Predicting extrauterine growth restriction at the parenteral-enteral nutrition transition: a model for risk identification in very preterm infants.

INTRODUCTION: This study aims to identify the key predictors of longitudinal extrauterine growth restriction (EUGR) in very preterm infants during the parenteral-to-enteral nutrition transition and subsequently develop an effective predictive model for early clinical identification. METHODS: A total of 260 infants born at 28-31+6 weeks of gestation were included in this retrospective study. Perinatal and early postnatal clinical data were collected. The univariate regression analysis (p < 0.1) and LASSO algorithms were employed to comprehensively screen for variables. The screened variables were utilized to construct the optimal model. The development of an optimal predictive model involved performance comparison and validation across five algorithms, including logistic regression and four machine learning algorithms. This process was coupled with an explanation of machine learning algorithms using the Shapley Additive exPlanations (SHAP) to ensure interpretability. RESULTS: The study identified six key predictors: gestational age, sex, time to regain birth weight, gestational diabetes, hypertensive disorders of pregnancy and blood urea nitrogen. Among the models compared, the logistic regression model exhibited the best performance, and the nomogram developed from this model demonstrated strong predictive efficacy. CONCLUSION: In this study, we developed and internally validated a prediction model for longitudinal EUGR in very preterm infants specifically during the parenteral-to-enteral transition. The model demonstrated robust predictive performance, strong clinical utility, and exceptional interpretability. It offers a practical tool for risk stratification during the parenteral-to-enteral transition, enabling clinicians to identify high-risk infants who may benefit from timely, personalized nutritional support.