It is estimated that by the year 2040, globally there will be 642 million persons with diabetes mellitus, and half of this population will develop diabetic retinopathy. 8 Despite recent advances in diagnostic capabilities and the availability of highly effective evidence-based treatment options, vision loss from complications of diabetic retinopathy remains the leading cause of legal blindness in working-age adults in most Western societies. 9 This is due to multiple factors, including that early stages of disease are asymptomatic and patients are unaware of the need for screening, poor access to care, and patient inconvenience. 10 As a result only about half of all patients with diabetes mellitus undergo annual retinal examinations as is recommended by widely accepted guidelines. 11 Appropriate screening and timely treatment of vision-threatening diabetic eye disease significantly affect vision outcomes in patients with diabetes, and most cases of severe vision loss are avoidable. Ocular telehealth technology is now well established as an effective adjunct method to increase adherence to guidelines for assessment of diabetic retinopathy and achieve all three Triple Aim goals.
Numerous programs have demonstrated that diabetic retinopathy telemedicine programs with acquisition of retinal images at the point of care in primary care locations can effectively increase rates of eye examinations among patients with diabetes and even reduce the rate of blindness and vision loss. 12,13 The benefits of ocular telehealth programs to remotely evaluate for diabetic retinopathy are perhaps best exemplified by the United Kingdom National Health Service (UK NHS), which reports that, for the first time in 5 decades, the leading cause of blindness in working-age adults in England and Wales is no longer diabetic retinopathy. 14 This outcome was largely achieved with the nationwide telemedicine diabetic retinopathy screening program, which is in place in England and Wales and has attained a very high uptake rate.
Both the American Academy of Ophthalmology (AAO) and the American Telemedicine Association (ATA) have stressed the need for validation of diabetic retinopathy screening programs. 15,16 The accepted gold standard for identifying diabetic retinopathy remains the Early Treatment Diabetic Retinopathy Study (ETDRS) 30-degree, stereoscopic, seven-standard field, color, 35-mm slides and, more recently digital images, evaluated by experienced readers. 17 Diagnostic accuracy is often measured with sensitivity (true positive rate) and specificity (true negative rate) with comparison to a gold standard reference. Additional standard statistical measures to assess reliability and reproducibility include kappa values for agreement of diagnosis, false-positive and false-negative readings, and positive predictive value and negative predictive value for identifying levels of retinopathy and macular edema. 16
In the ATA Telehealth Practice Recommendations for Diabetic Retinopathy, four categories of validation are described for diabetic retinopathy telehealth programs using ETDRS seven-standard field photographs as the reference standard. 16 Category 1 validation indicates a system that can identify patients with no or minimal diabetic retinopathy from those who have more than minimal diabetic retinopathy. Category 2 validation indicates a system that can determine accurately if vision-threatening diabetic retinopathy (diabetic macular edema or severe nonproliferative or worse levels of diabetic retinopathy) is present or not. Ocular telehealth programs in the United States are most often Category 2 programs. Category 3 validation indicates a system that can identify ETDRS-defined levels of diabetic retinopathy and macular edema sufficiently accurately to allow patient management and treatment. A Category 4 validated system matches or exceeds ETDRS photographs in the ability to identify levels of diabetic retinopathy and diabetic macular edema. The category of validation required for an individual telemedicine diabetic retinopathy program depends on its specific goals and objectives.
In the UK NHS, a sensitivity of 80% and specificity of 95% has been suggested as a minimum performance threshold. 18 However, sensitivity and specificity measures will vary by the clinically relevant diagnostic target (ie, sensitivity and specificity for the presence of vision-threatening diabetic retinopathy versus presence of any diabetic retinopathy will be different). A meta-analysis of validated diabetic retinopathy telemedicine programs described in the literature revealed that the pooled sensitivity of telemedicine exceeded 80% in detecting the absence of diabetic retinopathy and low- or high-risk proliferative diabetic retinopathy. 19 It exceeded 70% in detecting mild or moderate nonproliferative diabetic retinopathy and clinically significant macular edema. The pooled sensitivity was 53% in detecting severe nonproliferative diabetic retinopathy. The pooled specificity of telemedicine was 89% or better. Diagnostic accuracy is generally higher with digital images obtained through dilated pupils and, in particular, with wide-angle imaging devices. 2,20 Although it is not the intended purpose of telemedicine diabetic retinopathy programs to evaluate for other ocular diseases, nondiabetic retinopathy ocular findings are relatively common. 21 Protocols must be in place to determine which findings would require referral and evaluation even in the absence of diabetic retinopathy.
