Diabetes Microvascular Disease Diagnosis and Treatment After High-Deductible Health Plan Enrollment

Diabetic microvascular disease is the leading cause of blindness and renal failure in the U.S. (1–5). Microvascular diseases, such as retinopathy and nephropathy, occur when long-standing hyperglycemia damages small blood vessels, leading to progressive loss of organ function (6–9). Effective measures exist to detect and slow the progression of retinopathy and nephropathy (10–12), including regular screening, medication intensification, and disease monitoring (12,13). Timely care for microvascular disease is crucial; treatment delays of even several days for complications such as retinal detachment can profoundly worsen vision (14,15).

People with diabetes thus require regular secondary preventive testing and subsequent downstream services, but access to care might be impeded by cost. More than half of commercially insured people in the U.S. now have high-deductible health plans (HDHPs) that can lead to annual out-of-pocket spending of ∼$1,000–$7,000 per person for most care not classified as primary prevention. In 2020, 57% of workers had deductibles of ≥$1,000 and 26% had deductibles of ≥$2,000 for single coverage (16). The Affordable Care Act requires no out-of-pocket costs for common primary preventive services, but importantly, does not exempt most evidence-based secondary prevention from out-of-pocket costs. Chronically ill patients in HDHPs therefore often face high cost sharing for expensive services such as diabetic retinopathy screening, specialist visits, and procedures to prevent or limit vision loss.

Research has found that patients with diabetes respond to HDHPs by delaying or reducing important care (17–19). These actions have been associated with short-term morbidity among low-income members (17,19) but no apparent increases in longer-term adverse macrovascular outcomes, such as myocardial infarction or stroke, in an overall population with diabetes (20). Studies have also demonstrated that exempting preventive services from cost sharing under HDHPs preserves use in the short-term (19,21–25).

However, effects of HDHPs on longer-term microvascular disease care and outcomes remain unknown. We studied a unique natural experiment in which patients with diabetes in HDHPs had low out-of-pocket barriers to nephropathy care but higher out-of-pocket barriers to retinopathy care. We hypothesized that retinopathy screening and diagnosis would be delayed while nephropathy measures would remain unchanged among HDHP members with diabetes.

We drew our study population from ∼48 million commercially insured members in a large national commercial (and Medicare Advantage) health insurance claims data set who were enrolled between 1 January 2003 and 31 December 2014. The data set contains enrollment information and all medical, pharmacy, and hospitalization claims. We included only members with employer-sponsored insurance because people in the nongroup market can self-select into insurance types and Medicare Advantage beneficiaries do not have HDHPs.

We considered an insurance plan to have a low annual deductible if ≤$500 and a high deductible if ≥$1,000. At smaller employers, we determined the deductible amount from a benefits table obtained from the health insurer. This table mostly included employers with <100 individuals but also had a modest number of larger employers. For employers not represented in the insurance benefits table (mostly large employers), we imputed employer-level deductible amounts from actual out-of-pocket spending by individuals who used health services, applying an algorithm with a sensitivity and specificity of >95% (Supplementary Table 1) (20). To minimize bias that would occur if subjects were allowed to choose their deductible level, we included only individuals whose employers mandated low-deductible then HDHPs (intervention group) or that offered only low-deductible plans year-on-year (control group).

We defined the index date for employers that switched to HDHPs as the beginning of the month when the switch occurred. For employers that did not switch plans, the index date was the beginning of the month when their yearly account renewed. Some members had multiple eligible index dates (e.g., multiple low-to-low-deductible years or both low-to-low- and low-to-high-deductible years). In the cases of members with both low-to-low- and low-to-high-deductible years, we randomly assigned enrollees to the HDHP pool or the control pool. For members assigned to the control pool that had multiple low-to-low-deductible spans, we randomly selected one of their potential index dates (and their corresponding before-after enrollment years). Employers had index dates between 1 January 2004 and 1 December 2014.

