How AI is redefining clinical documentation [PODCAST]

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes

We welcome Matthew Allen, a medical student, as we delve into the significant findings of the American Medical Informatics Association (AMIA) 25×5 Task Force, which aims to reduce the clinician documentation burden by 25 percent in five years. Matthew will share his unique perspective on how generative AI can transform health care documentation, shifting the focus from creation to information retrieval and synthesis. We’ll discuss the challenges of managing the ever-growing data from patient records and wearable devices, and explore innovative solutions. Matthew will help us understand the potential of generative AI in creating dynamic, intuitive ways of presenting medical information, far beyond traditional text-based methods.

Matthew Allen is a medical student.

He discusses the KevinMD article, “Are we missing the mark with generative AI?”

Our presenting sponsor is Nuance, a Microsoft company.

Together, Microsoft and Nuance are leveraging their rich digital technology and advanced AI capabilities to tackle some of health care’s biggest challenges. AI-driven technology promises to revolutionize patient and provider experiences with clinical documentation that writes itself.

The Nuance Dragon Ambient eXperience, or DAX for short, is a voice-enabled solution that automatically captures patient encounters securely and accurately at the point of care. DAX Copilot combines proven conversational and ambient AI with the most advanced generative AI in a mobile application that integrates directly with your existing workflows.

Physicians who use DAX have reported a 50 percent decrease in documentation time and a 70 percent reduction in feelings of burnout, and 85 percent of patients say their physician is more personable and conversational.

Discover AI-powered clinical documentation that writes itself. Visit https://nuance.com/daxinaction to see a 12-minute DAX Copilot demo.

VISIT SPONSOR → https://nuance.com/daxinaction

SUBSCRIBE TO THE PODCAST → https://kevinmd.com/podcast

RECOMMENDED BY KEVINMD → https://kevinmd.com/recommended

GET CME FOR THIS EPISODE → https://earnc.me/agTWS2

Powered by CMEfy.

Transcript

Kevin Pho: Hi, and welcome to the show. Subscribe at kevinmd.com/podcast and get CME for this episode by clicking on the CME link in the show notes. Today we welcome Matthew Allen. He’s a medical student, and today’s KevinMD article is titled “Are we missing the mark with generative AI?” Matthew, welcome to the show.

Matthew Allen: Glad to be here.

Kevin Pho: So let’s start by briefly sharing your story and journey to where you are today.

Matthew Allen: In undergrad, I interned at a health tech startup company that was doing eConsult implementation. I was young and naive, but we’d go out to medical clinics and train doctors on how to use the system that connected the primary care docs with the specialists. This health system was having an issue: There are not enough specialists, a lot of specialty visits are not necessary, and the primary doc can handle the patient’s condition. So if we can connect them virtually, we could solve that problem. That kind of opened my eyes: Wow, doctors really are unhappy with their computers. Instead of seeing them as a help, they see them as a hindrance, something between them and patients. I was really confused by this. I was thinking, “Man, this should make medicine a lot easier.”

So I kind of got introduced to that whole world, and when I was applying to medical school, I was lucky enough to get accepted to UC San Diego. They are a national leader in informatics and how clinicians are using computers, and so I was just so excited to take the opportunity to come to UC San Diego and get more involved in that world.

Kevin Pho: Perfect. Let’s jump into one of the hottest topics, generative AI, things like ChatGPT and Google Bard. Right now we have so many health care startups kind of tripping over themselves to see how generative AI can apply to physicians today. I’m interested in hearing your insight in your KevinMD article titled “Are we missing the mark with generative AI?” So tell us about this article.

Matthew Allen: Yeah. The reason I wrote the article is because early on, a lot of the excitement was, “Hey, if I type a question into this ChatGPT, it’ll totally produce an answer. We could use that to answer patient messages. Doctors’ inboxes are overflowing; let’s use it to solve that problem. Doctors don’t like taking time to write these really long notes; let’s use it to solve that problem.” And those are all worthwhile use cases; I want to emphasize that.

But I was also reading a lot of articles that said, “Hey, the chart keeps getting longer and longer.” The average medical chart now is actually half the length of Shakespeare’s Hamlet, and a lot of the information in the chart is duplicate. Copy and paste is our friend, but it’s also our enemy. So really, we have an overabundance of low-quality information in the medical chart, and we’re not wanting some machine to just produce a bunch more. So I worried: Are we just going to use this to automate a bunch of stuff and make notes even longer and make the charts even more bloated, and then clinicians down the road are going to have to just sort through all this information? So that kind of got me thinking about the article.

Kevin Pho: All right. So you mentioned that there’s a task force, the AMIA 25×5 Task Force, regarding the current state of clinical documentation burden. Tell us about the key findings from this report, and use that as a jumping-off point in terms of how generative AI can or cannot help us.

