METHODS
We performed a retrospective cohort study of all deliveries with a documented QBL from January 1, 2018, to January 1, 2020. Institutional review board approval was obtained from Boston Medical Center and Boston University School of Medicine.
At our institution, an interdisciplinary team implemented a postpartum hemorrhage bundle protocol in 2016 [Appendix S1]. It includes but is not limited to the following: unit preparation for obstetric hemorrhage, including simulations, hemorrhage risk assessment on admission [Appendix S2] and throughout the intrapartum course, quantitative blood loss (QBL) for all deliveries, and a staged response to the QBL.
Information was gathered from department birth logs that are abstracted daily from the electronic medical record for all deliveries. This includes information about maternal demographics, maternal medical and obstetric history, antenatal course, intrapartum course, anesthesia type if any, and QBL. Additional information was abstracted by trained chart abstractors into a standardized chart abstraction form using a secure database.
We created an OBH-M composite of all cases associated with any of the following: (i) transfusion of any blood products, (ii) intensive care unit admission, (iii) clinical, laboratory, or radiologic evidence of end-organ damage (iv) hysterectomy, (v) operative intervention in the immediate postpartum period, (vi) uterine artery embolization or (vii) maternal death. Individual deliveries were categorized into ten equally sized deciles based on the total QBL. We then compared the proportion of obstetric-hemorrhage related morbidity within each decile. Among the deciles with the highest proportions of OBH-M, we stratified deliveries into seven groups with ascending intervals of 250cc QBL to interrogate a threshold of QBL associated with an increase in association with a composite of obstetric hemorrhage-related morbidity (OBH-M). A significant point in change of risk for OBH-M was established by sequentially conducting the ranked analysis and stratified QBL-based analysis.
We then compared the standard definition of OBH (≥1000cc) to the new definition extrapolated from our stratified analysis (≥1500cc). These tests included positive predictive value (PPV), negative predictive value (NPV), sensitivity, specificity, and positive likelihood ratios (+LR). We utilized a McNemar test to compare sensitivity and specificity and calculated a post-test probability calculation based on the +LR values. We also calculated an area under the ROC curve (AUC) for each definition and compared the AUCs with the DeLong test.6 An alpha level of 0.05 was used to determine significance. All analyses were completed using STATA MP, version 17 or IBM SPSS Statistics 27. 7,8