Machine learning model-based preterm birth prediction and clinical nomogram: A big retrospective cohort study

Area of Science:

• Reproductive Health
• Data Science in Medicine
• Clinical Prediction Modeling

Background:

• Preterm birth remains a leading cause of neonatal mortality and morbidity worldwide.
• Existing predictive models often lack robust external validation, limiting their clinical applicability.
• Early identification of at-risk pregnancies is crucial for implementing timely interventions.

Purpose of the Study:

• To develop and validate a multifactorial predictive model for preterm birth risk.
• To identify key risk factors associated with preterm birth using machine learning algorithms.
• To provide clinicians with a tool for early prevention of preterm birth.

Main Methods:

• Retrospective cohort study using National Vital Statistics System (NVSS) birth data (2022 and 2018).
• Four machine learning algorithms (logistic regression, adaptive lasso, bootstrap forest, boosted trees) were employed for feature identification.
• A logistic regression-based nomogram was constructed using consensus features and validated internally and externally.

Main Results:

• The study analyzed over 5.2 million mother-infant pairs.
• Machine learning models achieved high accuracy (AUC > 0.7) in preterm birth prediction.
• The developed nomogram demonstrated consistent performance across training, internal, and external validation sets (AUCs ~0.70).

Conclusions:

• A reliable and externally validated preterm birth prediction tool was successfully developed.
• The model effectively identified nine common risk factors for preterm birth.
• This tool addresses the need for validated prediction models to support clinical decision-making and early prevention efforts.