Early risk prediction of gestational diabetes using routine antenatal care clinical data using machine learning

Abstract

Gestational diabetes mellitus (GDM) is a major contributor to adverse maternal and neonatal outcomes, particularly in low-resource settings where timely diagnostic testing is not always feasible. Early identification of high-risk pregnancies using routinely collected antenatal information could support targeted screening and preventive care. This study evaluated the feasibility of early GDM risk prediction using machine learning models trained on routine antenatal care data from 3525 pregnancies in Uganda (2153 non-GDM and 1372 GDM cases). Logistic regression, Random Forest, XGBoost, and a soft-voting ensemble were developed and assessed using a held-out test set. Model performance was evaluated using receiver operating characteristic area under the curve (ROC AUC), precision–recall AUC (PR AUC), accuracy, precision, recall, and F1-score, and interpretability was examined using SHapley Additive exPlanations (SHAP). Tree-based models demonstrated strong discrimination, with Random Forest and XGBoost achieving ROC AUC values of 0.997 and PR AUC values above 0.995, while logistic regression achieved ROC AUC of 0.982. Random Forest achieved high sensitivity (recall = 0.994), missing only two GDM cases in the test set. SHAP analysis identified body mass index, high-density lipoprotein cholesterol, blood pressure, and prior metabolic history as influential predictors, with feature effects consistent with established clinical knowledge. The proposed approach is intended as an early risk-stratification and triage support tool rather than a diagnostic system. All predictors were available prior to confirmatory glucose testing, and model outputs are designed to prioritise women for further evaluation and follow-up. Although performance estimates reflect internal validation within a referral-hospital cohort, the findings demonstrate that routinely collected antenatal variables contain clinically meaningful predictive information. With external and prospective validation, interpretable machine learning models may support more efficient screening and monitoring of gestational diabetes in resource-constrained antenatal care settings.

CrossRef