Abstract
Purpose:
To investigate prevalence of diabetes (DM), diabetic retinopathy (DR), and areas with highest rates of undetected DR. To quantify and map locations of disparities as they relate to poverty and minority populations.
Methods:
Retrospective cohort study from large regional health data repository (HealthLNK). Geographic Information System (GIS) analysis mapped rates of DM and DR in Chicago area ZIP Codes.
Results:
Of 1,086,921 adults who met the inclusion criteria, 143,790 with DM were identified. ZIP Codes with higher poverty rates were correlated with higher prevalence of DM and DR (Pearson’s correlation coefficient 0.614, p<0.05, 0.333, p<0.05). Poverty was negatively correlated with likelihood of DR diagnosis (-0.638, p<0.05). Relative risks of DM and DR were calculated in each ZIP Code and compared to actual rates. 36 high-risk ZIP Codes had both high-risk of DM and low DR detection. In high-risk ZIP Codes 85.4% of households self-identified as ethnic minority and 33.0% were below the Federal Poverty Level (FPL). Both percentages were significantly higher than the Chicago average of 50.5% minority and 19.9% below FPL (p<0.05). 67 ideal ZIP Codes had both low risk of DM and high DR detection. In ideal ZIP Codes 32.6% of households self-identified as minority, and 10.2% were below the FPL (p<0.05).
Conclusions:
A health care disparity exists with regards to DM and DR. High-risk ZIP Codes are associated with higher poverty and higher minority population, and they are highly concentrated in just 17% of the ZIP codes in the Chicago area.
Keywords: Health Care Disparity, Diabetes, Diabetic Eye Disease
Introduction
Visual impairment and blindness present a serious public health challenge in the United States. The leading cause of acquired vision loss in economically active people is diabetic eye disease. The Centers for Disease Control and Prevention (CDC) estimates that diabetes causes 12,000–24,000 new cases of blindness in the United States each year (1). It is projected that the number of individuals with diabetes in the United States will increase to 54.9 million (a 54% increase) by 2030 (2). The prevalence of diabetic retinopathy (DR) in the United States will nearly triple from 5.5 million cases in 2005 to 16.0 million cases by 2050 (3). More severe vision threatening diabetic retinopathy (VTDR), will have an expected increase from 1.2 million cases in 2005 to 3.4 million cases by 2050 (3–4). The increasing prevalence of diabetes, DR, and VTDR in the United States is consistent with worldwide trends. The World Health Organization (WHO) projects that the number of people with diabetes worldwide will more than double between 2000 and 2030, leading to more cases of DR and VTDR (5).
In the United States, the rates of diabetes, DR, and VTDR are significantly higher in minority populations (particularly in non-Hispanic Black and Hispanic populations) and among individuals of lower socioeconomic status (SES) (6–7). The CDC reports that the prevalence of diabetes among non-Hispanic Whites in 2017 was 7.4%, while non-Hispanic Black and Hispanic populations experienced significantly higher rates of diabetes (12.7% and 12.1%, respectively). The prevalence of DR among non-Hispanic Whites (24.6%) was lower compared to non-Hispanic Black (38.8%) and Hispanic (34.0%) populations. The disparity is even greater with VTDR. The prevalence of VTDR in non-Hispanic Whites is 3.2%, while it is more than double in the Hispanic population (7.3%), and almost triple in the non-Hispanic Black population (9.3%). These statistics clearly demonstrate disparities in diabetes and extent of eye disease, particularly in minority populations (6).
In addition to higher rates of diabetes, DR, and VTDR, low income and minority populations have lower rates of screening for diabetic eye disease and poorer glycemic control (measured by hemoglobin A1c levels), putting them at higher risk for vision loss. It has been well documented that screening, early detection, and timely treatment of diabetic eye disease, coupled with glycemic control dramatically decreases visual impairment and blindness in individuals with diabetes (5, 8). Approximately half of patients with diabetes in the United States receive the recommended annual eye screening exam (9). Profound disparities exist, however, with only one-quarter to one-third of individuals receiving appropriate screening in some low income and minority populations (10–11)
Vision loss from diabetes significantly impacts quality of life and has a significant economic burden on society. Reducing this disparity in screening rates by ensuring increased access to regular diabetic ocular screening and timely treatment for DR could save an estimated $624 million and preserve 400,000 person-years of sight each year in the United States (5).
