Teleophthalmology and Inequities in Diabetic Eye Disease at Safety Net Hospitals

Abstract

Introduction:

Teleophthalmology has emerged as a convenient and cost-effective intervention to increase access to screening for diabetic retinopathy (DR), a disease that disproportionately affects socially disadvantaged communities. However, a few studies have directly compared the detection of eye disease by teleophthalmology between socially and geographically diverse communities. This study compared the rates and severity of diabetic eye disease, as detected by teleophthalmology, between safety net and non-Safety Net Hospitals (non-SNHs).

Methods:

Retrospective chart review of patients screened for DR at county Safety Net Hospitals (SNHs) and non-SNHs in 150 cities and 30 states. The rates of DR, macular edema, suspected cataract, suspected glaucoma, and suspected age-related macular degeneration were compared. Relative risk and severity of disease in the county SNH population were calculated. Images were graded by the same group of IRIS readers, who used at least one image per eye with a 45° field centered between the optic disc and the macula. Participants with ungradable screening images were excluded.

Results:

Ninety-four thousand three hundred twenty-nine participants were screened for eye disease from September 1, 2016 to August 31, 2017. Among the screened participants (54% female; mean [SD] age, 58.7 [12.9] years), overall disease detection was 31% in the county SNH population and 23.6% in the non-SNH population. Compared with the non-SNH population, the county SNH population was twice as likely to screen positive for three or more concurrent eye conditions (1.2% vs. 0.7%) and had increased prevalence of DR (20.2% vs. 16.2%), macular edema (4.9% vs. 3.4%), suspected glaucoma (9.1% vs. 4.3%), suspected cataract (9.6% vs. 4.8%), and proliferative DR (2.1% vs. 1.0%).

Conclusions:

Increased diabetic eye disease prevalence and severity among people seen at SNHs highlights the need for continued resources to screen, treat, and manage disease. Teleophthalmology continues to be an important tool in efforts to mitigate health inequities and address barriers faced by underserved communities.

Introduction

Diabetic retinopathy (DR) is the leading cause of visual disability and permanent blindness in working-age Americans.¹ A national cross-sectional study estimated that of the 10.2 million Americans who had diabetes mellitus (DM) in 2005–2008, 40.3% and 8.2% had DR and vision–threatening DR (VTDR), respectively.² Virtually 100% of individuals with Type 1 DM and more than 50% of those with Type 2 DM develop vision impairment within two decades of disease onset.³

Cases of DR and DR-related blindness are expected to rise due to the increasing prevalence of DM and the life expectancy of patients with DM; 40–50 million Americans may be affected by DM by 2030.⁴ However, the prevalence and disease burden of DM and DR varies across different patient population groups within the United States.

A growing body of evidence suggests that the burden of DM disproportionately affects the socially disadvantaged. Socioeconomic inequality, systemic racism, geography, sexism, and other social factors belie health inequities, which result in socially disenfranchised groups suffering shorter life expectancies, more severe disease, and earlier onset of chronic disease and disability.^5,6 Whitehead defines health inequities as “differences which are unnecessary and avoidable but, in addition, are also considered unfair and unjust.”⁶

For example, a meta–analysis by Galea et al.⁷ concluded that the number of U.S. deaths in 2000, which were attributable to poverty or income inequality, surpassed those due to an acute myocardial infarction. In turn, good health is critical to overcoming social disadvantage, and poor vision is independently associated with poor health outcomes and reduced earning potential.⁸

Several studies suggest that health inequities also exist in the prevalence of DR and VTDR.⁹ Marginalized race/ethnicity appears to be a significant risk factor for developing DR and VTDR, even after controlling for HbA1C and duration of diabetes,^10–14 and although the Baltimore Eye Survey found that while blindness and vision impairment from all causes impacted blacks at nearly twice the rate of whites,¹⁵ this gap shrank after adjusting for education level and household income.⁸

Despite the existing published data, there is still much to be learned about health inequities in the prevalence and severity of DR. Existing studies had small sample sizes and did not directly compare the prevalence of disease across different communities. For example, a study of six Los Angeles county safety net clinics reported on the prevalence of DR in their population determined through teleretinal screening,^16,17 but it did not compare their prevalence with other populations.

