Hierarchical Condition Category

Hierarchical Condition Category in Medicare Risk Adjustment

A hierarchical condition category, usually shortened to HCC, is one building block of the diagnosis classification that the Centers for Medicare & Medicaid Services (CMS) uses to pay private Medicare Advantage plans. Each enrolee's recorded diagnoses are sorted into categories, the less severe category is dropped wherever two related ones coexist, and every category that survives adds a fixed amount to a relative risk score. Because those scores decide how much a plan receives for each person, the design of the categories governs both the accuracy of payment and the reward for recording diagnoses. The general logic of turning codes into a prediction belongs to the diagnosis-based risk score, and HCCs are the grouping CMS built and keeps revising for that purpose.

How diagnosis codes become condition categories

Pope and colleagues describe the classification behind the 2004 model as a series of aggregations. More than 15,000 ICD-9-CM codes were each mapped to one of 804 diagnostic groups, which were combined into 189 condition categories (CCs), similar in clinical content and cost. Hierarchies were then imposed among related CCs, so that a person is coded only for the most severe manifestation of a disease, and the categories that survive are the HCCs.

In the coronary artery disease hierarchy of that version, a code for acute myocardial infarction excluded the three lower categories even when codes for them were present. Unrelated conditions accumulate instead, so a man with heart disease, stroke and cancer carries an increment for each. MedPAC's March 2026 report cites four chronic kidney disease HCCs, ranked stage 5, stage 4, stage 3B and stage 3 (except stage 3B), of which only the highest applicable one counts.

$$H_{ih} = C_{ih} \prod_{g \in A(h)} \left(1 - C_{ig}\right)$$

where $H_{ih}$ equals 1 if person $i$ is assigned HCC $h$ and 0 otherwise, $C_{ih}$ equals 1 if any of the person's eligible diagnosis codes maps to condition category $h$, and $A(h)$ is the set of categories whose presence excludes $h$ under the CMS hierarchy rules.

Ten design principles and their trade-offs

Pope and colleagues list ten principles behind the classification, and the CY 2024 Rate Announcement confirms that CMS still uses them to group codes and choose payment HCCs. Categories should be clinically meaningful, predict spending and be large enough for stable estimates, with hierarchies inside each disease process and unrelated diseases adding up. Vague codes are grouped with lower-paying categories, and repeating a code or adding closely related codes does not raise the prediction.

Three principles were followed absolutely. No category may carry a negative weight or pay less than a category it excludes, the ranking must be transitive, and every code must be classified. Where estimated coefficients broke the first rule, constraints were imposed. The tenth principle keeps out of the payment model any category particularly open to discretionary coding or not credible as a cost predictor. Clinical detail, stable estimation and resistance to gaming pull against one another, and the authors settled those trade-offs with empirical evidence and clinical judgement.

How the CMS-HCC payment model was built

CMS chose the Diagnostic Cost Group (DCG) and HCC family, according to Pope and colleagues, largely for transparency, ease of modification and clinical coherence. The transitional PIP-DCG model it replaced used only inpatient diagnoses, so plans that kept people out of hospital through good ambulatory care could look healthier and be paid less. Of the 189 categories, 101 survived the first exclusions, and CMS cut the list to 70, balancing reporting burden against predictive power, generally keeping the severe, costly categories at the top of each hierarchy.

The model uses prospective risk adjustment, predicting spending in one year with diagnoses recorded in the year before. The 2004 version was calibrated by weighted least squares multiple regression on 1999 diagnoses and 2000 spending for a 5% sample of fee-for-service beneficiaries, and its R-squared was 9.8%. On earlier data the authors report R-squared of 1.0% for an age-sex model, 6.2% for PIP-DCG and 11.2% for the full DCG/HCC model, and only 43% of DCG/HCC payment rested on demographics.

Relative factors and the payment risk score

Coefficients are published as relative factors, each a cost increment expressed relative to average fee-for-service spending in a chosen denominator year, so that the average fee-for-service beneficiary scores 1.0 in that year. A person's raw score is the sum of the factors that apply, and CMS then applies a normalisation factor and the coding pattern adjustment.

$$R_i = \frac{1 - c}{N} \left( \sum_{a} \delta_a D_{ia} + \sum_{h} \beta_h H_{ih} + \sum_{k} \gamma_k I_{ik} \right)$$

where $R_i$ is the payment risk score of enrolee $i$, $D_{ia}$ indicates the demographic cells and statuses that apply, with relative factors $\delta_a$, $H_{ih}$ is the HCC indicator defined above, with relative factor $\beta_h$, $I_{ik}$ indicates disease interactions or condition counts, with relative factors $\gamma_k$, $N$ is the normalisation factor and $c$ is the coding pattern adjustment as a proportion. Payment is broadly the plan's base rate times $R_i$, as on the capitation page.

