Assessing the MA Risk-Adjustment Model’s Accuracy Among Subpopulations

  • This page as PDF

Summary

The accuracy of the CMS-HCC model differs by beneficiaries’ race and ethnicity.

The Centers for Medicare & Medicaid Services (CMS) uses the CMS-Hierarchical Condition Category (HCC) risk-adjustment model to pay Medicare Advantage (MA) plans a capitated amount per member per month. The CMS-HCC model includes historic utilization, diagnoses, and costs for fee-for-service (FFS) Medicare to predict costs for MA plans. While the model accounts for certain demographic details, including age, sex, and dual and disability status, it does not account for beneficiaries’ race/ethnicity.

Stakeholders have raised concerns with various elements of the current CMS-HCC model, including the appropriate use of beneficiary demographics data and decisions on the inclusion of chronic diseases in the model. Previous research shows that, because the model predicts healthcare costs based on FFS Medicare spending, it can underpredict costs for beneficiaries who have had low spending as a result of systemic access barriers. In other words, the model coefficients may build inequities into the MA plan payment structure and perpetuate health disparities.

Approach

Avalere conducted an analysis of the CMS-HCC community model to assess its accuracy in predicting costs for groups of beneficiaries categorized by race/ethnicity. To do so, Avalere estimated predicted costs from the model compared to actual costs. After using claims data to first identify and then calculate the actual costs of FFS beneficiaries, Avalere compared the predicted costs for this sample using the CMS-HCC community model in use for plan year 2023.

The ratio of predicted costs to actual FFS Medicare spending is known as the predictive ratio. A predictive ratio of 1.0 means a beneficiary’s (or a group of beneficiaries’) expected healthcare costs are accurately predicted. The further the predictive ratio diverges from 1.0, the more significant the underprediction or overprediction (i.e., a predictive ratio greater than 1.0 means the model estimated more than actual costs and a predictive ratio less than 1.0 means estimated costs were less than actual costs).

Overall Results

Avalere’s analysis found that the CMS-HCC model, on average, closely predicts healthcare costs for beneficiaries who are Black and non-Hispanic White; it overpredicts for beneficiaries who are Asian/Pacific Islander and those who are Hispanic; and it underpredicts for beneficiaries who are American Indian/Alaska Native (Figure 1).

Figure 1. Predictive Ratios by Race/Ethnicity in the CMS-HCC Model
Figure 1. Predictive Ratios by Race/Ethnicity in the CMS-HCC Model

Race/Ethnicity and Dual Status

Avalere also analyzed the predictive ratio by beneficiaries’ race/ethnicity to stratify by beneficiaries’ dual-eligible status. Results for this analysis showed the CMS-HCC model underpredicts costs for full and partial dual beneficiaries who are Black compared to non-dual beneficiaries who are Black. Further, the model underpredicts across every category for American Indian/Alaska Native beneficiaries. Although the model more accurately predicts costs for dual-eligible beneficiaries who are Hispanic, the model demonstrates a greater overprediction of costs for non-dual beneficiaries who are Hispanic (Figure 2).

Figure 2. Predictive Ratios by Race/Ethnicity and Dual Status
Figure 2. Predictive Ratios by Race/Ethnicity and Dual Status

Disease/Condition Category

Avalere’s analysis showed the CMS-HCC model accurately predicted overall costs for beneficiaries who are Black. However, for beneficiaries with the most common HCCs, the model underpredicts for Black beneficiaries with eight of the top ten conditions. The model generally overpredicts for all beneficiaries who are Hispanic; however, the predictive ratios for specific conditions are generally lower than for the full group of beneficiaries (Table 1).

