Machine learning-based prediction algorithm of spontaneous preterm birth using multi-source data.

Area of Science:

• Obstetrics and Gynecology
• Data Science in Healthcare
• Environmental Health

Background:

• Spontaneous preterm birth (sPTB) presents unclear etiology and significant neonatal risks.
• Early prediction of sPTB is crucial for timely interventions and improved outcomes.
• This study integrates electronic health records (EHR) and environmental factors for sPTB prediction.

Purpose of the Study:

• To construct and evaluate machine learning (ML) models for predicting spontaneous preterm birth (sPTB).
• To identify key predictors of sPTB by integrating multi-source data.
• To assess the clinical utility of ML models in sPTB prediction.

Main Methods:

• A retrospective cohort study of 54,132 singleton pregnancies was conducted.
• Multi-source data including EHR, air pollution, meteorological, and greenness factors were collected (82 predictors).
• Extreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), and logistic regression (LR) models were employed, with performance assessed by AUROC and AUPRC. SHAP values were used for feature importance.

Main Results:

• The XGBoost model achieved the highest performance with an AUROC of 0.926 and AUPRC of 0.502.
• Significant predictors identified include eosinophils percentage, albumin, uric acid, amniotic fluid pocket, and sulfur dioxide exposure.
• The study highlights the predictive power of integrating clinical and environmental data.

Conclusions:

• Combining EHR data, environmental factors, and ML methods enables highly accurate sPTB prediction.
• The developed models demonstrate strong discriminatory power for sPTB.
• Further refinement is needed to enhance prediction precision for clinical application.