#118 AI Chat and Clinical Decision Support with Jen Weaver

One year ago, we released an OT potential podcast episode exploring what AI could just conceptually mean for clinical decision support. Fast forward to today, and AI chats have become a regular part of the decision-making process for many healthcare providers. In fact, open evidence in AI chat now reports that 40% of US physicians use their platform daily. At OT Potential, we recently launched our own AI chat and have just learned so much in the process. With the world of knowledge translation changing at just this incredible pace, it feels more important than ever to discuss what critical questions should we be asking to harness this technology. What are the strengths of chat-based tools as they currently exist? What are their vulnerabilities and how can we move forward strategically? To have this conversation, we are just incredibly lucky to be joined today by Jen Weaver, PhD, OTRL. Jen is a leading voice-in knowledge translation within occupational therapy. Together, her and I will discuss what these rapid changes mean for you in your OT practice, so let's dive in. Welcome to the OT Potential podcast. I'm your host, Sarah Lyon, OTRL. And before we begin, I wanted to let you know that this podcast may qualify as continuing education for you. You are probably listening to this podcast on a free podcast platform, but to gain CEU credit, you will need to be a member of the OT Potential Club, our OT Continuing Education Platform. After listening today, you can log into OT Potential to take a five-question quiz and we will generate a certificate for your time today. So, as I mentioned at the top, we are welcoming to our studio Jen Weaver, PhD, OTRL. Jen started her journey in healthcare as an occupational therapist. She attended the University of Southern California and was motivated to implement evidence-based practice and inspired to do research. However, right away, it was recognized to be easier said than done. Her first stop as an occupational therapist was in an inpatient rehabilitation facility in Southern California. She was supported by her colleagues to advance her education and pursue research endeavors. She earned her PhD in translational health sciences from George Washington University in 2021. She now serves as an assistant professor and researcher at Colorado State University. So, without further ado, I will patch Jen into our live studio. Great. Thank you so much for having me, Sarah. I am so thankful that you are here today. I want occupational therapy professionals to be such savvy users of AI. I want them to reject bad uses, to push back on things that we wish were different about the applications are being presented to us. I just feel confident in the basic understanding of these models and how we want them to work. I presented at the Nebraska Speech Therapy Conference and I had people raise their hand if they were using AI in practice. And 90% of hands went up. But then I asked them if they were using purpose-built AI solutions, like things specifically for healthcare. And then only like two hands went up in the space. So I think the big difference between now and a year from now is that in our EMRs, just in our practices, we're going to see more AI chats that are made specifically for us. And I think that is good, but we also have to be critical of those and recognize the strengths of AI chats for healthcare providers versus something like chatGPT should be helpful. Before we get to all that though, I just want to learn more about you when I think of knowledge translation. For years, I've just thought of your name. We're meeting for the first time today, but I've just known your research and your work. How did you get interested in knowledge translation just overall pre-AI and knowledge translation? Yeah, definitely pre-AI, pre-chatGPT. When I first graduated from occupational therapy school, I was able to take a position where I was one of the two lead occupational therapists for the inpatient rehabilitation unit. A lot of my research now focuses on traumatic brain injury, but that is actually the area where I felt the least confident as a clinician. And that's where all of my extra training went. And eventually I did have my certification as a brain injury specialist. I was curious about different approaches, even ones that were said to not be evidence-based. And I was able to attend those conferences in person. And I was able to read journal articles, but I also recognize that a lot of my colleagues were not always able to attend in person conferences. And reading the journal articles, it was on my own time. We did not have time built into our day to read the new evidence. I was fortunate to work with another person who was really motivated to bring new knowledge back with me. And we would set up different systems in the occupational therapy gym so we could use new assessments that we learned about or think about incorporating different treatments. But I realized pretty quickly that there needed to be a better way of transferring this new information and translating evidence-based information. Now I think of it being even more critical because I feel like we do have more occupational therapy researchers. And the rate of publication is just, that's changed too. With the development of AI, I think we're having more and more publications come out that are relevant and being able to synthesize that information takes a long time if you do a systematic review. So hopefully AI can help us in some of these