(293) Development and validation of a screening tool to predict severe neonatal morbidity

Objective:
To develop and validate a clinical screening tool to predict severe neonatal morbidity (SNM) at both labor floor admission and delivery using routinely available maternal, fetal, social, and intrapartum factors.

Study Design:
This retrospective cohort study included singleton deliveries at ≥32 weeks of gestation at two tertiary hospitals in New York (2019–2023). Patients were randomly assigned to training (70%) and testing (30%) cohorts. SNM was defined using a composite of diagnoses and procedures indicating life-threatening neonatal complications. Six multivariable logistic regression models were developed using stepwise selection. Maternal variables were derived from obstetric comorbidity index (OB-CMI) components. Fetal factors included fetal size abnormalities, amniotic fluid abnormalities, major structural anomalies, and abnormal umbilical artery Doppler measurements. Additional models incorporated social and intrapartum variables. Models were compared using AUC and Brier score; final models were validated on the testing cohort.

Results:
Among 53,159 deliveries, 2,554 (4.8%) newborns experienced SNM. Models based on maternal or fetal factors alone had moderate discrimination (AUCs 0.72 and 0.68, respectively). Combining maternal and fetal variables improved discrimination (AUC 0.73); adding social factors yielded the best model at admission (AUC 0.74, Brier 0.057). Gestational age, major fetal anomaly, and polyhydramnios were the strongest predictors available at admission. After delivery, cesarean birth and chorioamnionitis were highly predictive, and a final model incorporating intrapartum variables achieved the best performance (AUC 0.76). All models demonstrated good calibration and consistent validation.

Conclusion:
We developed and validated a practical tool for SNM prediction at maternal admission and delivery. Incorporating defined fetal characteristics alongside maternal and social factors improved predictive performance and may support early risk stratification and neonatal care planning.