This Week Health

Don't forget to subscribe!

April 16: Today on TownHall, Brett Oliver, Family Physician and Chief Medical Information Officer at Baptist Health talks with Hamed Abbaszadegan, MD, Physician Executive at Stanson Health. We reconnect with our most popular guest from last year as he navigates us through the complexities of AI and machine learning, natural language processing, and the advancement of large language models. What does AI actually look like in the healthcare industry currently? How are health systems integrating it into their processes and where are they being hesitant? Is the cloud an essential part of this technological evolution and what does it mean for AI in healthcare this year? Could we be stepping into an era of 'cautionary evaluation'? Listen in to get answers to these thought-provoking questions and much more

Subscribe: This Week Health

Twitter: This Week Health

LinkedIn: Week Health

Donate: Alex’s Lemonade Stand: Foundation for Childhood Cancer

Transcript

This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

Today on Town Hall

we're really going to get into this component of time, where you can't really do things without the cloud.

And to make a reference to short circuit, the movie from the 80s where he would read the books very quickly, . It's the same concept. All the books of the universe are going to be in the cloud.

So if I can compute that quick, boy, are we going to make some new crazy conclusions about healthcare.

My name is Bill Russell. I'm a former CIO for a 16 hospital system and creator of This Week Health.

Where we are dedicated to transforming healthcare, one connection at a time. Our town hall show is designed to bring insights from practitioners and leaders. on the front lines of healthcare. Today's episode is sponsored by ARMIS, First Health Advisory, Meditech, Optimum Health IT, and uPerform. Alright, let's jump right into today's episode.

Welcome everyone again to an episode of Town Hall. I'm Brett Oliver, a CMIO for Baptist Health in Kentucky and Indiana. And I'm excited to have back for a second time Hamed Abizadeh again. Hamed was our most downloaded Town Hall episode last year. So I thought, I think I'm obligated to bring him back for a second go around.

So I want to welcome you once again. And if you'd remind folks of your background and sort of your present role with Premier.

Absolutely. Thanks for having me again. First off, the award was like super unexpected and it was probably the neatest thing. recognition or award I had all year of last year. So, definitely an honor to be back for an Encore presentation is great.

So my official title is Physician Executive with Stansen Health, which is a subsidiary of Premier Inc. and support Premier and Stansen in a number of different ways from product development to customer success and then, of course, backing up as we showcase our products for health systems.

And you were just mentioning to me before we got started, you're still practicing.

You were pulling your shift this weekend. What's your medical background?

Yes. Sorry, I forgot to mention. Internal medicine is my primary training. Pretty much most of what I do is hospitalist medicine about one weekend a month. This is something that I really emphasized when I first started that I really got to maintain my clinical skills and it brings a lot more value when you're having these Conversations about software tools for healthcare settings.

And also I have a background in clinical informatics and essentially my career has basically been clinical informatics. Glad to have you. Thanks again for being here. Last year we did talk AI. And, but so much has changed since then. I feel like it was a decade ago, and you're having discussions with non technical people all the time clinicians that, want to use different tools.

How do you like to differentiate maybe the terms AI and large language model and GPT, and is there an importance in doing so? And then maybe if there's a product that you would be examples of each that you run across.

Sure, definitely. It's important to understand the differences, and I think this year we should definitely take some time to educate the audience on understanding some key.

Differences. I think before we begin it's really important to understand three key technical advancements that have really maintained our push towards AI, and that just continues to be the volume of data, especially with devices, recording systems, sensors. It's estimated, and I unfortunately don't have a reference for this.

I've tried to look it up, but It's estimated that the volume of data that we have in the last four, five years is more than all the data we've had in all of human history up until this point. And that seems really impressive, but if you think about it, especially as our devices are getting smarter with sensors, it makes a lot of sense to me.

So volume of data, access to that data, and then the ability to compute ultra fast, especially with on demand computing, those computer resources that are made available as needed. So keep these in the back of your mind. But in terms of definitions, I want to really set the tone for how I think of AI. So if you're going to talk about a full bore AI, it really is broken down to me in four components.

So the data pattern analysis, which is really machine learning, The natural language processing, such as your Siri or Google Home that kind of just text component. And then the final two differentiators is planning and reasoning. These are the final two components that we'll often see a lot of machine learning and natural language processing, but we don't always see reasoning and planning.

