hccinfhir (HCC in FHIR)

A Python library for extracting standardized service-level data from FHIR ExplanationOfBenefit resources, with a focus on supporting HCC (Hierarchical Condition Category) risk adjustment calculations.

Features

• Extract diagnosis codes, procedures, providers, and other key data elements from FHIR EOBs
• Support for both BCDA (Blue Button 2.0) and standard FHIR R4 formats
• Pydantic models for type safety and data validation
• Standardized Service Level Data (SLD) output format
• Multiple HCC model support (V22, V24, V28, ESRD V21, ESRD V24, RxHCC V08)
• Flexible input options: FHIR EOBs, service data, or direct diagnosis codes

Installation

pip install hccinfhir

Why FHIR-Based HCC Processing?

Risk Adjustment calculations traditionally rely on processed claims data, leading to information loss and reconciliation challenges. hccinfhir processes FHIR resources directly because:

• FHIR represents the source of truth with complete clinical and administrative data
• Risk Adjustment requires multiple data elements beyond diagnosis codes
• Direct processing eliminates data transformation errors and simplifies reconciliation

Data Model & Flexibility

While built for native FHIR processing, hccinfhir works with any data source that can be transformed into the SLD (Service Level Data) format:

sld = [{
    "procedure_code": "99214",
    "diagnosis_codes": ["E11.9", "I10"],
    "claim_type": "71",
    "provider_specialty": "01", 
    "service_date": "2024-01-15"
}, ...]

Or, for direct risk score calculation from a list of diagnosis codes, you only need the model name, diagnosis codes, and basic demographic factors:

from hccinfhir.model_calculate import calculate_raf

diagnosis_codes = ['E119', 'I509']  # Diabetes without complications, Heart failure
age = 67
sex = 'F'
model_name = "CMS-HCC Model V24"

result = calculate_raf(
   diagnosis_codes=diagnosis_codes,
   model_name=model_name,
   age=age,
   sex=sex
)

For more details on the SLD format, see the datamodels.py file.

Core Components

1. Extractor Module

Processes FHIR ExplanationOfBenefit resources to extract Minimum Data Elements (MDE):

from hccinfhir.extractor import extract_sld, extract_sld_list

sld = extract_sld(eob_data)  # Process single EOB

sld_list = extract_sld_list([eob1, eob2])  # Process multiple EOBs

2. Filter Module

Implements claim filtering rules:

• Inpatient/outpatient criteria - Type of Bill + Eligible CPT/HCPCS
• Professional service requirements - Eligible CPT/HCPCS
• Provider validation (Not in scope for this release, applicable to RAPS)

from hccinfhir.filter import apply_filter

filtered_sld = apply_filter(sld_list)

3. Logic Module

Implements core HCC calculation logic:

• Maps diagnosis codes to HCC categories
• Applies hierarchical rules and interactions
• Calculates final RAF scores
• Integrates with standard CMS data files

from hccinfhir.model_calculate import calculate_raf

diagnosis_codes = ['E119', 'I509']  # Diabetes without complications, Heart failure
result = calculate_raf(
   diagnosis_codes=diagnosis_codes,
   model_name="CMS-HCC Model V24",
   age=67,
   sex='F'
)

4. HCCInFHIR Class

The main processor class that integrates extraction, filtering, and calculation components:

from hccinfhir.hccinfhir import HCCInFHIR
from hccinfhir.datamodels import Demographics

# Initialize with custom configuration
hcc_processor = HCCInFHIR(
    filter_claims=True,                                    # Enable claim filtering
    model_name="CMS-HCC Model V28",                       # Choose HCC model version
    proc_filtering_filename="ra_eligible_cpt_hcpcs_2025.csv",  # CPT/HCPCS filtering rules
    dx_cc_mapping_filename="ra_dx_to_cc_2025.csv"         # Diagnosis to CC mapping
)

# Define beneficiary demographics
demographics = {
    age=67,
    sex='F'
}

# Method 1: Process FHIR EOB resources
raf_result = hcc_processor.run(eob_list, demographics)

# Method 2: Process service level data
service_data = [{
    "procedure_code": "99214",
    "claim_diagnosis_codes": ["E11.9", "I10"],
    "claim_type": "71",
    "service_date": "2024-01-15"
}]
raf_result = hcc_processor.run_from_service_data(service_data, demographics)

# Method 3: Direct diagnosis processing
diagnosis_codes = ['E119', 'I509']
raf_result = hcc_processor.calculate_from_diagnosis(diagnosis_codes, demographics)

# RAF Result contains:
print(f"Risk Score: {raf_result['risk_score']}")
print(f"HCC List: {raf_result['hcc_list']}")
print(f"CC to Diagnosis Mapping: {raf_result['cc_to_dx']}")
print(f"Applied Coefficients: {raf_result['coefficients']}")
print(f"Applied Interactions: {raf_result['interactions']}")

The HCCInFHIR class provides three main processing methods:

1. run(eob_list, demographics): Process FHIR ExplanationOfBenefit resources

   • Extracts service data from FHIR resources
   • Applies filtering rules if enabled
   • Calculates RAF scores using the specified model

2. run_from_service_data(service_data, demographics): Process standardized service data

   • Accepts pre-formatted service level data
   • Validates data structure using Pydantic models
   • Applies filtering and calculates RAF scores

