Deepscribe podcast recording – 2024_10_03 09_59 EDT – Recording
===

Amy: [00:00:00] Welcome, everybody, to Mastering Medicare with Alex and Amy. We're back with another amazing episode. I think in the spirit of 2024, where everything has the word AI in it, Alex, I thought it would be really important for us to have an completely centered on AI podcast. What do you think?

Alex: Totally. And it was so funny what happened because Amy called me, texted me, I, want you to talk to these people. I, this company called Deep Scribe. There's a Dr. Dilly. I, I was like, I know Dean, like . He's, we've already hung out before. It's too small of a world. This was serendipitous and exciting and really excited to have.

Dr. Dean D'Alili here from DeepScribe. And so why don't we just jump right into it? Dean, welcome. Tell us who you are, what's your story before DeepScribe and then let's get into the DeepScribe story. 

Dean: Yeah, sounds good. Alex. Thank you so much. Amy. Great to see you guys both again.

Nobody who's anybody in health care doesn't move around these circles without running into Alex at least once or twice over [00:01:00] the course of a year. So no small coincidence that we reconnected. Yeah. So happy to be here. I'm chief medical officer of deep scribe. I've actually been with the company for going on 6 months.

And had a rather circuitous route on my way to joining an AI the AI healthcare craze. So just a little bit of autobiography about myself. I'm an internist was trained in internal medicine at Johns Hopkins where I actually focused on geriatric primary care. And through a series of you know, weather related and also just professional events.

My career led me to become a hospitalist where I started off in academics at Johns Hopkins, teaching residents about health care and really being a student of it myself. I was there back in the days when I trained, we were in a paper based system and just on the verge of going into electronic health records.

This was before all of the health care policy changes had been made. And You know, learned a lot in that early phase about [00:02:00] thinking about process, thinking about technology, thinking about all the various stakeholders who were involved in making patient care happen and really learning what I didn't know about the complexity of health care You know, I would say that it was a very pivotal time for me and my professional development.

I had taken and I made this allusion to weather related events. So I'm originally from New Orleans. And when I left Hopkins from residency was going to be moving back there. And the timing of it just happened to be. Within a day or two after Hurricane Katrina hit and that caused the first real pivot in my career where I spent some time on Capitol Hill.

And so leaving residency, what that meant was, is I went from being in the trenches, the day to day, you know, patient care journey to really trying to learn about healthcare at the national level. And so spent some time in Capitol Hill, working on the Senate in the help committee. At the same time that many of these questions related to value based care pay for performance.

The electronic health record movement was really being started. And so when I then reentered medicine as [00:03:00] an academic, and as a teacher, you know, I was really grappling with. You know, the day to day complexity of it, but also thinking about the 000 foot view of what's going on in health care and how we could do things differently.

And so I would say it was really that period of time in my career that I started to think more about the health care system more broadly and how I could play a role in trying to architect something that was, higher quality, more efficiency. And more value based before value based care really kind of became a buzzword.

And so from there I ended up leaving academics going into the community really re confronting some of the research that we've all read about, which is about the significant lag. Between the discovery of new knowledge and its wide proliferation into normal practice. It's often said it is a 17 year gap.

I would say, in my personal experience, moving from Hopkins to Florida could have been longer than that. But It caused me and enabled me to sort of start my career around a trajectory of leadership and trying to think about process change and team building [00:04:00] and how to really drive behavior change and clinicians and the intersection between that and healthcare with technology has been a recurrent theme of my career over the course of several decades.

The bulk of my career has been in the physician practice management space. So I worked for companies like. Hospital physician partners for Schumacher clinical partners, and then ultimately for envision which is the largest of the physician practice management companies in the hospital medicine, emergency medicine space.

And from there went to dispatch health, where I got my foot in the door on the healthcare startup and the digital health kind of landscape. And then that led me to deep scribe. So my journey has kind of been everything from academics, community based. administrative leading at scale digital health and now AI.

And so it's with all of those perspectives that I sort of bring to work day to day and yeah, happy to share with you guys what we're doing, what we're building, what we're excited about, but just also tell you that backstory as a way of context. 

Alex: Amazing. so Dean, did you ever actually have an in person scribe in any of the work that you did?

Dean: So [00:05:00] rarely. most hospitalists did not work with scribes throughout the bulk of the roles that I've played. We did have a lot of in person scribes that I managed. I one time did walk around with a scribe as a hospitalist and didn't find it to be super helpful in that type of an environment, but yeah, that had more limited experience myself with live scribes than some of my colleagues have.

Alex: What you, Amy? Did you ever have a scribe? 

Amy: Oh, always. I mean, from the minute that I hit the floor, even as I feel like, a resident in my, last year residency, I feel like I sometimes, one of the attendings was like, I'm not going to deal with this. Just go work with Amy. I actually did get a scribe starting my senior year as a resident, and then I always had one moving forward.

Alex: For those people who don't know what a scribe is, maybe we should explain what that 

Amy: is. I was thinking we need to go all the way back to the beginning a little bit. Yeah. So 

Alex: before you get into it I've had patients ask me like, who's that person? I say, oh, that's my scribe. And they're like, they think I'm talking about something like from the Teutonic Knight era or something like somebody with a special hat or whatever.

Yeah. So what is a scribe? 

Amy: yeah, that's [00:06:00] a 

Dean: great question. Oh, and Amy, feel free to answer. 

Amy: I'm going to jump in to just say that without a scribe in certain settings. it was very difficult to keep track when things are flying at you. There was an actual person that follows the doctor around that basically takes notes, makes sure that the chart looks just so and kind of make sure that it gets all tidied up in a way that physician can.

