Projected Impact of the Medicare Part D Senior Savings Model on Diabetes-Related Health and Economic Outcomes Among Insulin Users Covered by Medicare

Insulin is a lifesaving medication for all people with type 1 diabetes and many with type 2 diabetes. One in every three Medicare beneficiaries has diabetes, and more than 3.3 million Medicare beneficiaries use one or more of the common forms of insulin (1). From 2003 to 2013, the annual list price of insulin increased threefold and has continued to rise (2). The annual cost of insulin per person to the health system was estimated to be $4,500 in 2017 in the U.S. (3).

The rapid increase in insulin price has led to a substantial increase in patient out-of-pocket (OOP) costs (4,5). Higher OOP costs are associated with lower insulin adherence and dose rationing (6,7). According to a recent survey, 25% of insulin users had cost-related insulin underuse (8), leading to worse glycemic control, more diabetes-related complications, and higher medical costs (9). Moreover, an increase in OOP costs may increase health disparities, as changes in OOP costs disproportionally impact socioeconomically disadvantaged populations (8).

To lower the OOP cost and improve access to insulin, the Centers for Medicare & Medicaid Services developed the Medicare Part D Senior Savings Model (SSM), and the new policy took effect on 1 January 2021 (10). The SSM was designed to keep the OOP costs for insulin at $35 or less per month in the deductible, initial coverage, and coverage gap (i.e., “donut hole”) phases. In capping the OOP cost, the SSM is expected to increase insulin use and improve patient adherence among those who currently have trouble affording insulin. It is expected that increased medication adherence would improve glucose control and lead to better long-term health outcomes (10). At the same time, the SSM may increase total insulin expenditures because of increased insulin adherence and increased use overall. However, it may also lower the cost of treating diabetes-related complications for Medicare due to improved health outcomes.

The number of Medicare beneficiaries who would benefit from the SSM and the long-term health and cost implications have not been assessed. Yet such information is critical for an understanding of whether the policy could achieve the intended effect as designed and further refinement of the policy. The results of this study fill this evidence gap with estimations of the number of Medicare beneficiaries potentially affected by the SSM and quantification of the potential health and economic impact with use of a simulation experiment.

We used data from the 2018 Medical Expenditure Panel Survey (MEPS) to build the study sample of Medicare beneficiaries with diabetes who were on insulin treatment and paid >$35 per month. This sample included Medicare beneficiaries with self-reported diabetes and documentation of insulin use.

Due to the hierarchical benefit design of Medicare Part D, significant variations in insulin OOP costs were observed among the beneficiaries. Such variations would, in turn, lead to heterogeneous responses to the OOP cost reduction from SSM. According to a recent study published by Trish et al. (11), for Medicare beneficiaries who end the year in the catastrophic phase, additional purchases of insulin would not have a significant impact on their OOP cost burden, making their insulin usage insensitive to the insulin OOP cost change. On the other hand, for Medicare beneficiaries who end the year in the coverage gap, additional use of insulin can potentially result in a significant increase in their OOP cost burden. Thus, following an approach similar to that of Trish et al. (11), we grouped the study sample into three subgroups with potentially different responses to OOP cost change: 1) individuals who ended the year in the deductible/initial phase, 2) individuals who end the year in the coverage gap, and 3) individuals who ended the year in the catastrophic phase. The 2018 Medicare Part D payment policy (12) was applied for determination of which payment phase (i.e., initial/coverage gap/catastrophic) the last insulin prescription landed in (details provided in Supplementary Appendix 1). We used the 2018 MEPS survey weights and Medicare beneficiaries report (2018) (13) to populate the study sample to a nationally representative simulation sample, with the number of individuals in the simulation sample matching the actual number of individuals in the U.S.

