Can AI Predict Preterm Birth in Diabetic, Hypertensive Pregnancies?

Retrospective study finds machine-learning model identifies clinical factors associated with preterm delivery in a high-risk obstetric population.

By
Julia Cipriano, MS, CMPP
February 24, 2026
3 min

Conexiant

Overview

A Naive Bayes machine-learning model, developed from a retrospective dual-center study, effectively predicted preterm birth in pregnant women with gestational diabetes and hypertensive disorders. This model offers a promising tool for personalized obstetric risk management in high-risk pregnancies.

Background

Gestational diabetes mellitus (GDM) and hypertensive disorders of pregnancy (HDP) frequently coexist, significantly increasing the risk of preterm birth and adverse perinatal outcomes. Current predictive tools for this high-risk group are limited, with GDM affecting approximately 6-9% of pregnancies and HDP affecting 2-8%. This highlights the need for innovative approaches. The development of machine-learning models may enhance individualized risk assessment and management in obstetric care.

Data Highlights

Model	AUC	Accuracy	Sensitivity	Specificity
Naive Bayes	0.777	0.801	0.792	0.804
LASSO	0.802	N/A	N/A	N/A

AUC values above 0.7 are generally considered acceptable for clinical use.

Key Findings

The Naive Bayes model was selected for its balance of discrimination, interpretability, and robustness among the thirteen variables assessed, leading to five key predictors for preterm birth: alanine transaminase, aspartate transaminase, albumin, lactate dehydrogenase, and systolic blood pressure.
The model maintained strong generalization in an external validation cohort with an AUC of 0.777.
SMOTE was utilized to address class imbalance, enhancing model performance.
Future studies are recommended to validate the model in larger, multicenter cohorts.

Clinical Implications

The Naive Bayes model can assist clinicians in early identification of high-risk pregnancies affected by GDM and HDP, facilitating personalized risk management. Implementing such predictive tools may improve clinical decision-making and outcomes in obstetric care, potentially reducing preterm birth rates.

Conclusion

The development of a machine-learning model for predicting preterm birth in high-risk pregnancies represents a significant advancement in obstetric risk management. Continued validation and integration into clinical practice are essential for maximizing its utility and ensuring patient safety.