Skip to main content

Comment on HHS-operated risk adjustment technical paper on possible model changes

Editor's Note:

This analysis is part of the USC-Brookings Schaeffer Initiative for Health Policy, which is a partnership between Economic Studies at Brookings and the University of Southern California Schaeffer Center for Health Policy & Economics. The Initiative aims to inform the national health care debate with rigorous, evidence-based analysis leading to practical recommendations using the collaborative strengths of USC and Brookings. 

The Center for Consumer Information and Insurance Oversight (CCIIO) at the Centers for Medicare and Medicaid Services recently published a technical paper that lays out potential changes to the risk adjustment program that operates in the individual and small group markets. We recently submitted comments to CCIIO on the proposals outlined in the technical paper.

Timothy Layton

30th Anniversary Associate Professor of Health Care Policy - Department of Health Care Policy, Harvard Medical School

Faculty Research Fellow - National Bureau of Economic Research

The first section of this letter comments on the changes to methods for calculating risk scores that are described in the main text of the technical paper. We make three main points:

  • The transfers generated by the current risk adjustment methodology are most likely too small, not too large: Because health status is only imperfectly observed in risk adjustment, fully mitigating selection incentives is likely to require transferring more from insurers with low-risk enrollees to insurers with high-risk enrollees than is suggested by directly observable differences in risk. But it is doubtful that the existing mixture of under- and over-prediction documented in the technical paper is, on net, producing transfers that meet this standard. Indeed, the large role that narrow network, tightly managed plans currently play in the individual market is consistent with the view that risk adjustment transfers are too small and insurers retain strong incentives to pursue low-risk enrollees.
  • In this environment, changes that increase risk scores for low-risk enrollees are likely to worsen market outcomes by reducing plan quality and, potentially, competition: All else equal, steps that increase the risk scores of low-risk enrollees (or reduce the risk scores of high-risk enrollees) will tend to reduce risk adjustment transfers. Since we believe that transfers are already too small under the existing risk adjustment methodology, we expect that such changes will tend to increase selection incentives, thereby reducing plan quality and, potentially, making it harder to sustain robust plan competition.

Raising risk scores for low-risk enrollees is also unlikely to have the countervailing benefit of increasing enrollment, as posited by the technical paper. Since most enrollees receive the premium tax credit and the tax credit’s value depends on the premiums of the lowest-cost silver plans, changes that reduce the premiums of existing low-cost plans or spur insurers to market lower-cost products aimed at low-risk enrollees would leave net premiums of the lowest-cost plans little changed. Thus, increased enrollment is unlikely; indeed, since the quality of the lowest-cost plans could fall, enrollment could actually fall.

In light of these concerns, we recommend that CCIIO not adopt its proposed two-stage procedure to estimate model coefficients, which the technical paper shows would increase predicted spending for low-risk enrollees and reduce it for high-risk enrollees, thereby exacerbating selection incentives. By contrast, the proposals related to enrollment duration factors appear beneficial along this dimension (and may have benefits along other dimensions). We are uncertain what to recommend with respect to the proposed severity-HCC-count interactions since this change would increase predicted spending for both the lowest- and highest-risk enrollees, and the net effect on selection incentives is thus unclear.

  • Combining steps to improve model fit with changes to the transfer formula could potentially reduce selection incentives and improve market outcomes: The fact that the existing risk score model fits poorly on some dimensions does suggest that there may be opportunities to improve risk adjustment. In particular, improving model fit could allow the model to produce a more accurate rank-ordering of insurers’ risk mixes, even if it also led to smaller, less appropriate transfers. Coupling changes to improve fit with changes to the transfer formula that magnified transfers could make it possible to achieve the benefits of a more accurate risk score model without shrinking already-too-small transfers.

If CCIIO does elect to go down this road, we recommend that it give further consideration to improving model fit via a non-linear risk score model. Adopting a non-linear model has the potential to improve fit along some of the dimensions of interest without sacrificing fit along other dimensions (as the two-stage procedure would) and without creating new coding incentives (as the severity-HCC-count interactions approach would).

We close by commenting on the potential changes related to cost-sharing-reduction (CSR) plan variants that are discussed in the appendix of the technical paper. We make two main points:

  • Modifying risk adjustment to account for “silver loading” is appropriate, but CCIIO should refine its specific proposals: Explicitly incorporating the cost of providing CSRs into risk scores and modifying the transfer formula rating term to account for the higher effective actuarial value (AV) of silver plans would mitigate incentives to avoid CSR enrollees (and higher-risk CSR enrollees in particular). However, rather than continuing to construct CSR-variant risk scores as a multiple of silver-tier risk scores, as the appendix proposes, CCIIO should explore creating separate risk score models for the CSR variants. Additionally, in modifying the silver AV used in the transfer formula rating term, CCIIO should use state average effective silver AVs, not the national average CCIIO is currently considering, as there are large cross-state differences in average effective silver AV arising from state Medicaid expansion decisions and various other factors.
  • It is unclear whether risk scores should incorporate a special adjustment to reflect the distinctive spending patterns of CSR enrollees: The technical paper presents persuasive evidence that CSR enrollees spend less than would be expected based on their observed health status and cost-sharing, and it suggests that it may be appropriate to align CSR enrollees’ risk scores with their observed spending. However, CCIIO’s explanation for CSR enrollees’ unexpectedly low spending—that lower cost-sharing does not cause higher utilization in this population—is implausible. And, importantly, some plausible alternative explanations for this pattern, like that plans are currently doing a poor job serving their low-income enrollees or that low-income enrollees opt for plans with narrower networks or tighter utilization controls, would argue against making a special adjustment to align CSR enrollees’ risk scores with their observed spending.

While we do not currently have a firm view as to whether CCIIO should or should not make such an adjustment, we do believe that determining the right policy response requires better understanding why CSR enrollees exhibit unexpectedly low spending. We note that a final decision would also need to weigh how adjusting risk scores for CSR enrollees would affect premiums of silver plans relative to plans in other metal tiers and the effects those shifts would have on net-of-subsidy premiums and federal outlays.

The full letter is available here.


The Brookings Institution is financed through the support of a diverse array of foundations, corporations, governments, individuals, as well as an endowment. A list of donors can be found in our annual reports published online here. The findings, interpretations, and conclusions in this report are solely those of its author(s) and are not influenced by any donation.

Get daily updates from Brookings