Understanding Breast Radiology Reports: The Role of ChatGPT in Patient Communication

This heading encapsulates the essence of the study, highlighting both the focus on breast radiology and the involvement of ChatGPT in enhancing patient understanding.

10 AI as a Bridge: Can ChatGPT Help Patients Understand Their Breast Radiology Reports?

Background/Significance

In an era where large language models like ChatGPT are becoming integral in healthcare communication, understanding radiology reports poses a significant challenge for many patients. Research indicates that most patients operate at an 8th-grade reading comprehension level, complicating their ability to grasp complex medical jargon. This study investigates the potential of ChatGPT to facilitate better comprehension of breast imaging reports (specifically BI-RADS categories 3, 4, and 5), while ensuring that accuracy is maintained.

Materials and Methods

To explore this, we retrieved fifteen consecutive BI-RADS 3, 4, and 5 source reports from the radiology database. ChatGPT was prompted with direct requests: “please put the report into layperson terms” and “please provide recommendations for the patient.” This approach aimed to generate layperson-friendly reports and actionable recommendations for all 45 source reports analyzed.

The readability of source reports, ChatGPT-generated layperson reports, and accompanying recommendations was measured using the Flesch-Kincaid readability score—a tool that assesses health literacy by determining grade levels. Higher word counts typically indicate poorer readability. A radiologist evaluated the accuracy of ChatGPT outputs for any misleading information and checked for supplementary insights beneficial to patients, particularly focusing on biopsy recommendations.

Statistical analysis employed 2-sided unpaired, equal-variance t-tests to scrutinize the results for significance.

Results

Our findings revealed a statistically significant decrease in the mean grade level for BI-RADS 3 and BI-RADS 5 reports when evaluated against the layperson adaptations generated by ChatGPT. Specifically, BI-RADS 3 reports showed a mean grade level drop from 10.9 to 8.8 (P = .003) and BI-RADS 5 from 12.2 to 9.6 (P = .003). No significant differences appeared between source reports and ChatGPT-generated recommendations within all three BI-RADS categories.

Moreover, there was a notable reduction in word count for both BI-RADS 3 and BI-RADS 4 source reports when compared to ChatGPT-generated recommendations (P = .001 and P = .02, respectively). Importantly, the ChatGPT-generated reports were free from inaccuracies or misleading statements, and they provided valuable supplementary information, particularly regarding biopsy suggestions, which were absent in the original source reports.

Conclusion

Our research concluded that ChatGPT substantially improves readability for BI-RADS 3 and 5 reports while generating more concise recommendations for BI-RADS 3 and 4. The AI-driven reports maintained accuracy essential for patient management. These findings present a promising avenue for utilizing AI to produce patient-friendly reports, potentially alleviating patient anxiety surrounding breast imaging. By bridging the gap between complex medical language and patient understanding, ChatGPT could transform how patients engage with their health information, paving the way for improved patient outcomes and enhanced emotional well-being.

As we continue to integrate AI into healthcare, the role of accessible, comprehensible communication remains critical for patient empowerment and informed decision-making.