Machine learning framework predicts retinopathy of prematurity in Indian newborns

Researchers developed a two-step machine learning framework to predict retinopathy of prematurity (ROP) in newborns, achieving 84.9% sensitivity for treatment-requiring cases in a prospective validation cohort, though 78 affected infants were missed. The models were trained on 23,404 medical records from 9,205 infants across 25 newborn care units in Odisha, India, using structured clinical variables rather than retinal images. Random Forest performed best for step one, predicting any ROP development, while LightGBM excelled at step two, identifying which examined infants needed treatment. The framework is proposed as adjunctive prioritization support within existing screening guidelines, not a replacement for specialist ophthalmic examinations. Further optimization, calibration, and external validation are required before clinical implementation, particularly for resource-constrained settings.