Latest / Style Points / Anesthesia and Artificial Intelligence with Marcus Lehman

Transcript

Anesthesia and Artificial Intelligence with Marcus Lehman · Style Points

• 0:00
  Hello, anesthesia enthusiasts and welcome to Style Points. I'm chat GPT, your AI host,
• 0:16
  and this is a special episode where we delve into the world of artificial intelligence
• 0:20
  in medicine. Join us as we explore how AI is reshaping healthcare from diagnostics to
• 0:25
  patient care. Let's get started.
• 0:27
  Thanks, that was a really good intro.
• 0:30
  You're welcome. I'm glad you liked it. If you need any more help or adjustments, just let me know.
• 0:37
  Well, anesthesia enthusiasts, it appears that AI has taken my job. So, I guess I'll hand it over to Nate.
• 0:45
  Thank you, John. I have the pleasure of introducing Dr. Christina Zhang Miller today. She is one of our chief residents here at the University of Cincinnati,
• 0:53
  and while I'm sad that she'll be leaving us at the end of this year, I am glad that she's only going across the street where she'll be doing her pediatric fellowship at the Cincinnati Children's Hospital.
• 1:01
  Christina will be teaching us about qualitative versus quantitative twitch monitoring. Take it away, Christina.
• 1:07
  Thanks, Nate. In 2023, the ASA released its practice guidelines for monitoring an antagonism of neuromuscular blockade.
• 1:15
  These guidelines issue a strong recommendation for quantitative monitoring over qualitative monitoring, and our institution is actually in the process of transitioning over to quantitative monitors,
• 1:26
  so I figured it would be helpful to highlight the similarities and differences between the two.
• 1:31
  In qualitative twitch monitoring, an electric stimulus is applied to a peripheral nerve, and the resulting movement, or twitch, of its corresponding muscle is assessed by the clinician.
• 1:42
  Usually a train of four stimulus is applied, which involves the delivery of four equal two-hertz stimuli.
• 1:49
  To assess the depth of paralysis with qualitative monitoring, the user uses a peripheral nerve stimulator to deliver a train of four stimulus,
• 1:58
  counts the number of twitches elicited, and correlates that to a level of neuromuscular blockade.
• 2:03
  So, for example, zero out of four twitches reflects a deep level of blockade, while four out of four twitches corresponds to a shallower depth of blockade.
• 2:13
  So four out of four twitches can be further subcategorized into with or without fade based on tactile detection.
• 2:22
  The presence of fade compared to no fade indicates a greater level of residual weakness.
• 2:28
  Quantitative monitoring, on the other hand, measures a train of four ratio,
• 2:33
  which is the amplitude of the fourth twitch divided by the amplitude of the first twitch.
• 2:38
  As you can imagine, this allows for more accurate representation of neuromuscular blockade, which consequently reduces the risk of residual neuromuscular blockade.
• 2:49
  In fact, at four out of four twitches, up to 70% of nicotinic acetylcholine receptors may still be blocked and fade may not be perceived by the user until the train of four ratio is below 0.4.
• 3:03
  There are several types of quantitative monitoring. The most common modalities are acceleromyography and electromyography.
• 3:11
  Acceleromyography measures the acceleration of the stimulated muscle.
• 3:16
  The downside is that it requires free movement of the muscle, so it is not reliable for measuring thumb movement where the arms are tucked.
• 3:24
  Electromyography or EMG places electrodes over the path of the nerve and measures the muscle action potentials.
• 3:32
  While it does not require free movement of the thumb, it is susceptible to electrical interference.
• 3:37
  The other two modalities aren't as clinically relevant.
• 3:41
  Mechanomyography involves placing a flexible piezoelectric film in the groove between the thumb and index finger and measures movement of the muscle.
• 3:50
  As such, it can only be deployed at the adductor pollicis muscle.
• 3:55
  And finally, mechanomyography is not even available commercially, but directly measures isometric muscle contraction force.
• 4:04
  Side note, the 2023 guidelines also had a strong recommendation for monitoring the adductor pollicis muscle over other muscles because it has the longest time to reach a train of four ratio greater or equal to 0.9.
• 4:18
  In other words, the adductor pollicis takes longer to recover, so monitoring it ensures a more complete recovery of all muscles.
• 4:27
  So now we have applied our quantitative monitor and obtained a train of four ratio.
• 4:32
  How do we interpret these results and devise a reversal strategy?
• 4:36
  The 2023 guidelines recommend using Sugamidex as the chosen reversal agent.
• 4:42
  A train of four ratio greater than 0.9 corresponds to acceptable recovery and should be confirmed prior to extubation.
• 4:51
  There are no official recommendations yet on the proper Sugamidex dosing based on train of four ratio measurements.
