Transcript
Marla: Welcome everyone, and thank you for joining us for today's webinar, debunking AI, miss the Truth about AI and Physical Therapy. I'm Marla Ranieri, head of Clinical Innovation and Clinical Strategy at Prompt, and I'll be your host for this really important conversation today. Artificial intelligence has really been a hot topic.
I hear everybody talking about it on LinkedIn throughout their conversations, hiring new clinicians or just in general with other business leaders. And it's a little scary. There's a lot of confusion about it. Is it something I wanna jump into as a clinician I wanna jump into as a clinic owner or even therapist coming into the field.
They wanna know. What's the misinformation about it? What is accurate? What should they be using? How should they be implementing it? And that's what our conversation is gonna be about today. So some hot questions and topics that come up is, will AI replace therapists? Is it too complicated for our clinics?
Can it really deliver personalized care? So we're gonna ask these questions to our panelists of experts. We are very fortunate to have some phenomenal panelists that I will be introducing in a few minutes, and you'll hear directly from them, clinic leaders, clinic owners, and also AI developers. We're gonna ask them about how they use AI to streamline operations, enhance patient care, empower their teams, not replace them, and also just make it very interactive and dynamic.
So we wanna hear from you as well. We have a chat bar where you're more than welcome to talk amongst each other there. But if you do have a question. If there is a q and a at the bottom, please use that for any questions, so this way we can absolutely address your question and make sure we answer it during this webinar today, or if we don't, we can always follow up afterwards.
And today will be recorded as well. So if there's any points or topics that come up that you wanna review, we will have that recording for you too. So before we begin, let's launch a quick poll. I wanna get to know the room in the audience and hear from you. So you should see it popup on your screen. Just gonna ask you where you are in your AI journey.
Are you just starting to learn ai? Are you actively researching options? Are you already using it in some areas? Are you using it extensively or are you still skeptical or unsure? So feel free to answer that. I'm gonna answer it myself as well. Great. And this will give us a nice perspective. And as I said, feel free to chat amongst each other in the chat too, as we really want everybody to be engaged and to be able to be contributing today.
Great. So now that we know our audience, let's meet our panelists, so we have some great ones here for you. Amanda Brewer, thank you for jumping on. Amanda, good to see you.
Amanda: Of course. Happy to be here. Good to see you too.
Marla: And Amanda is a physical therapist, entrepreneur, and the founder of Brewer Physical Therapy, which is a multi-location outpatient practice in Louisiana, established in 2007.
She holds a doctorate in physical therapy from the University of Mississippi Medical Center. And brings nearly 20 years of clinical and leadership experience. She's a certified clinical instructor, dry needling specialist, and served as the vice president of A PTA Louisiana for over six years. She currently sits on the board of directors for the A PTA Physical Therapy Provider Network of Louisiana serves as a key contact for A PTA and an alternate delegate to the A PT House of Delegates.
Thank you so much for all that work that you are advocating for all of us, Amanda. Of course. Of course. Despite having no formal tech training, Amanda has become a self-taught leader in integrating AI into her outpatient care, using tools like chat GPT to reduce administrative burden, streamlining operations, enhancing patient outcomes, and she's here to share what real world implementation of AI looks like for clinicians and practice owners alike.
It is a pleasure to have you. So now we have Sterling l Carter who is found, who's the founder of Sterling Physical Therapy and Wellness, and a co-founder of Sterling Staffing Solutions. Thank you for joining us Sterling.
Sterling: It's great to be here, Marla.
Marla: Sterling is a decorated US army veteran healthcare entrepreneur, and nationally recognized physical therapist with over 25 years experience.
He retired as a major after 26 years of service, and he led over 500 soldiers as an executive officer of a US Army Medical Hospital. Hospital. Wow. Thank you, Sterling. That's fantastic.
Marla: He holds a doctorate in physical therapy from Simmons College and a master's from Texas Women's University.
Sterling founded Sterling Physical Therapy and Wellness and developed the Sterling Treatment Method, achieving an 85 success rate in pain reduction. His clinic boasts over 505 star reviews and is nationally recognized for excellence. He's a co-founder of Sterling Staffing Solutions, a three-time Inc. 5,000 honoree, and is active in healthcare, real estate and business development.
His accolades include the Presidential Lifetime Achievement Award and Houston Business Journalist most admired CEO. Wow. Very. Very fantastic and a passionate mentor and speaker. He's dedicated hundreds of hours to student mentorship, community service, while continuing to lead and innovate in the field of healthcare and entrepreneurship.
Thank you, Sterling. Can't wait to hear more from you.
Sterling: Excited to be here. Thank you.
Marla: And finally, we have Rohan Nanu, who is not a PT. He's the lead AI engineer and strategist at Prompt Therapy Solutions. Thank you for joining us, Rohan.
Roshan: Happy to be here.
Marla: and he is a leader in AI innovation. His unique expertise in transforming outpatient rehabilitation practices is underpinned by a PhD in neuroscience and a rich multidisciplinary platform of physics, machine learning and engineering.
He has helped launch AI initiatives with numerous enterprise companies, and he is now the lead AI engineer and strategist at Prompt Therapy Solutions. Pioneering AI solutions to challenges and opportunities arising within physical therapy. His work focuses on leveraging AI to streamline operations, enhance clinical productivity, and deliver data-driven strategies that improve patient outcomes and drive business growth with deep expertise in healthcare technology and implementation.