Similar studies have been conducted among Hispanic¹⁸ and non-Hispanic black communities,^10,15,19 but a few have directly compared these with each other or with other communities. It can also be difficult to accurately interpret and compare disease prevalence when diagnoses are determined by a variety of providers in different settings—for example, telehealth, traditional clinic visits.

Accurate prevalence estimates could help characterize health inequities (if present), monitor the effectiveness of public health interventions, and aid in policy decision making and resource allocation. To mitigate some of the limitations of previous studies on health inequities, the authors chose to analyze data from a single, cross-country, teleophthalmology service utilizing the same pool of reading providers. As a proxy for socially disadvantaged communities, the authors used individuals who sought care at county Safety Net Hospital (SNH) systems compared with all other non-SNH systems.

According to the Institute of Medicine, SNHs are defined as providers that (1) provide care regardless of a patient's ability to pay and (2) deliver a significant level of health-related services to the uninsured, Medicaid, and other vulnerable patients.²⁰ Safety nets include Federally Qualified Health Centers, Rural Health centers, Disproportionate Share Hospitals, and Community Health Centers.

This study compared the prevalence and severity of DR and other nondiabetic eye diseases in a very large, geographically diverse population screened by the same group of eye providers.

Methods

All data were provided by Intelligent Retinal Imaging Systems (IRIS, Pensacola, FL), a company that provides and supports an end-to-end teleophthalmology screening program for multiple health care clinics and systems across the country. At the time of this study, Ben Taub Hospital (Houston, TX) and Grady Memorial Hospital (Atlanta, GA) were the only safety net systems contracted with IRIS. The safety net status was determined by IRIS.

The IRIS screening program was FDA-approved for the detection of diabetic macular edema (DME) and DR, but readers also commented on suspected cataract, glaucoma suspect, suspected vein occlusion, suspected macular hole, suspected dry and wet macular degeneration, and suspected hypertensive retinopathy. Image acquisition protocol was at the discretion of each imaging site, but most followed IRIS recommendations (the majority used a single posterior pole photograph of the nerve and macula).

Grading was performed on at least one image per eye with a 45° field centered between the optic disc and the macula. Images were read by the same group of trained and certified Eye Care Professional readers contracted by IRIS. De-identified data from imaging sessions occurring between September 1, 2016 and August 31, 2017 were analyzed for demographic data, location of screening (SNH vs. non-SNH), and presence of disease as determined by screening.

Statistical Analysis

Analysis was performed with Graphpad Prism statistical software, and Chi-squared tests were used to compare samples as appropriate. The ungradable image rate was calculated by using the total number of images and nongradable images per patient study in the IRIS data. Disease prevalence (Eq. 1) and relative risk of a disease in the county SNH population (Eq. 2) were calculated as follows, with similar calculations occurring with the non-SNH population:

$$Diseaseprevalencei=\frac{\left[Numberofpatientswithdiseasei\right]}{\left[Totalnumberofpatients\right]-\left[Ungradablepatients\right]}$$ Eq. 1

$$RelativeriskofdiseaseiinFQHC=\frac{{\left[Diseaseprevalencei\right]}_{FQHC}}{{\left[Diseaseprevalencei\right]}_{non-FQHC}}$$ Eq. 2

Results

This project was deemed quality improvement in nature and abides by SQUIRE guidelines, thus institutional review board approval was not required. Participants were not required to sign an informed consent, and all patient information was de-identified. This project conformed to the tenets of the Declaration of Helsinki and complied with the Health Insurance Portability and Accountability Act.

Patient Demographics

A total of 94,329 screening exams were performed and interpreted by IRIS during the period of September 1, 2016 to August 31, 2017. No patient received more than one screening exam during this period. Thirty-nine percent of these exams (37,006 exams) were performed at SNHs. The ungradable image rate was 5.9% among all patients, and this rate was higher in the non-SNH group than the SNH group (7.7% vs. 3.0%). Compared with those in non-SNHs, the patients seen in SNHs were older (60.9 vs. 55.2 years old, p < 0.0001, Table 1). Further, patients seen in SNHs were more likely to be female (59.8% vs. 50.8% female, p < 0.0001, Table 1).

Table 1.

