Free papers detail | 11th Asia Pacific Tele Ophthalmology Society Symposium

Title
Real-world Artificial Intelligence-assisted Diabetic Retinopathy Screening in Primary Health Care Setting in Jakarta, Indonesia

Authors
Yeni Dwi Lestari, Ratna Sitompul, Indah Suci Widyahening, Dicky Levenus Tahapary, Muhammad Bayu Sasongko

Presenting
Yeni Dwi Lestari

PURPOSE:
Diabetic Retinopathy (DR) is one of the most common microvascular complications of Diabetes Mellitus (DM) and a leading cause of blindness. Early detection and timely intervention are essential. This study aimed to evaluate the integration of Artificial Intelligence (AI) to support DR screening in Primary Health Care (PHC) facilities in Jakarta, Indonesia.

METHODS:
DR screening was conducted at Non-Communicable Disease (NCD) Clinic at PHC by trained general practitioners or nurses. DM patient underwent basic eye examination and retina examination using fundus camera. The images were analyzed by AI software (Radr) to determine the need of referral for further management. Efficacy was measured by screening coverage, proportion DR patients and other eye diseases, and AI accuracy. Feasibility was measured by examination duration.

METHODS:

DR screening was conducted at Non-Communicable Disease (NCD) Clinic at PHC by trained general practitioners or nurses. DM patient underwent basic eye examination and retina examination using fundus camera. The images were analyzed by AI software (Radr) to determine the need of referral for further management. Efficacy was measured by screening coverage, proportion DR patients and other eye diseases, and AI accuracy. Feasibility was measured by examination duration.

RESULTS:
This implementation study was piloted in two PHCs, Pasar Rebo and Tanjung Priok, within one month in August. The coverage of DM patients screened was 31,1% at Pasar Rebo and 47,2% at Tanjung Priok. AI performance showed 75.7% sensitivity, 72.0% specificity, 9.9% positive predictive value, 98.6% negative predictive value, and 72.1% overall accuracy. The mean screening duration was 20.8 minutes (SD ± 8.7). Feasibility assessment using CSQ-18 questionnaire showed a mean score of 60.19 (range 18-72), indicating good satisfaction.

RESULTS:

This implementation study was piloted in two PHCs, Pasar Rebo and Tanjung Priok, within one month in August. The coverage of DM patients screened was 31,1% at Pasar Rebo and 47,2% at Tanjung Priok. AI performance showed 75.7% sensitivity, 72.0% specificity, 9.9% positive predictive value, 98.6% negative predictive value, and 72.1% overall accuracy. The mean screening duration was 20.8 minutes (SD ± 8.7). Feasibility assessment using CSQ-18 questionnaire showed a mean score of 60.19 (range 18-72), indicating good satisfaction.

CONCLUSIONS:
The AI-assisted DR screening model in PHC setting in Jakarta demonstrated good effectiveness, efficacy, and satisfaction score. Further study involving broader setting with longer period is needed to assess the impact. This finding can be used as a supporting data for DR and DM comprehensive services.