Incidence of Referable Retinal Disease in Diabetic Patients at a Primary Care Practice

Abstract

Purpose:

This work tests the feasibility of remote ophthalmic imaging to identify referable retinal abnormalities and assesses the effectiveness of color fundus photography (CFP) vs optical coherence tomography (OCT) for this purpose.

Methods:

This prospective, nonrandomized study included 633 patients with diabetes at Duke Primary Care. Undilated patients underwent screening with CFP and OCT camera (MaestroCare, Topcon). Images were graded independently for interpretability and the presence of predetermined retinal disease. Retinal disease was classified as diabetic retinopathy (DR) referable to a retina specialist or incidental findings referable to either a retina specialist or a general ophthalmologist, depending on severity.

Results:

Mean (SD) age of screened patients was 66 (13) years, and 49% were women. The average glycated hemoglobin A_1c level was 7.6 % (SD, 1.7%), and 30% of the patients were on insulin. The average duration of diabetes was 5.9 (SD, 7.3) years. Remote images from OCT were significantly more interpretable than CFP (98% vs 83%, respectively; P < .001). Referral rates were 9% for DR and 28% for incidental findings. Among patients with DR, OCT and CFP were helpful in 58% and 87% of cases, respectively (P < .001).

Conclusions:

Remote diagnosis of ophthalmic imaging at the point of service may allow for early identification of retinal disease and timely referral and treatment. Our approach showed that OCT had significantly better interpretability, while CFP was more helpful in identifying DR. These findings may be important when choosing the screening device in a specific context.

Introduction

The most common eye diseases affecting the adult-aged population worldwide are diabetic retinopathy (DR), cataracts, glaucoma, and age-related macular degeneration (AMD). ¹ These diseases are frequently diagnosed in the advanced stages when treatment might have limited potential and unfavorable visual outcomes. Therefore, delayed presentation may often result in vision loss. In addition, late disease presentation leads to an unnecessary increase in health care costs. Early and regular screening and monitoring of these patients are paramount to ensure improved clinical outcomes and cost-effectiveness.

Teleophthalmology offers a way for early screening and regular monitoring of patients. ^{2 -4} The merits of teleophthalmology for the early diagnosis of retinal disease, particularly DR, have been known for a long time. ^{5 -7} Unfortunately, several barriers, including access, cost, and reimbursement guidelines, prevent its wider acceptance and use.

Duke Ophthalmology started a program involved in the remote diagnosis of retinal disease a couple of years ago with a few simple goals: improving access, identifying retinal disease early to improve visual outcomes, and reducing health care costs. The idea was to place simple-to-use, nonmydriatic color fundus photography (CFP) and optical coherence tomography (OCT) cameras in Duke’s medical clinics and to screen patients before they leave. As previously published, we piloted our remote diagnosis concept and distinguished it from traditional teleophthalmology with respect to several vital elements. ⁸ Briefly, traditional teleophthalmology involves CFP on dilated pupils taken by expert photographers with expensive equipment in dedicated brick-and-mortar imaging centers. Remote diagnosis can be defined as the acquisition of CFP and OCT on nondilated pupils by nonexpert imagers, possibly using less expensive equipment permanently located at the point of service (eg, primary care and endocrinology clinics).

To achieve our goals, we partnered with different companies manufacturing ophthalmic imaging devices (eg, Optovue, Zeiss, Topcon, and Optos) to validate remote diagnosis’s feasibility in nonophthalmology outpatient clinics and to identify the unit that best serves our needs. Our first study was a noninferiority study that tested and demonstrated this approach’s feasibility and accuracy in 2 environments with a relatively high prevalence of DR and AMD: Duke Endocrinology outpatient clinics and assisted living centers, respectively. ⁸ To test the remote diagnosis approach in a more “real-world” setting with a lower retinal disease burden, we tested Maestro 3D-OCT (Topcon) on an adult population with diabetes in a Duke primary care clinic.

In the present study, we tested the feasibility of a retinal image capture using a system located at the point of service in an environment with a relatively low prevalence of retinal disease. Our primary outcomes were the incidence of DR and other referable retinal diseases. Our secondary outcomes included the differences in the contribution of CFP and OCT in identifying referable retinal disease, the characterization of affected patients’ demographics, and improvement in The Healthcare Effectiveness Data and Information Set (HEDIS).

