Validation of STARZ scoring in very low birth weight (VLBW) neonates in the United States

Abstract

Abstract Background Acute kidney injury (AKI) is a significant complication in very low birth weight (VLBW) neonates, associated with increased morbidity, mortality, and prolonged hospitalization. Early identification of at-risk infants enables timely interventions. The STARZ score, developed and validated in India in near-term infants, is a clinical tool to predict neonatal AKI risk, but its applicability in early preterm United States (US) populations remains unknown. Methods A single-center retrospective cohort study of VLBW preterm infants ( ≤ 1500 g, ≤32 weeks gestation) was performed. The original STARZ model was applied to assess its predictive performance. AKI was defined using modified KDIGO criteria. Predictive performance of the STARZ score was evaluated using ROC curve analysis with sensitivity, specificity, positive and negative predictive values. Results In this VLBW cohort, the median gestational age and birth weight were 28 weeks (IQR 26–30) and 1100 g (IQR 790–1300), respectively. AKI occurred in 26 (9.6%) infants. A STARZ score ≥31.5 was observed in 47% of the cohort. While the score demonstrated acceptable discrimination (AUC 0.806) and high sensitivity (84.6%) with a high negative predictive value (97.2%), specificity (55.7%) and positive predictive value (16.8%) were low. Conclusions In a US VLBW population, the unmodified STARZ score demonstrates limited clinical utility for AKI risk stratification, thus limiting its routine clinical use and highlighting the need for population-specific validation and redevelopment of AKI risk models. Impact In a cohort of very low birth weight (VLBW) infants, the STARZ score demonstrated high sensitivity and negative predictive value for acute kidney injury (AKI) detection, but poor specificity and positive predictive value, limiting its clinical utility for risk stratification. These findings suggest that several components of the STARZ model may be too common among VLBW infants, resulting in overclassification of infants as “high risk” despite a low incidence of AKI. These results highlight the need for refinement of neonatal AKI prediction tools and support the development of VLBW-specific risk models tailored to their unique physiology and clinical characteristics.