Early Prediction of Hypertensive Disorders of Pregnancy Using Machine Learning and Medical Records from the First and Second Trimesters

Abstract

Hypertensive disorders of pregnancy (HDPs) are a leading cause of maternal and perinatal morbidity and mortality worldwide, particularly in low-resource settings. Early and accurate prediction of HDPs is essential for timely intervention, which can improve outcomes for both mothers and infants. However, existing artificial intelligence models rely on costly or invasive methods-such as blood and urine biomarkers, genetic testing, or ultrasound—that are typically collected later in pregnancy, thereby delaying actionable care. This study aims to develop a practical and cost-effective machine learning (ML) model for the early prediction of HDPs using only non-invasive, routinely collected data from the first and second trimesters. Specifically, we used maternal demographic information, blood pressure (BP) measurements, and body mass index values to train and evaluate the ML model. The developed CatBoost model achieved an area under the receiver operating characteristic curve of 0.77, an accuracy of 0.67, a sensitivity of 0.70, and a specificity of 0.67. The dataset consisted of 6,422 records related to 6,060 pregnant subjects, collected from academic medical centers in Atlanta, Georgia (2009–2022). In conclusion, this approach offers a practical solution for the early prediction of HDP, by relying only on inexpensive tools, BP measurement devices, and weight scales, with appropriate testing and validation. The model has the potential to support early risk stratification and improve the management of adverse pregnancy outcomes, especially in resource-limited environments.