Rohan advises organizations on how to adopt smooth integrations of AI into workflows for a whole new perspective on managing and delivering care. It's a pleasure to have you. We can't wait to hear more about the AI and developmental side of what we're talking about today. Thank you, Rohan.
Roshan: Thanks for the intro, Marla.
I'm looking forward to talking about it.
Marla: Great. So now we're gonna get right into our questions. And like we said, there's been a lot of chatter about ai. It's a hot topic, a lot of clinics wanna implement it, might not know how, and also a little bit fearful. So I'm gonna start with Miff number one. Which is will AI replace physical therapist and rosin?
I'm gonna punt this over to you a little bit so you can talk about how AI is designed to assist but not replace clinicians. And a little bit about why the human connections still matters so much.
Roshan: Yeah, I guess I'll start with my own, like misgivings about AI as a generalized tool. We've seen numerous cases over the past decade of people trying to use AI to fully automate things, and the truth is, in most situations it just doesn't work out.
So a few months ago I came across a new product out there that boasted a large action model. And the concept was you just have this little phone with a button on it and you talk to it and tell it to do whatever, and it goes and does it like planning a trip or making new music or creating a playlist on Spotify.
And the first time I used it, I asked it to look up some recipes and plan out a meal plan for me and. It spent 45 minutes trying to do this, but kept trying to log into something failing and then trying again over and over again and just going off on its own little tangent for over half an hour and I couldn't get it to stop.
So long story short, I don't think AI is anywhere near ready to fully automate things, even at the level of meal planning, right? It's definitely not ready to fully automate PT jobs. And that's why, when we design AI tools and we're looking at it, the trick is always going to be how do we turn something that's not fully reliable into something that can reliably be used in PT.
And we always see it as an assistive tool, right? We focus on AI for specific purposes to give practitioners insights that they need to speed up their job and help them do it in a more data-driven way. Like we're no way trying to remove that human connection or remove them from the loop. It's just not there yet.
And I doubt it will ever be there because there always is that human component, especially in medicine, where you have to have that human interaction.
Marla: Yeah, I love that and it really is true that it's progressed so much and there are so many uses of AI, which we'll talk about today, but it's not to be your physical therapist.
It's to automate and to assist so that your physical therapists, your occupational therapists, or speech therapists can do their job even better and have the time to be able to focus with the patient and focus on those human connections. So Amanda, I'm gonna. Bring that next question to you and saying, how has AI helped enhance your team's interactions with patients instead of replacing them?
And if you could tell us a little bit about that.
Amanda: Yeah. So from my perspective, and as a busy PT clinic owner, I actually want AI to replace some of the tasks that I do not find meaningful or fulfilling. There's data entry writing appeal letters, repetitive documentation. Those aren't any reasons that any of us went to school to be a pt.
So I feel like we love to use AI to take some of those things off of our plate or make it just easier. And that also, like it's been alluded to before, gives us an opportunity to spend more of our time doing what we do love, which is coming up with the best patient experience that we could possibly deliver.
And also, of course, outcomes.
Marla: Thank you. And that's exactly what I think everybody on this call wants to hear about and find out a little bit more, and we'll dive into more about how you're doing that and the different uses for AI. So thank you Amanda. And Sterling, what about you in terms of your practice and how you have seen that AI components help enhance connections versus removing them or taking them away?
Sterling: Yes, I agree with Roshan and Amanda on so many of those points. It's been amazing. I think right now in healthcare we are on the cusp of something really big because it allows us to do our jobs better, faster and more effectively. We're able to communicate with our patients in an effective way and get the outcomes that we need from developing a really concise evaluation. And really creative goals that are specific for our patients. All that is possible with AI. Even as a CEO when I want to communicate with my team, if I wanna send an email and I just wanna make sure that the email is politically correct or it's not too abrupt because I'm prior military.
I can put that in ChatGPT and it helps me to smooth out the edges a little bit. So there's so many different things that, and when we're talking about healthcare. And the fact that right now we're having to do more with less reimbursement rates are going down. We're having to keep our quality and outcomes up, we need AI to help assist us with that.
And I really think that this is one of the industries in which. It can only assist. It cannot take over. We have what's known as our biopsychosocial method, our approach to treatment, and we need the hands on touch. We need the communication, the motivation, all of those things in order for our patients to get better.
But AI just helps with all the other aspects of healthcare.
Marla: Yep. Yep. And just as you guys said, you want it to replace some parts of that administrative Yeah. Or that busy work that nobody really wants to focus their time on. They'd rather be focusing their time with the patient, interacting the biopsys, asocial model, and creating your therapeutic alliance.
Great answers and really appreciate that insight. So let's move on to myth number two is AI too complicated for PT clinics? And I'm gonna send that back over to you, Amanda. Tell me a little bit more about how you feel implementing AI, implementing AI and teaching it to your clinicians and using it in your practice.
Did you find it too complicated or give us some tips and tricks on how you did that?
Amanda: So one very important thing to note is I am not a software engineer. I am a mom of four. I have four clinic locations and about 50 employees. And so I really just don't have any time to waste. So as soon as I started hearing about, ChatGPT was the first thing that I was aware of.
