Evaluation of potential sources of bias in a genetic epidemiologic study of breast cancer

Systematic errors, or bias, can arise at several stages of a study, including selection of subjects, measurement of exposure and disease, and data analysis. Little attention appears to have been paid to potential sources of bias in genetic epidemiologic studies, despite the fact that the study of units (families) and members (relatives) within those units produces unique opportunities for bias to be introduced. The ability to evaluate whether selection bias has occurred is rare. In 1944, a case control family study of breast cancer was initiated at the Dight Institute for Human Genetics at the University of Minnesota. A follow-up study of these 544 families is currently being conducted on sisters, daughters, nieces, and granddaughters of the probands, and a control group of women who are spouses of male first- and second-degree relatives. Updated data are collected on females who are 18 years or older through telephone interviews and questionnaires. The availability of detailed family history information on these families at baseline provided an opportunity to evaluate several potential sources of bias. Analyses were performed to determine if families lost to follow-up differed from those who were successfully located or excluded (ineligible), and whether participation rates within a family differed by relationship to proband, age, and family history of cancer. The latter participation rates for individuals were examined with respect to a telephone interview, a mailed questionnaire, and screening mammography. There were no statistically significant differences in the cancer histories of families that were excluded, those that were lost to follow-up, and those that participated. Within families, degree of relationship to the breast cancer proband was significantly associated with age-adjusted participation rates on mailed questionnaires (P < 0.005) and mammography (P < 0.005), but not telephone interviews (P = 0.29). After adjustment for age, marry-ins with a family history of breast cancer were not significantly more likely than marry ins without a family history to undergo mammography (P = 0.1) or return mailed questionnaires (P = 0.74). Although non-participation is a potentially serious source of bias to a genetic epidemiologic study and the effect of variable participation rates should be explored when investigating the genetic component of a disease, it does not appear to be a problem for this particular study.