Podcast

The Future of AI in Medical Documentation with Walter Yuan

About the Episode

"AI can become every clinician's co-pilot." -Walter Yuan, CEO and Co-Founder of Empathia AI


Imagine a world where artificial intelligence isn’t just a tool but a trusted co-pilot in every clinician's day. In this episode, we explore how AI is reshaping healthcare by enhancing physician workflows, improving the quality of documentation, and paving the way for a more efficient medical practice. We dive into the nitty-gritty of AI hallucinations in medical records, the steps needed to build trust between AI systems and healthcare providers, and how AI might transform the future of patient care. Whether you're a medical professional or simply curious about the role of AI in healthcare, join us for a thought-provoking discussion on this powerful partnership redefining modern medicine.

Objectives and Discussions

03:02 - The Impact of AI on Physician Workflows
Exploring how AI is transforming daily tasks, allowing physicians to focus more on patient care.

05:52 - Enhancing Patient Interaction and Documentation Quality
Improving patient communication and record accuracy through AI-enhanced documentation tools.

09:00 - Navigating EMR Integration and Documentation Control
Strategies for seamless integration of AI into Electronic Medical Records for better data management.

12:04 - Improving Note Quality and Metrics
Using AI to enhance clinical note quality, ensuring accurate data and better healthcare metrics.

14:54 - Addressing Hallucination in AI Documentation
Tackling inaccuracies and "hallucinations" in AI documentation for trustworthy medical records.

17:59 - Customizing Templates and User Interaction
Adapting AI templates to user needs for more efficient, personalized documentation workflows.

20:57 - The Future of AI in Medical Documentation
Predictions on AI's evolving role in creating precise and streamlined medical records.

30:23 - The Evolution of AI in Healthcare
How AI advancements are shaping the future of healthcare with better analytics and insights.

34:10 - Integrating AI with Medical Reasoning
Exploring the synergy between AI and physician expertise for more accurate medical decisions.

36:08 - The Role of AI as a Co-Pilot
AI as a support tool, helping physicians make informed decisions while maintaining clinical autonomy.

39:05 - The Future of AI in Patient Interactions
Looking ahead at AI's potential to enhance patient engagement and relationship-building.

44:28 - Navigating Competition in AI Healthcare Solutions
Insights into how AI providers are innovating to stand out in the competitive healthcare landscape.

52:51 - Building Trust and Consent in AI Usage
Establishing ethical guidelines for AI in healthcare to maintain patient trust and consent.

More Resources:

Empathia AI: https://ca.cherry.health/partnerships/empathia

Podcast: https://podcasters.spotify.com/pod/show/leaders-in-healthcare

LinkedIn: https://cherryhealth.co/linkedin

Facebook: https://www.facebook.com/cherryhealthinc/

Instagram: @cherry.health

Twitter: @cherryhealthinc

Do you have a topic or speaker you would like considered for the Leaders in Healthcare podcast? Suggest a speaker to alitta.tait@cherry.health

Transcript:

Dr Jordan Vollrath (00:01.805)
All right, today we're back talking with my friend Walter Yuan. He's a serial entrepreneur and the CEO of Empathia AI, one of the leading medical AI scribe tools. Walter, thanks for joining me again today.

Walter Yuan, Empathia AI (00:14.712)
Hey Jordan, it's so great to see you again. Thank you for inviting me back.

Dr Jordan Vollrath (00:19.983)
Always a pleasure.

Walter Yuan, Empathia AI (00:22.171)
okay, Can I start with a quick field story?

Dr Jordan Vollrath (00:27.557)
Hit me what do you got, what's new in the world of AI and Empathia.

Walter Yuan, Empathia AI (00:32.383)
So Jordan, A few episodes ago, you interviewed Dr. Amjad Kaheed Saleh or Dr. AJ as he goes by in the founder of another AI scribe startup, Pippin. And so I met Dr. AJ and Mary at the Ontario MD Digital Health Conference.

about a month ago. so as I walked by their booth, I thought, no, not another AI scribe startup. I know. Every AI scribe company was there. so something Ontario MD definitely did right. Mary was so friendly and started telling me how Dr. AJ began his journey with Pippin.

Dr Jordan Vollrath (01:04.099)
It is hot, there is so much activity in this space.

Walter Yuan, Empathia AI (01:25.208)
I kept on wondering why the name Pippin resonated with me so much. So I turned around and asked Dr. AJ, is Pippin that Pippin? He smiled and said, yes, you're the first to guess it. And so to think about it, For those who are from my generation and who remember Michael Jordan, his fairness, and his reign over basketball, winning six championships with Chicago Bulls,

And we know he couldn't have done it without his ultimate psychic or the co-pilot and to use more contemporary AI term, Scottie Pippen. So if AI health tech, including Empathia AI, and do it right, AI can become every clinicians, everyday co-pilot and really help clinicians to enjoy practicing medicine again. So I thought it was a good story to share with you.

Dr Jordan Vollrath (02:23.418)
definitely. We don't have basketball here in the prairies where basketball is. So yeah, that was, reference went way over my head.

Walter Yuan, Empathia AI (02:28.352)
All right.

Dr Jordan Vollrath (02:35.153)
But yeah, I mean, it was on a panel talk last night with some other family physicians and an ER doctor and most of them were saying, yes, they're using the AI tools on a daily basis. Just the amount of time it's been saving them. Like they were saying like, okay, regardless of like the stats and the studies and sort of the claims, he was like, yes, it definitely saves me at least an hour a day. And this type of like,

is just becoming more and more commonplace. Like again, I use it literally every single day when I'm doing my virtual care practice. Yeah, it's just, I don't know how people would be not using them at this point, other than I don't have a good reason there, fill in the blank. Like I don't know why somebody would not be using a tool like these at this point. It just doesn't make any sense.

