Development and Validation of Risk Prediction Models for Gestational Diabetes Mellitus Using Four Different Methods

Area of Science:

• Perinatal medicine
• Reproductive health
• Computational biology

Background:

• Gestational diabetes mellitus (GDM) is a common condition during pregnancy.
• GDM increases risks for both mothers and newborns.
• Early identification of GDM risk is crucial for effective management.

Purpose of the Study:

• To develop and validate early-first-trimester risk prediction models for GDM.
• To compare the performance of score-scaled, logistic regression, and machine learning models.
• To identify pregnant women at high risk for GDM for prompt intervention.

Main Methods:

• Meta-analysis of 42 studies to create a score-scaled model.
• Development and validation of logistic regression, decision tree (DT), and random forest (RF) models.
• Utilized a cohort of 1075 Chinese pregnant women for model development and validation.

Main Results:

• Score-scaled model achieved an AUC of 0.772.
• Logistic regression model showed AUCs of 0.799 (training) and 0.834 (validation).
• DT and RF models demonstrated AUCs of 0.825/0.823 (training) and 0.816/0.827 (validation), indicating good predictive performance.

Conclusions:

• Developed GDM prediction models (score-scaled, logistic regression, DT, RF) demonstrate good performance.
• These models can identify high-risk pregnancies using common clinical indicators.
• Early risk identification facilitates prompt and targeted interventions for GDM.