So machine learning, we've kind of discussed this, especially last year, that pattern recognition of data and the natural language processing I feel everyone has a lot of familiarity with. But the planning is really kind of making sense of things and saying that this is what will happen if you do this or don't do that.

And last year we talked a lot about the map application, but kind of that planning and estimation of time, knowing what will happen, And that reasoning about that is really where those decisions come in. So if you're driving and suddenly the artificial intelligence is making certain reasoning about what will happen, then it's going to plan a new route for you and it might audibly speak to you.

So IBM defines AI more of any system capable of stimulating human intelligence and thought process. I think that's very good definition, but to me it doesn't really, it doesn't really expand what it means in the full context. So the way that I look at it is, AI to me is data pattern estimates and predictions of what can or could happen.

that are stated with some sort of guidance in this interface that's natural. So again, if the map while you're driving is actively talking to you and rerouting and planning a new path, that's a good example of that reasoning and planning. So you really got to incorporate those two in terms of reasoning and planning into your definitions.

And just to round out some of the latest trends with large language models, really, those are LLMs, CHAT GPT. The GPT is a generative pre trained transformer and the large language model is really notable for the ability to achieve that general purpose language generation and understanding. So, in terms of examples, we kind of mentioned the AI side, but the large language models where we're seeing it in use is ability to generate text, which we know, but generate notes.

And we saw at the EPIC conference this last year that they demonstrated some of this dictation and DC summaries pre prepared. So if you go back and look in the chart and kind of generate text summarizing what happened to a patient in a hospital, that would be an example of that. Basically the large language model and kind of GPT kind of bringing that, that in together.

So, as you mentioned with some of my weekend clinical work as a hospitalist, I have residents, so I kind of chuckled at that. It's basically having your own residents on tap. To do some of your notes for you this will kind of pick that up.

Very fair. Well, one of questions I was most anxious to ask you today, because of the breadth of health systems that you're exposed to with your job with Premier or with Stanson where do you see, in terms of AI, where do you see health systems being aggressive?

Maybe that's not the right word, but I'm going to use the term aggressive. And then where do you see hesitancy? With AI right now. Sure. And we use the broad term AI. You define it as you want to see fit.

Yeah. Again, I think the differentiator is there's lots of machine learning going on. There's lots of NLP that is really starting to break out.

But the reasoning and planning are the big differentiators. And that's where we're not necessarily seeing as much as you might expect, but really a lot of these components of AI are honing in on cancer detection and navigation. So we'll use that word kind of broadly. I don't think we're really mainstream with a lot of imaging components with cancer detection.

There are pockets of that, but I just, it's not mainstream yet. The navigation side and ability to. Pull the text and different components of what's in a patient's chart to aid and help. Certain chronic disease management, especially cancer navigation, we're starting to see that more and more. There's a lot of companies that do that.

But I would say the biggest aggressive push right now would be in the coding front. So leveraging the different elements of AI To be able to understand patient's conditions and ability to give the right code at the right time to the provider to select. So, inferring suspect conditions, validating conditions.

This is where there's been a lot of push. This is certainly where our company is doing a lot of aggressive development in that. And this is where the need is. And I think a lot of that is driven by really hard ROIs that are seen instantly. The payers are happy with that because they're getting the right code, and the providers are happy with that because they're getting the right code.

At the end of the day, you want the true clinical picture for patients. So, leveraging technology to help your clinical documentation improvement specialists, help your coders. Notice I'm not saying get rid of. This has been a huge push. I've seen this a lot in the last year. Thank you. And we still have switchboard operators in the hospital, so these tools are really being used to augment those decision makings, just as you use the map for you to augment decision making with your path.

Yeah, no, that's a great analogy with switchboard operator and AI. So as far as those systems that are being aggressive, do you think in that coding space, part of it is the ROI, 100 percent agree with you there when you're versus looking at, say, some radiology based AI that it makes them more efficient or they feel like they're more efficient, but it's really hard to put an ROI to it.

Do you also think, though, that in that coding space, it's a place where you can be more aggressive because there's not that risk to patient care per se? Yeah, It's more of a back office functionality. I'm just curious as to, or do you think it's just more mature in that space and that's why you're seeing more adoption?

