Latest / CareTalk: Healthcare. Unfiltered. / One Giant Leap for Healthcare AI w/ Dr. Robert Wachter, Author, A Giant Leap

Transcript

One Giant Leap for Healthcare AI w/ Dr. Robert Wachter, Author, A Giant Leap · CareTalk: Healthcare. Unfiltered.

• 0:07
  John: Welcome to Care Talk America's home for incisive debate on healthcare, business, and politics.
• 0:12
  John: I'm John Driscoll, the chairman of the Yukon Health System, and today I am delighted to have my friend, Dr. Robert Wachter, the famed author and prognosticator about technology and healthcare, the father of hospitalist medicine, and most recently, the author of A Giant Leap.
• 0:32
  John: How AI is transforming healthcare and what it means for our future.
• 0:37
  John: Jumping on the AI bubble, Bob, congratulations.
• 0:41
  Dr. Wachter: Of course.
• 0:41
  Dr. Wachter: Thank you, John.
• 0:42
  Dr. Wachter: If anything, I'm a decent trend spotter.
• 0:45
  John: Well, it's an interesting time to be doing this, but maybe we'll go back a little bit.
• 0:50
  John: 'cause when you're also the author of the Digital Doctor, uh, it, it, that seemed, you seemed a little bit more of a technology skeptic, uh, at that phase of, of the intersection or the crash of technology into healthcare and medicine.
• 1:07
  John: Um, what's changed now with ai.
• 1:11
  Dr. Wachter: Well, let me start with, I think I was a technology optimist in the beginning because healthcare was so late to the dance in technology that for a decade or two I'd been waiting for technology to enter our world.
• 1:23
  Dr. Wachter: And then it did in about 2010 or so when we all got electronic health records.
• 1:28
  Dr. Wachter: And that's when I turned into a technology, uh, grump because I saw the EHR come in and with high hopes.
• 1:37
  Dr. Wachter: And then I saw all of my colleagues unhappy at, as it had turned them into pretty expensive data entry clerks.
• 1:44
  Dr. Wachter: And then patients got their own version of the electronic health record, the patient portal.
• 1:48
  Dr. Wachter: And it gave them tons of information, but no useful intelligence, but it gave them a little button that they could send a little message to their doctor.
• 1:55
  Dr. Wachter: So their doctors now were spending all night answering patient messages.
• 1:58
  Dr. Wachter: So, um, I was not so much a, a skeptic, but I was disappointed.
• 2:03
  Dr. Wachter: In how technology had failed to transform healthcare in the ways that I had hoped.
• 2:08
  Dr. Wachter: And that's the, I think that's the zeitgeist that came through in the digital doctor 10 years ago.
• 2:13
  Dr. Wachter: At the end of that book, I have a chapter that people said to me, uh, who wrote this because this chapter's really optimistic.
• 2:18
  Dr. Wachter: And I said, no, it's me.
• 2:20
  Dr. Wachter: I can see where this is going.
• 2:22
  Dr. Wachter: But it's clear that the electronic health record was not the answer.
• 2:25
  Dr. Wachter: It was the scaffolding.
• 2:26
  Dr. Wachter: And the answer was going to be taking all this data.
• 2:29
  Dr. Wachter: That we have in now in digital form and figure out how to analyze it, understand it, and use it in ways to make healthcare better and safer and cheaper and more accessible.
• 2:40
  Dr. Wachter: And as soon as on November 30th, 2022, when I tried chat GPT for the first time, it was like, okay, this is it.
• 2:46
  Dr. Wachter: And so I kind of went into this journey over the last couple years.
• 2:51
  Dr. Wachter: Uh, leaning a little bit toward optimistic, partly because this technology is unlike anything we've had before, but partly because the system is so incredibly screwed up that I really feel like the only way that we're gonna make it better is with the use of, of this new ai.
• 3:06
  Dr. Wachter: And so that's kind of how I got into the writing of the book, and I came out of it even more optimistic than I, than I was when I started.
• 3:14
  John: So technology kind of created, you know, the 12 to 15 extra hours today.
• 3:19
  John: I would argue that that's even light given, um, what we're seeing of pajama time that doctors need to do order entry, and that's outside of the disconnection that many of them patients and doctors feel when doctors and PAs and techs and nurses are constantly rattling away, entering information is in sort of the, the.
• 3:43
  John: Sort of the, the sort of tech te health tech.one.
