Abstract / Summary
Background: To evaluate the diagnostic accuracy and grading concordance of smartphone-based non-mydriatic fundus photography (SFP) compared with slit-lamp biomicroscopic indirect ophthalmoscopy (BIO) for screening diabetic retinopathy (DR) in a hospital-based diabetic population in India. Materials and Methods: In this prospective observational study conducted over 12 months at a tertiary-care teaching hospital of NMCH, we enrolled 186 consecutive patients with type 2 diabetes mellitus (T2DM). Each patient underwent SFP using a portable smartphone-based fundus imaging device followed by standard 90-diopter slit-lamp BIO by a masked retinal specialist. DR was graded using the International Clinical Diabetic Retinopathy (ICDR) Disease Severity Scale. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and weighted kappa (κ) were computed with BIO as the reference standard. Results: Of 186 participants (mean age 54.3 ± 9.7 years; 52.7% male), 74 (39.8%) had DR on reference examination. Image quality was adequate in 94.1% of eyes. For detection of any DR, SFP yielded sensitivity 87.8% (95% CI 77.8–94.3), specificity 89.3% (81.8–94.4), PPV 84.2%, NPV 91.7%, and accuracy 88.7%. For moderate-to-severe NPDR and PDR, sensitivity exceeded 91% and specificity exceeded 97%. Overall weighted κ for severity grading was 0.891. On multivariate logistic regression, longer diabetes duration and higher HbA1c were the strongest independent predictors of DR. Conclusion: Smartphone-based fundus photography demonstrates high diagnostic accuracy and near-perfect grading agreement with slit-lamp BIO for DR detection. It represents a cost-effective, portable alternative for DR screening in resource-limited settings of India