To propose three methods to estimate associations between pregnancy exposures and outcomes before term, including the association between preeclampsia and preterm fetal-infant mortality, while avoiding the selection bias found in conventional analytic designs. Population-level analysis of 1,099,839 women who delivered live or stillborn infants in Quebec hospitals from 1989 to 2012, covering nearly a quarter of Canadian births. The exposure of interest was preeclampsia at 20-29, 30-33, 34-36, and ≥37 weeks of gestation. We compared preeclamptic with non-preeclamptic pregnancies in parametric survival models with a Weibull distribution using three analytic designs: (1) fetuses-at-risk; (2) bias correction factors; and (3) analysis of pregnancies at high risk of preterm delivery. Main outcome measures were occurrence of fetal or postnatal infant death before discharge. In models affected by bias due to selection of preterm births, women with preeclampsia early in gestation had paradoxically lower risks of fetal and infant mortality than women without preeclampsia. All three analytic approaches reversed the associations, showing a harmful effect of preeclampsia before term. At 20-29 weeks, for example, preeclampsia was associated with 1.35 times the risk of infant mortality (95 % confidence interval 1.14-1.60) and 1.40 times the risk of stillbirth (95 % confidence interval 1.18-1.67) with the fetuses-at-risk approach. Paradoxically protective benefits of exposures such as preeclampsia before term are artifacts of inappropriate analytic design. Outcomes before term should be analyzed with care, using methods that address bias due to selection of preterm deliveries.
Keywords: Epidemiology; Perinatal mortality; Preeclampsia; Preterm birth; Selection bias.