Patientsim

For specific clinical cohorts, load the appropriate cohort skill from the table below.

Overview

PatientSim generates realistic synthetic clinical data for EMR/EHR testing, including:

• Patient demographics
• Encounters (inpatient, outpatient, emergency, observation)
• Diagnoses (ICD-10-CM)
• Procedures (CPT, ICD-10-PCS)
• Medications (with RxNorm codes)
• Lab results (with LOINC codes)
• Vital signs

Quick Start

Simple Patient

Request: "Generate a patient"

{
  "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"
  }
}

Clinical Cohort

Request: "Generate a diabetic patient with complications"

Claude loads diabetes-management.md and produces a complete clinical picture.

Cohort Skills

Load the appropriate cohort based on user request:

Generation Parameters

Output Entities

Patient

Demographics extending the Person model with MRN.

Encounter

Clinical visit with class (I/O/E/U/OBS), timing, location, providers.

Diagnosis

ICD-10-CM code with type (admitting, working, final), dates.

Medication

Drug with RxNorm code, dose, route, frequency, status.

LabResult

Test with LOINC code, value, units, reference range, abnormal flag.

VitalSign

Observation with temperature, HR, RR, BP, SpO2, height, weight.

See data-models.md for complete schemas.

Clinical Coherence Rules

PatientSim ensures generated data is clinically realistic:

1. Age-appropriate conditions: No pediatric conditions in adults, geriatric conditions require appropriate age
2. Gender-appropriate conditions: Prostate conditions for males only, pregnancy for females only
3. Medication indications: Drugs match diagnoses (metformin requires diabetes)
4. Lab coherence: Values align with conditions (elevated A1C with diabetes, BMP reflects renal status)
5. Temporal consistency: Diagnoses before treatments, labs after orders
6. Comorbidity patterns: Realistic comorbid conditions cluster together (e.g., diabetes + hypertension + CKD)

See validation-rules.md for complete rules.

Safety Guardrails

All PatientSim data is 100% synthetic (fictional and simulated). Enforce these rules at all times:

1. No real patient data. Never copy real medical records, real MRNs, or real SSNs into output. All identifiers must be generated.
2. No clinical advice. Output is test data, not medical guidance. Never phrase output as a recommendation for actual patient care.
3. No real provider NPIs in patient context. Use synthetic NPIs (prefix with 9999) unless explicitly pulling from NetworkSim reference data.
4. Validate code systems. Only emit ICD-10-CM codes from the current valid set (e.g., E11.65 not E11.999). Same for CPT, LOINC, and RxNorm -- use real codes, not invented ones.
5. PHI boundary. If a user supplies real patient details, refuse and explain that PatientSim generates synthetic data only.

Negative Examples (What NOT to Generate)

Edge Case Handling

Output Formats

Data Integration (PopulationSim)

Add geography (5-digit county FIPS or 11-digit tract FIPS) to ground generation in real CDC PLACES, SVI, and ADI data. See data-integration.md for full patterns, data sources, and provenance tracking.

Examples

Example 1: Basic Patient with Encounter

Request: "Generate a 45-year-old male with an office visit for hypertension"

Output:

{
  "patient": {
    "mrn": "MRN00000001",
    "name": { "given_name": "Michael", "family_name": "Johnson" },
    "birth_date": "1980-06-22",
    "gender": "M"
  },
  "encounter": {
    "encounter_id": "ENC0000000001",
    "patient_mrn": "MRN00000001",
    "class_code": "O",
    "status": "finished",
    "admission_time": "2025-01-15T09:30:00",
    "discharge_time": "2025-01-15T10:00:00",
    "chief_complaint": "Blood pressure follow-up"
  },
  "diagnoses": [
    {
      "code": "I10",
      "description": "Essential hypertension",
      "type": "final",
      "diagnosed_date": "2024-06-15"
    }
  ],
  "medications": [
    {
      "name": "Lisinopril",
      "code": "104376",
      "dose": "10 mg",
      "route": "PO",
      "frequency": "QD",
      "status": "active"
    }
  ],
  "vitals": {
    "observation_time": "2025-01-15T09:35:00",
    "systolic_bp": 138,
    "diastolic_bp": 88,
    "heart_rate": 72,
    "temperature": 98.4,
    "spo2": 98
  }
}