other areas. I really so much should just set early interest in knowledge translation. I started working as a generalist where I was going between hospital, outpatient and SNF and then seeing all the ages and just really feeling the struggle of like how in the world am I supposed to keep up with MS, Down Dead, Kids with Autism, just the whole spectrum. And really just feeling like it is truly impossible and that was what motivated me to start OT potential and just explore the different ways of doing it. And I love continuing education. I love hearing people's stories and how they've successfully done knowledge translation. I think that's like such an important way that we learn is to like hear from individuals and I think that helps us in code information so much better. But after years of doing that and also like even that's not enough, the questions OT's get are so specific depending on where you practice, depending on the patient and feel like there's no way that I can provide enough or continuing education is only one piece of the puzzle. So I've just been just like really genuinely excited about these new AI tools because I'm like what we we need something like this like what we were doing before just wasn't sufficient. I know this is your area of expertise and I'm sure we all filled this kind of intuitively. But can you just speak to those barriers that we've all experienced like how do you think about them currently, how do you articulate them? Yeah, why is this so hard? I think in terms of barriers, it really to me, it starts at this system at system and organizational level. I think back to when I practice clinically and you would be at 90% productivity. You really have time to like take care of your own needs like eating lunch or going to the bathroom in between your patients or your clients. So one at that level, there's no time built in for you to keep up with the existing evidence. I think it's a good thing to do that on your own time and I think people critique when people document on their lunch hour because you're then not getting paid. I think that's the same critique right that time does need to be built into people's schedules. People in organization level, maybe it's not a daily thing that's built out but a monthly thing again that time is not provided to us and my needing to do those CEOs, we do it on our own time and we have to make sure that those CEOs count. So I think that's a major barrier and then another one access, I think about when I first became an occupational therapist, I was registered and licensed and luckily at that point in time, there was a search engine built into our national board certification and when had access to journal articles, but a lot of the new evidence that is in a research articles behind a paywall. Not all of it is open access even though that's the trend that things are more moving towards. So again, there's two really system and organization level barriers, I think. And when I. I was in practice. I think it took me like my whole first year to really feel comfortable and confident in my routine and then re-engage with evidence platforms where I could bring evidence back to the clinic. But what you said, we see such a variety of patients were trained as a generalist. We could become specialists in one area, but that doesn't necessarily mean that's the only client base that you're seeing. You're still going to see people with other diagnoses or just the personal factors around them where the contextual factors are really going to change things in how you approach treatment. That's where I do think in terms of knowledge translation, we know that for existing literature, some evidence that exists today, if it was published today, we expect that it will take about 16 or 17 years to fully integrate into clinical practice. But a lot of that evidence doesn't make it through those translational hoops. That's something that I think about a lot is how can we improve the speed at which clinicians know the existing evidence and can use it as a reason through how they want to evaluate their client and how they want to develop a treatment plan with their client as well as monitor and adjust it. Just like you said, you can't keep up with it all. You're going to see a lot of patients and you might be a specialist in acquired brain injury, but then you're still going to see people with a hip fracture. So many clients. Yeah, this is, I mean, I've just definitely come to the place where I'm like, our work, I mean, I want to use the word impossible. The challenge is so great for us as OTs to be taking in all this information and then synthesizing it and then building it out into an hour long session. I've been thinking a lot about the difference between how doctors make decisions versus how OTs take that knowledge. So often not to oversimplify the work of a doctor, but they're looking at what level should I do? How do I respond to that lab level? It's much more procedural and like five minute questions that they're looking for support from. And we're looking at how do I take this line from the research and build an hour session around it? That is very complex like the challenge for us. It's hard for all medical professionals, I think for OTs and our fellow people in therapy. It's particularly hard because of no journal article says, here's our finding, here's what an hour long session looks like. Or ideally, it would say, here's what 12 hour long sessions look like. That is so hard like to take that information and build it out into 12 hours