Artificial Intelligence and Patient Education

Real Talk: Eosinophilic Diseases

Co-host Ryan Piansky, a graduate student and patient advocate living with eosinophilic esophagitis (EoE) and eosinophilic asthma, and co-host Holly Knotowicz, a speech-language pathologist living with EoE who serves on APFED’s Health Sciences Advisory Council, have a conversation about artificial intelligence (AI) and patient education, with guest Dr. Corey Ketchem, a third-year Gastroenterology Fellow at the University of Pennsylvania.

In this episode, Ryan, Holly, and Dr. Ketchem discuss Dr. Ketchem’s interests, and his research into using an AI chatbot to provide patient education on eosinophilic gastrointestinal diseases. He shares, in broad terms, the methodology and conclusion of the research and what current and future research he is pursuing about using artificial intelligence to improve patient education and care.

Listen to this episode to learn about the current limitations and potential future benefits of using AI to help patients.

Disclaimer: The information provided in this podcast is designed to support, not replace the relationship that exists between listeners and their healthcare providers. Opinions, information, and recommendations shared in this podcast are not a substitute for medical advice. Decisions related to medical care should be made with your healthcare provider. Opinions and views of guests and co-hosts are their own.

Key Takeaways:

[1:17] Ryan Piansky and co-host Holly Knotowicz introduce the topic, artificial intelligence and patient education, and their guest, Dr. Corey Ketchem, a third-year Gastroenterology Fellow at the University of Pennsylvania.

[1:30] Dr. Corey Ketchem has an interest in allergic inflammation of the gastrointestinal tract, particularly eosinophilic gastrointestinal diseases (EGIDs), as well as artificial intelligence and epidemiologic studies.

[2:01] Dr. Ketchem did his residency at the University of Pennsylvania following medical school. There he met Dr. Evan Dellon, a world expert in EoE. Dr. Dellon became a mentor to Dr. Ketchem.

[2:24] As Dr. Ketchem learned more about EoE, he was fascinated by the many unknowns and opportunities for discovery within the eosinophilic GI field. He wanted to make an impact on patient care.

[2:51] Under Dr. Dellon’s mentorship, he did epidemiologic studies. Seeking specialized training, he ended up at the University of Pennsylvania where he is getting rigorous training in epidemiology to study EGIDs.

[3:18] As ChatGPT was gaining its buzz, Dr. Ketchem saw a lot of clinical applicability. He views AI as an asset in epidemiology and hopes to use it to accelerate his research.

[4:30] AI usually references using computers to mimic human abilities, estimate decisions, or predict outcomes. An example is Natural Language Processing (NLP), to analyze and understand human language. Large Language Models (LLM) use NLP.

[5:08] ChatGPT is based on a LLM. LLMs use NLP techniques to understand vast amounts of text that they are trained on and generate responses in a chat format.

[5:25] Machine learning is another subset of AI that uses statistical techniques to give computers the ability to learn with the data and predict outcomes.

[5:50] The hope is to use these AI techniques to speed up discovery and also minimize human expense or labor.

[6:28] Dr. Ketchem co-authored a paper in Clinical Gastroenterology and Hepatology about an AI chatbot and EoE. He had been inspired by a cardiology paper on whether ChatGPT would create accurate, appropriate answers about cardiology disease health.

[7:19] Dr. Ketchem wondered if ChatGPT could be applied to EoE education. He discussed it with Dr. Dellon and Dr. Krystle Lynch, Dr. Ketchem’s mentor at the University of Pennsylvania, and with Dr. Joy Chang, at the University of Michigan. They came up with a study design.

[8:06] The study asked ChatGPT questions about EoE, focusing on patient education and the therapeutics, and seeing if it gave accurate responses or not.

[8:45] The four doctors developed 40 questions that they gave ChatGPT as prompts and evaluated the responses. They proposed the questions in two ways: each question in an individual chat and 40 questions in a single chat.

[9:41] Analyzing the responses, the study demonstrated that ChatGPT responded with multiple inaccuracies to questions about EoE on general topics, complications, and management. Over half of the responses mixed correct and incorrect information.

[10:09] To evaluate the readability of the responses, the doctors used the Flesch-Kincaid reading ease tool. To understand the output from ChatGPT one would need high school and two years of college. That poses a potential health literacy barrier.

[11:40] The questions ranged from general: “What is eosinophilic esophagitis?”, to complications: “What is a food impaction?”, “What is a stricture?”, to therapeutics: “What are steroids for eosinophilic esophagitis?”, “Can I use a proton pump inhibitor for EoE?”

