Improving postpartum hemorrhage risk prediction using longitudinal electronic medical records

Area of Science:

• Obstetrics and Gynecology
• Maternal Health
• Medical Informatics

Background:

• Postpartum hemorrhage (PPH) is a primary cause of preventable maternal mortality in the US.
• Current risk assessment tools for PPH lack optimal accuracy.

Purpose of the Study:

• To develop and validate a novel risk assessment tool for PPH.
• To compare the accuracy of the new tool against existing clinical risk assessment methods.

Main Methods:

• Utilized a previously validated PPH digital phenotype to identify 6639 PPH deliveries from a cohort of 70,948.
• Trained a gradient boosting model using pre-delivery electronic medical record data.
• Compared the model's performance against three clinical tools and one published model using a held-out test set.

Main Results:

• The 24-feature model achieved an AUROC of 0.71, outperforming existing tools (research-based AUROC 0.67, clinical AUROCs 0.55-0.61).
• Identified novel predictive features including red blood cell indices and admission infection markers.
• Determined critical thresholds for vital signs and lab values, noting a significant risk increase with intrapartum systolic blood pressure >132 mm Hg.

Conclusions:

• Developed a novel, accurate approach for predicting PPH risk.
• Identified clinical feature thresholds to guide intrapartum monitoring for PPH.
• The model shows potential for prospective evaluation to reduce PPH-related morbidity and mortality through early detection and prevention.