• 4:58
  One common approach is to dose Sugamidex at two milligrams per kilogram if the train of four ratio is between 0.4 and 0.9.
• 5:07
  If there are one to two post-titanic twitches, then Sugamidex can be dosed at four milligrams per kilogram.
• 5:13
  If there aren't any post-titanic twitches and immediate reversal is needed, then 16 milligrams per kilogram of Sugamidex is given.
• 5:21
  Neostigmine can be used as an alternative to Sugamidex if there is only a minimal depth of blockade, which corresponds to four out of four twitches with absence of fade.
• 5:30
  In that case, the dose of neostigmine is 20 mikes per kilogram.
• 5:34
  And those are the basics of quantitative twitch monitoring.
• 5:37
  Thank you for coming to my Twitch Talk.
• 5:39
  Thanks, Christina.
• 5:41
  Marcus Lehman is an assistant professor of anesthesiology that works with me here at the University of Cincinnati
• 5:46
  with a particular interest in artificial intelligence and how it can be integrated into the medical field.
• 5:51
  This grew out of his desire to make work-life balance better, or at least 10 percent less annoying.
• 5:57
  Coincidentally, he is also a reality television show star made famous by his appearances on Survivor and Labor of Love.
• 6:04
  Marcus, thanks for joining us.
• 6:06
  Yeah, glad to be here. Excited to talk about healthcare AI and just join the podcast.
• 6:11
  We've been looking forward to it for a long time.
• 6:13
  So to start, I want to ask you one of my favorite standard get to know you questions that I like asking everybody.
• 6:19
  If you could invite anyone in the world over for dinner, including somebody from the past or even from fiction, who would you invite over and why?
• 6:27
  So this is a fun question. I actually think Benjamin Franklin would be my answer.
• 6:32
  Interesting.
• 6:33
  Not only was he kind of one of the first Americans, patriots that gave all of this time and effort into creating what we now live in as the United States of America,
• 6:45
  but he was just always curious, always creating, always inventing, and just a probably super interesting guy to get to know.
• 6:54
  From musical instruments to science, he contributed in all these different ways.
• 6:59
  And yeah, just totally sounds like a guy that's always thinking about how to make life a little bit better, which is super cool.
• 7:05
  That's really neat. Yeah, I could definitely imagine Benjamin Franklin being a good dinner guest.
• 7:10
  He'd probably talk most of the time.
• 7:13
  Yeah.
• 7:14
  That's what I would be hoping for.
• 7:15
  Yeah.
• 7:17
  So we're talking about artificial intelligence, AI.
• 7:21
  This is something that a lot of people have been talking about in the medical field or just in general.
• 7:27
  But let's take a step back. What is AI? Can you talk about it in a broader sense?
• 7:34
  What are people talking about when they're talking about AI?
• 7:38
  Yeah. So, I mean, AI has been this vague term that's been around for a long time.
• 7:42
  It's basically an outgrowth of computer science vocabulary, if you will, as a way to bucket some of the different things that are being done in the industry.
• 7:52
  I think for a long time, it was more of a science fiction term that was being played out in movies that we've seen, like war games and this idea of a sentience computer system.
• 8:05
  So a non-human that is able to do human things.
• 8:09
  And in reality, this version of AI has basically gotten a little closer to that.
• 8:16
  And that's what's gotten everybody excited and talking about it.
• 8:19
  So is AI just chat GPT at this point? Have we achieved artificial intelligence with this technology?
• 8:27
  Yes. So the chat GPT is what they call a generative AI.
• 8:30
  And so there's a subset of artificial intelligence and there's a bunch of different things that you've probably heard as you listen.
• 8:36
  You've heard of machine learning and neural networks and there's deep learning and there's weak AIs.
• 8:42
  And then what we're talking about with chat GPT is that generative AI term.
• 8:47
  But Netflix uses AI to figure out what awesome movie you want to see or have never heard of or didn't even know you wanted to see.
• 8:56
  You've got other different patterns like swarm intelligence, which some companies use to help know when trucking should be done down a certain highway versus other times.
• 9:07
  And so there's all these different patterns.
• 9:09
  The super intelligence or the general strong AI is what we still have is the Holy Grail where you have a computer that can learn and can think and can incorporate things at a level that goes way beyond anything that we've ever been able to do.
• 9:25
  So one thing that I have read or heard recently is that our computers now are passing the Turing test.
• 9:34
  Can you tell me what that is and what that means?
• 9:37
  Yes. So Turing was a scientist philosopher, if you will, that was sort of thinking deeply about, OK, so what would qualify a computer to have its own consciousness?
• 9:52
  And he created a series of tests that you'd sort of have to pass in order to do that without diving too deeply.