I think that really, it's not a matter of risk because to be honest, if you get the wrong code, there's downstream implications. There's automated follow up labs and management and specialty consultation. So really, It, to me, in many regards, getting the wrong code has its own subset of problems. I think the maturity with what software can do, namely in the machine learning and natural language process components, as we've seen with the large language models and chat GPT, it's not only accepted, but it's really, I think it, the maturity with what we can do from the healthcare side is already there.

You can read the notes. It's all natural language processing. We can find the patterns and make those inferences based off how you develop the model and what certain clinical factors you look for. So, I'm not a data scientist, but what I would say is from a data scientist perspective, it's not as complex as all these deep neural networks and unsupervised learning.

We're not really in that. We're more at the basic aspect. Still not the full bore like reasoning and planning component, but again, scan the field, look, and prompt these appropriate codes based off of labs that are there, based off of meds that are there, based off of what the clinician wrote in the note, based off of the What the patient is stating, so reading the notes, one side, looking at current diagnoses, past diagnoses, structured fields like meds and labs, and just bringing that together, computating it, boom.

I think it's not too complicated to spit out the right code.

  📍 📍   In the ever evolving world of health IT, staying updated isn't just an option. It's essential. Welcome to This Week Health, your daily dose of news, podcasts, and expert commentary.

Designed specifically for healthcare professionals like yourself. Discover the future of health IT news with This Week Health. Our new news aggregation process brings you the most relevant, hand picked stories from the world of health IT. Curated by experts, summarized for clarity, and delivered directly to you.

No more sifting through irrelevant news, just pure, focused content to keep you informed and ahead. Don't be left behind. Start your day with insight at the intersection of technology and healthcare. This Week Health. Where information inspires innovation. 📍 Increase 

 Makes a lot of sense. Have you guys seen any difference? Have you looked at the difference between your coding tools, your AI assistants with coding, and those ambient technologies that are being used, whether it's DAX or something like that? Is it, have you seen a difference one way or the other?

Great point. So what I can tell you is I've seen a lot of difference with more stagnant embedded content versus active cloud use and leveraging these tech like machine learning and NLP and some extent of regioning planning. So there's definitely, it's more specific and it's smarter in when it's firing.

In terms of ambient technology, really that is being amplified, especially because you can leverage these tools with text and natural language, and then leverage the cloud with that massive amount of data. Just think about how much data even just this podcast is and this is just one. So if you had this constant.

flow of data and information, how are you going to process that? So it's becoming simpler to do because you can handle the volume of the data, but there hasn't been too big of growth in the ambient tech yet. Because it's expensive but a lot of, I've seen a lot more vendors rise in the ambient listening.

I mean, it makes sense, you just listen to me and then you jot off to the side, you generate, right, the GPT, you generate that language and text for the note ahead of time. But again, a lot of companies are leveraging, base models of open AI and other components. And then going from there. So, I think there will be more proliferation of that, especially as the costs come down.

And naturally, the costs will come down as more vendors are in that space. So,

Yeah, I'll be curious to see if, the ambient summary. better with coding or not as you get more experience. But so, whether it's in a back office functionality or a clinical functionality, when we're talking about AI, where do you see gaps or maybe the largest gaps in health systems, both their intake, their evaluation, and then the ongoing monitoring of AI?

Yeah, it's a good question. And I just realized I didn't really touch upon the hesitancy. But this is a good place to discuss it, where those gaps are. So, I would say that the hesitancy, which is associated with the gaps, is the full bore AI, okay? From what I've seen, machine learning is pretty much now commonplace pattern recognition to predict certain diseases and then get those images done for early lung cancer screening or get your diabetic patients in if you're predicting that they're going to have bad disease states.

But, the reasoning and planning and taking what the AI is telling you to do is another component of it that is not yet rooted for practical clinical medicine. We still want the clinician to make the decisions, and I don't think that's going away anytime soon. I think the augmentation factor will grow.

But what we're seeing in terms of the gaps is building these tools is getting more difficult, requires more technical skill. And my opinion is one of the hottest fields in the country is data science because everybody from the FHIR department to the hospital, to the bank, to Amazon. Everyone has some component and need of a data scientist to help make sense of this, to help hook their tech, combine it with large language models and other components of what we've seen to then make their product even better.

