WHITE PAPER:
Harvard-Inovalon Medicare Study: The Importance of Plan Design in Medicare Advantage
Summary
- Medicare Advantage (MA) plans can vary substantially from one another in terms of their coverage design and patient out-of-pocket cost exposure, such as in terms of premiums, co-insurance, co-pays, and deductibles. However, it’s not well understood how different plan design features affect who enrolls in a given MA plan, whether and how certain features may address socioeconomic-related health disparities, or how they impact a plan’s healthcare utilization. This lack of understanding hinders carriers’ efforts to optimize MA plan designs, and policymakers’ efforts to optimize the overall makeup of the MA program.
- Harvard and Inovalon have partnered to shed light on these critical questions, using Inovalon’s large and nationally representative data assets. Inovalon’s claims data cover a broad range of MA plans and capture plan members’ clinical diagnoses, healthcare utilization, and quality outcomes. These data track outcomes both before and after members’ enrollment in Medicare. The data’s scope enables us to account for any pre-existing differences between plans’ members, and thus measure the impact of plan features and plan performance after adjusting for who enrolls in each plan.
- Examining the impact of financial characteristics of MA plans, we find that zero-premium plans attract a more socioeconomically disadvantaged population than higher-premium plans. This underscores how MA plans’ appeal to disadvantaged populations stems partly from the lower out-of-pocket costs that patients in these plans experience.
- Examining the impact of plan coverage design, we specifically compare plans employing more versus less managed care: Health Maintenance Organizations (HMOs) vs. Preferred Provider Organizations (PPOs). We find that HMO plans have over 25% lower utilization than PPOs for the same population, indicating that managed care provisions can be a highly impactful plan design element and can substantially reduce costs.
- This work is intended to support health plans and policymakers through valuable analytics and new insights. Specifically, this research provides the foundational capabilities to help plans better understand the different plan types and features they can offer, and hone in on an optimal set of offerings. This research can be leveraged to help plans benchmark performance at a high level of granularity to maximize actionability, including in terms of specific markets, subpopulations, and design elements such as premiums.
Introduction
For the past 40 years, Medicare beneficiaries have been able to freely choose between two different types of insurance coverage: publicly administered coverage through traditional Medicare Fee-for-Service (FFS), and privately-administered coverage through a managed Medicare Advantage (MA) plan. Along with dramatic MA enrollment growth in past decades, from 17% of Medicare beneficiaries in 2000 to over 50% in 2023,1 there has been an expansion in the types of MA plans available to beneficiaries. The growing options have come partly through government policy, such as the Medicare Modernization Act, which has given MA insurers greater latitude in how they design their plans and the types of supplemental benefits they can offer.1 For example, MA insurers can offer an increasing number of supplemental benefits that are not covered under traditional Medicare, such as dental, hearing, vision, fitness, and transportation.2 Additionally, MA insurers have flexibility over the financial and coverage design of every benefit plan they offer, including premium amounts charged to patients, copay and cost-sharing rates for different services, as well as adoption of different managed care provisions.3
Despite the increasing size of the MA market, and the increasing variety of benefit plan types that MA insurers can and do offer, there’s limited data on how outcomes vary under different types of plans. Specifically, there is limited understanding of how specific plan features, such as plans’ out-of-pocket costs and coverage design, impact who ends up enrolling in those plans and the healthcare utilization of those who do enroll in specific plan types.
Inovalon and Harvard researchers have partnered to fill this information gap around the impact of different plan design features and different plan types in service to key stakeholders who could benefit substantially from that information, including Medicare beneficiaries, policymakers, and especially MA plans. This joint effort is powered by Inovalon’s unique and highly comprehensive MA data assets, which cover a large and broad sample of MA plans, comprising around 30% of all MA lives at any given point in time. Inovalon and Harvard’s efforts are also supported by Inovalon’s other non-MA data assets, including its large private claims dataset covering about 30% of U.S. under-65 commercial lives, and its Centers for Medicare & Medicaid Services (CMS)-sourced FFS Medicare data accessed through a research Data Use Agreement.
