Validation of a screening score model to predict the development of retinopathy of prematurity

TL;DR

This study creates and validates a screening model to predict retinopathy of prematurity in preterm infants, aiming to improve detection in low- and middle-income countries.

Contribution

The study introduces two locally validated, risk-based screening models for ROP using Indonesian neonatal data, optimized for resource-limited settings.

Findings

The combined model achieved 84% sensitivity and 81% specificity in predicting ROP.

Pre-test probability of ROP increased to 0.76 after a positive screen and decreased to 0.13 after a negative result.

The models include predictors like oxygen exposure, intrauterine growth restriction, and socioeconomic status.

Abstract

Retinopathy of prematurity (ROP) is a leading cause of preventable blindness in preterm infants, with a disproportionate burden in low- and middle-income countries. Screening criteria from high-income settings may not be directly applicable in these contexts. We developed and validated two pragmatic risk-based screening models using multicenter Indonesian neonatal data: Model A (FiO₂-based) and Model B (SpO₂-based). Significant predictors included intrauterine growth restriction, oxygen exposure, exchange transfusion, and socioeconomic status. Internal validation showed moderate discrimination (AUC 0.719–0.732) with sensitivities of 77–86% and specificities of 44–58%. The bedside operational score form is presented for clinical use. External validation in 163 infants (gestational age 25–37 weeks, birth weight 600–2000 g) confirmed robust performance, with the combined rule (positive if either model was positive) achieving a sensitivity 84%, a specificity 81%, positive predictive value 76%, and negative predictive value 87%. The pre-test probability of ROP was 0.42, increasing to 0.76 after a positive screen and decreasing to 0.13 after a negative result. These findings support the use of locally validated risk-based scores as a practical complement to gestational age and birth weight criteria, optimizing ROP case finding in resource-limited settings.

The online version contains supplementary material available at 10.1038/s41598-025-24303-1.

Figures