Comment from IQVIA
AnonymousSupportOther
Summary: The commenter provides specific technical suggestions to improve the draft guidance on AI in regulatory decision-making. They request more examples of "operational efficiency" exclusions and seek clarification on data traceability, validation expectations for federated learning, and validation approaches for unsupervised models.
Page 3, Scope: Clarify "Operational Efficiency" exclusions by providing more examples, such as AI in regulatory writing, AI in DM query management, and AI in clinical trial monitoring.
Page 12, Line 367: Improvement Suggestion: Clarify how federated learning impacts FDA evaluation for AI credibility, particularly regarding:
Data traceability: What does FDA expect from sponsors to demonstrate data quality when it is not centrally stored (using federated learning)?
Validation expectations: Should models be evaluated per site or only on the final aggregated model or both?
Page 13, Line 377: The document focuses on supervised learning performance metrics (e.g., ROC curves, F1-score) but does not clarify how unsupervised models (e.g., clustering, anomaly detection) should be validated for regulatory decision-making. Improvement Suggestion: Provide guidelines on how unsupervised models should demonstrate credibility, including acceptable validation approaches.