Christine Buttorff, PhD

 The use of generics in Medicare Part D generates cost savings for plan sponsors, beneficiaries, and the federal government. However, there is considerable variation in generic use across plans, even within a therapeutic class. Our objective is to understand the extent of variation in generic use in Part D and to understand factors associated with generic use.

 We used an observational study design using Medicare Part D claims from 2006 to 2016.

 We used descriptive statistics and regression analysis to examine the variation in generic and brand use across plans and the extent to which patient, plan, and area characteristics are associated with the choice of medication within a therapeutic class.

 Although generic use has increased markedly over time in Part D, substantial variation across plans persists in a number of common therapeutic classes. Beneficiary characteristics such as gender and health status are associated with higher/lower generic use, as are plan characteristics such as plan type (stand-alone prescription drug plan or Medicare Advantage), premium, and parent company.

 Because we cannot study the impact of brand-name drug rebates on generic use, we can study the variation in generic use across Part D plans as an indirect way to assess pharmacy benefit manager and plan incentives. We find circumstantial evidence that, in certain classes, rebates may play a role in influencing brand over generic use, although the exact relationship is unknowable given the proprietary nature of rebates.

Generic use has increased markedly over time in Medicare Part D, but substantial variation across plans persists in a number of common therapeutic classes.

Beneficiary and plan characteristics are associated with generic use.

We find circumstantial evidence that, in certain classes, rebates may play a role in influencing brand over generic use, although the exact relationship is unknowable given the proprietary nature of rebates.

Changing Part D’s underlying benefit design to reduce incentives to use high-cost drugs or encouraging the use of more tightly managed formularies could increase generic use.

The growing gap between a drug’s list (or “prerebate”) price and what manufacturers actually receive after discounts or rebates (the net price) is at the center of an increasingly public debate over rising drug costs.

 Pharmacy benefit managers (PBMs) typically negotiate rebates or discounts from drug manufacturers on behalf of plan sponsors (“plans”) in return for preferred formulary placement. For example, off-patent brands such as Lipitor (atorvastatin), Nexium (esomeprazole), and Neurontin/Lyrica (pregabalin) remained top-selling medications in Medicare Part D despite widely available generic versions or over-the-counter substitutes.

 A possible explanation for this is that PBMs or plans may place higher-cost drugs on lower cost-sharing tiers to increase the volume used of the drugs in exchange for larger rebates.

 Lower cost sharing for enrollees (involving lower co-payments or coinsurance, for instance) increases utilization.

 Favoring high-cost brands over lower-cost generics may reflect strategic behavior by plans or their sponsors (PBMs) to steer enrollees to heavily rebated drugs.

Medicare Part D is the federal prescription drug program for the disabled and those 65 years and older. Part D plans must use a pharmacy and therapeutics committee to establish formularies.

 Plans must cover at least 2 drugs in each class and must cover substantially all drugs in 6 protected classes (antidepressants, antipsychotics, anticonvulsants, immunosuppressants for transplants, antiretrovirals, and antineoplastics).

 The standard Part D benefit includes a deductible, an initial coverage limit, a coverage gap, and a catastrophic phase that beneficiaries enter when they have spent $6350 in out-of-pocket costs in 2020. In each benefit phase, the extent of cost liability shifts among the plans, manufacturers, beneficiaries, and the federal government.

Although Part D plans are required to report rebates and other discounts to CMS, recent analyses find greater use of high–list price, high-rebate drugs in Part D than in commercial markets.

 This raises concerns that several structural features of Part D may increase plans’ incentives to encourage costly, high-rebate drugs. A majority of drugs in stand-alone plans (but not Medicare Advantage plans with drug coverage) now require coinsurance rather than a co-payment

; in this situation, high prerebate prices increase the beneficiary’s out-of-pocket cost, moving beneficiaries into the catastrophic phase of the benefit sooner—the fastest growing portion of the program’s cost.

 Once a beneficiary reaches the catastrophic phase, plan liability is low and federal liability is high: The federal government pays 80% of the drug costs on average, the plan pays 15%, and nonsubsidized beneficiaries pay 5% of the prerebate price.

Additionally, the Affordable Care Act and the Balanced Budget Amendment of 2018 restructured the Part D benefit to reduce beneficiary cost sharing in the coverage gap.

 The primary change required manufacturers to provide discounts on brand drugs purchased in the coverage gap, with the discount reaching 70% of the prerebate price in 2019. The manufacturer discount counts toward beneficiaries’ out-of-pocket spending, which also pushed more beneficiaries into the catastrophic phase where plan liability is low.

Because rebates are proprietary, we cannot directly assess their impact on plans’ formulary design and medication use. (CMS requires plans to submit information on Direct and Indirect Remuneration in order to reconcile expenditures at the end of the year. However, this information is proprietary and is not available to researchers.) Rather, we assess this indirectly by examining the extent of variation in generic use across Part D plans. We use Medicare Part D data from 2006 to 2016 to examine differences in brand and generic drug use within a therapeutic class and the extent to which beneficiary, plan, and area characteristics explain the variation.

We used Part D Prescription Drug Event (PDE) data from January 2006 through December 2016 for a 20% random sample of Medicare beneficiaries enrolled in Part D. Part D plans are sold either as stand-alone prescription drug plans (PDPs) or in conjunction with a Medicare Advantage plan that also includes medical benefits (MAPD). The PDE data capture drug name, National Drug Code (NDC), dosage, quantity, date of service, and payments made by the beneficiary, plan, and other third-party coverage. We aggregated individual NDCs into therapeutic classes based on First Databank (FDB) definitions, focusing on the 10 most commonly used classes in Part D: antiasthmatics, antihistamines, antidiabetics, antineoplastics, anticoagulants, antihypertensives, antihyperlipidemics, anticonvulsants, antidepressants, and antiulcerants. Antineoplastics, anticonvulsants, and antidepressants are 3 of the 6 protected classes. We linked the PDE data to the plan characteristics file, which contains detailed information on each plan’s formulary and benefit design at the NDC level. This information was then linked to plan formulary files containing additional information on whether a given plan requires prior authorization or step therapy for a specific drug.

Most Part D plans use tiered formularies to steer beneficiaries to preferred medications. In 2015, the vast majority of Part D enrollees were in plans that used 5 cost-sharing tiers: preferred and nonpreferred tiers for generic drugs, preferred and nonpreferred tiers for brand drugs, and a tier for high-cost specialty drugs, with average co-payments of $1, $4, $38, and $80 for the first 4 tiers, respectively, and coinsurance of 29% in the specialty tier.

