HealthSim - Synthetic Healthcare Data Generation

Quick Start

Generate Clinical Data

Request: "Generate a 65-year-old diabetic patient with hypertension"

Claude will produce a patient with:

• Demographics (age 65, realistic name/address)
• Diagnoses (E11.9 Type 2 diabetes, I10 hypertension)
• Medications (metformin, lisinopril)
• Labs (A1C, BMP with values in expected ranges)
• Comorbidities (likely hyperlipidemia, possible obesity)

Generate Claims

Request: "Create a professional claim for an office visit"

Claude will produce:

• Claim header (provider NPI, member ID, service date)
• Service lines (CPT 99213/99214, charges)
• Diagnoses (ICD-10 codes)
• Adjudication (allowed, paid, patient responsibility)

Generate Pharmacy Data

Request: "Generate a pharmacy claim that triggers a drug interaction alert"

Claude will produce:

• Prescription details (NDC, quantity, days supply)
• Pharmacy claim (BIN, PCN, cardholder ID)
• DUR alert (DD code, clinical significance, recommendation)
• Claim response (approved with warning or rejected)

Scenario Skills

PatientSim Scenarios

Load these for clinical data generation:

See: skills/patientsim/ for detailed skills

MemberSim Scenarios

Load these for claims and payer data:

See: skills/membersim/ for detailed skills

RxMemberSim Scenarios

Load these for pharmacy and PBM data:

See: skills/rxmembersim/ for detailed skills

TrialSim Scenarios

Load these for clinical trial data generation:

See: skills/trialsim/ for detailed skills

PopulationSim Scenarios

Load these for population intelligence and cohort definition:

Key Differentiator: PopulationSim analyzes real population data (Census, CDC) and outputs specifications, not synthetic records. These specs drive realistic generation in PatientSim, MemberSim, and TrialSim.

See: skills/populationsim/ for detailed skills

NetworkSim Scenarios

Load these for provider network knowledge and entity generation:

See: skills/networksim/ for detailed skills

Generative Framework

Load these for batch generation and specification-driven data creation:

Key Differentiator: The Generative Framework enables specification-driven generation at scale. Build a profile, define a journey, then execute to generate hundreds or thousands of correlated entities across all products.

See: skills/generation/ for detailed skills

Common Skills

Cross-product infrastructure skills:

See: skills/common/ for detailed skills

Output Formats

Default: JSON

By default, Claude outputs data as JSON objects that match the canonical data model.

Healthcare Standards

Request specific formats:

See: formats/ for transformation skills

Analytics Formats

See: formats/dimensional-analytics.md for star schema details

Export Formats

Generation Parameters

Demographics

Clinical (PatientSim)

Claims (MemberSim)

Pharmacy (RxMemberSim)

Reproducibility

For consistent results across sessions:

Request: "Generate 10 patients using seed 42"

Claude will:

1. Use seed 42 for all random selections
2. Generate identical output if same parameters used
3. Note the seed in output for reference

Validation

Claude automatically validates generated data for:

• Structural: Required fields, data types, formats
• Temporal: Date ordering (discharge after admission, etc.)
• Referential: Foreign key relationships
• Clinical: Age-appropriate conditions, gender-appropriate conditions
• Business: Valid code combinations, realistic pricing

Request explicit validation: "Validate this patient data"

Reference Data

For code lookups and documentation:

Format Transformations

Transform generated data to healthcare standards:

Healthcare Standards

Export Formats

Integration & Testing

Examples

Example 1: Simple Patient

Request: "Generate a patient"

Output:

{
  "mrn": "MRN00000001",
  "name": {
    "given_name": "John",
    "family_name": "Smith"
  },
  "birth_date": "1975-03-15",
  "gender": "M",
  "address": {
    "street_address": "123 Main Street",
    "city": "Springfield",
    "state": "IL",
    "postal_code": "62701"
  }
}

Example 2: Clinical Scenario

Request: "Generate a 58-year-old female with poorly controlled Type 2 diabetes"

Output:

{
  "patient": {
    "mrn": "MRN00000002",
    "name": { "given_name": "Maria", "family_name": "Garcia" },
    "birth_date": "1967-08-22",
    "gender": "F"
  },
  "diagnoses": [
    { "code": "E11.65", "description": "Type 2 diabetes with hyperglycemia" },
    { "code": "I10", "description": "Essential hypertension" },
    { "code": "E78.5", "description": "Hyperlipidemia" }
  ],
  "medications": [
    { "name": "Metformin", "dose": "1000 mg", "frequency": "BID" },
    { "name": "Glipizide", "dose": "10 mg", "frequency": "BID" },
    { "name": "Lisinopril", "dose": "20 mg", "frequency": "QD" }
  ],
  "labs": [
    { "test": "HbA1c", "value": "9.8", "unit": "%", "flag": "H" },
    { "test": "Glucose", "value": "245", "unit": "mg/dL", "flag": "H" },
    { "test": "Creatinine", "value": "1.2", "unit": "mg/dL" }
  ]
}

Example 3: Professional Claim

Request: "Generate a paid professional claim for an office visit"

Output:

{
  "claim": {
    "claim_id": "CLM20250115000001",
    "claim_type": "PROFESSIONAL",
    "member_id": "MEM001234",
    "provider_npi": "1234567890",
    "service_date": "2025-01-15",
    "place_of_service": "11",
    "principal_diagnosis": "E11.9",
    "claim_lines": [
      {
        "line_number": 1,
        "procedure_code": "99214",
        "charge_amount": 175.00,
        "units": 1
      }
    ]
  },
  "adjudication": {
    "status": "paid",
    "allowed_amount": 125.00,
    "paid_amount": 100.00,
    "deductible": 0.00,
    "copay": 25.00,
    "coinsurance": 0.00
  }
}

Example 4: Pharmacy Claim with DUR

Request: "Generate a pharmacy claim that gets rejected for early refill"

Output:

{
  "claim": {
    "claim_id": "RX20250115000001",
    "transaction_code": "B1",
    "ndc": "00071015523",
    "drug_name": "Atorvastatin 20mg",
    "quantity": 30,
    "days_supply": 30,
    "service_date": "2025-01-15"
  },
  "response": {
    "status": "rejected",
    "reject_code": "79",
    "reject_message": "Refill Too Soon",
    "dur_alert": {
      "type": "ER",
      "message": "Refill 12 days early (before 80% used)",
      "previous_fill_date": "2024-12-27",
      "days_early": 12
    }
  }
}

Tips

1. Be specific: "diabetic patient with A1C of 9.5" beats "sick patient"
2. Request format early: "Generate as FHIR..." rather than converting after
3. Use seeds: For reproducible test data across sessions
4. Batch large requests: "Generate 100 in batches of 20"
5. Validate sensitive data: Request validation for production-like scenarios

Disclaimer

HealthSim generates synthetic test data only. It is not a clinical decision support system and does not provide medical advice, diagnosis recommendations, or treatment guidance.

Intended uses:

• Software development and testing
• System integration validation
• Training and educational demonstrations
• Performance and load testing

Not intended for:

• Clinical decision support
• Medical advice or treatment recommendations
• Actual patient care
• Processing real PHI

The clinical patterns, medication regimens, and lab values reflect general healthcare conventions suitable for test data. They do not account for individual patient circumstances or the full complexity of clinical practice.