Can ChatGPT Aid in GERD Management for Clinicians?
Artificial intelligence chatbot, ChatGPT, has shown potential as a valuable source of information for patients and a helpful tool for clinicians managing gastroesophageal reflux disease (GERD), according to a recent study. By using a conversational format, this AI tool could improve clinical efficiency and reduce the number of patient messages and calls, potentially alleviating clinician burnout.
However, researchers led by Jacqueline Henson, MD, from the Division of Gastroenterology at Duke University School of Medicine, Durham, North Carolina, caution that some inconsistencies and content errors observed in ChatGPT’s responses require careful clinical oversight.
GERD is a common and increasingly complex condition to manage, affecting nearly 30% of US adults. While AI technologies like ChatGPT have showcased their potential in medicine, their ability to provide guidance specifically for GERD management remains uncertain.
To assess the accuracy and specificity of ChatGPT’s responses regarding GERD care, Henson and colleagues created 23 prompts based on published clinical guidelines and expert consensus recommendations. These prompts consisted of five questions about diagnosis, 11 on treatment, and seven on both diagnosis and treatment.
The researchers submitted each prompt to ChatGPT three times on separate occasions without offering any feedback. They then enlisted three board-certified gastroenterologists to rate the appropriateness and specificity of the responses.
The results showed that ChatGPT provided appropriate responses to 63 out of 69 (91.3%) queries, with approximately 29% considered completely appropriate and 62.3% mostly appropriate.
However, the consistency of responses to the same prompt was often lacking, with 16 out of 23 (70%) yielding varying levels of appropriateness. Some prompts received both inappropriate and appropriate responses, while others were inconsistent.
For example, ChatGPT failed to suggest considering Roux-en-Y gastric bypass for persistent GERD symptoms with pathologic acid exposure in obese individuals. Additionally, the chatbot presented certain risks associated with proton pump inhibitor therapy as factual information.
Nevertheless, the majority (78.3%) of responses included specific guidance, particularly for prompts related to diagnosis (93.3%). In all cases, ChatGPT advised contacting a healthcare professional for further advice.
Feedback from eight patients with varying educational backgrounds revealed that the ChatGPT responses were generally understandable and useful.
Overall, Henson and colleagues concluded that ChatGPT provided largely appropriate and at least some specific guidance for GERD management, demonstrating its potential as an information source for patients and an aid for clinicians. However, the presence of inconsistent responses and inappropriate information suggests that further improvements are necessary before its application in healthcare, specifically for GERD.
In summary, while ChatGPT shows promise in supporting clinicians and providing patients with information, its current limitations prevent its widespread use in managing GERD.