 Plans may also combine formularies with other rules—such as step therapy and prior authorization—to manage beneficiaries’ use of prescription drugs. Restrictive formularies that list fewer drugs in each therapeutic class will generate higher rebates because plans can drive volume toward these drugs.

We relied on FDB definitions to classify drugs as either single-source brand, multisource brand, or generic. Multisource brands have generic equivalents with the same active ingredient(s), and almost all prescriptions for multisource brand drugs are dispensed as generics.

 Single-source brands have no generic substitutes, so prescriptions written for single-source drugs limit Part D plans’ opportunities for generic substitution.

Our key outcome is the generic dispensing rate (GDR), defined as the percentage of generic fills of all fills in a therapeutic class per year, weighted by the days supplied. We additionally examined the rate of single-source drug use within a class, which is the inverse of the generic utilization rate minus the multisource use rate. We did not use the generic substitution rate because this is typically 95% or higher. Therapeutic substitution is where a generic is substituted for a brand of a different molecule. The appropriateness of therapeutic substitution is complex and beyond the scope of this paper.

We regressed a set of beneficiary and plan characteristics in a logit model to assess the odds that a given fillwas for a generic drug, within a therapeutic class. The unit of analysis is the PDE (or claim). We adjusted for several beneficiary characteristics that come from the Master Beneficiary Summary and Enrollment files, including age, gender, race/ethnicity, and beneficiary status (dual eligible, low-income subsidy [LIS], non-LIS). We used the number of different drug classes that a beneficiary took in a year as a proxy for health status.

We included several plan characteristics such as each plan’s monthly premium and whether it provides some coverage in the coverage gap (also known as the “donut hole”). Managed plans tend to have more control over physician prescribing behavior and typically make greater use of utilization management tools, so we include a binary indicator for MAPDs. Given that 3 firms—UnitedHealth, Humana, and CVS Health—account for more than half of all Part D enrollees and the 10 largest sponsors account for nearly 90% of enrollment, we include binary indicators for each of the 10 largest plan sponsors to assess variation in generic use across parent companies.

 Finally, we included state, year, and drug class fixed-effects to control for any unmeasured differences in generic use across states, years, or drug classes. We used Stata/MP 14.1 (StataCorp) for all analyses. Significance levels were determined using clustered standard errors at the plan level.

Our primary regression analyses report the years 2010-2016 because in those years the indicator about whether the fill required prior authorization was available. We provide the results for the full set of years (2006-2016) without the utilization management indicator in the 

Following trends in the commercial market, the GDR in Part D reached 90% for MAPDs and 88% for PDPs in 2016 across all drug classes, up from 63% and 53%, respectively, in 2006 (

). Generic use remained higher in MAPDs compared with stand-alone PDPs, but the difference narrowed over time.

 shows the distribution of GDRs in 2016 at the plan-class level for the 10 classes we examined in detail, separately for PDPs and MAPDs. The ends of the boxes show the interquartile range of GDRs (25th and 75th percentiles), and the whiskers represent the 10th and 90th percentiles of plans. The variation in generic use across drug classes is largely due to different product cycles. Classes such as antiasthmatics were brand dominant in 2016, with average generic dispensing rates of 20% to 25%. Plan-level variation in generic use tends to be lower in generic-dominant classes, such as antidepressants, and higher in classes with over-the-counter substitutes, such as antihistamines and antiulcerants.

More importantly, Figure 2 highlights variation in generic use 

 a therapeutic class. For example, in the treatment of type 2 diabetes, 10% of PDPs have GDRs of 50% or less, whereas 1 in 10 plans have GDRs higher than 70%. Given that each plan faces the same set of treatment options, large differences in generic use within a class may be due in part to nonclinical factors, such as coverage generosity, formulary restrictions, or differential strategies to favor high-rebate drugs. Not all plans cover the exact same mix of drugs in each class; however, there are 3 protected classes (antidepressants, anticonvulsants, and antineoplastics) shown in Figure 2 for which plans must cover substantially all of the active ingredients in the class, and there is still variation in generic use for these classes.

To better understand how plans differed in their generic use, the 

 shows class-level market shares within selected classes among Part D plans in the top and bottom quartiles of generic dispensing in 2016. In each of the 4 classes, plans in the bottom quartile of GDR have higher use of a brand drug relative to the top-quartile GDR plans. For example, esomeprazole (Nexium) was the third most widely used antiulcerant in low GDR plans, with an average market share of 12.7%, but it was rarely filled in the highest GDR plans.

Another possible explanation for these differences is that some plans favor highly rebated drugs over lower-cost alternatives by placing them on a lower tier with lower patient cost sharing to encourage use. Although we cannot infer causality, we observe this pattern occurring in some classes. The Table includes a column for the most common tier placement for the highest utilized single-source drugs among plans in the bottom quartile of the distribution of GDR. Esomeprazole was often not covered in the highest GDR plans; most low GDR plans considered it a “preferred brand” (tier 3).

Thus far, our results have been descriptive and have not controlled for a range of beneficiary, plan, provider, and area characteristics that can also affect medication choices, ranging from health status to perceptions that generics are less effective.

 displays the odds ratios of filling a generic prescription from class-level regressions that control for patient, plan, and area characteristics. Dual-eligible beneficiaries and those receiving the LIS were 24% and 28%, respectively, less likely to receive a generic compared with nonsubsidized beneficiaries, conditional on receiving a drug in the class.

Plan-level factors also affect the mix of brand and generic drugs. Beneficiaries enrolled in plans with higher premiums were less likely to fill a generic drug, presumably because coverage of brand drugs was more generous in higher-premium plans. Those enrolled in stand-alone PDPs were 10% less likely to fill a generic prescription than similar beneficiaries in MAPDs.

As we noted earlier, the Part D market is highly concentrated,

 and some of these plans also operate PBMs. We hypothesized that market concentration and other organizational features may play roles in influencing generic usage. Accordingly, we found that beneficiaries in Kaiser Permanente, a tightly managed staff model health maintenance organization, were 49% more likely to fill a generic than beneficiaries in other plans, and all of Kaiser’s plans are MAPD. They were followed closely by WellCare, a large insurer predominantly operating in government markets (Medicaid, Medicare Advantage, and Part D). On the other end, GDRs were lowest in plans operating as both PBMs and Part D plans (Express Scripts, CVS Caremark, and UnitedHealth). Beneficiaries in UnitedHealth and CVS plans were 19% and 20%, respectively, less likely to fill generics than other beneficiaries.

