Hello everyone, and welcome to ACPM's introduction to AI and technology with our hosts from GOMO Health, Bob Gould and Jessica Anderson. To lead us in this conversation, I welcome Dr. Michelle Soltis, who is the actual chair of the Education Committee at ACPM and very involved with the AI fundamentals for physicians, both in ACPM's curriculum and in exploring new information, which Bob and Jessica will present to us today. With that, I give you Colonel Soltis. Thank you, Cara. Welcome to our esteemed presenters, ACPM members, session attendees, and other guests. My name is Michelle Soltis, and as Cara noted, I have the distinct pleasure of serving as the chair of the ACPM Education Committee and as the chair of the Innovations in Preventive Medicine track for this year's Preventive Medicine 2024 conference. Thank you for joining ACPM today for this live event, AI Fundamentals for Physicians. In today's rapidly evolving healthcare landscape, staying abreast of the latest technological advancements is absolutely paramount. Artificial intelligence has the potential to offer physicians powerful tools to improve diagnostics, care management, and patient outcomes, as well as to enhance the patient experience. This webinar today is designed to equip you with the foundational knowledge to understand this dynamic. ACPM is very pleased to partner with GOMO Health in bringing this webinar to you. Today's speakers are Bob Gold, founder and chief behavioral technologist at GOMO Health, and Jessica Anderson, GOMO Health's director of clinical performance improvement and organizational excellence. Bob Gold brings over 20 years of applied research and developmental experience in the behavioral health and cognitive sciences of human motivation, activation, and resiliency. A pioneer and innovator in the fusion of technology and healthcare delivery, Bob specializes in designing clinical care plans and digital therapeutics that integrate psychosocial factors and determinants into physical care plans for chronic and complex population health management. Jessica Anderson is a dedicated healthcare professional specializing in maternal health, clinical performance improvement, and organizational excellence. A nurse by training, Jessica has extensive experience at the intersection of clinical care and digital therapeutic programming. Despite the volume of material and our limited time today, today's speakers will aim to address any questions. We encourage you to pose any questions as they arise using Zoom's chat feature. We'll try to cover as many questions as possible in the time allotted. We'll provide you with access to the presentation slides and to additional resources post-webinar, as well as to responses to any questions we weren't able to get to. Kindly ensure your microphones are muted to minimize any disruptions. Should you encounter any technical issues, please communicate with ACPM staff using the chat feature. And as a quick shameless plug, I'd be remiss if I didn't mention the Innovations in Preventive Medicine track will host both a concurrent session and workshop focused on the application of artificial intelligence in the healthcare and public health spaces during Preventive Medicine 2024. We also look forward to your thoughts as to how we can continue to shape our educational content and continue to meet our ACPM members' needs. Once again, thank you for joining us today. Together, let's explore how artificial intelligence is reshaping the landscape of healthcare and empowering physicians to deliver exceptional care. And with that, I introduce you to Bob Gold and Jessica Anderson, who will present this informative content on AI fundamentals and clinical practice. Thank you, Michelle. You should be a DJ. You're awesome with that. Thanks everybody. So, you know, it's interesting when you talk about artificial intelligence. Let's just talk about humans and intelligence for a second. There are approximately 100 billion neurons in everyone's brain, give or take a few. I know plenty of people who don't use a lot, but there are about 100 billion neurons. And fascinatingly, there are 100 trillion synoptic connections in the brain, as much as the internet, right? And if you don't know this, children have 10 times the synoptic connections that are lost and become muscle memory as you get to an adult. So artificial intelligence has a long way to go. I just want to be clear before it could do what humans do. So we should all feel good about that. A couple of points and then we'll get to it. One is our goal is to demystify this and AI is out there. It's been used by Google and search for 10 years or so. Uber service uses it, FedEx and those types of things, logistics companies, Amazon uses it on shopping and other sorts of things. So it's really out there and we'll go through some examples of non-health care and health care as we go. And just want to point out that real quick, since health is 19% of our economy, on one hand, there's a lot of clinician burnout. On the other hand, we need better effective ways to help an aging, growing population who are facing more mental, physical, and social challenges than ever. So there are ways that technology and AI can be used very effectively. Okay, go ahead. Was that one of my chooses versus Jess versus next? So there's three modules. We're going to go quickly. Just want to be clear, this is a fundamentals class. So the idea is to give you kind of a big picture universe. We can have follow-on sessions that could go deeper into modeling