Whether it's for ambient technology to then generate the notes and then prompt the clinician. I heard you discussing colonoscopy wasn't done. Do you want to order a colonoscopy? So, connecting these dots and monitoring that internally and understanding how it works, I think we're a bit a ways away. In fact, I had a physician that contacted me just the other day that they had just this basic dashboard and one structured field is now not pulling.

And they found out it hasn't been pulling this data for three years. So my response was, you got to look on the back end, right? You've got to find out where maybe a patch derailed how that is read now. There's something there on the back end that you've got to go through identify. So, simple enough for dashboards that were created by the hospital.

But what if you have a glitch in your system of ambient tech or other, component of full bore AI? You're going to need some massive technical expertise for that. And they're pretty much getting doctor salaries. So it's hard for hospitals to find these people and keep them on staff.

Oh, great point.

Great point. It's one thing if it's a dashboard, it's another when you're delivering clinical care itself. Yeah. Let me shift gears a little bit to how the adoption of AI algorithms by health systems will actually function from a technical perspective. I'd like to understand your thinking about how health systems should be thinking about leveraging the cloud as they think about AI options and are there cost implications with that?

Yes. Great question. I think we're basically at a position in society we can't survive without the cloud. So we got to a position with the internet now that we basically can't survive without it. But let's just, before we get into healthcare, let me give you a quick story of Spotify, which is a music app.

So Apple Music, Spotify. These are different types of music applications that are subscription based that basically access the songs you want for the cloud. And Apple's own advertisements back in 2001, they were the horn that, okay, You can have 1000 songs on an iPod in your pocket. I don't really have an iPod.

I don't think anybody does anymore. But fast forward 20 years in late 2022, they made the announcement and over the years, it had been progressing. But now you can access 100 million songs on your wrist, which means if you have an Apple Watch, you just press some buttons and you can access it. The story I had was my father was in the car with me and I was showcasing Spotify.

And we have an Iranian Persian ancestrally background, so I told my dad, name any song off the top of his mind. And he mentioned a song from the 1980s, okay, that was from Iranian background, and those of you that have the same heritage probably even know what the song is. Anyway, I just searched for it with the name of it, phonetically, the best I could, and we found it.

1980s song was found at ease. So that song is stored and lives in the cloud. I certainly hadn't downloaded it. I didn't have a tape or CD or any of that. So that's really the power of the cloud. And this last year, I went to the Cerner Annual Conference and the head of Oracle, which bought Cerner was talking about their cloud solutions and the lightbulb went off in my mind at that moment that now it makes sense why Oracle bought Cerner.

This is my opinion, of course. But they now have a lot of access to healthcare data, and they can make sense of it with their on demand computing, cloud storage, and software that can start to make better predictions and full bore AI with planning and reasoning, even with that health information.

So we're really going to get into this component of time, which I actually think we are in some regards. where you can't really do things without the cloud. Now a lot of hospitals have the different EHR vendors and they have their software and storage of things within the cloud, but it's by storing everything in one area where you can handle all this volume, you can rapidly computate that.

And to make a reference to No. Johnny 5, the short circuit, the movie from the 80s where he would read the books very quickly, and suddenly have knowledge and what to do about circumstances. It's the same concept. All the books of the universe are going to be in the cloud. To some extent they are. We just mentioned we have a hundred million songs there.

So if I can compute that quick, boy, are we going to make some new crazy conclusions about healthcare.

Right, and not just from your own organization, right? From potentially multiple organizations and millions of patients like the one you're seeing in front of you. time.

It really feels like it. Well, this is like Before

you mention cost, I'm sorry. So the cost actually with cloud I feel is kind of getting lower and lower. It doesn't cost a huge amount to store. And then the subscriptions you pay for gets you that access. Yes. So, I think the cost will actually be not so bad to be able to tap into that.

Tap into that, yeah, that makes sense. Well, I have one last question for you. It may not be fair given the speed with which what we've been talking about changes, right? But what predictions would you make for the rest of 2024 in regards to where we'll be with AI and healthcare by the end of the year?

