With diabetes rates projected to rise sharply, automated retinal screening may represent an attractive low-cost option to meet the growing demand for routine screening services.
Nearly 4 in 10 Americans with diabetes currently fail to undergo recommended annual retinal exams, resulting in tens of thousands of cases of blindness that could have been prevented. Advances in automated retinal disease detection could greatly reduce the burden of labor-intensive dilated retinal examinations by ophthalmologists and optometrists and deliver diagnostic services at lower cost. As the current availability of ophthalmologists and optometrists is inadequate to screen all patients at risk every year, automated screening systems deployed in primary care settings and even in patients’ homes could fill the current gap in supply. Expanding screens to all patients at risk by switching to automated detection systems would in turn yield significantly higher rates of detecting and treating diabetic retinopathy per dilated retinal examination. Fewer diabetic patients would develop complications such as blindness, while ophthalmologists could focus on more complex cases.
Am J Manag Care. 2014;20(11 Spec No. 17):eSP48-eSP52Automated image-based detection of retinal disease is rapid and can obviate the labor-intensive method of image interpretation and data entry by physicians.
Given the rapidly growing prevalence of a number of common chronic conditions, especially diabetes, it is difficult to conceive how the resulting sharp increase in the demand for recurring diagnostic and therapeutic services can be met without significant increases in physician productivity. Information technology (IT) in particular has been advanced as a potential source of productivity increases in healthcare.1 Another approach centers on the implementation of electronic health records (EHRs). Yet, some healthcare providers remain hesitant to adopt EHR systems because these typically entail large upfront costs of installation, training, and work flow redesign; can be difficult to use2,3; and to date, reports of the success of EHRs in increasing productivity have been mixed.4,5 Another approach centers on the increased use of computeraided diagnostics, which involves sophisticated computer algorithms that analyze a patient’s medical images and mark suspicious elements of the images, thereby assisting the examining physician’s detection of disease.6 Because these systems continue to require image interpretation and data entry by physicians, however, their productivity potential is limited.7
Automated Detection of Retinal Disease
The current standard for detection of diabetic retinopathy is a dilated retinal examination by an ophthalmologist or optometrist, using indirect ophthalmoscopy and a handheld lens (Figure).8 A patient’s typical visit takes about 2 hours to allow time for pupil dilation, consultation with the ophthalmologist or optometrist, the patient’s recovery, and documentation of the results in the patient’s medical chart or EHR.
By contrast, automated image-based detection of disease takes less than 2 minutes9 and does not require the labor-intensive interpretation of medical images and data entry by ophthalmologists or optometrists. Low-cost, compact retinal imagers have been prototyped that allow patients to perform retinal scans on their own.10 Moreover, “smart” retinal imagers are equipped with image-analysis algorithms that evaluate the camera-generated digital images for retinal abnormalities and automatically flag patients for a followup assessment by a physician, including a confirmatory dilated retinal examination and possibly treatment by an ophthalmologist.11 To date, patient-operated “smart” retinal imagers have only been used on an experimental basis in research settings. Once approved by the FDA, however, these devices could be made available for sale, for lease, or even for free in primary care offices, minute clinics, pharmacies, and other healthcare facilities, where trained personnel could instruct patients on how to use them properly. The widespread availability of a device that enables patients to screen themselves for retinal abnormalities will offer the potential of reaching more persons with diabetes more frequently than is possible by following the current practice of relying exclusively on dilated retinal examinations.
Although the precise integration of results from automated disease-detection devices into existing work flow designs remains to be determined, one option would follow the model of current diagnostic home-use tests for HIV or pregnancy: patients would periodically screen themselves for retinal abnormalities and save both the images and their automated interpretation, possibly following the Continuity of Care Document standard12 as required for meaningful use.13 Patients will be able to access the results electronically and share them with an ophthalmologist for further examination and for inclusion in their EHRs. Analogous to the follow-up care for positive HIV and pregnancy tests, if the device detected a retinal abnormality, patients would first consult with their primary care physicians who in turn might refer them to an ophthalmologist for a confirmatory dilated retinal examination and possibly treatment.
