Machine learning prediction model based on first-trimester TyG index for gestational diabetes mellitus in advanced maternal age: a retrospective cohort study

Emerging evidence indicates a correlation between the triglyceride-glucose (TyG) index and gestational diabetes mellitus (GDM). However, its predictive efficacy for GDM in pregnancies with advanced maternal age has not been conclusively established. This study sought to assess the predictive utility of the TyG index for GDM specifically among pregnant women of advanced maternal age. This retrospective cohort study included 635 advanced maternal age pregnant women (164 diagnosed with GDM). First-trimester clinical and biochemical indicators were collected, and a machine learning method was employed to construct a prediction model to evaluate the role of the TyG index in GDM risk prediction. The Gradient Boosting model incorporated seven key predictors: TyG index, maternal age, pre-pregnancy body mass index, platelet count, uric acid, low-density lipoprotein, and high-density lipoprotein. The optimized model demonstrated exceptional discriminative performance, achieving an AUC of 0.963 and a prediction accuracy of 0.901. DCA confirmed that the model provided significant net clinical benefit for identifying high-risk GDM cases in advanced maternal age pregnancies. It demonstrated favorable clinical application potential. The first-trimester TyG index exhibits robust predictive capability for GDM onset in pregnant women of advanced maternal age, underscoring its potential as a valuable clinical biomarker.