Introduction Rural versus urban disparities have been observed in diabetic eye screening, but whether the level of disadvantage in rural versus urban areas is related to these disparities is unclear. Our goal was to determine the role of level of disadvantage in explaining the effect of health systems on rural and urban disparities in diabetic eye screening.
Research design and methods This is a retrospective cohort study using an all-payer, state-wide claims database covering over 75% of Wisconsin residents. We included adults with diabetes (18–75 years old) who had claims billed throughout the baseline (2012–2013) and measurement (2013–2014) years. We performed multivariable regressions to assess factors associated with receipt of diabetic eye screening. The primary exposure was the primary care clinic’s combined level of rurality and disadvantage. We adjusted for the health system as well as patient-level variables related to demographics and comorbidities. Health system was defined as an associated group of physicians and/or clinics.
Results A total of 118 707 adults with diabetes from 698 primary care clinics in 143 health systems met the inclusion criteria. Patients from urban underserved clinics were less likely to receive screening than those from rural underserved clinics before adjusting for health system in the model. After adjusting for health system fixed effects, however, the directionality of the relationship between clinic rurality and screening reversed: patients from urban underserved clinics were more likely to receive screening than those from rural underserved clinics. Similar findings were observed for both Medicare and non-Medicare subgroups.
Conclusions The effect of health system on receipt of diabetic eye screening in rural versus urban areas is most pronounced in underserved areas. Health systems, particularly those providing care to urban underserved populations, have an opportunity to increase screening rates by leveraging health system-level interventions to support patients in overcoming barriers from social determinants of health.
- Diabetic Retinopathy
- Rural Health
- Healthcare Disparities
Data availability statement
Data are available upon reasonable request. Data are available upon reasonable request to those who obtain regulatory approval for data access from the University of Wisconsin Health Sciences Institutional Review Board.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
WHAT IS ALREADY KNOWN ON THIS TOPIC
Rural versus urban disparities have been observed in diabetic eye screening, but it is not clear how socioeconomic disadvantage in rural versus urban communities is related to differences in receipt of diabetic eye screening.
WHAT THIS STUDY ADDS
Using state-wide claims data, we examined the effect of health system on the association of primary care clinic rurality and disadvantage with receipt of diabetic eye screening.
After removing the effect of the health system, patients from urban underserved clinics were more likely to receive screening than their rural counterparts.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Health system-level interventions provide an important opportunity for increasing diabetic eye screening rates and supporting patients in overcoming barriers from social determinants of health.
Despite landmark clinical studies showing that early detection and treatment can prevent 90% of blindness,1 2 receipt of diabetic eye screening remains around 50% among patients in the USA.3 Screening rates are even lower in underserved communities.3–7 Most studies seeking to identify factors associated with receipt of diabetic eye screening have focused on patient-level characteristics.8–10 While there are examples of robust system-level interventions in the literature, there has been limited adoption of such interventions and the focus in clinical practice has been primarily on patient-level interventions.11 Consequently, majority of the interventions to increase screening have centered on improving patient education among high-risk populations, producing limited improvements in screening receipt.12–15
Our prior work and that of others suggest that health systems and primary care clinics may represent underutilized targets for interventions to increase diabetic eye screening. A primary care provider’s recommendation is the strongest factor influencing a patient’s decision to obtain diabetic eye screening.16 17 Primary care provider behavior has been strongly linked to organizational goals, incentives, and infrastructure at the health system and clinic level.18–21 Previously, we observed that the median magnitude of the effect of health system on the odds of screening receipt was 1.24 (IQR, 1.11–1.48) using an all-payer, state-wide claims database that covers over 75% of Wisconsin residents (Wisconsin Health Information Organization (WHIO) All-Payer Claims Database) and including all health systems in the database.11 Patients who obtained care from urban primary care clinics had a higher likelihood of screening receipt than those who obtained care from rural primary care clinics after adjusting for the effect of the health system, but those from urban clinics had a lower likelihood of screening receipt prior to adjusting for the effect of the health system. This reversal suggested that the health system strongly impacted whether patients received diabetic eye screening.