This study captures a large, diverse patient population to compare prevalence of diabetes and diabetic eye disease in different ZIP Codes in the Chicago metropolitan area and identifies areas with lower than expected rates of detection in order to guide the implementation of future screening initiatives. By combining health care information with data related to income and ethnicity, the relationship between these demographic factors and disparities in screening rates for diabetic eye disease can be assessed. With this approach, we seek to demonstrate a method by which the impact of diabetic eye disease can be understood in the context of the geographic distribution of sociodemographic factors.
Methods
A retrospective cohort study was conducted using information from the Chicago HealthLNK Data Repository (HDR), an assembly of non-duplicated, de-identified patient medical records. The HDR includes nearly six million unique patients, of which 2.7 million reside in Chicago (12). These data are restricted to adults aged 19–89 years and contain primarily structured data elements. The HDR includes electronic health records from six health care institutions: Cook County Health and Hospital Systems, Loyola University Medical Center, Northwestern Medicine, Rush University Medical Center, University of Chicago Hospitals and Clinics, and University of Illinois at Chicago Medical Center (12). For this study, the variables included demographics, diagnoses, health insurance type (uninsured, Medicaid, Medicare, or private insurance), visit date, visit location, and patient home ZIP Code. We obtained data from January 1, 2006 through December 31, 2012. The study was reviewed and approved by the Northwestern University Institutional Review Board and the research adhered to the tenets of the Declaration of Helsinki. United States 2010 census data were used to analyze income and race/ethnicity in each of the Chicago area ZIP Codes.
Inclusion/Exclusion Criteria
Patients aged 19–89, living in the Chicago area (determined by patient’s home ZIP Code at time of last visit), having “established care” (defined as two or more visits within the last three years), with any International Classification of Diseases (ICD-9) diagnosis for diabetes (defined as at least one visit with an ICD-9 code: 250.xx (diabetes mellitus), 357.2 (diabetic neuropathy), 362.01 – 362.07 (diabetic retinopathy), or 366.41(diabetic cataract)) were considered for the study. The patients with diabetes in the study were considered to have diabetic eye disease if they had any of the ICD-9 diabetic retinopathy diagnoses 362.01–362.07 or sequalae of diabetic retinopathy 364.42.
Statistical Analysis
We utilized geographic mapping methods to determine statistically significant ZIP Codes with higher rates of diabetes and diabetic eye disease. Details of the patients’ ZIP Code of residence, age, gender, race, and insurance status were retained for geographic adjustment. The observed number of patients in the HDR with these conditions were then compared to the expected values obtained from published diabetes and DR rates, controlled by ZIP Code demographics with Tele Atlas™ (13).
The ratio of actual cases of DR to actual cases of diabetes was compared with the expected ratio to test the null hypothesis of no difference between the observed and expected ratios of diabetic retinopathy. A health disparity or underserved area was identified when the observed rate was lower than the expected rate while controlling for age, gender, race, and insurance status. The adjustments employed a spatial adjustment factor for estimated cases (denominator) in each ZIP Code relative to its gravity of influences of age, gender, race, insurance status, and population density. The population-weighted centroid for each ZIP Code was calculated using 2010 US Census total population estimates by geographically associated US block groups in 250 ZIP Codes in the Chicago area.
ArcGIS 10.7.1 software was used for all calculations and cartographic work (TeleAtlas). An inverse distance-weighted approach was then applied within the two-step methodology to calculate gravity scores for each ZIP Code. The average score value (measure of health care access) was then determined and divided into the scores for each study area ZIP Code (20–23). This produced a multiplier for estimated cases by ZIP Code that was less than 1 in ZIP Codes with less gravity when compared to the average, and larger than 1 when ZIP Code weight was greater than the average. Estimated cases (denominator) by ZIP Code within the Chicago area were then adjusted using the gravity multiplier for diabetes and diabetic retinopathy. All rates were calculated for all ZIP Codes within the Chicago area and statistical differences between observed and geographically adjusted expected rates (p<0.10, p<0.05, p<0.01) were highlighted as underserved areas.
