Development and validation of a prediction model for gestational diabetes mellitus based on clinical characteristics and laboratory biomarkers among Chinese women.

Abstract

Background and aims: Early detection of gestational diabetes mellitus (GDM) is critical for maternal and child health. Although several prediction models exist, their complexity and reliance on less clinically accessible biomarkers have limited generalizability. This study aimed to develop and validate a clinically practical GDM prediction model.

Methods and results: Based on a retrospective cohort containing 30 480 pregnant women from China (2014-2019), three prediction models (basic, full and optimal) were developed using logistic regression to select predictors. Predictive accuracy of prediction models was evaluated by the area under receiver operating characteristic curve (AUC). The nomogram was established to predict individual probability of GDM, with decision curve analysis (DCA) assessing clinical utility. A total of 8161 (26.8 %) women were diagnosed with GDM. The optimal model, incorporating nine clinical characteristics and biochemical indicators, had a good predictive effect for GDM with AUCs of 0.74 (95 % CI: 0.74-0.75) in the training cohort and 0.74 (0.73-0.76) in the validation cohort. The performance of the optimal model was significantly greater than the basic model (AUC of 0.62) and close to the full model (AUC of 0.75). The calibration curve showed that the established nomogram had good accuracy to predict individual probability of GDM. The DCA showed that the prediction model had a positive net benefit at threshold between 0.1 and 0.8.

Conclusion: The nine-item optimal prediction model yielded high predictive accuracy, facilitating the identification of high-risk women, and the refinement of personalized diagnostic and treatment modalities.