Methods

Patient Population

This prospective, nonrandomized study was part of the recently established Duke Ophthalmology Quality Assessment and Quality Improvement project. We imaged 635 patients with diabetes at one of the Duke primary care clinics during 6 months of screening. Study participants had to be older than 18 years and able to sit for image acquisition. All eligible patients at the site were invited to participate by their primary care physician.

Study Design and Outcome Measures

Remote diagnosis imaging of nondilated pupils was performed at the point of service by trained but nonexpert imagers using a US Food and Drug Administration–approved, noncontact, portable retinal imaging device. The unit was the Maestro 3D-OCT multimodality OCT and fundus imaging device (Topcon Inc). CFP had a picture angle of 45° or 30° (or their equivalent when using digital zoom) on the nondilated pupil. The B-scan horizontal range for the OCT was 3 to 12 mm (Figure 1), and whole-volume scan was used for grading. Images were taken after the patient’s primary care physician visit to not disturb the clinic’s flow. Categories of the prespecified outcomes were the same as previously published ⁸ : (1) DR, (2) incidental findings referable to a retina specialist (ie, presence of macular lesions requiring urgent intervention or frequent follow-up by a retina specialist), (3) incidental findings referable to general ophthalmology or optometry departments (ie, retinal or other ocular disease that requires a nonurgent referral to an eye specialist), (4) no diseased condition, or (5) uninterpretable images. Any ungradable images were grouped with the referable disease, as these eyes would have failed image screening.

Figure 1.

Good-quality images: (A) optical coherence tomography of diabetic macular edema and (B) color fundus photograph of presumably inactive proliferative diabetic retinopathy. Uninterpretable images: (C) optical coherence tomography and (D) a color fundus photograph demonstrate a poor view of the posterior pole.

Image Grading

Image grading was conducted by consensus and was performed by 3 Duke medical retina fellows at the time (J.-B.K., Kishan Govind, MD, and May Marquez, MD) with at least 5 years of ophthalmology experience and an attending medical retina specialist (M.H.). CFP and OCT images were considered uninterpretable if no clear identification of macula was available (ie, poor positioning) or if they had inadequate exposure owing to media opacity. As previously published, the CFP images were assessed for DR severity using the standard Early Treatment Diabetic Retinopathy Study scale and for AMD using the Age-Related Eye Disease Study 9-Step Severity Scale. ^9,10 Similarly, the OCT images were graded according to the presence of the following: significant vitreous debris, cystoid intraretinal fluid, subretinal fluid, hyperreflective foci, pigment epithelial detachment, drusen, retinal pigment epithelium or retinal atrophy, macular holes, epiretinal membrane (ERM), and any other abnormal macular findings. ⁸

Training of Imagers

Certified medical assistants at the primary care office were identified as imagers for this study. Before collecting images, each imager read the instruction manual for the imaging device. Next, imagers received practical instructions on the use of the device to help develop their proficiency. After the initial training, imagers were observed and coached during their first imaging sessions with at least 5 study participants. A periodic review of remote images was conducted to direct the retraining of imagers if suboptimal image quality due to poor technique was suspected. Topcon’s personnel (Greg Hoffmeyer, BFA) and M.H. performed the training and supervision.

Statistical Analysis

All statistical analyses were performed in Python (version 3.7.6; Python Software Foundation, https://www.python.org). Descriptive statistics were calculated to assess baseline demographics and medical comorbidities. Chi-square and Fisher exact analyses were conducted to assess the association among categorical variables, t tests were used to compare means of continuous variables, and ordinary least squares logistic regressions were used to assess the association between predictor variables and a binary outcome. Cohen κ was used to test agreement between remote diagnosis and standard examination by retina specialists at a tertiary care center. ¹¹ Sensitivity and specificity metrics were calculated to assess how well remote diagnosis evaluation correctly detected referable retinal disease compared with criterion-standard evaluation by retina specialists at a tertiary center. Two-proportion z tests were used to compare the usefulness of OCT and CFP findings in making referral decisions. The threshold for statistical significance was set at an α value of .05 for all analyses.

