Skip to content

leelesemann-sys/market-access-intelligence

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Market Access Intelligence

Language: English | Deutsch

Cross-country Health Technology Assessment (HTA) analysis powered by AI.

Compare drug approval decisions between G-BA (Germany) and NICE (UK) through interactive analytics, side-by-side drug comparisons, and an AI-powered knowledge base chat.


The Problem

Market Access teams in pharma spend weeks manually comparing HTA decisions across countries. G-BA and NICE evaluate the same drugs but use fundamentally different frameworks (clinical benefit vs. cost-effectiveness), making cross-country analysis complex and time-consuming.

Studies show only 54-72% concordance between agencies for the same drugs - understanding why decisions diverge is critical for market access strategy.

The Solution

Market Access Intelligence automates the entire workflow:

  • 2,300+ assessments from G-BA (998) and NICE (1,312) in a unified database
  • Cross-country matching linking the same drug across agencies by INN
  • RAG-powered chat that answers questions using assessment data + PDF decision documents
  • PDF analysis of G-BA "Tragende Gruende" (decision rationale) for clinical detail beyond metadata

Features

Analytics Dashboard

  • KPI overview: assessment counts, matched drugs, concordance rate
  • Outcome distribution (positive/restricted/negative) by agency
  • Concordance matrix (G-BA vs NICE)
  • Assessment timeline by year

Drug Comparison

  • Side-by-side comparison for any matched drug
  • All assessments per agency with indication, rating, outcome, date
  • Direct links to source documents

RAG Chat (AI Knowledge Base)

  • Ask questions in German or English
  • Hybrid search: vector similarity + keyword matching + semantic reranking
  • SQL routing for aggregate queries ("all orphan drugs", "how many")
  • PDF-backed answers citing clinical endpoints and decision rationale
  • Source citations with links to original documents

Example queries:

  • "Compare G-BA and NICE decisions for pembrolizumab"
  • "Welche Orphan Drugs wurden 2024 negativ bewertet?"
  • "What mortality endpoints were evaluated for nivolumab in NSCLC?"

Architecture

                  Data Sources                    Processing              Frontend
              +-----------------+
              |   G-BA XML      |---+
              | (POST -> S3)    |   |      +-------------+
              +-----------------+   +----->|  Importer   |---+
              +-----------------+   |      | (UPSERT)    |   |
              |  NICE Excel     |---+      +-------------+   |    +------------+
              | (HTTP GET)      |                             +--->| Azure SQL  |
              +-----------------+                             |    | Database   |
              +-----------------+          +-------------+    |    +-----+------+
              |  G-BA PDFs      |--------->| PDF Parser  |----+          |
              | (Tragende       |          | + Chunker   |          +----+----+
              |  Gruende)       |          +-------------+          |         |
              +-----------------+                              +---+---+ +---+---+
                                                               |Embedder| |SQL    |
                                                               |256-dim | |Routing|
                                                               +---+---+ +---+---+
                                                                   |         |
                                                               +---+---+     |
                                                               | Azure  |    |
                                                               |   AI   |    |
                                                               | Search |    |
                                                               +---+---+    |
                                                                   |         |
                                                               +---+---------+--+
                                                               |    RAG Chat    |
                                                               |  (GPT-4o)     |
                                                               +-------+-------+
                                                                       |
                                                               +-------+-------+
                                                               |   Streamlit   |
                                                               |   Dashboard   |
                                                               +---------------+

Tech Stack

Layer Technology
Database Azure SQL (Serverless)
Search Azure AI Search (hybrid: vector + BM25 + semantic)
Embeddings Azure OpenAI text-embedding-3-small (256-dim)
Chat Azure OpenAI GPT-4o
PDF Parsing pdfplumber
Frontend Streamlit
Language Python 3.11+

Data Sources

Source Agency Records Content
G-BA XML G-BA (Germany) 998 assessments Benefit ratings, subgroups, endpoints, comparators
NICE Excel NICE (UK) 1,312 appraisals Recommendations, categorisations, comments
G-BA PDFs G-BA (Germany) 2,376 available "Tragende Gruende" decision rationale documents

Harmonization

G-BA rates clinical added benefit (Zusatznutzen). NICE evaluates cost-effectiveness (ICER threshold). These are fundamentally different measures.

We preserve original ratings and only harmonize to a ternary outcome for cross-country comparison:

positive restricted negative
G-BA erheblich, betraechtlich, gering, nicht quantifizierbar, gilt als belegt nicht quantifizierbar (insufficient evidence) ist nicht belegt, geringerer Nutzen
NICE Recommended (incl. CDF, IMF) Optimised, Only in research Not recommended

Setup

Prerequisites

  • Python 3.11+
  • Azure SQL Database
  • Azure AI Search service
  • Azure OpenAI deployment (text-embedding-3-small + GPT-4o)

Installation

git clone https://github.com/leelesemann-sys/market-access-intelligence.git
cd market-access-intelligence
pip install -r requirements.txt

Configuration

Copy the environment template and fill in your Azure credentials:

cp backend/.env.example backend/.env

Required variables:

HTA_SQL_SERVER=your-server.database.windows.net
HTA_SQL_DATABASE=your_db
HTA_SQL_USER=your_user
HTA_SQL_PASSWORD=your_password
HTA_SEARCH_ENDPOINT=https://your-search.search.windows.net
HTA_SEARCH_KEY=your_search_admin_key
HTA_OPENAI_ENDPOINT=https://your-region.api.cognitive.microsoft.com/
HTA_OPENAI_KEY=your_openai_key

Database Setup

Deploy the schema to your Azure SQL database:

# Apply schema.sql and schema_phase1d.sql to your database

Run the Pipeline

# Import G-BA + NICE data
python backend/run_pipeline.py --source all

# Build search index (embed + index)
python backend/run_pipeline.py --index

# Process G-BA PDF documents (Tragende Gruende)
python backend/run_pipeline.py --pdfs --pdf-limit 10   # test with 10
python backend/run_pipeline.py --pdfs --skip-scrape     # full run (use cached URLs)

Start the Dashboard

streamlit run backend/app.py

Project Structure

market-access-intelligence/
├── backend/
│   ├── sources/          # Data source adapters (G-BA XML, NICE Excel)
│   ├── pipeline/         # ETL: import, embed, index, PDF processing
│   ├── chat/             # RAG: hybrid retrieval + GPT-4o response
│   ├── app.py            # Streamlit dashboard (3 tabs)
│   ├── run_pipeline.py   # CLI orchestration
│   ├── config.py         # Central configuration
│   └── db.py             # Database connection
├── sql/                  # Database schema
├── docs/                 # Architecture documentation
└── requirements.txt

License

This project is licensed under the MIT License.

G-BA data is subject to G-BA usage terms. NICE data is publicly available under NICE terms.

About

Cross-country HTA analysis: G-BA (DE) & NICE (UK) — Dashboard + RAG Chat

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors