Construction of a predictive model for gestational diabetes based on routine clinical indicators in early pregnancy: a three-hospital retrospective development study in Urumqi, China

Gestational diabetes mellitus (GDM) is a common metabolic disorder during pregnancy and is associated with adverse maternal and neonatal outcomes. Early identification of high-risk individuals is crucial for timely intervention. This retrospective study was conducted in three hospitals located in the same area of China (Urumqi, Xinjiang), covering 1,391 pregnant women from these three hospitals during the period from January 2024 to June 2025. Data were collected at the first prenatal visit (gestational age <14 weeks). Nine machine learning algorithms were developed and compared. The performance of the models was evaluated through multiple indicators, including the area under the receiver operating characteristic curve (AUC) and the F 1 -score, etc. Calibration curves were plotted to evaluate the consistency between predicted probabilities and observed outcomes, and decision curve analysis (DCA) was performed to assess the clinical net benefit of each model. Model interpretability was achieved using Shapley Additive exPlanations (SHAP), and an online prediction tool was constructed using the R Shiny framework. Among the nine models, the adaptive boosting (AdaBoost) performed well, with an AUC of 0.744 (95% percentile interval: 0.703–0.762), a sensitivity of 0.790, and an F 1 -score of 0.511. Calibration curves and decision curve analysis indicated acceptable calibration and favorable clinical net benefit. The SHAP analysis indicate that pre-pregnancy body mass index (BMI), fasting plasma glucose (FPG), creatinine (Cr), white blood cell count (WBC), and the triglyceride-glucose (TyG) index were important predictors. To facilitate clinical application, an online interactive prediction tool was developed ( https://earlygdm.shinyapps.io/workrun28/ ). The AdaBoost model demonstrated moderate discriminative ability (AUC 0.744) for early GDM risk stratification, with acceptable calibration and clinical net benefit. The results indicate that this model may have potential value as an auxiliary tool for risk screening. The online prediction tool based on this model is helpful for individualized risk assessment, but external validation is required before any clinical application.