Yeah, really good question. And my answer might surprise you. I actually feel that from the towards the end of 23 until now, we've taken a couple steps backwards, and I'll mention why. I think we're really advanced in awareness. And people's like understanding of AI has really come a long way. So what is AI, how does it work?

How does it fit for me? I think we're going to continue to advance more. People will get on the bandwagon of subscriptions or using data, especially your own data from monitoring your heart rate in the gym to maximize your performance. I mean, we've seen this really explode in the sporting world. But I think there's also, we're in this kind of.

time of caution and ease in terms of double checking what is this even going to do. And so, largest provider of healthcare, largest health system in the country is the VA. And they've really been very cautious towards the end of last year, identifying what the different hospitals and facilities are doing, and they kind of want to shape that.

So I think we're in this mold phase right now. Okay, what is this going to look like for healthcare? 24, I think, is going to be a big year. We might not have the latest gizmos and tools , that, do crazy things that we might think. I'm not so sure if that's going to take place in 24, but some of these steps backwards that I've seen, and I'll get into automotive examples, so one is Tesla.

If you or someone you know has a Tesla, you probably have realized that autopilot, I mean, again, my opinion really has been ruined by this point. Because the number of accidents that were associated, this was on the news just a couple months ago, was very high, they've had to reel back and make the software watch your eyes more, give you more warnings, and deactivate.

And I was one of those people. I had five deactivations within whatever number of time, and my autopilot has been completely shut off. for a week. So basically, Tesla has slapped my hand and said, you're not allowed to use this for a week, which has, driven me nuts, but maybe it's a good thing because I was not being safe, admittingly.

But even the Google owned or Alphabet owned Waymo, they had a first recall of their software because two of their robo taxis hit the same tow truck just a few minutes apart. So some of these glitches and lack of refinement. are starting to be identified, and I think that has caused a huge surge of caution and a huge surge of, let's look into what this is.

So recognition is going to skyrocket, use in some areas skyrocket, caution implementation, way more involvement of human input, human evaluation, and maybe to some extent even governmental assessment and evaluation.

Yeah, your Tesla example is To me, it's not fair. We're setting the bar for some of these AI algorithms to be 100%, and yet, of those times that you were deactivated from your self drive function, you didn't have an accident.

Look at the accident rate of a Tesla versus another car, a standard car, and they're not comparable, but yet, Because it's a new technology, it gets put in the spotlight, or maybe they just don't like Elon, I don't know, but I think you're right. We have to sort of weigh out those, the pros and cons.

I've had the same kinds of discussion with some AI applications in healthcare where they're like, oh, but it missed this one. It caught 99, but it missed this one. Like, what did the human do? What's, where's the bar we're starting from, and so those are the discussions. I'm with you that I think this is a year that we're having a lot of those and figuring out, where that threshold is.

To AIR is Human, the famous book to me that basically created the subset of quality improvement in healthcare as we know it, yeah. I mean, we don't need to get into it, but I'm sure we've all seen some errors and things happen in healthcare. And then like, what is that? I mean, I would presume the human is more than the machine, but again, I think we're in a year of caution, cautionary evaluation is what I just literally thought of on the spot.

I like it. We'll go with that and we'll stop there. Hey, Hamid, thank you so much for taking your time out to be with us again. I appreciate your insights and I know the listeners will as well.

Always appreciate it. Thank you very

much. All right.

Thanks for listening to this week's Town Hall. A big thanks to our hosts and content creators. We really couldn't do it without them. We hope that you're going to share this podcast with a peer or a friend. It's a great chance to discuss and even establish a mentoring relationship along the way.

One way you can support the show is to subscribe and leave us a rating. That would be really appreciated. And a big thanks to our partners, Armis, First Health Advisory, Meditech, Optimum Health IT, and uPerform. Check them out at thisweekhealth. com slash partners. Thanks for listening. That's all for now..

Thank You to Our Show Sponsors

Our Shows

Newsday - This Week Health
Keynote - This Week Health2 Minute Drill Drex DeFord This Week Health
Solution Showcase This Week HealthToday in Health IT - This Week Health

Related Content

1 2 3 258
Transform Healthcare - One Connection at a Time

© Copyright 2024 Health Lyrics All rights reserved