New model may predict preeclampsia in late pregnancy, suggests study

A machine-learning model developed by Weill Cornell Medicine investigators may provide clinicians with an early warning of a complication that can occur late in pregnancy.

Preeclampsia is a sudden onset condition that involves high blood pressure prior to delivery. It affects about 2% to 8% of pregnancies worldwide and can have serious consequences for both parent and child. A new study, published March 6 in JAMA Network Open, describes a machine-learning-based computer model that provides continually updated predictions of preeclampsia risk based on electronic health record data recorded late in pregnancy. The study was co-led by Dr. Fei Wang, associate dean for AI and data science and the Frances and John L. Loeb Professor of Medical Informatics in Department of Population Health Sciences at Weill Cornell Medicine, and Dr. Zhen Zhao, professor of clinical pathology and laboratory medicine at Weill Cornell Medicine and central laboratory director at NewYork-Presbyterian/Weill Cornell Medical Center. Clinical expertise in obstetrics was provided by Dr. Tracy Grossman, assistant professor of clinical obstetrics and gynecology at Weill Cornell Medicine and a maternal-fetal medicine specialist at NewYork-Presbyterian Brooklyn Methodist Hospital.

Existing models that assess preeclampsia risk during the first trimester are primarily used as early warnings, allowing clinicians to prescribe aspirin as a preventive medication early in the pregnancy and provide additional monitoring throughout at-risk pregnancies. While these approaches may reduce the risk of early-onset preeclampsia, their predictive accuracy is limited for late-onset and term cases, which account for the majority of preeclampsia diagnoses. As a result, few tools are available to help predict short-term preeclampsia risk during the last trimester of pregnancy when most cases arise. To fill this gap, co-first authors Dr. Haoyang Li, a postdoctoral associate in population health sciences, and Dr. Yaxin Li, a postdoctoral associate in pathology and laboratory medicine, worked with Drs. Wang, Zhao and Grossman to develop and test a preeclampsia modeling tool using deidentified electronic health record data on almost 59,000 pregnancies at three NewYork-Presbyterian hospitals. The team created the model using data on 35,895 pregnancies of patients who delivered at NewYork-Presbyterian/Weill Cornell Medical Center between October 2020 and May 2025. The model most accurately predicted the likelihood of preeclampsia around 34 weeks, potentially giving clinicians time to take preventive measures.

The team then validated their model using data from 8,664 pregnancies at NewYork-Presbyterian Lower Manhattan Hospital and 14,280 at NewYork-Presbyterian Brooklyn Methodist Hospital. The model showed the pregnant patient’s blood pressure was the strongest predictor of preeclampsia. However, early in the third trimester, abnormal results from routine testing of the patient’s blood may also suggest potential risk. These laboratory results may suggest that emerging problems with the placenta, which provides nutrients and oxygen to the fetus, could be contributing to preeclampsia at this stage. Later in the third trimester, the patient’s age and white blood cell count became more important indicators, suggesting inflammation may be playing a role at this time.

The model may help clinicians identify patients in the third trimester of pregnancy most likely to develop preeclampsia and provide them additional lead time to take timely clinical action, including enhanced monitoring, blood pressure management, and decisions around delivery timing. Unlike earlier approaches that provide a single, static risk estimate, this model continuously updates preeclampsia risk with current electronic health record data as pregnancy progresses, aligning prediction with real-world clinical decision-making in late pregnancy. More study is needed to determine if preeclampsia at different stages of the third trimester has distinct causes, like placental dysfunction or systemic inflammation. But if those patterns are confirmed, they may help clinicians develop more targeted preeclampsia interventions that address the root causes.

Reference:

Li H, Li Y, Zang C, et al. Machine Learning for Dynamic and Short-Term Prediction of Preeclampsia Using Routine Clinical Data. JAMA Netw Open. 2026;9(3):e260359. doi:10.1001/jamanetworkopen.2026.0359