Can Artificial Intelligence Create an Accurate Colonoscopy Bowel Preparation Prompt?

Abstract

Background and Aims

Colorectal cancer is the third most common cancer in the United States, with colonoscopy being the preferred screening method. Up to 25% of colonoscopies are associated with poor preparation which leads to prolonged procedure time, repeat colonoscopies, and decreased adenoma detection. Artificial intelligence (AI) is being increasingly used in medicine, assessing medical school exam questions, and writing medical reports. Its use in gastroenterology has been used to educate patients with cirrhosis and hepatocellular carcinoma, answer patient questions about colonoscopy and provide correct colonoscopy screening intervals, having the ability to augment the patient–provider relationship. This study aims at assessing the accuracy of a ChatGPT-generated precolonoscopy bowel preparation prompt.

Methods

A nonrandomized cross-sectional study assessing the perceptions of an AI-generated colonoscopy preparation prompt was conducted in a large multisite quaternary health-care institution in the northeast United States. All practicing gastroenterologists in the health system were surveyed, with 208 having a valid email address and were included in the study. A Research Electronic Data Capture survey was then distributed to all participants and analyzed using descriptive statistics.

Results

Overall, 91% of gastroenterologist physicians determined the prompt easy to understand, 95% thought the prompt was scientifically accurate and 66% were comfortable giving the prompt to their patients. Sixty four percent of reviewers correctly identified the ChatGPT-generated prompt, but only 32% were confident in their answer.

Conclusion

The ability of ChatGPT to create a sufficient bowel preparation prompt highlights how physicians can incorporate AI into clinical practice to improve ease and efficiency of communication with patients when it comes to bowel preparation.