In combination, Inovalon’s data assets represent the largest longitudinal claims data source in the U.S., surpassing the data assets of even the federal government and CMS. This data provides the Harvard and Inovalon research team with unprecedented visibility into healthcare outcomes within the MA space, as well as across American healthcare broadly. The breadth of Inovalon’s data enables detailed comparison across different MA carriers, plans, and specific plan benefit package designs. This ability is unique, given that MA data is typically limited to just a single carrier or a narrow slice of the MA market. Another unique feature of Inovalon’s data is that it follows each individual across different types of insurance coverage, giving visibility not just on outcomes under different MA plans, but also on that same person’s outcomes before they enrolled in the specific MA plan or traditional Medicare.
This white paper is the fourth in a series authored by Harvard and Inovalon researchers. The preceding white papers compared who enrolled in MA vs. FFS, finding that MA attracted a socioeconomically disadvantaged population. The studies additionally compared post-enrollment healthcare resource utilization and quality outcomes under MA vs. FFS, finding significant utilization reductions and quality improvements under MA. This paper builds on the previous studies, which considered Medicare as a whole, by instead looking at differences within the MA program. This research examines whether certain plan types attract an especially disadvantaged population and whether certain plan types are especially effective at reducing healthcare utilization.
This study is meant to correct a common misconception — that MA is a monolith — by focusing on the differences that exist across MA plans. The study is also meant to fill an informational blind spot for health plans themselves, to provide some insights into how plan design choices impact enrollment and utilization outcomes. We look at two key dimensions along which MA plans can differ: financial generosity and non-financial coverage design. We examine the impact of these characteristics on enrollment composition and utilization. In this study, we limit our focus to select key plan design features, but our research framework can be used in the future to look at a much broader set of features.
First, we study plans’ financial generosity, in terms of how much they charge in patient premiums. While premiums are not the largest category of out-of-pocket costs facing patients, they are likely to be the most salient. We compare MA plans that do not charge any premium whatsoever to plans that do charge premiums. Next, we look at plans’ coverage design, in terms of whether they adopt an HMO or PPO design. HMOs enroll roughly 60% of MA beneficiaries, and employ more stringent managed care provisions than PPOs, which enroll the remaining 40%.3 When comparing plans to one another in terms of certain outcomes, we employ various statistical techniques to adjust for differences in enrollment composition across plans, so that our estimates only reflect the impacts of the plans themselves.
The few past studies on this subject have been restricted to narrow sets of measures and limited sets of plans. Inovalon’s data enables us to conduct a much more comprehensive and credible examination, thanks to the data’s coverage of a broad spectrum of plans, for a large and nationally representative population. Even more problematically, past work has been unable to fully adjust for enrollment differences across plans when comparing key outcomes such as utilization (for example, adjusting for utilization differences that arise from plans enrolling populations that are healthier vs. sicker ex-ante). We overcome this perennial challenge by leveraging the availability of pre-65 commercial data on beneficiaries, from prior to their enrollment in different MA plans. This data effectively lets us net out any differences that already existed when beneficiaries first enrolled in MA plans, and hone in on what’s attributable to the plans themselves.
Objective
In this white paper, we look at plan design differences within MA, and the impact of specific plan design features on enrollment composition and utilization. Looking first at plan financial generosity, we examine whether more financially generous, zero-premium MA plans attract a more socioeconomically disadvantaged population. Looking second at plan coverage design, we examine whether HMO plans achieve greater utilization reductions than PPO plans.