Potential SSM enrollees were assumed to be individuals with an average monthly payment of insulin in the deductible, initial, and coverage gap phase >$35 who could realize lower OOP costs by choosing a participating enhanced Part D plan. This logic follows the classic rational choice theory (14), in which individuals make choices that result in the optimal level of benefit (i.e., lowering OOP cost). Note that patients covered by low-income subsidy programs were not considered potential SSM enrollees in our study. Although we could not identify them directly from the MEPS data, those covered by the low-income subsidy had insulin OOP cost <$35 per month. Thus, they were automatically excluded from the SSM group. Demographic characteristics (age, sex, race), diabetes duration, annual total and OOP cost on insulin, insulin adherence, and insurance enrollment were compared between potential SSM enrollees and non-SSM enrollees in each of three subgroups, respectively, with t tests and the χ² test.

We used the Building, Relating, Assessing, and Validating Outcomes (BRAVO) diabetes model (15) to assess the long-term health and economic outcomes resulting from the OOP cost reduction for Medicare insulin users. The BRAVO diabetes model is a person-level, discrete time, microsimulation model. It included patient characteristics—demographics, biomarkers, disease history, and treatment—to predict the lifetime clinical and economic outcomes. The model has been extensively validated and calibrated against 18 international trials (16,17) and calibrated to a U.S. representative cohort (18). It has been used for economic evaluation (19,20) and program assessment (21).

The SSM was assumed to improve blood glucose control by increasing insulin adherence, leading to risk reductions for both macrovascular and microvascular complications. Thus, our simulation experiment includes five macrovascular complications (myocardial infarction [MI], congestive heart failure [CHF], stroke, angina, revascularization), three microvascular complications (end-stage renal disease [ESRD], blindness, and severe pressure sensation loss), cardiovascular disease death, all-cause mortality, life expectancy, and quality-adjusted life-years (QALY). The simulated economic outcomes include total medical costs and OOP costs. Medical costs include care for diabetes-related complications (or savings associated with averted cases) and diabetes management, including insulin and other medications.

Two scenarios were simulated to measure the impact of SSM: one without SSM and the other with SSM. We used information collected from MEPS 2018 to simulate the scenario without SSM (details provided in Supplementary Appendix 1). For the scenario with SSM, we replaced the observed OOP insulin cost in 2018 with a monthly OOP cost rate of $35 for each individual.

The change of A1C under SSM will be the main factor that drives the change in health outcomes in the simulation. We estimated the A1C reduction as a result of the cap of OOP cost in three steps: 1) assess how much OOP cost was reduced as a result of SSM, 2) assess how many additional days of insulin use as a result of the lowered OOP cost, 3) assess how much A1C would be reduced as a result of the increase in insulin use. We built a series of equations measuring how the number of days covered by insulin would change in response to OOP cost change (i.e., price elasticity for demand) for different population subgroups (details provided in Supplementary Appendix 2). We used these equations to convert OOP cost change into the change of the number of days of insulin use and then used the equation of Schectman et al. (22) to convert the changes of the proportion of days covered (PDC) of insulin use to changes in A1C.

OOP cost was defined as the amount the patient paid out of pocket, which included deductible, co-pay, and coinsurance. The total medical cost was defined as the total amount paid for the corresponding care, including insurance payment, OOP cost, and other sources. The OOP and total medical costs were estimated for insulins only and overall (i.e., the sum of Medicare Parts A, B, and D). OOP costs and total costs for insulin were estimated with data from MEPS 2018. Costs of different complications were previously reported (23). QALY decrements associated with different complications were based on the Health Utility Index diabetes complication equation (19). Insulin adherence was measured with the PDC, which is calculated based on the total days of supply the person purchased in 2018 over 365 days. For individuals on both basal and bolus insulin, we estimated the PDC separately and took an average. The economic evaluation was conducted from a health system perspective, with use of 3% discount rates for cost and QALY. All costs were inflated to the 2020 values based on the consumer price index for medical care services (24). A 5-year initial policy horizon and an extended 20-year horizon were evaluated for short- and long-term effects of SSM.