However, binary classification of areas as either rural or urban may be overly simplistic and obscure variable levels of health insurance, healthcare access, education, income, and health status that contribute significantly to patient outcomes. Previous reports have demonstrated the need for more nuanced classification of rurality informed by social determinants of and barriers to healthcare access.22 23 Thus, the Wisconsin Collaborative for Healthcare Quality (WCHQ), in collaboration with the University of Wisconsin Health Innovation Program, developed six groups or geolocations to better describe the healthcare access and socioeconomic characteristics of rural and urban zip codes across the state of Wisconsin: (1) urban advantaged, (2) urban, (3) urban underserved, (4) rural advantaged, (5) rural, and (6) rural underserved.22 24 Rurality and level of disadvantage in this study refer to these six WCHQ geolocation categories which were created in consideration of access to healthcare providers, insurance status, employment status, median household income, educational attainment, and health outcomes among people residing within these areas. These rural and urban underserved areas are associated with 6%–10% lower rates of breast, cervical, and colorectal cancer screening than urban advantaged areas, but the association between rurality and level of disadvantage and receipt of diabetic eye screening has not been previously evaluated using this categorization.22
In this study, we used the WHIO All-Payer Claims Database to determine the effect of a patient’s health system on their likelihood of screening receipt based on primary care clinic rurality and level of disadvantage. We also compared these results between Medicare and non-Medicare subgroups.
We conducted a retrospective analysis of de-identified data from the WHIO All-Payer Claims Database, which includes 75% of the Wisconsin population. The WHIO was established in 2005 as a state-wide collaboration among Wisconsin insurance companies, healthcare providers, large employers, and public agencies. The WHIO database contains medical, dental, and pharmaceutical claims information, along with de-identified patient demographics from commercial insurers, Medicare, and Medicaid in the state of Wisconsin. Insurance claims were included from commercial, Medicaid Health Maintenance Organization (HMO), Medicaid Fee-For-Service, and Medicare HMO payers. The WHIO also provides information regarding the clinic location at which care was provided, which is not available from most claims databases.
Inclusion and exclusion criteria
Individuals included in this analysis had 24 months of medical insurance coverage throughout the baseline and measurement periods. The baseline period was from October 1, 2012 to September 30, 2013. The measurement period was from October 1, 2013 to September 30, 2014. We included patients who were 18–75 years of age as this age group is recommended to receive diabetic eye screening in accordance with national guideline-based quality metrics (ie, National Committee on Quality Assurance (NCQA)/ Healthcare Effectiveness Data and Information Set (HEDIS)).3 Patients carried a diagnosis of diabetes according to the Hebert et al definition25 (ie, requiring at least two diagnoses of diabetes billed in an outpatient setting or one diagnosis of diabetes billed in an inpatient setting). The data set excluded those enrolled in hospice as they would not be eligible for diabetic eye screening.
We then excluded patients with a primary care provider located outside Wisconsin because the WHIO All-Payer Claims Database only includes detailed primary care clinic and provider information for health systems located in the state of Wisconsin (figure 1). We chose to focus on primary care physicians since 90% of patients with diabetes see their primary care physician at least once yearly26 and primary care physicians are typically those responsible for referring patients for preventive screening tests such as diabetic eye screening. We also excluded patients of primary care providers who were unassigned to a health system, patients whose primary care clinic was not assigned a WCHQ geolocation category, and those with a primary care provider who had fewer than 20 patients with diabetes in our data set as these providers may be less familiar with screening guidelines because diabetes management formed a very limited part of their clinical practice.
Patients were considered adherent with diabetic eye screening guidelines if they were screened once within the measurement year, following the American Academy of Ophthalmology (AAO) guidelines.27 Included patients were billed for a claim for an inperson eye exam with an eye care provider or teleophthalmology-based retinal imaging,28 since either fulfills diabetic eye screening guidelines and both methods are endorsed by the American Diabetes Association.2 29 Claims included an Evaluation and Management (E&M) code (992XX) or eye visit code (920XX) with an eye care provider (identified based on their Medicare taxonomy code) or billing performed for teleophthalmology imaging (Current Procedural Terminology (CPT) 92227, 92228, 92250).
We included the following patient-level predictor variables: age, sex, and hierarchical condition category (HCC) risk score. The HCC risk score provides a surrogate measure of medical comorbidities originally generated to predict healthcare expenditures in the subsequent year based on the previous year’s utilization.30 The average HCC risk score of a Medicare beneficiary is set as a reference score of 1.00. Insurance payer and health system were assigned based on the payer listed at month 12 of the baseline study period, and no patients were assigned to multiple payers or health systems. Patients dually eligible for Medicare and Medicaid were assigned to Medicare in the WHIO data set.
