- Diabetic retinopathy (DR) is a leading cause of preventable blindness globally, particularly in individuals with diabetes, with an expected rise to over 550 million diabetic patients by 2030.
- Early detection of DR is critical to prevent vision loss and severe visual impairment.
- IDX-DR is the first FDA-approved autonomous artificial intelligence (AI) system for detecting diabetic eye disease, granted De Novo clearance in 2018.
- IDX-DR utilizes deep learning algorithms to analyze retinal images, enhancing screening efficiency and accessibility, especially in resource-limited settings.
- Primary care integration: IDX-DR provides immediate results without clinician interpretation, enabling its use in primary care settings for routine diabetic care.
- Potential benefits include improved early detection, expanded access in underserved areas, and reduced healthcare costs by minimizing specialist referrals.
- Diagnostic accuracy metrics:
- Pooled sensitivity: 0.95 (95% CI: 0.82-0.99), indicating a high ability to detect true DR cases.
- Pooled specificity: 0.91 (95% CI: 0.84-0.95), showing a moderate rate of false positives.
- Area Under the Curve (AUC) of the Summary Receiver Operating Characteristic (SROC) curve is 0.95, indicating excellent diagnostic performance.
- Q index*: 0.90, where sensitivity equals specificity, reinforcing robust diagnostic capability.
- Heterogeneity in sensitivity (I² = 99.4%) and specificity (I² = 99.1%) suggests variability influenced by study design, population demographics, DR severity, and imaging protocols.
- Subgroup analysis:
- Prospective/cross-sectional studies: Sensitivity 0.92 (95% CI: 0.61-0.99), Specificity 0.90 (95% CI: 0.80-0.96).
- Retrospective human sample-based studies: Sensitivity 0.99 (95% CI: 0.18-1.00), Specificity 0.92 (95% CI: 0.88-0.94).
- Retrospective image sample-based studies: Sensitivity 0.97 (95% CI: 0.77-1.00), Specificity 0.91 (95% CI: 0.69-0.98).
- Study characteristics:
- Included 13 studies from countries like the USA, Switzerland, Poland, Italy, Austria, Spain, China, and the Netherlands.
- Total participants: 12,233 adults (≥18 years) with type 1 or type 2 diabetes.
- Reference standards: Early Treatment Diabetic Retinopathy Study (ETDRS), International Clinical Diabetic Retinopathy (ICDR) scale, or EURODIAB criteria, graded by retinal specialists.
- Imaging: Primarily Topcon non-mydriatic fundus cameras (e.g., NW-400, TRC NW200, TRC NW100); some used optical coherence tomography (OCT).
- Limitations:
- High heterogeneity across studies due to differences in DR severity, imaging protocols, and reference standards.
- Lack of data from high-diabetes-burden countries like India.
- No long-term clinical outcome data (e.g., vision preservation, quality of life).
- Ethical considerations: Concerns include patient data privacy, consent, and the need for secure, anonymized data handling to maintain public trust.
- False positives can lead to patient anxiety, unnecessary tests, and increased costs, particularly in resource-limited settings.
- Mitigation strategies: Pair IDX-DR with secondary evaluations by healthcare professionals or a tiered system for high-risk cases.
- Socioeconomic impacts: Initial costs (~$13,000 for IDX-DR system, $25 per screening) may limit adoption in low-income regions, necessitating public-private partnerships or subsidies.
- Cost-effectiveness: IDX-DR may reduce healthcare costs by ~23.3% compared to traditional methods, with an incremental cost-utility ratio (ICUR) of ~$258,721.81 over 5 years.
- Comparison with other AI tools:
- EyeArt: Sensitivity 96%, Specificity 88%.
- AEYE-DS: High accuracy for DR and diabetic macular edema (DME) detection.
- Retinalyze and Medios AI: Notable for utility in various settings.
- Future research needs:
- Long-term outcomes (e.g., vision preservation, treatment adherence).
- Standardization of imaging and diagnostic protocols.
- Cost-effectiveness in low-resource settings.
- Comparative studies of all FDA-approved AI tools for DR screening.
Citation address of the uploaded paper:
Khan, Z., Gaidhane, A. M., Singh, M., et al. (Accepted for publication February 17, 2025). Diagnostic Accuracy of IDX-DR for Detecting Diabetic Retinopathy: A Systematic Review and Meta-Analysis. American Journal of Ophthalmology, 273. Supplemental material available at AJO.com.