Scores tend to rise over time with changes in population and coding. The CY 2027 Rate Announcement explains that the normalisation factor projects the average fee-for-service score forward to the payment year and that applying it holds that average at 1.0, and sets it at 1.079 for the 2024 CMS-HCC model. The coding pattern adjustment addresses the separate fact that Medicare Advantage plans record more diagnoses than fee-for-service providers for similar people. The statutory minimum is 5.9%, CMS applies that minimum for 2027, and MedPAC notes that CMS has never imposed a larger reduction.

Model versions and the 2024 reclassification

The CY 2024 Rate Announcement adopted a model, often called V28, rebuilt on ICD-10 codes and recalibrated on 2018 diagnoses and 2019 spending in place of 2014 and 2015 data, which raised the number of payment HCCs to 115, against 86 before. It was phased in at one third of the score in 2024, two thirds in 2025 and the whole score in 2026, and for 2027 CMS keeps it rather than adopting the recalibration it had proposed.

The revision applied the tenth principle directly. All three diabetes HCCs were constrained to one weight, because the category for diabetes with chronic complications could be coded from a laboratory finding with no implications for treatment, and the extra cost now flows to well-defined complications such as chronic kidney disease. Angina pectoris, whose criteria allow coding when a patient shows only chest pain, moved to a non-payment HCC while unstable angina stayed in. MedPAC estimates that V28 reduced coding intensity by 2.9 percentage points in each year of its phase-in.

Coding intensity and the rules on diagnosis sources

A diagnosis counts only if it comes from a hospital inpatient stay, a hospital outpatient visit or a face-to-face visit with a physician or other health professional, and is supported by the medical record. MedPAC reports that Office of Inspector General audits of high-risk diagnoses for at least 30 contracts found 70% of the audited codes unsupported by medical records.

Fee-for-service physician and outpatient claims are paid mainly by procedure, so providers there have little financial reason to record every diagnosis, and the coefficients are calibrated on that sparser coding. MedPAC projects that Medicare Advantage risk scores in 2026 will be about 10% above those of comparable fee-for-service beneficiaries, leaving about 4% after the 5.9% adjustment, worth USD 22 billion in higher payments. For 2027 CMS also excludes diagnoses from unlinked chart review records, except for beneficiaries who switch plan organisation.

The HCC approach outside Medicare Advantage

Kautter and colleagues describe the HHS-HCC model for the Affordable Care Act individual and small group markets, with 264 HHS-HCCs in the full classification and 127 used in its risk adjustment models. It predicts medical and drug spending together for commercial insurance and uses same-year diagnoses, the design examined on the concurrent risk adjustment page. In the Medicare Shared Savings Program, federal regulation (42 CFR 425.605) uses an accountable care organisation's prospective HCC risk score to adjust its benchmark for changes in severity and case mix.

HCC scores in health economic studies

In outcomes research and economic evaluation an HCC score is mainly a covariate or matching variable for case-mix adjustment in US claims data analyses. Ellis shows a concurrent DCG/HCC model narrowing a raw 20% spending gap at age 40 between US private plan types to under 5%, most of it selection. The score measures expected spending, not severity or need, so it can balance groups on predicted cost while leaving confounding by unrecorded illness in place. Diagnoses recorded after an intervention begins may include its complications, so a score from a pre-treatment window is the safer baseline measure.

Comparisons across time need the same model version and denominator year, and the CY 2024 Rate Announcement warns that comparing HCC coefficients across the old and new models would not be appropriate.

Worked example of one enrolee's score (illustrative)

All relative factors here are invented for teaching and are not CMS coefficients. Only the normalisation factor of 1.079 and the 5.9% coding adjustment are published CY 2027 values, and the base rate of USD 1,100 a month is illustrative. A 79-year-old woman living in the community has base-year codes for diabetes, chronic kidney disease at stages 4 and 3, congestive heart failure and chest pain.

1. Apply the hierarchy. Stage 4 and stage 3 kidney disease sit in one hierarchy, so only stage 4 is kept. Chest pain is a symptom outside the payment model.