Demographics of Patients Screened in Nonsafety Net Hospitals and County Safety Net Hospitals

	NONSAFETY NET	SAFETY NET	p-VALUE
Age, mean (SD)	55.2 (13.3)	60.9 (11.4)	p < 0.0001
Female gender, n (%)	29,097 (50.8)	22,129 (59.8)	p < 0.0001

These exams were performed in 150 cities in 30 states (Fig. 1). Non-SNHs that provided images to IRIS were found in all 30 states. The median number of exams performed in non-SNHs for each state was 652 with a range of 8 to 17,557. The SNHs that had contracted with IRIS during this period were located only in Georgia and Texas, where 5,293 and 31,713 patients were seen in each state respectively.

Fig. 1.

Map of United States showing location of county SNHs and non-SNHs. Insets showing states with county SNHs: Georgia (above) and Texas (below). Circles indicate non-SNHs, and crosses indicates county SNHs. SNHs, Safety Net Hospitals.

Increased Overall Prevalence of Ocular Disease in SNHs

The percent of patients who screened positive for any eye disease was 26.5% overall, and this was significantly higher at SNHs compared with non-SNHs (31%, vs. 23.6%, p < 0.0001; Fig. 2A). Significantly increased rates of DR, DME, suspected glaucoma, suspected cataracts, and macular hole were detected in the SNH population compared with non-SNH (Fig. 2B). Reduced detection of vein occlusion, hypertensive retinopathy, and both dry and wet forms of age-related macular degeneration were found in the SNH population compared with non-SNH (Fig. 2B).

Fig. 2.

(A–D) Graph of overall disease prevalence for county SNHs and non-SNHs. (A) Percentage of patients with screening positive for disease, delineated by safety net status. (B) Relative prevalence of each disease (prevalence in county SNHs/prevalence in non-SNHs). Dashed line indicates relative ratio for overall prevalence of county SNHs over non-SNHs. (C) Percentage of patients screening positive for eye disease in each decade. (D) Percentage of patients screening positive for eye disease separated by gender. AMD, age-related macular degeneration; DME, diabetic macular edema; DR, diabetic retinopathy; HTN, hypertensive.

Overall rates of disease detection by age are shown in Figure 2C. There was a significant increase in disease detection in elderly patients at SNHs compared with non-SNHs. No patient younger than 10 years was seen. In patients aged 10–60 years old, there were disease detection rates of 24.7% and 19% in SNH and non-SNH, respectively (p < 0.0001).

This increased prevalence in the 10–60 age bracket corresponded to the overall increased prevalence for all ages (relative ratio 1.31). However, in patients aged >60 years old, there was significantly increased disease detection relative to detection for all ages (42% vs. 26%, p < 0.0001), which corresponded to a 55% increased relative risk.

The effects of gender on overall eye disease detection are shown in Figure 2D. There was no difference in detection rates between men and women in non-SNHs (23.7% vs. 23.6%, p < 0.7252). There was increased screening prevalence in men compared with women in SNHs (33.9% vs. 29.1%, p < 0.0001) corresponding to a 16% increased relative risk.

When comparing disease detection prevalence in women, there was an increased prevalence of 29.1% in SNHs compared with 23.6% in non-SNHs (p < 0.0001), corresponding to an increased relative risk of 23%. Disease detection in men screened at SNHs was 33.9% compared with 23.7% for men screened at non-SNHs (p < 0.0001), corresponding to an increased relative risk of 43%.

The number of concurrent suspected diseases in each patient was calculated (Fig. 3). The vast majority (>90%) of screening exams identified zero to one suspected diseases. The most commonly concurrent diseases suspected were DR and macular edema (53% of exams that identified at least two diagnoses). A small percentage of exams identified more than three diseases, and there were significantly more of these exam results in the SNH population than in the non-SNH population (1.2% vs. 0.7%, p < 0.0001).

Fig. 3.

Number of patients screened with multiple concurrent diagnoses in non-SNHs and county SNHs.

Increased Prevalence of Diabetic Eye Disease in SNHs

Of the patients seen in SNHs, there was a significantly increased prevalence of DR (Fig. 4). There were significantly more patients with NPDR (18.2% vs. 15.2%, p < 0.0001) and PDR (2.1% vs. 1.0%, p < 0.0001) in SNHs than non-SNHs (Fig. 4A). This corresponded to a 20% increased risk of NPDR. The increased prevalence of NPDR was mostly due to an increased risk of moderate NPDR (1.26 relative ratio), with slight increases in mild and severe NPDR (1.06 and 1.07, respectively; Fig. 4B).

Fig. 4.