in a way. And we have to engage the patient too. So I'm like, that's another, but yeah, that is also incredibly difficult and something that a doctor, the patient doesn't have to be as fully bought in to get their lab done or to take a medication. They have to be somewhat bought in, but they have to be really bought in for OTs. As we talk today, I mean, this is going to be just more conversational than some of my episodes because like I'm in the process of building this AI chat and I wanted to share about it. And we are kind of going into this like conceptual like, what does this mean realm where no one's fully an expert in this topic yet because it's emerging currently. So we've laid out, here's the many barriers of knowledge translation. I'm sure our listeners can just think of their own list and really what that feels like each day. It's a tough feeling to feel like you're not keeping up. How do we feel like AI chats could help us solve some of these things? And what is it poised to really help us solve versus not? Yeah. I'll let you go from there. So in terms of what AI chats could help us solve, one thing, there are different models of AI chat. So some of them, that generative AI model, I think that's where we have to be more cautious because they could make up answers that you might want to hear. You could potentially give it a list of interventions you're thinking about using. And we could ask AI chat, how can I apply X intervention in YOT setting? Because how you apply it in outpatient therapy versus inpatient rehabilitation might look different. I feel like in inpatient rehabilitation, I had a lot of 60 minute sessions, whereas in skilled nursing facilities outpatient sometimes 30 or 45 minutes. So we could even give the AI chat some parameters for build me out a session using this intervention, this timeframe. Another thing that I think could be really great is summarizing new evidence. We just have to make sure that depending on the model of AI that's being used, you just ask it for citations and make sure they're real. Because it will make up citations. Or I know you have a lot of experience, some newer experience with the RAAG or retrieval augmented generation tools. And so again, that holds information from specific databases, places, then you might feel more certain about what is generating for you. One place where I hope to kind of explore soon and some of my future work is we're always asked even in research to provide like a lay language abstract summary. And I think of that in terms of creating handouts for patients. And I remember a lot of the handouts I used looked pretty old. And so is there a way where you could get some really nice handouts and tailor it to the person in front of you? And having lived in many places from the west coast to the east coast as well as now somewhat in the middle here in Colorado, I think about all of the patients' interests. And sometimes I felt like I wasn't creative enough as an OT. But I knew what I wanted to work on. And if I could say I want to work on these skills, create a session that integrates my patients' interests in the rodeo, something that I know less about, but is a big culture here in Colorado, I think that would have been so helpful to tailor my interventions across different patient interests. And would it mean like there would be intervention even more fun for them? Yeah, I think of this, you just kind of alluded to these two basic types of models. And general purpose, generative AI, which is chat GPT, Gemini, and they are so fast, they're so good, they're so satisfying. And I'm like drawn to them for sure, just because they're so satisfying to use. But I definitely think that those have, we should choose, go to those only for a specific purpose, knowing that they are pulling from the sloth of the internet. They may not be as reliable. Sometimes you can't trace their thinking. You can still ask it for citations, but just overall not as reliable. And then we've had the open evidence of the world, these chats that are made specifically for healthcare providers. And a lot of EMRs are actually building in their own chats to like, ethics rolling them out. Some of the scribes have their own chats built in. And those tend to be that rack model that you mentioned where it goes. And it has to look up the information and confirm it. And then it uses the LLM to like polish it for you. Like, so it still comes back saddest, like it's more built out than just like a search. But it's going and finding something that it's confirming before it gives you an answer. The downside is that they are like tend to be slower and are more narrow. Like you will hit things where it says, I don't know. Whereas chat GPT will always pretend like it knows and keep you going. Something I've just been thinking about big picture is one of my hopes for these chats is like on one hand, it will help us. There's some things in our practice that I would love to see a little more standardized. Like that we tend to that we would do like more of the same assessments for different populations or at least like have core measures that we all do, even if you like, if they're part of a bigger mix. But it would just be really nice if we all did like a quality of life measure for chronic pain patients. like diversify what we're doing and really use our full scope and really tailor to individual patients. I'm curious just from your person, this is a very big picture, but like, what do you hope it helps us standardize versus gives us more ideas for? - Yeah. So