Beyond beneficiary and plan characteristics, prior work has shown that Medicare spending on pharmaceuticals varies widely across areas of the country such as hospital referral regions, even after adjustment for demographic characteristics, insurance coverage, and individual health status.

 Across the country, average GDRs across the 10 classes range from 0.83 to 0.88 (

). The states with the lowest and highest predicted GDR are New Jersey and Massachusetts, respectively.

Despite high overall use of generic drugs in Part D, generic utilization rates within a therapeutic class vary across plans. Some of the variation is due to the mix of beneficiaries enrolled in a plan. Dual-eligible beneficiaries and those receiving the LIS face little or no cost sharing and thus are less price sensitive to high-cost drugs. Plan features also matter. Generic dispensing rates are higher in MAPDs compared with PDPs, although the gap is narrowing over time. Generic rates are lower in plans with higher premiums that typically have lower cost sharing.

Nonetheless, significant variation persists after controlling for beneficiary and plan characteristics, raising concerns that the program’s design and institutional features may incentivize plans to favor high–list price, high-rebate drugs over lower-cost alternatives. A widely cited example of this is esomeprazole (Nexium), which accounted for the most spending in Medicare Part D in 2013 at $2.5 billion, despite widespread availability of close therapeutic substitutes and over-the-counter antiulcerants.

 Even after the drug lost patent protection in 2014, we still found that plans in the lowest quartile of GDR still had fills for the brand-name version.

In a competitive market, why would Part D plans favor a higher-cost drug? One possible explanation is the growth in rebates.

 Rebates are most common for high-cost brand drugs in competitive therapeutic classes, where there are multiple choices of similar products. Within Medicare Part D, rebates have more than doubled as a share of total spending since the program’s inception, from 9.6% in 2007 to 21.8% in 2017.

 Previous work found that 70% of Part D plans place at least 1 brand drug on a lower tier when a generic version is available, which is often cited as evidence of increasing rebates in return for more favorable tier placement to increase volume.

 Other work has found that a slight majority of generic drugs (53%) are placed on nongeneric tiers.

 More recent work found that by 2019, plans generally covered the generic version of a drug instead of the brand, when both were available.

 Besides more recent data, these findings may differ from those of prior work because the authors analyzed a small number of active ingredients and excluded brand drugs that did not have an exact generic equivalent in terms of form and dose.

While rebates lower plans’ overall costs, the tier placement also affects beneficiary liability, which in turn affects plan spending in Part D. A majority of drugs in PDPs (but not in MAPDs) now require coinsurance rather than a co-payment.

 Coinsurance on drugs with a higher prerebate price means that beneficiaries hit the catastrophic phase sooner, increasing costs to both beneficiaries and the federal government, but decreasing plan liability.

Consolidation at the parent level may allow the largest insurers to negotiate better discounts with manufacturers, but the potential for rebates to drive utilization becomes greater.

 We found lower generic utilization among firms operating as both a PBM and drug plan, specifically UnitedHealth, CVS, and Express Scripts. The Part D marketplace is dynamic and continues to undergo consolidation. In recent years, several large mergers have been completed or proposed between CVS Health and Aetna,

 Although Anthem and WellPoint merged in 2004, WellPoint did not begin using the Anthem brand name until 2014.

 Torchmark sold its Part D plans to SilverScript, a subsidiary of CVS, in 2016.

 Increased consolidation and vertical integration in Part D may restrict competition and raise total program costs.

Stand-alone PDPs manage prescription drug coverage only, whereas MAPD plans are at risk for both medical and drug expenditures. Recent work suggests that MAPDs provide more generous coverage of drugs that reduce downstream medical costs (including generic maintenance drugs), which is consistent with our findings of modestly higher generic dispensing rates in MAPDs.

2016 is the most recent year of claims data available to us, meaning more recent changes in generic use are not reflected here. Some drugs may be more effective or have fewer adverse effects than other medications in the class, at least for some patients, and physicians may have different experiences that affect their prescribing behavior. These unmeasured factors could bias our results if they differ systematically differ across plans, but there is no evidence to suggest that it is a concern.

Despite some widely publicized examples of excess profiteering in generic markets, new medicines will continue to drive drug spending. However, there are still opportunities for significant cost savings from generics, given the variation in GDR we found. One option is for CMS to prohibit Part D plans from giving branded products more favorable formulary placement than generic products. Another option would be the Trump administration’s proposal to apply rebates at the point of sale that was abandoned last year.

 One potential impact of that policy would be to reduce the incentive to place branded drugs on lower cost-sharing tiers, thus increasing generic use rates.

Variation in generic use has real consequences for plans, beneficiaries, and taxpayers. We find that overall generic utilization is increasing in Part D, but variation across plans within a class persists. Beneficiary and plan characteristics, state, and parent organization are all factors contributing to differences in GDR within a therapeutic class. We find evidence that rebates may be playing a role in discouraging generic use in particular therapeutic classes, but there is no public information on the extent of these rebates. Focusing on changing the underlying benefit design to reduce incentives to use high-cost drugs or encouraging the use of more tightly managed formularies could increase generic use.

 RAND Corporation (CB), Arlington, VA; IQVIA (YX), San Francisco, CA; University of Southern California, School of Pharmacy (YX, GJ) and Schaeffer Center for Health Policy & Economics (GJ), Los Angeles, CA.

 Dr Buttorff acknowledges support from the Schaeffer-RAND initiative; Dr Joyce was supported by a grant from the National Institute on Aging, R01AG059234.

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (CB, YX, GJ); acquisition of data (YX, GJ); analysis and interpretation of data (CB, YX, GJ); drafting of the manuscript (CB, GJ); critical revision of the manuscript for important intellectual content (CB, GJ); statistical analysis (YX, GJ); and administrative, technical, or logistic support (CB).

 Christine Buttorff, PhD, RAND Corporation, 1200 S Hayes St, Arlington, VA 22202. Email: buttorff@rand.org.

