PRISMATIC: Prescription Risk Inspection System for Multi-Agent Tactical Interaction in Clinical Decision

A multi-agent architecture leveraging patient statements and clinical knowledge for prescription risk inspection.

Medication prescribing errors remain a critical challenge in clinical practice, often stemming from incomplete patient understanding, ambiguous documentation, and suboptimal decision support. In this paper, we propose PRISMATIC, a 3-layer multi-agent prescription risk mitigation framework designed to generate safe, interpretable, and traceable drug regimens by analyzing unstructured patient clinical note texts. To enhance adaptability and safety, PRISMATIC integrates two mechanisms: (1) Dynamic Self-updating Weight Adjustment (DSWA), which tunes risk factor weights over time, and (2) Differential Feedback Calibration Mechanism (DFCM), which learns from discrepancies with gold-standard prescriptions to improve future outputs. Evaluated on a curated dataset from MIMIC-IV, PRISMATIC outperforms raw LLM outputs and prompting-based baselines (Few-Shot, Chain-of-Thought, ReAct, Tree-of-Thoughts) in reducing prescription risks. These results highlight the potential of multi-agent systems for improving clinical medication decision support.

Setup Instructions

This project uses Poetry to manage dependencies and the virtual environment.

1. Install Poetry

You can install Poetry via pip:

pip install poetry

Or using the official installer (recommended):

curl -sSL https://install.python-poetry.org | python3 -

2. Install Dependencies

Clone & Navigate to the project directory and run:

git clone https://github.com/ZaneLing/PharmAid.git
cd PharmAid
poetry install

This will install all required packages and set up a virtual environment.

3. Set OpenAI API Key

Open .env file and set your OpenAI API key as an environment variable:

OPENAI_API_KEY = 'your_api_key'

4. Run the Workflow

To execute the main script within the Poetry environment, run:

poetry run python workflow.py