Sources of bias in studies of time to pregnancy.

Among sexually active couples who are not using contraception, there is considerable heterogeneity in fertility, even among those who eventually achieve pregnancy. The number of menstrual cycles required, measured in integer time, is greatly overdispersed compared to the geometric, suggesting that the per cycle probability of conception varies considerably among couples. Some of this variability may reflect the effect of reproductive toxicants on fertility, and studies of time to pregnancy can be useful in identifying such toxic effects. We describe models for analysing time-to-pregnancy data, and discuss seven sources of bias that can lead the reproductive epidemiologist to spurious conclusions. Certain analytic and design strategies can help protect against some of the pitfalls.