Teleretinal Screening for Diabetic Retinopathy in Six Los Angeles Urban Safety-Net Clinics: Initial Findings

Abstract

Diabetic retinopathy is a leading cause of blindness in US adults. This paper presents initial results of a teleretinal screening project for diabetic retinopathy involving six Los Angeles safety net clinics. A total of 1,943 patients have been screened for diabetic retinopathy by three ophthalmologist readers, with 416 receiving a recommendation for referral to specialty care. Of the cases recommended for referral, 24 had proliferative diabetic retinopathy, 62 had severe non-proliferative diabetic retinopathy (NPDR), 60 had moderate NPDR, 19 had mild NPDR, 138 had a non-diabetic condition, such as glaucoma, 63 had clinically significant macular edema without retinopathy and 50 had non-gradable images. Between 3% and 12.2% of retinal images taken at the clinics were assessed by readers as inadequate for any interpretation. The study shows the feasibility and challenges of teleretinal screening for diabetic retinopathy in urban areas facing specialist shortages and an overburdened, under-resourced safety net care-delivery system.

Introduction

Diabetes affects an estimated 25.8 million people in the US or 8.3% of the population; 7.1% of non-Hispanic whites, 8.4% of Asian Americans, 11.8% of Hispanics and 12.6% of non-Hispanic blacks have been diagnosed with the condition.¹ A complication of diabetes mellitus is diabetic retinopathy, damage to the blood vessels of the retina, which is the leading cause of blindness among US adults between the ages of 20 and 74 years.^{2, 3} Providing individuals who have diabetes with timely retinal screening examinations, diagnosis and treatment of retinopathy would reduce the incidence of loss of sight from diabetic retinopathy. Laser photocoagulation surgery is an effective means of treating retinopathy if detected early (e.g., through routine yearly screening).^{4, 5}

Safety net clinics in the United States offer primary health care services to over 16 million patients nationwide, 2.3 million of whom live in California, whether or not those patients have the ability to pay for health care services.⁶ For urban, medically underserved patients in Los Angeles, primary care safety net clinics provide monitoring and other services for diabetic patients but they are usually not equipped to provide retinal screening examinations or specialty care. The reference standard for detecting and diagnosing diabetic retinopathy is seven-field 35-mm stereoscopic color fundus photographs and grading protocols, as defined by the Early Treatment Diabetic Retinopathy Study (ETDRS).⁷ However, ETDRS photography is impractical for use in safety net primary care settings for the following reasons: (a) the need for skilled photographers comfortable with patient pupil dilation, (b) the high cost of film processing and archiving, and, (c) the logistical difficulties involved in disseminating processed film to an ophthalmologist who does not work at the primary care clinic. Thus, diabetic patients in Los Angeles who visit safety net primary care clinics are typically referred to Los Angeles Department of Health Services public county hospitals for yearly ophthalmic screening examinations and treatment of diabetic eye complications. Unfortunately, this results in a large number of medically underserved patients being channeled into a public health system facing severe resource constraints for both routine ophthalmic screening examinations and treatment of diabetic eye disease.

The use of telemedicine to screen for diabetic retinopathy (as well as retinopathy of prematurity) has shown great promise both in the US and internationally.^8–16 For example, a randomized trial of point-of-service screening for diabetic retinopathy in a rural primary care clinic using nonmydriatic digital cameras and transmission of digital images for review by local ophthalmologists produced a six-fold increase in the rate of retinal examination compared to the usual referral protocol.¹⁷ Although many studies of teleretinal screening focus on its benefits for rural areas or developing countries, where eye care providers are few and patient access to providers is often limited by geographic distance, several studies have demonstrated that inner-city African Americans and Hispanics with diabetes also have limited access to appropriate eye care and that this is tied to delayed diagnosis and treatment of diabetic retinopathy.^18–20 This makes the assessment of diabetic retinopathy in medically underserved urban areas a good candidate for telemedicine.