Walter Yuan, Empathia AI (03:07.938)
There's no going back.

Walter Yuan, Empathia AI (03:27.16)
I absolutely agree. And when you, when you think about it, you know, how did we get here? Right. You know, physicians spend on average, you know, five hours a day documenting everything from medical notes.

Dr Jordan Vollrath (03:30.137)
What are you guys? Don't go ahead. What are we saying?

Walter Yuan, Empathia AI (03:45.388)
to patient handouts, to referral council letters, know, half of the physicians, you know, from the states and to report a burnout and it's costing billions of dollars, $4.6 billion annually due to turn and reduced work. It's a very serious problem. And so hopefully AI technology and AI scribe, AI for documentation and can play a major role in changing that.

Dr Jordan Vollrath (04:14.101)
What are you hearing from your clients? This was one of the questions the med students were asking us last night. They said, what's been like the best pieces of feedback we'd heard about our different companies? What are you guys actually hearing from the users of Empathia?

Walter Yuan, Empathia AI (04:29.26)
Right. think, you know, now that we think about AI scribe,

or AI for documentation in general has gone mainstream for probably two years. I mean, there are earlier attempts, but I think large language models, the OMS have fundamentally, or generative AI has fundamentally pushed AI scribe, AI for documentation to mainstream. so it's probably a good time to reflect on what AI scribe has done for physicians or clinicians in general so far. I think there are lot of positives, as you mentioned,

There's tremendous amount of time saved. But when you look deeper, that time saved actually come from two angles. One is intuitive, one is not. The intuitive angle is the reduction on documentation time. You don't have to spend so many hours per day catching up on those. have worked with, personally worked with so many doctors who have been in their clinic till evening, sometimes midnight or weekends.

just catching up on their charts. But now they're by 430 or 5. And that's tremendous life transformation, and to allow them to really enjoy their work and have a better...

at the same time have a better work-life balance. But the other dimension that often overlooked is actually, and it has shown in several trials and studies and in some, and pathogen were involved, was involved, is the amount of time spent on per encounter. It turned out that you also don't have to talk to patients as much while having a better quality of interaction with patients. Because think about it intuitively, you don't have to type at the same time anymore. And so,

Dr Jordan Vollrath (06:21.895)
I know the engagement with the patient. Now you're paying attention as opposed to, and I'm even pretty handy at typing. Like I can get pretty reasonable typing without looking, but even then it's still very distracting because you're like, trying to hit the backspace. And yeah, it lets you actually like be way more present.

Walter Yuan, Empathia AI (06:34.754)
Now, very distracting.

Walter Yuan, Empathia AI (06:40.844)
Yeah, yeah, and then You often have to pause and to think about what to type and correct some of spellings. Now you can be much more focused, more empathetic, and yeah, having a higher quality and shorter consultation. So time saving can be from multiple angles. And then there's also, as we mentioned, improved patient.

physician interaction and physicians now can see that you're focusing on them, right? That's improved patient care. And then going down the list, reduced cognitive load, right? Just by having, by knowing that there's a technology there listening and generating medical journal for you, you can always go back to minimally by looking at the transcript. You already have that peace of mind. You don't have to worry as much. And I think that cognitive load reduction is also pretty

significant. And the surprising finding that, at least to me, is some physicians, you know, we get asked a lot about integrating with EMR EHR systems, but some physicians reported that they actually enjoy having the independence of reviewing, you know, the AI generated notes outside of EMR system. Because that approach, I think, offers this ease of review and editing and it reduces risk of unnecessary

or unwanted log trails within the MR and finally, it gave them full control over their documentation. So there, I think these are all the positive things that early adopters of AI Scribe have reported or shared with us.

Dr Jordan Vollrath (08:22.033)
It's interesting, just the kind of the logistics of the workflow of like what gets entered into the note. And I know I've like had a few times where like the med student has gone and started entering the note and I'm like, okay, editing this later, like, whoa, whoa, we don't, I don't want that written that way as part of this like legal record for, know, I don't know if they just say something kind of like off color or if it's something that like makes the diagnosis seem different or like,

that the patient's in danger or something like that. And then the EMR is often just like auto-save things, right, as they go too, which makes it kind of tricky. You're like, okay, if everything is just going straight in there, I don't actually have full control over how I want it to be looking like exactly as I would phrase it for my documentation, for my like personal protection in terms of like medical legal risk. And so having it outside of the EMR initially,

Walter Yuan, Empathia AI (09:00.984)
Yeah.

Dr Jordan Vollrath (09:20.316)
I think definitely is a perk.

Walter Yuan, Empathia AI (09:22.188)
Yeah, absolutely. as you continue to build up that trust, I think having that opportunity to allow physicians to easily edit and adjust their note and before they copy into the EMR, I think can be viewed as a side benefit instead of as the initial constraint that, here's your system, integrate with my EMR.

I think that may change more and more as AI scribe, AI for documentation going mainstream.

Dr Jordan Vollrath (09:55.857)
And then, can I have one second? I just gotta go kick my dog away. She keeps knocking on the door and like, you're probably hearing this and I'm like having to mark the clip every 30 seconds here. Give me one sec.

Walter Yuan, Empathia AI (10:06.628)
Yep, go for it.

Dr Jordan Vollrath (10:38.371)
Usually she's a very respectful knocker. She'll give it like one or two knocks and then if there's no answer, she'll just leave it alone and I'll come get her later. But she's like hell bent on getting in here right now. I apologize.

Walter Yuan, Empathia AI (10:50.034)
Maybe my voice was very attractive to her. You know, the background color, you are so, you always strike me as someone so meticulous about everything. Look at the background color, how your side, the red and it just, it's yeah, it's very cohesive and it's stylish. Yeah, man. You know what you're doing.