Results
Of approximately 2 million adult patients in the HealthLNK database, 1,086,921 met the inclusion criteria, and 143,790 patients with diabetes were identified (13.2%). Among patients with diabetes, 11,058 (7.7%) had some form of DR.
Diabetes
Patients were grouped according to ZIP Code to assess the prevalence of diabetes and DR geographically. Figure 1 depicts the prevalence of diabetes in the Chicagoland area. Many areas with high rates of diabetes correspond with contiguous ZIP Codes in Chicago’s West Side, South Side, and south suburban areas.
Figure 1: Diabetes Prevalence in Chicago by ZIP Code.
This figure shows the prevalence of the rate of diagnosis of diabetes by ZIP Code. The highest rates of diagnosis of diabetes (large dots) are concentrated in the west and south sides of Chicago, and the west and south suburbs of Chicago.
Superimposing the Figure 1 map of prevalence of diabetes by ZIP Code onto a map showing the percent of households below the Federal Poverty Level (FPL) reveals that a higher rate of diabetes (larger dots) correlates with ZIP Codes with higher rates of poverty (darker blue) (Figure 2). The was a correlation between poverty and diabetes with a Pearson’s correlation coefficient of 0.614 (p < 0.05).
Figure 2: Prevalence of Diabetes by Percent in Poverty by ZIP Code in Chicago.
This figure represents prevalence of a diagnosis of diabetes by ZIP Code superimposed a map demonstrating the percent of the households in each ZIP Code that is below the Federal Poverty Level (FPL). ZIP Codes that have large percentages of the population below the FPL correspond with high rates of diabetes diagnosis. (Pearson’s correlation coefficient of 0.614, p< 0.05.)
Diabetic Retinopathy
The prevalence of DR (Figure 3) was calculated for each ZIP Code and correlated to poverty. ZIP Codes with higher rates of DR correlated with higher percentages of households below the FPL. There was a correlation of poverty to diabetic retinopathy with a Pearson’s correlation coefficient of 0.333 (p < 0.05).
Figure 3: Comparison Retinopathy Prevalence with Poverty by ZIP Code.
The prevalence of a diagnosis of diabetic retinopathy in each ZIP Code is superimposed on Figure 2. Generally, areas with high rates of DR diagnoses had a greater percent of the households below the FPL (Pearson’s correlation coefficient between diabetic retinopathy and poverty was found to be 0.333, p< 0.05.)
Assessment of Expected versus Actual Diabetic Retinopathy (Areas of Underdiagnosed Diabetic Retinopathy)
Zhang et al. estimated prevalence of DR and VTDR in the United States in adults with a diagnosis of diabetes to be 28.5% and 4.4% respectively (6). Figure 3 demonstrates that the prevalence of DR is indeed high in ZIP Codes with an increased prevalence of households below the FPL. However, given the higher prevalence of diabetes in ZIP Codes of more poverty, the rate of DR was often lower than expected.
Figure 4 demonstrates this with the ratio between the prevalence of diabetes and the rate of diagnosis of DR for each ZIP Code. We would expect areas with high diabetes to also have the highest rates of DR. Smaller dots correspond with areas where the prevalence of DR closely matches the expected rate of DR diagnosis given the rate of diabetes in each ZIP Code. Conversely, larger dots correspond to areas where the rate of DR is lower than expected given the rate of diabetes, i.e. underdiagnosis of DR. In order to highlight the areas with the highest rates of underdiagnosis, only the top 20% of ZIP Codes with highest rates of underdiagnosis (large dots) are shaded. These larger dots tended to be located in ZIP Codes with a higher percent of households below the FPL. Areas of presumed underdiagnosis (high prevalence of diabetes, low prevalence of DR diagnosis), tend to be located in ZIP Codes with more poverty, with a statistically significant Pearson’s correlation coefficient of -0.638 (p<0.05). This demonstrates that poverty is negatively correlated with likelihood to be diagnosed with DR. In other words, people with diabetes who live in the poorest ZIP Codes are the ones who are least likely to be diagnosed with diabetic eye disease.
Figure 4: Ratio of prevalence of diabetes to diagnosis of diabetic retinopathy.
This figure compares the prevalence of a diagnosis of diabetes to the rate of diagnosis of DR superimposed on a map of percent of households below the FPL by ZIP Code. Smaller dots correspond with areas where the prevalence of DR closely matches the expected rate of DR diagnosis given the rate of diabetes. Larger dots correspond to areas where the rate of DR is lower than expected given the rate of diabetes (underdiagnosed DR). To better visualize the disparity, only the top 20% of ZIP Codes with highest rates of underdiagnosis (large dots) are shaded. These areas with greatest underdiagnosis of DR tend to be in ZIP Codes with a higher percent of households below the FPL (Pearson’s correlation efficient of −0.638, p < .05).
To further validate the data regarding the ratio of prevalence of diabetes to prevalence of diabetic retinopathy, a separate analysis was performed comparing the relative risks of diabetes and DR for each Chicago area ZIP Code to the actual rate of diagnosis. Each of the 221 ZIP Codes was analyzed and sorted into one of four categories (ideal, high-risk, as expected, and not significant). For the purposes of this study, further analysis was done for the ZIP Codes in the ideal and high-risk categories.
There were 67 ZIP Codes in the “ideal” category consisting of populations with a low relative risk of both diabetes and DR and a high rate of DR diagnosis (Table 1). In other words, the rate of diagnosis of DR matched or exceeded the expected relative risk of DR. U.S. Census data show that the population contained within these ideal ZIP Codes represents 31.1% of the total Chicago area population. The average percentage of racial/ethnic minorities identified in these ZIP Codes is 32.6% (compared to the Chicago area mean of 50.5%) and, on average, only 10.2% of the households in these ZIP Codes fall below the FPL (compared to 17.9% below FPL in the Chicago area). These differences were both statistically significant (p<0.05).
Table 1. Ideal and High-Risk ZIP Code Demographics.
Ideal ZIP Codes have lower numbers of households that self identify as racial/ethnic minority and lower percent of households at or below the FPL compared to the entire Chicago area population (p<0.05). High-risk ZIP Codes have significantly higher numbers of households that self identify as racial/ethnic minority and higher percent of households at or below the FPL. (p<0.05) Chicago area averages can be found on the bottom row. (DM=diabetes mellitus, DR=diabetic retinopathy, FPL=federal poverty level.)
Category | # of ZIP Codes (%) | Population | % Minority | % Households Below FPL |
---|---|---|---|---|
Ideal (Low DM, Low DR, High Rate of Diagnosis) | 67 (30.3%) | 2,107,393 (31.1%) | 32.59% | 10.72% |
High Risk (High DM, High DR, Low Rate of Diagnosis) | 36 (16.2%) | 1,200,387 (17.7%) | 85.36% | 33.02% |
Total Chicago Area | 221 | 6,776,616 | 50.47% | 17.94% |
P<0.05 | P<0.05 |
A second category of “high-risk” ZIP Codes was identified containing populations with high relative risks of diabetes and DR and a low rate of DR diagnosis (Table 1). Low rates of DR in areas with high prevalence of diabetes can be explained by one of two things: either the detection of cases of DR is low due to lack of screening, or, less likely, patients with diabetes in these areas develop less diabetic eye disease than the general population. The population in these 36 high-risk ZIP Codes consists of 17.7% of the total Chicago area population. On average, 85.4% of individuals in these ZIP Codes identify with a minority racial/ethnic group (exceeding the Chicago area average of 50.5%) and 33.0% belong to households below the FPL (higher than the Chicago area average of 17.9%). Both relationships were statistically significant (p<0.05) (Table 1).
These 36 high-risk ZIP Codes (high relative risk of diabetes, low rate of DR diagnosis) were compared to the ZIP Codes from the first GIS analysis of the HDR where the prevalence of DR was lower than expected given the actual rate of diabetes. These 36 high-risk ZIP Codes identified in our second analysis closely corresponded to the areas identified with the GIS approach (Figure 4), detailing significant underdiagnosis of DR. All but four of these 36 ZIP Codes overlapped with areas from the GIS analysis.
