A machine learning-based risk prediction model for early preterm birth: development and prospective validation

Area of Science:

• Obstetrics and Gynecology
• Machine Learning in Healthcare
• Perinatal Medicine

Background:

• Preterm birth (PTB) is a major global cause of neonatal morbidity and mortality.
• Existing risk prediction tools for PTB during mid-pregnancy are limited in accuracy and clinical applicability.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting PTB risk using mid-pregnancy clinical and laboratory data.
• To assess the model's performance, interpretability, and clinical utility for risk stratification.

Main Methods:

• A Light Gradient Boosting Machine (LightGBM) model was developed using data from a derivation cohort.
• Predictor selection involved LASSO regularization; model performance was evaluated in a prospective validation cohort.
• SHapley Additive exPlanations (SHAP) were used for model interpretability.

Main Results:

• The LightGBM model achieved an AUC of 0.842 in the derivation cohort and showed stable discrimination and good calibration in the validation cohort.
• A risk stratification approach clearly separated PTB incidence among high-, intermediate-, and low-risk groups.
• SHAP analysis identified inflammatory markers, liver indices, pregnancy complications, and nutritional indicators as key predictors.

Conclusions:

• A validated ML model effectively predicts mid-pregnancy PTB risk.
• The model offers stable performance, good calibration, meaningful risk stratification, and transparent interpretability.
• This tool shows potential as a screening-oriented decision-support system for enhanced antenatal surveillance.