Simplifying Radiology Reports for Patients Using ChatGPT: A Systematic Review.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Radiology Communication

Background:

• Radiology reports are often complex for patients.
• Improving patient comprehension of medical information is crucial for informed decision-making.
• Large language models offer potential solutions for simplifying medical text.

Purpose of the Study:

• To systematically review the benefits and risks of using ChatGPT to simplify radiology reports.
• To assess the impact of AI-driven simplification on report accuracy and completeness.
• To evaluate the potential for enhancing patient understanding of radiological findings.

Main Methods:

• Systematic review of studies published until December 17, 2024.
• Searched PubMed/MEDLINE, Embase, Web of Science, and CENTRAL databases.
• Included 16 studies evaluating ChatGPT (GPT-4, GPT-4o, GPT-3.5) for report simplification.

Main Results:

• Simplified reports reduced reading grade levels from 9.6-14 to 5-9.4.
• Factual accuracy ranged from 78%-100%, completeness from 83%-100%.
• Potentially harmful errors decreased from 0%-36% to 0%-8% after outlier exclusion; only 19% assessed layperson comprehensibility.

Conclusions:

• ChatGPT shows potential for patient-friendly radiology reports, but requires radiologist oversight and adherence to privacy laws.
• Further prospective studies are essential to evaluate patient comprehension, satisfaction, and workflow impact before clinical implementation.