and more specific. So we're going to cover these three lessons and the third one will be very interesting. It gets deeper into the clinician role in building an actual AI application. Go ahead. Yeah, so I'll jump in here. The overall objective of this session is to really give a foundational understanding of the universe of AI and specifically what can physicians do and have a role in creating those AI solutions. It's really important that we have physicians have a seat at the table. So we're really going to get into the nitty-gritty of what that looks like. Go ahead. Go ahead. Okay, so this is sort of a, you could click through so it has all those questions right there. So yeah, so this is kind of the universe of AI in general. There may be a couple other things, but this is what is in general. And today we're going to focus more on the first two, machine learning and natural language processing. We'll mention things around expert systems, but we're not going to get into more of the robotic side of how AI could be, is being used in robotics and surgery and those types of things. We're going to focus more on other types of uses that I think are in line with preventive medicine. Okay. So just to give you an idea of how things work, AI is a big umbrella term. So a subset of AI is machine learning, and we're going to get into that. And just because people are familiar with Chad GPT, that is a form of machine learning called generative AI, and we'll get into that a little bit. And you'll see how there are different techniques and approaches to how AI can solve various problems, especially given some of the administrative burnout required in today's day and age, where there's been a lot of technologies that have increased work for clinicians. I think we have an opportunity now to help them decrease a lot of the administrative tasks necessary. Go ahead. Okay. I'm not going to go deep on this, but basically machine learning tries to have different techniques that imitate how the brain functions and how neurons pass information to one another in the brain to reach conclusions, right? So deep learning has what they call these hidden computational layers that are set, and that's an area that clinicians could play into very well. So as an example, Chad GPT is a form of generative AI, and it's been trained just on hundreds of billions of words. That's how it does it. And we're going to get into how it's reinforced to learn in a minute, you know, how it gets better at what it does. The regenerative API is still an emerging field where if, for example, in Chad GPT, if you say, hey, what's a simple way to explain to a patient this disease, like what's a good way to explain, it will do that. Regenerative AI may come back to you and say, hey, doc, here's maybe a game or something else you should do rather than just have a simple way of explaining it, right? So regenerative API actually actively participates in the refinement of both the question and the answer. So we're not there yet. So we're going to focus on machine learning, generative AI, and we'll get a little bit into what a neural network or deep learning is. Go ahead. Okay, so just to describe machine learning, I'm going to use an example first of a non-health example, then we could use health. So let's say you're at a bank and you're trying to predict who's going to default on a mortgage, you know, who's more likely to default on a mortgage. So supervised AI is you're going to train the machine by giving it these thousand names and data who defaulted and these thousand profiles of people who didn't default. And then so you're supervising the clusters and model, and then it's going to come back and do its thing. And then as you add a person, if they default or not, it's going to, you're using reinforcement to reward it or tell it it made a mistake. But that's sort of supervised AI where you're creating the cluster yourself. Unsupervised AI is you just give it all your bank's loans and people, and it will form its own clusters. Oh, and it may form clusters you wouldn't even think of. People who live, who buy homes in rural areas or people who have this profile or people who are second generation Americans or whatever data you feed it. And then from those clusters, it would then analyze the next step of the default. So it would form, it would help you form its own clusters. That's sort of unsupervised. Reinforcement is, if you use chat GPT, reinforcement is, it uses a reward mechanism model where you tell it, oh, you got it right, you didn't get it right. The tone was off, not off. So it's, you're reinforcing the model. So if you have treatment decisions to make and AI is giving you data, what it thinks based on the choices, you could then, after the fact, feed it back and reinforce what the outcomes were, and then it would get better and better. So that's reinforcement learning from mistakes. Go ahead. Okay, so yeah, we talked about supervised learning. So you're training AI on models that you create. I think Jessica will just go over quick examples. Yeah, so a real world example that we have been used to, we probably didn't know that it was AI in the background, but is your email spam being auto sent to your junk folder. So the systems have been taught that if there's specific words in a subject line of an email that may be inappropriate or even certain emojis, it's all automatically gonna filter that and send it to your junk mailbox. How it can help healthcare, and we're seeing literature come out now where AI is