(A–C) Graph showing prevalence of diabetic eye disease in county SNHs and non-SNHs. (A) Graph showing prevalence of NPDR and PDR in county SNHs and non-SNHs. (B) Graph showing prevalence of mild, moderate, and severe DR in county SNHs and non-SNHs. (C) Graph showing prevalence of diabetic macular edema in county SNHs and non-SNHs. NPDR, nonproliferative diabetic retinopathy; PDR, proliferative diabetic retinopathy.

There was increased detection of DME in the SNH population compared with the non-SNH population (Fig. 4C), especially pronounced for moderate and severe DME (relative ratios of 2.05 and 4.50, respectively, p < 0.0001). However, the detection of mild DME was less likely in those screened at SNHs compared with non-SNHs (relative ratio 0.89).

Suspected Prevalence of Other Eye Disease in SNHs

Of the patients seen in SNHs, there was a significantly increased prevalence of suspected cataracts, suspected glaucoma, and suspected macular hole (Fig. 2A). The risk of suspected cataracts and suspected glaucoma was significantly greater than the overall risk of eye disease. Patients in SNHs had a significantly lower prevalence of suspected wet and dry age-related macular degeneration, suspected vein occlusion, and suspected hypertensive retinopathy.

Discussion

Teleophthalmology identified higher rates of diabetic eye disease in people seen in county SNHs compared with non-SNHs, suggesting social disparities in eye disease burden. In addition, this teleophthalmology program screened 94,329 people (37,006 at SNHs) in 1 year, adding to the growing literature citing examples of SNHs (and non-SNHs) successfully utilizing telehealth to reach individuals who may not otherwise have had the means to seek care.

Teleophthalmology programs such as this may help address geographic and socioeconomic disparities over time. Bastos de Carvalho et al.²¹ found that after establishing nonmydriatic fundus photography in SNH community clinics, their patients’ odds of being screened doubled. Teleophthalmology implementation in a large Los Angeles County SNH network increased screening rates by 28.2% and reduced wait time for an appointment by 89.2%.²²

After the widespread implementation of teleophthalmology in the Veterans Health Administration (VA), DR screening rates increased from 48%²³ to 74% (Molly J.E. Snider, unpublished data, 2021). Bridging the care gap through telehealth may mitigate transportation, distance, time, health literacy, and trust barriers, furthering the supposition that teleretinal imaging is not only feasible but also has tangible benefits for the most vulnerable communities.

Previous work demonstrates that teleophthalmology offers a cost-effective solution^24,25 in enhancing public health capacity and health care access at safety nets, where limited federal funds and resources are needed to serve the largest number of people. A prospective study in Singapore predicted that total savings, if teleophthalmology were implemented nationwide, would reach $21.6 million.²⁶ The Manitoba Study demonstrated an average savings of $1,007 CAD per tele-ophthalmology examination.²⁷

Generally, savings are attributed to the fact that patients with normal findings may bypass in-person examination,²² thus mitigating burden on eye clinics. Further, the potential benefit of reaching patients earlier and catching pathology early in the natural history of disease may translate to vision preservation, thus saving the federal government hundreds of millions of dollars²⁸ in health care costs and social security for disabled and unemployed patients.

A systematic review by Pasquel et al.²⁵ found that teleophthalmology is the most cost-effective in communities that would derive the most benefit from the advantages of telehealth (e.g., otherwise less likely to receive screening, large population size, heavy disease burden).

The detection of disease overall was 31% in SNHs compared with 23% in non-SNHs, corresponding to a 31% increased risk ratio. People screened at SNHs were also 105% more likely to be identified with moderate DME and 26% more likely to be identified with moderate NPDR. These differences may be attributable to the older average age²⁹ and/or greater prevalence of underlying metabolic disease found in SNHs compared with non-SNHs.²²

The severity of eye disease at the point of screening was also unequal: Patients evaluated at SNHs were 350% more likely to have severe DME and 83% more likely to have three or more concurrent suspected eye diseases than those at non-SNHs. These findings likely suggest screening later in the natural history of disease.

Factors related to delayed screening for DR have been studied extensively, and they disproportionately affect underserved communities; these include inadequate health insurance,^30–32 transportation limitations,³³ depression,³⁴ financial burden,^34,35 health literacy,^32,35 education level,³⁶ distrust of the health care system,³⁵ and competing priorities.³⁷ Although telehealth may be used as a tool to mitigate health inequities, it should not be expected to overcome precipitating barriers.

