Clinical AI investigation · November 17 study

The discharge summary a doctor would sign

A family physician opens a discharge summary. One sentence asks primary care to repeat a blood test on Monday. Was that the hospital team's plan?

  • Kanwar Partap Singh Gill, MD
  • Family medicine physician & independent policy analyst

Published

Sources checked through , 5 p.m. Pacific · facts checked and fully reviewed · 2 min read

A family physician opens a discharge summary. One sentence asks primary care to repeat a blood test on Monday. Was that the hospital team's plan? Has anyone placed the order? Has anyone told the patient where to go? In this hypothetical handoff, the sentence can read beautifully while leaving three people with different ideas about who owns the next step.

A study published November 17 in JAMA Network Open helps explain both the appeal and the limits of automated summaries. Twelve physicians reviewed 1,294 unique summaries against hospital records from five London hospitals in one NHS trust. Some summaries came from routine clinical care. Others were generated retrospectively from admission records using GPT-4o mini and a prompt clinicians helped design. In the adjusted analysis, reviewers were 22.2 percentage points more likely to say they would sign the machine-generated version with minor or no edits (95% confidence interval, 19.0 to 27.5 points). Reviewers also identified consequential omissions and errors. Read the study and its supplement.

That result deserves a place in the debate about documentation burden. A physician faced with several days of notes, tests and consultations may benefit from software that can assemble a coherent history. The experiment gave reviewers access to the underlying record. It measured their willingness to sign. It did not follow patients home to find out who received the promised phone call or whether a missed result delayed treatment. It also did not test ambient recording of outpatient conversations, Microsoft DAX or Epic's handling of proposed diagnoses. Each is a different clinical task.

A practical evaluation could start with a small sample of discharges. Ask a reviewer to trace every follow-up request to the underlying order, consultation or decision in the record. Ask the receiving team who actually received the request, when it arrived and who accepted responsibility. Where a patient declined a proposed intervention, check whether the summary preserved that decision. A reviewer who can see only the signed prose should be able to record “source unavailable” alongside a disputed statement. That is valuable information about the organization's ability to audit its own system.

The patient can ask a simpler question before leaving the hospital: “If nobody calls me about this result, whom should I contact, and when?” A useful discharge summary answers that question even when the patient has no access to the software that produced it.

Related coverage. Who Owns the Decision? · DAX, Dragon Copilot and physician review · Organizing medical records.

Where this continues on the site

Editorial and informational notice: this analysis of health care, technology, consumer and legal policy is published for education and public awareness. It is not individualized medical, legal or financial advice and does not replace your own clinician, lawyer, bank or benefits office. Court orders, settlements and filings are described for what they decide; allegations are not findings. Kanwar Partap Singh Gill, MD is a physician, not a lawyer, and this site is not a law firm. In an emergency, call 911.

Prepared 19 November 2026 with AI assistance from the sources linked on this page. Facts checked and the page fully reviewed. Evidence cutoff 19 November 2026, 5 p.m. Pacific. Corrections: editorial standards.

Search KPSGill.com

Search the full publication — patient guides, policy, investigations, enforcement records and research.