🏥 HealthTechOn Time📋 Fixed Price

HealthTech AI Clinical Documentation Platform — UK Acute Trust

UK6 min readUpdated August 2025
Region
UK
Contract
Fixed Price
Tech Stack
9 Technologies
IP
100% transferred

Project Overview

An NHS Acute Trust with 420 consultants and 1,200 clinical staff wanted to deploy an AI clinical documentation...

Technology Stack

ReactNode.js/FastifyPostgreSQLWhisper (OpenAIfine-tuned)Claude API (Anthropic)HAPI FHIR R4SystmOne HL7 v2AWS eu-west-2

Compliance & Standards

MHRA Class I SaMDNHS DSP Toolkit standard 4DTAC all 5 domainsDCB0129UK GDPR Article 9NHSX AI Lab guidanceWCAG 2.1 AACyber Essentials Plus
Step 01

The Challenge

An NHS Acute Trust with 420 consultants and 1,200 clinical staff wanted to deploy an AI clinical documentation assistant — transcribing clinical conversations (ward rounds, outpatient consultations, emergency assessments) into structured clinical notes and automatically populating the EPR (Electronic Patient Record). MHRA SaMD considerations (is the AI a medical device?), NHS DSP Toolkit standard 4 (AI governance), DCB0129 clinical safety, DTAC, UK GDPR Article 9 (health data processed by AI), NHSX AI Lab ethics guidance, and WCAG 2.1 AA were the requirements. Budget: £100,000.

Step 02

Our Approach

MHRA SaMD Classification for Clinical AI

The AI transcription tool processes clinical audio and generates clinical note text — the notes are reviewed and edited by a clinician before saving to EPR.

MHRA SaMD classification

clinical documentation assistant with mandatory clinician review = Class I (software does not drive clinical decisions — note is a draft for clinician to complete).

ClickMasters produced the MHRA software classification justification documentation

intended purpose (draft note generation, not clinical decision), clinician in the loop (cannot save without clinician review/approval).

AI Speech Recognition and Clinical NLP

Whisper (OpenAI) speech-to-text: medical vocabulary fine-tuned (NHS clinical vocabulary — procedure codes, anatomy, drug names).

SNOMED CT entity extraction

clinically-relevant entities (symptoms, diagnoses, medications, procedures) extracted and mapped to SNOMED CT UK codes.

Structured note generation

Claude API (Anthropic) — clinical conversation transcript → structured SOAP note (Subjective, Objective, Assessment, Plan).

Hallucination mitigation

AI cannot add information not present in the transcript — output is constrained to transcript content only.

UK GDPR Article 9 AI Processing Governance

Clinical audio is Article 9 special category health data.

AI processing legal basis

  • Article 9(2)(h) — provision of health care.
  • NHS DSP Toolkit standard 4 (Data-Driven Technology Assurance): algorithmic impact assessment, model documentation (model card — training data, limitations, performance metrics), and bias testing (demographic parity across patient age, ethnicity, and accent groups).

Bias testing

Whisper speech recognition accuracy across UK regional accents — additional fine-tuning for accents with below-threshold accuracy.

Mandatory clinician review

AI-generated draft cannot be saved to EPR without clinician editing and explicit approval.

Review interface

side-by-side (transcript and draft note), diff highlighting (AI additions highlighted for clinician attention), and one-click entity correction (SNOMED CT suggestion corrections).

EPR integration

FHIR R4 Clinical Impression resource → SystmOne or Epic via HL7 v2 MDM (Medical Document Management) message.

Audit

every draft generated, every edit made, every approval — immutable audit log.

Step 03

The Results

DTAC approved all 5 domains.

MHRA Class I classification confirmed.

Platform live at 16 weeks, £92,000 — under budget.

Clinical note documentation time: 8.2 minutes2.4 minutes (71% reduction).

Clinician edit rate: 64% of AI drafts required significant editing (expected — AI is a first draft, not a final note).

SNOMED CT coding accuracy: 91% entity match.

Transcript accuracy (Whisper fine-tuned): 97.3% word error rate below 5%.

NHS DSP Toolkit standard 4: assessment passed.

Client Testimonial
71% reduction in documentation time. 420 consultants × 71% time saving = real capacity for patient care. SNOMED CT 91% accuracy was the clinical governance committee's threshold. MHRA Class I classification correct — the mandatory clinician review is what makes this a documentation tool, not a decision support tool. DTAC first submission." — Chief Medical Information Officer, NHS Acute Trust (name withheld)
ClickMasters Case Study Team
Reviewed by James Whitmore, CTO

Project Details

Sector
HealthTech
Country
UK
Status
On Time
Contract
Fixed Price
Tech Stack
9 Technologies
Reading Time
6 min
IP Ownership
100% transferred
Last Updated
August 2025
Written By
ClickMasters Case Study Team
Reviewed By
James Whitmore, CTO

Related Case Studies

View All

Ready to Transform Your Business?

Let's discuss how our technical expertise can help you achieve remarkable results.