• 9:58
  Without diving too deeply, you basically have people who are either trying to get Internet famous for a bit or who believe in some way that we've crossed a certain threshold and they use that Turing test as evidence of that being done.
• 10:12
  Now, whether that's actually the case or not, I think is is always up for debate because there's nothing that's definitive.
• 10:20
  You can't go to chat GPT and do your own Turing test and be like, oh, OK, this totally is it.
• 10:26
  Certainly things like you've maybe seen being medical, like the you know, the the chat GPT that passed the medical licensing exam, USMLE step one.
• 10:35
  So those are things that kind of start to approach because it's because humans think basically, wow, this is something that would have taken me so much time to do.
• 10:44
  And this computer can do it. And it's incorporating the ability to read questions that does start to feel human.
• 10:51
  And so I think there is just this tension now as we get closer and closer to something like that to start calling it and being the first to call it is kind of the holy grail in and of itself for being a scientist who, you know, first person to really say it was the true generative
• 11:09
  AI deep thinking conscious machine versus somebody who doesn't get to have that little first sentence in history. That's that's kind of the excitement I think around this type of stuff that are making people say that.
• 11:22
  And then usually get fired from wherever they say it. That makes sense. So how does this all work?
• 11:29
  How what is a large language model? Can you break it down for me how this sort of thing came about? How does it work? Yes. So a lot of this stuff is basically
• 11:46
  companies that have a ton of access to a lot of data, and they started creating these patterns of training that basically allow a computer system to take huge libraries of information and basically run it through an algorithm.
• 12:03
  And if you think about your own learning, it's actually pretty similar. It's just that they did this on a huge scale. And what they're doing is they take in the supervised learning model, they take some input data, they basically have the model read about
• 12:21
  something that model then gives an outcome that's expected. And then the supervisor says, No, that's incorrect. And then the model goes back, changes it, and it does it over and over again until it gets the right output every time.
• 12:36
  In an unsupervised model, basically, they take the input data, and the model looks at it and then generate something and that's sort of like, it's like math and creativity. So there are two different things where sometimes the output kind of doesn't matter. It's just that it's creating something.
• 12:52
  And then other times where the output does actually matter. And so running this over millions and millions of cycles, they basically have changed the original computer system into something that has none of that ability into something that's more and more finely tuned.
• 13:08
  And so what you know, what we think of as talking to a human who maybe makes a vocabulary mistake, the computer starting to get to point where you talk to the computer and it actually is able to really keep track of the conversation and have some, I hate to use the word intuition, but that's kind of what it is.
• 13:27
  And so we're getting closer and closer to that idea of kind of walking up to something and having a conversation with it as opposed to someone and having a conversation with them.
• 13:36
  I think that sounds both exciting and somewhat creepy.
• 13:41
  Yeah, no, I would I would argue there's, there's an aspect of this that, I mean, we've always had kind of monkey in the machine type of computer programs that you know could be a chat bot and it's not really there and you can tell.
• 13:56
  And now, I think what these large language models have been able to do is practice so many times that they get awfully close to what feels like talking to another person and that is a very exciting thing for some people and for others it's it really kind of unsettles you.
• 14:15
  Well, I'd like to confirm here that we are both persons that are having this conversation I have not invited an AI to take part in the conversation yet. Yes. As far as we can tell I mean, you know, I mean, maybe we're just real.
• 14:32
  You look real to me but maybe we're just two AIs talking to each other. You have amazing robot like hair.
• 14:40
  So what are people using AI for these days, generally speaking, what are what's the use case for. Yeah, AI.
• 14:49
  Are you using it. So there's a, there's a ton of different ways I mean people have just gobbled up all the ideas, and I think fundamentally, you know, the first idea items that you saw in the news were like, AI curing cancer AI passing the exams and all these things
• 15:08
  and I think an AI, and it has an incredible ability of writing or composing things that come across in a much more professional fashion than some people can do and I think, ultimately, where AI is right now and the way I'm using it is giving someone
• 15:29
  or myself who understands a subject, an ability to write out initial drafts more quickly, or to create certain types of documents that don't really require a lot of understanding.
• 15:42
  And that's kind of like where, where I'd say the line is drawn for me.
• 15:49
  There are other companies that are using it to understand audio data, and then compile that data and it's like a meeting summary for instance, there's all kinds of uses. And then the part that nobody really is like dives into at least yet in the medical space is the
• 16:05
  kind of like visual stuff so like creating photos and all that stuff is pretty wild to see what it can come up with and I'm sure people have seen already you know the Pope and the Balenciaga jacket kind of photos that are being generated and that's actually where some of the
• 16:20
  conflict kind of comes into because it's a very powerful tool, and it's like Adobe Photoshop, you know, on steroids and those are the types of things that can get out of hand before anyone kind of organizes the whole kind of set of principles that are needed to
• 16:36
  probably guide some of this utilization. So I confess actually the logo for the podcast was something that I designed with AI. It took me a lot of time and iteration but using the AI photo generation I was able to come up with something
• 16:49
  that I thought was appropriate and that I didn't have to pay an artist for frankly.