being leveraged to look at x-rays and help diagnose. It can be applied to diagnose diseases based on those labeled medical images or patient records. Go ahead. Okay, so unsupervised learning, basically simple is you're gonna give it data and there's models that exist where it's gonna discover its own patterns and structures within unlabeled data. So you're feeding it to the machine and it's gonna come up with patterns that it may have taken a human years to do. And sometimes it doesn't require enormous, enormous amounts of data depending what you're trying to do. Something like a chat GPT did, but not everything needs all that. So Jess will give you an example now. Yeah, so Bob talked about the market analysis that we see when you're shopping on Amazon of the customers who bought this also bought this. You see this in YouTube uses this to feed you the next video that's aligned with whatever you've been watching. In terms of healthcare, so I think there's a unique opportunity here. It can help identify subgroups of patients based on similar characteristics or behaviors that as clinicians, you may not have time to identify those things. So for an example, if you have a patient population with diabetes, it may help you identify, okay, if this person is a mother who cooks for an entire family, it may be more challenging for that person to change the eating habits because they're not just cooking for themselves or cooking for multiple different food styles within their own house. So that's a unique way that it can help you identify those subpopulations. Okay, and then reinforcement learning is just that, that there's a human that's rewarding or telling the machine it made a mistake or it got something right. And those things are typically ranked and those rankings help it get better at what it does. So no different than how we learned in school. We're rewarded, we're told there's a mistake and hopefully we get better next time. Yeah, so for your real world examples, there are tools out there already in existence that help businesses with resource allocation. And in healthcare, while it's less commonly applied currently, that reinforced learning can be used for optimizing treatment plans or adaptive therapies. And so as this technology continues to grow, it'll start producing those suggested treatment plans. And if it gets it wrong, it gets a little boop on the nose and then it'll refocus and try again. Go ahead. All right, so I think we laid out basically kind of the universe of AI. We quickly went over a couple of the areas, right, on how it's used and how some of the techniques that help it simulate how the brain works. So we'll get more into that in the next module. Go ahead. Okay, so now we're gonna go through some concrete, more specific examples, right? And then you can go ahead. And then in module three, we're gonna show the role of the physician and actual the steps to create an AI application. So I think there's a lot of simple administrative things that AI could deal with and also act as treatment decision support. So it doesn't necessarily need to be patient facing, but it can really help the staff, all different roles within your organization cut down on a lot of time and effort. That's the idea in general. And by the way, if you're at a hospital, the EMRs use AI in certain things, your case management system, if you're running a third party clinical informatics system, it probably is using elements of AI within it to reach conclusions. Go ahead. Go ahead. Jess, you wanna take this? Yeah, so we're already seeing ways that AI can enhance efficiency by unburdening the clinical staff with some of those administration tasks. We see this where clinicians can have a phone while they're in with patients, getting that health history and the phone is taking the notes and dictating everything so that the physician after the visit can review what's written and then make edits instead of all the thousands of clicks that we have to do in the EMR system. It can also take these large volumes of data and compute those much faster than humans, which increases efficiency in healthcare operations. Next. Yeah, we'll get to some of the cybersecurity stuff. We're not gonna spend a lot of time. You'll see in a second. I saw a question pop up. So on treatment decision support, this is a great area to use because you can contain it, you could safeguard it and it could help clinicians and physicians with understanding things quicker. So if you have 15 minutes with a patient, you could maybe spend it on more impactful things, whether that's a tele-video health visit or in-person. And also tailoring preventive strategies. So to give an example is in, if you're recommending as, after you go through the ACLM course, you prevent a medicine, let's say a plant or better diet to folks, there are ways to train the machine that based on the profile, Betty Jo is just not gonna, not likely to follow certain things. So you could recommend it all you want, but I would recommend plan B, AI could help you because it may not be as good, but it's better than what she's doing and she's more likely to follow it. Like those are very interesting things AI could discover in people that it's just hard for a human to do that. Go ahead. Yeah. Yeah. So with a lot of these little subcategories that we're talking about, there's this huge component of data analysis and anyone that was around when paper charts were still in play and we were starting to switch over to EMR systems, there was a lot of