ChatGPT efficacy in simplifying cardiac magnetic resonance reports

Abstract

Abstract Introduction Cardiovascular Magnetic Resonance (CMR) imaging reports, though rich in detail, are often dense and highly technical, making it challenging for patients to fully grasp their condition and treatment options. Artificial intelligence (AI) has emerged as a promising tool on simplifying such reports with data arriving from other medical fields demonstrating an accurate and complete translation of medical reports. However, not much is known about AI performance in patients suffering from premature ventricular beats (PVBs) or cardiomyopathies. Aim This study examines the efficacy of ChatGPT, a large language model (LLM), in simplifying CMR medical reports for patients in order to enhance patient comprehension empowering them in their healthcare journey. Methods We designed a prospective multi-center cohort study, enrolling 50 consecutive patients from July 1st to October 31st 2024, undergoing CMR during diagnostic work-up following PVBs or non-sustained ventricular arrhythmias. For each exam, we asked ChatGPT (GPT-4o) to simplify the original CMR report. After reading the two reports, patients had to answer two respective questionnaires regarding context comprehension, condition acknowledgment, overall satisfaction and reassurance from the report. Responses were graded from 1 (strongly disagree) to 5 points (strongly agree), while overall satisfaction was graded from a scale 1-10. ChatGPT reports were checked for possible mistakes. Results Patients’ average age was 62.5 ± 15.6 years, including 14 (28%) women. The majority of patients were high school graduates. Participants were much more likely to understand CMR findings in the simplified report (3.41±1.02 vs 4.38±0.53, p <0.01), as well as the respective described condition (3.41±1.00 vs 4.47±0.54, p <0.01) with highly statistically significant difference. Patients were also able to explain report context in they own words more easily after reading ChatGPT report (3.18±0.97 vs 4.10±0.77, p<0.01). Moreover, they felt more reassured (3.46±1.01 vs 4.24±0.72, p<0.01) and overall, more satisfied from the simplified report (6.64±1.96 vs 8.87±1.15, p<0.01). Discussion In this prospective multi-center study, we demonstrated a highly augmented patient understanding, reassurance and satisfaction of CMR reports when simplified by ChatGPT. AI appears to be a powerful tool allowing improved medical communication, establishing patient-centered care.

Ähnliche Arbeiten