CMS Pub. 100-16, ch. 7, § 70

The CMS-HCC risk adjustment models are used to calculate risk scores, which predict

individual beneficiaries’ health care expenditures, relative to the average beneficiary.

Risk scores are used to adjust payments and bids based on the health status (diagnostic

data) and demographic characteristics (such as age and gender) of an enrollee. Both the

Medicare Advantage and Prescription Drug programs include risk adjustment as a

component of the bidding and payment processes. CMS uses risk adjustment to:

• Standardize bids so that each plan has a bid for the average Medicare beneficiary

• Compare bids based on populations with different health statuses and other

characteristics

• Adjust plan payment based on the characteristics of the enrolled population

CMS has developed separate risk adjustment models for the Parts A and B benefits

offered by plans under Part C and for the Part D benefits offered by prescription drug

plans. Within each benefit, CMS also developed segments of the models for

subpopulations with distinct cost patterns.

The Part C model has segments for the following subpopulations of beneficiaries:

• Aged/disabled Community

• Aged/disabled Institutional

• Aged/disabled New enrollee

• ESRD Dialysis

• ESRD Dialysis New Enrollee

• ESRD Transplant

• ESRD Functioning Graft – Community

o Add-on for 4-9 months

o Add-on for 10+ months

• ESRD Functioning Graft – Institutional

o Add-on for 4-9 months

o Add-on for 10+ months

• ESRD Functioning Graft – New Enrollee

o Add-on for 4-9 months

o Add-on for 10+ months

From 2006 through 2010, the Part D model uses a base model with multipliers for:

• Low Income (partial)

• Low Income (full)

• Long Term Institutional (aged)

• Long Term Institutional (disabled)

Starting in 2011, the Part D model has the following segments:

• Aged, non-low income

• Aged, low income

• Disabled, non-low income

• Disabled, low income

• Institutional

• New Enrollee, non-low income

• New Enrollee, low income

• New Enrollee, institutional

Table 2 below summarizes the common characteristics across all HCC-based risk

adjustment models.

Table 2. HCC Specific Characteristics

Characteristic Descriptions

Selected Significant

Disease (SSD)

Model

Model considers serious manifestations of a condition rather than all levels

of severity of a condition.

Include most body systems and conditions.

Models are Additive Individual risk scores are calculated by adding the coefficients associated

with each beneficiary’s demographic and disease factors.

Prospective Model Uses diagnostic information from a base year to predict Medicare benefit

costs for the following year.

Site Neutral Models do not distinguish payment based on a site of care.

Diagnostic Sources Models recognize diagnoses from hospital inpatient, hospital outpatient, and

physician settings.

Multiple Chronic

Diseases Considered

Risk adjusted payment is based on assignment of diagnoses to disease

groups, also known as Condition Categories (CCs).

Model is most heavily influenced by Medicare costs associated with chronic

disease.

Hierarchies Condition Categories are placed into hierarchies, reflecting severity and

cost dominance. Beneficiaries get credit for the disease with the highest

severity or that subsumes the costs of other diseases. Hierarchies allow for

payment based on the most serious conditions when less serious conditions

also exist.

Disease and

Disabled

Interactions

Interactions allow for higher risk scores for certain conditions when the

presence of another disease or demographic status, e.g., disabled status, is

indicative of higher costs. Disease interactions are additive factors and

increase payment accuracy.

Demographic

Variables

Models include five demographic factors: age, sex, disabled status, original

reason for entitlement, Medicaid or low income status.

These factors are typically measured as of the data collection period.

70.1 - Calibration of the CMS-HCC Risk Adjustment Models

(Rev. 118; Effective: ICD-10: Upon Implementation of ICD-10, ASC X12: January 1, 2012 (for ASC X12 5010); Implementation: ICD-10: Upon Implementation of ICD-10,

ASC X12: January 1, 2012 (for ASC X12 5010))

The CMS-HCC risk adjustment model is used to adjust payments for Part C benefits

offered by MA plans and PACE organizations to aged/disabled beneficiaries. The CMS-HCC model includes both diseases and demographic factors. There are separate sets of

coefficients for beneficiaries in the community, beneficiaries in long term care

institutions, and new enrollees. The CMS-HCC model was first used for payment in

2004 and has been recalibrated two times since then (2007 and 2009).

When CMS recalibrates the CMS-HCC risk adjustment model, it uses data from fee-for-service (FFS) claims, using one year’s diagnoses to predict the following year’s

expenditures. When developing the model, CMS consulted with a panel of outside

clinicians to review the diagnosis codes in order to group them with other clinically

similar diagnosis codes. These diagnosis groupings were then mapped to condition

categories based on similar clinical characteristics and severity, and cost implications.

Both the panel of clinicians and analyses of cost data informed the creation of condition

categories.

Coefficients for condition categories were estimated by regressing the total expenditure

for Medicare Parts A and B benefits for each beneficiary onto their demographic factors

and condition categories, as indicated by their diagnoses. Resulting dollar coefficients

represent the marginal (additional) cost of the condition or demographic factor (e.g.,

age/sex group, Medicaid status, disability status).

While all diagnosis codes are mapped to a condition category, not all condition categories

are included in the model used in payment. The decision to include a condition category

in the model is based on each category’s ability to predict costs for Medicare Parts A and

B benefits. Condition categories that don’t predict costs well – because the coefficient is

small, the t-value is low, the number of beneficiaries with a certain condition is small so

the coefficient is unstable, or the condition does not have well specified diagnostic coding

– are not included in the model.

In a final step, hierarchies were imposed on the condition categories, assuring that more

advanced and costly forms of a condition are reflected in the risk score.

In order to use the risk adjustment model to calculate risk scores for payment, CMS

creates a relative factor for each demographic factor and HCC in the model. CMS does

this by dividing all the dollar coefficients by the average per capita predicted expenditure

for a specific year (i.e., the “denominator year”). See Table 3 below for a list of data

years and denominator years in each version of the risk adjustment model. The relative

factors are used to calculate risk scores for individual beneficiaries, which will average

1.0 in the denominator year for the FFS population.

Each time the risk adjustment model is recalibrated, the relative factors can change.

Changes in the dollar coefficients resulting from the regression – the marginal cost

attributable to an HCC – can change relative to the average cost. For example, the

coefficient for diabetes can increase, reflecting higher costs for the disease; but if the

average cost for Medicare beneficiaries has increased even more than for diabetes, then

the relative cost of diabetes will decrease. This decrease in relative cost will be reflected

in a decrease in the relative factor, even though the costs associated with diabetes have

increased.

Although recalibrated models retain an average 1.0 risk score, individual beneficiaries’

risk scores may change, as may plan average risk scores, depending on each individual

beneficiaries’ combination of diagnoses.

Table 3. Data Years and Denominator Years

Payment Years Diagnoses Year Costs Year Denominator Year

2004, 2005, 2006 1999 2000 2000

2007, 2008 2002 2003 2005

2009, 2010, 2011 2004 2005 2007