Inferring effects of time-varying prenatal exposures on pregnancy loss from live-birth-identified conceptions: A simulation study Article Swipe
YOU?
·
· 2023
· Open Access
·
· DOI: https://doi.org/10.1289/isee.2023.op-069
· OA: W4390994329
BACKGROUND AND AIM: Identifying the determinants of pregnancy loss is a critical public health concern. However, it is difficult to enumerate the outcome, and so past studies have been limited to medically-identified losses or small, highly selected cohorts. Instead, we show through a simulation study of the effect of nitrogen dioxide (NO2) on pregnancy loss that researchers can use records of live births and gestational ages to identify live-birth-identified conceptions (LBICs)—the difference between the total number of conceptions and those lost for a given time window—to infer effects about pregnancy loss. METHOD: We simulated ten years of conceptions, pregnancies, losses, and births under several confounding patterns (no confounding, seasonal conceptions, temperature-influenced loss, long-term conception trends, and the previous three combined), and two NO2 effect forms (no effect and moderate effect). We used a time-series design and fit quasi-Poisson distributed lag models adjusted for season, year, and temperature. We considered two approaches to estimate the week-specific and 40-week cumulative effects of NO2 on pregnancy loss: 1) direct interpretation of estimated regression coefficients from the quasi-Poisson model and 2) g-computation to estimate the corresponding additive effects. RESULTS: Across all scenarios, our models, on average, correctly identified the critical window with appropriate coverage (range: 90% to 100%) and low percent bias (range: -0.51% to 1.5%). For example, under a moderate NO2 effect and combined confounding structure, the average bias and coverage for the additive cumulative association was 1.1% (standard deviation: 6.28%) and 96%, respectively. CONCLUSIONS: We demonstrated through simulations that our method relying on LBICs (typically available in administrative datasets) offers a viable approach to infer effects on pregnancy loss under realistic confounding scenarios. FUNDING: National Institute of Environmental Health Sciences R01 ES029943 and US Environmental Protection Agency RD-835872.
