ML Framework for Aggregating Individual-Level and averaged clinical data.

Area of Science:

• Pharmacometrics
• Artificial Intelligence in Drug Development
• Biostatistics

Background:

• Pharmacokinetics-Pharmacodynamics (PK/PD) data analysis is crucial for drug development, but individual patient data (IPD) is often unavailable or aggregated.
• Existing meta-analysis (MA) methods struggle with limited data, hindering detailed PK/PD parameter characterization.

Purpose of the Study:

• To develop and validate a generative AI approach for reconstructing IPD from population-level statistics.
• To enhance PK/PD modeling by overcoming the limitations of aggregated data in drug development.

Main Methods:

• A generative AI model was trained on a small set of existing IPD.
• The trained model regenerated IPD for other cohorts using only their population statistics.
• Simulated clinical trial data across multiple studies were used to test the methodology.

Main Results:

• The generative AI algorithm successfully learned and reapplied original data relationships.
• Regenerated IPD captured information lost in aggregated population statistics.
• Performance tests demonstrated good agreement between simulated ground truth and AI-generated data.

Conclusions:

• Generative AI offers a powerful solution for reconstructing valuable IPD from aggregated data.
• This method can significantly improve PK/PD analysis and drug development by leveraging sparse data.
• The approach holds promise for more robust efficacy and safety assessments in pharmaceutical research.