Project Overview
An NHS IAPT (Improving Access to Psychological Therapies) programme covering 280,000 referrals across 28 NHS T...
Technology Stack
Compliance & Standards
The Challenge
An NHS IAPT (Improving Access to Psychological Therapies) programme covering 280,000 referrals across 28 NHS Trusts needed an AI-assisted initial assessment triage tool -- using conversational AI to guide patients through PHQ-9 and GAD-7 assessment before allocation to the appropriate therapy step. MHRA Class IIa SaMD (AI aids therapist step allocation decision), DCB0129, NHS DTAC all 5 domains, NHS AI Lab 14 Standards, NICE ESF Tier 2 (informing clinical discussion -- not independent decision), UK GDPR Article 9 (mental health special category), ICO DPIA. Budget GBP120,000.
Our Approach
MHRA Class IIa SaMD
AI that aids therapist step allocation is Class IIa (assists clinical decision).
Class IIa pathway
UK Notified Body technical file (BSI Assurance UK), IEC 62304 Class B.
Conversational AI constraints
- 1AI never diagnoses (AI guides through validated questionnaire -- PHQ-9, GAD-7 -- it does not interpret results),
- 2AI never recommends treatment (AI produces scored questionnaire results for therapist review -- not a treatment recommendation),
- 3AI safety escalation (if patient expresses suicidal ideation during assessment -- immediate safety escalation to crisis service with human response),
- 4therapist review always required (AI triage aids step allocation, therapist confirms step),
- 5override mechanism (therapist can override AI-suggested step -- all overrides reviewed by clinical supervisor). PHQ-9 and GAD-7
Conversational Interface
Conversational AI for PHQ-9 and GAD-7: (1) NHS Login P5 authentication (patient authenticates before assessment), (2) conversational UI (React + WebSocket -- natural language question flow rather than form-filling), (3) PHQ-9 administration (9 validated depression questions -- AI presents in natural language, patient responds in natural language, AI maps response to PHQ-9 Likert scale 0-3), (4) GAD-7 administration (7 validated anxiety questions -- same pattern), (5) safety question (PHQ-9 Q9 -- thoughts of self-harm -- if positive, immediate safety escalation workflow).
NICE ESF Tier 2
AI informs clinical discussion rather than making independent decision -- compliant with Tier 2 evidence standard.
NICE evidence requirements for Tier 2
user acceptability, clinical safety, technical security (DTAC) -- no full clinical trial required.
AI Safety Escalation and Crisis Management
- Mental health AI safety -- critical design: (1) safety keyword detection (if patient types crisis keywords -- suicidal, self-harm, harm -- during assessment: immediate escalation), (2) safety escalation workflow (AI assessment pauses -- patient shown crisis resources: Samaritans 116 123, Crisis Text Line 85258, A&
- E), (3) therapist crisis notification (PagerDuty alert to duty therapist when crisis keywords detected -- therapist acknowledges within 30 minutes), (4) crisis assessment recording (all crisis escalations logged -- clinical supervisor review within 24 hours), (5) PHQ-9 Q9 positive protocol (Q9 score >
- 0 -- follow-up safety assessment questions -- clinical triage pathway, not generic AI response).
DCB0129
- crisis escalation is the primary clinical safety feature -- DCB0129 hazard log item 1.
- NHS DTAC Domain 1 --
NHS DTAC Domain 1 for mental health AI
- 1ICO DPIA mandatory (AI assessment of mental health = systematic profiling of special category data under Article 35),
- 2lawful basis (healthcare treatment -- Article 9(2)(h) -- IAPT is NHS healthcare provision),
- 3data minimisation (only PHQ-9/GAD-7 scores and session metadata stored -- not verbatim conversation transcripts),
- 4separate database schema (mental health AI assessment data encrypted with separate KMS key -- most sensitive NHS data category),
- 5retention (
NHS guidance
mental health records 8 years post-discharge).
ICO mental health guidance
clinical notes can be withheld from Subject Access Request if disclosure would harm patient -- AI assessment records are subject to the same protection.
The Results
MHRA Class IIa UKCA obtained (BSI).
DTAC all 5 domains approved.
NICE ESF Tier 2 confirmed.
Platform live at 22 weeks, GBP112,000. 280,000 referrals across 28 trusts.
AI assessment completion rate: 84.2% (vs 62.4% paper).
PHQ-9/GAD-7 accuracy vs therapist manual: 97.8% scoring agreement.
Step allocation agreement: 88.4%.
Crisis escalations: 48 in year one, all acknowledged by duty therapist within 30 minutes.
DCB0129 zero incidents.
AI Lab all 14.
UKCA obtained.
DTAC all 5.
NICE ESF Tier 2.
Crisis escalations all acknowledged within 30 minutes.
PHQ-9/GAD-7 accuracy 97.8%.
Step allocation agreement 88.4%.
The crisis escalation design -- AI assessment pauses at safety keywords, patient sees crisis resources, therapist gets PagerDuty alert within 30 seconds -- was the clinical safety design that DCB0129 required.
Mental health AI that does not detect and escalate suicidal ideation is not safe to deploy.
We made crisis detection the first design requirement, not an afterthought. -- Clinical Safety Officer, NHS IAPT Programme
Project Details
Related Case Studies
View AllNHS-Connected Mental Health App β DTAC Approved
NHS-Connected Mental Health App β DTAC Approved. A UK mental health technology company needed to build a patient-facing digital therapy support app connected to NHS syst...
Workforce Management SaaS β UK Healthcare Staffing Agency
A UK healthcare staffing agency placing 2,800 NHS-registered nurses, doctors, and allied health professionals ...
NHS Outpatient Referral System β FHIR R4 Integrated
An NHS Foundation Trust needed to replace a 2007 paper-based outpatient referral system. GPs referred patients...
Ready to Transform Your Business?
Let's discuss how our technical expertise can help you achieve remarkable results.