CMS Modifies Risk Adjustment for ACOs

Submitted by jonpearce on Thu, 2011-10-27 12:38

In the draft ACO regulations, CMS established a single ACO risk adjustment factor using historical data from all participants, and did not modify it throughout the participation period.  This policy reflected CMS’ concern that risk scores would increase due to changes in coding rather than actual changes in patient status, which would inappropriately lower the benchmarks and allow additional “savings” to be realized.  Therefore, the risk score of the initial group of members would be applied throughout the participation period regardless of changes in the actual membership, or changes to the health status of those members.

Comments on Draft Rules

This prompted a variety of comments that are addressed in the final rule.  Many commenters shared Singletrack Analytics’ opinion that this approach would discourage ACOs from recruiting the higher-severity who would be the greatest beneficiaries of coordinated care.  The Medicare Payment Advisory Commission (MEDPAC) even suggested that this approach “would create incentives for ACO providers to encourage existing patients who are costly to seek care elsewhere”. Other commenters noted that changes in a patient’s condition, for example discovering a previously-undiagnosed cancer, would not be accommodated.  In addition, CMS itself had noted that about 25% of members in the PGP demonstration project had changed providers each year, which would cause the risk factors to become increasingly unrepresentative of the current population.  One insightful commenter noted that the “ACO’s risk adjustment score should be determined by the population the ACO is actually treating, and should therefore be recalculated each year”, also noting that the “potential for … increased coding intensity {is} outweighed by concerns about creating incentives to avoid complex patients”. 

Revisions to Methodology

CMS appears to have recognized the perspicacity of these arguments because it has significantly revised its approach to risk adjustment, noting “commenters have persuaded us of the need to better account for risk associated with changes in the ACO’s beneficiary population”, although it remains concerned about the effects of upcoding the diagnoses used in the HCC-based risk scoring methodology.  While it hasn’t adopted the approach used by Medicare Advantage plans (which would be this author’s preference), it has taken steps to deal with the issues raised by the commenters.

For newly-assigned beneficiaries, CMS will now annually update the ACO’s risk score to adjust for the severity and case mix.  This will accommodate higher-severity members who may join the ACO (by utilizing its PCPs) to benefit from the care coordination services that the ACO should offer.  This will give the ACO the incentive to recruit such patients.

For continuously-enrolled members, the methodology is somewhat less clear.  CMS will compute HCC risk scores for these members, but will only adjust the members’ health status if the scores decline.  The rule is silent on what happens if the scores increase, so presumably there’s no adjustment.  The only adjustment will be for changes in age; as the population grows older its risk score will increase.

Effects of Revisions

This accommodates several of the problems, while leaving others still in place.  It handles the issue of new higher-severity members who may join the ACO and removes the incentive for the ACO to discourage these members.  However, it does not accommodate the issue of members having previously-undiagnosed conditions.  CMS appears to believe that upcoding will still occur for continuously-enrolled members for whom the ACO is doing the coding, but is willing to accept the coding for new members which would not have been performed by the ACO. 

This strikes us as a reasonable balance.  We understand CMS’ concerns about upcoding, but the HCC model simply incorporates all diagnoses that are fed to it and therefore shouldn’t respond to “gaming” the way that other payment models such as DRG grouping do.  In addition, it’s important to collect all relevant diagnosis data for all patients to maintain a comprehensive picture of their health status, so it should be expected that ACOs would gather a more complete set of diagnoses than individual providers.  Diagnosis data is often incomplete; in a recent data experiment Singletrack Analytics found that only about 20% of individual claims for known diabetics contained a diabetes diagnosis code. Gathering and reporting comprehensive diagnosis information should be viewed as a goal of an ACO, not as a process used to artificially increase payment.

The greatest opportunity for an ACO to succeed is to recruit and manage patients who have chronic conditions that can be managed by the ACO.  Through care coordination the complication rate of these patients can be reduced, which will also reduce their costs.  This is a win-win-win for the patient, the ACO and CMS.  Fortunately CMS has apparently recognized this and has modified the risk adjustment process to allow these opportunities to be realized.

Additional Resources

Singletrack Analytics has prepared several white papers and articles on utilizing risk adjustment in ACO analytics.  For more information see our “Analytics for ACOs” white paper series and our blogs on risk adjustment.