Accuracy and reliability of ocular telehealth programs for diabetic retinopathy are also affected by the unreadable image rate. Published reports of unreadable image rates in various telehealth diabetic retinopathy programs range from 3% to 35%. 22 Image quality is affected by multiple factors, including age, media opacity such as cataract and vitreous hemorrhage, and small pupil size. 20,22 Telehealth diabetic retinopathy programs may utilize pupillary dilation, whereas others perform imaging with a nonmydriatic camera and undilated pupils. Some programs also utilize selective mydriasis based on age or image quality. Generally, a higher unreadable rate has been reported with undilated pupils, especially in older individuals who also have some degree of cataract formation. 16 In addition to patient inconvenience and time constraints, the potential risk of angle-closure glaucoma is sometimes cited as a reason for avoiding pupil dilation in the primary care setting. However, the risk of inducing angle-closure glaucoma with low-dose mydriatics is minimal with no reported cases in a large meta-analysis. 23 Nonetheless, offices with mydriatic systems should have a defined protocol to recognize and address this potentially serious complication.
State-of-the-art retinal imaging devices generally do not require a trained ophthalmic photographer to obtain quality retinal images. However, screening in the primary care setting with imaging obtained by office personnel rather than a dedicated imager may result in imagers who acquire only a small number of images at infrequent intervals, which may also affect the overall image quality. Personnel acquiring images should have demonstrated qualifications for obtaining images of adequate diagnostic quality and need to be monitored on an ongoing basis to ensure consistent image quality.
Another essential element of telehealth diabetic retinopathy programs is the remote reading center where images are evaluated. Image readers range from nonlicensed technical readers to ophthalmologists with specialty training in retina. In the United States, technical readers have been successfully used for many years at academic reading centers to support clinical trials. They are less often used for community-based telemedicine diabetic retinopathy programs. Technical readers may add business and operational efficiencies to a reading center but also require careful supervision by eye care providers, standardized procedures for training, and effective quality measures. 24 Image readers need to be qualified and, ideally, certified to grade and interpret retinal images, especially if nonlicensed nonophthalmic personnel are utilized. Protocols need to be in place to assess image reading capabilities of individual readers with intragrader and intergrader agreement, in an ongoing fashion with a sample size appropriate to the size of the screening program.
Quality assurance processes provide guidance on methods to ensure ongoing quality and performance improvement. They may be quite complex and difficult to execute for telemedicine diabetic retinopathy surveillance programs. Although no uniformly adopted quality assurance guidelines are in place in the United States, the UK NHS has developed comprehensive protocols for their national diabetic retinopathy screening program to assess performance and safety, including reading centers. 25 Quality assurance metrics and performance indicators will vary based on program goals and objectives. Certain clinical outcome measures, however, are important for all programs, such as the eye examination rate among patients with diabetes, the ungradable image rate, and image reviewer performance measures. Reasonable turnaround times for image evaluation and return of the report to the ordering physician need to be set and monitored. Additionally, recommendations need to be established for referral urgency and timing for varying levels of disease severity. More difficult to measure but equally important for an effective diabetic retinopathy program is the rate of compliance with referral recommendations, rate of intervention for advanced disease, and ultimately, the rate of vision loss in the population being screened. Only if such data are systematically collected and analyzed will we realize the full potential of telemedicine for diabetic retinopathy and how it aligns with the Triple Aim health care policy.