The analyses included clinic-level and health system-level variables as well. The clinic where the patient saw their primary care provider was categorized as urban underserved, urban, urban advantaged, rural underserved, rural, and rural advantaged based on zip code from the WCHQ method of categorization, which uses an established model that includes access to healthcare providers, insurance status, employment status, median household income, educational attainment, and health outcomes.22 24 The health system where a patient was seen was included within a categorical variable indicated by one of 101 included health system names. Each of the 100 largest health systems was assigned to a single value (ie, health system A, health system B, etc). The 43 smallest health systems were aggregated under a single value. A health system was defined as an associated group of physicians and/or clinics and was determined for each patient based on the health system affiliation of the patient’s primary care provider.
We generated multivariable logistic regression models to determine possible factors associated with receipt of diabetic eye screening based on health system, primary care clinic rurality and level of disadvantage, patient demographics, and comorbidities. To account for the fixed effect of individual health systems, we included the 143 health systems as one categorical variable with 101 possible values (ie, largest 100 health systems plus the 43 smallest health systems aggregated and analyzed as a single health system) in the multivariable regression model. Predicted probabilities (PPs) were generated using LSMEANS SAS statement in the logistic regression.31 We also conducted a subgroup analysis among Medicare and non-Medicare beneficiaries. All statistical analyses were performed using SAS V.9.4.
A total of 118 707 adults with diabetes from 698 primary care clinics in 143 Wisconsin health systems were included in our analysis (table 1 and online supplemental table 1; available at https://www.aaojournal.org) after excluding 37 702 patients with a primary care provider located outside Wisconsin (10.9%), those with a primary care provider with fewer than 20 patients with diabetes in our data set (8.1%), those with a primary care provider unassigned to a health system (4.7%), and patients whose primary care clinic was not assigned a WCHQ geolocation category (0.4%). Patients excluded from our analysis (n=20 689) were slightly more likely to be female, younger, less likely to be insured by Medicare, and less likely to receive diabetic eye screening (online supplemental table 2; available at https://www.aaojournal.org).
Most patients included in our analysis were male (51.4%) and 74.3% were over 55 years of age (mean: 60.9±11.3 years). The most common insurance payers were Medicare (58.3%), commercial (30.9%), and Medicaid (10.1%). Majority of the patients (83.8%) obtained care at an urban primary care clinic (urban advantaged, 29.4%; urban, 43.6%; or urban underserved, 10.8%). The mean HCC risk score was 1.09. In comparison with the non-Medicare population, those insured by Medicare were older, more likely to be female, more likely to obtain care at a rural primary care clinic, and had a higher HCC risk score (p<0.001 for all comparisons).
The overall proportion of patients who received diabetic eye screening was 53.5%. The predicted probability of screening receipt was lowest among patients obtaining care from urban underserved primary care clinics (PP: 43.6%, OR 0.60, 95% CI 0.58 to 0.63) and highest among those obtaining care from urban advantaged primary care clinics (PP: 55.8%, reference group) before adjusting for health system in the model (table 2).
Patients from urban underserved clinics were less likely to receive screening than those from rural underserved clinics (PP: 50.1%, OR 0.79, 95% CI 0.71 to 0.88) before adjusting for health system. When health system fixed effects were added to the model, however, the directionality of the relationship between clinic rurality and screening receipt reversed. In these fully adjusted models, patients from urban underserved clinics were more likely to receive screening than those from rural underserved clinics (PP: 48.9%, OR 0.89, 95% CI 0.84 to 0.94 vs PP: 43.9%, OR 0.72, 95% CI 0.62 to 0.82, respectively).
Similar findings were observed for both Medicare and non-Medicare subgroups. Before adjusting for health system in the model, patients who obtained care from rural underserved primary care clinics were more likely to receive screening than those from urban underserved clinics (Medicare: PP: 51.1%, OR 0.76, 95% CI 0.67 to 0.87 vs PP: 45.9%, OR 0.62, 95% CI 0.58 to 0.65; non-Medicare: PP: 48.5%, OR 0.85, 95% CI 0.73 to 1.01 vs PP: 39.9%, OR 0.60, 95% CI 0.56 to 0.64, respectively) (table 3).