I. Instantly became intrigued by how could this help me? And especially because I don't have tech experience. So I just started very small and gave it de-identified spreadsheet information to help me analyze some of our data with our clinicians. Being able to use AI scribe from prediction Health inside of Prompt, I just, I can't say enough wonderful things. I was completely blown away. I used to joke long ago, probably 20 years ago, if I only had a body cam that could record everything that I did, while I worked with my patients and it could just do my note for me. And I feel like it's near that level.
It also helps you to be compliant as well, so that's wonderful. But, we just use it basically thinking about what are the problems that we're facing. What do we not like doing? And I just asked chat, GPT, how can you help me, lighten the load essentially. I also started out using it to ask it to teach me how to create effective prompts to accomplish the goal that I was trying to get to.
And it does, it tells you exactly what to do.
Marla: Yep. Yep. And that's, as you said, self-taught, self-made. And that's the beauty of it being there and being able to ask questions and also starting to dive into it. I like that you did dive into Prediction Health and Scribe and how you can use that with your therapist.
And it sounded like that was a pretty seamless transition. Sterling, I'm gonna punt the question over to you as well, and a little bit less on the clinical side, but maybe the admin side in terms of adding some of the features that are within your EMR online scheduling that utilizes AI and different of those pieces, how do you feel it was in terms of two difficult or too hard for your staff to adopt and for your company to be able to utilize?
Sterling: I think one AI is, has made it. Easier. It's not more difficult. The more that we use it actually easier it gets and the better that AI technology gets. It's learning as it goes along and just gets better and better. For my team, we have. Been more early adopters of a lot of the different AI tools that Prompt has had, which has been amazing.
One you were talking about as far as the online scheduling, patients can schedule their own appointments if they're not, if they haven't scheduled out their full plan care, they get alerts to letting 'em know that they're under, they, they have under scheduled for the week and then they can schedule their own appointments.
It's completely automated and integrated into our system to make sure that. We're maximizing our capacity. We're making sure that our patients are getting reminders when they need them. It gives the patients the ability to control their own schedule and pick their therapist.
So it's pretty awesome. And of course, we use Prediction Health as well. And literally we have the ability to come in on, during an evaluation and just hit a button. And it's listening to everything that we say. It only pulls in the stuff that's pertinent for the evaluation. And you can talk about the dogs and the birds flying down the, whatever.
It's not gonna pull up, pull those things. The other great thing too, about having an AI tool like this, or just like Chad g Bt, even. What I can do is I can take all of my signs and symptoms, all of my findings for my evaluation and put it in chat, GPT, and just ask the question and what do you think is a probable or some probable diagnosis of this?
I'm not asking for the answers, but I am looking for it to research all the evidence-based research and telling what's the best probable answer here and it really does help us as clinicians be, be more accurate and come up with a treatment approach that's better for our patients.
Marla: And I would just put a caveat to that, that obviously ChatGPT is not HIPAA compliant.
You're doing that de-identified with no patient information that would absolutely no patient
Sterling: information, just only signs and symptoms and general information. And it will give you just a general probability of problems or issues.
Marla: Great. Great. And Roshan, I wanna ask you, because now as we've talked about some of these different ways that it's not too complicated for them as clinic owners, I'd love to hear on your end as you're developing all of the AI tools about how you're thinking, to not only develop it to be in a compliant manner and also to make sure that it is easy to implement into the system, into prompts, so that our clinic and clinic owners can turn it on.
Roshan: Yeah, exactly. I think it's a huge concern right now actually because suddenly AI in healthcare is a huge hot topic in the field. So everyone is trying to jump into developing AI tools for healthcare. And a lot of those do have a lot of restrictions on how to be compliant. And things they need to consider before jumping into healthcare.
Right now, with the huge flood of AI tools coming towards healthcare, I think it's really important for people to be able to properly vet what tools they're bringing in and how to use them. Like they were saying with ChatGPT, everyone knows about it. It's great to use, it's important to make sure that all your info that's going in there is de-identified because.
With your free ChatGPT account or even actually with your paid ChatGPT account, they're training on your data unless you have a very specific HIPAA compliant agreement with them. So on our end, a lot of it is about making sure that we have all those agreements in place, we have all the right tools in place to de-identify that data to make sure it's used carefully.
But then on the other end, it's a lot about design and integration. I was developing AI systems with some large companies like Panasonic before Chat, GPT came out and all of the focus there was on right, making good traditional machine learning models like linear regression models or deep learning models.
Some of it was using like the precursor to chat GT GPT-3, which was just a basic large language model. It was okay at best, and then as soon as chat, GBT hit the market, every single company changed what they wanted. Because suddenly now there was this user interface where the everyday person can talk to their AI and they can get it to format their documents.
As they've been building on features, now it does spreadsheet analysis. Now it will search the web and do deep research and give you researched answers with sources which is all great if you are a user talking to the AI directly. That said, if you aren't familiar with good prompting techniques or you haven't taken the time to learn that, you can find it to be sometimes infuriating.
I've had several people come to me being like, I can't get Gemini to work, right? I can't get chat GPT to gimme the format or answer I want. It keeps inserting things like this and then really it takes a little bit of coaching on how to prompt, right? But when we're designing tools for ai, we need it to be much more reliable than that.
