Teleretinal screening programs prove a cost-effective option in preventing blindness related to diabetic retinopathy.
Diabetic retinopathy (DR) is the leading cause of permanent blindness in working-age adults, and the number of patients with diabetes worldwide is expected to reach 693 million by 2045.1 Consequently, it is necessary to find a more effective means of screening patients for diabetic retinopathy.2
A viable testing option is teleretinal screening for diabetic retinopathy (TSDR), a technology that has been used successfully in a variety of cultures and demographic settings, has been well reported in recent years,3-9 and whose cost effectiveness has also been confirmed.10-14
Typically, referable diabetic retinopathy (RDR) is defined as diabetic retinopathy worse than mild nonproliferative diabetic retinopathy and/or macular edema, based on the international classification of diabetic retinopathy.
In longstanding (10-20 years) TSDR programs, the incidence of sight-threatening diabetic retinopathy decreases over time, suggesting the program’s effectiveness and the possible need to refine the indications for screening in those programs.15-16
Recent publications have discussed the use of artificial intelligence (AI) as well as different cameras, including tabletop and handheld devices, to further TSDR programs.17-26
Multiple deep learning image assessment software have been developed.26-32 To date, two AI-based autonomous cameras have been approved by the FDA (IDx-DR, Digital Diagnostics; Eyenuk), and several other automated AI-based algorithms with varying degrees of sensitivity and specificity are available worldwide.33
Sensitivity is a measure of how well a test can identify true positives and specificity, a measure of how well it can identify true negatives. In both diagnostic and screening tests, there is usually a tradeoff between sensitivity and specificity, such that higher sensitivities will mean lower specificities and vice versa. Sensitivity (true positive rate) is determined by dividing the number of true positives by the total number of positives. Specificity (true negative rate) is the number of true negatives divided by the total number of negatives. In TSDR programs, a high specificity rate is more important than a high sensitivity rate to ensure that patients with the disease are seen by a specialist.
Although a high sensitivity rate could ideally keep individuals without disease out of the specialist’s office, it is better to see “false positives” (due to lower sensitivity) than not see “false negatives” (due to low specificity). TSDR programs with high specificity will do a better job of not missing patients with disease, thus meeting its main objective of identifying patients in need of a DR specialist.
Three different fundus images have been used in TSDR programs. The gold standard is the Early Treatment Diabetic Retinopathy Study (ETDRS) 7 standard field (7SF) image. However, the burden of multiple image capture makes 7SF photographs unrealistic for a TSDR program. The most common image capture technique is less than 7SF, which usually involves an image of the posterior pole that includes the disc, macula, and vascular arcades. Alternatively, ultra-widefield images, which capture over 200 degrees of retina, have also been used in some DR screening programs.34 These techniques were found to be essentially similar in effectiveness, although the less than 7SF may have a higher number of ungradable images.35
Autonomous AI software is gradually being integrated into TSDR programs, but the current standard is point-of-care image capture with a fundus camera in a primary care setting, followed by a review of the image by an eye care provider via a web-based platform. One advantage of this method over the FDA-approved autonomous interpretation algorithms is that it allows screening not only for diabetic eye disease but also for such other pathologies as cataracts (hazy view), glaucoma (increased optic disc cupping), macular holes, epiretinal membranes, and macular degenerations, making it possible for patients with sight-threatening disease to triaged and, depending on diagnosis, contacted for a timely examination.
Reading centres vary depending upon the health care system; some readers are located in the same community as the camera, and others are centralized in a distant centre. A study confirmed some utility in centralizing the reading centre.36 But there are benefits to keeping the reading centre in the same community as the cameras, including increased accountability among patients, primary care physicians, reading centres, and ophthalmologists. A systems-based approach allows all stakeholders to give input when determining processes that improve adherence rates. One study suggests that local collaboration may increase adherence rates.37 Local relationships and referral patterns enable better communication among stakeholders, and community pride may promote cooperation in reducing risk of vision loss in these patients.
At present, AI algorithms are FDA-approved only for the detection of diabetic eye disease (DED) and/or RDR. Consequently, other pathologies will be missed by screening programs that employ only the AI component. This limitation may become more important as teleretinal imaging expands to screen for other disease states.
A criticism of TSDR programs is lack of adherence to the recommended follow-up exams among patients with RDR, even though adherence rates range widely, from 9.5% to 81.9% depending on the study.37-38 Adherence rates for these follow-up exams are determined by dividing the ratio of the number of patients seen by a specialist by the number of patients diagnosed with RDR. In one study, automated retinal image analysis (performed in a primary care setting and allowing for immediate interpretive feedback to the patient) improved adherence by more than 36 percentage points, from 18.7% to 55.4%.22
Consequently, AI cameras may not only provide instantaneous diagnostic information to patients, but also increase adherence rates among patients with RDR. Conversely, barriers to full utilization exist in some programs that make implementation difficult.39 The incidence of RDR increases with failure to attend screening appointments: 5-9 consecutive misses increase RDR likelihood from 4% to 15%, and 10 or more consecutive misses increase RDR likelihood to 20%.40
Reports confirm TSDR is able to detect macular oedema, with one report confirming the presence of lipid, which increases the chances of macular oedema. However, sensitivity and specificity in diagnosing macular oedema varies widely among reports.14, 41-42
Initially, these TSDR programs were focused on underserved communities, be they uninsured, rural, or urban public health systems. More recently, insured population groups have been studied, likewise with promising results. TSDR is effective for all population subgroups. As one would expect, regions and/or communities with lower access to health care have higher DR rates and, consequently, potentially would benefit more from the deployment of TSDR.14
TSDR is likely to become the standard strategy for screening diabetic patients for DR in most clinical settings around the world. Ophthalmologists engaged in the fight to prevent blindness from DR will be well-served by understanding the different aspects of TSDR and their communities by being involved in its deployment.