For all individuals in the study, the beginning of study time (time zero) was 12 months before their employers’ index date, and we defined this 12-month period as the baseline year (Supplementary Fig. 1). The employer’s index date was the beginning of the follow-up period. For each individual, we measured months from time zero to first detected outcomes in the baseline period, and we measured months from the index date to first detected outcomes in the follow-up period.

Individuals were eligible for the study based on the following criteria: enrolled through an employer that had coverage for at least 1 year before and after the index date, aged 40–64 years, met criteria for diabetes before the index date (based on version 11.1 of the Johns Hopkins ACG System) (26,27), and were continuously enrolled for at least 1 year before and at least 1 month after the index date (Supplementary Fig. 1). These criteria yielded 53,247 individuals whose employers switched to HDHPs and 440,359 whose employers kept low-deductible plans (Table 1).

This study was approved by Harvard Pilgrim Health Care Institute’s Institutional Review Board with a waiver of informed consent and supported by grant R01DK100304 from the National Institute of Diabetes and Digestive and Kidney Diseases.

We compared matched cohorts in this observational, before-after study. All members were enrolled in a commercial health insurance plan between 1 January 2003 and 31 December 2014. The intervention group consisted of individuals in low-deductible insurance plans for 1 year who were switched to HDHPs and were then enrolled for an additional 1 month to 4 years (Supplementary Fig. 1). The control group included contemporaneously enrol led matched individuals who remained in low-deductible plans throughout the study period.

We matched based on the calendar year of the index date; the propensity of the employer to mandate high-deductible insurance and the propensity of individuals to work for such employers (each divided into tertiles) (Supplementary Material) (28,29); follow-up duration (months of follow-up, capped at 4 years); and the following member-level baseline variables: out-of-pocket spending category ($0 to $500, $501 to $999, $1,000 to $2,499, or ≥$2,500), total health care cost tertile, mild-to-moderate nonproliferative diabetic retinopathy diagnosis (Supplementary Table 2), mild-to-moderate chronic kidney disease diagnosis, severe nonproliferative diabetic or proliferative retinopathy diagnosis, severe chronic kidney disease diagnosis, encounter for loss of vision (comprising a diagnosis or a treatment to prevent further vision loss), and definitive procedure to treat end-stage renal disease. The employer and individual propensity score models included multiple characteristics that were potentially associated both with study group assignment and outcome measures (Supplementary Material). For example, to reduce bias that could occur if study group members differentially switched providers at the index date, we included baseline and follow-up provider network type (Supplementary Material) as an employer propensity variable.

We used coarsened exact matching (Supplementary Material) (30–32), an approach that is similar to exact matching but differs by using categories instead of exact values to match. The software for coarsened exact matching creates weights for each stratum that adjust for any differences between study groups in the proportion of individuals in the stratum.

We censored individuals if they dropped from the sample (e.g., as a result of disenrollment), reached age 65 years (when Medicare coverage begins), or reached the end of the baseline year or follow-up period (4 years after the index date) for baseline and follow-up analyses, respectively.

Supplementary Table 2 includes billing codes we used to flag all measures. Primary measures included retinopathy and nephropathy screening; diagnosis of mild-to-moderate nonproliferative diabetic retinopathy, mild-to-moderate chronic kidney disease, severe nonproliferative or proliferative diabetic retinopathy, and severe chronic kidney disease; encounter for vision loss diagnosis or treatment; and encounter for loss of renal function treatment. To capture the latter measure, we detected codes indicating dialysis initiation or renal transplant. To detect the first detected evidence of vision loss, we used both diagnosis codes for vision loss and codes for procedures used to treat vision loss.

We measured time to the first detected instance of diabetic retinopathy and nephropathy screening per member (if any) in both the baseline and follow-up periods. For all other measures, because they can happen only once per patient, we captured only a single first event per member after time zero. We then measured the interval between time zero and any first detected event during the baseline period and the index date and any first detected event during the follow-up period. To determine cost sharing levels for nephropathy and retinopathy screening, we calculated the out-of-pocket spending amount per retinopathy and nephropathy screening claim. Because clinicians might not always use specific diagnoses when coding diabetic retinopathy and nephropathy, we also created measures of any ophthalmologic or renal disease diagnosis in diabetes, comprising the first detected instance of either nonspecific or specific codes (Supplementary Table 2).

We calculated total out-of-pocket expenditures by summing deductible, copayment, and coinsurance values for all claims per member per month in order to visualize overall cost sharing changes.

To estimate the difference between groups in the time to each outcome (18) (Supplementary Fig. 2), we 1) calculated the interval during follow-up between the index date and the date when the control group reached half its event rate at the end of follow-up, 2) calculated the interval during follow-up between the index date and the date when the intervention group reached half the event rate that the control group achieved at the end of follow-up, and 3) calculated the difference between these intervals (Statistical Analysis, below). This difference provides an intuitive measure of any delays that an average patient with diabetes in our sample might experience after a mandated switch to HDHPs. We used the same approach to assess potential delays during the baseline year.

We applied the Johns Hopkins ACG System (26,27) to calculate members’ baseline period morbidity score (Supplementary Material). We used 2008–2012 American Community Survey data (33) to characterize census tracts (i.e., neighborhoods averaging ∼4,000 people) (34) in which members resided. Categories included neighborhoods with below-poverty levels of <5%, 5–9.9%, 10–19.9%, and ≥20% and below-high school education levels of <15%, 15–24.9%, 25–39.9%, and ≥40% (35). We used geocoding to classify participants as from predominantly White, Black, Hispanic, or mixed neighborhoods, and we used a superseding Hispanic or Asian categorization based on the E-Tech system (Ethnic Technologies) that analyzes full names and geographic locations of individuals (36). Other covariates included age category (Supplementary Material); sex; U.S. region (West, Midwest, South, Northeast); employer size used as a continuous variable or with categories of 0–99, 100–999, or ≥1,000 individuals (Supplementary Material); and calendar month of the index date.

We compared baseline characteristics of our study groups using standardized differences (37). We analyzed time to primary measures during the baseline year and the follow-up period in separate models using parametric survival time regression with a Weibull distribution (38). The term of interest was HDHP group status, and we adjusted for employer size category (given residual imbalance after the match) (Table 1). The coefficient for the HDHP term was an adjusted hazard ratio indicating the independent association of HDHP membership with the outcome of interest. We used these hazard ratios and marginal effects methods (39) to estimate differences in timing for the intervention and control group to reach half the control group’s rate at the end of the baseline year or follow-up period (Supplementary Fig. 2) (18).

We used generalized estimating equations regression with a zero-inflated negative binomial distribution to estimate changes in total out-of-pocket spending per member-year and retinopathy and nephropathy screening expenditures per event. The term of interest was an interaction between HDHP group status and study year, and we again adjusted for employer size.

Our matched diabetes sample included an HDHP group of 50,790 individuals and 335,178 individuals who remained in low-deductible health plans throughout the study period (Table 1). The mean age of intervention and control group members was 51–52 years (SD, 8.9–9.2), 44% were women, 57% lived in neighborhoods with below-poverty levels of ≥10%, 13% lived in neighborhoods in which ≥25% of the individuals had below-high school educational levels, and 11–12% were Hispanic. For both groups, the percentage with a follow-up time of 1 year was 75%; 2 years, 41%; 3 years, 25%; and 4 years, 15%. After matching and applying match-generated weights, all standardized differences between the intervention and the control group at baseline were <0.2 (Table 1), indicating minimal differences (37).

Individuals with HDHPs experienced total (medical plus pharmacy) out-of-pocket expenditure increases ranging from 23% (95% CI 21%, 26%) to 31% (26%, 36%) per follow-up year versus baseline relative to controls (Fig. 1 and Supplementary Table 3). HDHP members had low cost sharing for nephropathy screening (mean of $2.2 at follow-up) (Supplementary Table 4) but higher out-of-pocket costs for retinopathy screening (mean of $43.3 at follow-up). Control group members had corresponding mean out-of-pocket costs of $1.3 and $29.9 at follow-up.

