Selected Podcast

Artificial Intelligence in Medicine at Carle Health

In this episode, Dr. David Lovinger and Taylor Craig lead a discussion focusing on the current trends of AI in healthcare.

Artificial Intelligence in Medicine at Carle Health
Featuring:
Taylor Craig, BS | David Lovinger, MD, FMH, FACP

Taylor Craig, BS is the Director of Digital Engagement.


David Lovinger, MD, FMH, FACP is the Associate Chief Medical Officer, Health Informatics.

Transcription:

Dr Rania Habib (Host): Artificial intelligence is taking the world by storm, and it is expected to be a $305 billion industry by the end of 2024. Seventy-seven percent of all electronic devices now use some form of AI. But what is AI and how does it integrate into the medical field? Find out all about AI and Medicine on this episode of Expert Insights.


This is Expert Insights with the Carle Foundation Hospital. I'm your host, Dr. Rania Habib. Joining me today is Dr. David Lovinger, the Associate Chief Medical Officer for Health Informatics; and Mr. Taylor Craig, the Director of Digital Solutions. And they are here to discuss artificial intelligence in Medicine at Carle Health.


 Welcome, Dr. Lovinger and Mr. Craig, and thank you for joining me today on the podcast to discuss this very timely topic.


Taylor Craig: Glad to be here.


Dr. David Lovinger: Thank you. Glad to be here.


Host: Mr. Craig, let's start with you. Artificial intelligence, which is abbreviated AI, is probably the hottest topic in most industries right now. But what is AI?


Taylor Craig: I think AI is an umbrella term that gets thrown around a lot. I think AI meant something very specific probably a long time ago of trying to replicate human intelligence. And as the field has progressed, I think it's become more of an umbrella term that includes kind of subsets that we hear about often, like machine learning, natural language processing, computer vision.


What's happened the last two years, or two and a half years, however long ChatGPT's been out, is that in the realm of natural language processing, the advent of transformers in the model has really lowered the barrier for entry, being able to kind of converse in that natural language and ask AI to do things as you would a person has excited people and lowered the barrier for entry overall to the more complex components of AI like machine learning.


Host: Now, you said it's an umbrella term. So, obviously, AI might mean different things in different industries, but what are the current trends in AI right now, Mr. Craig?


Taylor Craig: In the healthcare world, the computer vision or overall image processing obviously has been around a really long time. That field continues to progress as does the overall data and analytics spectrum. What feels much newer is, again, going back to the GPT type models, those being able to sit over the top of the other applications, and not just limited to AI, but also automation. Carle here, I think we like to view automation in similar fashion to AI. They kind of go together hand in hand. You're able to kind of go from point A to point C, very quickly now by using ChatGPT-esque applications.


Host: Now, for those people who don't really understand ChatGPT, can you just give us kind of an overall understanding of how it was formed and how it works?


Taylor Craig: ChatGPT is a large language model. So, you can kind of think of it as something similar to how Google used to go and crawl itself around various web properties to return itself on search indexes. ChatGPT has consumed most of the internet, or in some way, shape, or form. And the windbreak moment, or what led to ChatGPT was again the creation of these transformer models, which was just a new approach in the realm of natural language processing and overall kind of how these systems are set up.


It's fair to say ChatGPT was one of the first that you could interact with. A normal person could go out and interact with it. It didn't have to be a developer that had it loaded on their machine, which is how most transformer models that existed the previous three to five years.


Host: So, ChatGPT essentially is consuming all this information in the internet. And then, when we're entering a question, it's using all that data to come up with an answer. Is that kind of the general understanding?


Taylor Craig: Yes.


Host: Perfect. Because it's confusing, especially to those in healthcare.


Taylor Craig: Well, and it's presented in a lot of different ways. But it's much more similar to, I think, the way that Google works than an actual intelligence kind of behind the scenes. It's very good at mimicking human interaction though, and getting better and better all the time.


Dr. David Lovinger: Large language models are essentially like the autocomplete in your phone, except, you know, many powers greater than that. So, you ask it a question, it gives an answer that is based on all of the information that is fed into it. But it also can do some strange things, because it doesn't really know anything underneath. It is, autocomplete on some level.


