Diagnostic Accuracy of An Artificial Intelligence Based mHEALTH Intervention for Cataract Detection: A Multi-Center Prospective Study in Tamil Nadu, India

Abstract

Background: E-Paarvai is an AI-based mHealth initiative piloted in Tamil Nadu, which enables frontline health workers to screen for cataract using a smartphone camera. The study aims to evaluate the diagnostic accuracy and reliability of e-Paarvai for cataract detection in primary care settings. Methodology: This prospective study was performed in 2022 in seven Upgraded Primary Health Centers across Tuticorin, India. Outpatients (age ≥ 50 years) without bilateral aphakia/pseudophakia, recruited by consecutive sampling, were each screened for cataract by e-Paarvai and an Ophthalmic assistant. Estimates of accuracy and reliability were reported along with their 95% confidence intervals (CI). Results: Among 337 participants (674 eyes) included in the analysis, 55 (16%) had unilateral and 168 (50%) participants had bilateral cataract on clinical eye examination. E-paarvai had a sensitivity and specificity of 83% and 53% at the subject level and 73% and 70.3% at the eye level respectively. Assuming 65% prevalence for cataract, PPV was 76% and 82%, while NPV was 62% and 58% in the per-subject and per-eye analysis respectively. The test-retest agreement was substantial with Kappa of 0.63. Conclusions: E-paarvai has an undeniable potential to improve detection and yield of cataract when implemented as a mass strategy in an eye care resource limited population.