Dr Jordan Vollrath (10:54.257)
call it a rid.

Dr Jordan Vollrath (11:12.027)
Thank you, I appreciate that.

So Tell me a little bit about how you guys are actually seeing things progress in terms of the note quality. How has that evolved over the months and over the years from when you started and How do you even assess note quality?

Walter Yuan, Empathia AI (11:31.148)
Yeah, that's, you know, when we first launched Empathia now two years ago, and one thing that

internally, we have all recognized is that until we have incredible note quality, we're not going to be able to win physicians and clinicians trust easily. But it's like you said, it's an incremental process. Initially, we focused on what we called or what I call direct metrics.

And these are metrics such as information accuracy and information capture. Are we capturing all the necessary or medically necessary information and information placement? And it's not just about accuracy. It's not just about completeness, but also about placing the right information in the right sections, right? And also reduce hallucination and not putting content

in your note where there's not evidence that the content was being mentioned during the encounter. So that's sort of the first stage. I think, you know, Empathia probably has just moved past that stage, and some of our peers are probably still in that stage. Now we're on to the second, I call focusing on more the meta metrics. So these are...

Are we translating the notes or the patient after visit handout in medical terms? For physicians, and they look at their notes, they feel this is medically rigorous. And for patients, they feel like, this handout is inclusive. I can understand. And this handout is useful for me to understand my health condition and the potential personal risk for the treatment plan I've been prescribed. And so that's medical translation, consistency. Like you said,

Walter Yuan, Empathia AI (13:29.352)
AI models are going to change. Sometimes we're not even notified as the application developers, the underlying AI models have evolved.

And so how we counter that so that we always maintain that node consistency so that we don't run into situations, you know, this week, you know, the nodes look great and next week somehow the nodes got caught code or some virus and then the code quality suffers so that we always have that consistency. But in order to have that consistency, you really have to have a really rigorous and automated evaluation framework so that you periodically run and see if you have regression tests to say,

OK, are the nodes being generated today have covered all the direct metrics? And so that's another thing. And then some of these other meta-metrics now we look at is, can we capture all the multilingual encounters?

not the multilingual where, for example, if you and I both can speak French, that's one coverage. But what if we don't have a common language? You can only speak French, and I can only speak English. And can we cover the mix and multilingual encounter? Traditionally, that has to involve a medical interpreter. And now we can develop an AI medical interpreter where a patient just scan the QR code and then have a conversation with their

healthcare provider and in real time. And then last but not least, we touched on this in our past conversations, can we have a seamless support on teleconsultation? Can we still capture the information in that virtual environment? And so these are some of the meta metrics that we focus on. So in the end, think having node quality is gonna be a direct predictor on the amount of time we can save for physicians.

Dr Jordan Vollrath (15:22.907)
Very cool. I mean, like for myself, was even years ago when some of this technology was still in its infancy. The note quality was like well above my personal baseline, but I guess it depends on you, the individual doctor and how verbose they are and how well they did in poetry class back in the day. But yeah, think the quality of the notes when I'm using the AI tools is just drastically higher than...

when I'm plunking away at it on my own. It's interesting that now we're starting to move into like the application of the initial note quality now to like the patient handouts and into like actually being able to translate things into an encounter. It's really cool to see things going down like kind of that next generation of adding on to the baseline function.

Walter Yuan, Empathia AI (16:07.958)
Absolutely, you have to expand. mean when you build a product you want to have sort of an anchor feature

That anchor feature defines the value proposition for probably the most salient pain point for target customers. I think arguably in our case, AI scribe documentation is that pain point. But you can't just be complacent by staying at that pain point. You have to go upstream to say, OK, can we move up so that we can cover more the friend desk chores from patient intake, from effects, digest,

from compiling an accurate patient past medical history and both relevant for the particular specialist and to the middle office where the anchor features are the AI scribe, the documentation, and the language interpreter, and all the way to the downstream back office where we can do coding and billing. And coding and billing may not be as significant

in Canada but it's very important for our American physicians.

Dr Jordan Vollrath (17:20.497)
Can we back up as well? Like, are you able to tell me more about the hallucination side of things? Cause this is like the number one quoted downside to these types of tools. And personally, like while I've been using them, honestly, I don't think I've ever caught it hallucinating. Like with the template that I've been using, there's often like errors of omission and things that it's missing out. like how often, for example, in Empathia, like how often does it...

try to hallucinate, how do you actually mitigate it doing so? And how often, like what are the actual rates of some wild hallucination actually making it through to the end note then?

Walter Yuan, Empathia AI (17:59.884)
Right, right. I think it all depends on how you want to use the underlying generative AI model. And if you want to turn the temperature up,

to give the model more freedom to help you, it sometimes can go overboard. You end up with hallucination. And then on the other extreme, you try to constrain the model to say, only do what I exactly tell you to do. So for example, by supplying a template and also build a system, say, hey, only include content that is specified in the template and don't include anything else, even though that content may be medically irrelevant.

the balance is somewhere in the middle.

You have the template. You can direct how Empathias AI model to generate your notes so that it contains your personal preferences. It contains a language that match your domain expertise. And it has the necessary checks. And then Empathias model just fill in the blank. But then also allow us to have a little bit of freedom so that in case something happens during that conversation, maybe in the form of we have macro supports, right?

few macros and you know for family medicine very rarely a patient comes in and I mean they take it takes weeks to weeks to make an appointment to see you and sometimes they bring up additional issues and so that's why it makes you know covering family medicine so different from you know covering specialties and oftentimes you know you have to add additional information because a single template may not be able to cover a single encounter and so I think where you have the right

Walter Yuan, Empathia AI (19:48.058)
temperature so that you allow Empathia, for example, to help you. And then at the same time, so we minimize hallucination that we don't include things that you didn't say or things that just are irrelevant. I think it's a constant, it's a continuing improvement. And I think, again, this is where having a solid automated evaluation framework is so important. It's so tough to do because it's different from traditional software development where I can just write hundreds of test cases

and say, hey, if your program spills out, it's correct. If it's one, it's wrong. But in this generative AI age, you really have to fundamentally understand the semantics of an output and to say, hey, this output actually looks good. And or if this output has hallucination. So by having that evaluation framework, I think would move not just on pathia, I think the whole field forward.

tremendously and then at the same time combat hallucination while letting physicians enjoy that AI's power to help you to do some kind of medical reasoning or additional decision support that AI is capable of doing now.

Dr Jordan Vollrath (20:57.795)
And how does a person actually change this? like there's the template that you set up and like you kind of tell it. But I think of a template more like you'd use a macro and an EMR, like what are the headings in different things? But because it's this big free text input and because it's like a large language model, you can like give it directions in like normal language. Like how should you actually direct your template? What are the different things that you should actually be including in there beyond that?

Like what are the headings and the subheadings for the categories of information I want to be using.

Walter Yuan, Empathia AI (21:33.774)
So, Empathia over the past months has now offered physicians a couple of different ways for you to interact with our underlying model. So, one way, obviously, is through template. And I think template and macros. And I think there is an art on how you use, let's just be more specific, templates for the moment.

So I have seen physicians on Empathia writing templates that are so detail oriented. So basically they say, I'm going to try to cover everything I can think of.

some potential present issues. then just use my template as verbatim as possible, just plug in these parameters on the fly. That's all you need to do. So that's one extreme. And then on the other extreme, I've also seen physicians on our platform write very skeletal templates. Basically say, hey, I'm happy, I just want these sections. And then here I stay what these sections for. And so basically have a simple statement include, you know, this kind of content.

in this section. So it's a very simple half-page template. think having these two extremes would allow us to serve physicians differently. Some prefer to have more control over their medical notes, and some are trusting our AI model more. And so that's one way to interact with our model. Now we also provide two different editors.

One is a traditional editor where you can type, you can dictate. And we also offer some shortcuts, autocomplete. We have the entire Canadian medication database in our system. So you can type a letter. It brings up some medication names and then the most current medication name, the CPT names. And then so you can just hit Enter and enter these terms automatically. so that's one way to interact with our system, the traditional editor. And the other editor we call a smart editor, where you can type in sort of a chat

Dr Jordan Vollrath (23:20.679)
cool.

Walter Yuan, Empathia AI (23:40.718)
GPT-ish prompt. And from the prompt, such as, I want this section to be point form instead of narrative form, or I want, I think this information should be part of the past medical history instead of HPI. But what we do behind the scene is, whichever editor you use, whether the traditional editor or the smart editor, we start curating a long-term memory for each of our physician. And so that's where the customization comes

going to come from, and from template and also from your long-term memory. So now when you have a new encounter, and we'll take inputs from several different angles, if you apply a template, we use your template as a direction. If you have long-term memory with us, so through your Addis, through your Smart Addis, we'll apply that long-term memory with the working memory, which is the current encounter you have. And then by combining all these inputs, we generate hopefully the most cohesive and comprehensive.

note for you.

Dr Jordan Vollrath (24:42.289)
How does that work then? That's an interesting, like the fact that it's kind of like tailoring things as you go based on my previous requests. Like how would you recommend somebody use the tool? I guess, like, should you just kind of like test it out with the default settings and see how it goes and start out with making those suggestion edits? Or do you say like, go back to like your template that you start out with, try and get it as close as you can and then go from there. Cause this is one of things I was actually working on.

literally this morning, I was using Scribeberry, forgive me, I apologize. It's got the voiceover IP access, which you were saying. like trying to figure out for my consulting patients, there's just like on the initial intake, these chunks of information that are always pretty much the same. It's like, how many meals do you have a day? How many snacks do you have a day? Like, do you skip meals? And so it's usually just like yes, no questions or like a.

multiple choice input, but then there's often like this kind of richer context that comes from like other details the patient puts in there. And I want to capture those, but I kind of want them like kept separately from those like categorical kind of standard inputs that I can go back and reference later. Like how do most physicians handle that versus kind of the more free form?

Walter Yuan, Empathia AI (26:04.174)
Yeah, I think it also depends on individual preferences. I think precisely because the difference, the patient encounter nature, the difference in patient encounter nature across family medicine to all the...

the different specialties, and where in family medicine you often encounter patients with multiple issues, or with unexpected issues. where when a patient sees a specialist, often both sides understand precisely what the issue is. so given that, I see a lot of family physicians are now really regular template users.

And so they use our system and we encourage them to take advantage of the editors to start constructing their long-term memories with us so that we can quickly customize and personalize the notes for them.

And then for specialists, they also, know, often within their circle, they share templates, and many of them are really template users and active template users. And for them, you know, we would encourage them to start uploading. have several different ways for you to start curating your own template library, and so you can quickly set up your template library and on Pathia and then use that to personalize, to improve your notes. And down the road, I really have this vision

that if we do it right at some point no one needs templates and you just need to construct a really or conduct a really good encounter because that's the root of all data.

Walter Yuan, Empathia AI (27:52.298)
If you have a really good encounter, let's leave it to AI, to our technology, to our innovations, to quickly learn your personal preferences, to discover your practice setting, whether you're clinic or hospital or ER, and what kind of specialty you're covering, family medicine, to psychiatry, to neurology. And once we learn all that and then you make a few edits, we pick up your style and generate your notes for you. So no one needs to spend hours, even days to develop

develop these templates. And then once you switch EMR, all your templates are wasted. And or you have to repeat that process, recreate all these templates. So I think when you think about it, templates are for generating the notes. But in the end, if we have solid encounters, we should be able to, through technology, to personalize for you and at the same time, generate quality notes. It's just one.

