AI imaging helps enhance early diabetic retinopathy detection and improve patient outcomes: JAMA

A recent retrospective cohort study published in the JAMA Ophthalmology highlighted the slow adoption of artificial intelligence (AI) systems for detecting diabetic retinopathy (DR) in the United States, despite their proven efficacy. The study analyzed the use of Current Procedural Terminology (CPT) code 92229, established in January 2021 to support reimbursement for AI-based DR screening, across a database of over 107 million patients spanning 62 healthcare organizations.

The findings revealed that, out of nearly 5 million patients with diabetes examined from January 2019 to December 2023, only 4.2% underwent any ophthalmic imaging for DR. Within this subset, the use of AI-based imaging represented just 0.09% of total screenings, with only 3,440 patients utilizing the AI code 92229 since its inception. By 2023, the frequency of AI imaging had seen only a marginal increase, from 58.0 to 58.6 instances per 100,000 diabetic patients which indicated a slow adoption.

Also, traditional imaging techniques such as optical coherence tomography (OCT, CPT code 92134) and fundus photography (CPT code 92250) were more commonly used. OCT was performed in 80.3% of patients with at least one type of ophthalmic imaging, while fundus photography was utilized in 35.0% of cases. Traditional remote imaging (CPT codes 92227 and 92228) remained minimally used, accounting for only 1.0% and 2.5% of patients, respectively.

While the overall use of remote imaging methods surged by 90.16% between 2021 and 2023, AI-based screening remained disproportionately low. The data indicated that AI-based imaging had a higher referral rate to OCT (7.74%) when compared to traditional remote imaging (5.53%) by showing its potential for more targeted and effective DR detection. However, adoption hurdles such as cost, lack of awareness, and integration issues may be limiting widespread use. More than 80% of patients receiving AI-based imaging were concentrated in the South, a region comprising only 40% of other imaging modalities. Additionally, nearly half of the patients screened with AI systems were Black, in contrast to roughly a quarter seen in other imaging methods.

Despite FDA approval for AI-based systems like LumineticsCore and EyeArt, the broader implementation will require improved support for workflow integration and collaboration between primary care providers and ophthalmologists. The programs such as the Stanford Teleophthalmology Autonomous Testing and Universal Screening initiative highlight the importance of streamlined processes and patient-centered scheduling. Overall, the study points to a need for targeted strategies to boost the uptake of AI imaging, enhance early DR detection, and improve patient outcomes through more accessible and integrated screening solutions.

Source:

Shah, S. A., Sokol, J. T., Wai, K. M., Rahimy, E., Myung, D., Mruthyunjaya, P., & Parikh, R. (2024). Use of Artificial Intelligence–Based Detection of Diabetic Retinopathy in the US. In JAMA Ophthalmology. American Medical Association (AMA). https://doi.org/10.1001/jamaophthalmol.2024.4493