DisDash AI (formerly Discharge AI)

AI-assisted discharge coordination for NHS ward teams. Helps clinicians see who may go home today, what is blocking discharge, who owns each action, and what documentation still needs review.

MVP complete — ward dashboard, patient workspace, mock EPR snapshots, structured questionnaire, AI discharge plans, task/blocker tracking, draft discharge summaries, human approval, and audit logging. All data is fictional.

Not production-ready. AI outputs are draft-only. Final discharge decisions require authorised human approval. Never use real patient data in development.

---

Quick start

Prerequisites: Node.js 20+, Docker

npm install
docker compose up -d
npm run db:push
npm run db:seed
npm run dev

Open http://localhost:3000 and pick a demo user from the header role switcher (Doctor, Nurse, Pharmacist, Discharge Coordinator, Admin, etc.).

Copy .env.example → .env:

Variable	Default	Notes
DATABASE_URL	matches docker compose	PostgreSQL
AI_PROVIDER	mock	Set openai + OPENAI_API_KEY for real LLM
OPENAI_MODEL	gpt-4o-mini	Optional
SESSION_SECRET	—	Required in production

---

Workflow

EPR snapshot + questionnaire + free-text notes
        ↓
AI readiness summary / discharge plan
        ↓
Tasks, blockers, domain RAG status (RED / AMBER / GREEN)
        ↓
Draft discharge summary → clinician edit → approve document
        ↓
Final discharge plan approval → audit log

Routes: /wards/4A (dashboard) · /encounters/[id] (patient workspace)

Workspace tabs: Summary · Questionnaire · AI plan · Tasks & blockers · Documents · Approval · Audit log

---

Demo patients (ward 4A)

11 fictional patients (999… NHS numbers). Full list seeded via npm run db:seed.

Patient	Encounter	Scenario
Jane Demo	enc-H001	TTO not screened — E2E test patient
Robert Sample	enc-H002	Awaiting transport
Margaret Fictional	enc-H003	Awaiting OT
David Example	enc-H004	Care package pending
Susan – Thomas	enc-H005–H010	Consultant, family, not fit, care home, pharmacy, tomorrow
Arthur Mockwell	enc-H011	Full mock EPR — post-op cholecystectomy, renal mass follow-up, POC delay

Arthur Mockwell walkthrough (enc-H011)

1. Sign in as Discharge Coordinator or Admin.
2. Open Mockwell, Arthur (bed 11).
3. Summary — bloods, imaging, clinical notes timeline from mock EPR.
4. Generate AI discharge plan — RED status, care-package blocker, renal follow-up flagged.
5. Resolve blockers → Documents → generate & approve discharge summary.
6. Approval → final plan approval → check audit log.

---

Reset database

npm run db:seed upserts baseline data but does not clear AI plans, documents, audit events, or resolved blockers.

Full reset:

docker compose down -v && docker compose up -d
npm run db:push && npm run db:seed

Jane Demo only: npm run test:workflow:reset

---

Testing

Requires PostgreSQL running and npm run db:seed.

npm test                    # Vitest unit tests
npm run test:integration    # API task/blocker workflow
npm run test:e2e            # all Playwright tests
npm run test:workflow       # Jane Demo full discharge E2E
npm run test:workflow:ui    # headed, slow-mo
npm run test:unblock        # doctor RED→GREEN E2E

---

Stack

Next.js 15 · React 19 · TypeScript · Tailwind · Prisma · PostgreSQL · Vitest · Playwright

src/
  app/              # pages + API routes
  components/       # dashboard, workspace, approval UI
  server/
    ai/             # mock + OpenAI providers, schemas
    modules/        # encounters, plans, tasks, blockers, documents, audit
    integrations/   # EPR adapter boundary
    policy/         # approval gates
prisma/             # schema, seed, Arthur EPR data
tests/              # unit, integration, e2e

---

Safety principles

AI may: summarise context, flag missing information, suggest tasks, draft plans and documents, show uncertainty and source evidence.

AI must not: decide medical fitness, submit final documentation, change medications, send external communications, or act without audit logging.

Every AI output is labelled draft-only, editable, and requires human approval before clinical use.

---

Future roadmap

Near term: persist readiness summaries · per-patient reset scripts · richer OpenAI prompts · GP handover / patient advice drafts · ward metrics dashboard.

NHS EPR integration — replace MockEprAdapter (src/server/integrations/epr/adapter.ts) with a trust connector:

Trust EPR (Epic, Cerner, System C, …)
    → FHIR R4 (Patient, Encounter, Observation, DiagnosticReport, MedicationStatement)
    → ClinicalDataSnapshot + SourceEvidence
    → Discharge AI workflow (unchanged)

Phase	Goal
v1 ingest	Read-only scheduled + on-demand snapshot refresh
Provenance	AI citations linked to SourceEvidence rows
Auth	NHS Smartcard / trust IdP replacing demo session
Safety	DCB0129 hazard log, AI feedback, SUI hooks
v1 export	Approved discharge summary → EPR via FHIR DocumentReference (no autonomous write-back)
Later	Multi-ward views · social care handoff · SLA escalation · trust-hosted LLM (Azure NHS) · pilot KPIs

---

Scripts

Command	Description
npm run dev	Dev server (Turbopack)
npm run build	Production build
npm run db:push	Apply Prisma schema
npm run db:seed	Seed mock patients
npm run db:studio	Prisma Studio
npm run lint	ESLint

---

Licence & warnings

Hackathon / demo project. Not NHS-approved, CE-marked, or clinically deployed. Trust DPIA, clinical safety assessment, and EPR integration agreements are required before any live patient data.

Use fictional mock data only.