1. Thomas K, Goodnough A. Trump’s efforts to rein in drug prices face setbacks. 

. July 11, 2019. Accessed October 8, 2020. https://www.nytimes.com/2019/07/11/health/drug-prices-rebate-donald-trump.html

2. Aitken M, Kleinrock M. Medicine use and spending in the U.S.: a review of 2017 and outlook to 2022. IQVIA. April 19, 2018. Accessed October 8, 2020. https://www.iqvia.com/insights/the-iqvia-institute/reports/medicine-use-and-spending-in-the-us-review-of-2017-outlook-to-2022

3. Dusetzina SB, Bach PB. Prescription drugs—list price, net price, and the rebate caught in the middle. 

. 2019;321(16):1563-1564. doi:10.1001/jama.2019.2445

4. Dusetzina SB, Jazowski S, Cole A, Nguyen J. Sending the wrong price signal: why do some brand-name drugs cost Medicare beneficiaries less than generics? 

. 2019;38(7):1188-1194. doi:10.1377/hlthaff.2018.05476

5. Gibson TB, McLaughlin CG, Smith DG. Generic utilization and cost-sharing for prescription drugs. 

 2010;22:195-219. doi:10.1108/s0731-2199(2010)0000022012

6. Goldman DP, Joyce GF, Zheng Y. Prescription drug cost sharing: associations with medication and medical utilization and spending and health. 

. 2007;298(1):61-69. doi:10.1001/jama.298.1.61

7. Hoadley JF, Merrell K, Hargrave E, Summer L. In Medicare Part D plans, low or zero copays and other features to encourage the use of generic statins work, could save billions. 

. 2012;31(10):2266-2275. doi:10.1377/hlthaff.2012.0019

8. Zhang Y, Lave JR, Newhouse JP, Donohue JM. How the Medicare Part D drug benefit changed the distribution of out-of-pocket pharmacy spending among older beneficiaries. 

. 2010;65(4):502-507. doi:10.1093/geronb/gbp111

9. Medicare Prescription Drug Benefit Manual: chapter 6 – Part D drugs and formulary requirements. CMS. January 15, 2016. Accessed October 8, 2020. https://www.cms.gov/Medicare/Prescription-Drug-coverage/PrescriptionDrugCovContra/Downloads/Part-D-Benefits-Manual-Chapter-6.pdf

10. Part D payment system. Medicare Payment Advisory Commission. October 2018. Accessed October 8, 2020. http://medpac.gov/docs/default-source/payment-basics/medpac_payment_basics_18_partd_final_sec.pdf?sfvrsn=0

11. Socal MP, Bai G, Anderson GF. Favorable formulary placement of branded drugs in Medicare prescription drug plans when generics are available. 

. 2019;179(6):832-3. doi:10.1001/jamainternmed.2018.7824

12. Fein AJ. Why Part D plans prefer high list price drugs that raise costs for seniors. Drug Channels. January 22, 2020. Accessed October 8, 2020. https://www.drugchannels.net/2020/01/why-part-d-plans-prefer-high-list-price.html

13. Pearson CF, Brantley K, Frieder M. Majority of drugs now subject to coinsurance in Medicare Part D plans. News release. Avalere; March 10, 2016. Accessed October 8, 2020. https://avalere.com/press-releases/majority-of-drugs-now-subject-to-coinsurance-in-medicare-part-d-plans

14. Chapter 14: The Medicare prescription drug program (Part D): status report. In: Medicare Payment Advisory Commission. 

Report to the Congress: Medicare Payment Policy

. Medicare Payment Advisory Commission; March 2018:397-442.

15. Bipartisan Budget Act of 2018, HR 1892, 115th Cong (2018). https://www.congress.gov/bill/115th-congress/house-bill/1892/text

16. Cubanski J. Summary of recent and proposed changes to Medicare prescription drug coverage and reimbursement. Kaiser Family Foundation. February 15, 2018. Accessed October 22, 2019. https://www.kff.org/medicare/issue-brief/summary-of-recent-and-proposed-changes-to-medicare-prescription-drug-coverage-and-reimbursement/

17. Cubanski J, Neuman T, Damico A. Closing the Medicare Part D coverage gap: trends, recent changes, and what’s ahead. Kaiser Family Foundation. August 21, 2018. Accessed October 8, 2020. https://www.kff.org/medicare/issue-brief/closing-the-medicare-part-d-coverage-gap-trends-recent-changes-and-whats-ahead/

18. Hoadley J, Cubanski J, Neuman T. Medicare Part D at ten years: the 2015 marketplace and key trends, 2006-2015. Kaiser Family Foundation. October 5, 2015. Accessed October 8, 2020. https://www.kff.org/medicare/report/medicare-part-d-at-ten-years-the-2015-marketplace-and-key-trends-2006-2015/

19. Effects of using generic drugs on Medicare’s prescription drug spending. Congressional Budget Office. September 2010. Accessed October 8, 2020. https://www.cbo.gov/sites/default/files/111th-congress-2009-2010/reports/09-15-prescriptiondrugs.pdf

20. Cubanski J, Damico A, Neuman T. Medicare Part D in 2018: the latest on enrollment, premiums, and cost sharing. Kaiser Family Foundation. May 17, 2018. Accessed October 8, 2020. https://www.kff.org/medicare/issue-brief/medicare-part-d-in-2018-the-latest-on-enrollment-premiums-and-cost-sharing/

21. Howard J, Frank G, Kiptanui Z, Hansen R, Qian J, Harris I. Identifying and understanding influencers of generic drug utilization among Medicare beneficiaries. 

. 2016;56(suppl 3):100. doi:10.1093/geront/gnw162.389

22. Howard JN, Harris I, Frank G, Kiptanui Z, Qian JJ, Hansen R. Influencers of generic drug utilization: a systematic review. 