The Public Health Potential of Automated Diagnostics
Although diabetic retinopathy can be detected early and treated effectively, it remains a serious and widely feared complication that accounts for more than 24,000 largely preventable new cases of blindness and nearly 50,000 new cases of vision loss annually in the United States.14 Although current guidelines recommend that all persons with diabetes in the United States undergo a retinal exam every year, in 2010 ophthalmologists and optometrists examined only about 63% of the 25.8 million Americans with diabetes, or about 16.25 million patients.14,15 Potential causes include the limited number and availability of ophthalmologists and optometrists, many of whom operate at full capacity; the limited number of and access to high-quality digital retinal cameras16; patients’ ignorance of screening recommendations; and the reluctance of asymptomatic patients to incur the monetary and time costs associated with an examination.17
Automated retinal imaging could be used to screen all persons with diabetes annually and thereby identify those who require further examination and possibly treatment. To assess the impact of such automated detection systems on patient health and the cost of care, it is useful to compare their sensitivity and specificity statistics to those achieved by ophthalmologists and optometrists. As with any screening process, minimizing false negatives (high sensitivity) is vital to ensure that patients excluded from further assessment as a result of the screen will not suffer from the lack of indicated treatment, which in this case would prevent vision loss and improve the patient’s quality of life.
More reliable early detection of the disease could also reduce the risk of a professional liability claim arising from a missed diagnosis. On the other hand, by minimizing false positives, highly specific detection systems reduce unwarranted dilated retinal examinations by ophthalmologists or optometrists, thus lowering the cost to patients without the disease (and their insurers). The sensitivity levels of detecting referable retinopathy achieved by ophthalmologists, optometrists, and retinal specialists have been reported to range from 34%18 to 91%,9 with specificity levels reaching 98%.9 By comparison, in a recent study the sensitivity and specificity levels achieved by an automated system reached 97% and 59%.9 The burden and cost implications are discussed next.
As ophthalmologists and optometrists currently examine only 63% of all persons with diabetes, or about 16.25 million, at a sensitivity level as high as 91% they are currently identifying 63% × 91% = 57.3% of all patients who require treatment, for a Likelihood Ratio Positive (LR+) of 45.50 and a Likelihood Ratio Negative (LR—) of 0.43 (Table). By contrast, if all persons with diabetes under went an initial retinal exam using an automated detection system, the number referred to an ophthalmologist for a confirmatory dilated retinal examination and possibly treatment would drop to 12.02 million; these systems would flag 97% of the estimated 2.58 million persons with diabetes who every year develop retinopathy that requires clinical management,19 or about 2.50 million, and 100% — 59% = 41% of the remaining 23.22 million patients who do not require treatment, or 9.52 million. In combination with the follow-up examinations by ophthalmologists, this hypothetical screening practice would, in assessing the overall process, yield both a higher Likelihood Ratio Positive (108.00) and a lower Likelihood Ratio Negative (0.12) than the current screening practice. Screening all persons with diabetes for retinal abnormalities using automated detection systems would cut in half the typical patient’s number of office-based dilated retinal examinations per annum. Perhaps most important, it would raise the proportion of patients requiring treatment who receive it from 57% to 88%, thereby preventing 54% more cases of partial vision loss or blindness.
If the adoption of automated detection systems left the proportion of persons with diabetes undergoing screening unchanged at its current level of 63%, the number of referrals to ophthalmologists would drop from 16.25 to 7.57 million, while the proportion of patients requiring treatment who receive it would decline minimally from 57% to 56%. Thus, automated detection systems would yield a significantly higher rate of detecting and treating diabetic retinopathy per dilated retinal examination. A full cost-effectiveness analysis of universal automated screening is beyond the scope of this report.
Even at the comparatively lower levels of specificity, automated screening allows patients to avoid on average every other office-based dilated retinal exam they would be scheduled to undergo under the current testing recommendations. At the same time, the relatively high rate of false positives under automated screening might lead some patients to discount a positive test result and discourage them from promptly seeking a confirmatory dilated retinal examination, even if they knew that a delay in treatment might result in impairment or total loss of vision. Educational campaigns, reminders, and possibly modifications to the patient’s health insurance benefit could be considered to minimize the risk that a patient will ignore a positive test result.
