Statistical analysis
We used modified Poisson regression to estimate RRs with 95% CIs between exposure groups before gestational day 238 and late-onset preeclampsia.32 Cox proportional hazard models with time since gestational day 238 were used to estimate hazard ratios (HR) for exposure after gestational day 237. We used robust standard errors to account for correlation within women who participated with >1 pregnancy in the PRIDE Study.33,34
The models were weighted using inverse probability of censoring weights and adjusted for a sufficient set of confounders. We used inverse probability of censoring weights to account for potential selection bias resulting from differential loss-to-follow-up,35 by fitting logistic regression models to predict not being lost-to-follow up using determinants of attrition. We used the models’ predicted probabilities to calculate inverse weights for loss-to-follow-up. Under the assumption that data were missing at random, we imputed missing data on confounders through multiple imputation (25 imputations; Text S1).
We conducted a number of sensitivity analyses to assess the robustness of the primary analysis. Firstly, we selected gestational hypertension as secondary outcome measure, distinguishing between exposure in gestational weeks 0-19 and after gestational week 20, as some studies indicate a protective effect of calcium supplements on this outcome as well.6 Women with chronic hypertension (N=85) were excluded from these analyses. Secondly, we used an externally validated prediction model to select women at high risk of developing preeclampsia, with a risk threshold of 3%,36,37 and replicated the main analyses in this population. Finally, we restricted the analyses to women who did not use calcium-containing supplements during pregnancy. All statistical analyses were performed using Stata/SE 16.0.