The European Medicines Agency opened an AI-assisted review pilot on Sunday for oncology drug applications, deploying validated tools that scan clinical study reports for statistical inconsistencies, endpoint switching, and safety signal patterns. Human assessors retain exclusive authority over approval recommendations.
Agency officials said the pilot targets bottlenecks in rolling submissions where sponsors file thousands of pages across multiple modules. Models trained on anonymized historical dossiers highlight sections requiring deeper expert review, not autonomous accept-or-reject rulings.
Validation Standards
EMA published technical documentation requirements aligned with the EU AI Act's high-risk classification for medical devices. Vendors must submit bias audits demonstrating consistent performance across demographic subgroups represented in European trial populations.
Industry Response
Roche and AstraZeneca volunteered two filings for the pilot. Smaller biotech firms asked whether AI tooling favors sponsors with standardized data formats, potentially raising barriers for novel trial designs.
Patient Advocacy
Health access groups demanded public summaries of when AI tools influence review timelines. EMA committed to publishing aggregate metrics after 12 months without disclosing sponsor identities.
Transatlantic Coordination
The U.S. Food and Drug Administration runs parallel machine-learning pilots on manufacturing inspection data. Regulators scheduled a joint workshop in September on harmonizing transparency standards for AI in drug review.