Dr Jordan Vollrath (28:43.963)
So yeah, the template itself becoming obsolete, right? I guess this is like the initial phase of all the different companies just figuring out how doctors want to work with the tool. And then it's going to be completely obsolete of even choosing a template. It's just gonna do it right for you.

Walter Yuan, Empathia AI (29:01.226)
And A plea to physicians who listen to this episode, be patient with the empathia or our AI peers. I mean, think about it. When we try to teach our own children, it's a process, right? You don't tell the child once and then to expect that child learns everything. You just told them and be able to just do the work, right? It's a learning process. And internally, our goal, we have a metric. And if a physician

added X number of notes, and we're going to be able to learn that physician style and that physician's domain expertise and this physician's language. so we hope to, you know, next year we hope to lower that benchmark and so that we can learn a little faster. But be patient with AI companies because it's your training the model to write like you. So it's a learning process. It's not going to happen overnight.

Dr Jordan Vollrath (29:57.393)
I hadn't really thought of it that way, but you're so right. Like just how fast the technology is changing. Like I wouldn't expect the new nurse that I work with to have everything dialed in just how I like it after one week or the new MOA. So it's fascinating just thinking about a piece of software from that regard of like it's learning. Give it a moment.

Walter Yuan, Empathia AI (30:07.03)
It's not close.

Walter Yuan, Empathia AI (30:20.035)
Yes, yes.

Dr Jordan Vollrath (30:23.457)
One of, sorry, what was I gonna say? The clinical reasoning side of things. You mentioned that it's like if given some leeway and the freedom to actually start getting involved, what does that look like with empathy? I have not actually been on the tool for a few months now. That sounds new.

Walter Yuan, Empathia AI (30:42.284)
So I think it's really, when you think about the history of AI, use the word history, made it sound like it's been around for a long time. Exactly. I think this whole powerful AI age really warped time for developers.

Dr Jordan Vollrath (30:57.147)
dozens of months.

Walter Yuan, Empathia AI (31:06.658)
potential adopters. mean, it seems like a long time, but really, it's only been a few years. And I think I can safely say that all AI

scribe or documentation companies or startups, even include some of the big players, Microsoft Nuance or Bridge. And we're at the stage where we just focus on capturing the information, right? Making sure we capture everything with things of size, a medically compliant pre-charging note, and then we help reduce physician documentation administrative burden. So we focus on that part. But I think there's more beyond

that and which is what you just mentioned, medical reasoning. Because I have spent a lot of time, even though I'm not a practitioner, I have a strong biology background and I study bioinformatics and so I can read the notes and try to understand what the notes are about. And then when you compare, for example, a note that's written by a physician without help from AI and then a note generated by AI, for example, empathia, they still look different.

they still look different because the note that's written by a really good physician or physician that's really good at writing her note, you read it as a story. And even from the first few sentences on, you get a feeling that the physician is telling a story. It's telling a detective story.

you know, from the patient history taking, I ask these questions. Here are why I ask these questions. And then based on these questions, I derive at these differential diagnosis. And then finally, I hold me on this diagnosis. And then from that diagnosis, I think about these potential treatment plans.

Walter Yuan, Empathia AI (32:58.412)
And then I picked this treatment plan precisely because I take into the patient-specific information, the demographic, the health condition, and then whatever additional patient information, chronic illness, and all that patient-specific information into consideration to arrive at a specific treatment plan or the best possible treatment plan under all these constraints. So the note looks like that.

It's a story, it's a pleasure to read that detective story. I think right now, almost all, if not all, AI strike companies are still at that mechanical information capture. So we're Empathia, our AI group is focusing really on bringing the medical reasoning to the new generation to help conduct, help physicians conduct better encounter to take patient history, patient medical history better.

and then to help with differential diagnosis, and then in the end to personalize treatment plan. If we can do that, we truly become not just an AI scribe agent, but become an intelligent medical agent for every physician. So I think that's where the field hopefully will be moving towards.

Dr Jordan Vollrath (34:10.529)
You mentioned that it's like now it's plugged in with the Canadian drug databases, like the latest in names. How long is it going to be until there's like a, like up to date plugin for the software and it just goes directly to the source for like the diagnostic criteria and the treatment plan and like follow up.

Walter Yuan, Empathia AI (34:29.998)
That will be Jordan. That will be so important because right now we're only taking a baby step. I'm sure there are peers that could do this area better than we do now. But right now we just offer these names, right? But imagine if we can also bring up.

not just the name, the medication name, the instructions, the most up-to-date instructions, the dosage, the frequency, the personal potential, personal risk, and relevant personal risk. I don't want to see a risk, hey, Walter, if you're pregnant, you shouldn't be taking this medication. It's irrelevant. I'm a guy, and I'm already at an age, and it's just completely irrelevant information. So I think that if we not only can bring up-to-date medical knowledge, but also in a way that can really help physicians, I think that

will be a huge value add for our doctors. And quick sharing, I was talking to a rural medicine physician, and not too long ago, she told me that she gets a helicopter in all the time to see a patient. And right there, and she's not in her office, she can't easily go. And she has to make a decision right there. And it's tough.

It's tough. if Ampathea has that medical tragic PT where she can easily query and by having the patient information by default there and conditional on that and have the most relevant treatment plan, instructions or recommendations, I think that kind of decision support would really bring the AI for health care to its next level.

Dr Jordan Vollrath (36:08.817)
How much longer until you replace me with AI? Like how many years is it going to take?