So a lot of it is about, how do we design this so that it actually. Integrates well with their EMR and their workflows without disrupting them. We don't want our users to have to iterate a bunch or keep saying no, not that paragraph. No, let's shorten this over here. We want it to be accurate every time, or else it's gonna take more time than it's worth.
So we spend a lot of time thinking about how that UX should be building up guardrails around language models or around even regular ML models so that we can get reliable outputs that can be trusted. And that's really, where most of my focus is like, how do we build these systems so that we can focus and direct AI towards specific purposes and automate workflows.
Marla: Yeah. And I would say Roshan, a lot of the, what you built inside the EMR, people don't even know AI is there doing it. So there's a lot of backend AI happening that again, is, makes it really easy to implement in your team because on your end, Amanda and Sterling, you can click and just say, yes, I would like this feature and be able to implement it.
And that goes hand in hand with a question that we have in the QA right now that says. How do you afford the additional costs of AI, especially as reimbursements are going down and salary needs are increasing, as we all know, we're feeling that. But we'd love Sterling. I'll ask you about that part.
I know you said you're using PredictionHealth and some of the Prompt Plus features or most of them. How are you able to afford that with your current EBITDA and the way that your clinic runs?
Sterling: Great question Marla. And a great question for that person that asked it, number one with PredictionHealth.
What it's helping us to do, and what it helps my therapist to do, is understand how to bill appropriately. So there are situations when we may build therapeutic exercise, which is one rate, but PredictionHealth is actually helping you to say, okay, I. You can actually build therapeutic activities and it's a higher paying code.
It's all based on what you're doing, that particular exercise. Why are you doing it? What are you doing it for? And so it really actually helps us with billing and getting higher reimbursements. So that's what we see there. It's also helping with auditing our notes to make sure our notes are being written appropriately so that we, if we do get audited by Medicare.
We're not getting money pulled back and that sort of thing. So it's very proactive in those things. And then looking at the prompt side of things, we the numbers don't lie. We actually have met with the team at prompt to ask how much have we saved or have we generated and savings based on the Prompt Plus system.
And what we're seeing is that we have a significantly higher show rate. We have patients that are completing most of their visits. So our goal is to get to 10 to 12, which is the industry average per episode, and we're hitting that, whereas in comparison, when we didn't have these AI related tools, that wasn't happening.
Marla: That's a great answer, and I appreciate that insight. And Amanda, you as well, just going into that, looking at the business side of it, how are you able to utilize those AI tools? Any additional information from what Sterling said?
Amanda: Yeah, he brought up some really great points, especially with Prediction health and helping our therapists be able to feel confident based on what they have documented in the note that.
It's reminding them essentially that there are other codes out there, besides 9 7 1 1 0. And Sterling said, we do get reimbursed more for that, and that's really important right now when we are facing reimbursement cuts or just stale rates for such long periods of time. I really view it as a really excellent return on investment and it really just pays for itself.
And not only that in monetary value, but even from the perspective of reducing burnout. We have therapists that say, I. I'll never work anywhere where I can't use Prompt and PredictionHealth, and so they just can't imagine going back to the old way of how things used to be.
And that has a pretty big value to me.
Marla: And that shows in the dollars, right? Go ahead, Sterling.
Sterling: I was just gonna chime in on that portion about burnout we have, our therapist is so much happier now. They can complete evaluations, the documentation in minutes, and they're not staying at work.
One and two hours after work trying to complete documentation. They're able to go home and spend time with their family and doing things that they love. So it absolutely improves culture. Just patients, our therapist satisfaction, and those are the things that you really can't put a price tag on.
Marla: Yeah. If you can't put a price tag on burnout, you can put a price tag on rehiring a new clinician and all the time and effort that goes into, but you're absolutely right. Thank you for that insight. And now we're gonna move on to the next myth, which is. AI can't personalize care. So Rashan, I'm gonna punt this one over to you in the sense of, I know we're doing so much at prompt with AI Insights and potentially how that helps personalize care even more can you tell us a little bit about AI Insights and why, and how we've added that to help personalize the care?
Roshan: I think we might call the feature something different on our end, but I think it's case summaries. Yes. But yeah. Yeah, with AI, the goal there really was to be able to give providers, 'cause we know, the same provider's not taking on every patient every visit. I. Or, sometimes a PA is stepping in or sometimes someone stepping in to cover.
Or even if it is the same provider, maybe it's been a little while since the last visit for that patient, or they're seeing a lot. So the goal there was to really make sure that when a provider is jumping into that visit, they can quickly get up to speed on. Hey, this is what's going on with this patient.
Hey, this is what's changed with this patient over the past few visits, or what hasn't changed, which might always be just as important, right? So we wanted to make sure that without having to read through all of the previous notes, a provider could very easily understand, all right, this is what, who this patient is, what's going on with them, and what I need to focus on for this visit.
Rather than having them spend a bunch of prep time, which they probably don't have time to do with visits, book back to back all day. My wife's a doctor and she's dealing with that all the time. So the goal there was to help use AI to help providers be more connected with their patients when they walk in and spend less time being, like getting up to speed.
