Discussion

AI discharge summaries won higher marks in a five-hospital study

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#14188

In a retrospective study across five London hospitals, discharge summaries drafted with GPT-4o mini were more often approved with no or minor edits than summaries written by clinicians. The results offer a promising case for AI-assisted paperwork, but do not yet show that it makes care safer.

Watch Desk analysis

What happened

Researchers compared 645 AI-generated discharge summaries with 649 clinician-written summaries, using reviews by 12 resident physicians. Reviewers approved 92.2% of the AI summaries with no or minor changes, against 70.0% of the clinician-written ones. The study met its pre-specified non-inferiority criteria.

Reviewers also rated the AI summaries higher for accuracy, completeness, tone and clarity. The study reported inaccurate or incomplete information in 26.2% of AI summaries, compared with 45.1% of those written by clinicians. MedPage Today reports that the research used records from five hospitals and calls for prospective evaluation.

Why it matters

Discharge summaries help carry important information from hospital to the next stage of a patient’s care. If AI can draft them well and save clinicians time, that could be useful. But a reviewer’s assessment of a document is not the same as evidence that patients had safer transitions or better outcomes.

The study was retrospective, tested one model, and used de-identified records that may have lost clinically relevant details. It also leaves a stubborn practical question: can clinicians verify a summary quickly when the underlying record is long and fragmented?

Our read

This is stronger evidence than a polished AI demo: the summaries were compared with routine clinician-written work and assessed by clinicians. Still, “preferred draft” is not “proven safe care”. The sensible next step is prospective testing in real workflows, with careful checks on errors and the time needed to catch them. No one needs another paperwork shortcut that quietly hands the hard part back to the doctor.

What to watch

  • Whether prospective studies reproduce the results in live clinical settings.
  • Whether later evaluations test more models and different patient populations.
  • Whether time saved on drafting survives the work of checking each summary.

Discussion spark: Should hospitals use AI to draft discharge summaries now, with clinician review, or wait for prospective evidence that the approach improves patient care?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.