Play 46 — Healthcare Clinical AI

HIPAA-compliant clinical AI — PHI de-identification with Presidio, ICD-10/CPT medical coding, drug interaction checking grounded in FDA data, patient risk scoring via FHIR, clinical NLP decision support with mandatory disclaimers, and immutable audit trails.

Architecture

Component	Azure Service	Purpose
Clinical NLP	Azure OpenAI (GPT-4o, HIPAA-eligible)	Clinical decision support, coding, Q&A
De-Identification	Presidio (local)	Strip PHI before any AI processing
Patient Data	Azure Health Data Services (FHIR R4)	Patient records, conditions, medications
Drug Database	FDA drug interaction DB (grounding)	Evidence-based interaction checking
Hosting	Azure Container Apps (private endpoints)	Clinical AI API
Secrets	Azure Key Vault (CMK encryption)	API keys, FHIR credentials
Audit Trail	Immutable Blob Storage	HIPAA-compliant access logging

📐 Full architecture details

Aspect	Play 35 (Compliance Engine)	Play 46 (Healthcare Clinical AI)	Play 01 (Enterprise RAG)
Domain	General compliance	Healthcare / clinical medicine	Enterprise knowledge
Regulation	GDPR, SOC 2, EU AI Act	HIPAA, BAA, PHI protection	General data privacy
Data	Policy documents	Patient records (FHIR R4)	Corporate documents
AI Task	Gap detection	Clinical coding, drug interactions	Q&A with citations
Safety	Compliance scoring	Clinical safety (0% harmful advice)	Groundedness
De-identification	PII detection	PHI de-identification (18 HIPAA types)	PII redaction

DevKit Structure


46-healthcare-clinical-ai/
├── agent.md                                # Root orchestrator with handoffs
├── .github/
│   ├── copilot-instructions.md             # Domain knowledge (<150 lines)
│   ├── agents/
│   │   ├── builder.agent.md                # Clinical NLP + de-id + FHIR
│   │   ├── reviewer.agent.md               # HIPAA audit + PHI + safety
│   │   └── tuner.agent.md                  # De-id recall + coding + cost
│   ├── prompts/
│   │   ├── deploy.prompt.md                # Deploy HIPAA-compliant pipeline
│   │   ├── test.prompt.md                  # Test with synthetic data
│   │   ├── review.prompt.md                # HIPAA compliance audit
│   │   └── evaluate.prompt.md              # Measure clinical accuracy
│   ├── skills/
│   │   ├── deploy-healthcare-clinical-ai/  # BAA + FHIR + Presidio + private endpoints
│   │   ├── evaluate-healthcare-clinical-ai/# De-id recall, ICD-10, drugs, safety, HIPAA
│   │   └── tune-healthcare-clinical-ai/    # Entity config, prompts, grounding, audit
│   └── instructions/
│       └── healthcare-clinical-ai-patterns.instructions.md
├── config/                                 # TuneKit
│   ├── openai.json                         # Clinical model (temp=0, deterministic)
│   ├── guardrails.json                     # PHI entities, audit rules, safety
│   └── agents.json                         # FHIR config, consent, context
├── infra/                                  # Bicep IaC
│   ├── main.bicep
│   └── parameters.json
└── spec/                                   # SpecKit
    └── fai-manifest.json

Quick Start


# 1. Deploy HIPAA-compliant infrastructure (verify BAA first!)
/deploy
 
# 2. Test with synthetic patient data
/test
 
# 3. Run HIPAA compliance audit
/review
 
# 4. Measure clinical accuracy
/evaluate

Key Metrics

Metric	Target	Description
PHI Recall	> 98%	PHI entities correctly detected (HIPAA critical)
ICD-10 Category Accuracy	> 90%	Correct code category assignment
Drug Interaction Detection	> 95%	Known interactions correctly flagged
Hallucination Rate	< 1%	Made-up clinical information
Harmful Advice Rate	0%	Clinically dangerous recommendations (non-negotiable)
HIPAA Audit Coverage	100%	All queries logged (de-identified)

Estimated Cost

Service	Dev/mo	Prod/mo	Enterprise/mo
Azure OpenAI	$50	$400	$1,500
Azure Health Data Services (FHIR)	$30	$150	$500
Azure AI Search	$0	$250	$800
Azure AI Content Safety	$0	$60	$200
Azure App Service	$15	$80	$250
Cosmos DB	$5	$75	$400
Key Vault	$1	$10	$25
Application Insights	$0	$30	$100
Total	$101	$1,055	$3,775

Estimates based on Azure retail pricing. Actual costs vary by region, usage, and enterprise agreements.

💰 Full cost breakdown

WAF Alignment

Pillar	Implementation
Security	HIPAA BAA, private endpoints, CMK encryption, PHI de-identification
Responsible AI	Clinical disclaimers, evidence grounding, 0% harmful advice
Reliability	Deterministic (temp=0), seed=42, reproducible clinical responses
Cost Optimization	gpt-4o-mini for grounded lookups, local Presidio, FHIR caching
Operational Excellence	Immutable audit trail, FHIR AuditEvent export, 7-year retention
Performance Efficiency	Patient context caching, batch FHIR queries

FAI Manifest

Field	Value
Play	`46-healthcare-clinical-ai`
Version	`1.0.0`
Knowledge	R2-RAG-Architecture, T2-Responsible-AI, O2-Agent-Coding, T3-Production-Patterns, R3-Deterministic-AI
WAF Pillars	security, responsible-ai, reliability, operational-excellence, cost-optimization
Groundedness	≥ 85%
Safety	0 violations max