Study design and population
We conducted an individual-level register- and population-based cohort study using data from the Danish Civil Registration System, the Danish Medical Birth Registry, and the Danish National Prescription Register [27,28,29]. All women with a registered singleton and first recorded pregnancies in Denmark between 1997 and 2017 were identified from the Danish Medical Birth Registry (n = 707,467), regardless of their previous pregnancies before the study period. After excluding 54,657 births with coding errors on gestational length, maternal drinking water estimates from private wells, or missing drinking NO3− data, our study population included a total of 652,810 mother-child dyads.
Data sources
The registers utilized in this study offer an exceptional resource with their depth and breadth of data collected. The unique civil personal registration (CPR) number, which has been assigned to nearly all (> 99%) Danish residents since 1968, was used as key identifier to link data between the national registries. In addition, the Danish Civil Registration System holds weekly updated information on date of birth, sex, and residential history [27].
The Danish Medical Birth Registry has information on all births in Denmark from 1973 onwards, including the CPR number of the child, the mother, and the registered father. It also holds information on gestational age based mainly on ultrasound measures and occasionally on last menstrual period, height and weight at birth, birth date, birth outcome (liveborn or death and cause of death), sex, parental age, and number of previous births [28].
The Danish National Prescription Register holds information on redemption of prescribed medication. This register has individual-level data on all prescribed drugs dispensed at any pharmacy in Denmark since 1995 [29]. Pharmacies are required to register all redeemed prescriptions, which are coupled with the reimbursement of expenses from the state. This ensures highly accurate prescription data and completeness has been estimated to be > 97% [30].
Nitrosatable drug assessment
Detailed procedures for classifying as nitrosatable and for categorizing them based on their functional groups and indications have been discussed in detail elsewhere [1]. We used nitrosatable medicinal compounds lists from published literature to classify nitrosatability, and we included the medication reported to be used between 1997 and 2017 [1, 4, 31]. We included only the medication reported to be used in our population in the study period. All drugs were categorized on the basis of the presence of secondary amine, tertiary amine, and amide functional groups. These categories are not mutually exclusive.
We identified the Anatomical Therapeutic Chemical codes for the categorized drugs and matched these with codes in the Danish National Prescription Register, which also contained information on the date of redemption of the prescription (see Supplementary Table S1 for a complete list of included nitrosatable drugs). All analyses are based on the first nitrosatable prescription drug redeemed from the estimated first day in the last menstrual period (week 0) until the end of week 22 of pregnancy. Women with no redeemed nitrosatable prescription drug during this period were defined as the unexposed comparison.
Drinking water NO3− assessment
The Danish national geodatabase Jupiter holds drinking water monitoring data [32]. The maternal residential drinking water NO3− estimates build on the approach developed and described by Schullehner et al. [33]. In short, annual drinking water production volume weighted average NO3− concentrations were calculated for public water supply areas and spatio-temporally linked with the geocoded residential history of every person registered in the Danish Civil Registration System. For this study, 107,821 NO3− samples in 3635 public waterworks between 1993 and 2017 were extracted. NO3− levels were imputed by interpolation for missing years, if a NO3− sample was available within 3 years, otherwise the year was assigned as missing NO3−. Individual exposure for each woman was computed as the time-weighted average concentration during pregnancy taking into account changes in residence during pregnancy [34]. The average NO3− concentration level (mg/L) for each woman was calculated in the period from the estimated first day in the last menstrual period (week 0) until the end of week 22. As private wells are not monitored to the same degree as public supplies, users of private wells (approximately 3% of Danish residents) were excluded from the study population.
Stillbirth assessment
Before 1 January 2004, stillbirth in Denmark was defined as the birth of a non-vital fetus after ≥28 completed weeks of pregnancy. In 2004 the cut point changed to 22 completed weeks, as in most other countries [35]. To harmonize the study population, we extracted all registered stillbirths born at 22 weeks of pregnancy or later between 1997 and 2017. Pregnancy period was based on the child’s date of birth and gestational age at birth in days.
Statistical analyses
We investigated, if drinking water NO3− in the population was associated with stillbirth by using Cox proportional hazards model with gestational age (days) as the underlying time scale. We estimated crude and adjusted hazard ratios (aHRs) and corresponding 95% confidence intervals (CIs) for stillbirth in relation to five categories of residential drinking water NO3− concentrations defined a priori according to those previously described cut-off in Coffman et al. 2021 [25]. The first category consists of exposures below the uppermost limit of detection at 1 mg/L. The four other categories were categorized as following: > 1- ≤ 2 mg/L, > 2- ≤ 5 mg/L, > 5- ≤ 25 mg/L and > 25 mg/L.
We also examined whether the association between any nitrosatable prescription drug intake and stillbirth was modified on the multiplicative scale by drinking water NO3− stratified into the five categories of NO3− concentrations. In a secondary analysis, secondary amine, tertiary amine, and amide intake were also stratified into the five categories of NO3− concentrations to further examine effect medication by drinking water nitrate in these functional groups.
Adjusted analyses included the following covariates identified a priori using existing literature and directed acyclic graphs and obtained from the Danish Medical Birth Registry: parity (continuous), urbanicity of maternal address at birth (five categories), maternal age at delivery (continuous), smoking during pregnancy (yes/no), pregestational BMI (continuous), and highest level of maternal education (six categories) and maternal occupation (seven categories), which were based on the International Standard Class of Occupation and Education codes (ISCO-88 and ISCED) obtained from Statistics Denmark [15, 19, 36]. The proportional hazard assumption and the linearity of covariates were accepted. All analyses were performed using Stata, version 15.0 (StataCorpLP, College Station, TX, USA) and R 4.0 (R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria).