Using Public Datasets to Identify Priority Areas for Ocular Telehealth

Abstract

Purpose: Telemedicine can expand access to ocular services, but barriers include restrictive policies and poor reimbursement. A tool to identify priority regions for interventions is needed.

Methods: Eye care provider (ECP) density, self-reported visual disability, and demographics were calculated using census data and professional registries. The relationship between visual disability and ECP density was explored in fractional regression models. These data were compared with state telemedicine policy favorability.

Results: For each additional ECP per 100,000 population, there was 0.0111% less disability in the county (95% confidence interval −0.0150% to −0.00719%) in an adjusted model. Of 3,142 counties, 1,078 (34%) were in the worst population-weighted quartile for ECP density and visual disability.

Conclusions: Low ECP density is associated with higher visual disability, suggesting an opportunity for ocular telehealth. Counties with favorable policy climates should be prioritized for telemedicine implementation. Public datasets can be used to survey wide geographic areas to identify areas worthy of detailed needs assessments.

Introduction

Vision loss has a major impact on individual quality of life and a high cost to society due to medical services and lost productivity.^1,2 Although most of the common causes of vision loss in Americans can be mitigated with early detection,³ not all people have adequate access to screening and management services. A 2008 study found that only 54.8% of adults with vision or eye problems and 62.9% of adults with diabetes had a dilated eye examination within the past year.^4,5 Multiple studies have shown that ocular telehealth can successfully widen patient access to high-quality health care while decreasing costs.⁶

While telehealth is beneficial to both patients and the health care system, technical and legal challenges must be overcome for successful implementation. While many of the restrictions on telemedicine discussed in this article have been relaxed during the coronavirus disease 2019 public health emergency, at least partial reinstatement of many restrictions is likely following the pandemic. Two major barriers to telehealth in the United States include physician licensing restrictions and limited coverage by insurance.⁷ It is imperative to permanently address these legal and policy barriers to expand patient access.

The objective of this study was to use publicly available datasets to identify regions (county or larger) of the United States with the highest potential benefit from expansion of ocular telehealth and highest likelihood of success given the local policy environment, as well as those states in which policy efforts will be required to facilitate telehealth expansion.

Materials and Methods

For this study, we merged several publicly available datasets to get estimates of visual disability, density of eye care providers (ECPs), and policy climate. Visual disability was estimated from the American Community Survey (ACS) conducted by the U.S. Census Bureau.^8,9 Responses to question 17b (“Is this person blind or does he/she have serious difficulty seeing even when wearing glasses?”) were accessed as a proportion of respondents at the county or county-equivalent level (all hereafter referred to as counties). To merge various datasets at the same time point, we chose 2016 as our index year, which was the latest date 5-year estimates were available from the ACS using the now deprecated American Fact-Finder tool.¹⁰ Population-weighted quartiles of visual disability were calculated to minimize the effect of very small and very large counties (i.e., each quartile contained 25% of the U.S. population, rather than 25% of counties).¹¹ Because of the high burden of ocular disease caused by diabetic retinopathy, data on diabetes mellitus (DM) prevalence in 2016 were accessed from the Centers for Disease Control and Prevention Diabetes Atlas.¹² Rurality was estimated using the U.S. Department of Agriculture 2013 rural–urban continuum codes.¹³

The total numbers of ophthalmologists and optometrists from the available year closest to 2016 were extracted for each U.S. county from the Area Health Resources File (AHRF),¹⁴ and population-weighted quartiles of overall ECPs per 100,000 population were generated. The AHRF is maintained by the U.S. Health Resources and Services Administration and provides current and historical data at the county level on health facilities, health professionals, and health training programs. This dataset is itself an aggregation of data from over 60 sources, including the American Medical Association (AMA) and Centers for Medicare and Medicaid Services National Provider Identifier (NPI) file.

Because retinal medicine and surgery is not formally recognized by the AMA as a distinct subspecialty of ophthalmology, the county's vitreoretinal workforce cannot be estimated from the AHRF. Therefore, the capacity to treat the most common vitreoretinal disorders was estimated from the Medicare Provider Utilization and Payment Data: Physician and Other Supplier dataset,¹⁵ from 2016, queried for Current Procedural Terminology (CPT) code 67028, “Intravitreal injection of a pharmacologic agent.” To merge these data with the ACS and AHRF, the county was incorporated into the dataset using the 2010 Zip Code Tabulation Area to County Relationship file from the U.S. Census Bureau.¹⁶

Finally, state policy climate was estimated using the 2016 American Telemedicine Association (ATA)-assigned letter grade (A, B, C, or F)⁷ based on insurance parity between in-person and telemedicine visits, coverage for store and forward imaging, physician practice standards, informed consent practices, and licensure requirements.

The relationship between provider density and self-reported visual disability was analyzed in univariable, fractional regression models with individual counties as the unit of analysis. Subsequent multivariable regression models were created, adjusting for potential confounders of rural status, DM prevalence, total population, proportion of the population that was eligible for Medicare (as a proxy for how much of the population was greater than 65 years of age), and per capita income. Estimates for visual disability and density of ECP were then compared with the state favorability ratings from the ATA Gap Analysis to assess the likelihood of successful ocular telehealth implementation. Analyses were performed in Stata 16 (StataCorp LLC, College Station, TX). This study was deemed “not human subjects research” by the University of Vermont Insititutional Review Board and exempt from review.

Results

Visual disability and ECP data were available for all 3,142 2016-era U.S. counties. The median visual disability in a county was 2.6% (range: 0–15.2, intraquartile range [IQR]: 1.6) and the median number of ECPs per 100,000 population was 13.3 (range: 0–132, IQR: 17) (Table 1). Seven hundred thirty-five counties (23.4%) with a median population of 6,817 (range: 89–43839, IQR: 8,359) had no documented ECPs.

Table 1.

Distribution of Self-Reported Visual Disability and Eye Care Provider Density Per Population-Weighted Quartile

	LOW QUARTILE	MEDIUM–LOW QUARTILE	MEDIUM–HIGH QUARTILE	HIGH QUARTILE
Population reporting visual disability	<1.8%	1.8–2.1%	2.2–2.6%	>2.6%
Ophthalmologist/100,000	<3.0	3.0–5.3	5.3–7.7	>7.7
Optometrist/100,000	<12.0	12.0–15.3	15.3–18.5	>18.5
Total eye care providers/100,000 (median)	<15.5 (5.9)	15.5–21.2 (18.2)	21.2–27.0 (23.4)	>27.0 (32.3)

Areas with higher ECP density had lower overall visual disability in univariable regression (p < 0.001). For each additional ECP per 100,000 population, there was 0.0286% (95% confidence interval [CI] −0.0334% to −0.02386%) less disability reported in the county. Since the median county had a population of 26,946 and would have 700 people (2.6%) reporting visual disability, a theoretical county of the same size with access to one additional ECP per 100,000 population would have 7 fewer people reporting visual disability.

When adjusting for rural status, diabetes prevalence, county population, per capita income, and proportion of the population eligible for Medicare, fewer ECPs in a county remained significantly associated with visual disability, although the effect was slightly blunted. In the adjusted model, there would be 0.0111% (95% CI −0.0150% to −0.00719%) less disability reported for each additional ECP per 100,000 population. This same adjusted model likewise suggested rural status, diabetes prevalence, population, per capita income, and proportion of the population eligible for Medicare to be independently associated with visual disability (Table 2).

Table 2.

Outcome of the Multivariable Fractional Regression Model, Adjusting for All the Listed Covariates

	CHANGE IN COUNTY DISABILITY PER UNIT INCREASE IN COVARIATE	p	95% CI
Eye care provider/100,000	−0.0111%	<0.001	−0.0150% to −0.00719%
Rurality	0.0815%	<0.001	0.0636% to 0.0993%
Population size/10,000	0.00102%	0.013	0.000213% to 0.00184%
Per capita income/$10,000	−0.453%	<0.001	−0.524% to −0.383%
Diabetes prevalence	0.0740%	<0.001	0.0616% to 0.0865%
Proportion of population eligible for Medicare	5.24%	<0.001	4.51% to 6.56%

Medicare utilization data for CPT code 67028 as a proxy for vitreoretinal capacity were also analyzed for association with self-reported visual disability. Utilization was only reported in 675 of 3,142 counties (21%). Among counties with reported Medicare 67028 billing, the median number of services was 1,443 (range: 12–62,386, IQR: 4,355), with wide variability, such that the mean number of injections was 3,869 with a standard deviation of 6,326. Among the counties with reported 67028 utilization, there was no significant association between either the absolute number of services billed or the per capita number of services and visual disability. Likewise, there was no significant association between visual disability and a county reporting no 67028 billing versus those that did.

We identified 1,078 of 3,142 (34.3%) counties as ocular telehealth opportunity counties, defined as being in both the lowest quartile of ECP density and the highest quartile of visual disability (Fig. 1—more than 25% of counties due to population weighting of quartiles). Several factors were independently associated with a county being labeled an opportunity county compared with those counties that were not. A county was more likely to be an opportunity county if it had lower per capita income (odds ratio [OR] 0.45 for each additional $10,000 per capita income, 95% CI 0.40 to 0.51), was older (OR 1.06 for each additional percentage of the population that was eligible for Medicare, 95% CI 1.04 to 1.08), more rural (OR 1.20 for each step increase in rurality, 95% CI 1.16 to 1.24), and had higher diabetes prevalence (OR 1.11 for each additional percentage of the population with DM, 95% CI 1.08 to 1.14) adjusted for each of the other factors.

Fig. 1.

Map of the continental United States with shaded opportunity counties. These counties were in the highest population-weighted quartile for visual disability and lowest population-weighted quartile for eye care provider density.

To explore which areas might benefit most from telemedicine interventions, we identified states with 10 or more opportunity counties and an A grade from the ATA for telemedicine policy favorability (New Mexico, Oklahoma, Virginia, Mississippi, and Tennessee). To see which areas might benefit not only from telemedicine but would also require concerted policy efforts, we identified states with 10 or more opportunity counties with C or lower grade (Indiana, Idaho, West Virginia, Florida, Arkansas, and North Carolina). Of note, the only 2 states with a 2016 F grade from the ATA, Rhode Island and Connecticut, had no opportunity counties.

Discussion

Our analysis shows that several freely available datasets can be used to characterize geographical gaps in the distribution of eye care resources. By combining these resources, we found that a higher density of ECPs in a county was independently associated with lower amounts of visual disability. We were also able to pinpoint counties that might present opportunities for ocular telehealth due to high prevalence of visual disability and low provider density. Cross-referencing these counties against a database of state favorability toward telemedicine at a policy level revealed those communities that might benefit from more immediate telemedicine programs and those that might require state-level lobbying efforts.

Other groups have used large datasets to explore associations between visual disability and sociodemographic factors. Chou et al.¹⁷ explored patterns of eye care use in the 22 U.S. states using the visual impairment and access to eye care module of the Behavioral Risk Factor Surveillance System surveys. This group found that county-level factors such as per capita income and racial/ethnic makeup were independently associated with eye care utilization and concluded (as we do) that contextual factors need to be addressed to improve vision health. The same group had previously shown that residence within a county with low ECP density was associated with lower likelihood of receiving an annual dilated eye examination among insured adults with diabetes, and likewise recommended expansion of eye care services.¹⁸ Geospatial and census data have also been used on a more granular level to explore the distribution of medical resources and the patient population using a telemedicine screening program for diabetic retinopathy in North Carolina.¹⁹ This group concluded that geospatial analyses allow for improved characterization of barriers to care and more precise expansion planning for their telemedicine network.

While our approach allowed us to look at the entire United States without exclusions based on insurance or disease status, our approach does have several limitations. Our measure of visual disability was self-reported and could have been interpreted differently by different respondents. Additionally, not all the data streams we used are updated in real time; there is a lag between when the data are collected and when they are made publicly available. Additionally, the prevalence of the severe level of visual impairment queried in the U.S. Census ACS was quite low and did not permit granular analysis of more mild visual disability. However, given the large sample size of all U.S. counties and robust associations we found, we believe the patterns of more mild vision loss would reflect the associations we found as well.

Interestingly, our attempt to correlate access to vitreoretinal services with visual disability was not successful. This is likely because Medicare CPT utilization by individual providers does not always reflect the actual geographic diversity of vitreoretinal capacity. Procedures are listed at the single site where a provider's NPI is registered even when they may have been performed at a satellite location in another county. Additionally, facility-performed injections may not always appear in the utilization dataset. Other methods may need to be developed to explore the capacity for retina and other ophthalmological subspecialties.

Conclusions

Any effort to improve access to eye care may likewise improve visual disability, and we used public datasets to identify those regions most amenable to improving access through ocular telehealth. Counties with low access to eye care services and with high visual disability may present the clearest opportunity for telemedicine intervention, and counties in states with favorable policy climates should be prioritized for implementation. Overall, our approach is not a substitute for detailed needs assessment before implementing telemedicine in a region, but it does provide a framework for rapid low-cost exploration of wide geographical areas to identify communities worthy of further efforts for ocular telehealth interventions.

Authors' Contributions

All of the listed authors meet criteria for authorship.

Disclosure Statement

None of the authors have any competing interests, personal financial interests, employment conflicts, or any other competing interests to report.

Funding Information

C.J.B. was supported by the National Institute of General Medical Sciences of the National Institutes of Health under Award Number P20GM103644. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. B.Y.K. and S.D. were supported, in part, by the Elliot W. Shipman Professorship Fund.