A groundbreaking machine-learning model developed by researchers at Weill Cornell Medicine promises to revolutionize the clinical management of preeclampsia, a dangerous hypertensive disorder occurring late in pregnancy. By harnessing vast electronic health record data and applying sophisticated artificial intelligence techniques, the model aims to provide clinicians with dynamic, real-time predictions of preeclampsia risk during the critical third trimester. This advancement addresses a long-standing gap in obstetric care, offering a potential tool to mitigate adverse outcomes for both parent and child.
Preeclampsia, characterized by sudden onset high blood pressure prior to delivery, complicates approximately 2% to 8% of pregnancies globally. The condition’s pathophysiology involves complex interactions among the placenta, maternal cardiovascular system, and immune responses. Its late manifestation, often after 34 weeks’ gestation, presents considerable challenges for timely diagnosis and intervention, with serious repercussions including maternal organ damage and fetal growth restrictions.
Traditional predictive models focus primarily on early pregnancy indicators, aiming to stratify risk during the first trimester. These algorithms enable preventive strategies such as aspirin prophylaxis and enhanced surveillance for high-risk patients. However, these early models lack precision in forecasting late-onset and term preeclampsia cases, which constitute the majority of diagnoses. Consequently, clinicians have had limited tools to anticipate onset during the late stages of pregnancy, constraining their capacity to deploy tailored interventions.
Addressing this clinical void, a multidisciplinary team spearheaded the development of a machine-learning algorithm trained on an extensive dataset comprising nearly 59,000 deidentified pregnancies from three NewYork-Presbyterian hospitals. The core training cohort consisted of 35,895 pregnancies delivered at NewYork-Presbyterian/Weill Cornell Medical Center between October 2020 and May 2025. By integrating temporal clinical variables and leveraging advanced computational methods, the model dynamically updates preeclampsia risk scores as new health data become available, closely mirroring the clinical workflow.
Key components impacting model performance include vital signs, laboratory findings, and demographic variables. Notably, blood pressure measurements emerged as the dominant predictor throughout gestation. Early third-trimester abnormalities in routine blood laboratory tests also contributed significant predictive value. These lab markers likely reflect evolving placental dysfunction, highlighting the model’s biological interpretability in addition to its statistical power.
As the pregnancy advances into the later third trimester, the influence of parameters such as maternal age and white blood cell count intensifies. Elevated leukocyte levels suggest systemic inflammation may be a critical mediator of preeclampsia pathogenesis during this period. This nuanced understanding provides a potential window for distinguishing phenotypic subtypes of preeclampsia, stratified by underlying mechanisms such as placental insufficiency versus inflammatory pathways.
Validation of the algorithm on independent cohorts from NewYork-Presbyterian Lower Manhattan Hospital (8,664 pregnancies) and Brooklyn Methodist Hospital (14,280 pregnancies) demonstrated robust generalizability and predictive accuracy. The model’s capacity to identify high-risk pregnancies around the 34-week mark offers clinicians a valuable lead time for intensified monitoring, pharmacologic blood pressure control, and strategic decisions regarding timing of delivery to optimize maternal-fetal outcomes.
Unlike prior static risk calculators that provide a single timepoint assessment, this continuously updated model dynamically incorporates newly acquired electronic health data. This feature aligns closely with real-world clinical decision-making processes in late pregnancy, enabling personalized risk stratification and responsive management plans as patient status evolves.
The work underscores the power of artificial intelligence to transform prenatal care by uncovering complex, temporally dependent risk patterns otherwise imperceptible through conventional analytic methods. It also opens avenues for further investigation into the differential etiologies of preeclampsia manifestations at distinct gestational stages. Clarifying these mechanisms could inform more tailored and effective therapeutic interventions targeting the root causes of the disorder.
Future research directions include prospective clinical trials to evaluate the model’s impact on patient outcomes and its integration into routine obstetric practice. There is also interest in deploying similar AI-driven predictive frameworks for other pregnancy-related complications, potentially broadening the scope of precision medicine in maternal-fetal health.
This innovative model represents a significant step forward in the quest to preemptively identify and mitigate preeclampsia during one of the most vulnerable periods of pregnancy. Its development reflects a successful confluence of clinical expertise, advanced data science, and translational research aimed at improving the safety and well-being of families worldwide.
Subject of Research:
Preeclampsia risk prediction using machine-learning models based on longitudinal electronic health record data in late pregnancy.
Article Title:
Machine-Learning Model for Dynamic Prediction of Preeclampsia Risk in Late Pregnancy.
News Publication Date:
6-Mar-2026
Web References:
Weill Cornell Medicine – Dr. Fei Wang
Weill Cornell Medicine – Dr. Zhen Zhao
Weill Cornell Medicine – Dr. Tracy Grossman
References:
Published in JAMA Network Open, March 6, 2026.
Image Credits:
Not provided.
Keywords
Pregnancy, Preeclampsia, Machine Learning, Artificial Intelligence, Electronic Health Records, Maternal-Fetal Medicine, Predictive Modeling, Placental Dysfunction, Inflammation, Blood Pressure, Third Trimester, Obstetrics