Walter Yuan, Empathia AI (36:13.198)
Ever. Never. And I think patients come to see you, not just for, I know this sounds cliche, but in all seriousness, I think, I mean, I'm a patient to my family doctor.

And even if Empathia can offer me some medical advice, if we decided to go that route at some point, and I still want to see my doctor. And I think having that empathetic interaction.

would allow me to feel more comfortable, confident about my life, and to make sure that whatever the advice I got has been validated by my doctor. And if he is not sure, I'm sure I trust that he would consult other specialists to give me the best possible advice. I think, again, there's a reason even the giants like OpenAI, Microsoft, Google call their AI models co-pilot, not pilot.

And I think, you know, it would only empower clinicians, not displace you.

Dr Jordan Vollrath (37:31.025)
It seems more and more apparent to me that it's like going that direction. Like we talk about co-pilot versus pilot. I mean, you think about an actual plane, like the pilot only does like the landing and the takeoff. Even that can be done by the actual like pilot software now. And so it seems to me like more and more, obviously the AI is not going to be able to suture up a boo boo. It's not going to be able to reset a fracture. It's not going to be able to

palpates your abdomen. But when it comes to just like a talk, like a conversation, which is most of what I do, maybe this is why I feel the most, not threatened, like I will just kind of move over to doing something different as the AI starts getting more and more intelligent. like virtual care is all talking. Like if you could log on to a website and just talk right to Empathia as the patient, like it just seems inevitable that in the coming years,

the technology will get to that point where you don't necessarily have to go straight to the doctor first. You can actually just talk to the website, to the program, and then maybe it directs you and says, okay, no, you gotta actually go get your knee examined or do this or that, or have this rash biopsy. But it seems just like it's going that direction and we're kind of have to shift what we actually focus on to be more procedural or more.

niche work compared to sort of that general conversational type of medicine.

Walter Yuan, Empathia AI (39:05.27)
Yeah, Jordan, maybe I'll share two thoughts quickly. mean, it almost gets into a philosophical discussion, but I think there's something concrete can be said here. I'll share two thoughts really quickly. One is that when you think about the whole patient visit workflow,

And as you mentioned, the most important, the reason you remember you're just talking to patients is the most important component is that encounter, is that consultation. I think if AI has done or AI HealthTech collectively, Empathy included, do its work, we can significantly alleviate all the downstream documentation administrative work. But you didn't sign up for that in the first place. In any case, you sign up to see patients.

There we go. So you really are passionate in talking to your patients, understand their needs, and help provide care for them. And then that encounter, that conversation is so important. I don't think even if AI can adopt your voice, I don't think I have had the same feeling, that same emotion, like what we're doing right here, a podcast.

Dr Jordan Vollrath (40:00.901)
why I became a doctor. I'm passionate about paperwork. I just love doing it.

Walter Yuan, Empathia AI (40:28.086)
Right? Imagine if I'm just talking to an avatar, I probably wouldn't be this engaged at all. Because I know deep down, you may not even care. You may be playing golf somewhere. Right? And so having you there and seeing that I'm getting your attention, I think it means a lot to your patients. Having that good encounter is the foundation of all the downstream documentations. Right? So that AI cannot do yet. And then the second thought is, again,

Dr Jordan Vollrath (40:55.921)
Yet, I like how you add that qualifier on there.

Walter Yuan, Empathia AI (40:57.454)
Yes, well that's the second thought that to mind. So You can say AI scribe only has been around for a couple of years. Within the next few years, it's going to advance so much it's going to replace doctors. But how long has automated driving been around?

They have been around for decades and billions of dollars have been poured into it. But still, they're nowhere near the state where you can just let a car just drive.

why I give you a simple case. I borrowed this from one of my favorite podcasts on economic side. And so an example, the two economists discussed David Otter from MIT and Steve Levitt from University of Chicago. And the example David gave, I thought it was just so inspiring. It's one of the reasons to make auto driving autopilot. And so difficult is, for example, if a cow just falls from the sky and dropped on the highway.

right? And you can argue to the autopilot predictive model, the AI model, gosh, we have never been trained on that. So we don't know what to act, so we crash. But think about how human brains work. When we see that black swan event, a cow dropped from sky and all of a in front of me, I slam the brake. Even though we have never seen that pattern, that's the power of human brain because we generalize.

And we can generalize across these different scenarios. And when it happens, we know how to react. I think it's still going to be decades, from now to have that medical. And as a encouragement, I think it's going to be inevitable that we're going to coexist with AI, with powerful AI or generative AI. But I think it's really far-fetched to say AI is going to replace us.

Dr Jordan Vollrath (42:40.835)
Really, really?

Walter Yuan, Empathia AI (42:59.15)
That's my encouragement. So continue to see your patients, and enjoy practicing, and then leave all the documentation stuff to us.

Dr Jordan Vollrath (43:09.009)
I say, I say replace kind of facetiously, but definitely like taking over a lot of the duties, like a lot of like the, different, I don't know, maybe it's the intake side of things and the initial history gathering or all the way down. Yeah. But so I am very actually excited for, you know, the following years and decades of just seeing how this continues to augment things. But yeah, it will be interesting. I think it's going to happen.

Walter Yuan, Empathia AI (43:22.766)
I'll repudiate

Dr Jordan Vollrath (43:36.337)
I don't even know. It's hard to say. It's hard to say. But I am very excited for the change.

Walter Yuan, Empathia AI (43:41.804)
I want to say it's going to be for a long, time before AI can truly adopt that medical.

reasoning to really emulate human intelligence. And it can do a lot of things. And I think there's a division of work. There are things that AI is really good at, even now. And I think as human beings, we just want to be mindful so that, for example, I have young children and my son is working, my daughter is going to be applying for college. When we have these conversations, I often just ask them to think twice on picking a career where human

plays a major role and that may be the forward-looking that you have to do now. Thanks to AI.