Given the simulation is at the individual level, instead of performing a conventional one-way sensitivity analysis without altering parameters, we conducted an extensive subgroup analysis to evaluate the impact of the SSM in patients with different characteristics. Most importantly, existing evidence suggests that insulin costs documented in insurance claims do not include manufacturer rebates, so the observed insulin cost is not a true reflection of the cost to the health system (25). Thus, we performed a one-way sensitivity analysis to test the impact of the rebate for insulin on the long-term economic impact of SSM, assuming the rebate offset 70% of the insulin cost to the health system (payer). In addition, to explore the impact of parameter uncertainties on the cost and health outcomes of SSM, we performed a probabilistic sensitivity analysis using distributions of costs, utilities, and risk-related parameters (Supplementary Appendix 3). The results of the probabilistic sensitivity analysis are presented as 95% probabilistic CIs of the point estimates from the base-case analysis.

We identified 4.2 million Medicare insulin users eligible for SSM, among whom 1.6 million (38.3%) had average insulin OOP costs of >$35 per month and were likely to benefit from SSM (Table 1). Among 1.4 million individuals who ended the year in the deductible or initial phase (subgroup 1), 0.5 million (33%) were likely to enroll in Part D plans participating in SSM because of the potential OOP cost reduction. Compared with non-SSM beneficiaries, potential SSM enrollees included fewer females (29.6% vs. 52.0%) and more Whites (71.5% vs. 54.4%) and were more likely to have only Medicare coverage (64.5% vs. 32.2%; all P < 0.05).

Among 1.5 million individuals who would end the year in the coverage gap (subgroup 2), 0.6 million (41%) were likely to enroll in Part D plans participating in SSM. Compared with non-SSM beneficiaries, SSM enrollees were more likely to be White (80.4% vs. 69.2%), more likely to have higher family income ($70,668 vs. $40,883), and more likely to have only Medicare coverage (58.6% vs. 34.9%, all P < 0.05). Among 1.3 million individuals with the last insulin prescription in the catastrophic phase (subgroup 3), 0.6 million (41%) were likely to enroll in Part D plans participating in SSM. Compared with non-SSM beneficiaries, SSM enrollees were older (mean age 70.3 vs. 64.8 years), were less likely to be female (56.5% vs. 72.6%), had higher family income (mean income $79,046 vs. $38,680), and were more likely to only be covered by Medicare (50.9% vs. 26.8%) or dually eligible for commercial plans (44.3% vs. 19.6%) (all P < 0.05). Individuals with dual eligibility for Medicare and Medicaid will not be likely to enroll in SSM because they already have very low OOP for insulin.

In all three subgroups, SSM enrollees all had significantly higher mean annual OOP costs on insulin than non-SSM beneficiaries (subgroup 1 $462 vs. $93, subgroup 2 $675 vs. $78, and subgroup 3 $1,303 vs. $91, respectively; all P < 0.05). However, we only observed lower insulin adherence in SSM enrollees than in non-SSM beneficiaries in subgroup 2 (PDC 57.8% vs. 70.9%, P < 0.05). We further examined the association between insulin OOP cost and insulin adherence in all three subgroups using the regression approach (Supplementary Appendix 2) and found consistent evidence that the change in insulin use as a response to OOP cost change only occurred in subgroup 2.

Our simulation sample consists of 1.61 million individuals from all three subgroups who are likely to enroll in SSM. The estimated impact of SSM on the insulin OOP cost, days of supply for insulin, and A1C is presented in Table 2. For subgroup 1, SSM enrollment was associated with lower annual OOP costs in both White (−$221 [−43.7%]) and non-White (−$141 [−40.3%]) populations. For subgroup 3, SSM enrollment was associated with average annual OOP cost reductions of $1,063 (−73.6%) and $626 (−60.6%) for White and non-White population subgroups. For subgroup 2, SSM enrollment was associated with average annual OOP cost reductions of $349 (−51.9%) and $417 (−60.7%) for White and non-White subgroups. These OOP cost reductions further led to 43- and 40-day increases in insulin compliance and 0.19% and 0.18% reductions in average A1C level for White and non-White subgroups, respectively, in subgroup 2. Overall, 43.7% and 29.6% of the White and non-White Medicare insulin users would benefit from SSM, with average OOP cost reductions of $531 (61.6%) and $424 (57.7%). SSM will likely reduce the average annual OOP cost of insulin by $502 (60.7%) for the overall Medicare beneficiaries.

