and 431 controls participated in the study. Means of frequencies differed between the agreement in the questionnaire's and the interview's classifications of study subjects into quartiles of monthly intake varied between 31 percent and 57 percent. Kappa statistics in all food groups were below 0.41. In a regression analysis, case subjects with low responses on the questionnaire about intake of meat, snacks, and coffee and tea gave higher responses on interview than did controls who had low questionnaire responses for these food groups. The reverse was also true: cases' responses that were high on the questionnaire were lower on interview for these food groups than were controls' responses. CONCLUSIONS We found few signs of recall bias, and the few indications of a differential misclassification that we found were not in food groups that have been publicly discussed as causes of breast cancer.