FDE Playbook
HealthcareHospital network · 5 hospitals · 800 clinicians
Healthcare Triage Assistant
Customer brief
A hospital network wants an AI triage assistant that helps clinicians prioritise patient cases in the ED based on symptoms, vitals and history. Must not give diagnoses — only triage priority. Must integrate with the EHR.
Discovery questions
Ask these before designing a solution.
- 1What is the EHR system and does it expose an API?
- 2What is the triage protocol (ESI 5-level)?
- 3Who is the user — triage nurse, attending physician, both?
- 4What is the acceptable false-negative rate for high-acuity cases?
- 5Is PHI in scope? If so, what is the BAA with the model provider?
- 6What is the regulatory classification (FDA SaMD, clinical decision support)?
- 7How will clinicians override the system, and is override logged?
- 8What is the uptime requirement — is this life-critical or assistive?
Constraints
- PHI in scope — BAA required with model provider
- No diagnosis — triage priority only
- EHR integration via HL7 FHIR
- Audit log of every triage decision and clinician override
- High-acuity false-negative rate < 1%
- 99.95% uptime SLA
- Latency < 3s for triage suggestion
Requirements
- FHIR integration to pull patient context
- Structured triage output with priority + rationale + confidence
- Confidence threshold for escalation to attending
- Override logging with reason
- BAA-covered model provider or on-prem model
- Input guardrails (no free-text diagnoses from clinicians into prompt)
- Output guardrails (no diagnosis language)
- Online monitoring of false-negative rate
FDE Simulator
Decision 1 of 4
System Scope
Is this a decision support tool or an autonomous triage system?