I definitely study the intersection of measurement and knowledge, really. - Yeah. - So when you're like standardize a core outcome set, I was like, oh, it's like music to my ears. (laughing) We really need them. And as much as we need that, and I know other professions like physical therapy, they do have a core outcome set. Again, it would be great if we had a core outcome measure set where if you see this diagnosis, you're likely to administer a set of outcome measures that go beyond what's really mandated of us, right? I think of host acute care, often the assessments we use, they're the ones that are mandated by our payers. So we use Section GG a lot across post acute care. So I would love for us to have that standardization of outcome measures, but then I really want it also to help us link it to our interventions. Because if you're going to standardize an outcome measure, then we would anticipate seeing the patient make an improvement on that measure. And so that would be another great way for how could you use AI to generate some ideas of interventions that would help you reach specific targets in the outcome measure. So we're not just doing outcome measures at admission and discharge or weekly, if we are doing them weekly or every other week, I know that occurs for some conditions. When doing them, how can we make sure that our interventions are linked to making a change on the outcome that's selected? So I think we do a lot of great research, say in the measurement world, as well as in the intervention world. And I'd like to see those two pair up a little bit better. That's so interesting, because it kind of leads to something I wanted to talk about, which is structured data sense. Part of the way that these rag models strive to be faster is they have a nice structure data set that they pull from. And the thing that's easiest for me to conceptualize is in the assessment realm where you could, well, and I should say I've started on this OT potential, because I just see the potential so much where you could, you were taking assessments. And we've already, we already have an assessment search. So we had like, here's a population it's used for. Here's how much time it takes. Here's where you download it. And then you could build it out more to do the calculations and then get the score, like get the score, have the minimal clinical difference, understand what that means. And with that data set, I can see that potential of like moving towards a little more standardization where like the most popular assessments with a population like always gets adjusted first. I mean, kind of nudge people towards. Like here's 10 you could choose for TBI, but we really recommend these two. Like that I can totally imagine. I can't quite imagine what a structured data set would look like for interventions. Yeah, how we get to that next phase of like, you use this assessment and we're linking it to this intervention. I know, yeah, that's where like my mind is. I think that's where some people have done some really great work on trying to improve the specification of rehabilitation treatments. And so I wonder if it's a matter of trying to adopt some more of that language in our documentation. And with the documentation, it'd be great if there were structured prompts so that it could be almost a drop down, right? What was the target of the treatment? What was the individual ingredients that were directly affecting the treatment target? So if your target was a rousal or awareness, what part of your intervention was actually trying to address the arousal or awareness? And so just like we have the World Health Outcome or organization ICF to kind of distill diagnoses and identify body structures, body functions, there's some work out there about the rehabilitation treatment specification system. And I think that can improve not only research design, but also clinical reporting to improve that replication and then also that tailoring. I think that tailoring is so important. And kind of bringing that in, and Sarah, and knowing that you recently created an AI chat, I know you're aware that AI has a lot of possibilities, but with your experience, how do you think we can build them so they're more trustworthy? Because I feel like that's one of the things we can talk about how we could use AI or how it might support practice and translate knowledge. But just this morning, I have automated PubMed searches and so one PubMed search is about clinical decision making. And it retrieved an article for me about AI, clinician, decision support system. So I'm happy to share that with you later today. But clinicians still often express skepticism of AI, clinical decision support systems. And I think that skepticism is important. And I'm just wondering about your experience seeing how you are in the process of building and refining one. - Yes, we're talking about like the potential where it could nudge us towards court outcomes sets or prompt us in certain ways. You're like, this is getting very powerful. And I think as clinicians, we need to be asking a lot of questions about that. I think that one, as far as trustworthiness, just discerning, is it using that rag model? Is it retrieving and then augment and then using the LLM to augment it at the end? But it's starting with like solid data. And then to what knowledge base is it drawing from? Like I think we need to be asking that specifically as we adopt new things because just like anything, there's like a garbage and garbage out. And that's part of the reason like chat