Antibiotic Duration and Critical Illness After Culture-Negative Late-Onset Sepsis Evaluation in Infants Born Preterm.

OBJECTIVE: To characterize antibiotic administration patterns among very low birth weight (VLBW, < 1500 grams at birth) infants born <33 weeks' gestation with negative late-onset sepsis (LOS) evaluations relative to life-threatening organ dysfunction. STUDY DESIGN: This retrospective cohort study used electronic health record data from 3 neonatal intensive care units (NICUs) in the Johns Hopkins Health System from June 2016 to April 2026. Eligible infants were born preterm, VLBW, evaluated for LOS at ≥72 hours of age, had no systemic antibiotics in the preceding 48 hours, and had no growth from blood or other concurrent clinically relevant cultures. Primary outcomes were antibiotic courses ≤ 48 hours and > 48 hours of consecutive systemic antibiotic administration associated with culture-negative LOS evaluation. RESULTS: Across 202 initial culture-negative LOS evaluations, median antibiotic duration was 47.0 hours (IQR, 37.0-97.5); 122 courses were ≤ 48 hours, and 80 were > 48 hours. Antibiotic duration was poorly predicted from observed clinical features using supervised machine-learning models (AUC, 0.48-0.60). Unsupervised K-means clustering identified 2 physiologic phenotypes defined by the neonatal Sequential Organ Failure Assessment (nSOFA)-derived organ dysfunction, but these clusters did not distinguish observed antibiotic duration groups. CONCLUSIONS: Antibiotic duration decisions were not associated with life-threatening organ dysfunction and were not learnable from structured electronic health record signals. When cultures remain negative and organ dysfunction is absent or improving, continued antibiotics should be supported by explicit, objective criteria. Such criteria are necessary for stewardship now and for trustworthy artificial-intelligence decision support in the future.