THESS ‐ ROP study: Development of a prediction model for treatment‐warranted retinopathy of prematurity and external validation of postnatal weight‐gain‐based models in Greece

Purpose To develop THESS‐ROP, a novel prediction model for treatment‐warranted retinopathy of prematurity (TW‐ROP) and compare its performance with four established models in a Greek cohort of consecutively screened preterm infants. Methods This ambispective cohort study included all 526 infants screened for ROP (2016–2022) at the 2nd Neonatology Department of Aristotle University of Thessaloniki, a tertiary neonatal intensive care unit in Greece. The primary outcome was TW‐ROP (type 1 or 2 or aggressive ROP). Secondary outcomes included type 1 ROP and ROP treatment. A penalized ridge logistic regression model was developed using gestational age, birth weight, sepsis, and hydrocephalus up to 4 weeks postnatal age. Missing data were addressed using 10‐fold multiple imputation. Internal validation used bootstrap resampling with optimism correction. External validation of CO‐ROP, G‐ROP, G‐ROP 180 g, and ROPScore was performed in the same cohort. Performance was assessed using area under the curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value. Results TW‐ROP developed in 34 infants (6.5%; 24 type 1, 10 type 2), with 29 receiving treatment. THESS‐ROP demonstrated excellent discrimination [optimism‐corrected AUC 0.935, 95% CI (0.905–0.965)], pooled sensitivity 98.0% and pooled specificity 58.0%. Sensitivity reached 100% for type 1 ROP and treated ROP. In external validation, CO‐ROP, G‐ROP, G‐ROP 180 g, and ROPScore achieved pooled sensitivities of 90.9%, 98.8%, 98.8%, and 100.0% with specificities of 63.4%, 34.4%, 26.1%, and 62.1%, respectively. Conclusion THESS‐ROP model, incorporating easily available clinical data, enables effective early risk stratification at 4 weeks PNA and may optimize ROP screening. External validation in independent populations is required before clinical implementation.