In the Los Angeles safety net setting, the potential benefits of using telemedicine for diabetic retinopathy screening include the following: (a) referral of diabetic patients to the public (county) hospital for eye complications is limited to those patients who have been triaged by ophthalmologists using telemedicine, reducing the burden on county public health services, (b) attention can be focused on getting those patients that require referral to ophthalmology timely treatment for their conditions, and, (c) primary care providers can schedule their patients for local retinal imaging at the primary care center and monitor patients directly to boost compliance. With these potential benefits in mind, investigators at the Charles Drew University (CDU) Center for Biomedical Informatics partnered with six South Los Angeles community health centers on an NIH-funded study of teleretinal screening for diabetic retinopathy in the safety net setting. The six clinics are independent safety net clinics that are part of a healthcare collaborative formed in 2006 to address specialty care and other concerns affecting members.

Four of the six clinics had contracted with optometrists in the past to provide teleretinal screening services using the EyePACS^{11, 21} web-based retinal image viewing software. Two of those four clinics shared a nonmydriatic digital retinal camera, with patients from one clinic getting their retinal images taken at the other clinic site. Although CDU has its own teleretinal viewing software,¹⁴ a decision was made to use EyePACS at all six sites due to: (a) the majority of the clinics’ familiarity with that software and the cost associated with retraining clinic staff on a new system that would be used only for the duration of the grant, and (b) the UC Berkeley EyePACS program’s provision of long-term support services (e.g., camera repair, staff retraining), unconstrained by a grant funding period. Altogether, the six clinics served 99,574 patients in 2010, including children, and had a total of 356,315 patient encounters. Out of this patient population, our study focuses on diabetic retinopathy screening for adults 18 years of age and older.

Overall study goals include assessing: (a) the proportion of diabetic patients at risk for retinopathy from each primary care clinic who have received a retinal screening examination in compliance with current evidence-based guidelines; (b) the proportion of diabetic patients from each primary care clinic that have required ophthalmologic referral or treatment and subsequently received it (through a retrospective medical record review); (c) the proportion and quality of readable images; and, (d) the acceptability of using teleretinal screening to evaluate diabetic retinopathy in inner-city clinics through a qualitative analysis of clinic staff views of its utility, workflow impact, and ease of use, as well as safety net patient perceptions of telemedicine post clinic-visit. This paper will emphasize initial results for the first three study goals, while a separate paper more thoroughly addresses the qualitative analyses of the last study goal.

Methods

Agreements between Study Partners

Institutional Review Board (IRB) approval for the study was obtained from Charles Drew University of Medicine and Science. Memoranda of understanding (MOUs) were signed by CDU, the CEOs of all participating clinics, and the director of the clinics’ healthcare collaborative outlining the scope of the study and responsibilities of each party. An additional MOU between CDU and EyePACS was also signed. The CDU Center for Biomedical Informatics agreed to:

• provide three board-certified ophthalmologists to review retinal images taken at the clinic sites regardless of the health coverage status of patients,

• have the ophthalmologists perform retinal image readings and provide their assessments of patient images to the clinics within 14 days of retinal image upload, and,

• provide two new nonmydriatic digital retinal cameras and staff training on the use of the cameras to clinics that had no prior experience with teleretinal screening (two out of the six clinics).

The six clinics agreed to:

• identify staff to be trained on the use of digital nonmydriatic cameras,

• allocate sufficient staff time for implementing the digital retinal screening project, including staff photographers to capture retinal images, and staff to manage electronic referrals to the county,

• make all follow up referrals to outside agencies (e.g., county health facilities) for patients with positive findings, and,

• provide access to patient medical records for chart review.

The director of the clinics’ healthcare collaborative agreed to:

• serve as project liaison between the research team and the participating community health centers,

• incorporate work on the study into the South Los Angeles Collaborative for Specialty Care Access work plan, specifically to address the need for intervention and treatment for patients with positive findings, and,

• provide participating community health centers with technical assistance to fast-track and prioritize patients with positive findings at a county health facility.

Digital Retinal Image Capture and Evaluation Using Telemedicine

New image takers were provided training and were evaluated on use of the retinal cameras and the purpose of a screening program by EyePACS staff (per MOU with EyePACS). They were also given information on steps to follow for image taking through the EyePACS handbook.²¹ Image takers had to upload ten non-patient retinal images and have them graded as being of sufficient quality for diagnosis in order to be certified as ready for patient contact. The four clinics with prior teleretinal screening experience used Canon CR-DGi nonmydriatic retinal cameras for image capture while two clinics used new Canon CR-1 Mark II nonmydriatic retinal cameras purchased for the study. Photographers captured four retinal images for each eye – one external picture and three internal pictures covering: (1) the optic disc and macula, (2) the optic disc alone, and (3) the macula and retina temporal to the macula.¹¹ They uploaded retinal images to the web-based system for ophthalmologist review.

