Selection bias associated with contraceptive practice in time-to-pregnancy studies.

In studies of subfertility using data on time to pregnancy (the number of noncontracepting menstrual cycles required to conceive), many pregnancies are not included in analyses because time-to-pregnancy data cannot always be collected. Birth control failures are one such group (couples who are using birth control regularly or sporadically, but become pregnant anyway). The more fecund couples are more likely to have birth control failures, and consequently, more likely to be excluded from analyses. If an exposure of interest is associated with irregular use of birth control or with the choice of less effective methods of birth control, the fecundability in the exposed (based on data from couples who can provide adequate time to pregnancy data) will be underestimated. Such bias may account for the findings in the literature showing reduced fecundability among smokers. More generally, when a subset of couples who differ in their fecundability are excluded from analysis, there will be opportunity for selection bias to produce under or overestimation of the strength of an association between an exposure and fecundability. This problem affects both retrospective and prospective studies. Time-to-pregnancy studies need to be designed to minimize this potential bias and to collect data to evaluate it adequately.