3. calculate_from_diagnosis(diagnosis_codes, demographics): Direct diagnosis processing

   • Processes raw diagnosis codes without service context
   • Useful for quick RAF calculations or validation
   • Bypasses service-level filtering

Each method returns a RAFResult containing:

• Final risk score
• List of HCCs
• Mapping of condition categories to diagnosis codes
• Applied coefficients and interactions
• Processed service level data (when applicable)

Testing

$ python3 -m hatch shell
$ python3 -m pip install -e .
$ python3 -m pytest tests/*

Dependencies

• Pydantic >= 2.10.3
• Standard Python libraries

Research: FHIR BCDA and 837 Field Mapping Analysis

Core Identifiers

Field	837 Source	FHIR BCDA Source	Alignment Analysis
claim_id	CLM01 segment	eob.id	✓ Direct mapping
patient_id	NM109 when NM101='IL'	eob.patient.reference (last part after '/')	✓ Aligned but different formats

Provider Information

Field	837 Source	FHIR BCDA Source	Alignment Analysis
performing_provider_npi	NM109 when NM101='82' and NM108='XX'	careTeam member with role 'performing'/'rendering'	✓ Aligned
billing_provider_npi	NM109 when NM101='85' and NM108='XX'	contained resources with NPI system identifier	✓ Conceptually aligned
provider_specialty	PRV03 when PRV01='PE'	careTeam member qualification with specialty system	✓ Aligned but different code systems

Claim Type Information

Field	837 Source	FHIR BCDA Source	Alignment Analysis
claim_type	GS08 (mapped via CLAIM_TYPES)	eob.type with claim_type system	✓ Aligned but different coding
facility_type	CLM05-1 (837I only)	facility.extension with facility_type system	✓ Aligned for institutional claims
service_type	CLM05-2 (837I only)	extension or eob.type with service_type system	✓ Aligned for institutional claims

Service Line Information

Field	837 Source	FHIR BCDA Source	Alignment Analysis
procedure_code	SV1/SV2 segment, element 2	item.productOrService with pr system	✓ Aligned
ndc	LIN segment after service line	item.productOrService with ndc system or extension	✓ Aligned but different locations
quantity	SV1/SV2 element 4	item.quantity.value	✓ Direct mapping
quantity_unit	SV1/SV2 element 5	item.quantity.unit	✓ Direct mapping
service_date	DTP segment with qualifier 472	item.servicedPeriod or eob.billablePeriod	✓ Aligned
place_of_service	SV1 element 6	item.locationCodeableConcept with place_of_service system	✓ Aligned
modifiers	SV1/SV2 segment, additional qualifiers	item.modifier with pr system	✓ Aligned

Diagnosis Information

Field	837 Source	FHIR BCDA Source	Alignment Analysis
linked_diagnosis_codes	SV1/SV2 diagnosis pointers + HI segment codes	item.diagnosisSequence + diagnosis lookup	✓ Aligned but different structure
claim_diagnosis_codes	HI segment codes	diagnosis array with icd10cm/icd10 systems	✓ Aligned

Additional Fields

Field	837 Source	FHIR BCDA Source	Alignment Analysis
allowed_amount	Not available in 837	item.adjudication with 'eligible' category	⚠️ Only in FHIR

Key Differences and Notes

1. Structural Differences:

   • 837 uses a segment-based approach with positional elements
   • FHIR uses a nested object structure with explicit systems and codes

2. Code Systems:

   • FHIR explicitly defines systems for each code (via SYSTEMS constant)
   • 837 uses implicit coding based on segment position and qualifiers

3. Data Validation:

   • FHIR implementation uses Pydantic models for validation
   • 837 implements manual validation and parsing

4. Diagnosis Handling:

   • 837: Direct parsing from HI segment with position-based lookup
   • FHIR: Uses sequence numbers and separate diagnosis array

5. Provider Information:

   • 837: Direct from NM1 segments with role qualifiers
   • FHIR: Through careTeam structure with role coding

TODO: Enhancement Suggestions

1. Consider adding validation for code systems in 837 parser to match FHIR's explicitness
2. Standardize date handling between both implementations
3. Add support for allowed_amount in 837 if available in different segments
4. Consider adding more robust error handling in both implementations

Data Files

ra_dx_to_cc_mapping_2025.csv

SELECT diagnosis_code, cc, model_name 
FROM ra_dx_to_cc_mapping 
WHERE year = 2025 and model_type = 'Initial';

ra_hierarchies_2025.csv

SELECT cc_parent, 
  cc_child, 
  model_domain, 
  model_version, 
  model_fullname
FROM ra_hierarchies
WHERE eff_last_date > '2025-01-01';

ra_coefficients_2025.csv

SELECT coefficient, value, model_domain, model_version 
FROM ra_coefficients 
WHERE eff_last_date > '2025-01-01';

ra_eligible_cpt_hcpcs_2025.csv

SELECT DISTINCT cpt_hcpcs_code
FROM mimi_ws_1.cmspayment.ra_eligible_cpt_hcpcs
WHERE is_included = 'yes' AND YEAR(mimi_src_file_date) = 2024;

Contributing

Join us at mimilabs. Reference data available in MIMILabs data lakehouse.

Publishing (only for those maintainers...)

$ python3 -m hatch build
$ python3 -m hatch publish

License

Apache License 2.0