• 16:54
  Yeah, I think, I think the ability for novel output to be created by a layman, literally in my case since that's my last name is pretty wild and, and I think in some ways. I think about AI as your life you have to deal with all these different programs.
• 17:14
  And those programs didn't exist before so like we could have never made a spreadsheet back in the day, or created an email because you'd have had to like literally write the program to then do the email so somebody did all that for us.
• 17:26
  But now, you have to be a certain level of expertise to really get value and produce out of these programs. And what generally is going to be doing I think is being able to take all those programs and manage them for us, which is kind of the next evolutionary step if you think
• 17:44
  about it because people have been frustrated or at least I've been with the idea of like, I know how to make a presentation, but I don't know how to make it look like the way I want it to.
• 17:56
  So, I have the knowledge for something, I just can't accomplish the end result without an incredible amount of either communication with professional expense, or, you know, hours and hours of training myself to get good at this stuff and you chat GPT has the promise or generally
• 18:14
  I has a promise of accomplishing that for us. So what's the most interesting thing that you've recently used chat GPT or AI for in your personal or professional life.
• 18:27
  So one of the most interesting things that I think I've looked at recently are some AI billing capabilities that have been brought to my attention. There's a company expert docs that does some really high quality billing processing for healthcare providers
• 18:50
  and that's just a really big deal because part of the, you know, joy at work is actually getting remuneration for your effort. And so it's a huge issue of course, in the industry, but part of that process is, you know, whenever, as you can imagine, whenever you do something if you get a denied
• 19:08
  response that's really frustrating. And it's been impressive to watch this large language model that can basically go in, it sort of already understands the basic framework and it's done enough reps on billing that it understands how to do things
• 19:23
  but every office has slightly different needs and each state has different requirements. And so it goes and it trains on the office needs a little bit more and gets those kind of lined up such that supervised learning we talked about. And within a couple of weeks, it's able to basically take the now rate and reduce it to like less than 1%, which is shocking.
• 19:45
  And then the speed at which it processes stuff is probably four or five times faster than the human billing operator. Now, probably not a good sign for the community of people that are expert billers, but you still need that expertise to be able to know whether this thing's completely bonkers or doing the right thing.
• 20:04
  But it just really it's incredible for imagining like a clinic that is, you know, at a small outside place could take this device, use it, and, you know, maybe they can't afford the kind of billing expertise that they need and now they're able to stay open because of something like this.
• 20:21
  And so that's fascinating to me because it's a real world application. And it's not trying to like say it's not trying to cure cancer, but it keeps the guy who is curing cancer showing up and doing the work and not having to spend all this extra time afterwards, you know, away from their family or any of the other pieces that we as physicians often experience.
• 20:40
  When I think about sort of the crossroads between AI and anesthesiology, it makes me think about set as this, which maybe you can help explain to our audience what set as this is or was.
• 20:56
  Yeah, set as this is a it's interesting because it actually a lot of users were in Ohio. So, you know, we're cutting edge state, even though most people don't think of it that way. But it was a system that was touted as being able to manage the sedation.
• 21:13
  It was a sedation system, hence the set as this name for effectively minor procedures, mostly focused on GI suites like colonoscopy and EGD. And really, they only had FDA permission to do really, really healthy patients.
• 21:31
  But the goal of it was, you know, this thing gets started, it's able to get feedback from the blood pressure and the heart rates and patient movements and then dials in some propofol.
• 21:46
  And, you know, the next thing you know, the procedure is done. And so now you don't need this expensive CRNA or anesthesiologist doing the procedure and you're done.
• 21:57
  And then the company benefits because they're using kind of a disposable aspect of their products. So then they get, you know, kind of every time you do a procedure, they get get some money for that.
• 22:08
  So the problem was it is and this isn't like a fault of the tech. I think really the technology itself was fairly useful. But, you know, if you think about how many purely exceptionally healthy people are showing up for colonoscopies or EGD, you can get healthier people in colonoscopies and EGDs.
• 22:29
  A lot of people need EGDs for certain reasons. So set us this couldn't do a lot of the patients in the first place. And then the way that you get paid for these procedures actually doesn't work very well for the system because it's it's paying the provider based on having an anesthesia component.
• 22:51
  And so you were you're basically not getting paid a big chunk of the reimbursement for doing the procedure because you didn't have an anesthesia provider there.
