From Questions to Evidence

The Real-World Data Attack Rate Explorer Dashboard and AskRWD Lite (AI chat agent) are complementary, self-service tools for the Infectious Disease Therapy Area.

They combine transparent visual analytics with flexible plain-language querying. Together, they help clinicians quickly answer questions for drug development using real-world data.

Background

The infectious disease (ID) therapy area is advancing rapidly with new vaccines and monoclonal antibodies in development. Real-world data (RWD) is increasingly used to inform clinical trials and is drawing growing interest from clinicians. However, accessing RWD remains a bottleneck due to technical barriers such as complex data structures and gaps in training on study design and coding. This reduces efficiency in the fast-paced environment.

Solution Overview

To help clinicians explore real-world data easily, AstraZeneca’s Data Science & AI teams created two complementary, self-service tools: the ID RWD Attack Rate Explorer Dashboard and AskRWD Lite (AI Chat Agent). Both use the same data but support different needs. The dashboard provides visualization with step-by-step transparency into how results are built. AskRWD Lite offers more flexibility, enabling users to ask questions in everyday language beyond predefined options in the dashboard, without requiring coding. Together, these tools lower barriers to RWD exploration and help clinicians translate questions into insights.

Data source

Both tools use the same datasets to ensure consistency. The raw data is Optum’s de-identified Clinformatics® Data Mart Database (Optum® CDM), a large US administrative claims database with patient-level medical and pharmacy claims. It covers over 78 million individuals. For the tools, three study populations were created from the raw data: immunocompromised patients, patients with bronchiectasis, and patients with Clostridioides difficile infection (CDI).

Methods

The dashboard is built using Python Dash on the Scientific Computing Platform (SCP) Posit Connect service. When users create cohorts in the dashboard, Python scripts update results in real time. AskRWD Lite is built in Streamlit on Domino platform using LangGraph agent powered by Claude 4.5 Sonnet. When users ask a question, the agent generates SQL queries, optionally invokes a distribution-plotting tool, and returns a plainlanguage answer based on the SQL results.