Predicting recurrent gestational diabetes mellitus using artificial intelligence models: a retrospective cohort...

Area of Science:

• Reproductive Medicine
• Medical Informatics
• Biostatistics

Background:

• Recurrent gestational diabetes mellitus (GDM) poses risks to maternal and fetal health.
• Early identification of women at high risk for GDM recurrence is crucial for timely intervention.
• Predictive models using early pregnancy features can aid in risk stratification.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) models for predicting recurrent GDM before 14 weeks of gestation.
• To compare the performance of various AI algorithms against traditional logistic regression for GDM recurrence prediction.
• To identify key early pregnancy features that are predictive of GDM recurrence.

Main Methods:

• A cohort of 588 women with a history of GDM in a previous pregnancy was analyzed.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for feature selection.
• Five AI models (SVM, XGB, LGB, DTC, RF) and logistic regression were constructed and compared.

Main Results:

• 55.4% of women experienced GDM recurrence in a subsequent pregnancy.
• AI models, particularly LGB, RF, and XGB, demonstrated superior predictive performance (AUROC 0.942, 0.936, 0.924) compared to logistic regression (AUROC 0.696).
• The Light Gradient Boosting (LGB) model showed exceptional accuracy, excellent calibration, and superior net benefits.

Conclusions:

• AI models, especially LGB, offer a highly accurate and reliable method for predicting recurrent GDM risk early in pregnancy.
• These predictive tools can empower healthcare providers to offer personalized advice and implement preventive strategies.
• Early prediction of GDM recurrence can significantly improve maternal and fetal well-being by enabling proactive management.