Discussion
With the projected rise in incidence of diabetes and DR, timely and accessible ocular screening is necessary for early detection and treatment of diabetic eye disease. This will reduce both individual and societal burdens of visual impairment and blindness. The data described here reveal disparities in the relative rates of diagnosis of diabetes and DR in the Chicago area. Using geographic mapping techniques, we have shown that the prevalence of diabetes and DR are correlated to areas with higher rates of poverty. However, in areas with high poverty and high diabetes, the prevalence of DR was significantly lower than expected. The most plausible explanation is that DR is underdiagnosed in these areas. The most common cause of underdiagnosis of DR is lack of eye examination/lack of screening.
A second analysis was performed using relative risks of diabetes and DR compared to actual rate of diagnosis which validated this correlation between diabetes and DR and poverty. ZIP Codes with more residents in poverty and who identified as Black or Hispanic were found to have higher relative risks of diabetes and DR but a lower rate of DR diagnosis. We hypothesize that the lower prevalence of DR in these populations does not actually indicate a lower burden of eye disease but rather inadequate diagnosis of DR in these areas, likely caused by a lack of access to regular vision screening in these communities.
Of the 221 ZIP Codes analyzed in this study 36 (16.3%) were found to be “high-risk,” with high rates of diabetes, a high-risk for DR, but fewer than expected patients diagnosed with DR. By identifying areas where diabetic eye disease is under-diagnosed, it is possible to precisely identify neighborhoods to deploy resources to detect previously undiagnosed diabetic eye disease.
Further research is needed to fully elucidate the reasons behind such disparities in diagnosis and treatment, but targeted interventions can be implemented in the meantime aimed at reducing disparities in patient outcomes related to diabetic eye disease.
Although this study focuses exclusively on the Chicago metropolitan area, the methodology may be used in other regions. By pinpointing ZIP Codes with the greatest unmet needs, disparities can be addressed more effectively than with a broader, city- or region-wide approach. Even though characteristics of other cities may differ from Chicago, stakeholders can use this approach to examine the patterns of disparities and implement measures to address gaps in DR screening in their communities.
Although this study provides a robust analysis of data from a diverse group of healthcare institutions in the Chicago metropolitan area, there are limitations. While this is a large multicenter study with over 1 million patients who receive regular medical care at the region’s five academic medical centers and a large county health system, some population bias exists, as not all patients in the Chicago area are seen at one of these sites. Other factors such as access or long distance to travel to nearest eye care may impact retinopathy detection, but in our study many of the eye care sites are in or immediately adjacent to the high risk ZIP Codes. Additionally, the prevalence of diabetes in our study population was slightly higher than that of the entire U.S. adult population published by the CDC in 2015, and the rate of DR among subjects in our study population was lower than the national prevalence.
There are several possible reasons for the higher rates of diabetes and the lower rate of DR in our study. First, our study has a higher proportion of patients without insurance. This may have been a barrier to accessing regular eye care screenings and exams, resulting in decreased levels of detection of DR. Second, estimates of the prevalence of DR assume that all patients with diabetes are screened for DR. Nationally, however, only about half of individuals with diabetes receive an annual eye exam, and this number is lower in minority and low-income populations (9). Third, it is possible that patients may have received their primary care at one of the medical systems in the HealthLNK database but received eye care with a provider outside of the database.
Diabetes and diabetic eye disease impact a significant percent of the population, and this is expected to increase over the next few decades. In our study, we used different methodologies to identify and characterize areas at a high-risk of vision loss due to diabetic eye disease. The data clearly demonstrates a health care disparity in diabetic eye disease in both scope and concentrated location. These high-risk ZIP Codes, representing just 16% of all Chicago area ZIP Codes, were associated with poverty and higher percent minority populations. This information can be used to precisely identify populations and areas at the highest risk for diabetic vision loss, deploy resources to areas of greatest need, and reduce this significant health care disparity.
Acknowledgements:
The authors would like to acknowledge Jess Behrens MS, Kathryn Jackson MS, and Theresa Walunas PhD for their assistance with data acquisition and analysis.
Financial Support:
This research was supported on an unrestricted grant from Research to Prevent Blindness, NY, NY and Department of Health and Human Services National Institutes of Health, NATIONAL EYE INSTITUTE Grant Number: 1R21EY024050–01A1 “Understanding Diabetic Eye Disease in the Underserved Using Health Data Exchange”.
Footnotes
Conflict of Interest: No conflicting relationship exists for any author.
