Drug Evaluation Committee Overview of the "Risk-Based Credibility Assessment of AI Models" Framework proposed by the FDA
Electronic Standard for Medical Information Expert Committee
This document provides practical guidance on AI quality assurance for practitioners and users who implement and operate AI models in the GxP domain in pharmaceutical companies, encouraging them to practice critical thinking that keeps asking "Why use this AI?" and "What are the risks if it is wrong?
FDA's " Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry The proposed comprehensive framework is based on the "7-step risk-based credibility assessment framework" proposed in the FDA's "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drugs and Biological Products Guidance for Industry and Other Interested Parties" draft guidance (January 2025) and extends to 12 steps covering the PoC to operational phases. This book features a comprehensive framework that covers all phases from the PoC to the operational phase.
The distinctive feature of this book is that it systematizes for CSV practitioners the exploratory and iterative development process of AI models, which is essentially different from conventional CSV. This systematization is structured in such a way that the background of FDA's requirements for risk assessment, ensuring data independence, etc. can be logically understood.
We hope that this document will help to promote the use of credible AI by all parties.
Japan Pharmaceutical Manufacturers Association Drug Evaluation Committee
Electronic Standard for Medical Information Expert Committee Task Force 4