Related Experiment Video
Updated: Jun 22, 2025
07:57
Modeling Neonatal Intraventricular Hemorrhage Through Intraventricular Injection of Hemoglobin
Published on: August 25, 2022
2.9K
Machine Learning Models for Prediction of Maternal Hemorrhage and Transfusion: Model Development Study
Homa Khorrami Ahmadzia1,2, Alexa C Dzienny3, Mike Bopf4
1Division of Maternal-Fetal Medicine, Department of Obstetrics and Gynecology, George Washington University, Washington, DC, United States.
JMIR Bioinformatics and Biotechnology
|June 27, 2024
Summary
Machine learning models show improved prediction of postpartum hemorrhage (PPH) compared to traditional methods. Gradient boosting demonstrated the best performance in identifying women at risk for PPH.
Area of Science:
- Obstetrics and Gynecology
- Medical Informatics
- Data Science in Healthcare
Background:
- Current postpartum hemorrhage (PPH) risk stratification relies on traditional statistical models or expert opinion.
- Machine learning (ML) offers potential for more complex and optimized PPH prediction.
Purpose of the Study:
- To enhance PPH prediction accuracy.
- To compare the performance of ML models against traditional statistical methods for PPH risk stratification.
Main Methods:
- Utilized the Consortium for Safe Labor data set (2002-2008) from 12 US hospitals.
- Developed prediction models using logistic regression, support vector machines, multilayer perceptron, random forest, and gradient boosting (GB).
- Included 50 antepartum, intrapartum, and hospital characteristics; primary outcome was transfusion or PPH (≥1000 mL estimated blood loss).
Main Results:
- The gradient boosting (GB) ML model achieved the best overall performance for predicting the transfusion-PPH composite (ROC-AUC=0.833).
- Key predictive features included mode of delivery, oxytocin dose, tocolytic use, anesthesia nurse presence, and hospital type.
- Models using antepartum and intrapartum features showed the best positive predictive values.
Conclusions:
- Machine learning models demonstrate superior discriminability for PPH prediction compared to logistic regression.
- The Consortium for Safe Labor data set may have limitations for risk analysis due to subgroup effects impacting accuracy and generalizability.