. 2018;14(7):619-627. doi:10.1016/j.sapharm.2017.08.001

23. Zhang Y, Baicker K, Newhouse JP. Geographic variation in Medicare drug spending. 

. 2010;363(5):405-409. doi:10.1056/NEJMp1004872

24. Begley S. Nexium, Advair led Medicare drug spending in 2013—officials. Reuters. April 30, 2015. Accessed October 8, 2020. https://www.reuters.com/article/us-usa-healthcare-medicare/nexium-advair-led-medicare-drug-spending-in-2013-officials-idUSKBN0NL2IK20150430

25. Use of pharmacy benefit managers and efforts to manage drug expenditures and utilization. U.S. Government Accountability Office. August 13, 2019. Accessed October 8, 2020. https://www.gao.gov/products/GAO-19-498

26. 2019 Annual Report of the Boards of Trustees of the Federal Hospital Insurance and Federal Supplementary Medical Insurance Trust Funds. CMS. April 22, 2019. Accessed October 8, 2020. https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/ReportsTrustFunds/Downloads/TR2019.pdf

27. Sloan C, Young J, Donthi S, Fix A, Meltzer R. For the first time, a majority of generic drugs are on non-generic tiers in Part D. Avalere. February 13, 2020. Accessed May 26, 2020. https://avalere.com/insights/for-the-first-time-a-majority-of-generic-drugs-are-on-non-generic-tiers-in-part-d

28. Dusetzina SB, Cubanski J, Nshuti L, et al. Medicare Part D plans rarely cover brand-name drugs when generics are available. 

. 2020;39(8):1326-1333. doi:10.1377/hlthaff.2019.01694

29. Alston M, Dieguez G, Tomicki S. A primer on prescription drug rebates: insights into why rebates are a target for reducing prices. Milliman. May 21, 2018. Accessed October 22, 2019. https://www.milliman.com/en/insight/a-primer-on-prescription-drug-rebates-insights-into-why-rebates-are-a-target-for-reducing

30. Schulman KA, Dabora M. The relationship between pharmacy benefit managers (PBMs) and the cost of therapies in the US pharmaceutical market: a policy primer for clinicians. 

. 2018;206:113-122. doi:10.1016/j.ahj.2018.08.006

31. CVS Health completes acquisition of Aetna, marking the start of transforming the consumer health experience. News release. CVS Health; November 28, 2018. Accessed October 8, 2020. https://cvshealth.com/news-and-insights/press-releases/cvs-health-completes-acquisition-of-aetna-marking-the-start-of

32. Cigna completes combination with Express Scripts, establishing a blueprint to transform the health care system. News release. Cigna; December 20, 2018. Accessed October 8, 2020. https://www.cigna.com/about-us/newsroom/innovation/cigna-completes-combination-with-express-scripts

33. Livingston S. Centene-WellCare deal nabs shareholder approval. Modern Healthcare. June 24, 2019. Accessed October 8, 2020. https://www.modernhealthcare.com/insurance/centene-wellcare-deal-nabs-shareholder-approval

34. Company history. Anthem. Accessed October 8, 2020. https://www.antheminc.com/aboutantheminc/companyhistory/index.htm

35. Torchmark Corporation: Form 8-K. Securities and Exchange Commission. July 15, 2016. Accessed July 23, 2019. https://www.sec.gov/Archives/edgar/data/320335/000032033516000088/a8-kpartdsale.htm

36. Lavetti K, Simon K. Strategic formulary design in Medicare Part D plans. 

. 2018;10(3):154-192. doi:10.1257/pol.20160248

37. Starc A, Town RJ. Internalizing behavioral externalities: benefit integration in health insurance. National Bureau of Economic Research working paper 21783. December 2015. Updated April 2018. Accessed October 8, 2020. https://www.nber.org/papers/w21783

38. Proposed safe harbor regulation. CMS. August 30, 2018. Accessed July 23, 2019. https://www.cms.gov/Research-Statistics-Data-and-Systems/Research/ActuarialStudies/Downloads/ProposedSafeHarborRegulationImpact.pdf

Articles

Generic use has increased over time in Medicare Part D, but substantial variation across plans persists in a number of common classes.

Variation in Generic Dispensing Rates in Medicare Part D

 Value-based insurance design (VBID) lowers cost sharing for high-value healthcare services that are clinically beneficial to patients with certain conditions. In 2017, the Center for Medicare and Medicaid Innovation began a voluntary VBID model test in Medicare Advantage (MA). This article describes insurers’ perspectives on the MA VBID model, explores perceived barriers to joining this model, and describes ways to address participation barriers.

 In spring/summer 2017, we conducted semistructured interviews with 24 representatives of 10 nonparticipating MA insurers to learn why they did not join the model test. We interviewed 73 representatives of 8 VBID-participating insurers about their participation decisions and implementation experiences. All interview data were analyzed thematically.

 Fewer than 30% of eligible insurers participated in the first 2 years of the model test. The main barriers to entry were a perceived lack of information on VBID in MA, an expectation of low return on investment, concerns over administrative and information technology (IT) hurdles, and model design parameters. Most VBID participants encountered administrative and IT hurdles but overcame them. CMS made changes to the model parameters to increase the uptake.

 The model uptake was low, and implementation challenges and concerns over VBID effectiveness in the Medicare population were important factors in participation decisions. To increase uptake, CMS could consider providing in-kind implementation assistance to model participants. Nonparticipants may want to incorporate lessons learned from current participants, and insurers should engage their IT departments/vendors early on.

This is the first empirical study of value-based insurance design (VBID) in the Medicare population.

Fewer than 30% of eligible insurers participated in the Medicare Advantage (MA) VBID model test.

Nonparticipating insurers cited a lack of information about VBID performance in MA, an expectation of low return on investment, potential implementation challenges, and model design parameters as barriers to participation.

Participants highlighted the appeal of the VBID test as an opportunity to innovate and explained how they overcame implementation challenges.

CMS and insurers could use study insights to facilitate adoption of VBID as its use expands.

Increasing cost sharing (eg, deductibles, co-payments, coinsurance) can reduce utilization of healthcare services.

 However, some services, such as chronic disease medications and preventive monitoring and screening tests, are both clinically beneficial and of high value. Value-based insurance design (VBID) reduces cost sharing for high-value services to increase their use and ultimately improve patient health and reduce healthcare spending; cost-sharing reductions, however, are offered only to the patients most likely to benefit—such as those with chronic diseases.

VBID initiatives have most recently been implemented in employer-based populations,

 where they have increased service utilization but shown limited impact on spending or patient health.

 VBID has not been tested in the Medicare population; it is not known how older beneficiaries would react to reduced cost sharing for targeted services.

In 2015, CMS introduced a voluntary VBID model test for Medicare Advantage (MA) insurers. MA VBID waived a uniformity requirement that precluded insurers from offering different benefits and cost sharing to enrollees in the same plan.

