NHS - Automated Medical Triage

Fabric Score 3.9

PublicHealthNational Health Service UKBoth GenAI and Non-GenAI modelsUser / Operator / Front-line Staff ×1

Workflow Diagram

Fabric Score

Task

Patient submits a triage request for a clinical problem. Based on the clinical safety parameters and system capacity, the AI system books an appointment for the patient or re-directs them to external services.

Intent

The AI system aims to control patient flow and reduce burden on admin requests.

AI Workflow

Input

With consent, the patient (or admin team, on behalf of the patient), fills out the online form for triage. This is submitted to the AI system.

Process

Using clinician generated safety parameters, and based on system capacity, the AI system triages the request, and decides whether the patient should be offered an appointment (routine, urgent, F2F, telephone, nurse-led) or signposted to external services (pharmacy/A+E).

Output

The AI system books an appointment for the patient.

Human Oversight Level

Conditionally Autonomous AI

Institutional Oversight Examples

•Clinical safety parameters
•Medical device approval (UKCA Class 1 approved)
•Clinical Risk management (DCB0129 compliant)

Risk

Patients could be wrongly triaged and not seen as urgently as they should be. Forms may be seen by patients as obstruction to care.

Output Modification Telemetry

Unfaithful outputs (policy, tone, or style misalignment)65%
Missing or incomplete information10%
Irrelevant information10%
Internal inconsistency10%
Factual errors / hallucinations5%

Transversal Metrics

Grouped by Fabric dimension.

Efficacy

Accuracy90 %

Efficiency

Modification Rate15 %

Modification Time10 min/output

Verification Time5 min

Rejection Rate5 %

Operational Friction5 hrs/wk

Implementation Overhead100 hrs

Governance Overhead100 hrs

Time to Launch4 wks

Value

Effort Reduction90 %

Risk

Reliability5 incident/month

Autonomy85 %