Oluwatobiloba (Tobi) Ayo-Ajibola

Introduction

Tobi has been diving deep into the world of AI in healthcare, co-authoring five research papers related to artificial intelligence in the last two years. Driven by a fascination with ChatGPT, he, facilitated by Dr. Richard Kravitz, conducted a national survey exploring its role in medical education and patient care, the results of which were presented at SGIM24. This research reveals surprising insights into how AI may complement or even substitute primary care.

Q: What inspired your research into ChatGPT, and how did this national survey come about?

A: My interest in ChatGPT arose from a personal fascination with the rapidly evolving tool back to its initial public release in December 2022. Having explored the tool for various informal uses, I was curious as to the level of accuracy for response to inquiries related to cardiac pathophysiology, the preclinical exam module I was in at the time. This quickly turned into an interest (and some concern) in seeing what sort of response could be garnered to medical advice inquiries. Seeing the capability of the tool for answering these inquiries, I was curious as to how early users were using the tool for this purpose. At this point in February 2023, I had put together a preliminary survey draft but had yet to find a mentor with the expertise to champion and guide our work. During my literature review of past works exploring the use of online health information among patients, I came across multiple studies by Dr. Richard Kravitz, MD MPH, the director of the University of California Center, Sacramento, and was drawn to his expertise in the area. After emailing Dr. Kravitz, we soon met and bonded over a shared interest in this topic and a shared background in the Sacramento area. From then on, Dr. Kravitz was the key facilitator in creating a detailed survey based on previously validated works related to similar adoption of online health information during critical stages, such as the dawn of the internet and social media, which helped to focus and boost the effectiveness of our survey.

Q: In the title of your poster, you pose the question, “Is ChatGPT a complement or a substitute for primary care?” Based on your results, what did you find the answer to be?

A: We found that access to a usual source of primary care decreased the likelihood of using ChatGPT-derived health information. More specifically, when we subdivided the type of online health information as passive sources (encyclopedias, WebMD, etc.) vs. active (forums, chatrooms, Q&A sites) with ChatGPT as a subset of active OHI, we found that receiving regular care in a community-based office or clinic setting was associated with more use of passive OHI and less use of both active OHI in general and ChatGPT specifically. We believe these results suggest that ChatGPT and other forms of more active participatory health information sources may be less valuable to those with continuous access to primary care. This may be due to their access to a physician who may answer the inquiries they would have otherwise posed to the personalized and conversational tool. These findings seemingly strengthen the importance of the patient-physician partnership during the early stages of this potentially disruptive technology.

Q: Was there anything from your survey results that surprised you?

A: Yes, we were greatly surprised by the substantial use of ChatGPT among those receiving care from their local Veterans Affairs Medical Center. While we have yet to explore a possible explanation for this, we hope to explore this relationship in future investigations.

Q: Do you plan to continue your research? If so, what are the next steps?

A: Definitely. We are excited about the prospect of adjusting our survey to specialty-specific topics such as women’s health or psychiatry to explore how ChatGPT OHI use may change depending on the topic or population. We are open to and actively seek collaboration with experts in these areas.

Q: What role do you believe generative artificial intelligence models like ChatGPT will have in the future of primary care?

A: We believe generative AI models like ChatGPT could play significant roles in primary care. Numerous studies have demonstrated that ChatGPT can provide accurate health information, though there are occasional issues with specificity and instances of “hallucinations,” where the model generates fabricated details. Despite these issues, ChatGPT’s ability to incorporate a knowledge base provided by the user offers an intriguing opportunity for primary care physicians. They would be able to share examples of their preferred sources of medical information reference and writing styles, tailoring the AI’s inputs to align with their practice. Imagine a tool, primed to follow American Heart Association guidelines for hypertension management guidelines, that could guide a patient through a home blood pressure check and offer learning points on the risks of high BP per their findings.

Moreover, as AI uses in medicine continue to expand from drafting personalized letters for insurance approval to real-time scribing using natural language processing, physicians may find multifaceted utility for generative AI models in their practice.