Performance of a Deep Learning Diabetic Retinopathy Algorithm in India
An erratum to this article was published (JAMA Netw Open, 1 Apr 2025 ↗)
Authors' conclusion
In this cross-sectional study investigating the clinical performance of ARDA, sensitivity and specificity for severe NPDR and PDR exceeded 96% and caught 100% of patients with severe NPDR and PDR for ophthalmology referral. This preliminary large-scale postmarketing report of the performance of ARDA after screening 600 000 patients in India underscores the importance of monitoring and publication an algorithm's clinical performance, consistent with recommendations by regulatory bodies.
Study context
- Population
- Patients screened using automated retinal disease assessment (ARDA) at Aravind Eye Hospital, 45 sites in Southern India, randomly selected January 1, 2019–July 31, 2023; mean (SD) age 55.2 (11.9) years, 50.1% male3,941 with gradable photographs
- Participants
- 4,537 patients
- Comparison
- ARDA vs adjudicated grades by US ophthalmologists
- Outcome
- Sensitivity and specificity of ARDA for severe NPDR or PDR
Methodological detailscross-sectional, postdeployment, multicentre (45 sites)
- Cross-sectional
- Postdeployment
- Multicentre (45 sites)
Performance
As reported in the abstract
| Test or model | Sensitivity | Specificity | PPV | NPV | Clinically important miss rate |
|---|---|---|---|---|---|
| ARDA | 97.0% (95% CI, 92.6%-99.2%) | 96.4% (95% CI, 95.7%-97.0%) | 50.7% | 99.9% | 0% |
ARDA
Further findings
| Outcome | ARDA |
|---|---|
| STDR (DME or severe NPDR or PDR) | |
| Sensitivity | 95.9% (95% CI, 93.0%-97.4%) |
| Specificity | 94.9% (95% CI, 94.1%-95.7%) |
| PPV | 67.9% |
| NPV | 99.5% |
Prevalence among patients with gradable photographs
- Any DR
- 683 (17.3%)
- Severe nonproliferative DR (NPDR) or proliferative DR (PDR)
- 146 (3.7%)
- PDR
- 109 (2.8%)
- Sight-threatening DR (STDR)
- 398 (10.1%)
Original publication
Composed by AI from the published abstract and checked against it. How Briefs are made →