Results

In our analysis, a total of 633 patients and 1266 eyes were included. The mean (SD) age of our cohort was 66 (13) years, and 49% of the patients were women. Of the 633 patients, 72% identified as White, 23% as Black or African American, 1% as Asian, 1% as multiracial, and 3% did not disclose or identify as any of the aforementioned categories. Five percent of the patients had a diagnosis of diabetes mellitus type 1 (T1DM), and the other 95% had a diagnosis of diabetes mellitus type 2 (T2DM). The average glycated hemoglobin A_1c (HbA_1c) level for all patients was 7.6% (SD, 1.7%), and 30% of the patients were taking insulin. The average known duration of diabetes was 5.9 (SD, 7.3) years. Despite White patients being older on average than Black patients (Black: age (SD), 63 (14) years, White: 68 (13) years, P = .001) as assessed with a t test, Black patients had higher average HbA_1c levels (Black: 8.1% [SD, 2.1%], White: 7.4% [1.5%], P < .001). Chi-square analyses also showed that Black patients were more likely to be taking insulin (Black: 42%, White: 27%, P = .001) and have diabetic macular edema (DME; Black: 12%, White: 6%, P = .032) or DR (Black: 16%, White: 7%, P = .001).

Baseline demographic information and the most common medical comorbidities and remote diagnosis outcomes for the overall cohort and subgroups stratified by type of diabetes (T1DM vs T2DM) can be found in Table 1. Notably, patients with T1DM were younger on average (P < .001) and had greater HbA_1c levels (P < .001). As expected, they also had a much greater duration of disease (P = .013) and were more likely to be prescribed insulin (P < .001) as 100% of T1DM patients were taking insulin.

Table 1.

Demographics, Comorbidities, and Remote Diagnosis Evaluation for the Overall Cohort Stratified by Type of Diabetes.

Variables	All, N = 633	T1DM, n = 32a	T2DM, n = 590a	P value
Demographics
Age, mean ± SD, y	66.1 ± 13.4	50.3 ± 19.0	67.1 ± 12.4	<.001b
Sex, No. (%), female	310 (49)	12 (39)	288 (49)	.343c
Race, No. (%)				.423c
White	456 (72)	22 (68)	425 (72)
African American	146 (23)	7 (23)	136 (23)
Glycated hemoglobin A1c, mean ± SD	7.6 ± 1.7	9.0 ± 2.1	7.5 ± 1.6	<.001b
Duration of diabetes, mean ± SD, y	5.9 ± 7.3	15.1 ± 15.2	5.1 ± 5.6	.013b
Insulin use, No. %	190 (30)	32 (100)	153 (26)	<.001c
Comorbidities, No. (%)
Hypertension	481 (76)	13 (41)	466 (79)	<.001c
Hyperlipidemia	342 (54)	16 (50)	324 (55)	.718c
Obesity	228 (36)	3 (9)	221 (37)	.001c
OSA	82 (13)	0 (0)	82 (14)	.015d
Coronary artery disease	82 (13)	2 (6)	77 (13)	.411d
Chronic kidney disease	63 (10)	3 (9)	59 (10)	≥.999d
GERD	63 (10)	2 (6)	59 (10)	.759d
COPD	59 (9)	2 (6)	57 (10)	.759d
Tobacco	19 (3)	2 (6)	14 (2)	.159d
Remote diagnosis evaluation, No. (%)
Referral to retina for diabetic retinopathy	57 (9)	7 (22)	47 (8)	.016d
DME noted	44 (7)	5 (16)	39 (7)	.068d
Referral to retina for incidental finding	123 (18)	4 (13)	119 (20)	.429c
Referral to general ophthalmology for incidental finding	52 (8)	1 (3)	51 (9)	.504d

Abbreviations: COPD, chronic obstructive pulmonary disease; DME, diabetic macular edema; GERD, gastroesophageal reflux disease; OSA, obstructive sleep apnea; T1DM, type 1 diabetes mellitus; T2DM, type 2 diabetes mellitus.

^a Type of diabetes was unknown in 11 patients.

^b Compared using 2-sample t test.

^c Compared using χ² test.

^d Compared using Fisher exact test.

Of 633 patients, 52 (8.2%) were referred to a general ophthalmologist to follow up on a non–sight-threatening incidental finding, 123 (18.4%) were referred to a retina specialist for follow-up of a retina-related incidental finding of presumed higher severity, and 57 (9.0%) were referred to a retina specialist for follow-up of different stages of DR. Of these 57, 34% were referred for mild DR, 43% for moderate, 0% for severe, and 9% for proliferative DR, and 14% were referred for uninterpretable fundus photographs; 70% had DME. Table 2 lists a more detailed characterization of the 57 patients referred to a retina specialist for DR.

Table 2.

Clinical Characteristics and Staging of Patients Referred to a Retina Specialist for Diabetic Retinopathy.

