Introduction
In prior blogs, we have discussed how the 21st Century Cures Act (CCA) helped propel the advance of health IT interoperability requirements. The CCA also contained Medicare reimbursement policy provisions, including several that substantially impact the risk adjustment model for Medicare Advantage (MA) payments. The CCA provisions related to MA risk adjustment were all scheduled to be evaluated and/or implemented by 2022.
In this blog series, we provide an overview of each of the changes and potential impacts to MA plans, starting with the addition of condition count variables to the MA risk adjustment model. Future blogs will address adjustments for specific types of conditions, adjustments for different types of MA enrollees eligible for Medicaid benefits, and the potential expansion of the amount of diagnosis data available for risk adjustment calculations. Before getting into the changes, we also review some key details about the risk adjustment models used in MA to help put the changes into context. For additional details on how the MA risk adjustment works, you can also visit a prior blog.
Background on Medicare Advantage Risk Adjustment
The Centers for Medicare & Medicaid Services (CMS) pays private health insurance plans that participate in the Medicare Advantage (MA) program a monthly capitated amount to administer Medicare benefits to its enrollees. There is a base capitated payment amount, with adjustments being made based on risk score. The base amount varies by county, reflecting the costs of providing benefits for an “average” Medicare enrollee in each county. In this case, an average Medicare enrollee means an enrollee with the health status of the national average beneficiary in the traditional fee-for-service (FFS) Medicare program.
An enrollee’s risk score indicates her/his expected costs relative to the national average FFS enrollee. The risk score for the national average enrollee is set to 1.0. MA enrollees that are expected to be sicker than average have risk scores greater than 1.0, while healthier enrollees have risk scores lower than 1.0. The risk scores are applied to the capitated payment amount as a multiplier on the base payment amount. Thus, MA plans receive larger payments for enrollees with higher risk scores and vice versa. This helps ensure that MA plans that enroll sicker Medicare beneficiaries have the resources needed to effectively manage the health care needs of their members.
CMS measures the health status of MA enrollees largely through diagnosis codes that are documented in medical records. CMS then categorizes the codes into groups of clinically related conditions called Hierarchical Condition Categories (HCCs). These categories largely reflect chronic conditions, with some of the more common HCCs being chronic obstructive pulmonary disease (COPD) and congestive heart failure. The HCCs, along with demographic variables (e.g., age and gender), are used in regression models—known as the CMS-HCC models—to estimate a risk score for each enrollee. In general, the more HCC codes that are documented for an enrollee, the higher the risk score for that enrollee and consequently the higher the payment to the MA plan.
Adjustment to CMS-HCC Model for Enrollees’ Total Number of Conditions
As described above, the current CMS-HCC regression models utilize demographic variables and HCCs to generate a risk score for each enrollee. The CCA requires that adjustments for the total count of diseases or conditions for each enrollee are also included in the models. In 2020, CMS began incorporating such counts by basing the risk scores for each enrollee on a 50/50 blend of a CMS-HCC model with additional count indictors and a model that does not include the count indicators (i.e., the prior model used). Risk scores will be fully based on the new model—known as the alternative payment condition count (APCC) model—by the 2022 payment year.
CMS incorporated the count indicators by counting the number of documented HCCs for each enrollee. The count is limited to just the HCCs that are used in the regression models (i.e., those that are used for adjusting payments). Currently, there are 86 HCCs in the models (we will discuss the recent increase in the number of HCCs in an upcoming blog). For illustrative purposes, assume an enrollee has medical diagnoses that map to HCC 19 (diabetes without complications), HCC22 (morbid obesity), HCC55 (substance use disorder, moderate/severe, or substance use with complications), HCC59 (major depressive, bipolar, and paranoid disorders), HCC 85 (congestive heart failure), and HCC 111 (COPD); in this instance, CMS would assign this enrollee six medical conditions.
The CMS-HCC models are additive in nature, with the risk scores essentially being the sum of all of the coefficients in the regression model for the relevant demographic variables and HCCs for an enrollee. Thus, the new CCA requirement means that enrollee’s illnesses and conditions are taken into consideration in two different ways: once with a regression coefficient for the specific condition included in the model, and a second time with a regression coefficient for a variable that counts the number of conditions for each enrollee.
