Subscribe to The Podcast by KevinMD. Watch on YouTube. Catch up on old episodes!
Internal medicine physician Melvin Speisman discusses his KevinMD article, “How AI is transforming medicine faster than ever before.” In this episode, Melvin explores the rapid advancements in artificial intelligence within health care, including the approval of over 950 AI medical devices by the FDA as of October 2024, the deployment of AI algorithms in hospitals, and the significant impact of AI in radiology and cardiology. He also delves into the pioneering work in ophthalmology and the recent Nobel Prize-winning developments that are accelerating medical research. Listeners will gain insights into the current trends, potential risks, and innovative solutions AI offers to improve patient care, along with actionable takeaways for integrating AI into medical practice.
Our presenting sponsor is DAX Copilot by Microsoft.
DAX Copilot, by Microsoft, is your AI assistant for automated clinical documentation and workflows. DAX Copilot allows physicians to do more with less and turn their words into a powerful productivity tool. DAX Copilot automates clinical documentation—making it available in the EHR within minutes—and clinical workflows, including referral letters, after-visit summaries, style and formatting customizations, and more.
70 percent of physicians who use DAX Copilot say it improves their work-life balance while reducing feelings of burnout and fatigue. Patients love it too! 93 percent of patients say their physician is more personable and conversational, and 75 percent of physicians say it improves patient experiences.
Discover AI-powered solutions for clinical documentation and workflows. Click here to see a 12-minute DAX Copilot demo.
VISIT SPONSOR → https://aka.ms/kevinmd
SUBSCRIBE TO THE PODCAST → https://www.kevinmd.com/podcast
RECOMMENDED BY KEVINMD → https://www.kevinmd.com/recommended
GET CME FOR THIS EPISODE → https://www.kevinmd.com/cme
I’m partnering with Learner+ to offer clinicians access to an AI-powered reflective portfolio that rewards CME/CE credits from meaningful reflections. Find out more: https://www.kevinmd.com/learnerplus
Transcript
Kevin Pho: Hi, and welcome to the show. Subscribe at KevinMD.com/podcast. Today, we welcome Mel Speisman. He’s an internal medicine physician. Today’s KevinMD article is “How AI is transforming medicine faster than ever before.” Mel, welcome to the show.
Mel Speisman: Kevin, thanks for having me on the podcast today.
Kevin Pho: So let’s start by briefly sharing your story and journey.
Mel Speisman: For sure. I’m internal medicine, and I’ve been in medical education for 30 years and an active member and fellow of the American College of Physicians. I’m from Chicago, did my undergraduate education at Northwestern University in biological sciences and chemistry, and then I did my medical school at the University of Illinois College of Medicine here in Chicago.
Recently, I took a position as an assistant professor at the University of Illinois in the College of Medicine, Division of Academic Internal Medicine, where I got exposed to many invited guest speakers who were brought in by the dean or the chairman of medicine to give grand rounds or special Nobel Prize–type winning lectures. These speakers, during my time there in the last few years, had a big focus on artificial intelligence and how to apply it to medicine.
So that’s how I got pulled into this. And then, being inspired by what ACP does as a professional society, I realized that for physicians like you and me, there’s really no professional home right now for AI in medicine specifically. There are a lot of organizations that may include doctors in this space, but not a real home. So I actually stepped down from my dream job—being an assistant professor at a medical school—to launch my mini version of the American College of Physicians, which we call the American College of Artificial Intelligence in Medicine. So that’s my story, and I’m sticking with it.
Kevin Pho: All right. So let’s talk about your KevinMD article, which discusses that intersection, of course, between AI and medicine. It’s evolved so quickly since ChatGPT was introduced. Tell us what this article’s about for those who didn’t get a chance to read it yet.
Mel Speisman: Yes, so the article is a very quick read, very concise, and tells you essentially all the key things you need to know as of today on how artificial intelligence is being used to care for patients. I really break it down into categories.
How it blew up into medicine is what you said: OpenAI released ChatGPT a little more than two years ago, in November 2022. Prior to that time, there was a decade’s worth of work in both ophthalmology and radiology on AI applications in medicine using machine learning, but it was very much under the radar—physicians weren’t really aware of it unless they were in those fields. Then, when ChatGPT came out and the entire world saw this tool that can transform whatever it is they do, the uptake in the health care field was phenomenal.
If you look at publications—say in PubMed—and you look at any specialty within medicine, prior to the year 2020 there were fewer than a thousand articles on AI in that area, maybe 700 at the most, since the entire history of medicine. Since 2020, those fewer than a thousand articles have blossomed into at least 5,000 papers—or 1,000 per year. The rate of growth of new data on AI and medicine is so exponential that literally no human being could keep up. Our organization is focusing on educating doctors at all levels—medical students, residents, fellows, attending doctors—anyone who wants to learn how to use AI to take care of a patient. We’re that professional home to help sort through this massive amount of knowledge and information.