conversation about the benefits of having access to all of this data. Every time you make a click, it's recorded. But at the end of the day, we don't have the time, most physicians don't have the time to look at that high level data. And so with treatment decision support, you can use AI to do that kind of grunt work for you and look at the data and then you're using your clinical best judgment on how to use it. And we're gonna, there's a slide later on where we're talking about what AI can do and what it can't do. It can never replace a physician and there's a lot of reasons why. One quick thing, like one way we use it is interesting. So we have digital therapeutics for cancer care, so it's complex, but by training our model, we figured out that if a person has so many social or behavioral issues with certain physical issues, there's an 80% chance they're gonna get fatigued, come off the therapy or end up in the emergency department or harm themselves. When we discover those things, we escalate to nurse triage or social work or pharmacy so they can intervene. And that's something that wouldn't have been discovered in the normal process. So it's helped save a lot of lives. Go ahead. Yeah, so I just covered one example of that. And then as home health moves to the home and the home may be the next mini clinic with intelligent appliances, devices, all types of things. I think what needs to happen is the devices like wearables and others need to coalesce with the behavioral and other characteristics. So is your blood pressure high because you have a lot of stress or because your blood pressure's high, you have more, like what's the cause and effect? So I think by using AI to take in various amounts of data and help determine things are very interesting. Whether you expose that to the patient or not, you can start by just using it for the clinical team. And then later, if you're comfortable, expose it to the patients. So there are ways of integrating both patient reported data and metabolic data to do that. Go ahead. Yeah, so the risk assessment and risk stratification is fascinating. You know, it's interesting. So we have some programs around substance use and preventive mechanisms in Montana, as an example. And we have six Native American Indian tribes that participate. And after seeing the data and some of the modeling and talking to some of the chiefs of the tribes and the shaman, we've altered the way we engage. We've altered our care plans for the folks who live on the reservation and who live off and have certain characteristics. So through the different learning, it actually helped us stratify the engagement, the protocol, and the risk levels of doing that that would have been very difficult for just a human to look at all that, right? And then you see genetic here, that's a growing field. As we know, in the whole area of genetics and epigenetics and those things, there's some fascinating uses that could help people in various stages. Go ahead. All right, so clinical informatics and data analysis, you know, definitely that's sort of an easy one for AI. You know, as Jessica mentioned a few examples, but I think AI could really help. And some of the deep learning, machine learning techniques that you're gonna see when, how does a physician really help a machine learn? You're gonna see where a physician can come into play. It could be pretty simple to really help it weight the computations and what they call the biases. And I don't even mean that from a negative ethics perspective, but physicians have a lot of experience based on real world data and what's going on in their own mind where they may wanna weight a bias on a computation. Go ahead. So here's basically, you know, on the left are things AI is good at, on the right are ways physicians could really engage in the formation, not just be the end user of what AI has to say, but really help build the AI algorithms. And we're gonna go through an example of how you do that in a second, especially like interpreting nuanced information, all the things that in the hundred billion neurons and an experience that humans have understanding other humans, ethical and moral decision-making, all those types of things is where physicians really need to step in. Cause if you don't step in, it's gonna happen and you're not gonna have as much of a voice anyway. Go ahead. So really, I think we went through, you know, quickly, we went through different examples of different uses of AI from treatment decision to personalized care, to overall population health management. And now I think we're gonna get into the process of creating an AI application and the role of the physician in there. Go ahead. Okay, so we're going to first talk a little bit about things you could do to safeguard and that has to do with cybersecurity per se and ethics and those things. And then let's get into the role of the physician very specifically and who do you need to develop an AI application on your own? Who else do you need to do that with? So let's go to the next slide. Okay, so this is where the humans need to inject how you use AI. So there has to be involvement in terms of your info security and how it's used in terms of ethics bias and discrimination in the models, the computational models, right? So there are ways of getting various inputs into that that could de-risk some of that. How does that get back into your EMR? How does that really get back into where it can help a physician in the moment make a decision? And the other limitation is