Example 2: Acute Inpatient Encounter

Request: "Create an inpatient admission for pneumonia"

Generates a hospital encounter with:

• Encounter class I (inpatient), admission and discharge dates
• Diagnosis: J18.9 (Pneumonia, unspecified organism) plus respiratory symptoms
• Procedures: chest X-ray (CPT 71046), blood cultures, CBC
• Labs: WBC, procalcitonin, BMP, blood gas
• Medications: antibiotics (ceftriaxone + azithromycin)
• Imaging results documenting chest infiltrates

Example 3: Complex Multi-Condition Patient

Request: "Generate a 68-year-old female with diabetes, hypertension, and CKD stage 3"

Claude combines patterns from multiple cohort skills to generate a coherent patient with:

• Multiple chronic diagnoses with appropriate onset dates
• Medications for each condition (metformin, lisinopril, etc.)
• Quarterly encounters over 2 years
• Labs showing disease progression (A1C, eGFR trends)
• Comorbidity interactions (CKD affecting medication choices)

Related Skills

All cohort sub-skills are listed in the Cohort Skills table above. Additional references:

• oncology-domain.md - Foundational oncology knowledge

Cross-Product: MemberSim (Claims)

PatientSim clinical encounters generate corresponding claims in MemberSim:

Integration Pattern: Generate clinical encounter in PatientSim first, then use MemberSim to create corresponding claims with matching dates, diagnoses, and procedures.

Cross-Product: RxMemberSim (Pharmacy)

PatientSim medication orders generate prescription fills in RxMemberSim:

Integration Pattern: Generate medication orders in PatientSim, then use RxMemberSim to model pharmacy fills with matching NDCs and appropriate fill timing.

Cross-Product: PopulationSim (Demographics & SDOH)

When geography is specified, PatientSim grounds generation in real CDC PLACES, SVI, and ADI data via PopulationSim. See data-integration.md for the full data-driven generation pattern, data files, and provenance tracking.

Cross-Product: NetworkSim (Provider Networks)

NetworkSim provides realistic provider and facility entities for clinical encounters:

Integration Pattern: Generate encounters in PatientSim first, then use NetworkSim to add realistic provider entities with proper NPIs, credentials, and hospital affiliations.

Cross-Product: TrialSim (Clinical Trials)

For patients enrolled in clinical trials:

• ../trialsim/therapeutic-areas/oncology.md - Oncology trial endpoints
• ../trialsim/therapeutic-areas/cardiovascular.md - CV outcomes trials
• ../trialsim/therapeutic-areas/cns.md - CNS trial assessments

Integration Pattern: Use PatientSim for clinical care journeys. When a patient enrolls in a trial, apply TrialSim skills for trial-specific data (RECIST, SDTM format, randomization).

Output Formats

• ../../formats/fhir-r4.md - FHIR transformation
• ../../formats/hl7v2-adt.md - HL7v2 ADT messages
• ../../formats/hl7v2-orm.md - HL7v2 Order messages
• ../../formats/hl7v2-oru.md - HL7v2 Results messages

Reference Data

• ../../references/oncology/ - Oncology codes, medications, regimens

Generative Framework Integration

PatientSim integrates with the Generative Framework for specification-driven generation at scale.

• Profile-Driven: "Use the Medicare diabetic profile to generate 100 patients" — samples demographics, generates clinical attributes, links to NetworkSim providers.
• Journey-Driven: "Add the diabetic first-year journey to each patient" — generates encounters over time, labs, medication changes, and complication branching.
• Cross-Domain Sync: Patient → MemberSim Member (via SSN), Encounter → Claim, Prescription → RxMemberSim Fill, Trial Subject → TrialSim Subject. See cross-domain-sync.md.