Favorable Selection Ups the Ante on Medicare Advantage Payment Reform

Editor’s Note: This perspective was originally published on Health Affairs Forefront on June 13, 2023.

Since September 2021, an ongoing discussion on Health Affairs Forefront has provided an important forum for topics related to Medicare Advantage (MA). Authors have noted the growth of MA and have expressed sharply divergent views about the extent of MA overpayments and MA extra benefits and care coordination, including the social value associated with more generous MA payments. We (Ginsburg and Lieberman) attempted to sum up the MA debate and identify common ground.

A prominent thread in these Forefront articles has focused on the role of more intense coding of diagnoses in increasing MA plan payments, a question which the Medicare Payment Advisory Commission (MedPAC) directly addressed in its March 2023 Report. Previous overpayment estimates have not captured the effects of favorable selection into MA, which could be even larger than the $27 billion (6 percent) MedPAC estimate of MA overpayments. Our new research on favorable selection compares expenditures of beneficiaries staying in fee-for-service (FFS) Medicare to the expenditures of those switching to MA during annual open enrollment periods during the period from 2006 to 2019 (the last year with available data not distorted by COVID). Based on this research, we have developed estimates of MA overpayment from biased selection that, while complementing MedPAC’s estimate of overpayments from coding and quality bonuses, are substantially larger.

Particularly because of the rapid growth of MA in recent years, MA overpayments pose a serious fiscal challenge for the federal Government. Below, we describe how MA currently operates and analyze available policy alternatives. We highlight the importance of reforming MA plan payments. We sketch out three very different alternatives to the current approach of basing benchmarks on county-level FFS spending, which is no longer viable.

MA dominates FFS in enrollment and growth

In the period from 2006 (when the current bidding system for setting MA rates began) to April 2023, total Medicare beneficiaries with both Part A and B increased by 20.3 million. MA enrollment grew by 24.6 million (373 percent) while FFS beneficiaries decreased by 4.3 million (-13.3 percent). Among beneficiaries with both Part A and Part B (an MA requirement), the 31.2 million participants in private plans as of April 2023 outnumbered the 28.1 million in FFS—52.6 percent of beneficiaries were enrolled in MA, a percentage projected to grow to 69 percent by 2030.

The Centers for Medicare and Medicaid Services (CMS) sets MA payments based on county-level FFS expenditures derived from FFS claims data (but not data from MA). Basing MA rates on FFS averages has become increasingly problematic and vulnerable to biased selection because of coding intensity issues and three additional factors: the decline in FFS enrollment, switching by millions of beneficiaries from FFS to MA, and the extremely skewed distribution of Medicare expenditures. Recognizing that the number of FFS beneficiaries needed to generate statistically stable expenditure estimates increases as the distribution of expenditures become more varied and skewed, Medicare set the minimum risk-pool size for accountable care organizations at 5,000 FFS beneficiaries. (The CMS approach of using multiple years of county-level expenditure data when setting MA rates only partially mitigates these concerns.) The growth of MA and decline in FFS means that in January 2023, 16.5 percent of counties had 1,000 or fewer FFS beneficiaries with both Part A and Part B, 49.4 percent of counties had 3,000 or fewer, and 65.1 percent of counties had 5,000 or fewer.

Medicare overpays MA plans

MedPAC estimates $23 billion of MA overpayments in 2023 arise from the difference between typical coding in FFS versus MA, after incorporating the across-the-board MA coding reduction imposed by CMS. Because risk scores do not affect FFS payments, providers lack incentives to aggressively enter diagnoses into the medical records of FFS patients. In contrast, payments to MA plans are higher for sicker beneficiaries; thus, many but not all MA plans engage physicians to aggressively code their patients’ diagnoses to generate higher risk scores from the CMS risk adjustment model (which is calibrated using only FFS claims). Scholarly research has found even more extensive upcoding than assumed by MedPAC, and the MedPAC overpayment estimate includes $4 billion from bonus payments tied to plan “star ratings”. Others suggest additional factors for MA overpayments (e.g., “induced utilization”), but none of these overpayment estimates include the effects of favorable selection.