As one of the Triple Aim health policy objectives, comprehensive quality assurance programs will also assess patient satisfaction. Multiple studies show high levels of acceptance or preference for telemedicine as an alternative to conventional dilated retinal examination for diabetic retinopathy based on multiple factors, including patient convenience, cost, and perceived quality. 26
The diagnostic accuracy of using digital imaging and the high sensitivity of telemedicine systems to detect clinical levels of diabetic retinopathy indicate that they may appropriately be widely utilized for diabetic retinopathy evaluation. A systematic review of the economic evidence for diabetic retinopathy screening reveals that all studies have demonstrated it is also a cost-effective modality. 27–29 Furthermore, cost avoidance for health systems may be appreciated if disease is detected early and prompt treatment is initiated. However, significant barriers to widescale implementation of telemedicine diabetic retinopathy screening remain in the United States. Economics are a substantial concern, with the high cost of acquiring and operating imaging equipment coupled with unreliable reimbursement for the diagnostic procedure. Although placing imaging equipment in all primary care settings would greatly benefit noncompliant patients and improve disease detection across the board, the expense of such an undertaking is not realistic in the current health care economies.
Reimbursement for telemedicine diabetic retinopathy screening remains unclear. Most commercial payors in the United States understand the benefits gained by timely evaluation for diabetic retinopathy and are covering telemedicine services for this purpose. The Centers for Medicare and Medicaid Services, however, do not have a uniform coverage policy. Most telemedicine diabetic retinopathy service providers are billing using the Current Procedural Terminology (CPT) code for fundus photography (92250) with varying reimbursement success. Retinal telescreening codes also exist, but are often not applicable or useful. CPT code 92227 describes “remote imaging for detection of retinal disease (e.g., retinopathy in a patient with diabetes) with analysis and report under physician supervision,” whereas CPT code 92228 is designated for “remote imaging for monitoring and management of active retinal disease (e.g., diabetic retinopathy) with physician review, interpretation and report.” 30 The descriptor for 92227 is not accurate for most telemedicine diabetic retinopathy screening programs, as the image interpretation is generally performed by a licensed eye care provider rather than by a nonlicensed technician under physician supervision. Because no physician work is involved, reimbursement for CPT 92227 is minimal and does not cover the actual costs of most programs. On the other hand, CPT code 92228 is assigned reasonable reimbursement but is only applicable if diabetic retinopathy (or other active eye disease) is present. As expected, most screening studies report a high percentage of patients have no appreciable diabetic retinopathy and therefore CPT code 92228 often does not come into play. Fortunately, as telemedicine diabetic retinopathy systems become more widely accepted and acknowledged as a useful adjunct to in-person examination, reimbursement is becoming more consistent. Additionally, pay-for-performance health care measures are aligned with the Triple Aim policies and provide incentives and/or penalties for achieving or missing target metrics for diabetic retinopathy screening rates among the population with diabetes mellitus.
Telemedicine diabetic retinopathy programs have matured rapidly in recent years and are being increasingly used worldwide to increase the rate of annual eye evaluations for patients with diabetes. Although a properly integrated telemedicine retinal imaging system does not replace a comprehensive eye examination, the resultant improved surveillance for diabetic retinopathy results in greater timely access to effective treatments and realization of improved clinical outcomes.
Due to the high current global prevalence of diabetes, it is estimated that seven retinal examinations must occur every second to fulfill guidelines of annual retinal screening for all diabetic patients. 8 Even though telemedicine has the potential of increasing efficiency of access for people, it can increase the burden of image interpretation by physicians and graders. Therefore, automated computer-based grading systems were developed to meet the increasing eye care demand. These systems are capable of distinguishing normal structures from diabetic lesions, such as hemorrhages or cotton wool spots. 28 Based on the type, severity, and extent, images are graded for the presence or absence of diabetic retinopathy. Furthermore, automated systems can be adjusted in order to improve the balance between sensitivity and specificity. To minimize false-negatives, a high sensitivity is a desirable characteristic of such a computerized system. The disadvantage would be a lower specificity, which requires more images to be regraded by physicians, reducing efficiency and increasing the cost of screening programs. However, such systems would serve a prescreening function and on a large scale, still decrease overall workload on the reading center staff. 31,32