How HCC Engine Works

HCC Engine is an AI-powered clinical coding assistant that automatically extracts Hierarchical Condition Categories (HCCs) from clinical documentation. It uses:

• Clinical NLP - Natural Language Processing to identify diagnoses, symptoms, and medical conditions in free-text clinical notes
• CMS-HCC V28 2025 Midyear Model - The latest official CMS mappings with 115 payment HCCs and 7,903 ICD-10 codes
• Evidence-Based Decision Trees - YAML-based clinical rules for accurate code selection
• AI Verification - Optional LLM-powered code verification for quality assurance

The HCC Engine processes clinical notes through multiple stages:

1. Text Analysis: Clinical note is parsed for medical terminology, negation detection, and section identification (Assessment, HPI, etc.)
2. Condition Extraction: Diagnoses are matched against our terminology database of 15,000+ clinical terms
3. ICD-10 Mapping: Conditions are mapped to specific ICD-10 codes using decision tree logic
4. HCC Assignment: ICD-10 codes are mapped to HCCs per CMS V28 specifications
5. Hierarchy Application: CMS hierarchy rules are applied (higher severity trumps lower)
6. RAF Calculation: Final RAF is computed including interaction bonuses

An HCC is a grouping of related ICD-10 diagnosis codes that indicate similar levels of expected healthcare costs. CMS uses HCCs to:

• Adjust capitated payments to Medicare Advantage plans
• Account for the health status and expected costs of beneficiaries
• Ensure plans caring for sicker patients receive appropriate payment

The CMS-HCC V28 model (2025 Midyear) includes 115 payment HCCs mapped from 7,903 ICD-10 codes organized into disease groups:

CMS-HCC V28 is the current version of the CMS Hierarchical Condition Category model, phased in starting 2024:

• V28: Version 28 of the HCC model, replacing V24
• 2025 Midyear: The midyear update with finalized ICD-10 to HCC mappings effective for 2025
• 115 HCCs: Reduced from V24's 86 HCCs, but reorganized for better clinical alignment

The RAF (Risk Adjustment Factor) is a numerical score representing a patient's expected healthcare costs relative to an average Medicare beneficiary:

• RAF = 1.0: Average expected cost
• RAF > 1.0: Higher than average expected cost
• RAF < 1.0: Lower than average expected cost

The RAF score directly impacts Medicare Advantage plan payments. A patient with RAF 1.5 generates 50% more payment than RAF 1.0.

Total RAF is calculated as:

1. Base RAF: Sum of individual HCC coefficients after hierarchy application

2. Interaction Bonus: Additional RAF when certain HCC combinations are present together (e.g., Diabetes + Heart Failure)

3. Count Bonus (D1-D10+): Bonus for having multiple HCCs:

CMS recognizes that certain disease combinations have synergistic cost impacts. The V28 model includes these interaction pairs:

• Diabetes + CHF: HCC 37/38 + HCC 221-226
• Diabetes + CKD: HCC 37/38 + HCC 326-329
• CHF + CKD: HCC 221-226 + HCC 326-329
• CHF + COPD: HCC 221-226 + HCC 280
• Diabetes + CHF + CKD: Three-way interaction bonus

Each interaction adds a coefficient to the total RAF when both/all conditions are present.

Hierarchies are CMS rules that prevent double-counting related conditions. Within a hierarchy group, only the highest-severity HCC pays RAF.

Key hierarchy groups in V28:

Important distinction: CMS counts unique HCC numbers before hierarchy for the count bonus, but applies hierarchy for RAF calculation.

This is why the display shows "5 Unique HCCs" even when one is marked as ↓HIER - the count bonus uses pre-hierarchy unique HCC count as a proxy for disease burden.

AI Verification uses Large Language Models (Claude or GPT-4) to validate extracted codes against the clinical note. The AI:

• Confirms the diagnosis is explicitly documented (not just mentioned)
• Checks for negation or hypothetical language
• Suggests more specific codes when available
• Flags codes that need human review

Verification statuses:

• Verified: Code is appropriate based on documentation
• Rejected: Insufficient evidence or negated
• Upgraded: A more specific code is suggested
• Needs Review: Requires human judgment

Each HCC family has a YAML-based decision tree that guides code selection:

1. Prerequisites: Basic criteria that must be met (e.g., "diagnosis must be chronic")
2. Steps: Sequential decision points (e.g., "does patient have complications?")
3. Branches: Specific paths leading to final code selection
4. Evidence: Required documentation elements (labs, medications, imaging)

Example for Diabetes (E11.x):