Patient Perceptions on Operative Reports Written by Surgeons versus Versions Simplified by Artificial Intelligence

The inherent complexity of medical documents remains a barrier to patient literacy and engagement. Aligning these documents with specific patient literacy levels and reading preferences without compromising semantic meaning can improve patient comprehension and participation in their care. This study evaluates the effectiveness of the GPT-4 large language model in simplifying operative reports, assess patient preferences for original versus artificial intelligence (AI) simplified reports, and elucidate attitudes on AI in medical documentation.

A two-phase prospective study was conducted from June to October 2023. We surveyed adults 18 years or older who had undergone spine surgery. A total of 3932 patients were identified, with 360 completing the initial survey, and 251 completing the subsequent evaluation. Phase 1 involved surveying patient preferences and attitudes toward operative reports and AI. Phase 2 involved comparing patients' perceptions of their original operative reports to those simplified by AI. Reports were also reviewed by 2 independent physicians for quality and safety. The primary outcome measured was patient preference for either the original or the AI-simplified operative report. Secondary measures included readability, comprehensibility, overall satisfaction with the document, linguistic metrics, and physician review.

The primary outcome showed that 55.4% (95% CI: 48.6%-61.2%) of participants preferred the GPT-simplified reports for being significantly clearer (mean = 1.53 ± 0.83 vs. 1.98 ± 1.92, P < 0.001), terminology being more accessible (mean = 1.50 ± 0.70 vs. 2.31 ± 0.89, P < 0.001), and overall, more understandable (mean = 86.40 ± 18.07 vs. 73.70 ± 20.82, P < 0.001). Those with a bachelor's degree and below preferred the GPT-simplified report in higher proportion (59.6%, 95% CI: 52.0%-67.0%) of cases.

This study suggests that operative reports simplified by AI can facilitate patient understanding. However, preferences varied according to patient demographics, indicating the need for individualized approaches to simplification. Further research is necessary to optimize the use of AI for personalized patient education and to evaluate the long-term impact on healthcare outcomes.