Prediction of low birth weight using machine learning-based analysis of environmental and maternal risk factors: insights from the Korean CHildren's ENvironmental health study (Ko-CHENS)

초록

Adverse birth outcomes such as low birth weight (LBW) increase the risk of metabolic disorders and hypertension later in life. Although previous studies have examined obstetric factors or environmental exposures in relation to LBW, few have assessed both simultaneously. This study employed cohort data to predict LBW using machine learning (ML) algorithms that integrate maternal characteristics and environmental factors. We analyzed nationwide prospective birth cohort data from the Korean CHildren's ENvironmental health Study (Ko-CHENS). A total of 47 variables were evaluated using nine ML models, including linear models and tree-based models. The best-performing model was selected based on the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AU-PRC). Among 4893 mother-infant pairs, 333 cases (6.8%) were identified as LBW. Random forest achieved the highest performance with an AU-ROC of 0.94 and an AU-PRC of 0.768. Feature importance analysis identified gestational age as the most influential predictor, followed by environmental factors. Additional analysis revealed trimester-specific associations between air pollution exposure and LBW risk. Third-trimester exposure to nitrogen dioxide (NO2) (odds ratio [OR]: 1.81, 95% confidence interval [CI]: 1.15-2.84) and particulate matter with an aerodynamic diameter <= 2.5 mu m (PM2.5) (OR: 1.49, 95% CI: 1.03-2.17) demonstrated statistically significant positive associations with LBW risk. This study identified significant associations between environmental exposures and LBW risk using data-driven ML analysis. Our findings underscore the role of air pollution in increasing LBW risk and may inform public health strategies to reduce LBW incidence through targeted environmental policies.

키워드