• 3:47
  John: Um, can AI actually help unscrew up what's screwed up healthcare from technology where really the patients and doctors are both unhappy with that, how technology sits between them and communication?
• 4:02
  Dr. Wachter: I think so.
• 4:03
  Dr. Wachter: I, I think if we get it right, and that's an open question whether we will, but I, I, my sense is that we.
• 4:08
  Dr. Wachter: At least so far are, um, if you think about it, uh, people blame, you know, your favorite electronic health record vendor, whether it's Epic or Oracle now, or, or any of the others.
• 4:19
  Dr. Wachter: It's, it really wasn't their fault.
• 4:21
  Dr. Wachter: It was now you had a, an electronic vehicle where the doctor was recording all of their information.
• 4:27
  Dr. Wachter: And of course, everybody who'd been looking over my shoulder over my entire career who wanted me to document things a different way, whether it was my billing department or the malpractice department now had a mechanism to make me do stuff.
• 4:40
  Dr. Wachter: And so the EHR became this vehicle of checklists and forcing functions to make people document without anybody taking a step back and saying, does this do a damn thing to help the patient?
• 4:52
  John: Well, I mean, honestly, honestly, Bob, one of the things that I, that struck me when I first started to look at this, you know, epic, Cerner or ECW, whichever, it, it really turns the patient doctor experience into a billing encounter
• 5:06
  Dr. Wachter: to some extent.
• 5:07
  Dr. Wachter: I mean, I, I think that's often the party line, I'd say, John, about, about how bad the systems are.
• 5:14
  Dr. Wachter: The systems really are useful clinically.
• 5:16
  Dr. Wachter: I mean, the fact that I can look at a chart at the same time, a colleague.
• 5:20
  Dr. Wachter: A specialist, seeing the patient can look at a chart is great.
• 5:22
  Dr. Wachter: The fact that I don't have to go to the x-ray department to look at an X-ray, I can see it on my computer.
• 5:26
  Dr. Wachter: There are a lot of things.
• 5:27
  Dr. Wachter: The fact that I can do an electronic prescription to, to Walgreens is, is great.
• 5:31
  Dr. Wachter: Uh, but you're right.
• 5:33
  Dr. Wachter: It became, it became a transactional engine, whether and billing being the dominant thing.
• 5:39
  Dr. Wachter: And, um, I think we kind of lost sight of what the point of all this is.
• 5:43
  Dr. Wachter: The point of all of this is how do you take care of human beings?
• 5:46
  Dr. Wachter: Uh, either prevent bad stuff or.
• 5:48
  Dr. Wachter: Treat the bad stuff that they have.
• 5:50
  Dr. Wachter: And so I do believe that the, that AI has the capacity, if we don't screw this up, to take away a, a bunch of the administrative burdens.
• 6:01
  Dr. Wachter: And, and just, I'll give you a few examples.
• 6:03
  Dr. Wachter: Today at UCSF where I practice, uh, every doctor has access to an AI scribe.
• 6:09
  Dr. Wachter: That's a tool that didn't really exist two or three years ago, three years ago, uh, our busiest outpatient clinicians had a human scribe.
• 6:16
  Dr. Wachter: Had a, had a pre-med student hanging out in the room with them for 30 bucks an hour typing away.
• 6:21
  Dr. Wachter: And now all of us have access to an AI scribe.
• 6:24
  Dr. Wachter: And it's not perfect, but it's pretty, pretty darn good.
• 6:27
  Dr. Wachter: And so I put the phone out and I guessed the patient's permission.
• 6:31
  Dr. Wachter: And then I have a conversation, human to human looking, the patient in the eye in a way I couldn't do before.
• 6:36
  Dr. Wachter: It saves a few minutes, but I think more importantly, it humanizes the conversation.
• 6:41
  Dr. Wachter: We now all have access now to a chart summarization tool.
• 6:45
  Dr. Wachter: One out of five patient records is longer than Moby Dick.
• 6:49
  Dr. Wachter: That is 600 pages, and the idea that I will be able to read and review your record.
• 6:54
  Dr. Wachter: Of 600 pages in the three minutes I have to do so is impossible.
• 6:58
  Dr. Wachter: So there you have, and we're beginning to experiment, probably not ready for prime time AI drafting your, the response to an A message you send through the EHR inbox.
• 7:08
  Dr. Wachter: For example.
• 7:09
  Dr. Wachter: Obviously we've all heard about the AI drafting prior auth requests to the insurance companies.