Ophthalmologists have been reviewing image quality to determine adequacy for diagnosis. Images of sufficient quality were examined for the presence of mild, moderate, or severe non-proliferative diabetic retinopathy, proliferative diabetic retinopathy, clinically significant macular edema, and other clinically significant conditions that would warrant referral for specialty care.⁷ The ophthalmologists’ reviews of image quality enabled us to make decisions about retaking retinal images, retraining image takers, evaluating camera equipment, etc.

Retrospective Data Collection

In order to meet the study goals, medical records of diabetic patients at the six clinics were abstracted for information on patient date of birth, height, gender, race/ethnicity, medical visit history, most recent foot exam, all specialist referrals within the last year, co-morbid conditions, marital status, household income, number of household members, and level of education. Glycated hemoglobin values, weight and blood pressure data were gathered at time points prior to and after the introduction of teleretinal screening. Data were collected and analyzed using PASW Statistics 19 (SPSS Inc., Chicago, IL) and STATA 11.0 (STATA Corporation, College Station, TX).

Results

Table 1 outlines clinic utilization, patient population characteristics and the number of diabetic patients for the six clinics participating in the study. The clinics treated a total of 9,432 diabetic patients in 2010. The large minority population at the clinics is consistent with an inner-city safety net setting. The four clinics out of six with prior teleretinal screening experience are Clinics C, D, E, and F.

Table 1:

Clinic utilization characteristics from 2010

	Clinic
	Clinic A	Clinic B	Clinic C	Clinic D	Clinic E	Clinic F
	Number of unique patients served	9,510	26,211	3,868	36,149	14,605		9,231	Total = 99,574
% of patient population that’s Hispanic	53.6%	88.3%	59%	85.4%	85.2%	27.9%
% of patient population that’s non-Hispanic black	4.7%	7.5%	26.1%	12.7%	4.4%	58.9%
% of patient population that’s Asian	0.7%	1.1%	2.8%	0.6%	0%	10.7%
Number of patients with type 2 diabetes	823	1,494	589	2,800	2,233	1,493	Total= 9,432
Number of patient encounters	39,530	110,577	11,361	105,924	56,630	32,293	Total = 356,315

Prospective Teleretinal Screening Study

Two ophthalmologists began screening retinal images for the study on September 8, 2010 and a third began screening retinal images on December 14, 2010.

Between September 8, 2010 and June 6, 2011, a total of 1,943 diabetic patients were successfully screened for diabetic retinopathy. Of these patients, 416 were recommended by the ophthalmologists reading images for referral to specialty care. Twenty-four patients had proliferative diabetic retinopathy (PDR), 62 had severe non-proliferative diabetic retinopathy (NPDR), 60 had moderate NPDR, and 19 had mild NPDR. Table 2 lists the number of patients identified as having conditions requiring referral for specialty care, including clinically significant macular edema. Additional conditions recommended for referral included cataract, glaucoma, non-diabetic maculopathy, and vascular occlusion. Fifty patients with non-gradable images were also recommended for referral to specialty care.

Table 2:

Number of recommended referrals to specialists from teleretinal screening for diabetic retinopathy at six South Los Angeles clinics

Total number of patients screened	1943	100%
Total number of patients recommended for referral to a specialist	416	21.4%
Conditions recommended for referral by ophthalmologist readers
PDR only	17	0.9%
PDR and macular edema or other conditions	7	0.4%
Severe NPDR only	28	1.4%
Severe NPDR and macular edema or other conditions	34	1.7%
Moderate NPDR only	22	1.1%
Moderate NPDR and macular edema or other conditions	38	2.0%
Mild NPDR only	19	1.0%
Macular edema only	52	2.7%
Macular edema and other conditions	11	0.6%
Other conditions	138	7.1%
Non-gradable images recommended for referral	50	2.6%