Data, Sample, and Methods
For this study, we leverage Inovalon’s Medical Outcomes Research for Effectiveness and Economics (MORE2) Registry® dataset, which tracks demographic characteristics and outcomes for about 30% of all MA members at any given point in time. Inovalon’s data is sourced from a large number of different MA insurers and covers a diverse and nationally representative set of different MA plans. These data include medical and pharmacy claims for MA enrollees, along with enrollment and demographic files. To supplement this MA data, we also employ data tracking beneficiaries just before they joined Medicare, while they were still under commercial coverage. This data is also sourced from Inovalon’s MORE2 Registry dataset and includes medical and pharmacy claims and enrollment files. The MA and commercial data are all longitudinally linked, making it possible to follow the same individual as they transition from commercial to MA coverage at 65.
We supplement Inovalon’s MORE2 Registry data with two additional datasets, which provide additional enrollment, demographic, and socioeconomic information on patients in our sample. First, we use Medicare enrollment and claims files from CMS, which track detailed information on patients’ enrollment status in different parts of Medicare. These data also are comprised of claims files, tracking utilization, and outcomes for beneficiaries in traditional Medicare FFS. Next, we source data on social drivers of health (SDOH) from Inovalon’s SDOH Data Warehouse, which tracks SDOH at a highly granular, 9-digit ZIP code level, corresponding to approximately 30 million “near neighborhood” geographic areas. We link these different datasets to Inovalon’s MORE2 Registry dataset using patient-specific identifiers. We then de-identify the combined dataset using the HIPAA Safe Harbor method. The Medicare FFS data was linked to SDOH at the 5-digit ZIP level because CMS provides only city and state addresses and not street addresses.
Using these datasets, we construct a sample of beneficiaries who turned age 65 between 2015 and 2019, and who subsequently entered MA. We limit the sample to those who are fully enrolled in the Medicare program (i.e., for FFS in Parts A, B, and D or MA in a Part C plan with Part D coverage) within three months of turning 65. In addition, we focus on beneficiaries who ended up in MA within one month of Medicare enrollment and who then remained in MA for the 24 months thereafter; implicitly, we restrict only to those enrolled in MA plans tracked in the MORE2 Registry. We excluded those simultaneously enrolled (i.e., “dual eligible”) in a Medicaid or a commercial plan during the 24 months after enrolling in Medicare. We further limited to those enrolled in an employer-sponsored health insurance plan tracked in the MORE2 Registry for all 24 months before turning 65. This guaranteed that we had enough time to establish a baseline for sample beneficiaries.
Using these datasets, first we construct a number of demographic and socioeconomic measures, such as income and net worth. For these continuous characteristics, we group beneficiaries into discrete cells based on their values. In addition, we generate utilization measures in terms of total healthcare expenditures across the full range of care settings. This expenditure measure is constructed following a price normalization wherein we measure the amount spent on any given service at the reimbursement rate paid by Medicare FFS. Therefore, our expenditure measure only reflects utilization intensity and is unaffected by unit price differences across plans or insurance market segments.
When comparing utilization and other outcomes across plans, we use a combination of our unique data assets and statistical tools to control for underlying enrollment differences across plans. While past studies have been unable to control for such differences effectively, doing so is a critical prerequisite for generating useful and unbiased measures of plan performance. Specifically, we employ a difference-in-differences research design, comparing groups who received different treatments (for example, those in MA HMOs versus MA PPOs), both after the treatment is applied (i.e., after Medicare enrollment at age 65) and before, when the treatment had not yet been applied (i.e., before the age of 65).
The difference-in-differences approach uses utilization outcomes, when all beneficiaries are 64 and enrolled in employer-sponsored insurance, to estimate how large the pre-existing differences are between beneficiaries who enroll in FFS and those who enroll in different MA plans. At age 64, differences between beneficiaries only reflect pre-existing concerns rather than the effects of the MA plan they ultimately enroll in. Any differences between beneficiaries at age 65 after enrollment in Medicare net of differences at age 64 before enrollment in Medicare reflect the consequences of their selected MA plan rather than pre-existing concerns. Statistically, we estimate these pre-existing differences using person-level fixed effects in our primary regression specification (which are meant to adjust for each individual person’s baseline characteristics and outcomes pre-65).