The condition count is included as a series of “1”/”0” variables indicating how many conditions an enrollee has. For example, an enrollee with 6 chronic conditions would have his/her “6 count” variable set to “1” with the other count variables set to “0”. The count variables start with counts between 4 and 6 depending on which MA enrollee population is being segmented. (More details on model segmentation will be provided in an upcoming blog on the CCA changes related to MA enrollees eligible for Medicaid. For now, it is essential to know that different risk score adjustments are made for different types of MA enrollees.)
In its explanation of the condition count variables, CMS emphasized the long-standing principle of not wanting to penalize MA plans for documentation of additional diagnoses as a key reason for not starting the condition count with the first HCC. In general, enrollees with fewer conditions are healthier and are associated with lower costs. This could result in a negative coefficient (signifying that an enrollee with no or very few HCCs is less costly than the average Medicare enrollee), which makes it possible for risk score decreases with the reporting of additional conditions. For example, if the count variables started from a count of one, it would be possible that, with a negative count coefficient, the increase from the coefficient of an additional HCC could be less than the decrease from the next count variable. CMS wants to avoid such scenarios. This practice of essentially constraining the coefficients of count variables with a negative coefficient to zero (thereby removing its impact on the risk score) is consistent to the approach CMS has taken with coefficients of specific CMS-HCCs (which are only allowed to be additive—i.e., positive—in the MA risk adjustment models).
CMS also capped the count variables at 10 conditions. Thus, the coefficient for the “10 count” variable applies for 10 and any more conditions the beneficiary may have. CMS did this partly for statistical reasons (e.g., the sample size becomes much smaller when estimating costs for people with more than 10 HCCs, which could lead to less reliable risk score adjustments), as well as clinical reasons. Clinicians with whom CMS consulted did not think it was clinically meaningful to distinguish between 10 and more than 10 conditions.
Impact to the Risk Adjustment Model
An evaluation by the Medicare Payment and Advisory Commission (MedPAC) found that adding the new condition count variables improves how well the CMS-HCC models predict the cost of enrollees who have no conditions indicated in the model and those who have many conditions. CMS also conducted an analysis showing an improvement of the predictive accuracy of the model for lower cost (healthier) and higher cost (sicker) enrollees. CMS used these findings to justify moving forward with the APCC model, which includes the condition count variables.
The main purpose of the CMS-HCC models is to predict the cost of MA enrollees, and the CCA adjustments were largely implemented for the purpose of improving the accuracy of the models. Historically, the CMS-HCC models, much like many predictive models, often resulted in the underprediction of costs for people with very high levels of actual spending and overprediction of costs for people with very low levels of actual spending. Thus, there has been an underpayment for very high-cost MA enrollees and an underpayment for very low-cost MA enrollees. It appears that that the inclusion of the condition count variables helps address this. However, MedPAC notes in its analysis that while the systemic underpayments and overpayments are less, they are not completely eliminated.
Impact to MA Plans
The CCA adjustments for the total number of conditions will make it even more important for plans to capture all of an enrollee’s conditions. Plans with relatively high average conditions per enrollee will see higher risk-adjusted payments. Thus, the value of capturing all of an enrollee’s risk-adjusted conditions are that much more valuable.
It should also be noted that with the inclusion of total count variables in the CMS-HCC model, the coefficients for many of the individual HCCs will decrease. Thus, depending on the mix of HCCs within an MA plan’s enrollee population, a plan may see a decrease in payments. For example, MA plans with a relatively larger proportion of enrollees that have HCCs with coefficients that have decreased and who have less than 4 HCCs in total (and hence will not have an addition to the risk score due to the new condition count variables) may be more likely to see a decline in payments.
Looking Forward
Future blogs will address adjustments for mental health, substance abuse and chronic kidney disease conditions, adjustments for different types of MA enrollees eligible for Medicaid benefits, and the potential expansion of the amount of diagnosis data available for risk adjustment calculations.