Article Text
Abstract
Background Poor sleep is associated with increased risks of adverse pregnancy and perinatal outcomes in some studies using multivariable analyses, but these associations may be due to confounding, or biased due to the use of self-reported sleep measurements. Mendelian randomization (MR) is less vulnerable to confounding than multivariable regression and using accelerometers to objectively assess sleep will not be affected by reporting bias. Our aim was to use MR to assess the impact of four accelerometer-based sleep traits (i.e. duration, efficiency, fragmentation and chronotype) on obstetric outcomes (pregnancy loss, gestational diabetes, postnatal depression and offspring birthweight).
Methods We conducted a two-sample MR in UK Biobank (UKB). Genetic variants were selected as those that were genome-wide significant in a previous genome-wide association study completed in a sub-sample of UKB females and males (N=85,670): sleep duration (11 variants), efficiency (5 variants), fragmentation (21 variants) and chronotype (6 variants). We estimated the association of each variant with self-reported obstetric outcomes among UKB female of European descent (N=30,310–265,680 for different outcomes with smaller Ns for data available only in online follow-up). In main analyses, we used inverse variance weighting (IVW) to estimate the influence of sleep traits on each outcome. Sensitivity analyses to explore instrument validity included MR-Egger, weighted median and weighted mode estimators. All analyses were performed using R.
Results In IVW analyses, an hour increase in sleep duration was associated with 45 g (95% confidence interval (CI): -80, -10) lower birthweight, but little evidence was found for effects on any pregnancy loss (odds ratio (OR): 0.97; 95%CI: 0.86, 1.10), stillbirth (OR: 1.24; 95%CI: 0.82, 1.87), miscarriage (OR: 0.92; 95%CI: 0.80, 1.07), gestational diabetes (OR: 0.94; 95% CI: 0.35, 2.55), postnatal depression (OR: 1.18; 95% CI: 0.79, 1.77). We also found little evidence for an effect of sleep efficiency, fragmentation and chronotype on any of the obstetric outcomes considered. Results of sensitivity analyses were largely consistent with IVW analyses. Estimates for sleep duration-birthweight effect were -42 g (95% CI: -90, 5) in weighted median, -23 g (95%CI: -84, 39) in weighted mode and 18 g (95% CI: -74, 110) in MR-Egger.
Conclusion We did not find consistent evidence supporting an effect of poor sleep on obstetric outcomes. However, even with large sample sizes some of our results are imprecisely estimated and have wide CIs, and we did not have data to test the relevance of our instruments during pregnancy and to assess other important obstetric outcomes, e.g. hypertensive disorders of pregnancy.