[12:15] It was not clear where ChatGPT pulled data from to respond to the questions. The data it was trained on was known to be in texts over a year old. Newer data may not have been accessible to ChatGPT.

[13:29] The doctors asked about things that were common knowledge in the eosinophilic GI realm, like dupilumab, and ChatGPT didn’t know much about it because it was a newer treatment option for EoE at the time of the study.

[13:42] The doctors scored the answers on their scientific accuracy and patient educational value. Simple questions got good responses. For questions about therapies and complications, “it wasn’t doing well.” They identified limitations to the study.

[14:14] The doctors asked ChatGPT if EoE is associated with cancer. From their best epidemiologic knowledge, the doctors don’t think that it is. ChatGPT falsely associated EoE with esophageal adenocarcinoma.

[14:34] ChatGPT also associated EoE with Barrett’s esophagus. To the doctors’ best epidemiologic data, they are not sure that there’s a connection.

[15:02] When the doctors asked the questions in individual chats, they asked ChatGPT for medical literature references for the information. It didn’t provide accurate references. Titles and authors were often incorrect and links often didn’t work.

[15:36] The incorrect references were a signal that ChatGPT wasn’t ready to answer complex medical questions. In the more updated versions of ChatGPT, instead of giving references, it says you should consult your doctor, which is the right thing to do.

[15:56] The researchers concluded that implementing this technology requires clinical oversight; it’s a tool that should be used with caution for patients in educating themselves and also from the perspective of a physician who is not an expert in EoE.

[16:29] Dr. Ketchem had been surprised by how long the responses were. He was expecting paragraphs but got pages and pages. He was also surprised by how quickly people were starting to use ChatGPT in other aspects of gastroenterology.

[16:57] While Dr. Ketchem and his team were writing the paper, another study came out about gastroesophageal reflux (GERD) that was somewhat similar to what Dr. Ketchem proposed for EoE. There is rapidly much being published about ChatGPT.

[17:14] Although the results were imperfect, there is potential applicability in patient-facing chats in the future for patient education but not yet there “for prime time.”

[18:33] These chats need to be transparent about where they’re getting data, especially in the medical field.

[18:41] There will always be a role for people in medicine. You can't replace a face-to-face connection with a nurse or a physician with a chat bot.

[19:11] Dr. Ketchem says everyone needs to be careful about using AI tools. He advises patients to always discuss any medical questions with their physician. AI tools are not yet able to provide accurate medical information all the time.

[19:50] Ryan reminds listeners that this podcast is for educational purposes. Always consult your physician before making any changes to your healthcare. If you ask ChatGPT, also consult with your doctor before making any changes to your healthcare.

[20:31] One of the problems with large language models is the potential for inaccuracy. Dr. Ketchem’s gold standard is the medical literature and you don’t know where the large language models are getting their information.

[21:04] Future benefits may include helping patients get answers quicker and becoming more educated. Dr. Ketchem hopes we will get to a point where we can trust these technologies and implement them safely.

[21:37] Government organizations like the National Institutes of Health (NIH) and the U.S. Food & Drug Administration (FDA) are bringing together experts to think about large language models and create regulatory frameworks for their use in healthcare. Dr. Ketchem tells how HIPAA (Health Insurance Portability and Accountability Act) rules are followed to protect patients.

[23:29] Dr. Ketchem sees potential in machine learning to predict which therapies an EoE patient will respond to. AI is also used in colonoscopies to identify hard-to-see polyps. It might be useful in endoscopies to see changes in the esophagus from EoE.

[24:35] AI image recognition could also be applied in pathology. Dr. Ketchem is interested in trying to apply it to work he wants to do in the long term. People are working with pathology specimens to automate the counting of eosinophils. Dr. Ketchem discusses the potential use of AI for epidemiology in pathology.

[25:43] Dr. Ketchem and Holly discuss the potential for using AI chatbots in medical screening questionnaires. There will always need to be a human element.

[27:57] Dr Ketchem speaks to the potential future development of educational videos prepared by AI. It is a complex scenario that would require a lot of training. If a camera is added, AI could analyze where patients are having problems in taking medications.

[29:55] Dr. Ketchem says there are many moving parts in healthcare and many stakeholders, making it difficult to implement AI. It could be used in many aspects, but its use must be safe. Dr. Ketchem thinks it will soon be useful in medical imaging.