Variables	N = 57
Demographics
Age, mean ± SD, y	60.0 ± 14.0
Sex, No. (%), female	26 (46)
Race, No. (%)
White	31 (55)
African American	24 (42)
Diabetes type, No. (%), T1DM	7 (13)
Glycated hemoglobin A1c, mean ± SD	8.5 ± 2.0
Duration of diabetes, mean ± SD, y	10.6 ± 12.3
Insulin use, No. %	33 (58)
Diabetic retinopathy stage, No. %
Mild	19 (34)
Moderate	25 (43)
Severe	0
Proliferative	5 (9)
Othera	8 (14)
Diabetic macular edema	40 (70)

Abbreviation: T1DM, type 1 diabetes mellitus.

^a Some of the color fundus photographs were uninterpretable, thus diabetic retinopathy staging was not possible.

Using logistic regression modeling, several variables were shown to affect the likelihood of DR in remote diagnosis evaluation (Table 3). Patients with greater HbA_1c levels were more likely to have DR, with an odds ratio (OR) of 1.32 for each 1% increase in A_1c (P < .001). Longer duration of diabetes also resulted in a slightly increased risk, with an OR of 1.07 for each additional year a patient had diabetes (P = .002). Relatedly, insulin use also predicted DR (OR: 3.77, P < .001) as did T1DM (OR: 3.22, P = .010) and being Black or African American (OR: 2.63, P = .001).

Table 3.

Logistic Regression Analyses to Assess Putative Predictive Variables for Detection of Diabetic Retinopathy on Remote Diagnosis Evaluation.

Predictor	Odds ratio	95% CI	P valuea
Demographics
Age, y	0.97	(0.95-0.99)	.001
Sex, male	1.13	(0.65-1.98)	.665
Race, White	0.38	(0.21-0.68)	.001
Glycated hemoglobin A1c	1.32	(1.16-1.51)	<.001
Duration of diabetes, y	1.07	(1.03-1.11)	.002
Insulin use	3.77	(2.14-6.64)	<.001
Comorbidities
Atrial fibrillation	0.82	(0.24-2.75)	.750
Hypertension	0.58	(0.32-1.04)	.065
Hyperlipidemia, %	0.62	(0.35-1.07)	.087
Obesity, %	0.75	(0.41-1.36)	.343
OSA, %	0.91	(0.40-2.08)	.820
Coronary artery disease, %	0.66	(0.26-1.71)	.396
Chronic kidney disease, %	0.49	(0.15-1.63)	.247
GERD, %	0.91	(0.35-2.37)	.842
COPD, %	0.84	(0.36-2.47)	.907
Tobacco use, %	1.48	(0.33-6.69)	.609

Abbreviations: COPD, chronic obstructive pulmonary disease; GERD, gastroesophageal reflux disease; OSA, obstructive sleep apnea.

^aBold indicates P < .05.

Our cohort included 175 (27.6%) patients with incidental findings, with 52 referred to a general ophthalmology department and 123 to a retina specialist, depending on the disease’s severity. The most identified incidental finding was dry or wet AMD, followed by abnormalities of the vitreomacular interface: ERM, vitreomacular adhesions, and vitreomacular traction (Figure 2). Logistic regression analyses of several variables assessing putative predictive factors for incidental findings are shown in Table 4. We found that older age (OR: 1.07, P < .001) was predictive of having incidental findings. Higher HbA_1c values were associated with having fewer incidental findings (OR: .89, P = .026). Logistic regression analyses also showed that several comorbidities were predictive of whether a patient had incidental finding on remote diagnosis evaluation: atrial fibrillation (OR: 2.18, P = .018), chronic obstructive pulmonary disease (OR: 1.99, P = .014), chronic kidney disease (OR: 1.96, P = .014), and coronary artery disease (OR: 1.99, P = .005).

Figure 2.

Examples of common incidental findings. Dry age-related macular degeneration represented in (A) optical coherence tomography and a (B) color fundus photograph. (C) Vitreomacular adhesion with traction and (D) an epiretinal membrane represented in optical coherence tomography.

Table 4.

Logistic Regression Analyses to Assess Putative Predictive Variables for Detection of Incidental Findings on Remote Diagnosis Evaluation.

