Abstract / Summary
Early-onset sepsis may increase the risk of intraventricular hemorrhage in very preterm infants and/or very low birth weight infants through inflammatory, hemodynamic, and coagulation-related pathways. We aimed to develop and internally validate a clinical prediction model for early intraventricular hemorrhage risk stratification in this high-risk subgroup using routinely available variables. This retrospective multicenter study included preterm infants with gestational age < 32 weeks and/or birth weight < 1500 g who met the clinical diagnostic criteria for early-onset sepsis and were admitted to three tertiary neonatal intensive care units in northern China between January 2023 and December 2024. At least one cranial ultrasound examination was performed within the first 72 hours after birth, and the primary outcome was any intraventricular hemorrhage (Papile grade I–IV) detected within the first 7 days of life. Candidate predictors were prespecified and restricted to variables available before intraventricular hemorrhage ascertainment. Predictor selection was performed using least absolute shrinkage and selection operator regression, followed by multivariable logistic regression to construct a nomogram. Model performance was assessed by discrimination, calibration, Brier score, decision curve analysis, and bootstrap internal validation with 500 resamples. A sensitivity analysis was performed using severe intraventricular hemorrhage (Papile grade III–IV) as the outcome. Additional analyses assessed nonlinearity, interactions, leave-one-center-out validation, and a simplified model excluding purulent meningitis. A total of 340 infants were included in the final analysis. Least absolute shrinkage and selection operator regression identified five candidate predictors, and the final multivariable model retained four: 10-minute Apgar score, arterial pH, purulent meningitis, and patent ductus arteriosus. The model showed acceptable apparent performance, with a C-index of 0.755 (95% CI 0.701–0.810), a Brier score of 0.184, and a calibration slope of 1.000. Bootstrap validation with the final predictor set held fixed yielded an optimism-corrected C-index of 0.747, an optimism-corrected Brier score of 0.190, and an optimism-corrected calibration slope of 0.947. In the exploratory sensitivity analysis for severe intraventricular hemorrhage, the apparent area under the receiver operating characteristic curve was 0.843 (95% CI 0.750–0.937), and the Brier score was 0.039. We developed and internally validated a clinical prediction model for early intraventricular hemorrhage risk stratification in very preterm infants and/or very low birth weight infants with early-onset sepsis. The final model, based on four routinely available predictors, showed acceptable predictive performance and may provide a preliminary framework for early risk stratification in this high-risk subgroup. However, it should be regarded as an adjunct to, rather than a replacement for, routine cranial ultrasound screening, and external validation is required before broader clinical use. Retrospective Observational Study. Not applicable.