A Complement to Current Screening Practice
Automated image-based disease detection could complement the important role that ophthalmologists and optometrists currently fill in diagnosing diabetic retinopathy: it could reduce the time required of these specialists to interpret medical images, automatically add the results to the patient’s record, and overcome possible biases that may skew the clinical assessments. Once approved by the FDA, such an initial examination could be performed by anyone with the device, including patients themselves, which would reduce the cost of universal automated screening even more. More frequent detection of retinal abnormalities would not just prevent substantially more cases of blindness; it would also help identify patients with badly managed diabetic disease more generally and could eventually aid in the early detection of cardiovascular disease, macular degeneration, glaucoma, malaria, and Alzheimer’s disease.20-23
Given that the prevalence of diabetes in the United States is projected to increase by 50% over the coming decade24 and given the expected increase in insurance coverage as provisions of the Affordable Care Act are implemented, automated initial screening may represent an attractive low-cost option to meet the growing demand for routine diagnostic services.
Paradoxically, greater use of automated detection systems, coupled with the rising prevalence of diabetes, could largely sustain the demand for ophthalmologist services. Even if automated detection systems replace all initial retinal examinations currently performed by ophthalmologists and optometrists, expanding these screens to all persons with diabetes implies that ophthalmologists will continue to perform about three-fourths of the current volume of examinations. Moreover, the significantly higher number of retinal abnormalities that automated systems help detect implies that patients seeking a follow-up exam will now be much more likely to require treatment, thus allowing ophthalmologists to concentrate on more complex, and possibly more rewarding, tasks.Author Affiliations: From Department of Health Administration and Policy, George Mason University (LAH), Fairfax, VA; Division of Health Sciences Informatics, Johns Hopkins Bloomberg School of Public Health (HPL), Baltimore, MD; and Department of Ophthalmology and Visual Sciences, and Stephen A. Wynn Institute for Vision Research, University of Iowa Hospital and Clinics (MDA), Iowa City, IA.
Source of Funding: Dr Abràmoff gratefully acknowledges support through NIH EY-017989, EY019112, EY018853 and VA I01 CX000119.
Author Disclosures: Dr Abràmoff is named inventor on patents, owned by the University of Iowa, related to some of the technology described in this manuscript, and he is the founder, shareholder, and president of IDx LLC, of Iowa City, Iowa, which has licensed these patents. He has pending grants from the National Institutes of Health (NIH), the Department of Veterans Affairs (VA), Beckman Foundation, and Research to Prevent Blindness, and has received grants from NIH, the VA, and Beckman Foundation. Drs Helmchen and Lehmann report no conflicts of interest.
Authorship Information: Concept and design (LAH, MDA); acquisition of data (MDA); analysis and interpretation of data (LAH, HPL, MDA); drafting of the manuscript (LAH, HPL, MDA); critical revision of the manuscript for important intellectual content (LAH, HPL, MDA); statistical analysis (MDA); provision of study materials or patients (MDA); obtaining funding (MDA); administrative, technical, or logistic support (MDA); and supervision (MDA).
Address correspondence to: Lorens A. Helmchen, PhD, Associate Professor, Department of Health Administration and Policy, George Mason University, 4400 University Dr — MS: 1J3, Fairfax, VA 22030. E-mail: email@example.com.REFERENCES
1. Buntin MB, Burke MF, Hoaglin MC, Blumenthal D. The benefits of health information technology: a review of the recent literature shows predominantly positive results. Health Aff. 2011;30:464-471.
2. Chiang MF, Boland MV, Margolis JW, Lum F, Abràmoff MD, Hildebrand PL. Adoption and Perceptions of Electronic Health Record Systems by Ophthalmologists: An American Academy of Ophthalmology Survey. Ophthalmology. 2008.
3. Wang SJ, Middleton B, Prosser LA, et al. A cost-benefit analysis of electronic medical records in primary care. Am J Med. 2003;114:397-403.
