Predicting Postpartum Hemorrhage Risk With Second-Trimester Data.

OBJECTIVE

To assess whether a postpartum hemorrhage (PPH) risk stratification tool using structured data in the electronic health record (EHR) could be developed and validated with information known before the third trimester.

METHODS

We performed a retrospective cohort study of individuals receiving prenatal care and delivering at a single academic institution between January 1, 2017, and December 31, 2023. The cohort included individuals who established care before 24 weeks of gestation, delivered at or after 24 weeks, and had a recorded estimated blood loss at the time of delivery. The cohort was subdivided into three groups for model development: 1) training (66.0% of deliveries between 2017 and 2022), 2) testing (33.0% of deliveries between 2017 and 2022), and 3) temporal validation (deliveries in 2023). We performed traditional and machine-learning modeling approaches in the training data. Discrimination and calibration were compared by calculating the area under the receiver operating characteristic curve and constructing calibration curves; a single approach was selected to identify a predicted probability to classify patients as high risk of PPH, which we defined as two or more times the increased risk of PPH and without exceeding a screen-positive rate of 20% (or twice the estimated baseline rate of PPH in the population).

RESULTS

A total of 17,201 patients were included and split between the three cohorts: 10,060 in the training set, 5,183 in the testing set, and 1,958 in the temporal validation set. Model discrimination and calibration were similar among all approaches. Model parsimony was used to select the final modeling approach: logistic regression with stepwise backward selection. This approach yielded an area under the curve of 0.689 (95% CI, 0.667-0.715) and 0.659 (95% CI, 0.621-0.694) in the testing and validation data. Among those classified as high risk, the screen-positive rates and positive predictive values were 19.2% (95% CI, 18.1-20.3%) and 20.1% (95% CI, 17.8-22.7%) in the testing set and 21.6% (95% CI, 19.9-23.4%) and 21.3% (95% CI, 17.7-25.5%) in the validation sets.

CONCLUSION

Using structured EHR data known by 24 weeks of gestation, we developed a prenatal risk stratification model for PPH. This model was designed to be used for predelivery planning and health optimization, particularly among those at the highest risk of PPH and its subsequent morbidity, in contrast to other tools that stratify risk after admission of for delivery.