GPT is not is trustworthy because part of the internet is garbage. So sometimes you are going to get garbage. I think that creating knowledge bases is something that's just going to become very, very important for our health tech companies. And as I mentioned before, was this created for rehab or was this created for doctors because our decision making process is so different? I don't think we can be using the exact same tools that doctors are using. That's my hunch. Like I think they could still be helpful for learning about diagnoses, but I'll be eager to see rehab specific ones. And then just from my, like some of the things I've learned from programming one too, like I'm able to say like do not offer to do things that you cannot do. Like there's like there's a trade off between like making it seductive so it keeps going versus telling it to stop and as a user it's frustrating when it stops or it's like I don't know, but that's also honest and I think we should expect some of the models to tell us that it doesn't know. I think that that's actually a good sign. But I just encourage people to ask so many questions 'cause I think we're all going to see these AI chats whether it's at OT potential or in your EMR or using open evidence, feel very empowered to ask these questions. Oh, there's one more thing about training too. Lots of people will be like this is trained for therapy and with that real, like ask questions about that because I know some of them are like, we got 10 notes from students and we trained our AI on that and I'm like that is not sufficient training to me for a clinical setting like really, like if something doesn't quite make sense, like dig in and ask those questions for sure. Yeah, I love that you said garbage in, garbage out. One of my favorite all-time articles talks about garbage in, garbage out. But just in terms of even thinking about the data they're built on, I always think about even as a measurement person, who's contributing to that data set? Or who are we missing, right? Like who's not in that data set? Who's not represented in the data? And that's where I think again, this is just another tool to help that clinical reasoning, but it's gonna come down ultimately to you as the clinician are still the one made up. making decisions, signing with your license and needing to take ownership for that evaluation and/or session note. And just thinking about you might be getting evidence back, that's not contextualized to your person. And that's why I do think AI is, it's not going to replace clinicians because I still think that clinical reasoning is so important. And this is just one tool, just like an outcome measure, it's one way of getting information, but there's so much other information that the clinician is distilling and synthesizing in the moment. - And there's something that we can check easily and something that are hard to check. Or I'm just thinking, Jen, as I've seen my chat, one of the ways I've been surprised is how much people ask it for translation. Like translate this into Croatian. Well, I guess if you were a native speaker, you would know that. But if I had a Croatian speaker, I would not know if it's accurately doing it. So I think you just have to be like savvy about, can I verify the output of this? - Right, right. - Yeah. - What do you think are the options available to us? Because you've created this AI chat bot, how are you going to be able to navigate the speed and the quality of the AI chat? - Yeah, I think that like we touched on the building out a structured data base is the way of the future. And for me, I'm just watching the types of questions that people are asking and starting to ask myself, do I need to incorporate more into my knowledge base? I've been surprised just anecdotally as I'm looking. Like only like 30% of the questions are easily answered in like pub med. And then the rest are like so specific. Like can I do this modality in the state of Texas? Or how do I make this more patient-friendly? Or what goal should I write for this? Like those are all different types of questions and it would have to be trained differently to be good at those. So for myself, I'm like, does that mean I need to prioritize adding regulatory information? There's a way to like train it on goal writing. I think I could definitely do that where I give it like a basic model and then it just goes for that. And something's chat GPT might be just better at. Like I think bringing into lay language maybe. I don't know. I think some of it is training my own and some of it is helping people navigate which model is different for best, which use case for different purpose. Yeah, knowing which model, for a certain purpose. And that's where, yeah, like the lay summaries, a generative AI might be useful. But then I would think the clinician's putting in the information that they want translated. And so that, them to even have to create that information of what they want their client and their client's family to know. I know that takes time to create. So sometimes I wonder about the clinical utility of it, just knowing that so many settings are just very fast-paced. Mm-hmm. I'm curious as you think about this topic, what research questions popped your mind? Like I look at the data that I get from just like my own chat that I just launched like five days ago. And I'm like this is incredible because people just ask such rich question and it's such a window into the clinical decision making process. What do we need to be studying about it from like a