One thing that’s really striking is that I’m in Chicago, and most major medical conferences are at McCormick Place. We’re actually going to hold one there on September 19th and 20th, our own scientific session for AI in medicine. The cardiology conferences like AHA and ACC are held here in Chicago at McCormick Place, the radiology conferences—RSNA—was in its 111th year this year, and I was there one month ago.
The entire session was about how AI is being implemented in radiology, and I got to meet leaders from all over the world who were there at Chicago McCormick Place to network and collaborate. There are almost 1,000 FDA-approved AI-for-medicine algorithms that could be used today, and every single one of those FDA-approved algorithms was being displayed in the exhibit hall. The doctors attending this conference and the health care leaders were all trying to decide, “Which one am I going to pick? Which one should I use?” So I’ll turn it back to you, Kevin.
Kevin Pho: For those physicians who may be outside the AI bubble, let’s say in a patient care setting—because you said that AI, before ChatGPT, was primarily used in ophthalmology and radiology—how is AI being used, say, in an internal medicine primary care setting today?
Mel Speisman: Kevin, I love that question so much. Thank you. In March of last year—we’re coming up on about one year—here in Chicago, for example, Rush University and Northwestern University, and now the University of Chicago, have all deployed in primary care medicine what’s called ambient listening systems. This is like an Alexa-type speaker—it’s not Alexa, but that’s the easiest analogy. So for those listening, I’m trying to make it basic. If you took the technology of an Alexa-type speaker and you put it in the patient room during the visit, the doctor no longer has to be glued to the keyboard or the screen.
We can go back to doing what we love, what brought us to medicine in the first place, which is building that doctor-patient relationship. We can maintain eye contact, read the emotions of our patient, and give counseling and feedback in a very empathetic way because we don’t have to type everything in and click during the visit. We can not look at the computer at all. It will listen to the entire conversation, and before we even leave the room, it will have generated, in the electronic medical record, the full documented note of what happened—plus the problem lists, the plans, the potential medicines that need to be prescribed, the referrals to specialists. This is amazing.
This has been implemented now for the last year, and one in every three hospitals and health systems in America are currently using one of these ambient listening systems. I’m fully not-for-profit; I don’t take any funds from any brands, but I’ll just tell you Rush partnered with a company called Suki (S-U-K-I). They purchased their ambient listening system from that company and deployed it in March. Northwestern, Abel Kho, who has his own department of AI in medicine, created their own system in conjunction with Microsoft. Many other university hospital systems—again, not influenced by branding—the one that’s probably had the best uptake across academic medical centers, such as the University of Chicago, is called Abridge, and they basically do the same thing. They listen to the doctor-patient conversation, link it to the EHR (like Epic), and generate an amazing document that is far better than anything a physician can do.
I’ve done primary care internal medicine at the University of Illinois for the last several years until I switched to being president of this AI organization. I will tell you, this is such an amazing thing because the burden on a physician to do EMR documentation at night after hours, in your pajamas—and I know, Kevin, you know this, you do it every day—it’s almost impossible. This is literally such an impactful thing. So this is how AI is currently being deployed and used. The other thing that’s currently done is many radiology departments have embedded AI in their scanners to help with reading chest X-rays, CT scans, MRIs. Those are the two ways it’s actually happening today. I’ll turn it back to you for more questions, Kevin.
Kevin Pho: So I hear you when it comes to ambient AI scribes. Moving on beyond that, do you see AI having value in terms of the diagnostic approach, or in terms of helping with inboxes, going beyond scribing? Where do you see AI infiltrating a primary care physician’s life past the documentation part?
Mel Speisman: Great. I’ll break it down into two parts. One is the inbox. Once electronic medical records were fully implemented in the U.S., patients had a portal where they could send a message to their physician. This put an incredible burden on the doctor who’s just trying to keep up with the notes from the visits during the day. They have to circle back to inbox messages. In practice, it’s almost impossible to do your visits eight hours a day, document for another 16 hours, and do your inbox. It’s to a point where this kind of burden has made many doctors literally leave medicine.
So I think we’ve solved it with AI, both with the note-writing and with the inboxes. Right now, most of those same medical centers that have deployed the ambient listening system for note-writing are also using it to generate a draft for the doctor to start with, to then respond to an inbox question from a patient. But the doctor, just like the doctor has to read every word of the note in the EMR before signing it, has to read that draft, modify it to that specific patient, and then sign it. Still a tremendous benefit. Both of these things are amazingly beneficial to the doctor and to the patient.
You can tell I have a lot of enthusiasm and optimism about the benefits. So you also mentioned diagnosis. The places that are really in the lead for decades on this are on both coasts. Harvard University, with Zak Kohane’s group in the Department of Biomedical Informatics and their New England Journal of Medicine AI. That group on the East Coast, and then Stanford, with its Center for Artificial Intelligence in Medicine and Imaging. They both have done a lot of research on using AI to diagnose rare diseases.
