On February 1, 2023, the Centers for Medicare and Medicaid Services (CMS) released the 2024 Advance Notice detailing plans to transition to a new risk adjustment model in payment year 2024. This model brings significant changes to Medicare reimbursements for coverage in the private sector. With the additional release of average 2021 risk scores to health plans, it became clear that a wide range of impacts were possible across plans, with most unfavorable and some experiencing headwinds of over 20%. A consensus developed that a nationwide average impact of full model transition was probably in the range of -1.5% to -2.5%, with dual plans experiencing an unfavorable impact two to three times worse. This leads to a question of great interest: Are these impacts appropriate, or can provider education and more accurate coding make a big difference?

Key changes in the 2024 rate announcement

In the calendar year 2024 rate announcement, CMS both finalized the transition to this model over three years, distributing the impact over this period, and released 50 pages characterizing the model updates.

The main changes to the model were:

1. Much more recent data was used to determine the coefficients. The 2020 model was calibrated on ICD-9 codes while the new model was calibrated on ICD-10 codes for the first time.
2. Shifted from 86 payment condition categories in the 2020 model to 115 in the new model — not by adding new diseases, but by modifying the breakout/hierarchy with increased granularity.
3. Substantially reduced the number of diagnosis codes that trigger payments: approximately 20% less ICD-10s now trigger a payment.

The impact of these changes can be characterized in three main components:

1. Nearly all demographic coefficients increase. These represent the component of a risk score determined by sex and age.
2. New enrollee scores increase both in aggregate and across the board. These are scores without an individual disease component given to members without full Medicare eligibility in the date of service period.
3. Several disease categories see far fewer members receiving payments for them due to ICD-10 codes dropped from the model (for example, vascular, psychiatric, or blood). This shifts more of the risk score to the demographic component and means that members who continue to get payments for these diseases tend to have more severe or complex conditions.
   

In the announcement, CMS emphasized that only 3.4% of the ICD-10 codes dropped from the model were motivated by Principle-10 alone (that is, considered discretionary or not credible as a cost predictor). They also note that these changes accounted for an outsized 18% of total model impact. Another 28% of code removals may have been partially justified by Principle-10, while 70% were removed for other reasons. There is copious justification given in the notice for the removal of these 70% of codes, even though they are likely found on a negligible portion of medical claims in practice. In fact, most of the model change is potentially explained by codes whose removal is either fully or partially justified by Principle-10. On actual medical claims, many of these are unspecified, non-specific, or lower severity codes.
 

The left-hand side codes result in payments under the 2020 model, but not under the 2024 model. The right-hand side codes, which are clinically related, result in payments under both models. For example, in the second row, if a patient has an alcohol-induced disorder that the clinician can identify and substantiate, but for which no specific code exists, then the patient should be coded with F10188 on the right. That code would result in a payment under the old and new models. If the clinician were to instead inadvertently code F1019, which is meant to represent the case when a disorder is present without the necessary detail in the documentation to assign a specific disorder, then there would be no payment under the new model. This type of mistake in documentation accuracy will not only mean the patient may not get the appropriate and needed care, but it will also result in lower, inaccurate reimbursement to the plan and risk-bearing providers.

Educating providers on the impact of coding accuracy is vital

Coding is a tedious task for a clinician that is repeated for every condition in every patient visit. We know this can lead to doctors choosing higher-level codes which misrepresent the severity level, specificity, or complexity of their patient’s condition. Given that the new model is tuned to require greater specificity/severity in many cases, the impact of this practice may materialize in far lower payments to Medicare Advantage Organizations (MAOs) and risk-bearing providers. It will be important to educate clinicians with the aid of robust analytics identifying where this may be occurring, and which alternate codes could be appropriate in these cases.

Along with provider education to prevent the miscoding described above, plans should update any risk adjustment forecasting models, intervention/program strategies, and chase activities to reflect the blend of the two models and closely monitor and evolve these efforts as results materialize through encounter data. As the Medicare Advantage program gets increasingly competitive, plans and providers cannot afford inaccurate clinical documentation or plan reimbursements. This could lead to worse patient outcomes, reduced margins, higher premiums and/or weakened product offerings over the next three years and beyond. We’re nearly at the end of the first year of service that will inform plan reimbursements under the new model, so plans and providers should take action immediately.