Predictor	Odds ratio	95% CI	P valuea
Demographics
Age, y	1.07	(1.05-1.09)	<.001
Sex, male	0.90	(0.64-1.27)	.552
Race, White	1.14	(0.75-1.74)	.535
Glycated hemoglobin A1c	0.89	(0.78-0.98)	.026
Duration of diabetes, y	1.00	(0.96-1.04)	.992
Insulin use	0.98	(0.67-1.44)	.936
Comorbidities
Atrial fibrillation	2.18	(1.14-4.17)	.018
Hypertension	1.24	(0.82-1.88)	.314
Hyperlipidemia	1.12	(0.79-1.58)	.523
Obesity	0.69	(0.48-1.00)	.049
OSA	1.54	(0.95-2.49)	.077
Coronary artery disease	1.99	(1.23-3.23)	.005
Chronic kidney disease	1.96	(1.15-3.35)	.014
GERD	1.37	(0.78-2.39)	.273
COPD	1.99	(1.15-3.45)	.014
Tobacco use	0.84	(0.27-2.63)	.761

Abbreviations: COPD, chronic obstructive pulmonary disease; GERD, gastroesophageal reflux disease; OSA, obstructive sleep apnea.

^aBold indicates P < .05.

All patients had CFP and OCT images taken. A total of 1235 (98%) of the OCT and 1049 (83%) of the CFP images were deemed interpretable (P < .001; examples of uninterpretable images and images with retinal disease can be found in Figure 1). Among patients that were referred to a retina specialist for DR, OCT was helpful in identifying retinal disease in 58% of cases, whereas CFP was helpful in 87% of cases (P < .001). Among patients referred to a retina specialist for retina-related incidental findings, OCT and CFP were helpful in identifying retinal disease in 86% and 68% of cases, respectively (P < .001). Among patients referred to a general ophthalmologist for other incidental findings, OCT and CFP were helpful in identifying retinal disease in 52% and 58% of cases, respectively (P = .396).

For the 15 patients that followed up for a standard evaluation at our institution, Cohen κ for the agreement was 0.29, which is deemed fair according to the classification set forth by Landis and Koch. ¹¹ Sensitivity and specificity of the remote diagnosis model in detecting referrable DR compared with criterion-standard evaluation by retina specialists at a tertiary center were 70% and 60%, respectively. The DR screening that we have conducted increased the primary care clinic’s C13 measurement of the Centers for Medicare & Medicaid Services’ HEDIS (Diabetic Care–Eye Examination) from a low 4-star (∼73%) to a 5-star (>83%) rating in only 5 months (the report was focused on the United Health Care patients because of the large attribution of patients and the incentives).

Conclusions

DR is the leading cause of preventable vision loss in working-aged adults; an estimated 285 million people worldwide have diabetes. ¹² There are currently 23 million adults living with diabetes in the United States alone, and approximately 1.3 million adults have T1DM, and 21 million have T2DM (0.5% and 8.5% of the adult population, respectively). ¹³ Roughly one-third of patients with diabetes will develop DR, and one-third of these will progress to its advanced stage, proliferative DR with or without DME. ^12,14 Finally, it has been shown that DR, once diagnosed, is significantly more likely to progress than to regress. ¹⁵ According to the American Diabetic Association’s Preferred Practice Patterns, patients with T1DM should have their first screening examination 5 years after diagnosis and every year after that. In contrast, patients with newly diagnosed T2DM should undergo a dilated retinal examination at the time of diagnosis and yearly after that. ^{16 -18} Regardless of these straightforward guidelines, it is estimated that only 40% to 50% of patients adhere to the recommended screening intervals. ^{19 -22} Teleophthalmology has been shown to increase compliance with the proposed guidelines. ²³

The duration of diabetes, poor glycemic control with a high HbA_1c level, and the presence of hypertension have been proposed as the most critical risk factors for the development of DR. ²⁴ Other reported risk factors are dyslipidemia, high body mass index, puberty, pregnancy, and cataract surgery. ^17,25,26 We assessed factors associated with DR as well as reported incidental retinal findings. Older age, higher HbA_1c levels, longer duration of diabetes, Black race, and T1DM were associated with an increased likelihood of being referred for DR. Similarly, older age, higher HbA_1c levels, and several comorbidities were associated with an increased risk of incidental retinal findings. Our screened population’s top 3 comorbidities were hypertension, hyperlipidemia, and obesity.

Various demographic and systemic risk factors for DR have been explored in the literature. ^{24 -26} Identifying these factors is important because they can enhance the pretest probability of disease, which in turn affects the posttest probability and the cost-effectiveness of the screening (eg, if the posttest probability of a screening program is poor, there will be many false-positive cases referred for evaluation, and the cost-effectiveness will suffer). ^{27 -29} The presented strong correlation between demographics, comorbidities, and HbA_1c levels with the incidence of DR might be useful in increasing screening sensitivity if these entities are used when examining patients with diabetes.