too much reliance on automation and what any machine would tell you to do, right? Versus what you think as a person, go ahead. So here are some ways you can safeguard uses of AI. One is the data and content universe. So for example, we have certain models that we define the universe of content. So let's say we're working with a hospital. It's only within the hospital data structure, the content that they've all bought into, that's the universe. It's not going out on the internet. It's not searching for everything in the world, right? So you could actually define what the data and content universe is. And that's a good way to help manage errant decisions that a machine may make. The second is you can test things and run it for a while as clinical support versus direct, let's say patient interactions. And when you're a lot more comfortable, you can let the machine for a limited number of things engage with the patients directly on certain topics, right? And then the other safeguard is the reinforcement learning of the machine. You definitely need to have checks and balances. So you're looking at what the machine is saying and you're having models that just tell it it's making mistakes or it's doing the right thing. So it keeps on getting better and that's no different than any other human being at all. So those are some safeguards you could put in place. Go ahead. Okay, I love this slide, by the way. I think this is really neat. So if you wanna create AI applications within your practice, hospital organization, there are some simpler ways you can do that. It doesn't have to be that complex, but let's just quickly go through this. So, and where the physician role could play. So if you look at the strategy, the physician role needs to say, okay, these are desired clinical outcomes. You know, this is what we want. Why are people not completing cardiac rehabilitation? Why are they not? Why are they abandoning a better diet and just using GLP medications like, you know, to lose weight? So the physician could define the objectives and the measurements that they're looking for. And then you move into the data. Okay, now what data are we gonna use? Is it just our own, our 30,000 patient data over the last five years? Is it other data? Are we gonna suck in data? How, you know, what is the data we're using there? And Jess, any thoughts on? Yeah, so there's a couple of questions that have come in about cybersecurity risk and the liability aspects of this. And I think in general, those things are still being discussed at large, both in the US and worldwide, right? This is a newer technology. And whenever you have new technology, you're essentially in a new frontier and laws need to be written as we kind of explore this new area, right? One of the things I wanna say is one way to think about this is as a physician, you have multiple sources of data input coming at you, you have your patient giving you their health history. You also have your nursing staff, whether you're in a clinic or a hospital setting, giving you their opinions on what they're seeing, right? And just because a nurse gives you an opinion and says, oh, I think it's this that's happening, you're taking that. But as a physician, it's ultimately your decision, whether to accept that input or say like, okay, thanks for your opinion, see you later, right? And we do the same thing with AI. Just because AI is available and it's suggesting a potential diagnosis, you're using your best clinical judgment to determine if you're gonna utilize that or not. So I think we're seeing where, and there's a comment in there about how wrong AI is getting diagnosis even with the best input. And I think we're taking AI at this point with a grain of salt. There's a lot that AI can do. There's a lot that AI cannot do yet. So hopefully that kind of makes sense with that example. Yeah, and look, it's like anything else. It's, you know, people learn by what they're trained and their set of experiences, so does a machine. And it's not the be all and end all, that's how we started. But just to keep on going, so after you define kind of the objectives and the data, then you're typically involved the data scientists who's used to creating a model, right? On how to look at things and there's different models to train the machine. And you as the physician can help understand those models and participate, especially around biases, weighting of the computational aspects of that, you know, then it moves past the data scientists where you could start benchmarking things, right? So then you could test the clinical outputs versus best practices, like, okay, we created a model, but are we wrong 80% of the time to the point of one question, right? Is it getting it wrong 80%? So there are ways you could do things to Jessica's point to just use them internally. It's like anything else. It's like teaching a new resident, like, you know, so, and then it keeps on going on and on. So I'm not gonna cover the whole thing, but the idea is that you as a physician could really participate in how the machine learns, what it learns from, what to look for, what are the outcomes you want? And it's a good exercise to just get involved in trying that on your own versus just being the end user of what other people have decided to put into an AI application. Go ahead. So I think in summary, yeah, there's a lot that deeper questions and things you can get into in terms of data and how that works and the efficacy involved in that. There are ethical considerations