(1227) Severe Early Post-Partum Hemorrhage: an Individualized Risk assessment by machine learning models

Objective: Severe early post-partum hemorrhage (SEPPH) is a common, life-threatening obstetric challenge. The risk factors of early postpartum hemorrhage had been researched, however, early, upon admission for labor, recognition of high-risk parturients by machine learning models (ML) has not been attempted.

Study Design: Analysis of a multi-parameter dataset of 227,424 births at a tertiary medical center. SEPPH was defined as - Hemoglobin loss more than 4 g/dl or post-partum hemoglobin level less the 8 g/dl or treatment of 2 or more pRBC post-partum. For the ML analyses, the data set was divided into two groups: SEPPH and no SEPPH; analysis was conducted using a gradient boosting model (CatBoost) algorithm, to create a model predicting the risk score of SEPPH. Furthermore, this type of analysis categorized the study population high-risk and low-risk groups for SEPPH by using the Youden's Index.

Results: Study population is depicted in Figure 1. Using traditional statistical methods dissected into the individual risk factors for SEPPH. Notably, we revealed that the primiparity is a major risk factor for extreme postpartum hemorrhage (OR 4.5 [4.0-5.0] CI, p< 0.001). In addition, Multiple gestation (OR 2.5 [2.0-3.1] CI p< 0.001), previous caesarean section (OR 3.0 [2.6-3.5] CI, p< 0.001), placentation anomaly (OR 4.0 [3.424-4.814] CI, p< 0.001), were significant risk factors.

The high-risk group identified by the ML model showed a significant higher risk for adverse maternal and neonatal outcomes including among the low and high risk for SEPPH: uterine rupture (OR-3.5), re-admission 42 and 90 days after delivery (OR -2.2, 2.2 respectively) and perinatal fever (OR-2.2), uterine scar dehiscence (OR-1.9). Newborn infection (OR-2.5), postnatal intensive care unit (NICU) stay >3 days (OR 2.4, 2.37 respectively) and low Apgar (1-7) at 1^st and 5^th min (OR 2.0, 1.26 respectively). Figure -2.

Conclusion: We built a machine learning model with defined measurable predictable factors to identify high risk for SEPPH. Prediction models for early PPH may allow appropriate risk reducing preventive care during and after delivery.