We presented the population-level simulation results for the overall SSM enrollees in Table 3. SSM was projected to avert 2,014 stroke cases, 935 MI cases, 315 CHF cases, 344 ESRD cases, and 1,132 deaths in 5 years while gaining 3,220 life-years (95% simulation CI 1,226–5,215) and 3,381 QALY (2,004–4,758). SSM was likely to increase the total cost of insulin by $3.45 (95% simulation CI $3.23–$3.67) billion and the total medical cost by $2.84 (95% simulation CI $1.94–$3.75) billion. If the rebate offset 70% of the insulin cost, the SSM was likely to only increase the total insulin cost and total medical cost by $1.04 (95% simulation CI $0.97–$1.10) billion and $0.42 (95% simulation CI −$0.35 to $1.19) billion. Throughout a 20-year time horizon, the SSM was projected to avert 3,513 stroke cases, 2,583 MI cases, 1,179 CHF cases, 1,601 ESRD cases, and 1,329 deaths. It would save 32,204 life-years (95% simulation CI 32,046–32,361) and gain 20,932 QALY (20,869–20,995). Meanwhile, the SSM would increase insulin costs by $9.22 (95% simulation CI $7.58–$10.85) billion but only increase the total medical cost by $5.56 (95% simulation CI $4.86–$6.25) billion due to lower complications. If the rebate offset 70% of the insulin cost, SSM would increase the total insulin cost by $2.77 (95% simulation CI $2.28–$3.26) billion and reduce the overall medical cost by $0.9 (95% simulation CI $0.24–$1.57) billion.

Based on our analysis and the study of Trish et al., we did not present results for subgroups 1 and 3 because SSM was not projected to impact their insulin use. We present the individual-level simulation results for subgroup 2 in Table 4. We found that SSM was projected to reduce risk of stroke (relative risk reduction 9.0%), MI (3.5%), CHF (1.4%), ESRD (1.8%), and death (1.0%) in 5 years. It was also projected to reduce patient OOP insulin costs by an average of $1,587 (95% simulation CI $1,650–$1,524), while increasing the insulin cost and total cost by $5,694 (95% simulation CI $5,488–$5,900) and $4,701 (95% simulation CI $3,547–$5,855), respectively, over a 5-year window. If the rebate offset 70% of the insulin cost to the health system, the SSM would only likely increase the total insulin and medical costs by $1,709 (95% simulation CI $1,646 to $1,772) and $916 (95% simulation CI −$95 to $1,927). Throughout a 20-year time horizon, the SSM was projected to reduce the risk of stroke (5.5%), MI (3.3%), CHF (1.5%), ESRD (2.7%), and death (0.3%). The implementation of SSM for 20 years is likely to extend life-years by an average of 0.052 (95% simulation CI 0.045–0.059) and QALY by an average of 0.035 (0.032–0.038) for patients from subgroup 2. It was also projected to reduce patient OOP insulin costs by an average of $4,069 (95% simulation CI $4,316–$3,822), while increasing the total insulin cost and medical cost by $15,237 (95% simulation CI $14,580–$15,894) and $9,196 (95% simulation CI $7,879–$10,513). If the rebate offset 70% of the insulin cost to the health system, the SSM would only likely increase the total insulin costs by $4,571 (95% simulation CI $4,373 to $4,769) and reduce total medical costs by $1,550 (−$3,314 to $34), respectively.