• 7:14
  Dr. Wachter: Uh, these are some of the ways that AI today is beginning to take away some of the paperwork burden, which I think will rehumanize medicine.
• 7:23
  John: So is it unfair though when you, in the book, you sort of suggest, look, we've actually invested in a, a fair amount of this technology.
• 7:30
  John: You really only point to a five to 10% improvement in productivity and a fair amount of money has been invested in AI, scribes and some of these productivity tools.
• 7:41
  John: Or is, is there just a natural limit to the productivity gains we get from these tools?
• 7:46
  John: Or are we just in the early stages of this tech transformation?
• 7:50
  Dr. Wachter: I think we're in the early stages and I think we're doing training wheels on the, uh, in these kind of administrative paperworky functions.
• 7:59
  Dr. Wachter: They're really important and the real mileage there will be not only saving me as the frontline physician time, but taking the thousand people in our billing depart.
• 8:09
  Dr. Wachter: And cutting that down to, you know, maybe a couple of hundred or taking the thousand people in our call center and, and, and, and making that much more efficient.
• 8:19
  Dr. Wachter: But I think there's an old saying in medicine that the most expensive piece of technology in a health system is the doctor's pen.
• 8:27
  Dr. Wachter: Uh, now it's not the doctor's pen anymore, it's now the keyboard.
• 8:30
  Dr. Wachter: The point is that physician salaries represent about 10% of healthcare costs, but the decisions that physicians make represent about 80%.
• 8:38
  Dr. Wachter: So where we really start seeing productivity gains and efficiency gains, and I think quality gains.
• 8:44
  Dr. Wachter: Or when we get to something that's harder than an AI scribe, which is really useful, thoughtful, um, uh, decision support that guides me to think of the right diagnosis.
• 8:57
  Dr. Wachter: Doesn't make it for me yet, because at this point we want to human making it.
• 9:01
  Dr. Wachter: We can talk more about that.
• 9:02
  Dr. Wachter: Uh, and then once we have a diagnosis says, you know what, there are two alternative medications here.
• 9:07
  Dr. Wachter: One's 5,000 bucks and one's 500.
• 9:10
  Dr. Wachter: And here are the 10 studies that say the 500 is just as good.
• 9:13
  Dr. Wachter: And so it guides me to the best, most cost effective care.
• 9:16
  Dr. Wachter: That's harder.
• 9:17
  Dr. Wachter: A lot of value judgements go into that, but I think that it was actually smart for us not to start with that.
• 9:23
  Dr. Wachter: You wanna start with low hanging fruit.
• 9:24
  Dr. Wachter: You wanna start getting the doctor's winning the hot doctor's hearts and minds.
• 9:28
  Dr. Wachter: And you do that with things that take a little work off their plate, like scribes and chart summarization.
• 9:32
  John: But one of the interesting, really, I think one of the most illustrative, um, examples you have in the book is the extraordinary failure of epic's, sepsis.
• 9:44
  John: Prediction and warning tool sepsis as many people know, is ex extremely, uh, an extremely toxic condition that if have grabbed earlier for patients who are at risk for sepsis, can be treated pretty easily with, or re pretty, pretty successfully with antibiotics.
• 10:01
  John: But if cut like can, can kill people and it kills a lot of people that it really shouldn't.
• 10:05
  John: Gosh, it seemed like that was such an obvious tool.
• 10:08
  John: I've seen lots of great predictive algorithms there.
• 10:12
  John: And yet the, perhaps you could go into kind of the failure and then, and then what, what appears to be, uh, the suc successful adaptation of the machine in the man to, uh, what's still a, a very, you know, dangerous human problem.
• 10:26
  John: But there appears to be some success with technology now.
• 10:29
  John: So how are we doing in sepsis?
• 10:31
  John: How did it, what did we, what did epic get wrong?
• 10:33
  John: And what are some of the doctors and systems getting right now?
• 10:36
  Dr. Wachter: Yeah, I mean, I think the sepsis example is a good example of, uh, that this is harder than it looks, and there's a lot of human factors and man machine interface that you have to get right in order to, to nail these things.
• 10:50
  Dr. Wachter: Um, this was, I think, premature on Epics part.
• 10:52
  Dr. Wachter: I think Epic had, you know, the, the dominant player in the electronic health record world, they're collecting all this data.
• 10:59
  Dr. Wachter: As you correctly frame sepsis as an important problem.
• 11:02
  Dr. Wachter: Sometimes one that's missed and one that's treated if treated early, makes a big difference.
