Research
Study finds clinicians accepted AI-generated discharge summaries more often
Clinician reviewers signed off on 92.2% of AI-generated summaries with no or minor edits, compared with 70.0% of clinician-written summaries.

A retrospective study in five London hospitals found that clinician reviewers signed off on a higher proportion of discharge summaries drafted by an AI model with no or minor edits than on summaries written by clinicians. The researchers reported that 92.2% of AI-generated summaries met that standard, compared with 70.0% of clinician-authored summaries. The result met the study’s prespecified noninferiority criteria.
The study included 1,294 summaries, 645 generated by GPT-4o mini and 649 written by clinicians. Reviewers rated the AI-generated summaries higher for accuracy, completeness, tone and clarity. The researchers also reported inaccurate or incomplete information in 26.2% of AI summaries and 45.1% of clinician-written summaries. The authors noted that only one AI model was tested and said prospective assessments at the point of care are still needed.


