AI Discharge Summaries Used in 57% of Cases in Hospital Pilot Study

Physicians at an academic health system integrated AI-generated discharge summaries into more than half of their patient records during a clinical pilot, according to research published in JAMA Network Open.

The 10-week study evaluated an agent-based workflow using a large language model to draft hospital course summaries from existing clinical notes. The system generated 1,274 summaries across 384 patient discharges on an inpatient medicine unit.

Adoption rates and safety profile

Clinicians chose to use AI-generated content in 57% of discharge summaries. Among cases where physicians provided feedback, most drafts were rated as unlikely to cause harm. Only one summary was flagged as potentially causing moderate harm—a case where the AI failed to mention a completed antibiotic course and the prophylactic indication for continued treatment. After adjudication, this was determined to pose no actual risk.

The most frequent issues involved missing information or incomplete clinical details rather than fabricated content. Physicians identified and corrected these problems during their standard review process before finalizing documentation.

The AI system operated automatically, pulling from admission notes and daily progress reports to create drafts that clinicians could optionally incorporate into their workflow. This design required no changes to existing documentation responsibilities.

Impact on clinician burnout

Burnout scores among participating physicians decreased from 1.75 to 1.20 on the Stanford Professional Fulfillment Index Work Exhaustion Scale, falling below the threshold for work-related exhaustion. The change was statistically significant (P=0.03).

Measured time savings were less clear-cut. While 71.4% of physicians recorded shorter completion times for discharge summaries, the overall reduction of up to 2.9 minutes was not statistically significant (P=0.13). However, clinicians reported perceived time savings in 67% of cases, with 32% estimating they saved more than 15 minutes per summary.

The researchers noted that the primary benefit appears to be cognitive offloading rather than clock-time efficiency, shifting the value proposition from pure speed to workforce sustainability.

Why it matters

Discharge summaries are tied to care transitions, quality metrics, and reimbursement workflows in healthcare organizations. This study provides prospective data on AI documentation tools in active clinical settings, unlike previous retrospective evaluations. The findings suggest AI assistance may reduce clinician burnout even when time savings are modest, though the single-unit design limits generalizability. Healthcare systems considering similar tools will need to weigh adoption rates against the resources required for physician review and error correction.

The study was conducted with a limited group of physicians at one academic medical center. Researchers emphasized that further evaluation is needed to assess long-term performance, scalability, and integration with broader clinical systems.

The findings were first reported by DocWire News, based on research published in JAMA Network Open by Grolleau and colleagues.