research perspective? So I think something you touched on was like goal setting or how can I take the result of an assessment and have something that I can quickly kind of put into my documentation. And so I then think of how can an AI chat model be that knowledge broker to support implementation of best practices? And when I think of the research that I do, I always partner with clinicians because I realize that every year I'm out of clinical practice, I'm just getting further away from it. So I think what you're learning is and what's amazing about what you're gonna learn from knowing what questions are being entered are what matters to clinicians because then by knowing what matters to active clinicians, I would then think some of our larger conference organizers could tailor the content to those needs of the everyday clinician. And I also think that research is changing and I've seen this at the national level so national and sea sheets of health, Department of Defense, as well as American Occupational Therapy Foundation requiring for some of the grant applications to be engaged with other shareholders or relevant parties to the research. So for example, one submission, the research question didn't necessarily stem from me, but the care partners and clinicians that I worked with and then I was able to create a research design around it. So I am really curious how AI can help us distill information. Systematic reviews take a long time and a lot of effort. Could that be done more quickly? Because again, those are knowledge tools that clinicians could use. A big piece though I think that we haven't quite touched on is the ethical guidelines. Like what ethical guidelines are needed for using AI chatbots or generative models like chat GPT. I've seen a couple of research products produced with a co-author of chat GPT. - Whoa. - I'm curious how that will even change documentation. Well, documentation say like supported by and then whatever AI chatbot or model that was used, it could be an in-house epic AI system, but I am kind of curious about the ethical guidelines that may or may not surround how it's used. Can't tell you what pet peeve there. Like every time I hear about ethical AI use, I'm like it's all about adding more regulations to us as therapists. And I gen feel super annoyed because like the therapy that I provide is so highly regulated. I have a board city of the state of Nebraska that could strip my license at any time. If I act outside of that, whereas these AI models like can say they provide therapy and do not have any regulations. So I feel annoyed that we're talking about more regulations for OT's. And I just also went to be talking about what does that mean that they're quote, unquote, providing therapy and how much does it have to be regulations on the models themselves too? - Yeah. - That's like a whole nother can of words. - Yeah, that is a whole nother can of words. - Yes, yes. I just feel so annoyed because I know how many regulations we have already to the point where it's ridiculous sometimes. - And that's where to be clear, my ethical guidelines are more about on the model because I think there's a lot of startup tech companies that are creating different uses of AI, whether it's even for mental health therapy. And again, what are the bounds that are gonna surround that? - Yes. - Yes. I mean, there have already been some pretty like sad news stories about some of the uses of AI. So I'm not gonna ignore that. That exists and it's out there. When I do think of ethical guidelines though for OT's, I do think it's on us to know what model we're using. And so that's why I think even webinars like this are important because to know if it's retrieving information only from PadMed, only from OT potential or can it make up an answer that you will appreciate reading. I use chat GPT to try to think about how to explain more complex concepts like the Rosh model. And it gave me an answer today about the Rosh model and I literally out loud to my computer and said, you're just giving me an answer. I wanna read. (laughs) If you think that because like what AI is gonna talk about the Rosh model. So it does learn what you ask and it's gonna like a model like chat GPT or Gemini. It's gonna learn the types of questions you ask and feedback that information to you. So just being kind of conscientious of what its scope could be, I think is really important. - Yeah, there are so many ethical considerations and for us to like, for us and for this tool, I would love to have someone on the podcast just to talk about that specifically. And what the options are even like that feels so important in all of this. One more thing I wanted to be sure to ask about with thinking of AI chats. I can't help but also think we're using AI chats, but so are our patients. And I saw a little anecdote the other day where someone went to see their cardiologist and paid $200 to see them. And the patient got died and they were like, "Well, you just told me what Chad GPT told me why I should pay for that." I think that people, our patients are using it too. I can see it changing expectations around what a therapy session should be like for so long, even though we tried to do different models. People still come to healthcare appointments to get knowledge from an expert quote unquote, but with so much more knowledge being accessible, how does that change people's expectations of therapy sessions? That's such a good question. Now, do I have the best answer, but I think over the past couple decades, even before