Concentrate on what they're supposed to be concentrating, which is on patient care. I mean, I think that's conceptually it and sort of addressing what I was thinking was the three prongs of medical documentation in my own little weird mind. You're gonna fix that for me, Dean. But would be you know, it's for clinical communications.

It's for justification for medical Insurance companies and for MedMal. mean, those are the three things that I always thought about when I was like, oh, my God, I have to fill out this awful chart. You know, I need to be able to document it so that if anyone asks for it later, it says what I think happened and what I think needed to have been thought about.

Dean: Yeah, I mean, I think that's fair. Number 1 Alex, I really enjoyed the medieval [00:07:00] reference because you're right. I mean, it's such an odd scenario where we live in a world where. medicine in America is the most expensive industry by a mile.

We spend over four and a half trillion dollars on health care in the U. S. We have outcomes that from a population health perspective, rank us just below Costa Rica. and we have, and I think these are great things and great attributes about the health care system. we have some of the most.

academically advanced people sitting in a role of decision making who were doing heroic professional work every day. But the reality of it is, is that the burden of what we've placed on backs of physicians and clinical practitioners more broadly, is clearly something that people struggle with to deal with.

And there, and as Amy, rightfully mentioned, there's a lot of different stakeholders in a modern note. What I actually did Alex, to sort of think about the historical framework was when I joined deep scribe, I started thinking about like, what was the history of medical documentation? And so I was surprised to find that like, ancient Egypt, there were medical scrolls that people had written.

And over the course, the [00:08:00] vast majority of history, like a lot of medical documentation was. Just like when people write books, things that people would write down to remember for themselves or things that they would write down to pass on history and learning and ideas and insight into others. Clearly we've come a long way in the way in which medical documentation has evolved.

And part of that has been, the rise of electronic health records, which I think was intended to be something really beneficial in terms of trying to make health care data more accessible to help make, at least medicine administration more. Reliable and safety oriented. But what it has become and morphed into is, for a lot of people a very high level of burden to kind of click through and navigate and capture information that is.

not necessarily purely in the best interest of the provider or of the patient for whom the information is being captured. And I'd say that's a whole nother level of like why we practice medicine the way that we do, why we document medicine the way that we do. It's for a variety of reasons that have evolved clearly over time due to both [00:09:00] economic and technology and clinical.

sort of forces that being what it is, there's the digital version of how we solve for scribing, which is what deep scribe does. And then the analog version, which is the live scribing that you talk about. And so for folks who don't yes, doctors do, as a result of needing more time and efficiency and the ability to sort of think and focus their time on patient care have in recent history brought into the patient exam room, a sacred space.

Another person whose job it is, is to listen and to write and to capture information and try to check all the boxes and to capture the care. Even though that they're not necessarily medically trained but that they can understand the terminology and complete the sort of paperwork administrative burden of what it means to practice medicine.

And so those are some of the folks. That we service, but actually it's ironically, it's a large group of providers who can't even afford a live scribe as a solution who are the most drawn to an AI based solution because of the cost savings and so, whereas in healthcare, the most common area to [00:10:00] see a live scribe is in the emergency room or an orthopedics or surgical based specialties where time is really a premium people want to move fast particularly in the surgical case, you know, they get paid to be in the operating room.

And they want to sort of see patients and their clinic schedules are much different than those of mine as an internist where, we may see 15 to 20 patients a day. An orthopedic surgeon can see 40 to 50 in a day. And it's, there's a lot of routine and there's a lot of speed and there's a lot of need to just have someone capture all of that in an electronic environment.

So, yeah, that's definitely. It's odd when you tell people outside of medicine, how the behind the scenes of medicine really works. And I think trying to explain to a lay person, the whole. Emergence of scribing is one of those things that makes you have to step back and reflect on your own profession and realize, man, how did we get here?

Amy: So I'm going to take a pause and I actually want to take us to a little bit of a higher level because I think one of the reasons that I was drawn to just even have you guys on the podcast and seriously, like it was just total coincidence after coincidence that got you here was because I have been, in this world of [00:11:00] house calls and then hospice and then home care and the emergency room and working as a hospital.

It's like, I've like sort of worked all around in all the different spaces and documentation of what you do every single day and every single time you see a patient is not even, you can't even describe the burdensome nature of it, whether you're writing it down and all the pertinent negatives you need to remember and all the different things and all the cognitive cueing that you need to have.

In addition to the fact that once the EMR came in, it took, you know, 36 clicks to get to this, that, and the next thing. it really has actually so disrupted what I think is the regular way of taking care of patients that should just be patient provider focus. I think it's sort of in some ways wrecked it.

Can you kind of like speak a little bit to the number of EMRs that You have seen or have worked with or that deep scribe is working with and kind of talk about how you interact with them. 

Dean: Yeah. I mean, I would say that there are literally hundreds of them. And so while we're fortunate that market is consolidated enough that there's a couple of key players that when we build deep relationships [00:12:00] with those relationships become impactful to a vast, plurality of the healthcare labor force.

The reality of it is, is that there are a lot of different EHRs out there. And then some of them are even homegrown by practice. And that does create a level of complexity for us because we are a tech integrated solution. And so the way I describe this for folks is that DeepScribe has basically, three different functional layers to what your experience is when you use DeepScribe.

One is, is what's the content. That we design to be specialty specific or that we allow a person to customize in terms of their preferences around how their documentation would appear. The second one is, is the engineering of how we create features and functionality that make the scribing experience.

More functionally relevant in your clinical environment. So things like, do you have the ability to, deal with patients that have multiple family members in the room? Can you deal with interruptions of your workflow? Can you allow multiple people to take care of the same patient because you work in an academic facility?

