Development and internal validation of a day-28 clinical risk stratification model for retinopathy of prematurity in preterm infants: A retrospective cohort study.

ObjectiveThis study aimed to develop and internally validate a day-28 clinical risk stratification model for retinopathy of prematurity in preterm infants.MethodsThis single-center retrospective cohort study included preterm infants who underwent retinopathy of prematurity screening between January 2019 and January 2026. Postnatal day 28 was defined as a fixed analytic prediction landmark rather than the scheduled date of the first ophthalmic examination. Candidate predictors were restricted to routine perinatal and early neonatal variables available from birth through day 28. The primary outcome was any-stage retinopathy of prematurity diagnosed during subsequent screening. Severe retinopathy of prematurity and treatment-requiring retinopathy of prematurity were exploratory secondary outcomes. Multivariable logistic regression was used to develop the model. Discrimination, calibration, internal validation using 1000-bootstrap resamples, risk stratification, and decision curve analysis were performed.ResultsIn total, 760 preterm infants were included. Any-stage retinopathy of prematurity occurred in 209 infants (27.5%), severe retinopathy of prematurity in 64, and treatment-requiring retinopathy of prematurity in 28. The apparent area under the curve was 0.868 (bootstrap percentile 95% confidence interval: 0.839-0.895), and the optimism-corrected area under the curve was 0.862. Apparent calibration in the development cohort was summarized by an intercept of 0.000, slope of 1.000, observed-to-expected ratio of 1.000, Brier score of 0.122, and Hosmer-Lemeshow χ2 of 5.32 (p = 0.723). Observed retinopathy of prematurity rates were 5.6%, 24.8%, and 62.8% in the low-, intermediate-, and high-risk groups, respectively. No infant in the low-risk group developed treatment-requiring retinopathy of prematurity (0/373; negative predictive value: 100.0%, 95% confidence interval: 99.0-100.0).ConclusionThe day-28 clinical risk stratification model demonstrated promising apparent and internally validated discrimination and acceptable apparent calibration for identifying preterm infants at risk of retinopathy of prematurity. It may support follow-up planning; however, it must not replace guideline-based retinopathy of prematurity screening. External validation is required before routine clinical use.