Dr Jordan Vollrath (44:28.657)
So as more and more tools come out, as more and more companies start to exist, how do you actually see the competition playing out? How do the different AI tools start to choose a lane or a niche and start to be, like where does everybody start to occupy their different levels in that ecosystem? I guess is what I'm getting at. Right now it seems like all of the AI tools are kind of like casting the net broadly.

It sort of works great for everybody and it probably does. It probably like no matter which one it would make your workflow a lot better from baseline. But like how do you see things actually starting to settle out into an equilibrium?

Walter Yuan, Empathia AI (45:11.598)
It's a tough question. I wake up every day getting emails from colleagues saying, hey, Walter, have you heard about this company? And the competition is fierce. But it may not be a bad thing for customers when there is a competition. Number one, that means there's a market. Number two, means that the best quality product can emerge at the best possible price.

And so it's going to be a win for customers. But for us as technological startup or tech startups, I think I just want to, before I say something about how Ampathia hope to differentiate from our competition, from our peers, I just want to encourage physicians to think from two angles to help.

understand where we come from as vendors. The first angle is that

Walter Yuan, Empathia AI (46:12.632)
The development of AI-empowered or AI-powered technology is fundamentally different from traditional software development. And traditional software development has the benefit of this well-known economic phenomenon called diminishing cost. Once you invest the initial capital to develop the software and the marginal

cost of maintaining or selling the software is approaching zero. And that's how Microsoft, Google, Facebook all become very, successful. But it's fundamentally different in this AI age software, because every time when we compute something, there's a cost. There's a variable cost.

whether we generate a note or we answer one of your prompts or we create a patient after visit handout, it costs the company. So there is a variable non-diminishing cost associated with the computation. So that's one fundamental difference. The other side is I plead to physicians to look at from value creation perspective. Because when you think about economics, there's this whole field called negotiation.

and it's all based on value creation. So when you think about, you know, a technology, what the technology has done for you is to look at the value created by that technology. If Amhathia or our peers really do the work that save you a couple of hours a day.

And you can actually calculate how much economic value it creates for you over a month. Let's say it saves you 15 hours a month. And then you know your hourly rate. So you can calculate the economic value the AI technologies have created for you. And then a sustainable division of that value sometimes is 50-50. But I think a lot of our startups, like Empathia, we agree that right now it's

Walter Yuan, Empathia AI (48:19.002)
consumer market, we need to compete so we lower the price. So that's where I think free is not going to work because they're just burning through their venture capital money, not their own funding. And to build a sustainable business, you don't want to use a vendor that would burn through. They try to get market share and burn through their VC capital. And then once that capital is exhausted, they can't sustain. So you want to choose a vendor that

that's responsible, that's on track to build a sustainable business. And so with that, differentiations, I see from, I won't name too many, I'll probably just mention two or three. One is the empathy team has been relentlessly focusing on node quality. I know we're not there yet, but I don't think any vendor is there yet. So there's still room for improvement. And even if you cover family medicine well,

there's still all these other specialties, right? And all the nuances, all the really complex conversations, how can you make sure not just, you know, forget about the basic metrics I mentioned. How do we combat information loss when you have a 60-minute long encounters? When you think about how these large language models are being trained, and almost none has been trained on a data set that involves really complex, let alone medical, conversations.

So you're basically telling a child to generate this medically-compliant comprehensive note, even though that child has never been exposed to these conversations before. That's a tall order. So the note quality, think, is really important. Number two, I've been telling the team is we have to expand. We touched on this earlier. We have to expand from AI scribe to covering physician workflow better.

You all have your own preferences. So I've run into physicians who insist on finishing their chart right after patient visit, right for good reason, because there's this phenomenon in cognitive psychology called retroactive interference. So if you ask me, I'm constantly on the road. If you ask me where I'm staying, the room number of my current hotel, I will be able to tell you. But even if I stayed at a hotel just two days ago, I wouldn't be able to tell you because I inferred.

Dr Jordan Vollrath (50:23.877)
you

Dr Jordan Vollrath (50:36.59)
What'd you have for lunch yesterday?

Walter Yuan, Empathia AI (50:37.966)
Exactly, right? So once that patient walks out the door, the next patient walks in, our memory plays that trick on us. We started focusing on that current patient we forget about, the last patient. So a lot of the physicians want to wrap up at least the key points of the previous patient right away. So how do we cover these different workflows based on personal preferences and different settings, practice settings, clinic, hospital, and ER?

And then finally, specialties. So I think this is the second area where Empathia has been really investing in creating this AI native solution where we can just allow physicians to put together any kind of workflow. If we do it well, if we support team-based consultation, we support different sources of information by combining them together, if we have these building blocks and let physicians create their own workflow use cases.

And so that's number two.

Number three, think is really important. I think we touched on this last time, self-learning. You have to build a system that's fundamentally different from the traditional software, which is static. You have to build a system that's self-learning, that's adaptive. So as physicians interact with your system, the more they involve, the more they invest, the more time they realize they started saving. And they realize, my goodness, the notes look more and more like I've written them myself.

And then the last thing I probably want to mention is I am so this is one area I'm proud of the team is we are probably the only pure tech company, but believe in ground level effort. I lead by example, have been spending countless hours by going across North America, knocking on doors, visiting clinics, shadowing doctors and conducting feedback sections and do onboarding.

Walter Yuan, Empathia AI (52:36.784)
And I probably personally have worked well over a thousand physicians in the past year alone just to understand like in this conversation understand what you need how we can serve you better. I think that's important that trust is important.