Beyond that, AI is actually going in a lot of directions to personalized care. Right from, I think one good one that Sterling brought up was helping find evidence-based research pertaining to specific patients. I know A PTA is putting out clinical practice guidelines all the time, and I know providers probably don't have time to read all of those or to, for every patient, find the exact clinical practice guideline, dive into those 76 page papers and say, Hey, this is exactly what's needed for my patient.
And so right now we're looking into using ai. To help curate that, that clinical, those clinical practice guidelines and get that information to providers for their specific patient that they're dealing with at the time. We're looking into using AI to help automate documentation and billing for those specific patients, but also to help start recommending things like.
Potential treatments that might help that other providers may be using or potential ROMs that they're taking for a particular patient that are the best to track the progress for that particular patient.
Marla: Yeah. And I know as a busy clinician and Amanda, I'm gonna have you chime in a little bit as well to be able to look at that case summary and know exactly what went on in the past about that patient and even some really key pieces of how their pain is progressing.
It does save time and helps you personalize and helps you be really attentive directly to that patient. Amanda, gimme some insights about how you are utilizing that or helping to make it help that personalized plan of care.
Amanda: Absolutely. Even you could take those AI summaries, the case trends from the documentation and put it in the prediction health sidekick or, chat GPT and say, here's this.
The status what should I be thinking about? As far as helping to progress the plan of care, especially for new clinicians even, that can just bring some things to top of mind as to what they could do to improve those outcomes for the patient. And like Roshan said, another thing that we did was.
We just got a shockwave. And of course it's not something that I learned about in school. So it's something I have to learn more about along the way. And there is another really great AI tool. It's called Open Evidence, and it is fantastic for clinicians. You do have to have an NPI number, I believe, and then it's a free account.
But it sources. And you can set different parameters of, what time period, what type of evidence you're looking for. So we were able to pull all of the recent evidence off of open evidence. I then uploaded that into chat GPT and was able to very quickly. Get an easy presentation essentially to educate the rest of our staff on, especially with the different parameters and dosing to help us use our shockwave more effectively with our patients because it's new for all of our providers.
So that was really helpful and created our laminated sheet that we could put up on the shockwave machine so that everything is right there and easily accessible for the providers.
Marla: And probably saved you quite a bit of time if you were to do that yourself. And Sterling, what about you?
How are you utilizing some of the features to help personalize that plan of care?
Sterling: Couple of things I wanna say one, Amanda, thank you for that resource. 'cause I haven't heard of Open Evidence, so I'm gonna use it. And Roshan, great point on in regards to case summaries, I think that's a really important AI tool that we're able to use for prompts.
If we have therapists that are working on the weekends that typically aren't a part of our full-time staff, they can see what's going on with these patients and see a summary before they see 'em. And it helps with more effective care. But going back to how we can have more personalized evaluations, we'll start there. What I love, and I, and every therapist out there is probably gonna be able to resonate with this. The old way of doing things was to have this template and you have okay, pain level, this constant, constant pain, intermittent pain, whatever. And it was all a template format.
And even though you were putting in customized information, it still had a very canned looking approach to it. Now that we're able to use Scribe, scribe the actual system is listening to what we're saying and it's putting it in a narrative format and it makes it completely different and completely custom for every single patient.
All of our goals are being assisted with goals, so it used to be rare. Every goal was to reduce pain from zero to, to, from five to zero in four weeks. That part is gone. Every single goal is customized. It's actually a very customizable approach to therapy.
So I think everyone would really love that.
Marla: Yeah, and it's the therapist's discretion to accept that or to change it or alter it, but it does help make you just sound so much more, I don't know, as we calling it personalized or fluent uhhuh early with those goals, instead of just doing exactly what you learned in school and verbatim writing the exact same goal, patient after patient. Because that was the easy way to do it.
Sterling: It's easier to read too. We're finding that our doctors are actually reading our evaluations, which is unheard of. Typically they see all these pages and pages of data and they just sign off on it or just, just breeze over it.
But you, it's easier to read and understand.
Marla: Great. Great. Perfect. And I've got a couple of good questions in here that I'll take 'cause it's relevant to what we're talking about. They, we have anonymous saying, I can see how AI will help with evaluations, but can you expand on what AI looks like for your daily treatments?
Are you recording all your interactions and what type of device are you using to record. Yeah, so we have a combination of iPads and MacBook errors, and basically when we pull the patient back from the lobby area, it's already on and recording. So if we're asking them how, how are they doing today?
Sterling: What's their pain level? How are they, have they been compliant with their home exercise program? We're getting all of that data from step, from them coming through the door. And it's actually recording into their system. Now, if you have two different patients, what you can do is you can open up, prompt in another window and you can start it there, but you have to go back and forth so it doesn't.
Record on with the wrong person, but it's definitely doable so you can see two patients at the same time. But it definitely helps to create a more personalized daily note as well. So the same things that it's doing for the eval system for the daily notes.
Amanda: And I'll add that we have some clinicians that will use it across all visit types, and then we have others that they really just find that significant value in their initial evaluations or reevaluations, or even a discharge visit.
And so they're not so much using it during the daily note, but like Sterling said, if they do, they may turn it on as they're going to welcome the patient back and get that initial. Report from the patient and then they'll turn it off and then they can turn it back on later when they get into another part that they just want to be documented for them.
