Doctors preferred AI hospital discharge summaries over human notes in 95 percent of cases

Software wrote better discharge summaries for long hospital stays than doctors did, according to a peer-reviewed study published in npj Digital Medicine.

Researchers at Stony Brook University compared discharge documents created by a large language model against those written by physicians. The team examined 60 internal medicine stays that lasted between 7 and 21 days.

Both hospital doctors and primary care physicians reviewed the paperwork side by side. The clinicians chose the AI-written summary in 95 percent of the cases.

The reviewers gave the machine versions higher marks for readability, overall quality, factuality, and completeness. Primary care doctors noted that the automated drafts made it easier to understand inpatient treatment and plan follow-up visits.

Medical notes written by humans missed more critical details than the model did. The automated system recorded fewer clinically relevant omissions, which drove its higher accuracy score.

The software also caught secondary medical issues flagged on scans. Out of 31 unexpected findings spotted in radiology reports, clinicians rated 93.5 percent of them as accurate and 87.1 percent as necessary to document.

Safety ratings between the two approaches remained balanced. Evaluators found no meaningful difference between human and machine summaries regarding potential patient harm.

Could routine hospital paperwork soon be handed off to software before a doctor signs off?

The study has clear boundaries. It looked back at records from a single hospital system across 60 patient cases, and human doctors still had to verify every finding before clinical use.