Results
Focusing first on how MA plans differ in terms of financial generosity, we look to a specific and highly visible plan attribute: whether the MA plan charges a premium. We compare MA plans that do not charge a premium to those that do.
Demographically, MA enrollees in zero-premium plans differ meaningfully from other MA enrollees, as well as those in FFS (Table 1). Compared to the other two groups, those in zero-premium MA plans are three times as likely to be non-white (i.e., much more likely to be Black, Hispanic, or Asian). Enrollees in zero-premium plans are also more likely to live in an urban area and much less likely to live in a rural area.
Our analysis also documents substantial differences in health-relevant socioeconomic characteristics, with zero-premium MA plan enrollees substantially more socioeconomically disadvantaged compared to enrollees in other MA plans, and even more so when compared to enrollees in FFS.
The average income of a zero-premium MA enrollee (based on their corresponding ZIP9) is $75,580 compared to $79,981 for other MA enrollees and $86,430 for those in FFS. This gap arises in part from the relative lack of MA enrollees in the most affluent segments: while 37.1% of FFS enrollees live in an area with incomes above $100,000, this is true for only 26.6% of non-zero-premium MA enrollees and 21.8% of zero-premium enrollees. The average enrollee in a zero-premium MA plan has a net worth that is 90% of that of other MA enrollees and only 70% of the net worth of FFS enrollees.
Those in zero-premium MA plans are more likely to face many other socioeconomic disadvantages in contrast to their counterparts (Table 1). They are more likely to have only a high school education or less. Compared to those in other MA plans as well as in FFS, they are less likely to own their home, to be married, to own a vehicle, and more likely to have difficulty speaking English. Finally, they appear more disadvantaged based on an aggregate measure of social risk, the Socioeconomic Status (SES) index.4
Next, we focus on coverage design differences across MA plans. Specifically, we compare HMO plans that employ relatively more stringent managed care provisions to PPO plans that employ relatively less stringent policies (Table 2). We find that those in MA HMO plans are three times as likely to be non-white as those in MA PPO plans. We also find that those living in the Northeast make up a substantially larger share of those in HMO as opposed to PPO plans, while those in the Midwest are much more likely to be in a PPO vs. HMO plan.
Looking next at a broad set of socioeconomic measures, we find that those in MA HMO plans face greater socioeconomic disadvantage than those in MA PPOs. However, the socioeconomic disparity between the two is not as great as between zero-premium and non-zero-premium MA enrollees. As shown in Table 2, MA HMO enrollees live in areas with 7% lower income and 4% lower net worth than their PPO counterparts. The socioeconomic disparity between MA HMO and PPO beneficiaries is also less pronounced than between FFS and MA beneficiaries, generally.
Perhaps the most striking difference between those in MA HMOs and PPOs is in terms of the type of coverage they were under, pre-65 (Table 3). Compared to MA PPO enrollees, those in MA HMOs were three times more likely to have been in a commercial HMO plan pre-65 (recall that 100% of both groups were in commercial coverage, pre-65). Conversely, those in MA PPOs were almost twice as likely to have been in PPOs pre-65 as those in MA HMOs. This suggests that beneficiaries prefer to enroll in MA plan types that mirror their coverage type pre-65. As such, beneficiaries with more prior exposure to managed care (possibly reflecting their prior preferences for managed care plans, or prior acclimation to managed care restrictions) also prefer MA plans with greater use of managed care tools such as care coordination.
Finally, we look at differences in utilization levels between MA HMO and PPO plans, following enrollment. In Figure 1, we plot utilization over time, for a matched set of beneficiaries in FFS, MA HMO, and MA PPO plans. We see a substantial jump in utilization for the FFS and MA PPO cohorts at the time of Medicare enrollment, while utilization for the MA HMO cohort remains relatively unchanged. Utilization levels for the FFS cohort see a continuous increase over the two years following Medicare enrollment, leading to a growing gap between this cohort and the MA HMO group. At the same time, utilization of the MA PPO cohort begins to fall starting one year following Medicare enrollment, with the gap between the MA HMO and PPO groups eventually narrowing. This suggests that MA PPO plans might have a more gradual impact on utilization than MA HMO plans, whose impact appears to be instantaneous. As of the end of our sample period, the gap between FFS and MA HMOs is about twice as big as with MA PPOs, suggesting that MA HMOs may reduce utilization by twice as much over the medium term.