It's if you had an in-person scribe, and you turned and you said, Hey, write this down. You just push the record button at that point.
Marla: And Amanda, are you using Impro where under each section there is a part where it'll write it for you, that AI piece there, so that if you've got that data, it does it.
You don't even need the prediction, health scribing. You actually can utilize that and tell us a little bit about that.
Amanda: Absolutely we've used that part. When it's summarizing the range of motion, the strength, those things for a very long time. And I will say it absolutely makes our documentation, I think, more compliant.
And so we like that feature as well. I. Great. And I've got another one, Rashan, I'll punt this to you. It's saying they've seen AI hallucinations with Google searching. And sometimes obviously that can cause the wrong information. So what are we doing, especially with our AI intake and our AI case summaries to safeguard, to catch those potentially dangerous hallucinations in our own tools.
Roshan: Yeah. That's one of the biggest concerns using large language models in our software, right? A lot of it is these models, they very much seem like they're doing reasoning, they're doing and thinking through things. But we also have to remember really what they're doing is predicting the most likely next word, based on everything they've seen and we're trained on.
So the important thing for us is making sure that when we take our inputs and turn them into very controlled prompts. That have all the context that they need in the right format, and then we're actually putting on additional guardrails to go through and see, hey, what are the claims this is making?
Are these in fact, based on this context? And then we also put a lot of thought into the design around those features. To make sure that they're not gonna give hallucinations in a potentially dangerous medical context. We want to make sure that if there's a risk of hallucination, it's not a dangerous or critical function, right?
And as it gets towards, hey, this has a high risk, if there's a hallucination, then that use case we're gonna make sure doesn't happen. Live or super fast, we also to take the extra time and put it through more revision and guardrails to check for accuracy.
Marla: Great. And how often are we doing that?
Roshan: All the time. So right now we have some live use cases with AI generated intake questionnaires for patients to try and get objective metrics around their level of function prior to and currently. Around their IADLs. And so when we do that has to be in front of a patient.
It's pretty quick. We put it through guardrails to make sure it's not asking patients to do tests that should be supervised by clinicians. We put it through guardrails to make sure it's not I. Saying anything inappropriate and make sure it's everything's PC or sensitive. And then when we get to evaluation, we're actually using that intake information to write documentation, and those are less time sensitive and we put 'em through more guardrails, making sure, hey, does this, in fact, is this based on the patient's intake?
Is it adding information that shouldn't be from the patient's intake even then, right? Those aren't a hundred percent, and we're still working to make those better and better as we go and as we get into more critical use cases, such as suggesting diagnosis, suggesting treatment options, right? As we get into those, we need to make sure that those are higher quality and aren't released until we have those fully evaluated.
Amanda: And I'll say, I'm very appreciative that y'all are doing that to pre, to prevent these, false statements from popping up in our notes as suggestions. There are about 40% I think of our viewers today that are using ai in their practice daily. And I'm sure you know if you're one of those people you understand.
It can give you things that are absolutely not correct or just make no sense. And so I really am appreciative of prompt putting in those guard guardrails. And going back to what Rashan said earlier, where you don't have to keep saying, no, that's not right. No, do it again. Those sorts of things.
So continuing to make us efficient as well.
Sterling: I'll just say too, just really just proud of what Prompt is doing in regards to setting up those guard well rails. I agree with Amanda. One thing that we have to understand is that ai, again, the whole overarching theme of this is not going to take over your jobs.
You still need to review your notes. You still need to make sure the information is accurate. You still need to make those. Intellectual decisions in diagnosing. This is only an assistant tool for you. So if you're thinking that you can just do all of your notes to be as scribed and it's gonna come out 100%, that's not the case.
There is one question here I see everyone's asking over and over again in regards to the objective. Portion of the evaluation. And how do you document that? Through Prediction Health. And you're really just talking through your objective portion of the evaluation. So if you measure and the shoulder is 120 degrees of flexion on the right side, that's exactly what you're saying as you're measuring And we just tell our patients, Hey, we're.
We have a scribe and so we're just talking through the evaluation. So as long as you do that, all the information ends up in the objective portion of the eval.
Amanda: And I think that's helpful too for the patients to hear that because prior to a tool like this, we maybe weren't calling that out so much and even discussing it at all.
And so I think that some of the patients find value in even just being able to hear those parts of our examination.
Roshan: And I actually think that brings up a really good point, Sterling, that you made that I kind of wanna raise is that for all of these AI tools a critical component is making sure people are trained with them right there.
There is a certain level of AI literacy that's needed and that really shouldn't be skipped by any users or any companies to just learn how to get the best out of your tools and not think that, hey, this is a magic bullet. No, it's. It's part of my workflow. I need to use it responsibly and I need to use it right in order to make sure that I'm getting the best use out of it, but also to make sure, you know my documentation's accurate, to make sure my patients aren't left out of the loop.
Great.
Marla: Yeah, exactly. And I think this is all what people are concerned about or fear, so it's nice to hear from all of you in terms of what we're doing on the technology side, Amanda and Sterling, how you guys are using it in your clinic. And I'd love to hear a little bit more about where you're using it and where you want it to go.
