Medical Student Harvard Medical School San Jose, California, United States
Disclosure(s):
Advait Patil, MS: No financial relationships to disclose
Introduction: The 21st Century Cures Act mandated patient access to their electronic health record, demonstrating a national commitment to patient empowerment. Despite this, clinical texts such as spine radiology reports contain complex medical terminology that can be challenging for patients to understand, impeding comprehension and informed consent. Recent advancements in large language models (LLMs) have shown promise in simplifying technical texts for non-expert audiences, which may make it easier for patients to engage with their healthcare information.
Methods: We used an LLM (GPT-4o) to simplify 149 spine radiology reports obtained from patients being evaluated for spine surgery. Each report underwent automated text simplification by the model, with modifications aimed at retaining essential clinical information while translating technical jargon into patient-centered language at an average American reading level. After simplification, we used natural language processing to compute pre- and post- simplification readability metrics.
Results: Readability improved from a high school upperclassman to a 7th grade level, within the average American’s reading level (8th grade). After LLM simplification of radiology reports, we find a significant decrease in mean Flesch-Kincaid Grade Reading Level (11.7 before vs. 7.4 after, p< 0.001), and improvement in reading ease (33.2 before vs. 69.0 after). The mean character count decreased from 1774 to 1539 (p < 0.001). Manual examination of a random sample of 20 LLM simplified reports revealed no loss in clinical accuracy or hallucination of information.
Conclusion : The successful use of LLMs for text simplification in spine radiology reports demonstrates a promising method for enhancing patient comprehension before complex surgical decision-making. This approach can be applied across electronic health record data at large towards increasing patient understanding, engagement, and truly informed consent by simplifying complex medical information.