Impact of GPT-4-Generated Discharge Letters on Patients' Medical Comprehension: Prospective Crossover Study.

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Patient Education

Background:

• Standard hospital discharge letters are difficult for patients to understand, increasing risks of medication errors and comprehension issues.
• Cognitive Load Theory (CLT) suggests complex texts overload working memory; AI-generated patient-centered versions may reduce this cognitive load.
• Evidence for the effectiveness of AI in improving patient understanding of discharge information is limited.

Purpose of the Study:

• To evaluate if GPT-4 generated patient-centered letters improve standardized patients' retention and understanding of safety-relevant medical information compared to standard discharge letters.
• To explore the potential effects of AI-generated letters on cognitive load, as described by Cognitive Load Theory (CLT).

Main Methods:

• A prospective, randomized, crossover study involving 48 trained standardized patients.
• Participants received both a conventional discharge letter and a matching GPT-4 generated patient-centered letter for assigned diseases.
• Comprehension was assessed by the proportion of predefined safety-relevant learning objectives reported, stratified by content field and Bloom taxonomy level.

Main Results:

• GPT-4 generated patient letters significantly improved overall comprehension compared to standard discharge letters (OR 1.74, P<.001).
• Patient letters led to higher rates of fully or partially stated learning objectives and fewer omissions.
• Improvements were most pronounced for 'Medication' and 'Organization' content, and for 'Remember' level objectives, but less so for 'Prevention of Complications' and 'Lifestyle/Disease Management'.

Conclusions:

• GPT-4 generated patient-centered letters enhance comprehension of safety-relevant discharge information, especially medication and organizational details.
• These AI tools show promise in reducing cognitive load and improving patient understanding within a CLT framework.
• Further research into multimodal, iterative AI supports is warranted to improve higher-order understanding and address remaining information gaps.