Early identification of women at risk of GDM may allow timely intervention and reduce adverse pregnancy outcomes. A study published in Diabetes Research and Clinical Practice developed and validated prediction models using routinely collected early-pregnancy clinical data.

The derivation dataset included 20,435 pregnant women enrolled at Beijing Obstetrics and Gynecology Hospital. Model performance was evaluated in a temporal validation cohort of 1,997 women and an external validation cohort of 100 women from additional subcenters. Two modeling approaches were assessed: logistic regression and XGBoost.

In temporal validation, both models demonstrated comparable discrimination. Logistic regression achieved an AUC of 0.738 (95% CI, 0.707 to 0.771), while XGBoost achieved an AUC of 0.737 (95% CI, 0.706 to 0.767).These findings indicate similar internal and temporal performance.

Differences emerged in external validation. Logistic regression showed an AUC of 0.674 (95% CI, 0.440 to 0.879), whereas XGBoost demonstrated an AUC of 0.737 (95% CI, 0.510 to 0.929). These results suggest that while both approaches perform adequately in internal settings, XGBoost shows greater robustness when applied to external populations.