Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Presentation Icons

Additional registration fee required

Faculty have requested this content not be shared outside of the session

CME Credit Offered

Abstract Award

Recording available 2/16-5/2

AIUM Credit

Foundation Awardee

Poster Icons

Abstract Award

Foundation Awardee

Poster Session 3

Category: Hypertension

Poster Session 3

(878) Development of novel machine learning model for the prediction of postpartum readmission for hypertension management

Thursday, February 12, 2026

10:30 AM - 12:00 PM

Submitting Author and Presenting Author(s)

Meralis V. Lantigua-Martinez, MD

MFM Fellow
Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, NYU Grossman School of Medicine
New York, New York, United States

Coauthor(s)

WL

Wenke Liu, PhD

Institute for Systems Genetics, NYU Grossman School of Medicine
New York, New York, United States
SF

Steven Friedman, MS

Associate Research Scientist
Division of Biostatistics, Department of Population Health, NYU Grossman School of Medicine
New York, New York, United States
AR

Ashley S. Roman, MD, MPH

MFM Division Director
Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, NYU Grossman School of Medicine
New York, New York, United States
DF

David Fenyo, PhD

Institute for Systems Genetics, NYU Grossman School of Medicine
New York, New York, United States
Christina A. Penfield, MD, MPH

Assistant Professor
Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, NYU Grossman School of Medicine
New York City, New York, United States

Objective:
Postpartum hypertension remains a leading cause of emergency department visits and readmissions in the US. Identifying patients at high risk for postpartum readmission may help identify candidates for closer outpatient follow-up. Machine learning may allow the evaluation of multiple variables to predict high risk patients with increased precision compared to traditional readmission risk models. We developed a machine learning algorithm that predicts patients at risk for postpartum readmission for hypertension management (PRHM).

Study Design:

We performed an observational cohort study of all births in an academic health system between 1/1/20 and 6/30/24. Predictive models for PRHM were developed using random forest and XGBoost. 54 variables associated with PRHM were considered in the model. Hyperparameters were tuned with a grid search over 5-fold cross-validation and model performance was evaluated with area under the curve for precision-recall (AUCPR) and receiver operating characteristic (AUCROC). The best hyperparameter set was selected as the combination that yields the highest mean AUCPR across folds. Feature importance was retrieved from the best model.

Results:
Of 56,147 deliveries among 50,467 patients, 730 (1.3%) resulted in readmission for hypertension management. Both random forest and XGBoost showed high AUCROC (random forest: 0.923 ± 0.005; XGBoost: 0.928 ± 0.004, mean ± standard error), but low AUCPR (random forest: 0.123 ± 0.007; XGBoost: 0.162 ± 0.010). At the optimal F1 score, specificity was around 98%, sensitivity ranged from 28.4%-43%, negative predictive value (NPV) was 97-98%, and positive predictive value (PPV) 15%-16% (Table 1). For both algorithms, a history of pregnancy related hypertension contributed the most to the final prediction (Fig. 1).

Conclusion:
Despite high overall accuracy and specificity, both models had poor positive predictive value in identifying patients at risk for PRHM, limiting clinical applicability. Our findings highlight the challenges of applying machine learning to rare outcomes like postpartum readmission.