Our study tested the feasibility of retinal disease screening at a primary care clinic with presumably lower retinal disease prevalence. One could argue that by screening only patients with diabetes, a higher prevalence of DR is expected. However, our assumption was that patients being followed up at the primary care clinic instead of the endocrinology department have less severe diabetes on average (ie, better HbA_1c levels, shorter duration of disease, and less likelihood of being prescribed insulin). The present results support our assumption: Captured rates of referral-requiring DR and incidental findings were approximately 9% and 28%, respectively. While there is a large range of referral rates from screening reported in the literature, the rate of referral for DR and incidental findings in our study are comparable with previous works. ^30,31

Interestingly, our study considerably improved the primary care office’s HEDIS C13 measurement (Diabetic Care–Eye Exam). In the United States, HEDIS assessment is becoming mandated for participation in the Health Insurance Marketplace through the Affordable Care Act, and thus it can be a great incentive for primary care and endocrinology clinics to participate in DR screening. ³²

In addition to previously reported better image interpretability of OCT relative to CFP on undilated pupils, ^{30,33 -35} the use of both imaging modalities independently could potentially lead to higher false-positive rates. We have demonstrated that CFP was significantly more supportive in identifying DR relative to OCT (P < .001), whereas OCT was significantly more supportive for retina-referable incidental findings (eg, AMD or ERM, P < .001). Albeit requiring deeper insight, these findings are important to guide the choice of imaging modality depending on the targeted retinal disease. To summarize, in an environment where low disease prevalence was expected, approximately 35% of patients required some level of further ophthalmic attention. This result suggests a similar screening process could be a feasible and efficient approach to identifying patients at risk.

The clinical implications and cost-effectiveness associated with screening for other retinal diseases still necessitate further investigation. We know that retinal screening in the adult population is associated with significant improvement in visual outcomes. ^36,37 In addition, it has been proposed that DR and general eye screening carried out concurrently might improve overall (clinical and economic) benefits. ³⁸ Numerous other important questions (eg, evaluation of completed referrals, long-term visual outcomes) remain unanswered and will be a part of our future research.

Advances in automated image analysis and interpretation have the potential to further improve screening for retinal disease. ^39,40 We have also developed an automated image interpretation tool to facilitate timely, accurate, and inexpensive image interpretation without direct human effort. The derived model was shown to have high accuracy and sensitivity to identify the presence of retinal disease (yes or no) with a relatively low false-negative rate. Using data obtained in this study, we have developed an automated, deep learning–based image interpretation model ⁴¹ with high accuracy and sensitivity to recognize referable retinal disease. Such an automated image analysis tool, combined with remote retinal imaging, could potentially deliver on the promise of a low-cost, fast, and effective retinal disease screening.

Limitations

Our study had limitations. One is our assumption that patients presenting to a primary care clinic would have less severe diabetes on average and a lower burden of diabetic eye disease than those presenting to an endocrinology clinic. Given that accessibility to endocrinology clinics remains difficult, with high waiting times in many parts of the country, there are likely many individuals with severe diabetes requiring endocrinology follow-up who are able to be seen only at their primary care clinic. Future studies may explore this assumption further by correlating a locale’s average wait times for endocrinologists with the burden of diabetes in primary care practices.

Unfortunately, most screened patients in our cohort did not follow up at our institution for a standard retinal examination but did so with a local ophthalmology clinic. Complying with Health Insurance Portability and Accountability Act of 1996 (HIPAA) regulations, we were unable to obtain their records and confirm the screening process’s accuracy. This resulted in only a moderate Cohen κ and relatively low sensitivity and specificity in this population compared with standard evaluation at our institution. Another important point is the device’s cost. Further investigation into available low-cost imaging devices is mandated to potentially deliver the promise of remote diagnosis of retinal disease.

Summary

Burdened with multiple comorbidities, patients with diabetes seem to be the most compliant with their primary care visits, making primary care clinics an ideal location to implement retinal image capture for DR screening. We have demonstrated the feasibility of using such a screening approach. Retinal screening in this setting has multiple advantages: accessibility, patient capture, cost-effectiveness, and improved patient triage. Emerging new imaging devices with simplified interfaces that can be used in nonophthalmic clinics and operated by staff with limited training, together with automated (deep learning–based) image interpretation, make remote diagnosis very attractive lead in early diagnosis and monitoring of many sight-threatening diseases.