Zak actually has a rare disease research program that’s nationwide, funded through the NIH, where anyone with an undiagnosed disease can enroll for free in this program. They will use AI to help diagnose someone who may have gone 10 or 20 years without a diagnosis. There are so many examples of patients who may have seen 10 or 15 doctors and still not had a diagnosis, but if that same information is entered into a large language model such as ChatGPT, all of a sudden the diagnosis is made. This is the true power of what we can do with large language models.
There’s one woman who’s very prominent in this regard—most people at my level who speak on this use her example—she saw 12 different doctors for her child, couldn’t get an answer. She put everything into ChatGPT herself out of frustration, got an answer, went to the proper specialist, her child underwent a surgery on his spine, and was cured when no one could help him before. That is a heartwarming story about the power of AI in helping people with undiagnosed or rare diseases.
One of my mentees, and I’ve taught over 1,500 students and residents, he has a son, 22, who’s had an undiagnosed condition since birth—no one could figure it out. He told me all the symptoms. I put them into ChatGPT. I had an answer in seconds. This is literally my own colleague, a physician, who could not, along with his physician wife, for 21 years get a correct diagnosis for their child—and in 60 seconds, I had the answer. I hope these examples I gave you on diagnosis inspire you to want to use it as a tool to really help patients. I’ll turn it back to you.
Kevin Pho: I was listening to Mark Zuckerberg on a recent podcast, and he said that the current state of AI is on par with a mid-level coder at Facebook. A statement like that put fear in a lot of computer science graduates, wondering if they’re going to have a job or if AI is going to take their job. Do you see the same thing happening in medicine, where the diagnostic abilities of AI become so good that health systems are going to replace physicians with cheaper AI alternatives? And what does that threat mean to our livelihood going forward?
Mel Speisman: Great question. I have a very clean answer to that: No, it will not replace physicians. I speak and teach about AI virtually every day at medical schools and hospitals, and what I tell my mentees is the same message I’d like to give to your listeners: AI is a tool, like a stethoscope. It’s not going to replace doctors. It’s simply going to be a powerful tool that will allow us to diagnose and treat patients in ways unimaginable, but it will not replace you.
It will always require physician supervision, and you and I both know about “hallucinations,” which means AI needs supervision. So my message is that you always have to supervise AI as if it were your medical student. A medical student can write a prescription, but they can’t sign it and it can’t be filled unless the attending doctor cosigns it. That’s how I want you to think about AI—whether it’s now or a decade from now—like a medical student. It can help you, but you have to read every word, check for mistakes, check for hallucinations. In the end, you’re signing off on it. It may not be correct, which is why you have to read every word of that ambient note and not just click “sign.” You must supervise it as if it were a medical student, knowing there could be a mistake. As a physician, your role is to oversee it as the teacher, as the faculty member, as the attending doctor, advocating for your patient and checking for errors.
This term “hallucinations” is a serious issue, and that’s why we’re all approaching AI in a way that’s cautious. We want to ensure safety. I’ve been advocating, through our organization—the American College of Artificial Intelligence in Medicine—for an evidence-based approach to this. Right now, out of the 1,000 FDA-approved AI algorithms, very few have been tested in the evidence-based medicine framework that you and I use for pills. You and I would never give a drug to a patient unless it withstood the rigors of a randomized trial. So our organization, and our national meeting in September, is going to emphasize an evidence-based medicine approach to using AI to care for patients.
Kevin Pho: We’re talking to Mel Speisman, an internal medicine physician. Today’s KevinMD article is “How AI is transforming medicine faster than ever before.” Mel, let’s end with some take-home messages that you want to leave with the KevinMD audience.
Mel Speisman: Great. Number one: Don’t be afraid. The more knowledge you gain about AI in medicine, the less fear you’ll have. The only people I know who are fearful of AI in medicine are those who haven’t educated themselves about it. As physicians, when new developments occur in medicine, what do we do? We read the articles, we learn the new data, and you have to apply that to AI now. Because thousands upon thousands of articles are being published every year on AI for medicine, use the same techniques you’ve used your entire career. Learn about it, and you won’t be afraid of it. Use it as a tool to help patients, improve care, and save millions of lives worldwide.
Kevin Pho: Mel, thank you so much for sharing your perspective and insight, and thanks again for coming on the show.
Mel Speisman: Kevin, it’s such a pleasure, and I want to thank you for the amazing work that you do to get the message out and to educate not only physicians but the general public about important topics in medicine. Really, bravo to you for this podcast, for your website, and for all that you do to teach. I’m honored to be a part of it today, so thank you for what you do.
Kevin Pho: I appreciate the kind words, Mel.
