Skip to content

asbinthapa99/nhs-wait-intelligence

BranchesTags

Repository files navigation

NHS Wait Intelligence

Open-source decision intelligence platform for NHS elective waiting list analysis, inequality mapping, and recovery tracking.

Live: https://nhs-wait-intelligence.vercel.app
Backend: Railway (FastAPI + PostgreSQL)
Stack: Next.js 14 · FastAPI · PostgreSQL · Recharts · Claude AI

What It Does

The NHS has 7.6 million people on waiting lists. Official data exists but is buried in spreadsheets. NHS Wait Intelligence turns that data into actionable intelligence — free, open, and built entirely on public NHS England statistics.

Feature What it shows
National Overview Live backlog totals, wait-band breakdown, worst-performing regions
Regional Inequality Map Interactive choropleth — inequality scores, deprivation index, trend per region
Elective Recovery Tracker NHS 18-week and 52-week targets vs actual performance, regional progress bars
Region Comparison Side-by-side radar + bar charts for any two NHS regions
Trends & Forecasting Historical trend lines with 6-month ML forecasts per region
Clinical Specialties Backlog and year-on-year growth by specialty (orthopaedics, cardiology, etc.)
ICS Benchmarks Integrated Care System performance, sortable by inequality / backlog / trend
Anomaly Alerts Statistical outlier detection — regions deviating significantly from expected
Strategic Simulator Model interventions: add surgical teams, test policy scenarios, optimise resources
AI Insights Claude-powered streaming analysis with multi-turn conversation history
Autonomous SQL Agent LangChain Text-to-SQL agent — ask ad-hoc data questions in plain English
Patient Pathway Guide Wait estimates, provider comparison, GP letter drafts, and stay/switch advice
NHS News Feed AI-triaged RSS headlines from NHS England, BBC Health, and The Guardian

Architecture

┌─────────────────────────────────────────┐
│  Next.js 14 Frontend (Vercel)           │
│  33 pages · Recharts · Framer Motion    │
└────────────────┬────────────────────────┘
                 │ HTTPS / JSON
┌────────────────▼────────────────────────┐
│  FastAPI Backend (Railway)              │
│  13 routers · Rate limiting · AI cache  │
│  LangChain SQL agent · CSV export       │
└────────────────┬────────────────────────┘
                 │ SQLAlchemy ORM
┌────────────────▼────────────────────────┐
│  PostgreSQL (Neon)                      │
│  regions · trusts · waiting_lists       │
│  processed_metrics · forecasts          │
│  ai_cache                               │
└────────────────┬────────────────────────┘
                 │ Monthly ETL
┌────────────────▼────────────────────────┐
│  Python Pipeline                        │
│  NHS RTT ingestion · ONS seeding        │
│  CQC ratings · inequality scoring       │
│  6-month ML forecasting                 │
└─────────────────────────────────────────┘

Local Development

Requirements

  • Node 20+
  • Python 3.11+
  • PostgreSQL 16+

Frontend

cd frontend
npm ci
npm run dev          # http://localhost:3000

Backend

cd backend
cp .env.example .env  # fill in DATABASE_URL and ANTHROPIC_API_KEY
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Pipeline

cd pipeline
pip install -r requirements.txt
python run_pipeline.py

Environment Variables

Frontend (Vercel)

Variable Value
NEXT_PUBLIC_API_URL Your Railway backend URL

Backend (Railway)

Variable Required Notes
DATABASE_URL Yes PostgreSQL connection string (Neon recommended)
ANTHROPIC_API_KEY No Powers Claude AI + SQL agent; falls back to local engine if absent
GEMINI_API_KEY No Gemini Flash — first in AI provider cascade
GROQ_API_KEY No Groq Llama — second in AI provider cascade
OPENROUTER_API_KEY No OpenRouter Mistral — third in AI provider cascade
PORT Auto Set by Railway

Optional (enhanced features)

Variable Notes
REDIS_URL Enables Redis-backed caching; falls back to in-memory without it
SMTP_HOST / SMTP_USERNAME / SMTP_PASSWORD Pipeline failure email alerts

Data Sources

All data is from official public NHS England and ONS sources. No patient-level data is ever used.