So those are all feature [00:13:00] functionalities that we engineer to like allow for a broader clinical application. And then the third and the most important one. Can sometimes be EHR integration, because one of the things that we do for convenience is, is we sync with a provider schedule. So that's how they access patients.

And then we also take the notes after we've recorded and process them and push them back into the EHR. So that way you're not. Having to use multiple platforms in the daily work for you to be able to, write a note. And the whole goal is to be more efficient, not to have to make people switch back and forth.

We've actually added on a lot of functionality that is context awareness based on those integrations. And so that's a really important feature for us. To create to reflect the reality that patient care is very rarely single visit, you know, episodic. It's usually part of a longer care continuum.

And so being able to read historical information and write a current note in the context of what's come before it is one of the things that makes our software much different and that's all EHR dependent. And so 

having those EHR relationships is really vital [00:14:00] for us.

Amy: That is very cool. So so just so that we kind of like, I also like to just go back to level setting. Basically, if I'm using deep scribe, I am a provider and I hate using that word, but I'm a provider in the room. I basically turn on an app on my phone and I put it in between myself and the patient.

And it basically listens to the entire interaction using, kind of get into how maybe like a little bit of your secret sauce and then it basically assists with the documentation. 

Dean: that's exactly what it does. It's the click of a button. Right. And so let's go back to the scenario that Alex was sort of describing of like the doctor bringing in a scribe.

I sort of spoke before about how scribing and documenting in general was just between a provider or clinician and a patient. And now we've got the computer, right? So there's a lot of different people in that sacred space. And where most clinicians, most physicians struggle is that they walk in, they shake hands with a patient and then they go log in and then they start typing what they're hearing.

Or they're having to listen and then try to translate that outside of the room into some other kind of. Electronic portal. What we do is we let the provider [00:15:00] walk in the room. We do use iOS devices because those microphones are some of the best in the business. We actually tried in our original iterations of deep scribe to create our own technology and devices and hardware, but we can never replicate.

The power and the accuracy of a microphone in an iPhone. And so there's for folks that are Android, we also accommodate laptop and desktop applications that are windows based. And so there's a lot of different ways you can use a deep scribe, but the functional sort of experiences is that you walk into a room and you hit a button.

And the microphone listens, it can capture all the voices in that room, even in a busy ER with alerts and with monitors and with interruptions and with overhead pages and that sort of stuff. And it can listen and also can contextualize. So it knows based upon the type of speech and the content of speech, who the provider is, who the patient is, who other family members are.

And so it's listening and recording. As soon as you stop and you hit complete again, one more click of the button, then it will take multiple recordings. So you can interrupt your workflow. [00:16:00] You can come back in and out of a room. You could switch over to multiple rooms or go from room three, but over to room two and back to room three.

So you can do this out of sequence and it will combine all of those recordings into a single unified transcript and then map those concepts that are within that transcript to. Medical concepts and restructure and summarize them in the form of a soap note. And so the basic soap note is actually pretty easy for AI to do, which is one of the reasons why we live in a very competitive landscape with a lot of other folks, now offering this up with GPT wrappers, et cetera.

What we have done is, is we've built our own. AI engine that we've trained over time through human scribes to code data and to label it. So that way we've been able to achieve the highest rate of accuracy in our industry. And that's not me saying. So that's actually comes from a evaluation that was done by McKinsey and company as they looked at the ambient scribe space.

And so our engine is not only producing more accurate data. We've now gone to the point where we've really focused on multiple [00:17:00] workflows so that the content is more pertinent to the clinical providers. And then we have all of those features and functionalities that allow it to be more useful in the clinical setting and create the type of documentation that providers really want.

And so, What this means from a provider's perspective is, you can go and see a patient and let's say your encounter visit is 15 minutes, 20 minutes, or in the orthopedics case, 5 to 7 minutes whatever additional time you would normally have taken to document is time that you now. And in aggregate, that can mean somewhere on the order of magnitude of an hour, hour and a half, sometimes up to two hours of your day that you're no longer spending typing into a computer.

Now, the other side of this is that for the patient, the patient doesn't see a handshake and then the back of someone's head as they focus in on the screen. You can stay engaged visually with your patient throughout the entire course of a visit. And we've seen that a spillover effect of this is that patient satisfaction, patient engagement has gone up.

So one of our largest health system partners is Oshner. And they studied this and they found a [00:18:00] statistically significant improvement in patient satisfaction numbers from the mid 80s to the upper 90s. on NPS scores for the providers who use DeepScribe. And we attribute this to the ability for a provider to re engage more with the patient than with the screen that's in front of them.

Amy: Amazing. So like, in my mind, I bought it for every single physician that I, like, know in my life. why would anyone want to be charting in their pajamas? And I think that's the whole thing. I mean, even in hospice, and I, and I have to go back to sort of other care settings, the documentation, and because the name of this podcast is Mastering Medicare, I think it's really important to kind of go back to this, which is the documentation is not just random, summary of the words patients say.

There are actually rules and requirements that go into medical documentation, and they are different. Over the course of whatever specialty, whatever setting, whatever type of. Medicine is being practiced or whatever type of service is being provided and There's documentation requirements in the cpt language, right?

If you do a nine nine blah blah blah It's going to tell you you have to do a b c and d and [00:19:00] document all these different things And you may have actually listened to those words being spoken and heard those words being spoken But unless they're written on that piece of paper if you ever get audited by an insurance company or medicare they're going to want that money back and Unfortunately, it's getting more and more terrible in terms of what is expected from a regulatory side.

So I think that this ambient AI thing almost like solves this enormous problem. I mean, it's sort of like looking at the answer, you know, I'm like ah, Dr. D'Lily figured it out. You know, it's one of those moments where you're like, this has to be the answer. How could we go.