2. Add the relative factors. With a demographic factor of 0.40, diabetes 0.17, kidney disease stage 4 0.52, heart failure 0.36 and a diabetes with heart failure interaction of 0.11, the raw score is 0.40 + 0.17 + 0.52 + 0.36 + 0.11 = 1.56. Without the hierarchy the stage 3 factor of 0.13 would also count, giving 1.56 + 0.13 = 1.69.

3. Normalise and adjust for coding. Dividing by the normalisation factor gives 1.56 / 1.079 = 1.4458, and multiplying by 1 minus 0.059 gives 1.4458 × 0.941 = 1.3605, the payment risk score.

4. Convert to payment. The monthly payment is 1.3605 × 1,100 = USD 1,496.55, and the diagnoses supply 74% of the raw score.

5. Add a code without a change in health. If a linked chart review adds a vascular disease code with an invented factor of 0.30, the raw score becomes 1.86, the payment score 1.86 / 1.079 × 0.941 = 1.6221, and the monthly payment rises by about USD 288, or about USD 3,450 a year, with no change in expected care.

6. Interpretation. The hierarchy stops two codes for one disease being paid twice, while unrelated diseases add up. Normalisation and the coding adjustment scale every score alike, moving the total paid without changing the ranking of enrolees. Each extra payment HCC is worth its factor, after normalisation and the coding adjustment, times the base rate whether or not care changes, which is the coding incentive in miniature.

Boundaries with neighbouring concepts

An HCC is a category, while a risk score is the number built from categories, demographics and interactions, and risk adjustment is the wider practice of paying or comparing with such scores. A diagnosis-related group classifies a hospital stay to price that episode, whereas an HCC describes a person's conditions over a year to predict later spending. Favourable selection, where enrolees cost less than their scores predict, is a separate problem from coding intensity, treated under adverse selection and cream skimming.

Misreadings and limits

Most errors come from reading a payment score as a clinical measure. Three recur.

1. Treating an HCC as a diagnosis. It is a payment category, and many recorded conditions, such as symptoms, map to no payment HCC.
2. Reading a higher score as a sicker person. The score rises with documentation as well as illness, which is why CMS applies a coding adjustment.
3. Expecting accurate prediction for individuals. An R-squared near 10% sets fair group averages but explains little of one person's spending, and Pope and colleagues found that even the full DCG/HCC model underpredicted for the most expensive beneficiaries.

Sources

• Centers for Medicare & Medicaid Services. Announcement of Calendar Year (CY) 2024 Medicare Advantage (MA) Capitation Rates and Part C and Part D Payment Policies. 31 March 2023. https://www.cms.gov/files/document/2024-announcement-pdf.pdf
• Centers for Medicare & Medicaid Services. Announcement of Calendar Year (CY) 2027 Medicare Advantage (MA) Capitation Rates and Part C and Part D Payment Policies. 6 April 2026. https://www.cms.gov/files/document/2027-announcement.pdf
• Ellis RP. Risk adjustment in health care markets: concepts and applications. In: Lu M, Jonsson E, eds. Paying for Health Care: New Ideas for a Changing Society. Wiley-VCH; 2008. Author manuscript dated February 2007. https://people.bu.edu/ellisrp/EllisPapers/2007_Ellis_Riskadjustment25.pdf
• Kautter J, Pope GC, Ingber M, Freeman S, Patterson L, Cohen M, Keenan P. The HHS-HCC risk adjustment model for individual and small group markets under the Affordable Care Act. Medicare & Medicaid Research Review. 2014;4(3):mmrr2014-004-03-a03. https://pmc.ncbi.nlm.nih.gov/articles/PMC4214270/
• Medicare Payment Advisory Commission. The Medicare Advantage program: status report. Report to the Congress: Medicare Payment Policy, March 2026, chapter 12. https://www.medpac.gov/wp-content/uploads/2026/03/Mar26_Ch12_MedPAC_Report_To_Congress_SEC.pdf
• Pope GC, Kautter J, Ellis RP, Ash AS, Ayanian JZ, Iezzoni LI, Ingber MJ, Levy JM, Robst J. Risk adjustment of Medicare capitation payments using the CMS-HCC model. Health Care Financing Review. 2004;25(4):119-141. https://pmc.ncbi.nlm.nih.gov/articles/PMC4194896/
• Code of Federal Regulations, title 42, section 425.605. Calculation of shared savings and losses under the BASIC track. eCFR, read 25 October 2026. https://www.ecfr.gov/current/title-42/chapter-IV/subchapter-B/part-425/subpart-G/section-425.605