File Source
pipeline/data/raw/rtt/*.csv NHS England RTT statistics
pipeline/data/raw/ons/population_by_region.csv ONS Population Estimates
pipeline/data/raw/ons/imd_by_region.csv English Indices of Deprivation 2019
pipeline/data/raw/ons/nhs_regions.geojson ONS NHS Regions Boundaries

API Reference

Core Analytics

GET  /health
GET  /api/overview
GET  /api/regions
GET  /api/regions/{id}
GET  /api/inequality
GET  /api/specialties
GET  /api/trends
GET  /api/anomalies
GET  /api/status/data
GET  /api/status/rtt-archive

Export

GET  /api/export/inequality.csv
GET  /api/export/specialties.csv
GET  /api/export/trends.csv

AI

POST /api/ai-explain          # Cached, multi-turn chat
POST /api/ai-stream           # Server-Sent Events streaming
GET  /api/ai-briefing         # Executive briefing (24 h cache)
GET  /api/ai-insights         # 3 proactive insight bullets
POST /api/agent-explain       # LangChain autonomous SQL agent

Patient Pathway

GET  /api/patient/choice-rights
GET  /api/patient/journey-guide
GET  /api/patient/preparation-guide
GET  /api/patient/local-summary?region=London
GET  /api/patient/area-compare?region=London
GET  /api/patient/providers?region=London&specialty=Cardiology
POST /api/patient/estimate
POST /api/patient/stay-switch
POST /api/patient/gp-helper
POST /api/patient/contact-guide

Simulation

POST /api/simulate
POST /api/optimize-resources
POST /api/scenarios

AI Features

Provider Cascade

When ANTHROPIC_API_KEY is set, AI endpoints use Claude for streaming analysis. The platform cascades through providers in order:

  1. Gemini 2.5 Flash (fastest)
  2. Groq Llama 3.1 (low latency)
  3. OpenRouter Mistral (fallback)
  4. Claude (Anthropic) (highest quality)
  5. Local NHS knowledge engine (always available, no key required)

Autonomous SQL Agent

The /api/agent-explain endpoint uses a LangChain Text-to-SQL agent backed by Claude Haiku. It autonomously queries the PostgreSQL database to answer ad-hoc questions in plain English:

"Which region has the highest percentage over 52 weeks this month?"
"How many trusts are in the Midlands with over 10,000 patients waiting?"

Requires ANTHROPIC_API_KEY and langchain-community + langchain-anthropic (both in requirements.txt).

Privacy

All questions passed to the SQL agent are scrubbed through a PII protection layer. Install presidio-analyzer and presidio-anonymizer to enable full Microsoft Presidio PII redaction in production. Without these packages the platform logs a warning and continues without redaction.

Testing

# Frontend build check
cd frontend && npm run build

# Backend tests (45 tests)
cd backend && pytest tests/ -v

# Pipeline tests
cd pipeline && pytest tests/ -q

Test coverage includes:

  • All core API endpoints (overview, regions, inequality, specialties, trends)
  • Patient pathway endpoints (choice rights, journey guide, estimates, stay/switch)
  • AI endpoints (explain, stream, briefing, insights, agent)
  • News service (RSS parsing, malformed feeds)
  • Services (inequality scoring, linear forecasting)
  • Empty and live-data states

Deployment

Service Platform Notes
Frontend Vercel Auto-deploys on push to main
Backend Railway Root directory: backend, Nixpacks builder
Database Neon (PostgreSQL) Serverless, connects via DATABASE_URL

Railway config (backend/nixpacks.toml):

[phases.setup]
nixPkgs = ["python312", "gcc", "libpq", "postgresql"]

[start]
cmd = "uvicorn app.main:app --host 0.0.0.0 --port $PORT"

Roadmap

  • Automated monthly RTT data pipeline (GitHub Actions — daily cron in .github/workflows/refresh.yml)
  • Postcode-level wait estimates
  • Email alerts when your region crosses a threshold (subscription endpoint live; scheduled job TBD)
  • Trust-level (not just region-level) breakdown on the map
  • GP referral pathway guidance with local provider lookup

Why This Project Matters

NHS Wait Intelligence addresses a high-impact national problem: long elective waiting times and unequal access across regions. The platform turns complex public datasets into practical decision support for service planning, policy analysis, and recovery monitoring.

Designed to be:

  • Public-interest focused — built on official open data, no patient-level data
  • Operationally useful — dashboards and exports support real planning workflows
  • Transparent — methods and sources are visible, reproducible, and auditable
  • Scalable — cloud deployment with continuous delivery and iterative improvement

Contributing

Built on public NHS data, published under MIT. No advertising, no data selling, no patient data.

If you work in NHS analytics, health policy, or public health tech — contributions, issues, and feedback are very welcome.

GitHub: https://github.com/asbinthapa99/nhs-wait-intelligence

About

AI-powered NHS waiting list inequality platform — full-stack Next.js + FastAPI + PostgreSQL

nhs-wait-intelligence.vercel.app

Resources

Readme

Contributing

Contributing
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages