(584) Beyond Comorbidities: Integrating Real-Time Clinical Data to Improve Prediction of Severe Maternal Morbidity

To compare the predictive performance of three models for severe maternal morbidity (SMM): (1) the Obstetric Comorbidity Index (OB-CMI), (2) the California Maternal Quality Care Collaborative (CMQCC) score, and (3) a novel model that integrates OB-CMI and CMQCC elements with real-time clinical data. Prediction of SMM without transfusion (SMM-NT) was assessed as a secondary outcome.


Study Design:

We conducted a retrospective cohort study including 160,515 deliveries from 132,189 patients across a large New York health system (2019–2024). SMM was defined using CDC criteria. Patients were randomly assigned to training (60%), validation (20%), and test (20%) sets, with all deliveries from a given patient grouped together. Logistic regression models were developed using backward elimination. The novel model incorporated comorbidities, vital signs (heart rate, mean arterial pressure, temperature), laboratory values (hematocrit, white blood cell count, platelets), non-linear terms, and missingness indicators. Model performance was assessed in the test set using area under the receiver operating characteristic curve (AUC-ROC), Brier score, and the Spiegelhalter test for calibration.


Results:

Among 31,943 test-set deliveries, 6,966 (4.3%) met SMM criteria and 1,373 (0.9%) met SMM-NT criteria. The novel model significantly outperformed both OB-CMI (AUC 0.6726) and CMQCC (AUC 0.7244), achieving an AUC of 0.7630 (p< 0.0001 vs. both) and better calibration (Brier 0.039, Spiegelhalter p=0.53). For SMM-NT, discrimination and calibration were even stronger (AUC 0.7868, Brier 0.0123). Model performance was consistent across racial and insurance subgroups.


Conclusion:

A novel model incorporating real-time admission vitals and labs improved prediction of SMM and SMM-NT compared to OB-CMI and CMQCC alone. Dynamic data may enhance early risk identification and support targeted maternal care interventions.