Development and external validation of a prediction model for poor prognosis before discharge in very preterm and/or very low birth weight infants with early-onset sepsis.

BACKGROUND: Clinical early-onset sepsis (EOS) is associated with death and major morbidity in very preterm and/or very low birth weight infants. We aimed to develop and externally validate a model for a composite adverse outcome before discharge.

METHODS: This retrospective multicenter study included infants with gestational age < 32 weeks and/or birth weight < 1500 g who met operational criteria for clinical EOS. Infants from two hospitals formed the development cohort (n = 249), and infants from a third hospital formed the external validation cohort (n = 84). The outcome comprised death, grade 3 bronchopulmonary dysplasia, severe intraventricular hemorrhage, cystic periventricular leukomalacia, or Bell stage II or higher necrotizing enterocolitis. Predictors were selected using least absolute shrinkage and selection operator and multivariable logistic regression. Internal validation used 500 bootstrap resamples.

RESULTS: The composite outcome occurred in 98 of 333 infants (29.4%). The final model included birth weight, blood lactate, and base excess. In the development cohort, the C-index was 0.8309, the Brier score was 0.1468, and the optimism-corrected C-index was 0.8270. The optimism-corrected calibration intercept and slope were - 0.011 and 0.968, respectively. In external validation, the AUC was 0.7819 (95% CI, 0.6602-0.9036), the Brier score was 0.1540, the calibration-in-the-large intercept was -0.254, and the calibration slope was 0.753, indicating overprediction and overly extreme predictions.

CONCLUSIONS: This three-variable model showed acceptable external discrimination, but calibration drift indicates that local recalibration and prospective multicenter validation are required before clinical use.