• 23:01
  And so you're not just paying for that lost income. You were actually then also paying set assist to use their machine. And so it's maybe a story of like a little too cool too soon.
• 23:12
  But, you know, it starts to hint at what some of the sort of this is more like a machine learning kind of characteristic AI would be able to accomplish for people. And again, I think it has a fate to of being way ahead way before generative AI.
• 23:30
  And so this this felt a little bit like an invasion into the art of medicine as well. So I think there were some users but it wasn't exactly like you would have needed massive adoption across the country or for this thing to have the legs.
• 23:45
  And so it eventually just petered out. And from what I recall they had some bad outcomes too. Is that right? Yeah. I mean it didn't take much.
• 23:54
  I think if you think of Tesla every time there's an accident it becomes like major world news. And the reality is like it's having far fewer accidents and human beings on a daily basis.
• 24:04
  But like that that is that activation energy that's required for new technology. It has to almost be perfect for a really long time before people will acknowledge that part of humanity is not as good as the machine basically.
• 24:22
  Sure. So what are some ways that we could use AI and anesthesiology?
• 24:30
  Yeah, there's great opportunities. I think we really, you know, I focus on the overall like work life balance of anesthesia and some of the requirements that are needed that we might not be accomplishing.
• 24:44
  I think one of the things is less to do with like putting the breathing tube into the patient. It's more to me about hey why don't we give everybody a letter about how hard or easy it was to put that breathing tube into the patient.
• 24:58
  That would be really helpful. First of all if they ever have it again. But second of all just so that they can understand a little bit more about their care.
• 25:05
  And so I feel like we have an obligation to do that. And personally, you know, that's not really accomplished very often even if we have a bad airway. It's still like another thing that has to get done.
• 25:16
  If you think about it and many of our patients don't necessarily understand vocabulary at a collegiate level.
• 25:24
  That's also making it hard for me to write a letter at sort of an eighth grade level. And you can imagine using chat GBT to take off this this burden and just write that letter for you.
• 25:37
  Oh, and by the way, if I just say now do it in Spanish, it'll do that in Spanish. I say let's do it in Farsi. It'll do it in Farsi. So in moments, I can imagine achieving a kind of communication with my patients that's totally unheard of in the past.
• 25:52
  And like is it perfect? No. But compared to zero, it's an incredible step forward for a lot of people. And I know it sounds silly, but it's a small thing that we kind of just don't do right now because as a clinician, you know, we tell ourselves, well, we're busy doing the work.
• 26:07
  And yet this is we've already done 95 percent of what we could do for the patient. And this little extra five percent might save lives down the road or at least give them a sense of what the process was in order for them to go through whatever procedure they're having.
• 26:21
  And then the next anesthesiologist who has to like re-explain intubation and they go through some anxiety like I had this over the weekend, you know, patient just doesn't realize what's happening to them after they're put to sleep worse.
• 26:34
  And we have an obligation to try and figure out how we make that easier and easier for folks. And this is definitely a great solution for that.
• 26:40
  One of the things that I think about that AI might be used for in the distant future or maybe in the not too distant future is recognizing some of the rarer complications that happen within anesthesiology.
• 26:54
  For example, there's a rare condition that I know that you're familiar with and a lot of our audience is familiar with called malignant hyperthermia, which in the middle of an operation, it can start with a high heart rate.
• 27:07
  And the patient's blood pressure will change and the temperature of the patient goes up and the end-tidal CO2 goes up.
• 27:18
  This strikes me as something that an AI who might be monitoring the patient with us could recognize perhaps even earlier than an anesthesiologist could.
• 27:30
  Maybe a pop-up could come on the screen and not necessarily tell us what to do, but, you know, point us in the right direction if it sees a particular pattern that it's seen before, if it's trained on a very large set of patients.
• 27:46
  Yeah, I mean, it definitely fits in the framework of, you know, you're going to learn or you're all learning all these different algorithms and centaur criteria and all across medicine.
• 28:02
  You have basically a series of checkboxes and having a system like this that can not just manage the ones that you have in your head, but manage every single one all concurrently is definitely a feasible thing.
• 28:15
  And even to build on your idea, you know, people that meet certain criteria for alcohol dependence, there's stuff that we just have floating back there that, oh, gosh, this guy maybe needs something else or we've got to be careful about this.
• 28:28
  But a system that is actually able to analyze the record and give you some of that stuff in a more official fashion so that our minds capture it and are much more able to treat it appropriately could be part of what you're talking about building.
• 28:47
  So what do you think are some things that you don't think AI could or should be used for within anesthesiology?
• 28:57
  An interesting question because I think it's it's fun to imagine where it could go.
