SAFECARD – Safety Assessment Framework for Evaluating Clinical AI in Realistic Decision-Making: Multicenter Expert-Validated Evaluation of the OCC-VIM Clinical AI Safety Framework

he SAFECARD (Safety Assessment Framework for Evaluating Clinical AI in Realistic Decision-Making) study protocol defines a structured evaluation methodology for AI-based clinical decision support systems (CDSS). Applying the OCC+VIM analytical model (Omission, Commission, Confession; Values in Model), it systematically identifies, classifies, and quantifies safety-relevant errors in AI-generated clinical recommendations. Version 3.6 incorporates the Bias & Equity dimension with perturbation methodology, conformity requirements under the EU AI Act, and references validated benchmarks including NOHARM and MedAgentBench. The protocol is designed for prospective multicenter application and serves as the methodological foundation for the SKAT observational study.

Scientific Coordinator: Olaf Göing. Principal Investigator: Prof. Dr. med. Johannes Brachmann, Sana Medical School Coburg.