Any farther with people being like, hang on, I'm just going to write down what you said later on in the day after you already said all those things in a way that Medicare told me that I'm supposed to write them down or with all the pertinent negatives so that the med mal attorneys don't say that I didn't do it right.

Or that like the guy up in the ICU doesn't say, Oh, they didn't document this. So I just sort of see this as like an, Oh my God moment. Can you speak a little bit to of How deep scribe addresses some of the Medicare requirements, if at all in [00:20:00] terms of helping to structure the information that it grabs during the ambient experience.

Dean: Yeah, a hundred percent. and I would say a couple of things in response to some of the comments that you made with regard to, if you don't write it, then it didn't happen. Right. And so that's one of the most important things that we grapple with. One of the things that I've started to see and then hear from users is how often ambient technology here, something in a conversation that the physician didn't hear or didn't remember.

That they heard. And so it's happened to me. And so what happens is you do an encounter, you go and you review a note, and you find a symptom in there, and you say, this must be an AI hallucination because I didn't hear that happen. And so what we have in our technology is the ability that you can click on any content within the note, And it will highlight for you in the transcript, what the source of that data was, and you can actually, because we record all of the conversations, you can listen to the conversation again, and it's happened to me where somebody said something about a cough, and I thought, you know, I'm role playing typically [00:21:00] when I use the technology, but I'm so focused in on a line of questioning or a diagnosis, or the thing that I have to go check the box for that.

I miss sometimes things that people say and then in reviewing it, I realize. Gosh, like, no, the I didn't make this up. Like, here it is in front of me and I can listen to the voice. And so it's been humbling for me to realize that one of the key things that we are learning is that the entire funnel of data that healthcare captures, which, by the way, health care is another statistic that I just learned this week is that health care generates like 40 percent of all of the data in the world.

And so a lot of this medical data that we're generating is just incredible volume of data. And I would say if we had to be honest with ourselves, the vast majority of it is noise and probably not super helpful. And so what does not appeal to me about the idea is that now there's even more of it, right?

But what does appeal to me is the idea that there is some signal in the data that already we're losing because we're relying on clinicians to [00:22:00] transcribe it for the first time into the record. So essentially like the top of the funnel is incomplete in some ways. Some physicians grapple with this in different ways and they ask us a lot of the times like, well, what do you do with the recording?

Do you keep it? Because does this recording put me at risk if something happens and it's discoverable? Unfortunately, we don't have a lot of case law to be able to cite in terms of data around what the risks are or the benefits are of having an ambient recording.

But the reality is, is that it could cut both ways, right? So if you're in a situation where the provider did something. And missed it and was negligent, then you're right, having ambient is not going to be helpful to you. I'd say the vast majority of the time what's happening is, is that people are catching things that they may have missed as a result of ambient.

And they're able to go back and either correct the record or correct outcome in terms of what happened with the patient. And they're also able and sometimes to use it as a defense. To say, no, these were the things specifically that we talked about it. I can show it to you here. And [00:23:00] so how will more robust source of truth play out in the legal system?

It's hard for anyone. I think to predict, you know, we've had lawyers tell us that because it's not part of the HR, it's not part of a standard discovery. I know, I'm sure Alex and you know that when you get involved in a malpractice event, a lot of things outside of the EHR become discoverable, like text messages, et cetera.

And so having any data to some degree for some folks can represent a risk. I'd say to the degree that providers are using this and they're using this with good faith and trying to take good care of patients and using it as an extra set of eyes to listen and to cure for things that they might have otherwise missed.

I would want to have ambient on my side. As a documentation tool. Now, how long you keep those records? You know, we allow institutions to sort of decide for themselves what the data retention policy is going to be, because there's no federal or state mandate that we have to keep it for a certain duration of time.

Our practice is to keep it for 7 years in accordance with the general. Medical record retention policy but because it's [00:24:00] not considered part of the EHR an institution could lower that threshold. Your question was specifically about Medicare and some of the burdens of documentation.

Part of what we are continuing to evolve is more and more. Content that helps people capture that type of very specific data. So for instance, in our primary care module, we built an annual wellness visit template that you can use as part of an ambient visit, either in the home or in the office to sort of go through that information and again, in an unstructured way, and then have it captured linking to vaccinations, you know, your, health screening exams, like what needs to happen terms of.

Upcoming tests or screening evaluations that need to take place. And so those are the types of things where we realize that one size fits all soap note is not going to do it for the vast majority of providers in the vast majority of clinical cases. And so building out that complexity of workflow is, really, I think, The key to being able to be more useful as a tool.

Amy: So I always like to realize who our listeners might be. So a soap note is, oh my God, I'm going back to like [00:25:00] third year med school is subjective, objective, obsessive assessment and plan. And that is sort of how the general medical thinking, this is how they train us to think of subjective, which is, this is what the patient thinks is going on.

You know, here's their story. Objective is like. Real live data like their blood pressure, their physical exam, and then there's an assessment, which is sort of a summary of what we think is going on. And then the plan is the plan. Now, Medicare has other types of ways that they want to see documentation, which is coding 95 and 97.

And then they got rid of some of all of that. So it sounds like you guys are really moving towards addressing what it is that the. Payers are asking for in order to validate the amount of money that they pay. 

Dean: Yeah, a hundred percent. So that, that specific part of your question is really a good one.

Deep scribe was founded in 2017. And as you both know, the original coding guidelines that we had operated under for the entirety of my career. Those E and M codes with evaluation management codes [00:26:00] were predominantly based upon the level of complexity of your subjective history and your objective and what sort of data.

So you used to get points for how complex each 1 of those data elements were. Now starting in 2023, that's essentially gone. And now all people care about from an perspective is, is the level of your decision making and the complexity of the patient's condition that you're treating and we've adapted to that.