• 29:05
  It's hard to talk about potentially losing your job in a sense to machine. I certainly think there's a component of people who are attempting to figure out how to use this to actually have medical care delivered by an AI and it could be as simple as diagnosing UTIs, which
• 29:28
  when I was in medical school, it became known that you could just diagnose it over the phone, which before it was a huge advance for people who really have problems with this.
• 29:38
  But that kind of makes you think, you know, is that next phase going to happen where we start replacing certain medical interactions using a generative AI that is fairly consistent?
• 29:49
  And, you know, there's a article that came out months ago about how it was even more empathic than many doctors because it has no ability to get frustrated, basically.
• 29:59
  So those are those pieces that I think you can imagine it doing those types of things.
• 30:05
  I don't see it being able at this point without a lot of tech being able to actually do any of the physical work that's required anesthesia.
• 30:14
  I mean, I think not not say, hey, we're safe and nobody else is, but there is a manual component to all of this that is really, really difficult to to overcome from a robotic standpoint.
• 30:26
  And so, you know, there's there's still going to need to be humans involved in a lot of this process.
• 30:31
  But, gosh, you know, it's hard not to imagine it really impacting overall workflows in the perioperative space just because of how many different places people are doing work that isn't necessarily a huge value add.
• 30:47
  Like a nurse asking the same question for the fiftieth time. Right. And stuff that could be auto imported using an AI that's just reading through the chart and consolidating that into the preop check checklist or check in.
• 31:01
  And so I think that's where I see it using becoming incredibly useful.
• 31:06
  And then, you know, again, you still have that that issue of adoption, I think, for people having a hard time letting a machine do something and you really have to make a concerted effort to show the data and the improvement for people.
• 31:22
  Well, I've never personally been frustrated with any of my patients, so I'm not sure what you're talking about there. But, yeah, I could imagine a scenario where it would be nice to have perhaps an AI augmented preop that would do some of the work or maybe some of the chart digging for me a little bit.
• 31:43
  Sounds like an interesting idea. So why can't we do this tomorrow? What are some of the barriers that we face before we start integrating AI into our practice models?
• 31:54
  So actually, there are some companies that are doing it now. There's there's a couple that are out there that are taking the roots of having a mic in the clinical room and then just recording the conversation and distilling that into a soap note.
• 32:08
  And, you know, I think there's varied success with that. I think one of the biggest challenges with this integration is that the companies that we use for medical records are not super ready to just let a bunch of companies loose in their chart.
• 32:27
  And you have, if you think about it as a patient, you have some concerns about these machines that are gathering information and are changed by the information. So which means they're holding some type of your information in their systems because that's how the language models grow.
• 32:45
  So you've got some data concerns as well. And then I think another big piece of this really is that idea that, OK, you're going to take this thing, you're going to use it.
• 32:59
  And I bet you're definitely going to look at the notes the first few times. But then as it gets easier and your confidence grows, you stop becoming vigilant about what something or someone else is doing.
• 33:11
  And we already do this, right? I don't look twice at my billing because I trust that the person that is doing it is doing it well.
• 33:18
  Well, the reality is that if there is a mistake and that comes out of something that you didn't review and it's a significant one, all of a sudden, like the experience you're going to have is going to be extremely negative.
• 33:31
  I think there's a huge hesitation around that still. And that's what the adoption curve is for, right? Like there's people who don't care at all. Maybe that's a clinic that doesn't have a lot of downside, like a dermatology clinic that can kind of capture a lot of information and is doing, you know, I don't know, gels for the face.
• 33:50
  There's not a lot of risk for doing that compared to a surgical clinic where, you know, you get this wrong and it's not in the note correctly and that's a huge, huge problem.
• 34:01
  Those are just kind of the pieces that I look at in the big picture of, you know, how does this work? And then what are the really hard things that we're going to have to overcome?
• 34:10
  And again, I'm sort of agnostic on like some of this stuff because I do think it's going to take longer to do it well and I would rather it be done well than to just be kind of overrunning the system.
• 34:23
  And you don't know where an AI is recording your conversation and where it's not and you don't know, you know, what's automated and what isn't.
• 34:31
  But that being said, you can look around every day, I bet, and see things that machines are doing for us that they don't do a great job of. If you've ever tried to do a text while you're driving by just talking to your iPhone
• 34:43
  and you probably see all these extra curse words because half the things it does is just completely wrong. And you're like, how can this iPhone in 2024 truly not understand like my words when I can, you actually, and I have talked about this
• 34:59
  because you can hop on chat GPT and have a conversation with it and it's unbelievable how clearly it understands what you're saying. But all of that rub is real. And it's it's if you look around, it's been happening to us, you know, already for a long time.
• 35:13
  So just an interesting adaptation of technology and all the pieces that come with it.