However, the directionality of this association reversed when health system was included (Medicare: 45.8%, OR 0.67, 95% CI 0.56 to 0.81 vs 50.3%, OR 0.83, 95% CI 0.77 to 0.89; non-Medicare: 43.3%, OR 0.78, 95% CI 0.63 to 0.97 vs 50.3%, OR 1.05, 95% CI 0.96 to 1.16, respectively). Additionally, non-Medicare beneficiaries experienced a greater reversal in the directionality of the relationship between screening receipt among rural and urban underserved clinics when the health system fixed effects were included in the model.
Finally, patient-level factors associated with receipt of diabetic eye screening were similar among patients insured by either Medicare and non-Medicare payers. These factors included older age, female sex, and greater HCC risk score (online supplemental table 3; available at https://www.aaojournal.org).
Primary care clinic rurality and level of disadvantage were associated with the likelihood of receipt of diabetic eye screening. Patients who obtained care at urban underserved primary care clinics were less likely to receive diabetic eye screening than those obtaining care at rural underserved primary care clinics, but the directionality of this relationship reversed when the model was adjusted for the effect of health system. Patients in both the Medicare and non-Medicare subgroups exhibited similar findings, although a more marked reversal was evident among non-Medicare patients from urban underserved clinics after adjusting for the effect of health system. Furthermore, patient-level factors associated with receipt of diabetic eye screening were comparable among Medicare and non-Medicare subgroups, and across WCHQ geolocation categories of the clinics, indicating our results are robust across health insurance types.
Our results suggest that there are factors related to a health system’s characteristics that substantially influence a patient’s likelihood of screening receipt. After removing the effect of the health system, there was a major change in the likelihood of screening receipt among rural versus urban underserved populations. Patients obtaining care from urban underserved primary care clinics would be predicted to receive more screening than their rural counterparts in the absence of health system-related factors.
While most studies have focused on patient-level interventions to increase receipt of diabetic eye screening,12–15 our results emphasize the importance of the health system and health system-level interventions, which have been under-represented in studies aimed at increasing screening. A recent Cochrane review of interventions to increase diabetic retinopathy screening found that health system-level interventions such as restructuring teams or workflows to support screening receipt, use of electronic patient registries and recall systems, and use of telemedicine-based screening effectively increased screening rates.12 A systematic review by Zhang et al32 identified five randomized controlled trials in which health system-level interventions, including telemedicine, significantly increased diabetic retinopathy screening, with relative risks ranging from 1.12 (95% CI 1.03 to 1.22) to 5.56 (95% CI 2.19 to 14.10). In our prior work, we found that tailored implementation of telemedicine-based screening using health system-level interventions, such as primary care workflow changes and patient recall systems, led to significant and sustained increases in screening rates in a rural health system.18 Targeted use of health system-level interventions in diabetes populations with lower screening rates, including rural and urban underserved populations, Black and Latinx communities, as well as younger populations,8 10 33 could be highly effective in increasing screening rates.
Our finding that the likelihood of receipt of diabetic eye screening was significantly different between patients obtaining care at rural and urban underserved clinics was likely due to differences in barriers to screening experienced by these communities. Both underserved populations (based on WCHQ definitions) have poorer health, are uninsured at higher rates, are more likely to be insured by Medicaid, experience more poverty, and have less educational attainment.22 However, rural underserved communities have less access to healthcare as a result of long travel distances and lower geographical density of healthcare providers, while those in urban underserved communities may have physical proximity to a greater number of healthcare providers but experience higher rates of unemployment, leading to reduced access to healthcare from lack of health insurance or underinsurance.22 34 35
Additionally, we observed that urban underserved clinics have a higher likelihood of screening receipt than rural underserved clinics after adjusting for health system, which suggests that patients from urban underserved populations may face greater barriers to eye screening related to social determinants of health, including structural racism,36 compared with rural underserved populations, which in Wisconsin are predominantly non-Hispanic and white. Urban underserved clinics in particular may have a substantial opportunity to increase screening using health system-level interventions. However, additional factors outside the health system may be limiting access for rural patients,35 and health system interventions alone may not be sufficient to address barriers to access in rural communities. Notably, our findings in this study differed from those which we previously reported in which patients obtaining care from rural, compared with urban, primary care clinics were less likely to receive screening when adjusting for the effect of the health system.11 37 This was likely due to the fact that the ‘urban’ and ‘rural’ categories in our prior study grouped together advantaged and underserved populations, which failed to capture the nuance of the substantial challenges faced by urban underserved communities.