Like what do you, what's next for ai? And. Amanda, I'm gonna have you answer that first about, tell us all the areas you're using it in right now, and then where you wanna see it go to.
Amanda: Gosh, it's hard. I don't know if I'll be able to remember all of the areas because we really are using it as many places as we can.
So one project that we looked at was. Just thinking about Sterling said earlier, our tone, our communication with patients even looking at the psychology behind what makes a patient more likely to pay their outstanding balance. So I used chat GPT to review some expert resources on that.
And to help us change the wording that we're including on our patient statements, in the little box that you can type something in. And I would say that we definitely did get some results based off of that. There is some science and definite psychology that you could apply there. And I am not a psychologist, so you know, I was able to get ChatGPT to help me out with that part.
So that was nice. But another thing that I really found beneficial is being able to. Have chat GBT again, with the de-identified information, but review all of our NPS comments. And so it can get to be a lot whenever you're looking at these really large spreadsheets. But chat, GBT is able to synthesize that information and present it to me in an organized fashion where we can actually take action on some of these things that that we need to improve for our patient experience.
Trying to think. Amanda, are you using the NPS comment through prompt where they reply to the NPS comments as well with ai?
Yes. Yes, we do. So when we get a Google review I just taught my clinicians this actually not too long ago. But I told 'em like, you can just. Pop it in and say, say, give me an appropriate response to this Google Review.
And it makes it so fast. Of course, they're gonna review it, make sure that it's genuine. They can even say, I'd like to say this, and it can just make their thoughts coherent. And with the right tone, you know that it's received in the way that you are wanting it to be received.
Marla: Great. And Sterling, what about you?
Some of the areas, maybe Amanda hasn't mentioned, any additional ones where you're utilizing it?
Sterling: Yeah. We're using it to help with writing blogs. So for our website you can do that. It can help with writing monthly newsletters. There's of course, I think Amanda mentioned this earlier, but we've had success with writing news, or excuse me, request for increase. Increase in our reimbursement. So letters to insurance companies asking for higher reimbursement rates. You would be amazed how well a letter can be written. Looking at all of the data that you have in regards to your outcomes and all that kind of good stuff. Putting in a cohesive letter explaining why you need higher reimbursement rates, which is important for everybody.
Policies and procedures. Believe it or not, we can take our policies and procedures. Put it in and ask for them to look for things that were missing in our employee manuals. How can we make this better, more efficient? The more you use ai, the more ways you find of utilizing it.
So it's just, it's a really helpful tool. I just advise everyone just to start using it. Just get out there and start using chat. GPT-4 0.0 is really a really good prompt. Has. So many tools and just get used to the technology. This is the new industry. We had the industrial Revolution that happened I think in the sixties we're now, and then we had, I think the what is it?
The website internet industry. Industry. Now we have AI technology and it is time. We can't avoid using it because it's the wave of the future. So the best thing to do is get out there and utilize it.
Marla: Great. And Roshan, for you, what is the future of AI? What are we doing here right now?
I know you mentioned a little bit earlier about the CPGs and the evidence-based research, but what else can you share with everyone of where it's headed?
Roshan: Yeah. I think Amanda and Sterling really covered like language models and where they're taking us. It's definitely a huge jump in accessibility to AI for most everyone.
But we also can't forget that legacy AI still exists and it's still very useful. And that's when, that's what comes from integrating directly with EMR software and things like that, is that we can leverage things like machine learning models to help. Automate our CM workflows by automatically categorizing and correcting claims.
We can automate prioritization of claims and which should be dealt with first, which are gonna give you the highest returns for your time. We're looking at using AI to actually guide business growth and analytics. And predict, Hey, should I hire a new PT or should I open a new clinic?
Or is this location good for a new clinic? Is this town growing? There's actually a website you can go to to take a look at projections for building height in certain cities. And it shows you a heat map. And actually it's a good indication of whether or not real estate value is gonna go up there.
And one of my favorites, one of the ones I'm really looking forward to is from a paper I recently read that they did this in psychology looking at only two disorders for now. It was depression and dementia. And they picked like six symptoms that they could actually quantify with various tests.
And they trained an AI model to actually model a patient's progression through all those symptoms with treatments. So they could actually model, Hey, this patient's gonna go in this direction if this treatment's applied at this point, this is how that direction's gonna change in that symptom space. And what I'm really looking forward to is seeing that type of technology develop out and expand to things like PT or to general healthcare, right?
To be able to look at a patient's progression in their symptom space with respect to different treatments and helping providers know at the beginning of their treatment plan Hey, you know what, if we do this for six weeks and then we do this for six weeks, this patient's gonna be better off than if we started off with B.
Right? Just from the get go. And I think that's probably one of the biggest impacting uses of AI that we can get to now. But outside of that, we're still looking at AI for IVR navigation and smart holding on phone calls, AI for fully automated calls for benefits verification. And. So much more.
Honestly, our roadmap is enormous. Actually. I would be really interested and I think that Prompt is focused with this mindset. Is any of these just repeated? I. Kind of meaning meaningless tasks that we have they have to be done. But they're just repeated over and over. So if you can find a way to solve those types of issues and, being able to apply certain rules in certain circumstances, I know that's how you go about it and listen to us and what are our problems, what are issues?