and, you know, who do you get input in how a machine learns from? There are institutional organizational considerations organizational cybersecurity things that you have to look for, but ultimately how it gets integrated into a workflow, a clinical workflow could be the role of the physician. And there are plenty of ways to safeguard it and start it in a small way. Go ahead. So I think basically AI is a tool AI is at the infancy, even though it's been used for 10 or 15 years and there's a lot humans are doing, but humans could train machines to do certain things that are useful to them and not have them do things that are not useful to them. So that's up to all of us as individuals or what organizations we're with. And also to understand what are the limitations of it, but regardless of some of the limitations, it shouldn't stop everyone from learning how to do this and getting better at it, because it can be very helpful if used properly and safeguarded. Go ahead. So that's it, I think for today, I know it was a quick session, but we're gonna now go through some questions if there are any in the chat. And also I believe at the end, you're gonna get a very short assessment survey to fill out. So thank you for participating today. Thank you so much, Bob and Jessica. I do have a few questions if I don't wanna waste any time. There is one that's specific, but Jessica has assured me that it's very good, solid question. And it reads, we're currently trying to immediately house emergency department patients experiencing homelessness. We have 36 beds at present that patients can go to in the community. We are scaling operations to reach 108 beds by the end of the year. Can we use AI to interview patients to find out exactly what services and level of care they need? These patients are very complex and we are not clearly meeting their needs at present. Yeah, so I think there's several ways. The answer is yes, okay? So basically a short answer is you create a model where you say, I believe these 20 questions and getting answers to these 20 questions from each person is important, right? Based on their social, mental, physical status, and then you create what you call some weighting to say, how do we weight that based on what their need would be? And then what is your facility's ability to handle? And then how do we risk stratify those folks to handle the certain cases versus others based on your current ability? And then third, you could also say, how do we partner with shelters or WISE or other organizations in your area to refer people to? You could definitely add that into your model of, hey, there are other alternatives besides our facility that could maybe help these people. So you could build an application that says, for these five, we're gonna refer them to ABC Housing Community Service. We're gonna handle these five in our facility because we're better equipped and so forth, yes. So that's how you would do that. Thank you, Bob. And one of the major questions I think that everyone should be aware of is that this presentation will be reposted as it's being recorded right now and featured on the ACPM Learning Management System. Some people are asking if they can use these slides again and repurpose them. Sure. Okay, well, that's another major question. I am, okay, here's one. While your AI machine is interacting with the patient, are you and your patient giving a hacker access to your electronic medical records or home file system? Yeah, that's a very specific implementation question for an organization. Typically, you're housing these things in the same controlled environment as your EMR or your clinical informatics or those things, right? So you're not typically housing the data set outside of the normal cybersecurity that an organization has, right? Because remember, AI is about, it doesn't matter per se where the data exists. You're building models and algorithms to exercise the data to reach a conclusion, right? So you don't have to necessarily move the data anywhere, but you just have to expose the algorithms to the data, which could be within your own infrastructure, info security infrastructure within your organization, right? The other interesting thing, I mentioned nutrition since this is preventive medicine. So let's say another interesting one is movement. So a lot of preventiveness is what happens in the brain, nutrition and movement or exercise. So one of the interesting things is, okay, we have 50 different types of movements from more strenuous to less. What is Tommy or Betty Jo more likely to do? So you could take all your patients, put them through this learning machine, see what people did or didn't do, see who's reinforced it, and then ultimately could come up and say, hey, for Tommy, Tommy's more likely to do these six things. So let's, of the 50 movements that would be good for Tommy, let's give him these six, right? And then you could look at, is that working, right? And then if it's not, you tell the machine, hey, no, we have to adjust, you know? So those are from preventive medicine perspective, you know, using certain techniques here about learning is just a fascinating opportunity. Perfect, thank you, Bob. I have another one that's pretty specific, but I think it will apply to everyone, and Jessica has approved. I have a data set of 18,000 patients collated over 20 years plus, showing the responses to a treatment program in one hospital for tobacco dependence. Evaluating the best outcomes from this real world clinical results could be