• 35:19
  Yeah, we're speaking to each other in August and my understanding that is when the next iPhone model comes out and the next operating system drops, there's going to be a large AI component to it as well.
• 35:33
  So maybe that'll fix all of our problems or create a bunch of new ones.
• 35:38
  Exactly. Maybe maybe it'll start writing the text for us and we didn't even ask you to.
• 35:44
  So I'm not sure if my wife would appreciate if I started having AI just respond to all of our texts and probably wouldn't be right.
• 35:55
  Probably not a good plan. Yeah.
• 35:59
  So the way that I sort of see this is we continue to move forward with technology within anesthesia. So you see ultrasound guided nerve blocks that have largely replaced the landmark technique video laryngoscopy is from what I'm seeing starting to replace direct laryngoscopy.
• 36:18
  So we're seeing better first pass success rate with more technology. Do you see AI starting to replace some of the decision making process for anesthesiologists? How much propofol or atomic that we give to a patient? What pressure we should put them on?
• 36:35
  It's a great question to consider. You know, I break this down a little farther. Like, why do I even give the amount of propofol I get right now? If you're honest about it, it's probably because it's what's in the syringe for most people.
• 36:49
  And I think you're you're certainly in the right calculation. So I usually do it. You know, we have students that you see. And so I do it, you know, based on the body weight and stuff. But you start understanding how other factors affect your decision making.
• 37:05
  And, you know, I read that in a book in terms of what is my actual dose for propofol. I believe the book because, you know, somebody else told me about it. And there's an apprenticeship aspect to all this.
• 37:16
  The addition of technology to me certainly has all the risks that we've talked about. But I think AI could do a lot for right sizing what we're doing. So it becomes much easier if, you know, imagine this in the morning, you show up and the AI not only helps you.
• 37:37
  We have a computer system that already takes the patient's record and puts it into our PIXIS that allows us to get the medicine out. You know, how nice if it tells you how many vials to pull out per patient.
• 37:49
  And it's a lot more accurate than what our current method is. And I think you'd need studies to show that. But I do think that the AI would potentially give us the ability to really get better at what we're doing.
• 38:02
  Because I think it would be built, if you build it, you know, by anesthesiologist for anesthesiologist, you're going to get that right sizing. Another example might be the ability to show on an ultrasound, like the nerve block, right?
• 38:15
  Here's the nerves and they can label the nerves for you. And it's going to be right because it's going to be built by us. We already know the nerves, but man, how much better if it highlights them.
• 38:27
  There is already AI technology being used by radiology to help accentuate their ability to pick up on issues and scans. And so I think you're going to have that validation over time and you're going to start seeing it in very small ways.
• 38:42
  It's not going to drive the overall piece. I think it's just going to right size, little bits and pieces of it over time. And I think that's going to be the real win for us.
• 38:52
  I think for people that are approximately our age and our late 30s, early 40s, etc. A lot of us are sort of grew up with the Terminator movie series and we've seen the Skynet sort of problem where AI gets out of control.
• 39:11
  Maybe for our younger audiences or medical students, they might recall Ultron from the Marvel movies, essentially where an AI gets out of control and almost kills all of humanity or does.
• 39:31
  This sounds scary and I think a lot of us have a deep seated fear of AI in some way based on some of the science fiction, but the science fiction is based on some sort of potential real scenario, right?
• 39:44
  I guess what I'm trying to get at is how do we weigh the upside versus the downside of integrating AI into our practice?
• 39:55
  I think that's a very difficult answer. Certainly not sure that I have one that's any better than the next person's, but I think a little bit about the it depends response because you have to look a little bit about the timeline.
• 40:10
  The first time you got on a bicycle did not go well, neither did the first five times after that, but then eventually riding a bike is incredibly liberating for you as a kid and as an adult.
• 40:25
  I think the riding wheels aspect to all of this stuff that is actually very uncomfortable for everybody and there may be times in the phase of humanity that a lot of people have more trouble than in other phases.
• 40:40
  The periods of the World Wars is a very difficult time that you and I have never had to experience. That aspect of this, and I'm not comparing this to a World War, unless the AI makes one for us maybe.
• 40:55
  I've seen War Games too.
• 40:57
  Exactly. But it's more like the application of this is not going to be simple or straightforward and it has that potential to be damaging. But I think there's just a lot of other things in life that are exactly that same scenario.
• 41:13
  And for many of us it's trying to understand if this is still within kind of the risk reward benefit that we are comfortable doing. Some people may not want their kid to ride a bike because they live in Manhattan, and that's just way more complicated than riding it out in Mason, Ohio, where you're in the middle of a neighborhood that has a car down the street maybe once every 20 minutes.
• 41:35
  And those things are going to be, I think, the scenarios where some people are going to start adopting this and other people are going to say, hey, this is too much for us.
