G CLI Agent ClinicalTrialMatching — Data skill for Claude Code

Agentic Clinical Trial Matching System

An AI-powered MVP that bridges legacy healthcare data and modern research protocols using **LangGraph**, **LlamaIndex**, and the **Model Context Protocol (MCP)**.

Scenario: The Clinical Recruitment Bottleneck

A pharmaceutical research organization manages hundreds of active clinical trials. Currently, specialized medical coordinators manually review thousands of patient electronic health records (EHR) to determine if they meet strict "Inclusion/Exclusion" criteria buried in 200-page Clinical Trial Protocols (PDFs).

The patient data resides in a legacy SQL-based Hospital Information System (HIS) that lacks modern APIs, creating a critical bottleneck in **Healthcare, Pharmaceuticals, and Biotechnology**. Delayed recruitment is the #1 reason trials fail, costing millions per day in lost patent life.

**Similar Scenarios:** Veteran affairs benefit adjudication, complex insurance underwriting, and legal discovery.

The Agentic Design

Agent Roles

• The Researcher (Medical Analyst): Uses RAG to parse the Trial Protocol PDF and an MCP Server to securely query the legacy patient database for specific lab results and ICD-10 codes.
• The Orchestrator (Clinical Lead): Receives the trial ID, sets recruitment goals, and manages the logic flow between data extraction and eligibility.
• The Executor (Regulatory Reporter): Compiles a "Match Justification Report" citing specific protocol pages and patient record timestamps.

The Math of Efficiency

Given $P = 10,000$ patients in a database:

• Human Manual Review: $H_{time} = 30 \text{ mins/patient}$ → 5,000 hours.
• Agentic Review: $A_{time} = 20 \text{ seconds/patient}$ → ~55 hours.
• Efficiency Ratio ($C_e$): Roughly 150:1 compared to professional human rates vs. token costs.

Technical Stack

• Languages: TypeScript (MCP Server), Python (Agent Logic).
• Orchestration: LangGraph (Stateful multi-agent workflows).
• **RAG Fra