Introduction
In our Part I exploration of Risk Adjustment Validation under Medicare Advantage (MA), we looked at how payments are risk adjusted to consider the varying health needs of people that enroll in MA plans. Specifically, we discussed how this process helps ensure that MA plans have enough resources to manage the health of all of their members, including those with multiple chronic conditions that may require more intensive treatments and care coordination. In Part II, we will focus more on how the audit sample is developed and how that sample is used by CMS to adjust payments to MA plans.
How is the Sample for RADV Determined?
The RADV process, including the sample determination, has evolved over time. However, one feature that has remained consistent is the reliance on the manual review of medical records, as discussed in Part I. Given the level of resources required, CMS has evolved the approach to target RADV audits on members that are most likely to have diagnoses that are not supported by medical documentation, as well as diagnoses associated with the largest amounts of MA payments.
There are two types of RADV audits; contract-level audits and a national audit. The contract-level audits can result in payment recoveries when diagnoses are not supported by valid medical record documentation, requiring MA plans to pay back some of the money received from CMS. This is because MA plans are paid on a prospective basis (prior to when the audits are conducted). In the course of these audits, however, CMS does not look for underreporting of diagnosis codes, so the result is only ever payment recoveries to CMS.
There are Two Distinct Samples Used for Contract-level Audits — Tier 1 & Tier 2
Tier 1
Tier 1 samples are based on MA members that are expected to have the highest probability for a payment error due to unsupported diagnosis codes. CMS uses statistical modeling techniques to predict the probability of a payment error for every member in the MA program (about 20 million members as of 2020). The modeling takes into account various attributes related to MA payments, including patient demographic factors, MA plan characteristics, and regional characteristics.
The top 300 members with the highest probabilities are selected for audit. All of the contracts (a MA plan may have multiple MA contracts with CMS) in which these members are enrolled are notified about the audit.
Tier 2
Tier 2 samples are limited to MA members with the top 10% highest probabilities (about 2 million MA members) for a payment error due to unsupported diagnosis codes. The same statistical modeling used for Tier 1 is used to determine the payment error probabilities for Tier 2. The audit of this sample is then focused on a condition that affects a high proportion of MA payments. For example, diabetes is the condition that CMS focused on for contract year 2014, which was the first year that the current RADV sampling approach was applied.
The Tier 2 contract-level selection process begins by CMS choosing a member with the highest probability of a payment error and who has the target condition (e.g., diabetes). CMS then identifies the MA contract for that member and randomly chooses up to 29 additional members enrolled in that contract (for a total of 30) who have the target condition and are also in the top 10% for predicted payment error. This process is repeated, beginning with the selection of another member from another contract and expanding to a total of 30 from that contract who satisfy all the conditions, until 6,000 total MA enrollees have been selected for audit.
For the Tier 2 contract level sample, CMS changes the target condition each year in order to avoid creating biases in coding practices.
In addition to the contract level samples, CMS audits a national random sample of 600 members in what is known as the national audit. This sample is for benchmarking purposes and does not result in any payment recoveries.
How are the Samples Used to Recover Payments Under RADV?
The audits of both contract-level samples described above typically result in payment recoveries. In the case of the Tier 1 audit, CMS would seek to recover payments for any relevant diagnoses that they deem are not supported by medical record review. Any payment recoveries would be limited to the audited sample.
The Tier 2 audit can also result in payment recoveries; however, in this case, the results are then extrapolated to a broader population of the MA contract. This extrapolation is limited to the enrollees in the MA contract who meet the Tier 2 threshold. For example, if a contract-level Tier 2 audit uncovered a 5% average overpayment, that 5% would be applied to all members enrolled in the MA contract with the target condition who are in the top 10% for predicted payment error. CMS would then pursue payment recoveries from each of the MA plans implicated by this extrapolation.
How can Advance Tools Help Improve this Process?
Because it would be time-and-cost-prohibitive to manually review the medical records of the entire MA population, extrapolation is a necessary component of the payment recovery process under traditional RADV. However, it’s easy to see why this is often a point of contention between the MA plans and CMS. Extrapolation makes assumptions about the wider MA population based on an audit of just 6,000 enrollees. If by chance the degree of payment error is overrepresented in the selected sample, justified payments for the rest of this population may end up subject to recovery.
RaLytics has developed approaches that alleviates this concern by making extrapolation no longer necessary. Our underlying policy-driven technology is capable of reviewing thousands of medical records in the time it would take one human coder to review a single document. This allows for a plan’s entire enrollee population to be reviewed, eliminating the need to extrapolate from a smaller sample and providing greater financial certainty to the plans.
Additional Information
Comparison of Tier 1 & Tier 2 Contract-level Samples for RADV
Tier 1 | Tier 2 | |
Target Population |
|
|
Sample Size |
|
|
Extrapolation Used for Payment Recovery |
|
|