🩺 Healthcare AI Clinical Decision Support System

An advanced, industry-grade clinical decision support system (CDSS) built on LangGraph, OpenAI, and PubMed, designed for high-precision patient monitoring, risk stratification, and evidence-based treatment synthesis.

🚀 Overview

This system serves as a world-class clinical co-pilot, empowering healthcare providers with real-time, data-driven insights. It leverages a sophisticated agentic workflow to analyze patient biometrics, predict multi-disease risks (Oncology, Infectious, Chronic), and generate personalized clinical road maps backed by rigorous medical literature.

🛠️ High-Performance Architecture

The system is powered by a Non-Linear StateGraph architecture, optimizing for both speed and clinical rigor through conditional triage and parallel processing.

🔀 Intelligent Workflow Orchestration

• Clinical Triage Router: Dynamically stratifies patients. High-risk profiles trigger an intensive clinical research track, while low-risk profiles are triaged to a rapid "Wellness Optimization" path.
• Parallel Treatment Tracks: Executes Medication Prescriptions and Lifestyle Advice generation simultaneously, significantly reducing latency and mirroring specialized clinical workflows.
• Drug Safety Guardrails: An automated validation layer integrating OpenFDA and RxClass APIs to check for boxed warnings, drug-drug interactions, and patient-specific contraindications (e.g., Metformin in hypoglycemia).

📚 Advanced RAG (Retrieval-Augmented Generation)

• Multi-Source Fetching: Integrates PubMed (E-Utilities) for academic literature and TavilySearch for real-time clinical guidelines.
• Corrective RAG (C-RAG): Implements a relevance-based filtering mechanism that automatically falls back to web retrieval if PubMed results are deemed ambiguous or insufficient.

🏗️ Core Components

1. Risk Stratification Node (early_disease_detection)

Leverages GPT-4o-mini with structured outputs to identify potential risks across:

• Chronic: Diabetes, Cardiovascular, Hypertension, Metabolic.
• Oncology: Hematological markers (WBC, Platelets) and constitutional symptoms.
• Infectious: Acute markers (Temp, SpO2, Resp Rate) for COVID-19 and viral/bacterial screening.

2. Clinical Evidence Engine (fetch_medical_literature)

A state-of-the-art literature synthesis pipeline:

• Vector Store: FAISS-based similarity search on top of PubMed abstracts.
• Refinement: Sentence-level decomposition and clinical summarization.

3. Safety-First Prescription (drug_safety_guardrails)

A rule-based and API-driven safety layer:

• RxNorm Interaction: Detects dangerous combinations (e.g., Anticoagulants + NSAIDs).
• Contraindications: Validates meds against patient allergies and biometric thresholds (BP/Sugar).

💻 Streamlit Interface

The application provides a premium, user-friendly interface for clinicians:

• Interactive Forms: Captures comprehensive biometrics, including acute clinical markers (Temp, WBC, SpO2).
• 🗺️ Clinical Road Map: A high-level, paragraph-style narrative that synthesizes the entire strategy.
• 📊 Real-time Risk Panels: Visual breakdown of disease risks and prioritized clinical flags.
• ⚠️ Urgent Alerts: Tiered escalation alerts (LOW to CRITICAL) with clear recommended actions.

⚙️ Setup & Installation

Prerequisites

• Python 3.9+
• API Keys: OpenAI, Tavily, OpenFDA

Installation

1. Clone the repository:

   git clone <repository-url>
   cd "Healthcare AI Clinical Support System"

2. Install dependencies:

   pip install -r requirements.txt

3. Configure Secrets:
   Create .streamlit/secrets.toml or set environment variables:

   [secrets]
   OPENAI_API_KEY = "your_key"
   TAVILY_API_KEY = "your_key"
   OPENFDA_API_KEY = "your_key"

Running the Application

streamlit run app.py

🔬 Scalability & Standards

• Modular Design: Every functional block is a LangGraph node, allowing for easy integration of new markers or APIs (e.g., Epic/FHIR).
• Pydantic Validation: Uses strict schema validation throughout the workflow to ensure clinical data integrity.
• Medically Cautious: Designed as a support system; it generates professional summaries while strictly avoiding diagnosis or unauthorized prescriptions.

Developed as a State-of-the-art Clinical Decision Support Tool.