Understanding the importance of social and environmental stressors in predicting maternal sleep quality during late pregnancy.

Poor pregnancy sleep quality has been linked to adverse pregnancy and birth outcomes, particularly in late pregnancy when sleep symptoms worsen. We aimed to identify potential important predictors of late-pregnancy sleep quality from social and environmental stressors and pregnancy health conditions, using a machine-learning approach. In the MADRES cohort, 687 mothers reported late-pregnancy sleep quality (Jenkins Sleep Scale [JSS]), with 26.3% reporting poor sleep quality (JSS≥12). Gradient boosting models were fitted using 59 predictors from 11 predictor groups, or each group separately, with relative influence analysis to identify the top predictors. Model stability of selected hyperparameters was assessed across 30 random train-test splits (80/20; n = 550/137), with 3-fold cross-validation; the optimal model achieved a cross-validation AUC of 0.647 and a test AUC of 0.704. Mental health during late pregnancy was the single greatest contributor to poor sleep quality, followed by sociodemographic and noise indicators (range of AUC: 0.53-0.72). Additional environmental exposures were also identified as top 10 predictors (e.g., ozone and organophosphate esters). We revealed that mental health interacted with environmental factors; mothers had an even higher risk of poor sleep quality if they had both higher-than-average mental health scores and environmental exposure. Late-pregnancy sleep quality was influenced not only by traditionally identified personal-level factors but also by a complex interplay with environmental exposures such as noise, air pollution, and chemical exposures.