Subscribe to The Podcast by KevinMD. Catch up on old episodes!

Join Maria Iliakova, a bariatric and general surgeon. We’ll dive deep into the world of artificial intelligence (AI) and its remarkable applications across various industries, including health care. Maria will shed light on what AI is, how it’s transforming medicine, and practical ways health care professionals like herself can leverage AI to enhance patient care.

Maria Iliakova is a bariatric and general surgeon.

She discusses the KevinMD article, “AI for health care professionals: How it is used now and how to use AI as a clinician.”

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/GCuXsS

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 back Maria Iliakova. She’s a bariatric and general surgeon. Today’s KevinMD article is “AI for health care professionals: How it is used now and how to use AI as a clinician.” Maria, welcome back to the show.

Maria Iliakova: Thank you so much, Kevin, great to be with you.

Kevin Pho: So Maria’s been on multiple times. Go to KevinMD.com/podcast to search her name and prior episodes to hear her story. But today let’s jump right into your most recent article about AI for health care professionals. So tell me what this article is about.

Maria Iliakova: You bet. So this article really is about artificial intelligence, or AI. And the reason why I wrote it, obviously, is because it’s been in the news like crazy lately, and I wanted for folks to get a sense of how it’s being used now in health care, because I think a lot of people don’t necessarily know that it’s already in use actively. And I definitely wanted to touch on how clinicians can actually use AI in their own practices, to their benefit and for the benefit of their patients.

Kevin Pho: All right, so how is AI being used today in health care?

Maria Iliakova: You bet. So actually, a lot of ways. It’s already built into our electronic health record systems, for instance. So if you’re using things like Epic or Cerner or some of the other bigger electronic health record providers, you’ll notice that you actually get recommendations built into your notes oftentimes. Even it says, oh, it looks like you’re writing this kind of note, here’s a template for instance recommended to you. Or here’s some language in your note that’s recommended to you. Or here even are some diagnosis, CPT codes that are recommended to you. So those are all examples of how recommendations are actually already being built into the software that we’re using.

Outside of that, there’s also quite a lot of object recognition kinds of AI that are being used in health care as well. So you’ll notice that pathology and radiology are really obvious ways in health care that we use images in order to guide diagnosis and treatment. And so there are artificial intelligence platforms that are already being used by radiologists and pathologists in order to identify everything ranging from cell structure to abnormal anatomy on a CT scan or an MRI.

I wouldn’t say that’s super widely used currently, that’s an area under a lot of heavy investigation right now. But there are some models that are even demonstrating they work as well as health care providers, and in some cases even better, which I think really gives us a run for our money in terms of determining what are the capabilities of the software now, and certainly where it goes in the future, versus human professionals that provide health care.

One other that I wanted to mention too, also something that probably a lot of folks have either heard of or used themselves, is speech recognition software. So even when you’re talking to your own Alexa or Google Home or something like that at home, or you’re using your phone, you’re able to say, find this location on a map for me, and automatically it will navigate you there. Or, pick up this song for me that I really want you to play on Spotify, for instance, and it’ll be able to do that.

In the health care setting we have something similar, in the sense of being able to dictate notes and then being able to get accurate dictations and accurate notes out of that. And some of that is human based still, but a lot of that is actually being converted to artificial intelligence speech recognition.

Kevin Pho: So AI of course has been popularized by these chatbots for the last year or so, but AI has been involved in health care, from what you’re saying, for a far longer time before this past year.

Maria Iliakova: Exactly. And so I’m really glad that it’s come to our attention a lot over the past year. So OpenAI is a company that developed ChatGPT, and they launched that product in November of, I believe, 2022, so now just over a year ago. And that really, I think, put a fire under a lot of people, of discovery and realizing that they could actually interact with artificial intelligence, or AI, in a way that they could do from their own phone, they could do from their home, and really for creative reasons.

But you’re absolutely right, these technologies have actually been in place for decades, and there’s been a lot of iteration of how it works. Some of the most obvious ones absolutely are happening at like the Netflix level or Spotify level, in terms of recommending songs or recommending movies or shows for us. But the exact same thing happens in health care too.

Kevin Pho: So how are you using AI in your daily, so you’re a bariatric general surgeon, tell us about that intersection with AI and your own professional clinical life.

Maria Iliakova: Absolutely. So I am no different than anyone else in the general public, in the sense of, when OpenAI announced ChatGPT I got right on and wanted to check it out and see how it worked. And pretty quickly actually I was starting to use it in my clinical practice on a regular basis.

And how does that look? So whenever you have to, for instance, write a letter to insurance based on a denial that you have to appeal, looking at a blank page or somebody else’s template can be tough. So I actually asked ChatGPT to create a template for me for a denial, given some citations and given some specific things I wanted in it. And what it created I think was a really workable, great template.

It’s not something that works from scratch. Think of it like a really intelligent assistant, or maybe a really intelligent intern. But it’s certainly not something that you would be able to send to an insurance company for an appeal without any kind of your own analysis of it and updating and so forth. But I think it absolutely eliminates that sort of the terror of the blank page, if you will, for any kind of documentation.

