This study, which analysed 50 discharge summaries, indicates that large language models (LLMs) can potentially render these summaries into formats that are much more accessible and understandable for patients than their original form in electronic health records. Despite this promise, hurdles need to be addressed, such as enhancing the accuracy, completeness, and safety of the translated summaries. Concerns over safety underscore the need for physician oversight in the early stages of implementation.
The core inquiry of this research was whether LLMs could effectively reformat discharge summaries to make them easier for patients to read and understand. The study’s findings revealed a notable improvement in the readability and comprehensibility of patient-friendly discharge summaries. The summaries were deemed completely comprehensive in 56 out of 100 evaluations, yet 18 reviews raised alarms about potential safety risks due to omissions and inaccuracies.
These outcomes are significant because they demonstrate that LLMs can indeed serve as tools for converting discharge summaries into patient-friendly language and structures. However, this potential comes with the caveat that before such a model can be implemented widely, it is imperative to address the concerns related to accuracy, completeness, and safety. The initial phase of introducing this innovative approach will necessitate a review by medical professionals to mitigate safety risks.
In essence, while the study underscores LLMs’ capacity to make discharge summaries more accessible to patients, it also highlights the importance of refining these tools to ensure they meet the highest standards of medical communication. The dual focus on innovation and safety reflects a cautious yet optimistic approach to leveraging AI in healthcare to enhance patient understanding and engagement in their care processes.
More information: Jonah Zaretsky et al, Generative Artificial Intelligence to Transform Inpatient Discharge Summaries to Patient-Friendly Language and Format, JAMA Network Open. DOI: 10.1001/jamanetworkopen.2024.0357
Journal information: JAMA Network Open Provided by JAMA Network
