Leveraging administrative health records and machine learning for population-level prediction of preterm birth.

Area of Science:

• Reproductive Health
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Preterm birth (delivery before 37 weeks) is a leading cause of neonatal morbidity and mortality.
• Existing prediction tools often require data unavailable early in pregnancy.
• Administrative health records offer a scalable solution for early risk prediction.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting preterm birth at 26 weeks of gestation.
• Utilize routinely collected administrative health records for prediction.
• Assess model performance using various metrics.

Main Methods:

• Retrospective population-based cohort study of 328,834 pregnancies in Alberta, Canada (2009-2018).
• Linked maternal records, physician claims, and prescription data.
• Compared logistic regression, random forest, neural networks, gradient-boosted trees, and transformer models.

Main Results:

• Gradient-boosted tree model achieved the best performance (AUC 0.7704).
• Transformer models also demonstrated strong predictive capabilities.
• Risk stratification identified distinct groups with varying preterm birth rates.

Conclusions:

• Machine learning models using administrative data can effectively predict preterm birth by 26 weeks.
• Models facilitate early, scalable, population-level risk stratification.
• These tools complement clinical assessment for identifying high-risk pregnancies.