Table 3 outlines the ophthalmologists’ ratings of image quality across the six clinics. The category of “insufficient for full interpretation” applied to images that had interpretable elements but were deficient in a way that made overall categorization difficult. For example, this would apply in a case where severe disease in need of referral could be clearly ruled out, but image quality was not high enough to discount subtle changes (i.e., a few scattered microaneurysms), or in a case in which image quality of one eye was adequate or better (and therefore sufficient for interpretation) while that of the other eye was not. Based on feedback from the ophthalmologists, three image takers at two of the clinics with prior teleretinal screening experience (Clinics D and E) were retrained in March 2011. Research also revealed that digital cameras at those two clinics (D and E) had developed problems that reduced the quality of the images taken and necessitated repair of the cameras in March of 2011.

Table 3:

Retinal image quality ratings by clinic

Image Quality	Clinic A	Clinic B	Clinic C*	Clinic D	Clinic E	Clinic F
Insufficient for Any Interpretation	9.0%	3.0%	6.2%	12.2%	10.2%	3.7%
Insufficient for Full Interpretation	20.5%	10.8%	25.5%	34.1%	20.8%	27.3%
Adequate	33.6%	36.0%	28.6%	28%	33.4%	33.2%
Good	14.8%	19.7%	15%	7.7%	15.3%	14.4%
Excellent	10.2%	12.8%	6.7%	0.9%	6.8%	9.1%
Not rated	11.9%	17.7%	18%	17.1%	13.5%	12.3%

^*Clinics C & F share a camera

The two clinics new to teleretinal screening (A and B) had markedly different experiences with photographer training even though the same initial camera training was provided at both clinics on the same day. At Clinic B, image takers needed several months and multiple camera retraining sessions to obtain retinal image certification. As a result, Clinic B began patient screenings in December 2010, while image takers at Clinic A began screening patients in August 2010.

Retrospective Study

Table 4 provides a summary of data abstracted thus far on diabetic patients at risk for retinopathy from the six clinics. It gives an overview of demographic characteristics, insurance status, patient population breakdown by clinic, and average glycated hemoglobin values. The data in Table 4 are abstracted from the same patient population as data in the prior tables outlining the screening results. However, since retrospective data collection is still in process, the data from the (prospective) screening results have not yet been fully matched to retrospective patient data.

Table 4.

Selected Characteristics of the Study Population (N=1074)

Patient characteristics	Number	Percent
Age (mean (s.d.), years) 54.0 (9.9)
Gender
Male	391	36.4
Female	683	63.6
Race/Ethnicity
White	1	0.1
Latino	810	75.4
African American	176	16.3
Asian	85	7.9
American Indian	1	0.1
Other	1	0.1
Missing	1	0.1
Education
No Education	7	0.6
<High School	139	12.9
Some High School	51	4.8
High School/GED	90	8.4
Some College	10	0.9
Associate Degree	23	2.1
Bachelors Degree	20	1.9
Masters	4	0.4
Missing	730	68.0
Marital Status
Single	277	25.8
Married	579	53.9
Divorced	71	6.6
Widowed	58	5.4
Separated	69	6.4
Missing	20	1.9
Household Income
0 to 9,999	697	64.9
10,000 to 25,000	333	31.0
25,001 to 40,000	21	2.0
40,001 to 70,000	0	0.0
70,001 to 85,000	0	0.0
85,001 to 100,000	0	0.0
100,001 or more	0	0.0
Missing	23	2.1
Insurance
HMO	25	2.3
Medi-cal	66	6.2
Medi-caid	13	1.2
Medi-care	26	2.4
Other	12	1.1
PPP	855	79.6
Private Insurance	1	0.1
Self Pay	76	7.1
Missing	0	0.0
Primary Language Spoken
English	257	23.9
Spanish	765	71.3
Other	52	4.8
Missing	0	0.0
Charts Abstracted (by Clinic)
Clinic A	165	15.4
Clinic B	159	14.8
Clinic C	148	13.8
Clinic D	105	9.8
Clinic E	153	14.2
Clinic F	344	32.0
Hemoglobin A1C in last year (range)
<= 7.0	145	13.5
>7.0	344	32.0
Missing	585	54.5
Hemoglobin A1C mean (s.d.) 8.4 (2.1)