 Starting in 2017, participating insurers in eligible states (

) could offer reduced cost sharing for high-value services or providers and/or offer supplemental benefits to beneficiaries with specific chronic conditions. Insurers could require that beneficiaries participate in care management activities before becoming eligible for VBID benefits. CMS did not provide extra financial incentives to participating insurers.

In parallel to the model test, CMS recently reinterpreted the uniformity requirement, giving MA insurers more flexibility to tailor benefits based on beneficiaries’ clinical needs.

 The change allows all MA insurers to adopt VBID approaches for Part C benefits beginning in 2019. Moreover, the Bipartisan Budget Act of 2018 expands the MA VBID model test to all 50 states in 2020.

Despite the dramatic increases in MA insurers’ ability to design more tailored benefits, VBID model uptake has been lower than expected: Only 10 (<30%) eligible MA insurers participated in the first 2 years of the VBID model test. In this study, we explored insurers’ perspectives on MA VBID, identified perceived barriers to joining the model test, and described ways to overcome them. Our findings suggest that implementation barriers and reservations about VBID in the Medicare population may hamper insurers’ enthusiasm about this type of flexible benefit design in the short term. Our findings may be useful for both CMS and MA insurers to facilitate the adoption of VBID as its use expands via both the model test and, more broadly, the uniformity requirement reinterpretation.

 We identified MA insurers eligible to participate in VBID in 2017 and 2018 by applying model eligibility criteria to publicly available MA insurer and enrollment data available as of December 2016. We also included 5 insurers interested in VBID but not meeting model eligibility criteria that contacted CMS during the first VBID application period. From this group, we contacted the largest 29 nonparticipating insurers, starting with national insurers, then reached out to larger regional or state-based insurers, aiming to speak with organizations from all eligible states. Of the 29 insurers contacted, 10 agreed to be interviewed, 14 did not respond to our invitation, and 5 declined to be interviewed. There were no significant differences in for-profit status or Blue Cross and/or Blue Shield (BCBS) affiliation between those nonparticipants who we interviewed and those we did not. However, the sample of nonparticipants we interviewed had more regional than national insurers, and there were more national than regional insurers among those we did not interview. The proportion of state-level insurers did not vary across the 2 groups.

Between February and March 2017, 2 researchers conducted 45-minute telephone interviews with each of the 10 nonparticipating insurers who agreed to be interviewed. We interviewed 24 representatives of 2 large national and 8 small regional insurers, including chief compliance officers, vice presidents for Medicare products, and medical directors for government programs, among others. We used a semistructured protocol to learn about the main reasons for not participating in VBID, barriers to participation, and VBID model changes that might make it more attractive. We also analyzed written comments that nonparticipating insurers had sent to CMS.

 Between June and September 2017, 2 researchers conducted individual or small-group interviews with 73 representatives from 8 of the 9 VBID-participating MA insurers. One participating insurer declined to be interviewed, stating a delay in its implementation. Each interview lasted 60 to 90 minutes. We interviewed representatives of 4 MA insurers during in-person site visits; the other 4 interviews were by telephone. Interviews followed a semistructured format covering topics such as the decision to participate, early implementation experiences, implementation barriers and facilitators, and feedback to CMS. We supplemented these semistructured interviews by reviewing the insurers’ VBID application materials.

Interviewees held a variety of positions in their organizations, including Medicare product specialists, Medicare compliance officers, actuarial directors, directors of regulatory affairs, care management directors and staff, informatics specialists, and/or medical directors of government programs.

All interviews were audio-recorded and transcribed. The RAND Institutional Review Board exempted the study from review.

Four experienced qualitative researchers used MaxQDA (VERBI Software; Berlin, Germany) to code each transcript and identify key themes, using a code book with codes derived deductively from the interview guides (eg, reasons for [not] joining the model test) and generated inductively based on unanticipated themes emerging from interviews (eg, information technology [IT] challenges).

 The coding team blindly double-coded 2 interviews with VBID nonparticipants and 4 interviews with VBID participants and discussed and resolved any discrepancies. All other interviews were coded by one person and reviewed by another. A few coding discrepancies, primarily related to the nuances of VBID model implementation, were resolved during team meetings.

We identified the most frequently mentioned reasons for either not joining or joining the model test and described strategies that participants used to overcome perceived barriers to joining MA VBID. 

 provides additional quotations illustrating the main themes we present in the following Results section. To protect the confidentiality of our study participants, we deidentified insurer names (we use “NPInsurer” for VBID nonparticipants and “PInsurer” for participants) and refrained from providing individual-level characteristics of our study participants.

Differences Between Participants and Nonparticipants

Only 10 insurers (<30%) participated in VBID during the first 2 years of the model test. Five participants were from Pennsylvania, 3 from Massachusetts, and 1 each from Indiana and Michigan. Nine participants were state-based insurers; 1 was a national insurer. Four were BCBS affiliates. Participants chose to enter plans primarily in their health maintenance organization contracts. (Of 12 contracts, only 3 were preferred provider organization contracts.) Compared with participants, nonparticipants were less likely to be not-for-profit and state-level (as opposed to regional or national) insurers. Participants were no more likely to be BCBS affiliates than were nonparticipants.

The MA market is dynamic and, in many geographic areas, very competitive. Because participants did not know who would apply and be accepted for VBID, it is difficult to know whether competitive pressures affected their decisions. Nonetheless, none of our study participants believed that the level of market competition or their market share had an impact on their decisions. Model test participants, however, stated that Pennsylvania and Massachusetts, the states where most participants were from, are “in [fore]front of healthcare [reform] in general” (PInsurer03) and are full of “forward-thinking” insurers (PInsurer07).

The nonparticipants we interviewed identified 4 main reasons for not joining the model test. First, 8 nonparticipating insurers felt that they did not have enough information to structure their VBID offerings. As part of the application process, insurers had to demonstrate that their designs would achieve savings over a 5-year period. To estimate savings, insurers needed better actuarial information on the likely changes in utilization that could be expected in the Medicare population. Many insurers did not feel comfortable estimating these impacts using data from the employed population younger than 65 years. Insurers wanted more information to develop realistic assumptions, particularly about changes in utilization and savings. VBID nonparticipants wanted to see how VBID participants “structure their VBID benefit” (NPInsurer10), “how the intervention works” (NPInsurer04), and what outcomes it would achieve (see eAppendix B for additional quotations describing all thematic findings reported here).

