AI in Healthcare Achieves 95% Preference Rate for Hospital Discharge Summaries Over Human-Authored Versions

AI in Healthcare Transforms Patient Discharge Documentation

Artificial intelligence is reshaping how physicians prepare hospital discharge summaries, with large language models demonstrating superior performance in accuracy, completeness, and readability. A Stony Brook University study published in npj Digital Medicine examined 60 Internal Medicine hospitalizations lasting 7 to 21 days, comparing AI-generated summaries against human-authored versions through paired assessments by hospitalists and primary care physicians1

1

2

Patient discharge summaries serve as critical roadmaps for follow-up care, containing essential information about treatment protocols, medication needs, changes in medication, unresolved medical issues, and outpatient requirements. Primary care physicians depend on these documents to maintain continuity of care after patients leave the hospital. For complex hospital stays with larger volumes of medical information, preparing comprehensive discharge documentation creates a significant burden for providers, consuming valuable time that could be spent on direct patient care1

1

Large Language Models Achieve 95% Preference Rate

The Stony Brook University study revealed that physician reviewers preferred AI-generated summaries in 95% of evaluated hospitalizations. Reviewers consistently gave AI-generated summaries higher scores across multiple quality metrics, including overall quality, readability, accuracy, and completeness. The AI-generated summaries also contained fewer missing clinical details compared to human-authored versions2

2

Primary care physicians found the AI-generated summaries easier to understand, simpler to explain to patients and caregivers, and more useful for follow-up care. "These results provide evidence that artificial intelligence can support discharge documentation for complex hospital stays," states Rachel Wong, MD, co-corresponding author and Clinical Associate Professor in the Departments of Internal Medicine and Biomedical Informatics at Stony Brook University's Renaissance School of Medicine1

1

Source: News-Medical

Clinical Workflows and Time-Saving Benefits

The research team developed their approach using a large language model designed to process and generate text based on input from clinicians. The LLM organized patient medical records and other information to create comprehensive patient summaries within just a few minutes. Clinicians helped design the approach based on the specific information physicians need after patient discharge, ensuring the AI-generated summaries addressed real-world clinical requirements2

2

This AI in healthcare application focused specifically on longer, complex hospital stays where the volume of medical information creates the greatest documentation burden. The time-saving benefits could allow physicians to redirect their attention to direct patient care rather than administrative tasks, though the study did not quantify exact time savings in this initial phase1

1

Related Stories

AI Poised to Revolutionize Hospital Quality Reporting, UC San Diego Study Finds

AI Outperforms Nurses in Predicting Hospital Admissions, Promising to Reduce ER Overcrowding

AI Diagnosis Outperforms Emergency Room Doctors in Harvard Study, Raises Questions on Patient Care

AI Model Quadruples Delirium Detection in Hospitals, Improving Patient Outcomes

Physician Review of AI Summaries Remains Essential

While AI improves accuracy and completeness, both AI-generated and human-authored discharge summaries contained errors in the study. Physician reviewers found no statistically significant difference in the estimated severity or likelihood of harm associated with errors in either type of summary. The authors emphasize that physician review and verification remain essential before any LLM-generated discharge draft enters a patient record2

2

Fusheng Wang, PhD, co-corresponding author and Professor in the Department of Biomedical Informatics and Computer Science at Stony Brook University, outlines the next steps: continue developing the LLM for complex patient discharge summaries with specific emphasis on assessing accuracy, safety, actual time savings for physicians, and integration into clinical workflows. A pilot study to evaluate the tool in everyday clinical practice at Stony Brook Medicine is under consideration1

1

Interdisciplinary Team Drives Innovation

The project involved an interdisciplinary team of researchers from Stony Brook University's Renaissance School of Medicine, including computer scientists, biomedical informaticians, physicians, and a communications expert. Lead author Tyler Osborne, a PhD candidate in the Department of Computer Science at Stony Brook, gathered large amounts of electronic health records data to complete the study. The Research Services & Innovation Team from Stony Brook Medicine Information Technology provided crucial support in this effort2

2