Finally, we employ regression analyses to precisely quantify the difference in utilization between MA HMO and MA PPO plans. These statistical analyses allow us to further adjust for differences in who enrolls in MA HMOs vs. PPOs, going beyond the cohort matching in Figure 1. These analyses specifically capture the impact of MA HMOs compared to an identical population enrolled in MA PPOs. Our regression results shown in Table 4 indicate that post-65 utilization under MA HMOs is 29% lower relative to a comparable population in MA PPOs, a result that is statistically significant at a 1% level; in absolute terms, this equates to $615 lower utilization on a person-quarter basis under HMOs than PPOs (in terms of price-normalized spend), given a post-65 FFS baseline of $2,157. For these results, we incorporate the full suite of controls, including individual fixed effects (to adjust for each beneficiary’s pre-65 utilization), and match-group-by-time–trend controls (to adjust for potential differential trends between match groups).
Discussion
While MA is mistakenly viewed as a monolith, individual MA plans differ markedly in terms of their financial generosity, coverage design, and other features. Our study tries to understand what this variety in plan design means, specifically in terms of a given plan design’s impact on enrollment composition as well as post-enrollment outcomes. The study leverages Inovalon’s unique MA claims data, which cover a wide variety of different MA plans across different markets and regions, enabling us to make a broad comparison. Inovalon’s data also allows us to effectively adjust for plan-specific enrollment mix when measuring the impact of different MA plans.
We find that the differences that exist between MA plans on paper, in terms of plan design, are highly consequential in practice. Looking first at plans’ financial characteristics, we find that whether or not a plan charges a premium substantially impacts who enrolls in it, with zero-premium plans attracting enrollees that are more socioeconomically disadvantaged than enrollees in MA broadly. Looking separately at plans’ coverage design, in terms of HMO/PPO status, we find that those in MA HMO plans are much more likely to have been in a commercial HMO pre-65 and those in MA PPOs are much more likely to have been in a commercial PPO pre-65. Altogether, beneficiaries appear to gravitate towards MA plans with similar coverage design to what they were exposed to pre-Medicare. Finally, we find that MA HMO plans produce substantially larger utilization reductions than MA PPO plans, even after adjusting for differences in who enrolls in the different plans and for differences in their baseline healthcare utilization before Medicare enrollment.
Our study findings suggest that the breadth of different MA plan types available to beneficiaries is a key part of MA’s appeal, and what allows MA to attract such a large proportion — more than 50% in 20231 — of all Medicare beneficiaries. Additionally, our study suggests that MA’s relative financial generosity is a key part of its appeal to socioeconomically disadvantaged populations given their larger representation in MA. Our second white paper, “Utilization and Efficiency Under Medicare Advantage vs. Medicare Fee-for-Service,” found that MA has substantially lower utilization and expenditures than FFS, even after adjusting for member enrollment differences across the two programs. This analysis demonstrates that the managed care provisions in MA HMOs are a key driver of the utilization reductions that MA plans achieve relative to Medicare FFS.
Altogether, our findings show that specific plan design features materially shape individual plan outcomes and that the makeup of plans under MA shapes overall program outcomes. Plan design is thereby highly consequential at both an insurer and a broader program level, warranting close attention from health plans as well as policymakers.
Medicare Advantage (MA) attracted
more than 50%
of all Medicare beneficiaries in 20231
How Can This Research Help Health Plans?
Our study and the analytics underpinning it can help MA plans navigate the different MA coverage design options they are allowed to adopt. This research can be tailored to specific health plans to help them understand the expected performance of different plan types and features, for both current and future Medicare age-ins based on the characteristics of the near 65 population in specific geographies.
