Diagnostic Accuracy of Wide-Field Digital Retinal Images in Retinopathy of Prematurity Detection: Systematic Review and Meta-Analysis

Abstract

Purpose To evaluate the diagnostic accuracy of the wide-field digital retinal imaging (WFDRI) for the detection of Retinopathy of Prematurity (ROP) in premature infants as compared to the binocular indirect ophthalmoscopy (BIO). Methods This systematic review and meta-analysis included the publications searched through PubMed (Medline), EMBASE, Scopus, Web of Science, Cochrane Library databases and Clinical Trials. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2, the hierarchical summary receiver operating characteristic, meta-regression, publication bias analyses, and the GRADE methodology for the certainty of the overall evidence were conducted. The pooled effect sizes of the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR) and diagnostic odds ratio (DOR) were calculated. Results Total sixteen eligible studies from 10 articles were included with total 2,537 image interpretations from 697 premature infants. Less than 50% risk of bias and low concern were found in each domain across all articles by QUADAS-2. The pooled effect sizes showed the sensitivity of 0.77 (95% confidence interval (C.I.): 0.69–0.84), specificity of 0.96 (95% C.I.: 0.92–0.98), PLR of 20.9 (95% C.I.: 10.2–42.5), NLR of 0.23 (95% C.I.: 0.17–0.33) and DOR of 89 (95% C.I.: 43–185) as compared to BIO. The income level, setting, mean/median birth weight and gestational age contributed to the significant differences in sensitivity (p < 0.001). No publication bias was found among these 16 studies. The GRADE quality of evidence showed moderate for the pooled sensitivity and high for the pooled specificity. Conclusions The diagnostic accuracy based on WFDRI is substantial and comparable to BIO, supporting its application in the ROP screening programs.