Breaking it down: proposed Medicare Advantage risk adjustment model changes for 2019

The Centers for Medicare & Medicaid Services (CMS) has released its advance notice of proposed changes to the Medicare Advantage (MA) risk adjustment model for the 2019 calendar year (CY 2019). These changes are in line with the agency’s continued commitment to ensure MA payments accurately reflect the relative health risk and actual expected healthcare costs of plan beneficiaries. In addition, they reflect risk adjustment improvements directed by the 21st Century Cures Act, signed into law in December 2016. Lesley Brown, Cotiviti's vice president of Risk Adjustment, breaks down the key updates.

Mental health, substance use disorder, and chronic kidney disease

CMS is proposing changes to the CMS-HCC risk adjustment model that focus on additional inclusions for chronic conditions related to mental health and substance abuse, as well as the severity of chronic kidney disease. Mental health disorders and substance abuse continue to be on the rise in the United States, with the recent opioid epidemic being a perfect example. CMS has determined that additional diagnoses expanding substance use disorders, mental health-related disorders, and the severity of chronic kidney disease are clinically meaningful and predict medical expenditures, and that the conditions can be diagnosed definitively. As such, CMS believes the diagnoses meet the essential criteria to be included in MA risk adjustment payment models. The table below summarizes the proposed CY 2019 HCC changes. 

Adding these specific HCCs is intended to better align MA plan payments to the actual costs of providing care to beneficiaries related to these conditions. Depending on a plan’s specific member population make-up, these changes could result in a significant increase to its risk adjustment payments.

Figure 1. Current 2017 CMS-HCC model vs. proposed 2019 CMS-HCC model.

Accounting for a beneficiary’s total number of conditions 

Under the 21st Century Cures Act directive to “take into account the total number of diseases or conditions of an individual beneficiary,” CMS has completed an extensive study to assess the impact on risk scores of adding a variable to the risk adjustment model that accomplishes this goal. Its analysis focused on which analytic construct to use within the newly proposed model:

• All Condition Count, which considers all conditions that a beneficiary has, encompassing both those that are included in the payment model and those that are not. CMS projects this would decrease overall MA risk scores by 0.28 percent.
• Payment Condition Count, which only considers the conditions included in the payment model. CMS projects this would increase overall MA risk scores by 1.1 percent. 

CMS extensively researched which conditions to include in the modeling counts (payment vs. non-payment conditions), whether to aggregate counts (e.g., “3–6 conditions”), whether to “top-code” counts (e.g., “10 or more”), and how to apply the counts to the HCC models. Models that included non-payment conditions in the counts were developed and evaluated separately from models that excluded them.

In the advance notice, CMS is proposing to use the Payment Condition Count, believing that it will improve the accuracy of the risk adjustment model. Conditions used in the payment model tend to be more clinically severe, and the majority are classified as chronic. In contrast, using the "all conditions" approach would count many conditions that do not meet the criteria of being clinically meaningful and cost predictive, such as those that are acute, typically being treated and cured in a single therapy course.

“CMS’s proposed plan to include payment condition counts in its risk adjustment modeling is a natural progression of the program,” David Costello, Verscend chief analytics officer, commented. “This will assist MA plans responsible for caring for sicker patients with multiple co-morbid conditions. In addition, the payment condition model manages the MA contract-level risk variation much more efficiently than the more broadly defined condition count approach. This reduction in variation will reduce the level of 'noise' and benefit providers that are servicing a higher-risk population. 

 

"The move addresses a well-known limitation of the HCC modeling technology that CMS has used to adjust Medicare Advantage payments since 2004. That early HCC methodology tended to under-predict at the high end of spending and to over-predict at the low end. This pattern can disadvantage MA plans with sicker populations and can induce perverse incentives to avoid enrolling higher-risk individuals.

 

"With this being said, CMS would be wise to contemplate Campbell’s law, which states that 'the more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.' Basically, once providers understand the underlying assumption of either approach, they will work to maximize their benefits by changing their coding patterns. Both approaches will lend themselves to potential corruption of the initial CMS intent to identify the truly higher-risk populations."

CMS’s intent is to put these risk adjustment model changes into place over a three-year period beginning in 2019, combining 25 percent of the proposed Payment Condition Count model with 75 percent of the existing 2017 CMS-HCC model. The phase-in would end in 2022, when the model would be 100 percent Payment Condition Count.

Also note that in 2019, CMS is proposing to calculate risk scores with 25 percent from EDS and 75 percent from RAPS. Only the EDS-based risk score would use the proposed Payment Condition Count model, while RAPS scores would be calculated using the 2017 CMS-HCC model.

Time to comment 

Interested stakeholders have until March 2, 2018, to submit comments on the proposal. We encourage MA plan stakeholders and other interested parties to comment on CMS’s methodology, specifically which condition count model they find to be the best option.