Dr Jordan Vollrath (52:51.727)
Yeah, just having known you for the last year or so, you are literally always on the road out to a clinic because I don't know if I've ever met anybody who's like out there in the mud with their like end users nearly as much as you have been.

Walter Yuan, Empathia AI (52:57.578)
sorry.

Walter Yuan, Empathia AI (53:06.36)
You know, Jordan, it's when I think about it, many of our peers are lab-baked physicians. We have a large physician team on the team as well, cross-specialties, our chief medical officer, Dr. Steve Sivak, who's a renowned internist and from the States. But I don't carry a medical title. So I don't have a practice or pride to be just in the trenches and working with physicians and be your medical student.

And in fact, if anything, I feel tremendously rewarding when I'm allowed to be in an exam room. And so just last week, I was in West Van and helping an elderly physician on board, Empathia.

And so she asked me to be in her exam room with her patients for four or five times until she feels comfortable with our technology. It's a tremendous learning experience. seeing how patients appreciate now, Dr. Susan can focus on them. And it's memorable.

Dr Jordan Vollrath (54:10.403)
every day is bring your CEO to work with you at the clinic.

Walter Yuan, Empathia AI (54:12.942)
Hey, blue collar.

Dr Jordan Vollrath (54:17.937)
One other thing I've got to ask you about. So the thing that I kind of find annoying right now with AI is that you have to do like an informed consent with your patients before you start using the tool with them. And so how long is that going to be the case for? Cause this seems kind of like just sort of colloquially known as standard practice right now, like new tool, like there's some risks involved. I don't consent to every single patient I use to like,

the EMR that I'm on, there's like general kind of privacy stuff that goes out and they like click on forms. But like in terms of having to like literally talk about the tool and the software with each and every patient before getting going, like how long until it's sort of just a given that the tools are safe and secure and trustworthy and accurate, that we don't need to be doing that.

Walter Yuan, Empathia AI (55:13.358)
I think there are a few angles, all positive. And as someone who has visited hundreds of clinics, health centers, hospitals, and I can name in one hand when a physician mentions that they're going to use technology to capture the clinical conversation and a patient says no. And I would say 99.99 %

times when a patient hears that, you can focus on me now instead of your keyboard. And that emotion is overwhelmingly positive and embracing. So that's one positive sign. And then the second is, there are recent trials, as I mentioned, even including the one that's finished by doctors, BC. And when asked physicians to pick priorities,

in terms of their AI adoption or concerns. And consent and privacy rank absolutely at the bottom. And right now, burnout is such epidemic and such a threat to their practice.

I think having the right technology to free them from that endless documentation administrative task, think, is something that physicians would prioritize over all the regulation and regulatory compliance requirements. And then from the top, even the health ministry of BC recently updated their AI adoption guidelines. I think it's overwhelming.

Dr Jordan Vollrath (56:58.257)
Thank

Walter Yuan, Empathia AI (56:58.572)
encouraging and I think sooner or later we had this discussion in our past conversations and I think it's not gonna be long that we're gonna have sweeping guidelines so that you don't have to go through that that consent step every single encounter

But I know a lot of the clinics, basically, they insert that in the new patient check-in packet. And so once a patient signs, and then all the follow-up visits, you don't have to mention it. So at least that's, I think, from a clinic all the way to health systems, I think the streamline is coming.

Dr Jordan Vollrath (57:26.362)
Mm-hmm.

Dr Jordan Vollrath (57:42.641)
I would echo what you're saying in terms of your experience. The vast majority of patients are like, yep, definitely no problem. And then I don't know, for me, it's maybe like one in 15 or one in 20, the patient doesn't consent to it. And I'm like, damn it, this, can I do this the old fashioned way? And then it's just like, I see where they're coming from in terms of like being a little bit wary around the privacy side of things and the security side of things. But on the other hand, I'm like, it's definitely going to

Walter Yuan, Empathia AI (57:57.583)
You have to write.

Dr Jordan Vollrath (58:12.699)
kind of change the quality of the documentation. It just suddenly like feels a lot more that like kind of disengage like they're telling me about something very deeply personal in like a psychotherapy session and they're just like the keyboards clicking along. And it's just, definitely is not ideal in my perspective.

Walter Yuan, Empathia AI (58:29.558)
Yeah.

Walter Yuan, Empathia AI (58:34.178)
Yeah, you brought up such a good point. know, I shadow doctors who also not use any technology. no, they're not using AI. And they're typing. And I could remember, I remember the awkward moments where the patient is sharing something really personal deep and really crave for that empathy.

And but then I look across the room, the doctor is just busy typing. And then with that sound. And yeah, I think this is where AI can really.

Dr Jordan Vollrath (59:14.645)
I hope things continue to go well. I hope to continue to see more of Empathia. I promised I'll go check it out again now that you guys have the voiceover IP support. I'll give her a whirl here and see how it goes and how it compares with the one I've been using. But I really appreciate you taking the time to chat again this afternoon, Walter. Thank you so much. It's been a pleasure seeing you.

Walter Yuan, Empathia AI (59:36.598)
Absolutely, Jordan. It's always good to see you again.

Dr Jordan Vollrath (59:39.286)
Again, if anybody's wanting to get hooked up, where can they find Empathia? What's the website?

Walter Yuan, Empathia AI (59:44.66)
It's empathia.ai and it takes less than a minute to sign up. We have a no string attached free trial. Try it. always believe, you know, let the product and value speak for itself.

Dr Jordan Vollrath (59:56.291)
Awesome. Okay. Thank you so much. Have a good one.

Search Jobs

Post a Job

About the Author

Cherry Health

Canada's Medical Network

About the Author

Cherry Health

Canada's Medical Network

About the Author

Cherry Health

Canada's Medical Network

Listen On

Listen On

Cherry Health 2024TM • All Rights Reserved