Amanda: So I think that it's great that y'all listen to your users. And are constantly seeking ways that you can improve the product. And I, for one, cannot wait. Every time there's an update, I know there's gonna be something better to help us out and make it easier for us to do our jobs. And I was gonna make one plug for use for AI and that would be for advocacy.
It can give you talking points when speaking to a legislator to help advance them. The problem areas that we face in this profession. And so I would encourage you to use it for that. And I had one other idea that came to mind that I find very useful is when we do have a patient complaint and the patient care advocate or office manager or myself even needs to call the patient and address the issue.
It can lay out fantastic conversational, natural talking points to make sure that I don't forget any part of something that's important that needs to be discussed.
Marla: Yeah Amanda, so many great tips and tricks that you've given our users today. Our attendees. Thank you so much.
I appreciate it. 'cause these are the little ways we always are only seeing one, which a lot is that Scribe AI people do know about that. But I feel like through today's conversation, we've heard from all of you from different areas and Rohan about what's coming, what's next? Being to able to look at a patient's progress over time and giving them that insight, if they follow their plan of care and how that expectation for healing to occur really helps with the buy-in and Sterling, all of the insight you've given of what you're doing from a training and education perspective.
So you're gonna see a little popup on the screen that is gonna say, if you wanna learn more about. Prompt about Sterling Physical Therapy or about Brewer Physical Therapy. Please put that in 'cause we're all here as a resource to you. So I would love to make sure that if you do wanna learn more and we didn't answer your question, we're gonna go through a few of 'em now.
But we will send out if you put it in the q and a with your name, we will send out an answer to your question follow up if we didn't get to it today. So thank you all for that. And now I'm just gonna look through a few of these and see if we can get some of these questions in here. So we have anybody using AI to help with billing or AR spreadsheets, data, anything like that.
Amanda: So I would say we, we definitely are gotta get rid of these windows real quick. Yes, we definitely are. So two other things I thought of were one, helping to write. Appeals or denials that you've received. It's very effective. And the insurance companies are actually using AI probably to deny some of these things that we have to appeal.
We need to be able to use the tool as well. And then additionally might have left me can't remember the other one about, we talked about, yes, you can have it analyze your claim denials. And so even not just responding to a denial, but where are your trouble issues? Is it with, is it with patient data and entry into the system? Are we having the wrong member id? Do we have the wrong insurance loaded as primary? Those sorts of things. So you can, if members, member's not found, I'm just thinking back on some of the things that I've had it analyzed for us, and then that helps us go back to the training with our front desk and insurance verification coordinator to know where do we need to make changes at to prevent these denials from happening in the future.
Marla: And Rashan, can you give a little update of what we're doing on that AR and RCM part with AI and denials?
Roshan: Yeah. I've just been put through RCM Bootcamp for the past month. So we're planning out a pretty big roadmap with RCN but we're looking at things like. First off, just routing claims better so that they get quickly resolved.
I've been seeing the majority of claims and remittance things that come in, just need some simple actions done and can be completed. So we're looking at fully automating things like that. We're also looking at handling some of the harder pieces, both at the source and at the RCM level, such as.
Correct. CPT codes, correct. Treatment notes to comply with those CPT codes. Making sure your DX codes and your modifiers are accurate, right? We're looking to see first if we can get it so that those are set at checkout. Properly. But then also when they come back to our CM to actually look over why a claim was denied, see if those need correction and suggest those corrections or to say, Hey, no, just resubmit this and go forward.
Or, suggest what needs to be made before resubmission or appeals. We're also looking at automation of compiling those appeals packets so that you can send 'em off quicker. Unfortunately we've been seeing a rise in insurance payers. Just denying some things that really should have been, shouldn't have been.
So really it's just a matter of getting that appeal in. And then further down the road, it's also adding on analytics and predictions to your RCM workflows to look at, what you've been losing, what could be, regained, what could be prevented in the future. Looking at trends for how to change activities in your practice to increase reimbursement.
Marla: Yep. It's fighting AI with their AI. Just as you said, Amanda, the insurance companies are using it to deny, we have to make sure we're utilizing it in the right way too to combat that.
Amanda: And then clinicians can use it as well to help them prepare for maybe a peer to peer review that they have coming up with and be able to.
Communicate those things that AI can, source and tell you what they are going to be expecting to find. And you can say, help me present this. In a way, it would give me the outcome that I'd like.
Roshan: Yeah, I'll give some advice to everyone when using it, especially the degenerative AI or the chat-based tools.
One of their biggest strengths is taking, here's all the info in my head, put it in the format I need for this audience. If you can phrase it like that it's gonna be wonderful every time. It saves so much time that I used to spend formatting.
Marla: Tons of use cases. Thank you guys.
Really appreciate it. I'll do a quick wrap up. Really, AI isn't here to replace PTs, it's here to empower them. You don't need a tech background to get started. Just a willingness to try something new. And personalization is actually enhanced through AI driven tools that help us better understand patient needs and progress.
And if you find the right partner. We're here and a lot of other people are here to help you get through it and to make your lives easier so you can focus more on the patient and patient care. Really appreciate all of our speakers and guests today. Thank you guys so much for spending your time with us and for bringing up this topic and being able to all answer each other's questions.