And to be honest with you, we even revised a lot of our policies in our program. So we are an MBSAQIP center of excellence, and that comes with a lot of standards and policies that have to be updated on a regular basis. So we even use ChatGPT in order to recommend updates and to accomplish some of the writing and scripting for some of the things that we use in our program as well.

So those are just some super basic, you don’t have to have any skills whatsoever in AI in order to interact with some of these technologies and really be able to use them for some really great writing and some creation of content.

Kevin Pho: So some of the things that you mentioned, they’re very helpful in terms of reducing that paperwork and bureaucracy and lessening the burden of electronic health records. I think AI is very powerful in helping clinicians with that. How about next level applications, in terms of helping with things like medical decision making? Where are we in terms of AI helping us with that?

Maria Iliakova: Absolutely. So I actually read an article not too long ago that was pitting AI against clinicians in making decisions, in triaging in emergency settings. And there is some evidence that in some cases the AI is actually capable of making decisions at the level of clinicians. And that honestly has been going on for, like we mentioned, decades in some cases. I think a lot of people may remember IBM’s Watson from the 90s and early 2000s, and there was some discussion there also of that being used for clinical decision making.

In my own practice, definitely we have things like risk calculators that we use, and all kinds of adjuncts to help us make decisions based on who’s a good candidate for surgery, who’s a good candidate for other things, who’s high risk, who’s low risk, and even sometimes in determining a diagnosis for patients.

There’s a lot of potential in using artificial intelligence there, because if you think about it, it’s a way to understand the 95 percent. It’s not great at giving you an accurate assessment of the 5 percent that’s complicated or challenging or really requires synthesizing a lot of potentially novel information. But if you think about the 90 to 95 percent of things that are fairly routine, well, that’s exactly what artificial intelligence does really well. It takes in a lot of data from a lot of different sources and is able to integrate it in order to be able to give you a really good outline of, hey, this is the most likely thing based on the information given, here’s the next likely thing. So it’s very, very powerful at doing things like clinical decision making and providing a baseline for that.

Kevin Pho: So in your world, what exactly does that look like when you say that you can use AI for risk assessment of patients? Do you use ChatGPT, do you have specific tools that are HIPAA compliant? So what exactly are the tools, and how would that work in your daily clinical life?

Maria Iliakova: Sure. So there aren’t any current tools that we use for that, there are some in development. In fact, in our own practice we’re developing a custom GPT that actually faces patients. So it’s not something that we use for our own clinical decision making, though that’s absolutely a possibility.

What we did is, we created a GPT. And what that means, for those who are uninitiated like I was before this, is you actually create a subset of, for instance, ChatGPT or some other model, and instead of feeding it all of the information that’s available on the internet, you feed in only the things that you want to.

So for instance, we put in our program materials, we put in videos and guidelines and papers and all kinds of sources that we wanted to, from vetted institutions and vetted professional organizations. And that created a baseline upon which then we started to train our actual model.

And what that means is, we started asking questions like, hey, can I eat an avocado three months after a bypass? Or other questions that patients would ask. Can I go running? How much do I need to be drinking? What kind of vitamins do I need to be taking? What about my beta blocker after surgery? Things like that. Based on our own program materials and other vetted materials, we were actually able to create an assistant that helps patients walk through their own decision-making and their own questions as they go.

We’ve tested that out on some patients and ourselves as well. I would say some of the most important things though to really consider when you’re creating those kinds of tools is boundaries. We established very clearly, if someone is asking beyond certain kinds of parameters, that it recommends that the person go directly to their health care provider, and that it not hallucinate and pretend to know information it doesn’t. Because that’s definitely been a feature of some of these AI models in general, and just to sort of play it safe, for it not to invent information or not to provide wrong information.

So that’s really how we’ve created a custom product essentially that’s validated, that’s accurate, and that’s 24/7 accessible to our patient population.

Kevin Pho: And how good is that information that it’s giving to patients? You mentioned the possibility of hallucinations in general. Are you happy with the output in general so far?

Maria Iliakova: We are. We haven’t really seen any hallucinations. When it goes beyond a boundary where it doesn’t have a direct ability to create an answer, it says, I don’t know, please refer to your health care provider at this contact information. Which definitely we prefer to it making up something entirely.

When you’re talking about health care there’s definitely a higher level of accuracy that has to be met. Giving incorrect information is a huge liability. And so there is a disclaimer there as well, that this is a product in testing, this is something that we’re developing but still working on.

Granted, you can give incorrect information or make the wrong assessment even as a really great qualified health care provider, but the bar is really high when we’re using artificial intelligence in health care, maybe more so than in other industries. So you’re absolutely right on the money there.

Kevin Pho: So let’s talk about some of the cautions with artificial intelligence, especially publicly facing ones like ChatGPT and other models that are out there. So there is of course a concern about the information that you’re feeding it, that it’s learning from the information that you’re feeding it. There are patient privacy concerns, of course. So talk more about those.

