Rehospitalization in Preterm Infants: Machine Learning Prediction Model and Associated Risk Factors

Abstract

Abstract – Introduction: Preterm birth is a major global health concern, with preterm infants (PIs) at increased risk of morbidity and rehospitalization in the first year after neonatal intensive care unit (NICU) discharge. Few studies have applied predictive modeling in this domain. This study aimed to develop a machine learning model to predict rehospitalization within 1 year of NICU discharge and to identify clinical characteristics associated with increased rehospitalization risk. Methods: This is a retrospective cohort study of 2, 226 PIs born between 2018 and 2023 at a tertiary-care pediatric hospital in Israel. Data were obtained from NICU and inpatient records, including clinical history, laboratory results, and hospitalization outcomes. A machine learning model was developed using the eXtreme Gradient Boosting algorithm (XGBoost) with 20 clinical predictors. Results: Rehospitalization within 1 year occurred in 358 of 2, 226 (16.1%) PIs, with one-third occurring within 30 days. The predictive model achieved an AUC of 0.69 (95% CI: 0.59–0.78), sensitivity of 0.38, specificity of 0.87, positive predictive value of 0.38, and negative predictive value of 0.87. Key predictors included early gestational age, lower birth weight, discharge weight >2, 000 g, prolonged NICU stay, trisomy, low socioeconomic score, gastrointestinal and neurological conditions, bronchopulmonary dysplasia, surgical interventions, and abnormal laboratory values. Conclusions: This study introduces one of the first machine learning models for predicting 1-year rehospitalization in PIs, combining predictive modeling with interpretability through the integration of novel parameters. The model provides insight into high-risk profiles and may support early targeted interventions in an underexplored clinical area.