Comparison of Interpretable Machine Learning Models Using Systemic Inflammation Index to Predict Preterm Birth in...

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Inflammation Research

Background:

• Gestational diabetes mellitus (GDM) significantly increases preterm birth risk, necessitating improved prediction models.
• Current prediction methods for GDM-associated preterm birth have limited accuracy due to exclusion of key inflammatory biomarkers.
• Machine learning (ML) offers advanced pattern recognition but is underutilized for predicting preterm birth in GDM pregnancies.

Purpose of the Study:

• To develop an interpretable ML model for predicting preterm birth in GDM patients.
• To integrate systemic inflammatory indices with traditional clinical markers for enhanced prediction accuracy.
• To enable early risk stratification at GDM diagnosis for timely clinical interventions.

Main Methods:

• Retrospective analysis of 389 GDM patients, divided into training (n=272) and external validation (n=117) cohorts.
• Development and validation of ML models using systemic inflammation indices, clinical indicators, and combined features.
• Application of Shapley Additive Explanations (SHAP) for feature interpretation and sensitivity analyses for robustness.

Main Results:

• The study identified seven significant predictors, combining systemic inflammatory markers and clinical parameters.
• An extreme gradient boosting (XGBoost) model demonstrated superior predictive performance (AUC-ROC: 0.932, AUC-PRC: 0.754) compared to other algorithms.
• SHAP analysis highlighted two clinical and three inflammatory markers as the most influential predictors of preterm birth.

Conclusions:

• The developed XGBoost model effectively predicts preterm birth in GDM by integrating clinical and inflammatory markers.
• This approach enables precise risk assessment, guiding clinical management and improving outcomes for high-risk GDM pregnancies.