Diabetic retinopathy has an illustrious biography encompassing many stages. It was born as a result of the triumphs of medical management of diabetes mellitus, which extended the diabetic life expectancy, unveiling longer-term complications. It was nurtured to the age of therapeutics by landmark clinical trials. Its young adulthood was heralded by the recognition that medical control prevents a large fraction of progression and visual loss. Its middle age is the ongoing coordination of the multidisciplinary efforts of education and screening, which have enhanced the use of these tools, resulting in a marked reduction of visual loss.
A key element driving these developments and efforts is the ability to visualize and image diabetic retinopathy for staging purposes in a given patient. Fundus photography enhanced (and arguably surpassed) ophthalmoscopy as a measure of retinopathy, leading to the development of a robust, standardized grading scheme that has served as a valid benchmark for studying and treating diabetic retinopathy for the past 40 years.
The digital age has left its imprint upon imaging in ways that need no elaboration. Film has become relegated to museums by digital photography and imaging. That digital fundus photography is at least equivalent to film has allowed continued application of the classification system. However, the rigorous 7-stereofield photographic set that was an invaluable research tool is generally recognized to be too cumbersome for everyday clinical applicability. Hence, a simpler iteration of the grading scheme has emerged for practical use in the clinic, requiring fewer images, usually not procured in stereo. Mosaic, composite photographs of the fundus are attractive for their relative simplicity in evaluating the stage of retinopathy. Furthermore, the ability to obtain representative images through an undilated pupil has further widened the potential scope of screening for diabetic retinopathy—effectively taking the diagnosing of diabetic retinopathy out of the ophthalmologist’s office and into general medical practice.
The study by Silva and associates demonstrates the feasibility of a commercially available, nonmydriatic, ultrawide-field retinal imaging system for detecting diabetic retinopathy. Their methodology is notable because of their use of a feature that contracts the acquired field to 100 degrees but enables better resolution and stereo images. Their hypothesis was that this provides suitable imaging for accurate screening. The 100-degree field closely approximates the area evaluated with the 7 stereofields evaluated in ETDRS protocols. Their findings are based on well-designed and well-executed study methods and advocate strongly for the utility of such screening, which is gaining traction especially in non-ophthalmology circles.
The study raises several questions. Its scope was to determine whether this system and strategy is a valid research tool in a cohort of patients known to have, mostly, nonproliferative diabetic retinopathy. The concern and responsibility of the clinician is whether it is a practical clinical tool. Part and parcel with the endorsement of this method is whether it is cost-effective.
The study’s principal purpose was to determine the accuracy of the nonmydriatic, stereo Optomap image relative to film-based ETDRS protocol photographs. The authors report an 84% exact correlation of stages, and 91% agreement within 1 stage, better than previous cited reports. While this is impressive, it must be remembered that experienced, trained graders interpreted the photographs—a level of accuracy that might exceed that of even an ophthalmologist, and almost certainly a non-ophthalmologist. Similarly, the images were procured by skilled photographers experienced with the rigorous standards of clinical trials; again, this is not likely to be duplicated “in the field,” especially in “underserved areas,” which is the authors’ stated target for such tools. Despite this rigor, it was particularly distressing that 27% of the 52 eyes with proliferative diabetic retinopathy, arguably the most important stage to identify in a screening exercise, were missed by this system. While this is similar to the accuracy of the clinician examining the eye, it is below the ETDRS photographic standard, a serious shortcoming as a research instrument and of modest concern as a clinical tool, especially considering the above-stated context about photographic and grader competence. While it is meritable that earlier stages of diabetic retinopathy had decent correlation to the standard, from a sight preservation standpoint this is less critical to distinguish than identifying eyes with undiagnosed proliferative retinopathy.
A general trend in medicine has been to rely increasingly on tests and imaging rather than using such tools to augment the totality of the clinical picture, the base of which is the clinical history and physical examination. Thus, it is important to emphasize that a “one-stop shopping” use may place diabetic retinopathy management out of its proper clinical context. Indeed, this has been a principal reason the ophthalmology establishment has generally resisted embracing the placement of such technology in non-ophthalmology offices. Reports that photographic screening in medical and endocrine clinics allow accurate detection of diabetic retinopathy abound. Algorithms to allow “automated” evaluation of these images have been proposed, similar to EKG machine interpretations, which, while helpful in practice, have a limit to their usefulness. While it is encouraging that otherwise undiagnosed patients might be brought under ophthalmic care as a result, or that the actual photograph might even be a suitable prop to encourage better compliance with the medical regimen, the risk of underdiagnosing other ocular conditions by skipping ophthalmologic evaluation cannot be overestimated.
Another consideration as to the clinical utility of this approach is how best to apply it. The current study was applied to patients known to have diabetic retinopathy; perhaps identifying certain higher-risk groups would be an important first step in optimizing clinical utility. These same considerations apply to the use of this approach in a telemedicine setting: the tools can be powerful, but applying them to the proper populations in the proper way is tantamount to usefulness.
A parallel consideration is the cost-effectiveness of applying this or any photographic screening, which, clearly, was beyond the scope of the Silva study. The sample space of this study was patients known to have retinopathy. Adding $50 to $100 to the screening of 26 million diabetics in the United States, to say nothing of the estimated 79 million with prediabetes (or the 200 million or more diabetics worldwide), would add at least a couple billion dollars to the aggregate health expenditures annually, which would represent at least a 1% increase in the entire expenditure for diabetes. This further emphasizes the importance of targeting appropriate cohorts.
Nevertheless, this study testifies to the capabilities and triumphs of the modern digital era. As with any new toy, we must now prudently address how to incorporate this into the medical armamentarium. Successful implementation might even signal the retirement of diabetic retinopathy as a threat to sight!