Second, 7 VBID nonparticipants cited potential lack of return on investment (ROI): “We just could not come to a positive ROI to where the program would at least cover its own costs in year 1,” said one representative (NPInsurer01). They felt that the implementation and administrative costs of VBID were too high, and they viewed potential returns as relatively low, because many stated that they were already offering high-quality care.

Third, representatives of 7 nonparticipating insurers worried about administering 2 sets of benefits to beneficiaries within the same plan based on the presence of an eligible health condition. In MA, all beneficiaries in a plan, which they select during the annual open enrollment period, get the same benefits regardless of their medical conditions. Under VBID, beneficiaries in the same plan may get different benefits, depending on their diagnoses. Being diagnosed with an eligible condition midyear could trigger a change in benefits. Nonparticipants worried about their IT capabilities and the ability of internal systems to identify, track, and administer VBID benefits. As NPInsurer08 put it, “How do you identify those [VBID-eligible] members and be able to administer those benefits to them specifically and not to the general population or vice versa? [How do you] make sure that we are able to track the claims? [How do we] make sure [the benefits] are administered exactly the way that we submitted in the bid, no more, no less?”

Finally, nonparticipating insurers raised concerns about the model test parameters. For example, some wanted to implement VBID in Chronic Care Special Needs Plans (C-SNPs). Others wanted to offer VBID to beneficiaries with conditions not allowed by the model test or to target a subset of beneficiaries, such as those with early- or late-stage diabetes. Five nonparticipants also worried that VBID marketing restrictions would not allow them to mention their participation in the model or specific VBID benefits in their pre-enrollment materials and outreach activities. As NPInsurer04 explained, “VBID almost looks like something you have to keep a secret for a while and…you can’t really use that to try to attract new members.” Some would have preferred to advertise their participation in VBID to further distinguish themselves from competitors.

Reasons for Joining VBID and Ways to Address Participation Barriers

Participating insurers tended to be willing to take risk and implement interventions that have not been tried before in MA. Five participants considered themselves innovators, willing to experiment with benefit design. Four participants stated that they joined the model test because VBID’s goals were consistent with their own organizational priorities of reducing spending and improving care quality. Finally, 3 participants commented on VBID’s potential to improve beneficiary outcomes by addressing structural barriers to care and increasing beneficiary engagement.

Participants did encounter the barriers described by nonparticipants. Their greater appetite for risk affected the way participants addressed these barriers. To illustrate, VBID participants handled the lack of evidence by reviewing literature on VBID in commercial plans and relying on their best clinical judgment. Many participants wanted to innovate: “We’re very innovative in a lot of the things we do. We try things. Anytime something new comes up, we tend to get involved in those things just because it’s an opportunity one way or another” (PInsurer03). Others considered VBID a useful benefit design experiment: “We’re certainly willing to go down the road of a demonstration to figure out if our hypothesis that by sending members to their specialists more we can reduce their inpatient hospitalization and their high-cost care is true or not” (PInsurer05).

Most VBID participants agreed that VBID cost savings would be minimal and focused on maximizing long-term outcomes, such as decreased hospitalizations and emergency department use, while reducing implementation costs. According to PInsurer03, VBID would yield benefits “if you can really do something that is going to help the population longer term, [such as] better quality of life [or] lower long-term costs, those are all good things.” To increase the chance of a positive ROI, several participants noted that they had designed their interventions to minimize implementation costs: “We needed to come up with something that would not add additional resources and cost to the actual program that we have now” (PInsurer03). Participants relied on existing programs and processes when possible, which helped them design interventions that were easier and less costly to implement.

VBID participants agreed that managing 2 sets of benefits within a plan, or what they called “a plan within a plan” (PInsurer06), was a serious implementation challenge. Managing VBID-eligible beneficiaries required substantial IT investments and extensive coordination across departments because multiple systems, such as claims processing or care management tracking systems, had to interact with each other. A representative of PInsurer07 noted that it had to modify about 15 applications before VBID rollout.

Insurers developed different approaches to tracking eligibility, participation status, and the correct payment amounts for beneficiaries eligible for reduced cost sharing. Some VBID participants “separated” beneficiaries or created different internal groups in their IT systems to flag VBID participants. As PInsurer06 explained, “We duplicated the existing structure of our benefits and made a separate benefit structure…a dedicated line for these members.” PInsurer07 created flags within internal IT systems to identify VBID-eligible and VBID-enrolled beneficiaries.

VBID-participating insurers also considered VBID marketing restrictions to be problematic, citing potential confusion among beneficiaries, many of whom were not notified about their VBID benefits until January 2017, months after receiving Annual Notice of Change and Evidence of Coverage documents that detailed all benefit changes. To address beneficiaries’ confusion, some MA insurers called eligible enrollees in addition to sending them letters describing new VBID benefits in January.

Although VBID participants understood CMS’ rationale for restricting advertising, they still noted that these restrictions may negatively affect beneficiaries’ awareness of and participation in VBID: “I understand the CMS’ concern around selection or cherry-picking…but [if we could market VBID,] we probably would have had more people say, ‘Hey, let me see if I’m eligible’ as opposed for us having to wait for things to hit the system” (PInsurer04).

VBID nonparticipants cited a perceived lack of information on VBID in MA, expectations of low ROI, potential administrative and IT hurdles, and concerns about the test’s design as the main reasons for not joining the MA VBID model. By contrast, participating insurers were interested in experimenting with benefit design, even if the ROI was uncertain; implementing interventions consistent with their organizational priorities; and improving beneficiary outcomes by addressing structural barriers to care and increasing beneficiary engagement. Risk tolerance among upper management and an entrepreneurial organizational culture that encourages innovation seem to differentiate VBID participants and nonparticipants. During the initial implementation period in 2017, most VBID participants encountered the administrative and IT hurdles feared by the nonparticipating insurers but overcame them. They also agreed that certain model test characteristics, such as marketing restrictions, may have limited their abilities to design their preferred interventions or made the implementation challenging.

Based on the feedback from both groups, CMS changed the model parameters for 2018 and 2019

 and relaxed some of the marketing restrictions.

 In particular, CMS added rheumatoid arthritis and dementia to the list of eligible conditions, allowed insurers to propose their own methods for identifying eligible beneficiaries, and made C-SNPs eligible to participate.