The rates of diabetic eye disease found in this study are comparable to other SNH populations in the literature. A Los Angeles safety net study²² using teleretinal imaging found that of the ∼20,000 patients screened in a 2-year period, 19.6% required referral for DR detection (compared with 20.2% in this study). The prevalence of diabetes was 19.9% in LA County and 20.2% in the SNHs studied here.¹⁸

Despite the differences in patient demographics among SNHs in LA county,²² Texas,³⁸ and Georgia,³⁹ the similarity in the prevalence of diseases suggests that these findings may be applicable to SNHs across the United States. The DR prevalence in non-SNHs here (16.2%) was also similar to those reported in the literature, including 15% in the West Los Angeles Veterans Affairs.⁴⁰

We identified two patient subsets that had a particularly increased relative risk. First, patients aged 60 years or older had a significantly higher prevalence of eye disease than those aged 10–60, which is not surprising as two of the diseases studied have age as a risk factor. Older patients are also more likely to have chronic conditions, transportation limitations, and live in rural/isolated areas, in addition to geographic and provider access barriers compared with younger, urban patients.²⁹

Second, men seen at SNHs had a greater relative risk of eye disease than men seen at non-SNHs, in addition to women seen at SNHs. Because the most common disease found in this study was DR, these gender differences align with others reported in the literature.² The difference found between men and women at SNHs but not at non-SNHs is a less commonly reported finding, and it may indicate the combined risk factors of low socioeconomic status and masculine gender.⁴¹

Teleophthalmology is an especially important tool in the COVID-19 era, where social distancing requirements have caused a 60% reduction in ambulatory patient visits⁴² and patients themselves may be hesitant to seek medical care until symptoms arise,⁴³ thus losing a valuable window of opportunity for early intervention.

By allowing normal patients to bypass busy clinics altogether, teleophthalmology may help clinics speed through backlogs of patients while also reducing exposure to the virus. The authors hope and expect teleophthalmology to facilitate the continuation of these vision-preserving screenings, especially in underserved communities that were disproportionately impacted by the virus, as the world struggles to recover from this global pandemic.

The limitations of this study include incomplete patient demographics and socioeconomic status. The authors acknowledge that the patient demographic breakdown may have shown significant difference between SNHs and non-SNHs. The race/ethnicity of patients was not available and therefore its association with disease prevalence could not be investigated, though SNHs care for a predominantly non-Hispanic Black and Hispanic population.⁴⁴

Another important demographic characteristic to collect in future studies is homeless status. In addition, there may be a discrepancy between suspected diagnosis and actual diagnosis, as the IRIS service is FDA approved for the detection of DR and macular edema but not for the other ocular conditions studied. Importantly, patient follow-up and outcome were not addressed in this study.

Only two states had county SNHs, whereas the non-SNHs were found in 30 states. The difference in geographic and/or population diversity may also skew the results. This study also did not differentiate between T1DM and T2DM, which have different natural histories. Future directions include studying the effects of patient ethnicity/race, verifying screening diagnoses, and correlating screening diagnosis with treatments and outcomes. In addition, no-show rate after positive DR screening should be followed, as timely follow-up care is crucial to obtain the expected visual benefits of early screening.

These findings suggest that underserved communities may have a greater burden and severity of diabetic eye disease, which would imply that SNHs require more resources not only in outreach and screening, but also for DR treatment. This article demonstrates that county SNHs in Georgia and Texas were able to successfully utilize a teleophthalmology program to capture diabetic and nondiabetic eye disease, and it is conceivable that deployment in other SNHs can also be successful.

Authors' Contributions

M.J.E.S. made substantial contributions to the analysis and interpretation of data, drafted and revised the work, confirmed the final approval of the version to be published, and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

D.L. and S.G. made substantial contributions to the analysis and interpretation of data, revised the work, confirmed final approval of the version to be published, and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

B.C., Y.K., and A.Y.M. made substantial contributions to the acquisition and analysis of data, revised the work, confirmed final approval of the version to be published, and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Disclosure Statement

No author has competing interests with respect to the research, authorship, and/or publication of this article. S.G. was an employee of IRIS at the time of data collection.

Funding Information

B.C. was supported by a fellowship from the National Eye Institute (F30EY025154).