A review of problems of bias and confounding in epidemiologic studies of cervical neoplasia and oral contraceptive use.

Abstract

At least 12 major epidemiologic studies on the association between oral contraceptive (OC) use and increased risk of cervical neoplasia reported inconsistent results. 3 found a significantly increased risk of cervical neoplasia in OC users, 7 found no statistically significant increase, 1 indicated either an increased risk for OC users or a decreased risk for diaphragm users, and another found that cervical dysplasia is more likely to progress to carcinoma in situ in women who continue to use OCs after the cancer diagnosis is made than in women who continue to use an IUD. A review of the methodologies of these studies revealed several reasons for the inconsistent findings: 1) the difference in Pap screening frequency between OC users and nonusers is an important source of bias which could be eliminated only by imposing a fixed screening schedule, 2) reliable information on the most important confounders is difficult to obtain, 3) extensive differential errors in the Pap test inevitably introduce misclassification bias, and 4) attempts to remedy these problems introduce new biases. A consistent pattern was nevertheless observed in the studies. Studies using incident cases yielded higher estimates of relative risk (RR) than studies using prevalent cases, and almost all studies yielded an RR estimate between 1-2 even though the RR estimate is not always statistically significant. The difficulty of separating a true elevation in risk from an elevation in risk due to bias and confounding is indicated.