Discussion

While a number of studies have shown that on average, only 60% of diabetic patients in the US receive timely eye examinations,^22–26 some studies of the urban safety net setting have shown annual eye examination rates for inner-city diabetic patients to be lower than 25%.^{20, 27, 28} The challenges documented in our study may help to illuminate some of the reasons for the disparity in screening rates while the successes demonstrate the opportunities that teleretinal screening provides in spite of these challenges. Prior to the introduction of teleretinal screening, diabetic patients at the six safety net clinics had to be referred to Los Angeles County facilities for routine eye examinations. Patient wait times to receive appointments for routine eye screening at county facilities ranged from four to eight months, due to the high volume of uninsured and underinsured patients served by those facilities. Teleretinal screening thus presents a means of addressing the poor annual eye examination rates for inner-city safety net clinics and beginning to close the screening-rate gap. In the almost nine-month period between September 8, 2010 and June 6, 2011, our study has screened 1943 patients for diabetic retinopathy, which represents 20.6% of the diabetic patients seen at these clinics. We expect that by the 12-month mark, the annual screening rate achieved will exceed the national average for inner-city safety net clinics.

Our initial results show that the vast majority of diabetic patients screened for retinopathy did not have retinopathy and were not recommended for referral to specialty care. This suggests that teleretinal screening for diabetic retinopathy in the urban safety net setting can be an effective way of highlighting precisely those patients who need specialty care while simultaneously reducing the burden on county public health facilities with regards to access to specialists. The study also illustrates challenges in implementing teleretinal screening in the safety net setting, including the need for effective monitoring of image quality to ensure its suitability for diagnostic recommendations as well as camera repair and retraining of photographers when necessary. It is striking that of the subset of patients that had a current glycated hemoglobin (hemoglobin A1C) value available in their medical records, over 70% had levels above 7% (Table 4), an indication of less than satisfactory glycemic control. This has substantial implications for diabetic complications, including diabetic retinopathy.

Although the EyePACS protocol recommends screening of diabetic patients at the time of the visit with their primary care providers, (i.e., integrating diabetic retinopathy screening with the primary care visit), the six clinics involved in the study have different approaches to handling diabetic retinopathy screening based on their existing workflow, staffing levels and other resources. None of the clinics currently integrate retinopathy screening into the primary care visit. All the clinics in this study use medical assistants as photographers, in contrast to similar studies in developing countries in which primary care physicians themselves perform camera duties.¹⁶ Given the shortage of primary care physicians and nurses in safety net settings in California, the clinics’ decision to use trained medical assistants rather than physicians and nurses for photography is a necessary accommodation to circumstances.

Our study results show that image quality varied among the clinics; there were several factors that may have affected this outcome. These factors include: (a) the presence or absence of a dedicated diabetes program, (b) whether or not a clinic had staff allocated exclusively for teleretinal screening, (c) camera equipment problems, (d) the need for personnel to be retrained for teleretinal screening, and, (e) a clinic’s image-taking schedule for teleretinal screening (daily versus weekly). Only one clinic (Clinic E) has a dedicated diabetes management program staffed by two medical assistants who act as the photographers for diabetic retinopathy screening, in addition to their other duties for the program. Other clinics use between one and three medical assistants as photographers, however, these clinic staff are also involved in a variety of other programs (e.g., HIV prevention programs). Image upload rates and quality scores were also affected by camera equipment problems at two clinics and personnel retraining.

As of March 2011, the two clinics (D and E) that needed camera repairs had the highest percentages of images rated insufficient for any interpretation, at 18.6% and 18.8% respectively. Photographers at these two clinics were retrained in March 2011. Post-certification personnel retraining on camera use was needed at clinics for two reasons: (a) turnover of medical assistants at participating clinics, necessitating training for new medical assistants, and, (b) difficulty in consistently gaining expertise with camera use for medical assistants who take patient images just once or twice a week. Since all images rated as insufficient for full interpretation had to be retaken, in the long term, the clinics may weigh the associated costs of rescheduling patients for image-taking with the costs of retaining dedicated photographers for diabetic retinopathy screening. In addition to the retraining, as clinic E began to screen patients almost daily, the medical assistant responsible for photography improved tremendously in skill over medical assistants at other clinics that screen patients just once or twice a week.

