Interpretable machine learning model for predicting low birth weight in singleton pregnancies: a retrospective cohort study

Background Low birth weight (LBW), defined as a newborn weighing less than 2500 grams, is an increasingly significant public health concern. Exploring the risk and protective factors for LBW is getting more and more important. This study aimed to utilize predictive models to identify the most critical factors associated with LBW in singleton pregnancies. Methods : A retrospective cohort study was conducted at the Binzhou Medical University Hospital, China, from 2022 to 2023. Singleton pregnancies with gestational age exceeding 27 weeks were included, while multiple pregnancies and fetal anomalies were excluded. Logistic regression (LR) and extreme gradient boosting (XGBoost) algorithms were employed to predict LBW and normal birth weight (NBW) outcomes. The LR model was interpreted using on odds ratio analysis and nomograms, whereas the XGBoost model was elucidated through Shapley Additive Explanations (SHAP) values to determine the factors most strongly associated with LBW. Results : In this cohort of 10,227 deliveries, 237 cases were classified as LBW. The XGBoost model demonstrated superior performance in predicting LBW, achieving an AUROC of 0.797. Both LR and XGBoost model identified maternal age, gestational age, BMI, hypertensive disorders of pregnancy (HDP),fetal distress as the critical factor associated with LBW. Additionally, a follow-up study of LBW identified that LBW infants encounter significant health challenges, including a high rate of hospitalization and the complex neonatal complications included congenital anomalies, NRDS and neonatal hyperbilirubinemia. Conclusion: This study demonstrated that the XGBoost model showed promising predictive accuracy for LBW deliveries. Pregnant women with a gestational age of less than 37 weeks, gestational BMI below 18 kg/m², maternal age younger than 25 years, or maternal comorbidities such as HDP or fetal distress are at an increased risk of delivering LBW infants. These findings highlight potential contributors to LBW disparities in China and underscore the utility of ML in maternal health research.