AI-Generated Discharge Summaries Score Higher on Quality, Readability and Accuracy, Study Finds

Published in npj Digital Medicine, the study examined whether an LLM designed to process and generate text could help clinicians prepare documents more effectively.

Artificial intelligence (AI) has shown potential to improve the quality of patient discharge summaries, with a Stony Brook University study finding that physicians preferred AI-generated summaries over those written by clinicians in 95% of the hospitalisations evaluated.

The findings highlight how large language models (LLMs) could support discharge documentation for patients with complex medical histories.

Published in npj Digital Medicine, the study examined whether an LLM designed to process and generate text could help clinicians prepare these documents more effectively.

Researchers at Stony Brook University developed an approach guided by clinicians' requirements for post-discharge information. The system processed patient records and organised relevant clinical details into a draft summary within minutes.

The research team compared AI-generated and physician-authored discharge summaries from 60 internal medicine hospitalisations at Stony Brook University Hospital.

The hospital stays lasted between seven and 21 days. Hospitalists and primary care physicians assessed the summaries in paired evaluations.

Physicians preferred the AI-generated summaries in 95% of the evaluated hospitalisations. They also assigned higher ratings to the AI-generated documents for overall quality, readability, accuracy, and completeness.

Reviewers identified fewer omissions of important clinical details than in the summaries prepared by physicians.

Primary care physicians additionally reported that the AI-generated summaries were easier to understand, explain to patients and caregivers, and use when planning follow-up care.

These findings suggest that AI could help organise complex medical information into documents that are more useful for clinicians managing patients after discharge.

"These results provide evidence that artificial intelligence can support discharge documentation for complex hospital stays," said Rachel Wong, co-corresponding author and clinical associate professor at Stony Brook University.

However, the study also identified limitations. Both AI-generated and human-authored summaries contained errors, and reviewers found no statistically significant difference between the two groups in the estimated severity or likelihood of harm associated with those errors.

The findings therefore do not establish that AI-generated summaries are inherently safer than physician-written documents.

The researchers have emphasised that clinicians must review and verify AI-generated drafts before they are added to patient records. Human oversight remains important to ensure that treatment details, medication changes and follow-up instructions are accurately represented.

Fusheng Wang, co-corresponding author and professor in the Department of Biomedical Informatics and Computer Science at Stony Brook University, said further research would focus on accuracy, safety, actual time savings for physicians, and integration into clinical workflows.

A pilot study evaluating the tool in routine clinical practice at Stony Brook Medicine is also being considered.

The study adds to research exploring how AI can support clinical documentation. Further evaluation in everyday hospital settings will be needed to determine whether these tools can consistently improve patient discharge summaries while maintaining safety and clinical accountability.

Stay tuned for more such updates on Digital Health News