AI chat, people have been Googling their diagnosis or their symptoms right before whether they're going to a healthcare provider or going to a specific therapy session? I do think because we know the evidence and there's more evidence suggesting that higher intensity is better. So whether that's higher intensity of the specific intervention or higher intensity, meaning just we don't have great ways of looking at intensity yet, but some studies use more minutes of therapy can result in improved outcomes. While we are trained as generalists, we do over time become specialists and we also, I just can't deny the fact that when you gain expertise, I tell all of our students whether they like it or not, they are in essence a researcher to some extent within N of one design every time, but our own brains will recognize patterns and there are times when you might go into evaluate a patient and you're like, oh, I can tell that they don't know what to do. They're underwear. It's going to go on their head. Right? Like you can almost know what's going to happen next because you've seen something time and time and again. That's something that your patient, your client does not necessarily have. They have not necessarily seen a condition over and over again or different performance pattern. So I do think we will still have a strong unique clinical lens to bring in terms of what's expected in therapy sessions. I think we talk about how to implement best practice, but I also think there is the idea of also de-implementing what's not best practice. So we may have been doing activities in the therapy gym or place that are no longer best practice based on new evidence. So figuring out how to de-implement those because they do somewhat become routine for the therapist to choose, right? It's a part of their routine to maybe go warm up on an arm bike, but there may be better uses of their time in the patient's time. So I do think the patients and maybe not just the patients, but also their families are going to have those expectations that their therapy time is used in a specific way. Of however high value looks to them, that's what they're going to expect and we're going to have those conversations upfront. That flows really naturally into two of the questions that we've had in our live chat. One is about, she called them quote unquote, old school therapists and that divide with this new generation. That divide has always existed like we haven't solved it and it's going to get worse with these AI chats. The question was how do we merge these quote unquote old school therapists with this new way of being, I see that being of huge divide and workplace, a huge divide in how we're translating knowledge. How do you think about that gap in merging it a little bit? Yeah, so I think that's one of the challenges even with occupational therapy research is a lot of information is tacit. And I am well aware that the evidence says that say NDT neurodevelopmental technique is not evidence based and even hearing that as a clinician, I still went into the two week course because I was so curious. It almost made me more curious because I was like, it's not evidence based, but we still have these trainings and my take away from that and even the individual patient cases that came from the facility that was hosting the training is I don't always think that what we measure or how we measure has been refined enough for everything that we do. So when I think of maybe an old school clinician, they might be using evidence that did exist and I think there is evidence that does not support NDT, but I can't let go that evidence is one part of how we make decisions, right? It's also informed by our own individual expertise and I would think of that, quote unquote old hot clinician, they have a lot of expertise. When I remember being, you know, the first couple years, I really relied on the evidence. And I think I read so much of the evidence because I wanted my clients to know that I knew what I was doing to some extent and I was evidence based, but clinicians with expertise, they can see patterns that have occurred in their practice over time. And that's another element to make an informed decision as well as taking in the patient's values and preferences. So I think in terms of that divide, I think hopefully the newer clinicians can show people ways of accessing information easier, more easily, and then hopefully the clinicians who have more experience in the field, they can also help caution the newer clinicians with, well, make sure you look into this, right? Or you never know what facet may be contributing to something like a cognitive decline. And so those are patterns that our clinicians with experience may have looked at that for years and can help guide the newer clinicians, been doing it for a while. So kind of marrying that new knowledge with also people with the expertise and pattern recognition, just built in. Yeah. Yeah, I think that's such a good reminder of just what I was thinking about at the top two where I'm like, there's so much that goes into a one hour therapy profession, like finding the evidence is like five minutes of it. And then there's a ton of soft skills of like guiding patients through that like arc of a therapy session. And so often our experience therapists are just really good at that. Yeah. Yeah. They've honed a lot of skills. We also had a question about AI tools impacting shared decision making, which that's such a great call out because I do see the potential for our clients to come in more