Machine Learning for Predicting the Transition From Gestational Diabetes to Type 2 Diabetes: A Systematic Review

Area of Science:

• Reproductive Endocrinology and Metabolism
• Computational Biology and Machine Learning

Background:

• Gestational diabetes mellitus (GDM) is a significant risk factor for postpartum type 2 diabetes (T2D).
• Early identification of high-risk women is crucial for targeted interventions and improved maternal health outcomes.
• Machine learning (ML) offers potential for developing predictive models for T2D transition post-GDM.

Purpose of the Study:

• To systematically review and evaluate the performance, predictive features, and methodological quality of ML models for predicting T2D risk in women with a history of GDM.
• To identify key features and assess the robustness of validation methods in existing ML models.

Main Methods:

• A comprehensive systematic literature search was performed across major scientific databases (PubMed, Scopus, IEEE Xplore, Web of Science).
• 13 studies were included after screening 178 records, with data extraction on study characteristics, ML algorithms, predictive features, performance, and validation.
• Risk of bias was assessed using the PROBAST tool.

Main Results:

• ML models demonstrated variable performance, with Area Under the Curve (AUC) ranging from 0.72 to 0.92.
• Models incorporating omics data showed superior performance compared to clinical data-only models.
• Key predictive features included age, body mass index (BMI), glycemic measures, and pregnancy-specific factors; however, only 38% of studies used robust external validation.

Conclusions:

• ML models, especially those integrating omics data, hold significant potential for predicting T2D risk post-GDM.
• Heterogeneity in validation methods and limited external validation necessitate standardized reporting and larger, diverse cohorts.
• Future research should focus on developing reproducible and generalizable ML models for personalized T2D prevention strategies.