Quality, Reliability, and Safety of a Commercial Evidence-Grading Artificial Intelligence System for Cardiovascular Decision Support: An Independent Proof-of-Concept Evaluation Using Simulated Cases

This dataset accompanies a vignette-based evaluation of AI-generated responses to 40 simulated cardiovascular cases covering coronary artery disease, heart failure, and arrhythmias.The Excel file contains case identifiers, age, sex, disease category, clinical severity category, age group, and individual scores from three cardiologist raters across eight evaluation domains: medical accuracy, factual accuracy, completeness, clarity, urgency, language quality, absence of false claims, and absence of harm. It also includes individual Global Quality Scores (GQS), calculated mean scores, and binary harmfulness assessments.The accompanying vignette document provides the clinical scenarios used in the evaluation. Case identifiers correspond across the files, enabling linkage between each vignette and its associated ratings.All cases are simulated and contain no identifiable patient information. Column abbreviations ending in R1, R2, and R3 denote the three raters; columns ending in Mean contain the corresponding average scores.