Summary
Provides consensus-defined methods and metrics for evaluating generative AI used to create patient discharge summaries, including documentation quality, factual coverage and consistency, usefulness, safety, transparency, and performance before and after deployment.
Healthcare Implications
Developers and health systems should test generated summaries against expert-reviewed records, evaluate missing or fabricated information, obtain clinician and user feedback, require human review before inclusion in the medical record, and monitor quality and safety across settings and patient groups.