analyzed by AI to create a protocol for use in the future. Where would I go to find a partner to analyze that data set? You know, they could, if you give it to offline, we could give them choices and several things, you know, not to get into it now. The other thing I wanna point out is, so, you know, just like all humans, AI is only as smart as what you feed it or what it knows, right? So let's say in that particular example, you could feed it your 18,000, and then we could also feed it other things that have been learned about why people don't stop smoking or they do stop smoking. So it could be a combination of your data set, plus we add in some other elements that are known, and that could really help you figure out a good protocol, you know, or what to say or how to motivate. Like in my area of science, how do you motivate, activate, and develop and foster resiliency in people, right? So yes, you could definitely get at things that help you with that. I think that's a great example of having all of this amazing data, but then we don't know what to do with it because it takes so much time. It would take so much like human hours to go through 20 years and 18,000 records to figure out what actually worked for that cohort of people. So it's a wonderful example of what can be done with AI. Exactly. And other than that, we have just been getting a flood of answer of thank yous from international constituents as well who are on the call, Bob and Jessica. That's nice, yeah. Opening the floor to anyone else who might have some questions, because as you can see, let's see. Are you familiar with automated electronic case reporting of reportable conditions? Can you comment on it as an example of AI? Yeah, so I don't know exactly the particulars of how that person's framing the question, but within case management, you know, there have been traditional algorithms that may be put out based on certain diagnoses or certain blood work or MRIs or things like that. Here are four things that may be included in it. Let's say how case management should handle this particular situation or what's best, right? So, and I think some of the case management tools are starting to really more effectively use some deeper learning models and algorithms to get better at that. So I don't know exactly what this person's getting at, but in general, case management is a perfect example. Like if you take three areas, let's say high risk, rising risk, and preventative care for people who are not in those categories, you know, how could you, what is the three triggers that are gonna move someone from fairly healthy in the preventative wellbeing to high rising risk? And what's gonna, what is it, how does it, why do people move from rising risk to high risk? Like, and if you can get ahead of that from a preventative medicine perspective, and that's where AI can help just identify these trends and patterns, right? And like, we didn't talk about all the areas, for example, there's techniques within AI for sentiment analysis, right? So if you're doing, it could get at sentiment where maybe an individual clinician is not as attuned to sentiment, it doesn't have the time to think about sentiment. And based on that, it could be, hey, we believe that Jessica may be very angry or become violent. Like now that may or may not be true. See, like Jessica said, you don't have to use that. Like as a human, you may know, but that could be interesting data. So there are these sub areas, like sentiment analysis, anomaly detection, like what are the anomalies of things that we would have never thought of, like to why certain triggers happen, right? Those are good uses of AI patterns, anomalies, right, those things. We do have a number of program directors and trainees on this webinar. We have about 70 participants still. Is there any experience that you could offer the group at large about how an academic could use AI at different levels of their teaching practice for students, for example, I would imagine they're a lot more savvy with AI as opposed to the more advanced resident as opposed to the practicing clinician. Well, depending on how the question is done, you can actually have a guest lecture or have somebody talk about AI. You could have a thesis if it's a master's or other PhD or other clinical program where you're giving them an exercise to create a model around certain things, right? So there's definitely different ways to utilize it within a class structure to help people think through. You could have little workshops around how this is being used. Again, I think that's what the question was. It was actually more of mine. I mean, we've come from an age where there was a Socratic method where there would be a preceptor and a mentee and it was very much back and forth. So I think it's just shifting the playing field a bit in terms of how to use the AI as an aid when it comes to things like clinic, when you have like a study. Yeah, I see what you're saying. So yeah, yeah. So it could be like the student answers a few questions. Let's say you have a model and AI would determine that this person is not grasping these two or three concepts. So here's, and in the content universe, the professor here has filled in what the AI machine is gonna say. It only could go against this content. So then the AI would tell the student, hey, here's ways of brushing up on these two areas, which could be varied by students. So it could really help reinforce where students need more help from an aid perspective, like a