OBJECTIVES: This study aimed to develop and externally validate machine learning-based models for predicting macrosomia and spontaneous preterm birth (sPTB) using multidimensional clinical data. METHODS: This retrospective cohort study included 14 238 pregnant women who received prenatal care between 2018 and 2023. Demographic characteristics, laboratory parameters, and delivery outcomes were integrated, with Least Absolute Shrinkage and Selection Operator regression used for feature selection. Predictive models were developed using logistic regression and machine learning algorithms (Light Gradient Boosting Machine, Adaptive Boosting, and gradient boosting decision tree), and performance was evaluated using the area under the curve (AUC), Brier score, decision curve analysis, and external validation results. RESULTS: Key predictors of macrosomia included pre-pregnancy body mass index, 1-h postprandial glucose, maternal education level, and comorbidities. The logistic regression model for macrosomia achieved an AUC of 0.742 (95% CI: 0.617-0.866) in external validation. For sPTB, significant predictors were comorbidities, aspartate aminotransferase, and uric acid, with the logistic regression model yielding an AUC of 0.706 (95% CI: 0.587-0.824) in external validation. Both models showed good calibration, with Brier scores of 0.182 (macrosomia) and 0.192 (sPTB). CONCLUSION: The predictive models effectively identify pregnancies at high risk of macrosomia and sPTB, facilitating early identification and targeted intervention. External validation highlights the need for prospective, multicenter studies in diverse populations to enhance generalizability.
Chen et al. (Fri,) studied this question.