Right? So, whereas in the olden days, Everybody wanted to have, very robust, nine organ system exam with all of these review systems. The need to do that now is much more limited. And we've seen providers say to us, like, we don't need as much verbosity in the note, like we want something that's much more concise and much more medically relevant which is good because as I alluded to earlier, I do think that there's a lot of noise in healthcare data you see, and I've seen Providers who copy forward notes where you go and you try to read a chart and it's literally littered with so many macros and things that are generic that you're trying to find signal in there of like what really [00:27:00] is going on with this patient and what does this provider really think?

One of the things I'm the most proud of about DeepScribe is when I use the technology is it really does synthesize information. Sometimes in ways that I hadn't done when I'm doing a visit. And I'm like, look, it's like, oh my 

Amy: God, you have a new, I've never heard of . 

Dean: The AI really is like thinking, I mean, it's not thinking.

technologists on the team know that that's not how large language models work. But it can fool you into thinking that it's being thoughtful. Because, you talk about things like your blood sugar or your insulin dosing and it's, understood that that's a concept that maps to diabetes.

And so just that type of thought process makes it feel much more intuitive and much more clinically useful. 

Alex: So, Dean, what you just shared a couple minutes ago It really brought back so many memories for me when I was using an in person scribe. So think this is really interesting because it's, I think tells us potentially where this is all headed.

So I'm in the exam room, like this probably happened, you know, a thousand times. I'm in the exam room talking to a patient and the [00:28:00] scribes there with me on the cow, on the computer on wheels. And the patient says something that seemingly is ridiculous. Ridiculous and doesn't make any sense.

Right. And I just make a slight glance over at my scribe and like a slight head movement, like don't document that. Right. And they understand. They've been working with me so my point is, I think as physician, as human physician, documenters, or. when we are taking a history from a patient, we actually have a lot of biases, and there's a lot of weird forces in play, like, as an ER doctor, you're, trained to instantly, they always tell us the first thing you walk in the room, you're making a decision.

Is this patient going home or not? And then whether you like it or not, a lot of the questions you're asking are biased towards pushing that patient in one direction or the other. Okay. and so you start to ask certain things and you start to ignore the seemingly noise that doesn't align with your judgment call as opposed to the other way around, which is listen to all the information and then make a judgment call.

[00:29:00] That's not how we're trained. We're actually trained to make the judgment call first, right? Amy knows this in emergency medicine. That is the first question. 

Amy: Make a decision about disposition instantaneously. Like they're either staying or going, we've got to decide now 

Alex: the note that we document is not a transcript, right?

It is actually a story that we've created, and the story has to make sense, right? So the story has to make sense for med mal reasons, because it's got to lead to some natural conclusion. Like anybody who reads that, the subjective and objective portion should come up with probably the same. Assessment and plan as you did, right?

So we're very selective what we document and I was and this is And I'm not saying this is necessarily good or bad. It has significant, both positive and negative implications. But there was a really interesting thing, a study that was posted on Twitter just in the last like 48 hours, which was the accuracy of diagnosis of doctors without AI was something like 74%.

Let's say, I don't remember the exact numbers, but it's roughly like that. With the assistance of AI, it was slightly [00:30:00] better at like 76%. Guess where AI was on its own. It was like 90 something percent. So the human piece of this is actually potentially the poison because we are so biased. We are so time restricted and we are trying to reach a specific goal.

Whereas if AI were taking the history on its own with unlimited amount of time with the patient and no biases. It might actually do a better job. So I I have lots of like really weird mixed feelings about this. And only now am I realizing how, like, the way that we would curate the note that we would write in the old era was, 

Dean: Yeah, I mean, I would say that I grapple one of the first questions I asked, the founders of describe when I joined was like, how far are we from really having AI just take over the role of the physician?

And they both smiled when I said it and didn't really have a great time based answer for, when it would happen or that it wouldn't happen. But I mean, I think where I've kind of come to [00:31:00] is, is that look, there's benefits to the technology. There's limitations. As well, one of the best frameworks that I've seen, written about AI in healthcare is from Neil Shaw.

He's the CMO of Maven. he was a friend of mine from long ago. We're still friends, but he wrote it in health affairs, an article about how AI. And use the metaphor for, I think we talked about this in a previous conversation you know, can be assistive all the way on a scale, like, 0 to 5, like, similar to self driving cars.

Right? And if you've been in San Francisco. Like I have recently where our headquarters are, you can get into a car with no human. And in some ways it's a better experience. It costs a little bit more money, ironically, but that's because there's, fighter jet level sensors all over these cars to make them roll down the street.

And apparently there's a whole team of humans behind the scenes helping make sure that this, seemingly virtual experience really is going the right way. You know, for me, I would say, okay. Doctors that use deep scribe and that use the ambient AI begin to learn how to navigate the story that they're telling to the [00:32:00] patient and with the patient to the AI.

Right? So, like an ER doctor who uses our software says, look, if I walk into the room and I think somebody who has chest pain has real chest pain. I evoke those symptoms from them. And again, it's sort of mirrors what Amy and you were saying with regard to, like, I've got the disposition in my head, but now I need to get the person to help me write the note that's going to reflect that story.

And then on 

Amy: the flip 

Dean: side, like if I don't think that they have real heart disease, right? Like I elicit those conversations. Look, I think that. I'm still of the camp where I want the vast majority of critical medical decisions to be made by a person. But I mean, I think where I land is like, some opinions in popular press that I hear is like, Hey, I is not going to take your job, but a person using AI is going to take your job.