[30:57] In the next decade, AI may be used in drug discovery, clinical decision-making, and healthcare administrative operations. The goal is to improve the care for the patient. Personalized care would be an aspirational goal of using artificial intelligence.

[31:29] Dr. Ketchem heard of a computer scientist at a government meeting suggesting a far-future scenario of doctors having digital versions of patients to test the patient’s reaction to a specific medication, based on comorbidities and other medications in use.

[32:30] Holly thanks Dr. Ketchem for sharing his research findings to help others.

[32:40] Dr. Ketchem’s last words: “The future is bright. There are many open avenues to apply these technologies to eosinophilic GI diseases – in diagnostic support, personalizing treatment, and predictive modeling – to make patient care better.”

[33:10] Dr Ketchem is building a research program to use epidemiologic training with artificial intelligence. He hopes to find how to take text from histology or pathology and apply epidemiologic methods, to build a cohort of patients to study diseases faster.

[34:03] Dr. Ketchem hopes to use AI to help predict patient outcomes, regarding who will respond to what therapy and who will have more complications from their disease; those are things he is interested in. There are so many unanswered questions.

[34:30] After Dr. Ketchem finishes his fellowship, he hopes to be an independent investigator, being curious and answering these questions somewhere. If you know of such a job, please let Dr. Ketchem know!

[34:53] To learn more about Dr. Ketchem’s research, please check out the links in the show notes. To learn more about eosinophilic gastrointestinal disorders, visit apfed.org/egids. If you’re looking for a specialist who treats eosinophilic disorders, use APFED’s Specialist Finder at apfed.org/specialist. [35:17] To connect with others impacted by eosinophilic diseases, please join APFED’s online community on the Inspire Network at apfed.org/connections.

[35:26] Ryan thanks Dr. Corey Ketchem for joining us today. Holly thanks APFED’s Education Partners, GSK, Sanofi, and Regeneron, linked below, for supporting this episode.

Mentioned in This Episode:

Corey Ketchem, M.D., M.S.

Penn Medicine

Abstract of paper in Clinical Gastroenterology and Hepatology: “Artificial Intelligence Chatbot Shows Multiple Inaccuracies When Responding to Questions About Eosinophilic Esophagitis”Medscape article about the paper in Clinical Gastroenterology and Hepatology: “ChatGPT Gives Incorrect Answers About Eosinophilic Esophagitis”, by Carolyn Crist

American Partnership for Eosinophilic Disorders (APFED)

APFED on YouTube, Twitter, Facebook, Pinterest, Instagram

Real Talk: Eosinophilic Diseases Podcast

apfed.org/egids

apfed.org/specialist

apfed.org/connections

Education Partners: This episode of APFED’s podcast is brought to you thanks to the support of GSK, Sanofi, and Regeneron.

Tweetables:

“We ultimately came to the conclusion that implementing this technology requires clinical oversight and it’s a tool that should be used with caution.” — Corey Ketchem, M.D., M.S.

“There will always be a role for people in medicine. You can’t replace a face-to-face connection with a chat. That’s just not going to work.” — Corey Ketchem, M.D., M.S.

“There will always need to be a human element to it. The goal is to make [AI for healthcare] as good as it can be. We’re certainly not there yet, but it’s probably closer to being here than we think.” — Corey Ketchem, M.D., M.S.

Bio:

Dr. Corey J. Ketchem, MD is a rising third-year gastroenterology fellow at the University of Pennsylvania, driven by a profound interest in allergic inflammation of the gastrointestinal tract, particularly eosinophilic gastrointestinal diseases (EGIDs). He has acquired a unique skillset in clinical epidemiology and biostatistics that equip him with the necessary tools to conduct rigorous research studies, culminating in a Master of Science in Clinical Epidemiology (MSCE) upon fellowship completion. Dr. Ketchem's passion for EGIDs has spurred a series of epidemiologic investigations focusing on both eosinophilic esophagitis (EoE) and non-esophageal EGIDs, yielding numerous publications in high-quality gastroenterology journals and earning him recognition through various research awards. Moreover, his academic path has included the incorporation of artificial intelligence into his research endeavors, aiming to enhance patient care and facilitate epidemiologic studies. Dr. Ketchem's trajectory is set toward becoming an independent researcher, dedicated to employing high-quality epidemiologic approaches to uncover pivotal insights into EGIDs, advance clinical knowledge, and optimize therapeutic strategies for patients.

Bio: Penn Medicine Division of Gastroenterology and Hepatology Fellows