To further address concerns about image quality, one of the ophthalmologists on the study (LP) met individually with photographers at different clinics to discuss image quality issues and to give them pointers for improving image quality. While Clinics D and E still have the highest percentage of images rated insufficient for any interpretation, by June 2011, those percentages had dropped to 12.2% and 10.2% respectively (Table 3). Interestingly, the two clinics (A and B) that had no prior experience with teleretinal screening had the highest percentages of images rated excellent. Clinic B, which had great difficulty in getting photographers certified in a timely manner and whose photographers required a large number of repeat trainings prior to eventual certification, ended up with the best image quality ratings across the six clinics by June 2011.

With regards to workflow, other researchers have observed that the introduction of telemedicine into a clinical setting alters existing work practices and power relationships.²⁹ The ideal situation would be to integrate photography into the diabetic primary care visit. However, the clinics’ decision not to integrate teleretinal screening into the primary care visit is based primarily on financial considerations: none of the participating clinics can afford to have a medical assistant dedicated solely to photography at this time. In light of this, our study aims to implement teleretinal screening in a manner that the clinics can afford in terms of financial and personnel costs and that they will be able to continue at the end of the grant funding period. Since there is a relationship between the workflow at a clinic and ability to screen, image quality, image upload rates, etc., we will continue to monitor how different safety net clinics manage this with the limited resources that they have.

With regards to the assessment of images, although some international studies of teleretinal screening for diabetic retinopathy have found that non-ophthalmic physicians, with adequate training and ophthalmology oversight, may be able to provide effective review of teleretinal images,^{13, 30} to date, the use of non-eye care providers to screen for diabetic retinopathy is not considered standard of care within the United States. For this and other reasons (the time needed to train non-ophthalmologist image readers, the complicated specialty referral system currently in place in Los Angeles County, and the aforementioned shortage of family practice and primary care physicians), we decided to use board-certified ophthalmologists as readers for our study. In this study, the peer review process using ophthalmologists ensures a greater level of trust in the recommendations of image readers by the receiving ophthalmologists and removes the need for a repeat screening examination in patients that need to be scheduled for a treatment appointment directly.

As stated in the introduction, the potential benefits of a teleretinal screening program for the Los Angeles safety net system include reducing the screening burden on county facilities and providing timely treatment for diabetic patients identified as having retinopathy. At the present time, patients from the teleretinal screening program who need referral for specialty care are input into the county’s referral processing system (RPS) by each clinic’s referral specialists or photographers, based on varying personnel responsibilities at each clinic. This can result in patient wait times on the order of several months or greater, depending on the total number of patients referred from different clinics to that county facility. Currently, the county facility has one retinal specialist who can provide half a day a week of complicated laser treatment for 8 patients and a number of general ophthalmologists able to perform uncomplicated retinal laser therapy. The waiting list for the laser clinics includes patients from several South Los Angeles primary care clinics, and in the past has included referrals from another county facility while its laser was undergoing repair.

In order to flag patients requiring immediate attention, the CDU team has set up a stopgap system with the Chief Medical Officer (CMO) of the closest county facility for the duration of the project to identify patients with moderate or severe non-proliferative diabetic retinopathy, proliferative diabetic retinopathy, and clinically significant macular edema so that they can receive expedited treatment. The stopgap approach is intended to address the provision of timely treatment to patients in the short-term while the clinics and county facility devise a long-term solution. Working with other stakeholders, the director of the clinics’ collaborative has identified gaps in the system for triage and presented the county with a report identifying those gaps and potential solutions. The county intends to replace the existing referral system (RPS) with a new system that addresses these gaps and facilitates triage, which would allow for better integration of future referrals from teleretinal screening into the Los Angeles County eye care clinics. To improve the specialty care referral process, the participating safety net clinics have recently hired a patient navigator who sits in the county facility to which most of their patients are referred and guides the patients through the process for receiving specialty care.

In summary, our initial findings demonstrate both the feasibility and the challenges of utilizing teleretinal screening in the safety net setting as a means of identifying and increasing timely treatment for diabetic patients with retinopathy. Future work will include an analysis of the cost effectiveness of teleretinal screening and an assessment of the impact of different patient outreach methods on improving screening rates.