Selection bias in Teratology Information Service pregnancy outcome studies

Abstract

Pregnancy outcome studies conducted through Teratology Information Services (TIS) rely on volunteer subjects. If these subjects tend to have different risk profiles than the population from which they are drawn, the results of TIS studies may have limited generalizability. We selected all subjects who enrolled in the California Teratogen Information Service (CTIS) pregnancy outcome study for prenatal exposure to carbamazepine or valproic acid between 1990 and 1997 and who received prenatal care through Kaiser Permanente of Southern California (n = 13). We compared these subjects to Kaiser patients identified through the Maternal Serum Alpha Fetoprotein Program with exposure to carbamazepine or valproic acid but who had not enrolled in the CTIS project. The controls were matched by Kaiser location and pregnancy year using a 2:1 ratio (n = 26). Medical records were reviewed and the prevalence of 14 pregnancy risk factors was compared between the two groups. There were no significant differences between the groups on any one risk factor; however, a notably higher proportion of women who did not enroll in the CTIS study used tobacco or had a positive family history of congenital anomalies. Although the sample was small, and results may not apply to other exposures in different health care environments, these data provide some evidence that women who enroll in TIS pregnancy outcome studies do not have a substantially different pregnancy risk profile than women who do not. Efforts to address possible selection bias should be incorporated in future TIS study design.