Maternal morbidity and mortality have gained major attention recently, spurred on by rising domestic rates even as maternal mortality decreases in Europe. A major driver of morbidity and mortality among delivering women is postpartum hemorrhage (PPH). PPH is currently phenotyped using the subjective measure of ‘Estimated blood loss’ (EBL), which has been shown to be unreliable for tracking quality. Here we present a framework for phenotyping PPH into multiple severity levels, using a combination of data-driven techniques and expert-derived clinical indicators. We validate the framework by predicting large drops in hematocrit and quantitative blood loss, finding that the framework performs better in predicting coded PPH than a hematocrit-based predictor or predictors based on other metrics such as blood transfusions, and does better in predicting quantitative blood loss, a gold standard metric for blood loss that we have for a subset of patients, than any predictor we could build using hematocrit drops alone. In all, we present a principled framework that can be used to phenotype PPH in hospitals using readily available EHR data, and that will perform with more granularity and accuracy than existing methods.
Learning Objective: To understand that there is a need to move beyond current inaccurate methods for determining postpartum hemorrhage, and that our severity phenotyping framework is an improvement on current methods in both granularity and accuracy.
Matthew Oberhardt (Presenter)
Alexander Friedman, Columbia University Irving Medical Center
Rimma Perotte, NewYork-Presbyterian Hospital
Jean-Ju Sheen, Columbia University Irving Medical Center
Alan Kessler, Weill Cornell Medical Center
David Vawdrey, NewYork-Presbyterian Hospital
Robert Green, NewYork-Presbyterian Hospital
Mary D'Alton, Columbia University Irving Medical Center
Dena Goffman, Columbia University Irving Medical Center