Machine Learning Model to Predict Pre-Term Delivery Using Electronic Health Records

Preterm birth continues to be a major global contributor to neonatal illness and death, presenting substantial burdens for healthcare providers. In this study, we develop and validate an innovative machine learning framework that utilizes standard electronic health record (EHR) data to predict preterm delivery risk. A retrospective cohort design was employed, including over 5,400 pregnancies, where key maternal, fetal, and obstetric factors were initially identified through univariate and multivariate analyses. We evaluated and compared multiple machine learning approaches-logistic regression, decision trees, support vector machines, Naïve Bayes, and the ensemble AdaBoost method-for preterm birth prediction. The AdaBoost classifier demonstrated superior performance, attaining 95.4% accuracy, 98.5% recall, and an ROC AUC score of 0.93 in internal validation tests, establishing it as the most effective model for this clinical application. Prospective external validation further underscored its clinical applicability, particularly in correctly identifying non-preterm cases while maintaining substantial sensitivity for high-risk cases. The integration of this predictive model into clinical workflows could facilitate early identification of women at risk for preterm delivery, enabling timely and tailored interventions. Furthermore, the research focuses the attention on the potential of effectively using the existing EHR data to create more scalable and cost-effective predictive tools in obstetrics. Despite proclaimed results, further related studies and multicentred validations are promised to enhance the model's applicability and to enhance its predictive accuracy for divergent populations.