Integrating Clinical, Imaging, and Genetic Data for Severe ROP Prediction

Abstract

Purpose : Retinopathy of prematurity (ROP) is a primary cause of blindness in children, posing significant healthcare challenges. Early screening and detection of the risk of severe ROP are crucial for improving treatment outcomes. Previous studies have demonstrated that combining the vascular severity score (VSS) from retinal fundus images derived by artificial intelligence with clinical demographics can enhance the risk predictive model. Additionally, a genome-wide association study (GWAS) has identified certain genetic loci associated with an increased risk for ROP. In this study, we aim to integrate the VSS, clinical demographics, and the most significant genetic locus, rs2058019, to develop an advanced risk-predictive model for severe ROP.

Methods : Our study was conducted across eight North American centers and included 920 infants with genetic risk information. The VSS was derived from retinal fundus images from the first eye exam, obtained at 31 to 33 weeks postmenstrual age. We included infants with body weight < 1501 g and gestational age < 31 weeks, and who were not diagnosed with TR-ROP within the examination window. There are 420 infants had genetic risk information and VSS, met the inclusion criteria. We developed six logistic regression models with L2 regularization to test the combinations of gestational age, VSS, and genetic locus rs2058019. Five-fold cross-validation, repeated 200 times randomly, was used to validate the prediction models and enhance their reliability. The area under the receiver operating characteristic curve (AUROC) and the precision-recall curve (AUPRC) were used to evaluate model performance.

Results : The model combining gestational age, VSS, and rs2058019 significantly outperformed the models using gestational age and VSS, and gestational age alone (Table 1, mean AUPRC = 0.549±0.011 vs. 0.543±0.012 vs. 0.467±0.013). In statistical analyses using logistic regression, gestational age, VSS, and rs2058019 showed a statistically significant correlation with severe ROP, each with p-values < 0.01.

Conclusions : In our study, we developed a predictive model for severe ROP by integrating gestational age, VSS, and the genetic locus rs2058019. This model significantly outperformed others in predicting severe ROP. Our findings underscore the value of combining clinical, imaging, and genetic data for a more accurate and personalized screening process for ROP.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.