Right. And so I still think that what we're trying to build towards is the magic of the marriage between. the clinical person and perspective and the AI enablement where [00:33:00] the AI is helping to make the person take care of the patient, better in the context of the limited amount of time that they have.

hadn't talked about this. So, like, the vast majority of what deep scribe has done in its history is really predicated on efficiency, trying to make that patient provider interaction more timely and less burdened by the documentation. In some ways, like we haven't won that game because we're still in the early innings of, doctors using ambient, but the game that we're now starting to play and where we're starting to spend a lot of time and product developing new feature functionality is what we call ambient intelligence, which is how do we bring information from outside of the visit, outside of the room, outside of the sacred space to the moment of where care is going to be delivered.

So the doctor who is either decided. that this should be a patient with chest pain that goes home or a patient with chest pain who should be admitted. Are they doing that based on instinct alone? Or are there things in the medical record or things in the patient journey that we can do to make sure that they're making that judgment, that critical, intuitive [00:34:00] judgment with the right amount of data so that the visit that follows the conversation that follows.

Steers the patient towards the right outcome. I think for us, that's where we want to go is trying to figure out the partnership between computer and person and how we bring those together in a thoughtful way, in a way that preserves the efficiency of what our core product is capable of delivering, but then layers in another.

Element of prioritization of information, because we all, like when you take a step back, Alex and Amy, like you both know this. You can't spend as much money as we spend trying to achieve good outcomes. land in the middle of the road of the pack of industrialized nations and realize that some of the root cause of what we're dealing with is, is that patients are getting sicker, right?

We have an aging demographic, an aging tsunami of baby boomers who now I just read or. between 60 78. And we all know that the, bird of illness when you're 60 versus when you're 78 changes dramatically. So we have aging patients. We have providers who are increasingly under time pressure in the [00:35:00] fee for service world.

We're in an environment where with rising inflation and labor costs going up, the only way for physicians to cover their overhead and preserve their income is to see more and more patients faster. And so part of what you have to do is you have to narrow the aperture of time where I get to learn about you and your conditions and move on to the next one.

And so what we want to do is we want to make sure that the synthesis of the information that you have available to you is as high quality and as comprehensive as we can make it. So that way you can make a better decision. Your patient can have a better outcome and you can move on to the next. 

Alex: One thing I just want to this ambient intelligence, Dean, I think is absolutely critical because what I'm seeing happening in the market is that especially in emergency medicine, more and more of the practitioner mix is becoming heavily weighted towards a P. P. S. You know, N. P. S. And P.

A. S. And I have worked with phenomenal P P's of all types. But their training is simply not in the especially like when when they're right out of their programs [00:36:00] is just not sufficient. They simply haven't seen enough patients. They haven't seen enough permutations of each disease state in order to be able to come up oftentimes, not always, but in order to consistently come up with the right and a complete differential diagnosis of what might be happening with the patient.

And if you can't even come up with an accurate differential diagnosis, then patients are going to die, right? Like it's going to happen. Like, are you considering brugada syndrome in that patient with syncope? If you're, if you don't even know what brugada syndrome is, how are you going to potentially catch that, right?

So that's where I see like incredible value for something like describe if you can listen better than a human can listen, and then you can take that full context information and not the biased context, right? Take that full information and come up with a more comprehensive and more accurate differential diagnosis.

And probably like, Prioritize that differential based on likelihood, right? Based on what it's picked up, I think that is a really, really powerful thing. That is something that [00:37:00] could take like a, you know, an APP with a mid or low level of experience and bring their diagnostic and treatment accuracy to to a higher level and same for physicians, right?

This is not limited to a P. P. This is Every clinician could be better using this sort of technology, which I love that, 

Amy: I think in some ways it reminds me of this idea that the provider becomes the telepresenter to the A. I. Right. Basically, the provider creates a conduit for the information to come into the A.

I. To be able to sort of like, 

the way that we've sort of structured some of the telemedicine that has happened since COVID and even prior to COVID is that you can take anybody that can hold up an iPad that can facilitate a conversation with a physician at the other end, right?

Like that, person who's holding The technology becomes the TelePre presenter. So now in this scenario, any provider is essentially the TelePre presenter to the ai. Now there's a higher level of thinking, so that you're basically, in some ways just kind of like moving the, computer around the room a little bit and letting Dr.

F [00:38:00] 7 63 make their diagnosis . 

Dean: It's funny. Yeah. So Amy, you remind me when I was a, medical student at Emory. And one of the senior faculty members there was J. Willis Hurst. So if you've ever read Hurst's The Heart, right? he was like the preeminent author. 

Amy: Oh yeah, I read it and it was fabulous.

I mean, it was so good. He was in 

Dean: his 80s when he when he was there and when when he taught me. And one of the things that he said was like, he always believed that computers were going to at some point, this was in the 90s, right? We were talking about this. He was like a, some point, the computers are going to be running the show and health care.

But he said, you know, the thing about a computer is that garbage in garbage out. And so to your point, I think, and going back to that story of the patient with the heart disease or maybe without, I think that there's something to say for that, right? Like, I think at some point you have to have a person who helps curate the information, the story, one of the things I like about deep scribe is that the doctor is not the only person writing the story anymore.

Now it's the patient. But the doctor does write a [00:39:00] story in your sense, right, which is to guide the conversation and to elicit the information and to queue up the information that is going into the, hearing, the listening, the thought process of the computer as well. and I don't want to say this in a way that's inaccurate for your listeners.

The doctors are the ones making the decisions, right? DeepScribe hears information, categorizes it, processes it, it structures it, and where the line is between doing that and thinking, you know, I'll leave to the technology folks and philosophers to sort of help delineate, but it's clearly going to get better.