In this study, we quantified the scale of the Medicare SSM and its projected impacts on a spectrum of long-term clinical and economic outcomes among Medicare beneficiaries treated with insulin. We identified 1.61 million Medicare beneficiaries who may benefit from SSM, constituting 38.3% of the overall insulin users covered by Medicare. The simulation showed that SSM was likely to reduce insulin OOP costs by an average of $502 per person per year. However, only one-third of the beneficiaries (i.e., subgroup 2) would have improved insulin adherence as a result of OOP cost reduction due to observed adherence elasticity in our sample. The improved insulin adherence among the SSM enrollees would avert thousands of complication cases and extend thousands of life-years. However, the SSM will increase the overall insulin spending by Medicare because of the increased insulin use. These increased costs will be partially offset by savings in medical costs associated with decreased complications.

As designed, the SSM appeared to influence certain demographics and socioeconomic population subgroups differently. For example, the White subgroups, beneficiaries not dually eligible for Medicaid, and beneficiaries with high family incomes are more likely to enroll in SSM than their counterparts. This is mainly attributable to existing programs, such as Medicaid dual eligibility and low-income subsidy program, that help patients from low socioeconomic subgroups access insulin by reducing their OOP costs. Thus, the SSM will be more likely to reduce OOP costs for individuals with middle and relatively higher socioeconomic status. Greater enrollment of high socioeconomic subgroups is concerning because whether high insulin OOP cost is a major barrier to insulin use for this particular population with high socioeconomic status is unknown. In the worst case scenario, patients with higher socioeconomic status and already high adherence will have reduced OOP cost from SSM but a limited improvement in health outcomes will be realized because high OOP cost may not be a barrier for insulin use in this population. The reduction in OOP cost also may not be large enough to motivate economically disadvantaged patients to increase their insulin use. In that regard, the SSM is more likely to benefit insulin users who are financially stable and already adherent, possibly creating large racial and economic disparities, as well as long-term burdens on Medicare. A payment reduction policy that does not account for heterogeneous responses can potentially widen the difference in diabetes management across different population subgroups. If SSM can accurately target the subgroups with a high response to OOP cost change, its economic efficiency would significantly improve.

An improved plan design that accounts for patients’ responses to OOP costs changes could potentially improve the SSM’s economic efficiency. For example, OOP reductions for individuals who have reached the catastrophic phase by the end of the year may not help to improve insulin adherence, as patients are only responsible for a small amount of OOP payment for covering additional insulin. This policy may be more useful if resources can be focused on further reducing the OOP by a greater amount among individuals who end their year in the coverage gap (“donut hole”), where OOP poses a significant barrier for them to access and adhere to insulin.

We found a significant association between insulin OOP cost and insulin adherence only for the beneficiaries who end the year in the coverage gap (subgroup 2). For Medicare beneficiaries who end the year in the initial, deductible, and catastrophic phases, the insulin OOP cost in coverage gap was not associated with insulin adherence. A mechanism that potentially led to such heterogeneous responses was explained by Trish et al. (11). Because of such heterogeneous responses to OOP cost change, the Medicare SSM will likely reduce the insulin OOP costs in two-thirds of the SSM enrollees without improving their health outcomes, leading to a potential reduction in the economic efficiency of Medicare Part D. Our results suggest that an SSM design that accounts for patient responses to lower OOP costs may help guarantee long-term economic efficiency.

Among Medicare beneficiaries in the subgroup who end the year in the coverage gap, SSM was estimated to reduce the OOP cost by $1,587 (53.6%) and $4,069 (53.5%) over 5-year and 20-year time horizons, respectively. This OOP cost reduction was projected to increase insulin adherence by 10% on average, leading to an average A1C improvement of 0.2%, which may be significant at a population level if sustained for long periods of time. Results from our microsimulation indicated that the implementation of SSM would result in tangible clinical benefits for Medicare insulin users as a whole. In the 5-year initial implementation period, SSM was estimated to save 1,132 lives, extend 0.03 million life-years, and avert thousands of macrovascular and microvascular complications. If the SSM were to be implemented for 20 years, it would extend 0.032 million life-years and result in a gain of 0.021 million QALY.

