Machine Learning and Statistical Models to Predict Postpartum Hemorrhage

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Postpartum hemorrhage (PPH) is a significant cause of maternal morbidity and mortality.
• Accurate prediction of PPH risk at labor admission is crucial for timely intervention.
• Existing risk assessment tools may lack sufficient predictive power.

Purpose of the Study:

• To develop and compare machine learning and statistical models for predicting PPH risk.
• To identify the most accurate model for PPH prediction using routinely available labor admission data.

Main Methods:

• Utilized data from the U.S. Consortium for Safe Labor Study (2002-2008).
• Compared logistic regression, lasso regression, random forest, and extreme gradient boosting models.
• Defined PPH as estimated blood loss ≥1,000 mL; assessed 55 risk factors.
• Validated models temporally and across sites using C statistics, calibration, and decision curves.

Main Results:

• Out of 152,279 births, 4.8% experienced PPH.
• Extreme gradient boosting achieved the highest C statistic (0.93), followed by random forest (0.92).
• All models showed good-to-excellent discrimination, with extreme gradient boosting providing the greatest net benefit.

Conclusions:

• Machine learning and statistical models can effectively predict PPH risk at labor admission.
• The extreme gradient boosting model shows superior performance for PPH prediction.
• Clinical application of these models can enhance preparedness and patient triage.