Development and Validation of an Early Prediction Model for Neonatal Hypoglycemia in Full-Term Appropriate-for-Gestational-Age Newborns within 24 Hours of Birth

Background: Standardized screening for neonatal hypoglycemia (NH) in full-term, appropriate-for-gestational-age (AGA) newborns is lacking. We aimed to develop a bedside prediction model for early NH (≤24 hours) in this population. Methods: This retrospective study included 1,543 full-term AGA newborns, randomly split them into training and testing sets at a ratio of 7:3. From 51 candidate variables, LASSO and multivariate logistic regression were used to select predictors and construct a nomogram. Model performance was evaluated via area under the curve (AUC), calibration, and decision curve analysis (DCA). Results: NH occurred in 15.4% (237/1,543) of the cohort. LASSO regression identified nine core predictive variables: maternal factors (gestational diabetes mellitus, gestational hypertension, mode of delivery, 1-hour glucose level on the oral glucose tolerance test, total bile acids, and hemoglobin) and neonatal factors (sex, gestational age, and feeding method). It demonstrated good discrimination, yielding an AUC of 0.943 (95% CI: 0.919–0.967) in the training set and 0.934 (95% CI: 0.896–0.971) in the testing set. Calibration curves and DCA confirmed robust predictive accuracy and high clinical net benefit. Conclusion: This bedside nomogram enables accurate, early risk stratification of NH in full-term AGA newborns, optimizing targeted screening in a frequently overlooked population.