References
- 1.Vision Health Initiative: Economic Studies. CDC. Accessed online on 2/14/2019. https://www.cdc.gov/visionhealth/projects/economic_studies.htm.
- 2.Rowley WR, Bezold C, Arikan Y, Byrne E, Krohe S. Diabetes 2030: Insights from Yesterday, Today, and Future Trends. Popul Health Manag 2017;20(1):6–12. doi: 10.1089/pop.2015.018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Vision Health Initiative Report: Projection of Diabetic Retinopathy and Other Major Eye Diseases among People with Diabetes Mellitus United States, 2005–2050. CDC. Accessed online on 6/14/2020. https://www.cdc.gov/visionhealth/publications/diabetic_retinopathy.htm. [DOI] [PubMed]
- 4.Saaddine JB, Honeycutt AA, Narayan KM, Zhang X, Klein R, Boyle JP. Projection of diabetic retinopathy and other major eye diseases among people with diabetes mellitus: United States, 2005–2050. Arch Ophthalmol 2008; 126(12):1740–1747. [DOI] [PubMed] [Google Scholar]
- 5.Ting DSW, Cheung GCM, Wont TY. Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review. Clin Expl Ophthalmol 2016;44:260–277. doi: 10.1111/ceo.12696 [DOI] [PubMed] [Google Scholar]
- 6.Zhang X, Saaddine JB, Chou CF, Cotch MF et al. Prevalence of diabetic retinopathy in the United States, 2005–2008. JAMA 2010;304:649–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Wang W, Yan W, Muller A, Keel S, He M. Association of socioeconomics with prevalence of visual impairment and blindness. JAMA Ophthalmol 2017; 135(12):1295–1302 Doi: 10.1001/jamaophthalmol.2017.3449 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Yau JW, Rogers SL, Kawasaki R, Lamoureux EL et al. Global prevalence and major risk factors of diabetic retinopathy. Diabetes Care 2012. Mar;35(3):556–64. doi: 10.2337/dc11-1909. Epub 2012 Feb 1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Lundeen EA, Wittenborn J, Benoit SR, Saaddine J. Disparities in Receipt of Eye Exams Among Medicare Part B Fee-for-Service Beneficiaries with Diabetes — United States, 2017. MMWR Morb Mortal Wkly Rep 2019;68:1020–1023. DOI: 10.15585/mmwr.mm6845a3externalicon [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Maclennan PA, McGwin G Jr, Heckemeyer C, et al. Eye care use among a high-risk diabetic population seen in a public hospital’s clinics. JAMA Ophthalmol 2014;132(2):162–167. doi: 10.1001/jamaophthalmol.2013.6046 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Owsley C, McGwin G Jr, Lee DJ, et al. Diabetes eye screening in urban settings serving minority populations: detection of diabetic retinopathy and other ocular findings using telemedicine [published correction appears in JAMA Ophthalmol. 2015 May;133(5):621]. JAMA Ophthalmol 2015;133(2):174–181. doi: 10.1001/jamaophthalmol.2014.4652 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.French DD, Behrens JJ, Jackson KL, Kho AN, Walunas TL, Evans CT, Mbagwu M, Margo CE, Bryar PJ. Payment Reform Needed to Address Health Disparities of Undiagnosed Diabetic Retinopathy in the City of Chicago. Ophthalmol Ther 2017. Jun;6(1):123–131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.ESRI [Internet]. Tele Atlas and Products for ArcGIS http://www.esri.com/data/streetmap. Accessed 31 Mar 2016.
- 14.Ting DSW, Tan GSW. Telemedicine for Diabetic Retinopathy Screening. JAMA Ophthalmol 2017;135(7):722–723. doi: 10.1001/jamaophthalmol.2017.1257 [DOI] [PubMed] [Google Scholar]
- 15.Jani PD, Forbes L, Choudhury A, Preisser JS, Viera AJ, Garg S. Evaluation of Diabetic Retinal Screening and Factors for Ophthalmology Referral in a Telemedicine Network. JAMA Ophthalmol 2017;135(7):706–714. doi: 10.1001/jamaophthalmol.2017.1150 [DOI] [PMC free article] [PubMed] [Google Scholar]