Matthew Allen: Totally. The American Medical Informatics Association saw that documentation burden was a huge issue, and they started this task force to try to reduce it by 25 percent within five years, which is ambitious. One of the first things they did was look at all the current efforts being done by organizations to reduce documentation burden. What was interesting is that all of those efforts focused on document creation, kind of what I was talking about before, automating notes or responding to messages, etc. Nothing focused on information retrieval, chart review. A new patient’s coming into my office; how am I going to familiarize myself with them? Or maybe I’m in an acute care setting, the emergency room. I don’t know this patient. I’ve got three to five minutes to scroll through old notes, try to find what’s important about this patient, and be ready for them. So information retrieval is a huge, huge part of documentation burden, but sometimes we don’t think of it as that. We just think about writing notes. So AMIA definitely highlighted that there was a gap there.

Kevin Pho: So in terms of document creation, we’re only about one year into generative AI. ChatGPT was released a year ago. Do you feel that document creation is kind of an iterative step, and perhaps the next step then would be information retrieval?

Matthew Allen: I would think that the reason document creation is being focused on is because it’s relatively low-hanging fruit. I agree, I do agree, and it’s not that there’s nobody working on this, and we should go after the low-hanging fruit initially. But I think people need to understand that documentation burden is about more than just creating the documentation, and if we just focus on that, we might make our other problem harder. We might make there be more stuff that we have to summarize and sort through. So I think it should be more of a focus even from the start.

Kevin Pho: So you mentioned that you’re at UCSD, which has a pretty strong clinical informatics program. What’s the general consensus there among the thought leaders in health care IT regarding that intersection between generative AI and health care?

Matthew Allen: I’d say, I mean, almost everyone here is very bullish on generative AI. They think it is going to be incredibly impactful in a lot of ways that we probably don’t even think about right now. I would say, as far as specific use cases, some people are split. Some people think answering patient messages is going to be totally revolutionary and help doctors a lot; others are very skeptical. With summarization, some people think the use case is obvious and it’s going to work. Other people think, “Yeah, there’s no way. Large language models, once the document that it has to summarize gets too long, it’s not very good at it.” So honestly, there’s a split about some of the use cases.

Kevin Pho: So you have a background. You said that you interned at a health care startup, and you’ve kind of seen the evolution of generative AI. So taking that next step, the intersection between generative AI and information retrieval, what do you see as an ideal scenario? Give some hypothetical case studies where you see that usefulness of generative AI in terms of information retrieval.

Matthew Allen: I mean, I think what would be awesome is something like a Wikipedia page for a patient. Instead of going into a patient’s chart and having tons of old notes and tons of old labs, and we could still look at that stuff if we want to, you have a Wikipedia page with different headings that everybody agrees are important. You could expand them, you could collapse them, and in three minutes you can have a really good idea of who this patient is, what’s important about them, and what’s important to your specific specialty or context.

One thing I’d also mention is we need to be rigorous about evaluating these tools. In digital health and generative AI, there’s a lot of excitement, which I think is good, but we need to make sure that we don’t get caught up in that excitement and start implementing a bunch of stuff without rigorous evidence that it helps patient outcomes. Patient outcomes should be at the center of everything we do. What’s nice about digital tools is that in a lot of ways, it’s easy to study them. You can turn on the tool for half the doctors and not for the other half, or half the patients and not for the other half, and you should be able to rigorously evaluate their real-world performance in real time fairly easily. So I think we can’t get so excited that we miss that evaluation piece, because then we’re going to lose doctors’ trust even more that technology is going to help them, and that would be the worst-case scenario.

Kevin Pho: What about the impact on physician and clinician burnout? A lot of these tools, I see, are marketed to make documentation easier, and documentation, as you know, is one of the pain points that a lot of physicians complain about. Should patient outcomes be the only metric? What if it improves clinician burnout and helps them stay practicing a little bit longer? How important is that?

Matthew Allen: You make an excellent point, and it is very important. Patient outcomes and physician burnout are tied together. If your physician is burned out, they’re not going to do as good of a job. So I absolutely agree with you that physician burnout should be part of the conversation. I do think that sometimes, when we’re talking about all this stuff, all the articles will be about technology and physician burnout, so I think maybe there’s just a disproportionate focus on that, and so we also have to think about patient outcomes. Maybe that’s because I’m young and naive and I’m not quite burned out, but I think that’s what most of the literature is on, physician burnout and how technology is going to help that. So I think maybe we broaden our perspective a little bit.

Kevin Pho: So what do you see in the coming months? I think it’s evolving so quickly. As I mentioned before, we’re about one year into the introduction of ChatGPT, and it’s created some seismic shifts in terms of what we can expect at that generative AI and health care intersection. What do you see the next few months bringing, just from what you’re seeing in your environment at UCSD?