Maria Iliakova: Absolutely, yeah. So think of it more as a tool than as a destination. So anything that artificial intelligence accomplishes is at bigger scale than people can as individuals, and on a level where it can just be done bigger, better, faster, if you will, than a human being could alone. If you think of it as an adjunct to human activity, where ultimately a human being is still involved in the final validation and final decision making, that’s probably the safest place currently in health care.

In terms of feeding in information, it’s a little bit of a garbage in, garbage out effect too. So if you don’t have really high quality data, or you’re using a combination of really high quality data and information as well as some mediocre, then the output will not necessarily be very high quality either.

So it’s really important to understand what is the source of the information you’re using. And that can be very difficult to identify if you’re using something that’s just a general large language model based software like OpenAI’s ChatGPT. If you’re creating your own custom one, it’s a lot easier to do that because you know what the sources are. And actually OpenAI and a bunch of these other companies do offer services directly to companies that develop products that face clinicians and that face patients.

But it’s really important to understand what sources are going in, building out parameters for safety, and absolutely taking into account things like privacy, HIPAA, and all the other standards that apply in health care that may not in other industries.

Kevin Pho: So I know the pace of innovation when it comes to AI has been just so fast, every week there’s almost a new announcement. I think we’re speaking on January 12th today, and ChatGPT has a store that comes out where we can look at other people’s custom GPTs. What do you envision the immediate future to bring in terms of the roadmap with that intersection between AI and health care?

Maria Iliakova: Yeah, I honestly think it’s going to be a bit of a democratization of health care, meaning it’s going to become a little bit more accessible for the average clinician to leverage some of these really high-tech tools.

To build an OpenAI custom ChatGPT is essentially free, you can acquire it through a professional subscription. There’s lots of other services. It’s a really cheap way to create tools that can alleviate the burden on your own practice, whether it’s through the administrative burden or financial burden, and to create some really excellent tools for yourself and your patients.

A lot of these tools also are in the process of improving. So where we were even a few months ago with some of the professional or subscription versions of some of these services is leagues different than where we are today, especially if you look at some of the image generation software, or when you feed in your own documents and it analyzes them.

So I really do think that there’s going to be an enormous improvement in how we do things like recommendation, object recognition, prediction, all of the different kinds of things that are already being done in other industries but now in health care too.

The other thing that I wanted to mention is object recognition, because that I think is massive. I’m a surgeon, my whole job is recognition. And I think there’s a really interesting application in terms of doing video analysis of when surgeons operate, being able to analyze anatomy and motion during surgery in order to be able to train others, for instance, in performing surgery, and doing quality assessment of how well surgeons are performing the procedures they’re doing, and even in developing better techniques to do things.

So imagine overlays that are integrated with our surgical tools and being able to understand a patient’s anatomy even before you’re in someone’s abdomen and operating. Some other things to think about are how we use our tools. So right now we’re working on a project, for instance, in object recognition for surgical tools themselves, and that enables us to learn how to automatically, robotically manipulate the tools, which again can alleviate some of the burden in tracking and using the tools themselves.

Kevin Pho: Now, AI has disrupted so many fields, and you talked a little bit about this earlier about pattern recognition and doing some of the things that pathologists and radiologists can do. Do you see AI being a threat to any specialty among clinicians?

Maria Iliakova: To be honest with you, I think it’s going to be part of everyone’s industry. So I think there are some that are lower hanging fruit than others, like maybe the radiology, pathology like you’re talking about, or maybe dermatology, and in some aspects from the surgical realm. But to be honest with you, I think everyone will have to deal with AI as part of their job, or potentially even taking part of the job away, the things that are easier to automate than others. That sort of 80 to 90 percent that is routine in a practice probably applies to all clinicians in a way.

I will say I don’t necessarily think of this as a negative, because we are understaffed, especially in rural communities, especially in times of when there’s conflict or when there’s financial problems, we are grossly understaffed. So being able to use artificial intelligence to do some triage, to do any kind of diagnostics as a baseline, and then have that informed by a human being and sort of complete the process with a trained professional, is really where I see a lot of AI utilization in all industries, in health care, going.

Kevin Pho: We’re talking to Maria Iliakova. She’s a bariatric and general surgeon. Today we’re talking about “AI for health care professionals: How it is used now and how to use AI as a clinician.” Maria, as always, we’ll end with some of your take-home messages to the KevinMD audience.

Maria Iliakova: You bet. So I think one of the things I would say is, don’t fear AI. Easy to say, tough to do. But if you’re interested, go play with it, practice it. Open up a ChatGPT on your browser and try it out. Or even if you do a Google search, you’re probably noticing some of that generative AI as a result of your search. Look at what it’s creating, see if it’s accurate. Take a look, and think about what kinds of applications you could have in your own practice, or what kinds of ways it’s already being used. That’s one.

And then two, I think the fear of AI taking our jobs, or really impacting our jobs, instead maybe encouraging folks to think about, how do we want to be involved as clinicians, how do we want to be involved as health providers, in creating tools that we’re willing to use and that are wonderful and great, rather than trying to stem a tide that I think is otherwise inevitable.

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

Maria Iliakova: Thank you so much, Kevin. It was a pleasure.