The goal of risk adjustment is to neutralize the financial effects arising from biased selection by creating risk scores that convert the expenditures of beneficiaries with different demographics and diagnoses to those of a “standard” beneficiary with a risk score of 1.0. As CMS reported to Congress:

“At the individual level, predicted medical expenditures can be lower or higher than actual medical costs, but at the group level, below-average predicted costs balance out above-average predicted costs…” [emphasis added]

This central tenet underlying the CMS risk adjustment model requires neutral selection to keep the distribution of expenditures at the group level unchanged when FFS beneficiaries switch to MA. Unfortunately, having millions of beneficiaries switch from FFS to MA while FFS enrollment declines makes it unlikely that the expenditure distribution for beneficiaries switching to MA exactly mirrors beneficiaries staying in FFS. The fact that the distribution of expenditures across FFS Medicare beneficiaries is highly skewed means that even modest degrees of biased selection can cause substantial amounts of overpayment. Favorable selection for MA occurs if switchers as a group have below-average expenditures after using risk scores to control for differences in expected spending. Changing FFS expenditures at the group level raises MA rates and increases total spending.

New research findings on favorable selection

Indeed, our new research comparing beneficiaries staying in FFS to FFS beneficiaries moving to MA during the 2006-2019 annual open enrollment periods found strongly favorable selection. Using the FFS claims data to compare expenditures of stayers versus switchers overcomes the incompleteness and other limitations in MA data, which effectively preclude comparing the health status and expenditures of beneficiaries in FFS and MA. The analysis reflects FFS claims data for a total of 402 million beneficiary-years from 2006 through 2019. Of the 16.9 million switchers during 2006-2019, 11.3 million remained in MA plans in 2020, comprising 46.9 percent of MA enrollment.

We assigned a risk score to each FFS beneficiary for each year based on the most current hierarchical condition categories (HCCs) model with CMS-provided software that accepts both ICD-9 and ICD-10 diagnostic codes (2016 v21), generating 183 different risk scores; expenditures were also converted to 2019 dollars. Even after grouping together beneficiaries with the same risk score in a year, expenditures varied widely and the distribution remained highly skewed. To measure the expenditures of individual beneficiaries with the same risk score in a year required arraying beneficiary expenditures from lowest to highest, assigning expenditures to percentiles, calculating the average expenditure for each percentile, and computing the ratio of each percentile’s average relative to the mean for that risk score. The resulting percentiles enabled quantifying how much above or below average each FFS beneficiary’s expenditures were for their risk score.

A logistic regression model analyzing the probability of switching to MA based on risk-score-adjusted expenditures found FFS beneficiaries with low expenditures are significantly more likely to switch to MA while beneficiaries with high expenditures are less likely to switch, findings consistent with descriptive results for expenditures and other characteristics of switchers versus stayers for each year from 2006 through 2019. On a risk-score-adjusted basis, annual expenditures for the 7.1 million MA beneficiaries who had switched from FFS in 2015-2019 averaged $5,409—half of the $10,865 average for all 29.0 million FFS beneficiaries in 2019. Removing the 2019 switchers from the FFS population increased the average spending for stayers by $91 (0.8 percent), to $10,956.

Out of 16.9 million beneficiaries switching from FFS to MA in 14 annual cohorts, 11.3 million switchers remained enrolled in MA in 2020. However, the most recent five annual cohorts (2015—2019) of switchers accounted for 7.1 MA beneficiaries in 2020 (29.5 percent of enrollment). Each annual cohort of switchers cumulatively influences the total impact of favorable selection, although below-average risk-score adjusted expenditures are likely to tend to move closer to average over time—regress to the mean.

As an order-of-magnitude estimate, overpayments from favorable selection equaled 14.4 percent of payments to MA plans in 2020; an alternative assumption of a quicker “regression to the mean”—the process by which an enrollee’s above- or below-average spending (adjusted for risk score) in one year often returns to the mean in subsequent years—lowers the estimated overpayment to 13.2 percent. The 14.4 percent overpayment estimate implies $40.9 billion in 2020 MA overpayments from favorable selection.

Applying the growth in MA spending projected by the Medicare Actuary increases the estimated overpayment in 2023 to $59.3 billion. Combining MA overpayments from favorable selection with the MedPAC estimate of overpayments from other factors suggests excessive payments to plans total about 20 percent, or $75 billion, per year. Importantly, the estimate of favorable selection reflects significant limitations and simplifying assumptions, making it an order-of-magnitude approximation rather than a Congressional Budget Office-style cost estimate.

Identifying substantial favorable selection into MA does not shed light on the factors behind it. Favorable selection could be driven mostly by individual beneficiaries choosing which model is most suited to them given their preferences and medical conditions. Or it could be driven mostly by actions by plans, some of which are designed to improve care, that attract relatively low-spending enrollees. Those beneficiaries used to substantial use of care, some of it highly specialized, to treat chronic illness may put greater emphasis on the broader provider networks in FFS than in some MA plans, as well as the much more limited use in FFS of utilization management tools such as prior authorization. Health plans’ substantial investment in primary care, which may accomplish a lot in the way of improving care as well as saving money, may be more appealing to relatively healthy beneficiaries than to those accustomed to needing care from many subspecialists.

Reforming MA rates to improve competition, address inequities and lower overpayments

Policies to improve the accuracy of MA rate setting could follow either of two fundamentally different directions: reforming administered payments or relying on competitive bidding.

Improving administered payments

The former option could either sever the link between FFS expenditures and MA or refine how FFS expenditures link to MA rates. For example, CMS could use its relatively open-ended authority to substantially increase the across-the-board coding reduction applied to all MA plans; over time, rates would be updated based on policy and budgetary considerations without regard to maintaining parity with FFS.

Alternatively, significantly correcting the relationship between FFS and MA might be achieved by making MA encounter data comparable to FFS claims data. This would involve mandating a major effort by MA plans to markedly improve the accuracy, completeness, and comparability of their data to address both coding and selection effects. Unfortunately, generating MA data comparable to FFS claims data would require significant behavioral change and investments on the part of the plans; differences in coding among MA plans and between FFS and MA would also have to be addressed. Differences in practice patterns between FFS and managed care—such as MA’s greater reliance on primary rather than specialist care or on enhanced skilled nursing facilities to minimize inpatient hospital care—would pose additional challenges.

Competitive bidding

A starkly different strategy would replace administered pricing with competitive bidding. If both MA and FFS were included, premiums would be set based on bids submitted by MA plans; average FFS expenditures would function as a notional “bid” for FFS. Even if risk-adjusted bids somehow succeeded in addressing coding differentials and favorable selection, significantly disrupting the FFS system relied on by tens of millions of seniors would effectively undermine the political viability of such “premium support” approaches.

An alternative approach would restrict competitive bidding to setting payment rates for MA, using market forces to determine what Medicare pays MA plans. To the degree that MA plans are more efficient, such competition would lead to some of the savings from this efficiency being captured by taxpayers instead of going to extra benefits for enrollees and to MA plan profits. Insurers would strenuously oppose competitive bidding unless the likely alternatives were more unpalatable. If competitive bidding were seriously considered, key questions would involve the relative generosity of MA and FFS benefits and transition rules and timing.

Fiscal considerations—such as massively overpayments to MA plans, the looming insolvency of the Hospital Insurance trust fund, and Congressional pressures to find savings—are likely to increasingly shape the policy environment. On the other hand, programmatic concerns arising from favorable selection, upcoding, and rates-setting based on increasingly problematic county-level FFS expenditures as FFS enrollment continues to erode will shape both reform proposals and industry preferences.

Authors’ Note

Paul B. Ginsburg served as a Commissioner and Vice Chair of MedPAC from 2016 to 2022. The Commission has taken a number of policy positions during that time period relevant to the content of this post.

Lieberman, S. M., Ginsburg, P. B., & Valdez, S. (2023). Favorable Selection Ups The Ante On Medicare Advantage Payment Reform. Health Affairs Forefront.

10.1377/forefront.20230606.520135

Copyright © [2023] Health Affairs by Project HOPE – The People-to-People Health Foundation, Inc.