Host: Oh, it's a great way to help people understand, because I still think people have a hard time wrapping their minds around exactly what is ChatGPT and AI. It's very confusing for some folks. Dr. Lovinger, what is the importance of AI specifically in Medicine, and what are some of the most common ways to use AI in Medicine?


Dr. David Lovinger: So, I'd say there's sort of two separate work streams or ways to use AI. And the one that comes to mind, I think, immediately is we could get AI to diagnose patients or to come up with treatment plans. And while AI is really quite powerful, there are still a number of things that humans do better. And I'd say that there's a lot of research into diagnosis and getting AI to "replace the doctor". But, in fact, I don't think that's the most valuable way to use AI.


The other stream is to get AI to take care of sort of the more mundane tasks like answering messages, summarizing a chart, finding this, that, or the other type of information. And AI can do that already much faster and probably better than humans can. So, one way that we're going to be using AI is to draft an in-basket message response to a patient. So, patient calls in and says, "Hey, doc, I'm out of my thyroid medication. Can you refill it?" I would type back or many docs would type back, "Okay, no problem. We refilled it." AI can put that in a much better package. They can say, "Thanks for calling in. I'm really glad that you're asking for a refill, because I don't want you to run out. We're going to refill. We're going to give you five refills over the next six months. "And it will be called into your preferred pharmacy, which is in the chart, but as a human, I would have to go look for, AI can have this at its fingertips. And we know from research that those computer-drafted responses, people like them better. They're complete and nicer than what a human would do, which is kind of ironic.


Host: It is. It is. But I think it's that. You know, we're so overrun with charting these days, right? We all log into Epic and we have to do all that. So, taking some of those tasks away will probably really help with just overall happiness at the hospital, which is great.


Dr. David Lovinger: Absolutely. And if I could add one other way that we use AI, and I think this gets to the heart of the conundrum, so we know that AI can read radiology studies, you know, x-rays, CTs, MRIs, at about 92% of the accuracy of attending radiologists, which, when you think about it, is incredible, but also who would want their study read 92% as well as a human could do it? Actually, nobody would, right? So, the way we use it is every radiologist will have a queue of studies that they're reading one after the other. The AI will read all of those studies ahead of time. And if they find a medical emergency, like a stroke or a pulmonary embolism, they'll move it up in the queue, so that the radiologist reads that next. And I think that's the heart of the issue, is, you got to have the right workflow. Because if you think about it, 92% is not good enough, but we found a way to make it really valuable.


Host: No, it is because, like you said, the emergency should take precedence, and there's no way for the human to know that. So, that's a really great example of how AI is supplementing care at Carle Clinic.


Dr. Lovinger, could you tell us what Carle health departments currently AI and describe how they are using it? I know you already told us about Radiology, but we'd love to hear about the other departments.


Dr. David Lovinger: There's Radiology. There's the draft response thing that I was just talking about. We're going to start piloting that next month. So, that will go to Primary Care. We also use other things, which I would think are technically not AI, but a lot of people would lump it in, which is we use predictive models for all sorts of things. We can predict who is likely to be readmitted to the hospital, and so we can devote more resources to helping them after they're discharged. We can predict who is likely to have a clinical deterioration, meaning instability in their vital signs, need to go to the ICU. And we can alert the team if they don't know this already, that there is something going on. We help predict sepsis. There's a number of things like that that we can predict. Those are all not AI today, but in the future may be, as Taylor mentioned. ChatGPT is a large language model. Large language models can be the basis of predictive tools that are even better than what we use today. So, I'd expect more and more happening in that realm.


Host: That's super exciting, actually. And the fact that you're really using this to predict sort of how we can take care and notify the teams about the most sick patients or those that might bounce back is just a huge advance for medicine in general, because it'll take those costs away from the patient and the hospital and hopefully prevent, you know, detrimental events.


Dr. David Lovinger: I think most teams don't need to know who their sickest patients are. It's the tier below that when they are starting to become sick. And sometimes those warning signs can go unnoticed, and the models can help us notice them.