4. Bhargava HK, Mishra A. Electronic Medical Records and Physician Productivity: Evidence from Panel Data Analysis. Social Science Research Network; 2011.
5. O’Reilly D, Tarride JE, Goeree R, Lokker C, McKibbon KA. The economics of health information technology in medication management: a systematic review of economic evaluations. J Am Med Inform Assoc. 2012;19:423-438.
6. Sonka M, Hlavac V, Boyle R. Image processing, analysis, and machine vision. Thompson Learning; 2008.
7. Fenton JJ, Taplin SH, Carney PA, et al. Influence of computer-aided detection on performance of screening mammography. N Engl J Med. 2007;356:1399-1409.
8. American Academy of Ophthalmology Retina Panel. Preferred practice patterns: diabetic retinopathy. San Francisco, CA: American Academy of Ophthalmology; 2008.
9. Abràmoff MD, Folk JC, Han DP, et al. Automated analysis of retinal images for detection of referable diabetic retinopathy. JAMA Ophthalmol. 2013;131:351-357.
10. Williams GA, Scott IU, Haller JA, Maguire AM, Marcus D, McDonald HR. Single-field fundus photography for diabetic retinopathy screening: a report by the American Academy of Ophthalmology. Ophthalmology. 2004;111:1055-1062.
11. Goatman K, Charnley A, Webster L, Nussey S. Assessment of automated disease detection in diabetic retinopathy screening using two-field photography. PLoS One. 2011;6.
12. D’Amore JD, Sittig DF, Ness RB. How the Continuity of Care Document can advance medical research and public health. Am J Public Health. 2012;102(5):e1-e4.
13. Blumenthal D, Tavenner M. The “meaningful use” regulation for electronic health records. N Engl J Med. 2010;363(6):501-504.
14. Huang ES, Brown SE, Ewigman BG, Foley EC, Meltzer DO. Patient perceptions of quality of life with diabetes-related complications and treatments. Diabetes Care. 2007;30:2478-2483.
15. Diabetes Report Card 2012. Atlanta, GA: HHS;2012.
16. Abràmoff MD; Garvin MK; Sonka M. Retinal imaging and image analysis. IEEE Rev Biomed Eng. 2010;3:169-208. doi: 10.1109/RBME.2010.2084567.
17. Abràmoff MD, Reinhardt JM, Russell SR, et al. Automated early detection of diabetic retinopathy. Ophthalmology. 2010;117:1147-1154.
18. Lin DY, Blumenkranz MS, Brothers RJ, Grosvenor DM. The sensitivity and specificity of single-field nonmydriatic monochromatic digital fundus photography with remote image interpretation for diabetic retinopathy screening: a comparison with ophthalmoscopy and standardized mydriatic color photography. Am J Ophthalmol. 2002;134:204-213.
19. Jones CD, Greenwood RH, Misra A, Bachmann MO. Incidence and progression of diabetic retinopathy during 17 years of a population-based screening program in England. Diabetes Care March. 2012;35:592-596.
20. Xu X, Niemeijer M, Song Q, et al. Vessel boundary delineation on fundus images using graph-based approach. IEEE Transactions on Medical Imaging. 2011;30(6):1184-1191.
21. Abràmoff MD, Alward WLM, Greenlee EC, et al. Automated segmentation of the optic disc from stereo color photographs using physiologically plausible features. Invest Ophthalmol Vis Sci. 2007;8:1665-1673.
22. Joshi VS, Maude RJ, Reinhardt JM, et al. Automated detection of malarial retinopathy-associated retinal hemorrhages. Invest Ophthalmol Vis Sci. 2012;53(10):6582-6588.
23. Xu X, Abràmoff MD, Bertelsen G, Reinhardt JM. Retinal vessel width measurement at branching points using an improved electric field theory-based graph approach. SPIE. 2012; 8314. doi:10.1117/12.911831.
24. Huang ES, Basu A, O’Grady M, Capretta JC. Projecting the future diabetes population size and related costs for the U.S. Diabetes Care. 2009;32:2225-2229.