aware of their options. Yeah. And for that to open the door to more shared decision making than ever before. What do you think of? So I love that. Yeah. I'm actually working with colleagues now on a study about shared decision making. And I do have my own views about shared decision making. So I focus a lot on assessment because my hope is that we have better data visualizations for clinicians to share back our rehabilitation assessment data with clients and their families. It don't necessarily think that's always done in a way that is easy to understand. And so even having that information exchange where the patient is telling you what matters to them and you're then able as a clinician to share back the findings from your evaluation, I think that can be the start for some really great opportunities in shared decision making. We also have different options and therapy that may not seem shared decision making to me has been studied traditionally in the medical model, right? Is drug A better than drug B? Do we want to do a CT scan or is it worth it to do the MRI? So when I think about shared decision making and rehabilitation, it might be that it's more in the moment and at the bedside. Like if you're trying to stimulate somebody's visual tracking, is there a way for you to engage with the family or with the patient and find out what motivates them to try to use a more preferred object or stimuli compared to the cups or toothbrushes that we have in the hospital base setting. So but I do think with AI, sometimes our treatment options are not as specific as they are in that medical model of shared decision making, but I do think there are ways that we can better engage our clients and our patients and also share back the data that we collect and by the data I mean our assessments and things of that nature. I do think they might come in more knowledgeable, but also maybe it will help us in sharing information that can seem really technical and an easier to understand or like hand out or maybe it's even it can help us with some data visualization that's specific to that client. I hadn't even thought about and I was so fine because I've actually seen it. I hadn't thought about data visualization as something that these AI tools will actually help us do and I've actually seen it in EMRs and I just hadn't thought through the implications of how helpful that will be. What if you learned about data visualization? I think really helped me the other time. I was like really going to ask about this. Yeah. One thing that I think impacts both clinicians and researchers is that most of our outcome measures are ordinal. We know that as the score increases the patient traditionally is getting better. However, what we don't know necessarily is the amount of change. That's one place where I think data visualization could really help us is that by taking that ornal data there are transformation tables that exist for some assessments to put it on an equal interval ruler. So where you can see that is zero to one might represent a lot of change where something in the middle like an 11 to 12 might represent a small amount of change. Now progress and of trend towards improvement is always excellent to see but I do think and hope that data visualization especially in the electronic medical record that's not built for clinical decision making. It's built for billing. Hopefully even our EMRs can help give information back to the therapist about what their patient looks like maybe compared to other patients that have come into that facility or place. So I think about data visualization a lot and I'm happy to chat more about that in the future because I think it's another important area where AI could really help us move forward. Yeah. We talk about the eventual transitioning to value-based care a lot on the podcast and what that will logistically look like and just as last week I saw my first EMR that was built for value-based care where it was built around tracking quality versus tracking billing and that like that shift towards value-based care plus AI is such a potentially exciting future and I really feel hopeful that we can harness the things that we're talking about today to really make therapy better. We are at the end of our time what's your last just like one minute wrap up take away of this conversation for us your final charge to our listeners. Yeah I think my wrap up is really just to embrace new innovative models and tools with caution. I want people to embrace things. I want people to try new things. I think it only makes us better as people the learning is really important for ourselves but also as new things come out and new things are developed I hope that clinicians continue to be skeptical and criticize it so that way they are judiciously using the new tools. Oh Jen that was so perfect I'm so thankful for this conversation I want to keep talking we'll have to have you back. Thank you Jen for this time for sharing I just appreciate it so much. Yeah thank you so much for having me. Thank you for joining us on the OT potential podcast to earn one hour of AOTA approved continuing education for your time today you will need to sign in or sign up at OTpotential.com. Once you're in the OT potential club you will find a five question post course quiz connected to this episode. When you pass the quiz with a score of 75% or higher you will be able to download a PDF certificate that certifies your completion of this course. Okay I want to thank you for joining us today and we'll see you next time. you