• 11:07
  Dr. Wachter: So Epic said, well, let's roll out this decision support tool.
• 11:11
  Dr. Wachter: Um, turn, you know, flip this switch, and, and the system will ping the doctor and say, doctor, it looks like your patient might have sepsis.
• 11:20
  Dr. Wachter: Sounds good, sounds easy.
• 11:22
  Dr. Wachter: Hundreds and hundreds of hospitals around the country turned it on.
• 11:26
  Dr. Wachter: Uh, it had been really un under, I'd say even untested.
• 11:31
  Dr. Wachter: Not, not that, not that the algorithm was untested, but the algorithm and how effective it would be and what it would do in the real world of clinical practice was untested.
• 11:41
  Dr. Wachter: And what people then found was this thing was sending off false alarms all the time about a ratio in one study from the University of Michigan, a ratio of about a hundred to one, which is unbelievable.
• 11:53
  Dr. Wachter: Now you might say.
• 11:54
  Dr. Wachter: Alright, I, I want to tune the algorithm so it never misses a case of sepsis and no big deal if it pings people unnecessarily, but that turns out to be wholly wrong.
• 12:05
  Dr. Wachter: It's wrong because if I have a patient with congestive heart failure and the algorithm says this patient has sepsis.
• 12:11
  Dr. Wachter: And I say, oh my God, I gotta infuse a whole lot of fluids.
• 12:15
  Dr. Wachter: I can kill the patient if they have congestive heart failure.
• 12:18
  Dr. Wachter: Um, the other problem, probably the bigger one, is the boy who cried wolf, which is after a hundred falls alarm, alert, fatigue, alarm alert fatigue.
• 12:24
  Dr. Wachter: You're just alert fatigue.
• 12:25
  Dr. Wachter: You're not gonna pay any attention to the, to the next alert.
• 12:28
  Dr. Wachter: And the elect next alert might be real.
• 12:29
  Dr. Wachter: So most.
• 12:30
  Dr. Wachter: Now what Epic did was made this easy, and in some ways it was too easy.
• 12:34
  Dr. Wachter: And I think the lesson that everybody learned, including Epic.
• 12:37
  Dr. Wachter: Is you don't roll these things out and turn it on in a health system with real patients before truly testing it.
• 12:44
  Dr. Wachter: Not in terms of what it looks like in a laboratory, but it looks like, looks like in real life.
• 12:48
  Dr. Wachter: You have to test it on the real patients in that system.
• 12:52
  Dr. Wachter: You have to be very sensitive to alert fatigue and false alarms and what, uh, a number of systems did, and I, I profiled the University of Colorado in the book is, uh.
• 13:03
  Dr. Wachter: The alert would go, first of all, they retune the algorithm based on their own data, which improved it.
• 13:08
  Dr. Wachter: Secondly, they then gave the alerts, not to the frontline doctors and nurses, but to a separate set of nurses who have watching every patient's room with a video monitor.
• 13:18
  Dr. Wachter: And through that, and third, those nurses went out to the wards and met the doctors and nurses and formed a relationship.
• 13:24
  Dr. Wachter: So that they recognized they were on the same team, and through those maneuvers they managed to tune this alert and alarm so that it's actually quite useful.
• 13:33
  Dr. Wachter: And I think the lesson there is, even for something that seems easy, like sepsis, this is tough stuff.
• 13:39
  Dr. Wachter: You have to be sure that you're getting it right.
• 13:41
  Dr. Wachter: You have to be sure that you actually test it in the real world with real patients, with your patients.
• 13:46
  Dr. Wachter: And I think we learned a lesson.
• 13:48
  Dr. Wachter: I think it's and, but it shows how hard decision support is and how important it was.
• 13:52
  Dr. Wachter: For decision support, not to be the first, uh, time a doctor or nurse ever sees ai, much rather the first time they see AI is me using open evidence on my phone, checking out, you know, what's the differential diagnosis of this patient with a fever and a white count and or using a something like scribes, which are pretty benign.
• 14:12
  John: I, I think what I find found fascinating about that example is you profile a, a pretty profound failure by the absolute dominant player in data and healthcare information.
• 14:24
  John: Epic.
• 14:25
  John: And yet you also, the it's, and it's an interesting combination of not just the doctor, but the nurses of learning, of adapting, of really going back to the data, of trying to make sure that the, that the machine is tuned to data that's relevant to the patients.