Matthew Allen: It’s a great question. I think the tools are going to get better very quickly. There are barriers to doing what we’re talking about. One would be cost: Using things like ChatGPT, every time you make a query, there’s a cost to that. There are also limits to how much information you can feed the tool to summarize. All of those things over the next six months will go down. The cost will go down, ChatGPT will be able to accept larger amounts of information to summarize, and all those metrics are going to improve. So some of the hurdles we’ve currently been facing with some of these use cases will become not as important with the advance of the technology. I think we’ll start to see more use cases than just patient messaging and automated note-taking, and I hope that comes with large academic medical centers running randomized clinical trials on different applications of generative AI in the next six months.

Kevin Pho: Talk about some of the downsides and pitfalls that we need to look out for as generative AI becomes more pervasive in our health care system.

Matthew Allen: That’s a great question. Some are technical. You could have new attack vectors for cybersecurity and HIPAA concerns. You could have deskilling of physicians if we rely on the computer even more, or when we don’t have access to it, are we going to be able to be physicians? I do think those will take a little bit longer. I think whenever you’re using technology to solve problems in health care, it can make it more human or less human. So we want to use technology that fades into the background and allows us to connect with patients and have that personal relationship. I think what we need to look out for is making sure that it doesn’t take more and more human connection out of medicine and make it more impersonal, more “I’m chatting with a bot instead of talking to the person,” if that makes sense.

Kevin Pho: So I want to shift the conversation. This isn’t actually part of your article, but just from your personal experience, you’re in your second year of medical school. How has generative AI affected medical education from your perspective?

Matthew Allen: That’s a great question. People do use it for sure to study and to answer questions, and I think it’s helpful. In some ways, medical students have already for a long time been using the internet and third-party resources to study and learn medicine, so I don’t think that it’s a fundamental shift, but it can make you a lot more efficient. I was actually talking to my wife the other day, who’s in law school, and she’s like, “Oh, you wrote that article. You didn’t even know I’ve been using that a ton in law school. When I have to do a lot of procedural history reading, I plug the huge, massive text into ChatGPT and have it summarize it for me, and it just helps me so much.” She gave the caveat that she has to teach the bot what is legally pertinent information to pull out of the lot of text. But I think the same goes for medical students: If we have to digest a ton of information, this could help us do it more efficiently.

Kevin Pho: Now, is it pretty commonplace among you and your fellow classmates?

Matthew Allen: For sure. Yeah, commonplace. I mean, initially everybody was like, “Oh, what is this tool?” and some people were more hesitant to try it. But now everybody’s familiar with it. I don’t think everybody uses it all the time, but a lot of people do, and everybody’s very aware of it and its implications.

Kevin Pho: How about the medical school professors? Have there been any guidelines in terms of how to use and how not to use generative AI in medical education?

Matthew Allen: Good question. Yes, professors bring it up. I would not say that there’s a consensus. Some of them are more, “Yes, use it for everything.” Others are more, “Hey, on this task, don’t use generative AI, because I worry that it will take away from your learning experience.” So it’s kind of professor dependent and situation dependent.

Kevin Pho: Any concerns the professors bring up regarding patient care, like patient privacy? Has that been formally brought up in your medical education, among your classes?

Matthew Allen: That specific topic, no. I mean, institutionally, definitely, UCSD is going to be saying, “Do not put any private patient information into open-source AI bots.” That is definitely a security risk. But specifically at the medical school, no.

Kevin Pho: We’re talking to Matthew Allen. He’s a medical student. Today’s KevinMD article is titled “Are we missing the mark with generative AI?” Matthew, let’s end with some of your take-home messages that you want to leave with the KevinMD audience.

Matthew Allen: Awesome. For my first takeaway, I’d say to clinicians: We need clinicians to be involved in this. If you’re talking about something like automated notes or chart summarization, we need what physicians want to be prioritized so that they can do a good job of taking care of their patients, and that means they need to be involved in the process. If medical-legal or regulatory or administrative or quality people are the ones who decide how these tools are going to be used, it might not look how your average clinician wants it to end up looking. So, clinicians, get involved in these tools that are being developed at your organization.

And then my other takeaway would be for my fellow med students. I’m sharing this in part to keep myself accountable, but I read recently a big report that a lot of people are talking about, that the majority of us medical students are planning on either leaving medicine or not treating patients. Everybody’s talking about burnout and medicine’s hard, and so a lot of medical students are like, “Man, maybe I’m just going to be a researcher, or I’m just going to work for a startup, or I’m going to be a thought leader,” or whatever. Those are good things, but we need people to take care of patients. So I’d encourage all of my classmates: Even if you’re not going to do full-time patient care, I don’t think we should ever give up on that as part of what we do. And we should trust that these tools and technologies are going to solve a lot of the issues that are taking the joy out of medicine, and so hopefully our careers will be less burnout-ridden and we can enjoy patient care.

Kevin Pho: Matthew, thank you so much for sharing your perspective and insight, and thanks for coming on the show.

Matthew Allen: No problem. Thanks for having me.

Listen and followApple PodcastsSpotifyYouTubeMore apps · Browse all episodes