While the SSM is projected to improve adherence and reduce diabetes-related complications, our simulation showed that there would be a substantial increase in total medical spending driven by increased insulin use. Who will bear these additional costs is not apparent, but the assumption is that it will mostly be Medicare. Centers for Medicare & Medicaid Services projections for SSM included savings of $250 million over 5 years contributed by the pharmaceutical industry via increased discounts during the coverage gap phase (10), which would be only a fraction of the spending increases. If such spending increases lead to increased Medicare premiums, additional barriers, especially for the economically disadvantaged population, may be introduced. Thus, continuous monitoring of the SSM implementation and its impact on plan premiums and the characteristics of the enrollees is vital. In addition, the proportion of rebates for insulin to payers to reduce the insulin cost is unclear in the U.S. Our study shows that, under the extreme assumption that 70% of the insulin cost was offset by rebates, SSM could potentially be a cost-saving policy in a 20-year window.

Our study has several limitations. First, the association between insulin OOP cost and insulin adherence was examined with use of cross-sectional data with a limited sample size. Thus, the endogeneity issue of OOP cost for patients on insulin cannot be adequately addressed in our study. To confirm our estimation of the demand elasticity for insulin, we externally validated our estimation with the results from the study of Trish et al. (11), where investigators examined the association between OOP cost of insulin and insulin adherence using claims data from 474,929 Medicare beneficiaries. While based on a different approach and a different database, the estimation of Trish et al. is highly consistent with ours, proving a good validity of our estimation. However, the endogeneity issue may still persist, which could bias our estimation of the policy’s impact. Thus, a more robust study with use of longitudinal data, a larger sample size, and a more advanced analytical approach (e.g., instrumental variable) is needed in the future to further investigate the demand elasticity of insulin. Second, the enrollment of SSM was determined based on whether OOP cost can be reduced through SSM. Patients might have other reasons to not enroll in the SSM than just OOP cost reduction (e.g., the inertia for changing plan), making our estimation on the scale of SSM a potential ceiling estimate. Lastly, concerns have been raised about using PDC to measure insulin adherence. However, adherence is not a key model input, and measurement error in adherence was unlikely to have a huge impact on our estimations of long-term outcomes of SSM.

The Medicare SSM may reduce the OOP costs of approximately one-third of the Medicare beneficiaries treated with insulin, who are more likely to be White and of higher socioeconomic status and who may have lower adherence sensitivity to OOP costs. Among these beneficiaries, SSM could reduce the OOP cost for insulin by 61% and may also substantially improve the health outcomes in one-third of this population due to improved insulin adherence. However, SSM will also increase overall Medicare spending for insulin and overall medical costs, which may impact future premiums and benefits.

Acknowledgments. The authors thank Dr. Ping Zhang from the Centers for Disease Control and Prevention for his consultation service.

Duality of Interest. H.S., L.S., and V.F. codeveloped the BRAVO simulation model and have an ownership interest in BRAVO4Health, a private company that aims to incorporate diabetes complications risk assessment in clinical practice. V.F. has received research grants (to Tulane University) from Bayer, Janssen, and Boehringer Ingelheim and honoraria for consultation work from Novo Nordisk, Sanofi, Eli Lilly, and AstraZeneca. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. H.S. and J.D.B. contributed to study concept and design. H.S., L.S., V.F., and J.L. contributed to acquisition, analysis, or interpretation of data. H.S., J.G., T.J., Y.Z., J.L., and J.D.B. contributed to drafting of the manuscript. All authors contributed to critical revision of the manuscript for important intellectual content. H.S. and O.G. contributed to statistical analysis. J.G., J.L., and J.D.B. contributed administrative, technical, or material support. H.S. and J.D.B. contributed to study supervision. H.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.