Exposures
Data on medication exposures were obtained from the three prenatal questionnaires and the first postpartum questionnaire. For a wide range of indications, including heartburn and acid reflux, women reported the name of the medication taken, time period of use, frequency of use, and quantity taken. Missing data on the duration of treatment, frequency, or quantity were replaced with the median cohort value for that variable specific for the medication of interest. Medications were coded according to the Anatomical Therapeutic Chemical (ATC) Classification System.21 Exposure to calcium-based antacids was defined as reported use of antacids (ATC code A02) containing any amount of calcium carbonate. The dose of calcium on each day was calculated by multiplying the amount of calcium carbonate in mg per medication unit by the number of units taken per day. The doses for multiple calcium-based antacids per day were summed. PPI exposure was defined as report of medication belonging to ATC group A02BC. Dosage was converted to Defined Daily Dose (DDD) per day. Following, daily doses were expressed as the average daily dose (milligrams per day for calcium-based antacids and DDDs per day for PPIs) per week. For calcium-based antacids, we considered a daily dose of ≥1 g calcium as high,6while a high dose for PPIs was >1 DDD. The sensitivity of the questionnaires was 0.89 (95% CI 0.86-0.93) for gastroesophageal reflux medication.22
We adhered to a recently published guidance on longitudinal methods for modelling medication exposures in pregnancy.23 We evaluated exposure binarily (any versus none) in the first 237 days of pregnancy, further subdivided into early pregnancy (gestational weeks 0-16) and mid-pregnancy (gestational weeks 17-33), reflecting the temporality of data collection in both sources (in weeks 17 and 34). Furthermore, we clustered women with similar individual trajectories of calcium dose or DDDs of PPIs in gestational weeks 0-33 usingk -means clustering with the R statistical software package “kml ”.24 This unsupervised learning approach makes no a priori assumptions about trajectory shape or membership.25 We considered daily and cumulative dose in each gestational week allowing for k = 2 to k = 8 clusters. We selected the number of clusters based on (a) optimization of three statistical quality criteria,25 (b) clinical relevance of the clusters, and (c) at least 100 pregnancies per cluster.
K -means clustering requires all pregnancies to have the same gestational length to avoid including exposure after the diagnosis of preeclampsia and on postpartum days.26 Moreover, immortal time bias could be introduced if we would apply the binary exposure categories after gestational day 237, as pregnancies without preeclampsia and pregnancies with longer gestations have more opportunity for exposure.27,28 Therefore, we modelled time-dependent changes in dose on each gestational day between 238 and the end of follow-up, defined as diagnosis of preeclampsia or delivery, allowing for daily changes in use (none/any) and dose (none/low/high). We determined exposure time by dividing the average number of person-weeks in each exposure category by the number of women with any exposure to each level after day 237. Women could contribute person-weeks to multiple dose levels.29