Table 1. Predictive Ratios by Race/Ethnicity and Disease Type
Top 10 HCCs* Overall Non-Hispanic White Black Asian/Pacific Islander Hispanic American Indian/Alaska Native
Overall Predictive Ratio 1.00 0.99 1.01 1.25 1.09 0.93
1. Diabetes with Chronic Complications 0.98 0.97 0.96 1.15 0.99 0.91
2. Vascular Disease 0.98 0.99 0.91 1.06 0.99 0.95
3. Specified Heart Arrhythmias 0.98 0.99 0.94 1.12 0.97 0.97
4. Chronic Obstructive Pulmonary Disease 0.98 0.99 0.91 1.12 0.95 0.96
5. Congestive Heart Failure 0.98 0.99 0.91 1.09 0.96 0.94
6. Major Depressive, Bipolar, and Paranoid Disorder 0.98 0.96 1.00 1.15 1.02 0.89
7. Diabetes without Complication 0.98 0.98 0.95 1.12 0.96 0.98
8. Rheumatoid Arthritis and Inflammatory Connective Tissue 0.98 0.98 0.97 1.13 1.00 0.95
9. Breast, Prostate, and Other Cancers and Tumors 0.98 0.97 1.00 1.14 1.03 0.95
10. Morbid Obesity 0.97 0.97 0.96 1.06 0.97 0.97

*Ranked by number of beneficiaries with the condition, largest to smallest.

The current model also underpredicts for enrollees who are Black and those who are Hispanic when they have at least 5 HCCs, and it overpredicts for beneficiaries who are non-Hispanic White who have at least five HCCs (Figure 3). For each of these groups, the finding for those with at least five HCCs is opposite of the finding for these groups of beneficiaries who have zero and up to four HCCs.

Figure 3. Predictive Ratios by Race/Ethnicity and Number of HCCs
Figure 3. Predictive Ratios by Race/Ethnicity and Number of HCCs

Discussion

An optimal risk-adjustment model accurately predicts costs overall and for subpopulations, including racial/ethnic minorities. Avalere’s research indicates that the CMS-HCC risk-adjustment model may incorrectly predict costs for certain subpopulations, which might perpetuate disparities by overpaying for some low-cost populations and underpaying for some high-cost groups of beneficiaries.

This study illustrates that, as beneficiaries have more diagnosed conditions, overprediction increases for those who are non-Hispanic White and underprediction emerges for those who are Black and those who are Hispanic. This finding may be due to the HCC count variable mandated by the 2016 21st Century Cures Act, which increased predicted costs for higher HCC counts. Further study may be warranted to better understand why trends in predictive accuracy differ across racial groups as the number as HCC counts go up, and whether the addition of the Cures Act may impact the accuracy of the model for specific subpopulations.

CMS enacted changes to the CMS-HCC risk-adjustment model as part of the 2024 MA and Part D Advanced Notice that would update the data years used for calibration, update the denominator year used to determine the average per capita predicted expenditures, and reclassify the HCCs using International Classification of Diseases, Tenth Revision, Clinical Modification codes. Further study would also be needed to evaluate the accuracy of this new model on payments for subpopulations of Medicare beneficiaries.

Funding for this research was provided by Arnold Ventures. Avalere retained full editorial control.

To learn more about social determinants of health, connect with us.

Methodology

Avalere identified all beneficiaries who were enrolled in both Medicare FFS Part A and Part B for at least one month in 2019. FFS beneficiaries were identified using the 100% files of Medicare FFS Parts A and B for 2019 and prior FFS enrollment in 2018. Medicare claims and enrollment data were accessed via a research collaboration with Inovalon Inc. under a CMS data use agreement.

Using 2018 and 2019 100% Medicare FFS claims data, Avalere identified all risk-adjustment eligible HCPCS codes in 2018 and claim expenditures, excluding hospice claims, in 2019. Avalere then estimated the CMS-HCC community model, which includes enrollees who have 12 months of Part B enrollment in 2018. The calibration included separate models for full duals, partial duals, and non-duals both under and over age 65 using 2018 diagnoses to predict Medicare Part A and Part B expenditures in 2019. For this model calibration Avalere used the CMS-HCC model that CMS used for 2023 payment (the V24 CMS-HCC model).

Webinar | Election 2024: What’s at Stake for Healthcare? 

On August 14 at 1:30 PM ET, Avalere experts and guests will discuss the 2024 elections, exploring the candidates’ health policy approaches and implications for stakeholders. 

Learn More

Register
From beginning to end, our team synergy
produces measurable results. Let's work together.

Sign up to receive more insights about Health Equity
Please enter your email address to be notified when new Health Equity insights are published.

Back To Top