AI-Generated Summaries Enhance Patient Understanding of Radiology Reports

Research conducted by the University of California, San Francisco (UCSF) highlights the potential of artificial intelligence (AI) in simplifying radiology reports for patients, potentially improving patient comprehension and reducing anxiety. This comes in response to the 21st Century Cures Act, which mandates timely patient access to health information, including radiology reports. While this transparency is beneficial, the complexity of medical terminology in these reports can often lead to patient confusion and unnecessary stress.

The study, led by Dr. Juan Serna and Dr. Jae Ho Sohn from UCSF, sought to address these challenges. "The risk is that patients won't understand what is normal and what is abnormal," remarked Dr. Serna. Dr. Sohn further noted the balance between the empowerment provided by information access and the challenges it presents, as many patients find the reports distressing without immediate physician interpretation.

To explore AI's potential in resolving these issues, the researchers used a secure version of ChatGPT to create simplified summaries of three deidentified lung cancer CT screening reports. These reports were chosen due to their potential to cause anxiety, featuring a range from negative findings to highly suspicious results. The AI-generated versions were reviewed for accuracy by a radiologist before being tested.

Using Prolific, an online survey platform, the study engaged 1,815 participants representative of the U.S. population. Each participant read the original and AI-generated reports and subsequently answered survey questions. Results showed that the AI-created summaries increased comprehension and reduced anxiety, positively affecting participants' willingness to wait for follow-ups rather than immediately contacting their doctors.

The study acknowledges limitations, such as not involving real patients reacting to their reports. However, Dr. Sohn emphasized the controlled study environment's importance for consistency. "We wanted to have a more controlled environment where everybody received the same reports," he explained, setting the stage for further research that might include real patients to refine AI summary use in clinical practice.

Drs. Sohn and Serna call for ongoing research and clinician involvement in the use of AI in patient communication. Dr. Serna highlighted the role academic institutions should play in guiding AI's safe implementation, emphasizing that radiologists need to lead evaluations to guide clinical use effectively.