Taken together, the results of our study demonstrate the significant gap that remains to be addressed with regard to increasing diabetic eye screening across the USA overall and in high-risk populations. Patient-level factors associated with increased likelihood of receipt of diabetic eye screening, including older age and female sex, were similar among patients insured by either Medicare or non-Medicare payers and were consistent with prior literature.9 33 38 Lower rates of screening have also been identified among younger patients who are Black or Latinx and those with lower household net worth.8 10 While our database did not include information regarding race, ethnicity or household income, these factors have been significantly associated with the likelihood of screening after adjusting for age and sex in some studies,39 40 but not in others.9 38 Inclusion of these variables would have enriched the data set and may have provided further evidence in support of the impact of social determinants of health on the likelihood of screening receipt.
While our study had many strengths, some limitations of this work included the study period of 12 months, which made our measurement of receipt of diabetic eye screening (based on the AAO guidelines) more stringent compared with other studies that considered screening within 15 months to be adherent.8 41 By using a 1-year measurement period to assess screening receipt, we may have systematically underestimated screening receipt. However, this would be applied equally to all health systems and would not affect estimated differences across health systems. In addition, current health system quality performance incentive guidelines (ie, NCQA/HEDIS)3 require that patients receive screening either within 1 year if they have a history of mild retinopathy or within 2 years for those with no history of retinopathy. Thus, we used a 1-year measurement period given that we did not have data on the presence or severity of diabetic retinopathy. As with other claims-based studies, we did not have records of each eye exam and this could also have systematically slightly overestimated screening receipt due to a small number of patients who may have deferred pupil dilation at their eye clinic visit and therefore did not obtain diabetic eye screening.8 10 Excluding patients from primary care providers with fewer than 20 patients with diabetes may have systematically excluded patients from smaller clinics and health systems. However, the number of patients meeting this criterion comprised a relatively small proportion of our sample (8.1%, n=12 710). Thus, it is unlikely that excluding these patients significantly affected our results.
Additionally, a single insurer was assigned to each patient in this claims database even though patients could have been covered by a non-Medicare insurer in the baseline period and Medicare in the study period. In our data set, only 2.4% of patients were 64 years old when the study period began, so this scenario would apply to a very small fraction of our study population and would be unlikely to have significantly affected our results. Further, given the limitations of our claims-based data set, we were unable to stratify our results by Medicaid eligibility and assess for differences based on this socioeconomic marker, nor were we able to generalize our results to populations not seeking care from health systems. Finally, the difference observed between the likelihood of screening in urban and rural underserved populations after adjusting for health system may be related, in part, to residual confounding, which could include unmeasured effects of social determinants of health.
In this state-wide, all-payer claims database, we observed significant associations between primary care clinic rurality and level of disadvantage and receipt of diabetic eye screening. Health systems had a substantial effect on the observed associations between diabetic eye screening and both rural and urban underserved populations. While most studies have focused on patient-level interventions to increase receipt of diabetic eye screening, our results emphasize the importance of health system-level interventions, such as restructuring teams or workflows to support screening receipt, use of electronic patient registries and recall systems, and use of telemedicine-based screening. Health systems, particularly those providing care to urban underserved populations, have an opportunity to increase screening rates by leveraging health system-level interventions to support patients in overcoming barriers related to social determinants of health.
Data availability statement
Data are available upon reasonable request. Data are available upon reasonable request to those who obtain regulatory approval for data access from the University of Wisconsin Health Sciences Institutional Review Board.
Patient consent for publication
The University of Wisconsin Health Sciences Institutional Review Board (IRB) staff reviewed this study and determined that the study involved secondary analysis of de-identified data from an existing data set and therefore did not constitute human subjects research. All research activities were conducted in accordance with the Declaration of Helsinki and all federal and state laws.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Contributors All authors made substantial contributions to the conception or design of the work; acquisition, analysis, or interpretation of data for the work; drafting the work or revising it critically for important intellectual content; 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. The principal investigator (YL) is the guarantor who accepts full responsibility for the finished work and/or the conduct of the study, had access to the data, and controlled the decision to publish.
Funding This study was supported by NIH/NEI (K23 EY026518; PI: YL), NIH/NEI (K23 EY030911-02; PI: RC), NIH NIDDK (1R01 DK132569-01; PI: MBB), Agency for Healthcare Research and Quality (K08 HS026279; PI: MBB), and in part by an unrestricted grant from Research to Prevent Blindness to the University of Wisconsin Department of Ophthalmology and Visual Sciences.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.