• 41:46
  I don't think it's going to be in our lifetime that you have complex computer systems, generative AI type models or whatever comes after running entire hospital systems or covering entire electric grids that are able to do kind of a, I decided to shut this down because you guys are misbehaving at night.
• 42:08
  But some people might argue that that in certain parts of the neighborhoods, like having no electric lights would actually benefit keeping people off the streets and hurting each other.
• 42:18
  And those are going to be the kind of debates that we're going to have when we do get to that point, and that's going to have to be kind of a decision point that probably like I said I don't see that in our lifetime even though people say we're going to have, you know, deep thinking
• 42:30
  AIs and all of the war games level type stuff by 2030. I think your society is not ready for that for another couple of decades, I'd say.
• 42:42
  Well, I hope you're right.
• 42:45
  It'll be interesting to see where all this goes but one thing that I think we can say for sure is that this isn't going away. I don't think AI is a fad I think it's a powerful tool, maybe akin to like when calculators first started coming on the market, except for
• 43:02
  language. So,
• 43:06
  if the AI is listening to this conversation in the future. Know that we were in general supportive of AI and
• 43:17
  give us mercy robot overlords if you're listening to this conversation in the future.
• 43:22
  So, lastly, what style points or points of personal preference have brought you success in your career in and out of the operating room.
• 43:32
  I think I have to give a plug to my reality TV experiences. I think it's fun to have played survivor and kind of understand a little more about how the world works through doing that.
• 43:47
  And one of the lessons I remember learning very quickly in my season was that, you know, merit is very important like doing well being there being present pitching in, but life's not completely about that. And I think there's there's a teamwork and a camaraderie.
• 44:03
  That's also required. So that's where you need to be personable and you know, learn about your, your peers and your partners. And then lastly, sometimes life doesn't care about any of that stuff. And so you may not get the promotion you're looking for or have things work out in your favor.
• 44:18
  And it just, it doesn't mean that, you know, we're on survivor that might get you voted out and that that's the end of the game, it just, there's another game that you'll then get to play.
• 44:27
  And so you just have to kind of keep working through that sense of reality. And I think before I had those experiences, I sort of felt like, okay, if I just work hard, keep my head down, like, don't do anything different.
• 44:40
  You know, everything's going to work out and I think you have to realize that a medical career is hard. It takes a lot of different parts of you as a person it takes your intelligence, it takes your hard work, but it also takes your personality and your kindness and your ability
• 44:56
  to be a good person to people. And then you've got to balance that with all the other things that are going on in your life like if you want to start a family. And I think for me it's kind of understanding that juggle is real.
• 45:09
  You know, it's not an easy thing, and it takes all the parts of you and you won't always do it well. And sometimes we get voted off. And then sometimes you'll stay on for another round and you just keep kind of keep going and making sure that, you know, when you show up to the
• 45:23
  stage, you'll get voted off of that. And so I think it's going to be a really special day that you've got some space for your patients to be cared about and that when you get home, you've got some space for your loved ones to be cared about.
• 45:36
  can become an anesthesiologist after that.
• 45:40
  So I definitely have moments where I will dream
• 45:43
  about being back on Survivor literally
• 45:45
  and trying to navigate the situation differently.
• 45:49
  I think if I had won a million dollars on Survivor,
• 45:52
  I'd have taken a really good hard look
• 45:54
  at whether residency or the rest of residency
• 45:57
  was really worth, juice was worth the squeeze.
• 46:01
  I do think it's worth considering if you ever,
• 46:06
  want to pursue something else outside of medicine.
• 46:08
  Like I did learn that it's safe to take a year break.
• 46:12
  It's better to know that you want to be there
• 46:14
  than to always wonder if you don't.
• 46:16
  And so that was one of the good lessons also.
• 46:18
  I took that time after my first year residency
• 46:21
  to explore that different direction.
• 46:23
  And I think it made it easier for me to kind of decide
• 46:28
  when I got back into the residency program
• 46:30
  that I was there for the right reasons.
• 46:31
  So definitely recommend that even though I wasn't there
• 46:34
  for the right reasons with a million dollars,
• 46:37
  I was there for the right reasons.
• 46:39
  Thank you so much for joining us.
• 46:41
  Glad to be here. Thanks for having me.
• 46:44
  Well, that's it for another episode of Style Points.
• 46:47
  If you enjoyed our discussion today,
• 46:49
  please don't forget to like and subscribe
• 46:52
  to stay updated with our latest episodes.
• 46:55
  Your support helps us keep bringing valuable insights
• 46:58
  right to you.
• 46:59
  Thanks for listening and we hope you join us again next time.
• 47:03
  Take a deep breath, open your eyes, we're all done.