External validation and updating of DIGIROP prediction models including parenteral nutrition for treatment-requiring retinopathy of prematurity in a Swedish national cohort

BACKGROUND/AIMS: This study represents an external validation with model updating of the online DIGIROP-Prescreen and DIGIROP-Screen 2.0 prediction models, incorporating ≥14 days of parenteral nutrition as an additional predictor, to identify infants requiring treatment for retinopathy of prematurity (ROP) in a contemporary Swedish cohort, estimate potential cost savings and compare DIGIROP performance with other models. METHODS: 2023 (n=1530) were evaluated. ROP and neonatal data were retrieved from Swedish quality registers. Sensitivity, specificity and the area under the receiver operating characteristic (AUC) curve were calculated. In the Västra Götaland Region cohort (n=245), the DIGIROP models were compared with WINROP and two Postnatal Growth and ROP (G-ROP) models. We estimated DIGIROP models' cost-saving. RESULTS: The mean gestational age was 27.7 (SD 2.2) weeks, birth weight 1029 (SD 317) g, 689 (45.0%) were girls and 85 (5.6%) infants received ROP treatment. For DIGIROP-Prescreen 2.0, the AUC was 0.89 (95% CI 0.86 to 0.92), sensitivity was 100% (95% CI 95.8% to 100%) and specificity 27.5% (95% CI 25.3% to 29.9%). For DIGIROP-Screen 2.0, the cumulative specificity increased from 27.7% to 67.8% between postnatal weeks 6-14, with a sensitivity of 100%. In total, ∼Int$750 000 could have been saved in screening costs in Sweden during 2021-2023 using DIGIROP models. DIGIROP-Screen at postnatal age 10 weeks had the same sensitivity but higher specificity than G-ROP. WINROP showed lower sensitivity but the highest specificity. CONCLUSION: DIGIROP 2.0 demonstrated high sensitivity and the most robust discrimination for treatment-requiring ROP in a contemporary Swedish cohort, compared with other models, with the potential to reduce unnecessary eye examinations and healthcare costs.