Clinical Intelligence Node

A multi-agent virtual M&M (Morbidity & Mortality) Committee platform. Analyses clinical cases against tiered medical evidence. Produces structured MDT minutes with full audit trails.

---

⚠️ Disclaimer

All clinical cases used in development, testing, and demonstration of this platform are entirely fictional. They were created solely to test software functionality and do not represent any real patient, clinical encounter, or medical record. No real patient data has been used or stored at any point.

This tool is not a substitute for clinical judgement. It is a decision-support aid intended to augment, not replace, qualified medical professionals. Always involve appropriate clinical specialists before acting on any output.

---

Architecture

Three LLM agents collaborate on each case:

Agent	Model	Temp	Role
Researcher	gpt-4o-mini	0.0	PubMed multi-pillar evidence search + synthesis
MDT Roundtable	gpt-4o	0.25	360° specialist debate, gap-finding, risk heatmap
Auditor	gpt-4o-mini	0.0	Formal Markdown M&M minutes + audit trail

Evidence Ladder

1. RCT / Meta-analysis / Systematic Review
2. Practice Guidelines (NCCN / AHA / ESC / ESMO / NICE / WHO)
3. Clinical Trial / Observational Study
4. Case Reports (rare/emerging conditions only)

When an initial query returns fewer than 2 results the Researcher automatically triggers Multi-Pillar decomposition:

• Pillar A — The Conflict: intersection of the two main complications
• Pillar B — The Guidelines: primary life-threat guideline search
• Pillar C — The Procedure: interventional / procedural evidence

---

Features

• Free-text or SBAR case input via chat interface
• Dynamic MDT panel selection from 25+ specialties
• Mandatory invitee list (force any specialist into the roundtable)
• Risk Heatmap (High / Medium / Low per clinical risk)
• Critical Information Gap banner
• Evidence Matrix tab with clickable PubMed links
• Formal M&M minutes with versioned audit trail
• 72-hour automated refinement cycle with contradiction detection
• EPR copy-paste buffer + Markdown download

---

Setup

1. Clone

git clone https://github.com/your-username/clinical-node.git
cd clinical-node

2. Create virtual environment

python -m venv venv
# Windows
venv\Scripts\activate
# macOS / Linux
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure credentials

cp .env.example .env

Edit .env and add your keys:

Variable	Source
OPENAI_API_KEY	platform.openai.com
ENTREZ_EMAIL	Your email — required by NCBI
ENTREZ_API_KEY	NCBI account — optional, raises rate limit to 10 req/s

5. Run

streamlit run app.py

---

Project Structure

clinical-node/
├── app.py                  # Streamlit frontend
├── clinical_engine.py      # Multi-agent MDT pipeline
├── pubmed_client.py        # PubMed / Entrez search client
├── cron_refine.py          # 72-hour refinement cycle
├── state_manager.py        # cases.json CRUD + minutes persistence
├── .streamlit/
│   └── config.toml         # Theme (light, Clinic palette)
├── .env.example            # Credential template
├── requirements.txt
└── README.md

Runtime directories (gitignored):

minutes/        # Generated M&M minutes (Markdown, one file per case)
cases.json      # Case registry (created on first run)
.refine_state.json  # Refinement cycle timestamp

---

72-Hour Refinement

Run manually from the sidebar or on a schedule:

# Windows Task Scheduler / cron
python cron_refine.py

When new PubMed evidence contradicts recorded minutes, the case is flagged Needs Review and a Dissenting Opinion is appended to its minutes.

---

Infosec Notes

• .env is gitignored. Never commit API keys.
• cases.json and minutes/ are gitignored. No case data is tracked in version control.
• No data is sent anywhere other than the OpenAI API and NCBI Entrez API.
• No authentication layer is included. Do not expose this app on a public URL without adding one.

---

License

MIT