Development and Internal Validation of an Early Prediction Model for a Composite Adverse Neonatal Outcome in Preterm Infants: A Prospective Cohort Study in Kazakhstan

Background: An estimated 13.4 million infants were born preterm worldwide in 2020, highlighting the need for early identification of those at risk of adverse neonatal outcomes. Preterm infants constitute a heterogeneous population in terms of maturity and clinical risk. A simple model based on variables available immediately after birth may facilitate the early identification of infants who require more intensive monitoring. Objective: This study aimed to develop and internally validate an early prediction model for a composite adverse neonatal outcome in preterm infants. Materials and Methods: This single-center prospective cohort study included 280 preterm infants born at 24 weeks to 36 weeks and 6 days of gestation. The primary endpoint was the occurrence, before hospital discharge, of at least one of the following events: death, grade III-IV intraventricular hemorrhage (IVH), grade III–IV periventricular leukomalacia (PVL), bronchopulmonary dysplasia (BPD), necrotizing enterocolitis (NEC), or culture-confirmed early- or late-onset neonatal sepsis. A binary logistic regression model was developed. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC ROC). Calibration was assessed graphically and by the Brier score, calibration intercept, and calibration slope. Internal validation was performed using 10-fold cross-validation and bootstrap optimism correction with model refitting in each of 2000 bootstrap samples. Results: The composite outcome occurred in 35 of 280 infants (12.5%). The final model included gestational age (odds ratio [OR] 0.736 per additional week; 95% confidence interval [CI] 0.614–0.882; p < 0.001) and the 5 min Apgar score (OR 0.447 per additional point; 95% CI 0.286–0.701; p < 0.001). The apparent AUC was 0.904 (95% bootstrap CI 0.848–0.952), the optimism-corrected AUC was 0.898, and the out-of-fold AUC from 10-fold cross-validation was 0.892. The out-of-fold Brier score was 0.066; the calibration intercept was −0.113, and the calibration slope was 0.925. The probability cutoff of 0.142 and the corresponding classification measures were considered exploratory. Conclusions: A two-variable model based on gestational age and the 5 min Apgar score demonstrated high internal discriminative ability. The model is intended for research evaluation and early risk stratification; independent external validation is required before clinical implementation.