Host: Now, Mr. Craig, what are some of the exciting new advances and research involving AI? You know, we listened in the introduction that this is a massive industry. So, what do we look forward to?


Taylor Craig: I think it's a combination of look forward to and be wary of. The trend right now is really every organization out there is trying to infuse AI in every product that's existed from creative tools to development tools to obviously tools that exist more on the clinical side of things, accounting, legal. All of their tool sets are getting infused with various pieces of AI. I think it's exciting. Obviously, work with developers, their mode of working is changing on a weekly basis as they get used to using AI in their workflow. It's certainly been incredibly helpful from the perspective that they can rapidly test their code without sending or shipping it off to a separate tester. We can kind of get that initial test out of the way early, which our cyber security certainly appreciates.


But I think, obviously, it is going to be a gold rush. I mean, I think the gold rush is already happening. So, companies look to infuse AI into their work processes or into the tools that they sell. I think it's going to be really important for organizations to have a good handle and a good policy on how AI is going to be allowed or should proliferate throughout the organization.


Host: Absolutely. And I think, you know, one of the questions that a lot of people in healthcare are worried about is, "Will AI take over our jobs as healthcare providers, as IT support staff, as support staff in general?" Is that a fear that we have in the back of our minds?


Taylor Craig: It's a fair concern to have. But I think, as Dr. Lovinger kind of illustrated, there's ways to use AI that kind of enhance workflows and enhance overall satisfaction. I think everyone in this industry works very hard. I think the studies kind of would indicate that this is not an easy industry to work in. And right now, it feels like AI has got a long road of chewing up just some of that extra work that falls to everyone in a healthcare organization. So, I think we're going to see jobs evolve very much over the next decade or so.


Host: Well, hopefully, as you said, it just makes us more efficient rather than replaces us is the goal. Now, you have both provided us with such a wonderful wealth of information about AI and how it's used in medicine. As we wrap up, what is your final take-home message for our audience regarding AI in Medicine? Dr. Lovinger, let's start with you.


Dr. David Lovinger: First, I think people should be very excited about what AI can do, particularly taking care of some of the lower level functions of your particular role, passing you the tool like a nurse to a surgeon in the OR. I think, you know, those are going to be great, and I think that they're going to be routinely loved.


I also think, though, that there should be some skepticism about AI is going to be everywhere, and it doesn't belong everywhere. And I think that doctors, nurses, other clinical staff should trust their intuition and ask for what's the value of this AI tool? How is this going to help me? How is this going to help patients more importantly in delivering care? But I think overall, I'm very excited about the things that AI can do and help us do better.


Host: Absolutely. I think that's a great point to make, is really understanding how it can help us be more efficient. Same question to you, Mr. Craig.


Taylor Craig: We have some tenets that we devise for our AI governance, which is be smart, be safe, and start small. I think if you're an organization that's looking to testing what's out there in the realm of AI, I think those are all really important and, obviously, very general statements. But as things are moving quickly, I do think if you are looking to bring AI in, and I do think it's important because I think this isn't a passing moment, I think this is more akin to something like the internet kind of emerging, it's really going to change our lives in ways that we don't probably fully understand. But AI is not something to rush into as an organization. So, plan on starting very small, and being very diligent about the parameters that you kind of set up around.


Host: Well, thank you both so much for joining us today and for really giving us a wonderful overview into the exciting uses of artificial intelligence in medicine.


Dr. David Lovinger: Thank you.


Taylor Craig: Thank you. Appreciate you having us.


Host: Once again, that was Dr. David Lovinger, the Associate Chief Medical Officer for Health Informatics; and Mr. Taylor Craig, the Director of Digital Solutions at the Carle Foundation Hospital. For more information and to get connected with one of our providers, please visit carle.org or for a listing of Carle providers and to view Carle-sponsored educational activities, head on over to our website at carleconnect.com. I'm your host, Dr. Rania Habib, wishing you well. That wraps up this episode of Expert Insights with the Carle Foundation Hospital.