mentor, preceptor, those types of things. Yes, yeah, yeah. You could train it on. That's an interesting exercise actually for an academic medical center to do to practice developing AI models. That's a good one. I don't doubt that. I'm just coming from surgery where most people are getting their practical knowledge in the operating room and in the clinic. We switched to the Socratic method of their grand rounds and incorporated a little bit of AI. But again, it's where do you house that after for repurposing even on today's webinar, for example, talking about this. People want continuous access, continuing updates in real time. So accurate data, right? Right. Any other questions from the audience? I don't see any right now. Okay, hold on. One new message. Biggest mistake you made. I don't get it. Anyone else? Any other questions? We are, as I said, oh, please go ahead. Just one other comment in how we use AI just as cognitive and behavioral scientists. So sometimes patients delay decisions or don't know what to do. So they'll go search the internet. They'll ask their friends. They don't just listen to necessarily what the doctor or healthcare team tells them. And the healthcare team doesn't have as much time to nurture the person to move them to things that can really make sense, right? Because a person's doing this anyway. So one interesting use of AI that we use is we learn about the person and we take what the doctor wants to convey and we nurture it over a period of time to increase the patient's trust and credibility in their own self-trust and credibility. This program's for them. It takes in consideration their lifestyle. They can actually do the tasks that the doctors recommend because they don't think they can do it. They're not gonna do it. So one way to use AI is not necessarily to make diagnosis decisions or come up with theories that may or may not be right. One use is to be a companion to the patient to help nurture what the doctor wants them to do, but using language, tone, and voice that the patient could relate to that the clinical team can't be in their lived environment every day. You follow, like, so there are ways of using this in a way that, you know, there's really not necessarily a downside at all in those examples. Okay, thank you. I'm opening the floor to any other questions from the audience. Again, this is being recorded and will be available once it's posted to our learning management system. Mary, you're good at this moderation kind of thing. What would your, in your experience in using AI, what would your most poignant message be to those on the call today as medical practitioners? I would say that the poignant thing, it's interesting, is like anything else in life, you gotta get in the game a little bit. There's simple ways to start. There's simple ways to think about it. There are ways you can develop a little model and use of this and test it. Like, you know, you don't have to go crazy. So I would say get in the game, you know, because what happens is history is repeated. I remember when the internet first came out, no one ever was gonna do any transaction on the internet because it wasn't secure. So people, some people shied away. And then if you look at EMRs 25 years ago, if I, any hospital exec would say, oh, we'll never have one EMR for the hospital. You're an idiot. Like the ED department's not gonna have the same medical record as surgery, as cancer. And now what do we have today? So my point is there are easy, there are some ways of getting in the game and trying things and that would be my recommendation and doing it in a safe way, like doing it within a little controlled environment, but learning what the downside is, learning what the risks are, learning what the biases are. But if you don't do it, you'll never get better at it. That would be my suggestion. So Jessica, please. The technology is starting to build upon itself at a rapid clip. So to Bob's point, getting to a point where you understand the technology, now's the time to do that because it's going so fast. Things are changing rapidly that it's very hard to keep up to date with the new technology and all the uses and the security issues and all that. So now's the time to pay attention, to start playing around with it. Even if you're just asking chat GPT to write your spouse a Valentine's letter, right? Try it out, see how it works, get a feel for it and then put your seatbelt on because it's taken off. That's right, exactly. So proceed, proceed a bit with caution. So maybe start with a Valentine, but not an evaluation, but maybe see how it works. The other thing to let the audience know is that we are still taking questions. If you couldn't pose your question right now during this webinar, feel free to direct them to myself, Cara Molinari or Carissa Becknell at ACPM. And we are happy to vet them to Bob and Jessica. Dr. Soltis, do you have any closing words or are there any other last minute questions? Oh, let's see. Just thank you a lot. Thank you again for taking the time this afternoon to spend with us and to share these very important foundational concepts. Thank you very much, Bob. Thank you, Jessica. You're welcome. Very well received in the comments. All right, thanks all. All right, everyone. Have a nice rest of your day. You too. Thank you, Bob and Jessica.

AI fundamentals for physicians

ACPM webinar

GOMO Health experts

AI in healthcare

diagnostics enhancement

care management improvement

patient outcomes

machine learning

ethical considerations