But the role of the clinician in my mind is always going to be preserved for the reasons why you say now, look, there are other companies in the AI sphere That directly interface with patients where through telehealth, you can get on and you could talk to a chat bot. You could give them your history, Alex and then they can start to steer you towards outcomes the way the FDA has regulated software as a medical device.

Implies that when the computer is really making a critical decision, that's where FDA approval needs to be received. And in our case, that's not what's happening, right? Like, we [00:40:00] very much preserve physician autonomy but this interplay is going to become more and more complex over time.

Amy: that's so fascinating. What you just said about software as a medical device. And I think that's probably I think one of the take home messages that I'm going to take from this conversation is just to be very aware of when is the physician decision making Not being usurped, but being replaced or even is it considered to be guided like what makes something a medical device then?

I mean, I don't want to get too far into the weeds there But so you're saying that deep scribe doesn't get into the diagnostics. It basically is here's what was said Here's how we've summarized it and that and doctor all of this and that is that it's not You cognitive queuing or anything like that?

Dean: we do, we nudge people over suggestions. We do synthesize like, again, I, to use the diabetes example, if you talk about insulin, if you talk about blood sugars, the computers decided that diagnosis is diabetes and that's how it writes the note. There's a human in the loop, which is the provider who both reads it, edits it, signs [00:41:00] it, and then acts on it.

Right? Like we're not taking orders, right? In another environment, person could say all this to a computer and the computer could say, you have diabetes. Here's a script for metformin. I would say that's where the line, I'd say, between where human in the loop becomes like the ultimate decision maker versus just the computer.

again, I'm not a regulatory and the FDA approval process isn't my subject matter expertise. But I would say, what we aim to do through ambient listening is to make the documentation process. more streamlined. What we are aiming to do with ambient intelligence is to make the conversation between the provider and the patient more informed and more valuable.

And to make sure that the actions that are taken are in accordance with trying to deliver the best possible care in that moment. 

Amy: So first of all, thank you. I couple more thoughts. I don't know if you guys see all these funny things that come over the web. Dr. Glaucon Flecken, the guy who always makes the really funny ones.

had a really funny one that had the rollout of the new EMR in the hospital and you know, there was sort of like the IT guy who's like, okay, so we're going to [00:42:00] talk to all the surgeons now. And like, he just started talking and some, nope, just like walked out of the room, right? Like people are quitting their jobs because of EMRs.

So, I mean, I don't mean to over stretch the amazingness of such a technology as ambient AI in terms of sort of solving some of that. But do you see this as a hospital system or a doctor's practice could implement to reduce burnout? I mean, that term all the time, but I just want to like really call that out.

Cause I see this as a burnout reduction agent. Yeah, 

Dean: a hundred percent. I mean, and other, competitors of ours have done, formal studies and research around some of the impacts on, provider, burnout, mental health et cetera. I'd say the data is there that like both time savings, physician satisfaction, their self reported, Propensity for burnout in our experience has all gone down.

I would say that's not unique to deep scribe. I think that is something that is just true of ambient broadly. What is, this year, I think, was the 1st year since the pandemic where self reported burnout [00:43:00] medical specialties more broadly declined, whether that's. True because of ambient or just because we've moved so far past the pandemic is sort of to be determined.

I don't think ambient is widely distributed enough to be solely responsible for that decline. But, yeah, I would just say. Look, I think we're in the early stages of this, like to use a baseball analogy. I'm living in Houston, so the Astros lost yesterday, but I think we're only in the second inning of the ambient story.

I think that the first couple of innings have been really positive in terms of. Trying to, help people with the burden of documentation, save time, improve patient satisfaction. Those are all unimpeachable outcomes. The next couple, stories that we tell, will be as we become more broadly adopted.

Like, do we become the new user interface to the EHR? Do we displace the EHR? Those are the types of questions that in 2025 and 2026, we're going to have to, begin to see like what the impact is. Clearly we don't live in a world where we get to just roll out technology and delight [00:44:00] people.

Like we live in a very competitive world. Other companies are chasing the same customers. The EHRs themselves are trying to sometimes, create this functionality, offer it sometimes for free. I would just say as a cautionary note, you sometimes get what you pay for in that scenario, but are going like EHRs in general are going to make.

are going to take steps to deliver products that preserve their market share. Like that's just the reality of this. And so how the winners and losers shape up over time, is very much a story that's so far been unwritten. But what we believe is that if you can make a very simple to use, tool, if you can save people time, you can delight providers and patients, and then if you can start to bring in a level of information that makes the value of the decisions that they make.

Greater both for the benefit of the patient, for the benefit of the buyer, the benefit of the user upstream, then everybody can win. And so, whether we're the only winner or one of many winners to me is really not what's as important as trying to build and deliver technology that makes this healthcare system of ours, better.

Alex: I'm going to add a [00:45:00] couple things since worked with. Many scribes over many years, and there's some interesting things that came out of that experience and it may offer a roadmap for deep scribe and other kind of scribe solutions. So, without a doubt Scribes help reduce the documentation burden, and that helps reduce burnout.

But there's other things that an in person scribe would do. For example, one of the things you mentioned, Dean, is the massive inflow of information that comes towards the provider, especially an ER doctor. that is interrupting their workflow, and those interruptions happen almost every 5 to 15 seconds.

And it is maddening. Like, imagine you're trying to work on an Excel spreadsheet, and every 12 seconds, somebody is tapping you on the shoulder and like smacking you in the face with a pie, right? That's what it feels like as an ERI. So now to give the EKG to the human scribe, you know, assuming it doesn't look like a STEMI or whatever.