This research can help plans project expected enrollment in hypothetical offerings, as well as project post-enrollment outcomes under the various plan types and features. Our analytics can help insurers predict the number and types of beneficiaries that would enroll in different potential offerings, allowing plans to identify the set of offerings that best appeals to their target markets.
Furthermore, our analytics can help MA plans predict utilization and quality under different designs, allowing plans to identify the set of offerings that would deliver the best overall quality and cost outcomes and appropriately plan for the resources needed to care for the new enrollees such as types of specialty providers; availability of hospitals, post-acute care, and other needed services; and the social risk factors in the target population that will need to be addressed to assure access to care and achieve good outcomes and health equity.
All these analytics can be delivered at a highly granular level, quantifying the impact of very specific plan features such as premium levels or HMO vs. PPO coverage design for specific geographies and subpopulations. These analytics can also help insurers benchmark the performance of their different offerings against one another and against competitors. As such, our analytics can offer health plans highly actionable insights in support of key strategy, sales, and product efforts.
How Can This Research Help Policymakers?
For policymakers, our study demonstrates that MA is not a monolith, and highlights the variety of offerings encompassed by MA. The study additionally highlights how different plan types end up attracting very different types of beneficiaries, while also achieving different outcomes. The study demonstrates that a lot of the value of the MA program comes from the broad variety of different plan types that it encompasses, as this allows it to appeal to a broader share of the Medicare population, especially the socioeconomically disadvantaged.
We show that different types of MA plans can support different worthwhile policy outcomes such as achieving the best outcomes possible for every Medicare enrollee under the most efficient and cost-effective models of care. Hence, by enabling and incentivizing the right mix of MA plan offerings, policymakers can maximize benefits to the Medicare program and potentially lower the overall costs to maintain the financial viability of the program.
The study highlights how different plan types end up attracting very different types of beneficiaries, while also achieving different outcomes.
Table 1. Demographic and Socioeconomic Factors by MA Zero-Premium, MA Non-Zero Premium, and FFS Enrollee Status
Table 2. Demographic and Socioeconomic Factors by Medicare Plan Type
Table 3. Enrollee Health Insurance Characteristics by Medicare Plan Type
Insurance Type Pre-65
Figure 1. Raw Time Series of Utilization by Medicare Advantage Plan Characteristics: HMO vs. PPO
Table 4. Regression of Total Healthcare Utilization on Medicare Advantage Enrollment: HMO vs. PPO
1 “The Evolution of Private Plans in Medicare,” Yash M. Patel, Stuart Guterman, The Commonwealth Fund, December 8, 2017, https://www.commonwealthfund.org/publications/issue-briefs/2017/dec/evolution-private-plans-medicare
2 “Extra Benefits Offered by Medicare Advantage Firms Vary,” Jeannie Fuglesten Biniek, Nolan Sroczynski, Meredith Freed, and Tricia Neuman, KFF, November 16, 2022, https://www.kff.org/medicare/issue-brief/extra-benefits-offered-by-medicare-advantage-firms-varies/
3 “Medicare Advantage in 2023: Premiums, Out-of-Pocket Limits, Cost Sharing, Supplemental Benefits, Prior Authorization, and Star Ratings,” Nancy Ochieng, Jeannie Fuglesten Biniek, Meredith Freed, Anthony Damico, and Tricia Neuman, KFF, August 9, 2023, https://www.kff.org/medicare/issue-brief/medicare-advantage-in-2023-premiums-out-of-pocket-limits-cost-sharing-supplemental-benefits-prior-authorization-and-star-ratings/
4 “Chapter 3: Creation of New Race-Ethnicity Codes and SES Indicators for Medicare Beneficiaries,” Agency for Healthcare Research and Quality (AHRQ), January 2008, https://archive.ahrq.gov/research/findings/final-reports/medicareindicators/medicareindicators3.html
