Consent Safety Board: FHIR-Aware Consent Safety Passport

Inspiration

Surgical consent failures are not just documentation problems. Missing laterality, missing serious risks, weak patient understanding, and poor handoff can create patient safety, compliance, and operational risk.

The hackathon theme emphasizes the healthcare AI “Last Mile”: turning raw intelligence into actionable deliverables. Consent Safety Board was designed to demonstrate that last mile. Instead of producing a generic summary, it creates a clinician-ready Consent Safety Passport that can support pre-op review, patient understanding, and compliance handoff.

What it does

Consent Safety Board reviews a synthetic/de-identified surgical consent scenario for total knee replacement.

The workflow:

1. Reads the current synthetic patient context from a FHIR Observation in Prompt Opinion.
2. Coordinates three linked specialist agents:
   • Clinical Intake & Risk Agent
   • Patient Understanding Agent
   • Compliance & Final Synthesis Agent
3. Calls an MCP tool named generate_consent_safety_passport.
4. Produces a structured Consent Safety Passport.

The passport includes:

• Patient & FHIR context trace
• Agent evidence trace
• Consent adequacy score
• Wrong-site / laterality safety sentinel
• Consent gap severity matrix
• Critical and high-priority missing disclosures
• Recommended revised consent wording
• Pre-op action checklist
• Compliance handoff note
• Safety boundary and review-support disclaimer

How we built it

The project was built on Prompt Opinion using a patient-specific synthetic FHIR workflow.

FHIR layer: The synthetic surgical consent scenario is stored in a FHIR Observation. The Orchestrator retrieves it through GetPatientData(resourceType: Observation) and treats the patient context as read-only.

Agent workflow layer: A Consent Review Board Orchestrator coordinates three linked specialist agents:

• Clinical Intake & Risk Agent normalizes the case and identifies procedure-specific risk disclosure gaps.
• Patient Understanding Agent reviews readability, patient comprehension risks, and teach-back needs.
• Compliance & Final Synthesis Agent prioritizes compliance issues and generates final review-support findings.

MCP layer: A lightweight FastAPI-based MCP server exposes one tool: generate_consent_safety_passport. The Orchestrator passes patient context, consent text, specialist findings, and FHIR context into this MCP tool. The MCP tool returns a structured Consent Safety Passport artifact.

Safety layer: The workflow uses synthetic/de-identified data only, does not update patient records, does not call PatientEditor, and does not determine legal validity, clinical approval, diagnosis, treatment, or surgical clearance.

Challenges we ran into

The biggest challenge was turning a broad clinical review task into a stable, auditable, last-mile workflow.

We had to make the workflow reliable inside Prompt Opinion by avoiding unnecessary patient search, document retrieval, or patient editing. We found that the most stable demo path was to store the synthetic consent scenario in FHIR Observation.valueString and retrieve it through the current patient context.

Another challenge was output quality. A generic consent report was not enough. We upgraded the output into a Consent Safety Passport with a wrong-site laterality sentinel, severity matrix, revised consent wording, pre-op checklist, and compliance handoff.

We also added a lightweight MCP tool so the final deliverable is generated as a standardized artifact rather than only as free-form agent text.

Accomplishments that we're proud of

We successfully built a working healthcare workflow inside Prompt Opinion that combines FHIR context, specialist agents, and an MCP artifact generator.

The system reads a synthetic patient-specific FHIR Observation, coordinates multiple specialist agents, and generates a structured Consent Safety Passport.

The final passport flags a critical laterality gap, identifies missing high-severity risks such as DVT and pulmonary embolism, generates revised consent wording, and produces a practical pre-op action checklist and compliance handoff note.

What we learned

We learned that healthcare AI workflows need more than a strong model response. They need patient context, role-specific review, safety boundaries, evidence traceability, and actionable output.

We also learned that MCP is valuable not only for data retrieval, but also for standardizing clinical artifacts. In this project, the MCP tool converts multi-agent findings into a reusable Consent Safety Passport.

What's next for Consent Safety Board: FHIR-Aware Consent Safety Passport

Next steps include:

1. Supporting additional procedures such as hip replacement, spine surgery, colonoscopy with sedation, and cardiac catheterization.
2. Expanding FHIR support from Observation to DocumentReference for consent documents and ServiceRequest for planned procedures.
3. Adding institution-specific consent policy templates.
4. Supporting multilingual patient explanations on request.
5. Adding clinician feedback loops to refine scoring and recommendations.
6. Exporting Consent Safety Passports as structured artifacts for quality, risk, and compliance review.

Built With

• fastapi
• fhir-observation
• healthcare
• hugging-face-spaces
• json-rpc
• linked-specialist-agents
• model-context-protocol-(mcp)
• prompt-opinion
• prompt-opinion-a2a-agent
• python
• synthetic/de-identified-patient-data
• workflow