And like, there's this filter on the input side. And that is tremendously helpful. The second thing that I experienced with a scribe. Was there's just another person with you having another person with [00:46:00] you provides a degree of emotional support that I didn't realize I really needed and it is something about that there's something like incredibly valuable and then thirdly, most scribes that I interacted with were pre med students. And so there was also this, like, opportunity to teach them and guide them. And many of them went on to become very successful physicians. And I, and they still message me years later. Because of like, maybe some of the experiences that we had together and what they learned from me.

So I think these are like really interesting kind of additional dimensions to think about. Like, what if your deep scribe agent actually had a name? What if it had a personality? Like what if it could occasionally send positive messages to the clinician? I don't know that I think there's, these sound kind of silly at first, but 

Dean: they're not silly.

So, incredible that Alex, because you and I had not talked about this preparing for today. Three weeks ago I was watching a news show and somebody who was an author on AI was doing a short segment about a book that they had written. And the story of the segment [00:47:00] was that he was talking about the chat GPT and how when chat GPT was first launched People would give it puzzles that it couldn't solve.

So, like, we've all been on websites where, you get the grid and you have to click on the boxes that show the motorcycle, crossing the street and it's fragments of a picture. So it's a picture puzzle and chat. She couldn't solve it. So what chat did was it went to a job board. And it hired someone and said, can you help me solve this puzzle?

And the human on the other side of the conversation said, no, you must be a robot if you're asking me to solve the puzzle and chat GPT said, no, I'm not a robot. I'm actually just visually impaired. And so then the person solve the puzzle and chat GPT got it. Right. So the point that he was making in telling the story was that, we talk about we're going through an election season right now.

And we talk about how AI has been really pivotal in spreading misinformation. And that the point that this researcher was making was that in 2020. [00:48:00] the for misinformation was attention span. So if you just sent out messages to people and you were scrolling through your newsfeed and you saw information that could be persuasive to you, but AI has moved beyond that in 2024, where now the battlefield is intimacy.

And to your point, I saw that and I said, know, like, we're very much in this prompting mechanical kind of information sort of realm in the way in which the product is sort of currently designed. What if it was oriented around intimacy and relationship building? And could we lever some of that as a way to positively influence change?

Right? Because at the end of the day, we have a broken healthcare system. Like this is the third time that like, we're revisiting that point. Making a broken healthcare system more efficient is not in and of itself. enough of a goal. Making providers less burned out in that broken health care system is a worthy goal, but it's not in and of itself a sufficient goal, right?

Like what we really want to do all of us who [00:49:00] want to see health care in America get better is we want to be able to change behavior. And the behaviors that we want to be able to change in some cases are clinical behaviors. In other cases, they're patient behaviors. And so you can also imagine, like, a 2 directional software interface, right?

Where patients are engaged in certain. Software technologies, providers are engaged and they use those engagements and those interfaces to try to help learn about each other and to influence each other. But ultimately, at the end of the day, like, can I play that role? Like what I always say to orthopedic surgeons is that deep scribe is not going to go get you a cup of coffee like it can't do that.

Right. But to your point, could deep scribe have the relationship that you mentioned about, like when you sit at the nurses station with your scribe and you're in between serious case, like there is benefit to that, right? Like, I know that because I've been in the healthcare environment. I've worked there and I know that being in the hospital.

despite being around all the people that you know, and all of the [00:50:00] familiar faces, like it could still be an emotionally lonely place, right? Because you're, kind of in your own moment. and you're surrounded by people and a lot of the patients that you see are people who you don't know, where you don't have continuity of care.

And so that lens it's not a crazy idea in my mind. Like, are we close to delivering that? We're not, but could we be if we intended to do that? I think, you know, that's a very interesting question that I grapple with myself. 

Amy: Well, it seems like Ambien AI and just DeepScribe and the whole general conversation we're having is another hello world.

is a new step, right? We're moving from, the paper charts to the E. M. R. to now sort of this whole new paradigm where things are more interactive. I cannot thank you enough for coming on the podcast today. This has been incredibly, I would say, intellectually stimulating. I think. Most of our listeners are really focused sometimes on the nuts and bolts of how does Medicare work?

And how does Medicare work? This is just going to become increasingly more relevant to anybody who is [00:51:00] interested in the world of aging, the health care space, period, end of story. I'm so excited because it's clearly more than just, you know, A virtual scribe. It is data utilizations. It's efficiency and workflow, and it's got some existential questions that go along with that.

Alex, any other reflections? 

Alex: think

you really opened my eyes that we truly are in just the 1st inning of this. This is just the very beginning and think I'm realizing how much flux there's going to be in everybody's workflow in healthcare. And this is just the early stages.

So Dean, where can people find you and your company online? anything you want to shout out? 

Dean: So yeah, deep scribe. If you Google us, if you whatever your search engine of the day is you'll find our, website, our landing page. We work with practices and providers of all sizes, types and varieties.

So while I mentioned earlier in the podcast that we large health system partners we have individual practitioners, ones and twos. Who call us all the time not just physician offices, but also We've been talking to other home health care companies that are doing care in the home some of which [00:52:00] that I previously worked for where I just know that the technology is going to work fantastically, like, just in terms of being able to capture the complexity of what happens in the home space as well as in the clinic space.

And so, yeah, I just encourage folks, if you are interested to learn more if you're not sure whether we would work with your EHR or not, our team is really happy to work with folks and to figure out where we can be helpful. It's not just a software that you buy and that you walk away with we have a whole customer success team.

We get on the phone with folks, we help them set it up and we do follow up phone calls to make sure that the, system is working the way we intended it to, and that your experience is delightful. And so, yeah, folks to out in our website. We'd be happy to make contact.

Alex: Amazing. Dean, thank you so much. 

Dean: Thank you. 

Amy: Thank you, Amy. 

Dean: Thank you, Alex, for helping me out today. And yeah, it was wonderful meeting you both.