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: Clinical Obstetrics

Poster Session 3

(935) Machine Learning Based Risk Stratification for Postpartum Hemorrhage: Insights from a National Dataset

Thursday, February 12, 2026

10:30 AM - 12:00 PM

Presenting Author(s)

Sage Smith (she/her/hers)

University of Chicago
Chicago, Illinois, United States

Submitting Author(s)

Ashley Zimmermann, MD (she/her/hers)

Resident
Northwell
New Hyde Park, New York, United States

Coauthor(s)

Mariah Buzzell, MD

Resident
Northwell
Northwell Lenox Hill Hospital, New York, United States
Angela Kown

University of Chicago
Chicago, Illinois, United States
YA

Yuzuru Anzai, MD

Attending Physician
Northwell
New Hyde Park, New York, United States

Objective:
The rate of postpartum hemorrhage (PPH), defined as blood loss of ≥1,000mL, has increased over the last decade and remains the leading cause of maternal mortality. This study aims to identify maternal and obstetric factors associated with increased risk of PPH.

Study Design:
Maternal and obstetric data from the CDC Natality dataset spanning 2021 to 2023 were analyzed. Deliveries complicated by PPH were identified based on the administration of blood transfusion with or without ICU admission. Clinically relevant variables such as maternal demographic information as well as obstetrical data were selected. Two different models were developed to predict PPH risk - a decision tree classifier model and a logistic regression model. Performance of these two models was evaluated using sensitivity, specificity, and area under the receiver operating curve (AUC).

Results:
With >95% power from our large cohort, the decision tree classifier model performed with an AUC of 0.83 (sensitivity 72.5%, specificity 77.9%, PPV 76.7%, NPV 73.9%), while logistic regression yielded an AUC of 0.80 (sensitivity 73.6%, specificity 71.5%, PPV 72.1%, NPV 72.9%). In the decision tree, delivery method was one of the strongest predictors of PPH, with risk decreasing in the following order: vaginal delivery after prior cesarean, primary cesarean, repeat cesarean, vaginal delivery without prior cesarean. Labor induction and prior cesarean delivery were also major contributors. Eclampsia was a negligible predictor, likely due to its low prevalence and overlap with other hypertensive variables. Regression analysis highlighted previous cesarean delivery, chorioamnionitis, and pre-pregnancy diabetes as the strongest predictors.

Conclusion:
Both models performed similarly and demonstrated good discrimination for identifying individuals at risk of PPH. Early identification of individuals at risk of PPH is critical to improve outcomes and guide preventative interventions.