TOP 33 - Predicting Onset and Severity of Necrotizing Enterocolitis (NEC)

Publication Number: 2784.TOP 33

Briggs S. Carhart-Veres, Johns Hopkins University School of Medicine, Baltimore, MD, United States; David Hackam, Johns Hopkins Children's Center, Baltimore, MD, United States; Khyzer B. Aziz, Johns Hopkins University School of Medicine, Baltimore, MD, United States; Amanda S. Hu, Johns Hopkins University School of Medicine, Baltimore, MD, United States; Lena M. Bode, Johns Hopkins Children's Center, Baltimore, MD, United States; Charbel Chidiac, Johns Hopkins University School of Medicine, Baltimore, MD, United States; William Fulton, John's Hopkins University, Baltimore, MD, United States

Background: Necrotizing enterocolitis (NEC) remains the leading cause of gastrointestinal-related mortality and long-term morbidity in premature infants. While well-established risk factors such as prematurity, formula feeding, and abnormal bacterial colonization are recognized, the ability to accurately predict which infants will develop NEC or progress to severe disease remains limited. Current clinical management is largely reactive, emphasizing treatment rather than prevention or early risk stratification, and mortality rates have shown minimal improvement over time. NEC represents a condition where systematic tracking of clinical and biological variables could meaningfully guide care, similar to how risk-based models inform management of other neonatal diseases such as bronchopulmonary dysplasia and early-onset sepsis. Identifying novel, reliable predictors of NEC onset and severity could enable more precise risk stratification, earlier intervention, and ultimately improved outcomes for this highly vulnerable population.

Objective: To develop and validate a machine learning–based predictive model to identify neonates at high risk for NEC onset and to stratify disease severity, including the likelihood of requiring surgical intervention.

Design/Methods: This study will employ a multi-center, secondary data analysis using electronic health record (EHR) data from six neonatal intensive care units (NICUs) in the United States. Eligible subjects include preterm infants admitted to participating NICUs during the study period. Demographic, clinical, laboratory, and imaging variables, including feeding timing, glucose, blood counts, inflammatory markers, radiologic studies, etc. will be extracted across sites. NEC onset and severity will be defined using standardized diagnostic and surgical criteria. Predictive modeling will incorporate logistic regression and supervised machine learning algorithms to identify variables most strongly associated with NEC development and adverse outcomes such as nutritional delivery, respiratory support, neurologic sequelae, and mortality. Model performance will be assessed using cross-validation analysis. The study was approved by the JHU IRB.