AI and Language models in RWE analytics

Adapting Language Models to Real-World Health Data

Traditional language models are not designed to process structured, high-dimensional, time-stamped health data. We bridge this gap by converting patient-level registry data into serialized, temporally structured text formats—so-called life sequences—that allow LMs to learn from data as narratives of care, treatments, diagnoses, and outcomes.

Focus Areas in Our AI Research

We actively explore and develop:

• Data transformation methods for converting registry data into formats suitable for generative models
• Explainability and uncertainty quantification, using tools such as SHAP and conformal prediction
• Model benchmarking of large and small LMs versus traditional ML and statistical methods
• Cross-country validation of AI models to support broader generalizability
• Sustainable and ethical AI, ensuring compliance with GDPR, the EU AI Act, and a commitment to smaller energy-efficient models

‍