At follow-up, HDHP enrollment was associated with relative delays in retinopathy screening (0.7 months [95% CI 0.4, 1.0]) (Table 2 and Fig. 2), severe nonproliferative or proliferative diabetic retinopathy diagnoses (2.9 months [0.5, 5.3]), and vision loss diagnosis/treatment (3.8 months [1.2, 6.3]). The follow-up delay for mild-to-moderate nonproliferative diabetic retinopathy diagnoses did not reach statistical significance (1.6 months, [−0.2, 3.4]). HDHP enrollment was not associated with statistically significant relative delays in chronic kidney disease-related measures. HDHP enrollees had slightly accelerated times to retinopathy and nephropathy screening at baseline versus controls (0.1 months), but we did not detect other baseline differences in the timing of microvascular disease diagnosis and treatment. Supplementary Table 5 lists adjusted baseline and follow-up hazard ratios from the parametric survival time regression models.

Our sensitivity analyses that included time to any ophthalmologic and renal diagnoses detected follow-up adjusted hazard ratios of 0.91 (95% CI 0.88, 0.95) (Supplementary Table 5) and 0.98 (0.93, 1.02) respectively, in the HDHP group relative to controls.

HDHP members with diabetes experienced delayed retinopathy care and vision loss diagnosis/treatment of up to 3.8 months compared with similar members who remained in low-deductible plans. In contrast, we did not detect statistically significant delays in chronic kidney disease care and outcomes. Although our data were unable to capture visual acuity outcomes, our findings raise concerns about visual health among high-deductible plan members given that even brief treatment delays for complications such as retinal detachment can profoundly worsen vision (14,40,41).

The delays we detected appear to be related to out-of-pocket costs for microvascular disease screening and clinician visits. Intervention group members faced relatively high retinopathy screening out-of-pocket costs and the need for expensive eye specialist care for retinopathy-related services. In contrast, nephropathy screening was inexpensive for HDHP members, and early-stage chronic kidney disease can largely be managed by primary care clinicians. Of note, we detected a nonsignificant 1.5- to 1.7-month relative delay in chronic kidney disease diagnosis and treatment among HDHP members. We therefore cannot rule out delayed chronic kidney disease care related to HDHP enrollment, and studies with larger sample sizes should examine these measures.

Results support policies toward health insurance benefit designs that reduce out-of-pocket costs for preventive care under HDHPs but also call attention to discrepancies in Affordable Care Act cost sharing exemptions. The existing exemptions apply largely to primary preventive services (42) and thus fail to protect chronically ill patients from high secondary prevention out-of-pocket costs. Our findings suggest that removing cost sharing for retinopathy screening could have substantial benefits among HDHP members with diabetes. In 2019, the Internal Revenue Service issued Notice 2019-45 that allows employers and health insurers to exempt retinopathy screening from deductibles for plans that have associated health savings accounts (43). However, uptake is voluntary and could well be low.

Our analyses were observational and therefore at risk for bias from unmeasured confounders. Nevertheless, our study included several features to minimize bias, including restricting the study to employers that do not allow enrollee choice of deductible level and matching to balance key employer- and individual-level characteristics. Our data source does not allow insights into whether the delays in retinopathy care we detected caused suboptimal visual outcomes or quality of life. Capturing these measures would require electronic helth record or survey data that were unavailable to us. Future studies should integrate such data sets and examine visual acuity outcomes among HDHP members with diabetes. However, treatment delays of even several days for complications such as retinal detachment can profoundly worsen vision (14,15). Given our inability to perform long looks back in claims data, our measures of mild-to-moderate and severe microvascular disease as well as vision loss might include nonincident events. Limited look back also implies that study groups could differ with respect to diabetes history and duration. However, we matched on factors that should balance these prebaseline characteristics such as age, sex, morbidity level, date of first appearance in our data set, date of first recognized diabetes diagnosis, and timing of baseline microvascular disease–related services. In addition, all non-screening measures captured at follow-up, our period of interest, were preceded by at least 12 washout months. We imputed deductible levels from claims for almost all large employers, but we do not believe that the imputation biased results because of its high sensitivity and specificity (Supplementary Table 1). Although we did not have exact health insurance benefit details describing cost sharing levels for various services, we confirmed our presumptions that nephropathy screening would have low cost sharing and that retinopathy screening would have higher cost sharing by examining out-of-pocket costs on the relevant claims that we had flagged as screening events.

Our study findings are not generalizable to individuals with uncommonly high deductibles, people with type 1 diabetes, newly insured patients, or vulnerable subgroups with diabetes. For example, only 3% of members in our study resided in predominantly Black neighborhoods. Results therefore do not generalize to this population, and future research should investigate impacts of HDHPs on microvascular complications among Black individuals with diabetes. In addition, although our study included a large national sample of HDHP members with diabetes, future studies should replicate our analyses among other national commercial health insurers.

Diabetes patients in HDHPs experienced delayed diagnosis of retinopathy and vision loss diagnosis/treatment compared with counterparts in low-deductible plans. Such changes were not detectable in chronic kidney disease–related care that generally has lower out-of-pocket costs for HDHP members. Our study raises concerns about unintended consequences of high cost sharing for chronically ill populations. Policymakers should consider extending Affordable Care Act cost sharing exemptions to evidence-based secondary preventive services such as retinopathy screening. Health insurers and employers should discourage HDHP uptake among patients with diabetes given the potential for worse vision outcomes. Results also support the effectiveness of current value-based insurance features (44–46) such as low out-of-pocket obligations for nephropathy screening. Finally, future research should assess whether patients with diabetes in HDHPs ultimately require more costly workups and more advanced treatments for retinopathy.

Acknowledgments. The authors thank Robert LeCates, MA, Beverly Adade, MBE, and Katherine Callaway, MPH, of Harvard Pilgrim Health Care Institute, for valuable assistance with literature searches, computer programming, data processing, and algorithm development.

Funding. This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases grant R01DK100304 (principal investigator: J.F.W.) and National Institute of Diabetes and Digestive and Kidney Diseases Health Delivery Systems Center for Diabetes Translational Research grant 1P30-DK092924.

The funder had no role in the design and conduct of the study, collection, management, analysis, and interpretation of the data, preparation, review, or approval of the manuscript, and decision to submit the manuscript for publication.

Duality of Interest. J.F.W., F.Z., C.Y.L, D.R.-D., J.W., and S.A. were employed in the Harvard Medical School and Harvard Pilgrim Healthcare Institute Department of Population Medicine. The Harvard Pilgrim Healthcare Institute is affiliated with Harvard Pilgrim Health Care, a not-for-profit health insurer that had no role in the study. J.P.N. reports personal fees from Aetna outside the submitted work. T.P.S. is a retina surgeon with Tallman Eye Associates (Lawrence, MA) and reports consulting fees from Pykus Therapeutics (Cambridge MA), Aldeyra Therapeutics (Lexington, MA), Brixton Biosciences (Cambridge, MA), Hubble Therapeutics (Boston, MA), and Alcon (Fort Worth, TX) unrelated to the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.F.W. contributed to conceptualization, data curation, funding acquisition, investigation, methodology, supervision, visualization, writing the original draft, and reviewing and editing manuscript drafts. J.W. and S.A. contributed to data curation, investigation, project administration, supervision, and reviewing and editing manuscript drafts. F.Z. contributed to data curation, formal analysis, investigation, methodology, visualization, and reviewing and editing manuscript drafts. C.Y.L. contributed to investigation, methodology, and reviewing and editing manuscript drafts. T.P.S. contributed to conceptualization, investigation, methodology, and reviewing and editing manuscript drafts. D.R.-D. contributed to conceptualization, data curation, investigation, methodology, supervision, and reviewing and editing manuscript drafts. J.P.N. contributed to conceptualization, funding acquisition, investigation, methodology, supervision, and reviewing and editing manuscript drafts. J.F.W. and F.Z. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 80th Scientific Sessions of the American Diabetes Association, virtual meeting, 12–16 June 2020.