 The VBID model test is occurring in a rapidly changing policy environment in which CMS is allowing more flexibility in benefit design through changes to the uniformity rule and the nationwide VBID model expansion. Our results point to 4 important considerations that may affect insurers’ willingness to adopt VBID in or outside the model test:

 Many insurers cited uncertainty about ROI as a key reason for nonparticipation. Generating evidence on VBID’s effects in the MA population may help insurers estimate the impact to their bottom lines and make an informed decision about MA VBID.

Insurers’ philosophy, rather than market characteristics, may influence participation.

Our study participants did not believe that market characteristics affected their decisions to join VBID. Instead, VBID participants welcomed the opportunity to experiment with benefit design to improve beneficiary health outcomes and care quality and viewed ROI as a secondary concern. Willingness to innovate with benefit design and be considered a leader in the MA space were more important for model participants than potential concerns about ROI. Nonparticipants, however, took a “wait-and-see” approach and wanted to avoid the unknown outcomes of VBID in the MA population. In some cases, they felt that they were already providing high-quality care and were reluctant to experiment given uncertainties.

Technological barriers can be significant.

 Many participants had to invest in IT systems to enroll beneficiaries in VBID, track their benefits, and pay the correct amounts to providers. Participants, especially those requiring beneficiaries to engage in care management, also needed to coordinate multiple internal departments that did not previously communicate or work together. Although such issues are not insurmountable, they may deter some insurers from offering VBID benefits until appropriate changes to their IT systems are implemented and tested.

 Nonparticipants indicated that they would be more likely to join the model test if they had even more flexibility to design and target benefits, and both participants and nonparticipants cited CMS’ marketing restrictions as an impediment to participation. CMS has already lowered participation barriers by allowing insurers to target additional conditions, extending eligibility to C-SNPs, and relaxing marketing restrictions. Additional flexibility will be permitted in 2020.

Our qualitative study has 3 limitations. First, insurers not responding to our interview requests may have had different perspectives on VBID than those who responded. This may be a particular concern for nonparticipants, given that only about one-third of nonparticipating insurers responded. To address this issue, we analyzed written comments from MA insurers who responded to CMS’ request for comments on the model test, but we did not see major differences in perspectives with those we interviewed. Second, we conducted interviews 6 to 8 months after the start of the model test, when MA insurers were still working toward finding solutions to some implementation challenges. Subsequent data collection may reveal additional implementation challenges and facilitators. Third, our study relied on self-reported data collected from model participants either by phone or in person and from model nonparticipants only by phone. Some participants may not have disclosed all implementation challenges they might have experienced, and, although unlikely, the mode of data collection might have affected responses.

This article describes perspectives of Medicare Advantage (MA) insurers about participating in the CMS value-based insurance design model test launched in 2017.

Insurers’ Perspectives on MA Value-Based Insurance Design Model

About a third of Medicare beneficiaries are covered by Medicare Advantage (MA) plans or accountable care organizations (ACOs). As a result of assuming financial risk for Medicare services and/or being eligible for shared savings, these organizations have an incentive to adopt models of delivering care that contribute to better care, improved health outcomes, and lower cost. This paper identifies innovative care models across the care continuum for high-cost Medicare beneficiaries that MA plans and ACOs could adopt to improve care while potentially achieving savings. It suggests policy changes that would accelerate testing and spread of promising care delivery model innovations.

Targeted review of the literature to identify care delivery models focused on high-cost or high-risk Medicare beneficiaries.â€¨

This paper presents select delivery models for high-risk Medicare beneficiaries across the care continuum that show promise of yielding better care at lower cost that could be considered for adoption by MA plans and ACOs. Common to these models are elements of the Wagner Chronic Care Model, including practice redesign to incorporate a team approach to care, the inclusion of nonmedical personnel, efforts to promote patient engagement, supporting provider education on innovations,and information systems allowing feedback of information to providers.The goal of these models is to slow the progression to long-term care, reduce health risks, and minimize adverse health impacts, all while achieving savings.These models attempt to maintain the ability of high-risk individuals to live in the home or a community-based setting, thereby avoiding costly institutional care. Identifying and implementing promising care delivery models will become increasingly important in launching successful population health initiatives.

MA plans and ACOs stand to benefit financially from adopting care delivery models for high-risk Medicare beneficia- ries that reduce hospitalization. Spreading these models to other organizations will require provider payment policy changes. Integration of acute and long-term care would further spur adoption of effective strategies for reducing or delaying entry into long-term institutional care.

This paper presents 9 delivery models for high-risk Medicare beneficiaries in different settings that Medicare Advantage (MA) plans and accountable care orga- nizations (ACOs) could adopt to achieve better care at lower costs.â€¨ Common to these models is a team approach to care, including nonmedical personnel, and providing care in home- or community-based settings to avoid costly institutional care.

Our review of the literature included intervention models such as Advanced Primary Care, Home-Based Primary Care, CAPABLE, MIND at Home, PACE, Hospi- tal at Home, Hospital Elder Life Program, Transitional Care, and INTERACT. Policy changes to permit MA plans, ACOs, and service providers to share Medicaid long- term care savings would further accelerate adoption.

About a third of Medicare beneficiaries are now covered by Medicare Advantage (MA) plans or accountablecareorganizations (ACOs).

 These organizations have an incentive to adopt innovations in care delivery that yield better care, improve patient outcomes, and lower costs.MA plans are at financial risk for the total cost of Medicare services, and Medicare ACOs are either at full or partial risk, or eligible for shared savings. Delivering care in a way that reduces the costly use of inpatient hospital care and emergency departments (EDs) helps these organizations to realize savings and gain market share.

Factors that place Medicare beneficiaries at risk for higher acute, post acute, and long-term care utilization and expen- ditures include age, number and type of chronic conditions, functional impairment, income level, and social support system (eg, living alone). Those Medicare beneficiaries with multiple chronic conditions, functional impairment, and low income, and/or who are living alone or in institutions, account for most Medicare expenditures and high rates of ED visits, hospitalizations, readmissions, and nursing home placements.

This paper illustrates how Medicare Advantage plans and accountable care organizations could benefit from adopting innovative care delivery models, and suggests policy changes to accelerate spread.

Christine Buttorff, PhD

Articles

Variation in Generic Dispensing Rates in Medicare Part D

Insurers’ Perspectives on MA Value-Based Insurance Design Model

Innovative Care Models for High-Cost Medicare Beneficiaries: Delivery System and Payment Reform to Accelerate Adoption