• 14:41
  John: In that particular market in Colorado that the nurses that, that their physical observation actually determining, how does the patient actually look And then, then working through, backing off of the alerts, but integrating man machine data and constantly, it, it's almost the art.
• 14:58
  John: Of ai, human healthcare interaction.
• 15:01
  John: It's, you, you have some pretty jarring examples of where we're missing stuff.
• 15:06
  John: That, what I love about that is the progressive success of, of, of the system actually adapting to the technology and the technology.
• 15:13
  Dr. Wachter: I think
• 15:13
  John: that's adapting to human system.
• 15:15
  Dr. Wachter: Super thoughtful way to see that, and I, I, I think that epic, you know, epic has become the dominant player in the world of building electronic health records.
• 15:25
  Dr. Wachter: That doesn't say they know.
• 15:26
  Dr. Wachter: The first thing about, about data analytics, about ai, about human machine interfaces when it comes to something different than using this big electronic filing cabinet.
• 15:38
  Dr. Wachter: Uh, to their credit, I think not only did all of us learn.
• 15:42
  Dr. Wachter: About all the things that you just raised, but Epic did as well.
• 15:45
  Dr. Wachter: And you know, there's a battle going on in the world of AI and healthcare, which is do you buy your AI tools from your EHR vendor?
• 15:56
  Dr. Wachter: Or do you buy your e your AI tools from a third party?
• 15:59
  Dr. Wachter: The third party could be everything from Google to a Silicon Valley startup.
• 16:03
  Dr. Wachter: And there's a lot of debate about that.
• 16:06
  Dr. Wachter: Epic right now is winning that game, in part because it's, it's the old IBM, it's the, if you remember, you know, nobody ever lost their job going with IBM.
• 16:16
  Dr. Wachter: Nobody ever lost their job.
• 16:18
  Dr. Wachter: I, I guess maybe except for this early substance example, going with Epic.
• 16:22
  John: But also it just to, but to give people context who don't live in our quirky world.
• 16:28
  John: Epic.
• 16:28
  John: Epic, you know, is you, you mentioned they're, they're in the 20 top academic medical centers in the country.
• 16:34
  John: I would argue that they're, if you look at it in terms of volume, they're probably at like a 70% share.
• 16:40
  John: I mean, they're something extraordinarily dominant and they've done a phenomenal job at building.
• 16:45
  John: Systems that at the time were the absolute best from a patient data perspective, but they're pretty do dominant.
• 16:52
  John: So, and, and there's, uh, almost three constituents.
• 16:56
  John: Bob, the way I look at it, one the epic, two, the hyperscalers, the Googles, the Amazons, the, the Apples, all of whom have seem to think that healthcare is easy and they're gonna come in here and fix it.
• 17:09
  John: And the third are these incredibly well-funded, if least, at least for a little bit Babylon.
• 17:15
  John: Mm-hmm.
• 17:15
  John: And forward being examples to the contrary.
• 17:18
  John: But there are plenty, there are hundreds more that have 10 up to, uh,$250 million in early stage capital, all fighting for the attention.
• 17:28
  John: And, and the, the op the opportunity to actually help improve healthcare.
• 17:33
  John: Okay.
• 17:33
  John: Ranking them.
• 17:34
  John: Uh, what's the probability?
• 17:36
  John: Where, where are the solutions gonna come from?
• 17:39
  John: Uh, mother Epic.
• 17:41
  John: Your, your neighborhood hyperscalers, Amazon and Google or the, or, the, or.
• 17:47
  John: The, the, the well-funded runts in, in Silicon Valley.
• 17:52
  Dr. Wachter: Yeah, it, it's a nice way of thinking about it.
• 17:54
  Dr. Wachter: I guess I, I push back slightly on, uh, on you saying that Microsoft, Google, et cetera, think healthcare is easy 'cause they have screwed up in healthcare enough times in the past that they have actually learned.
• 18:05
  Dr. Wachter: They did think it was easy and uh, but uh, I think they get it now.
• 18:09
  Dr. Wachter: They get how hard it is and that they need to be in it for the long game.
• 18:13
  Dr. Wachter: Uh, you know, the wimpy answer is there'll be winners in all three categories.
• 18:17
  Dr. Wachter: Um, and that healthcare is 20% of the GDP in a $5 trillion industry, and there's enough room for there to be winners in all the categories.
• 18:24
  Dr. Wachter: If I had to put my nickel down on one, I'd put it down on Epic and I would put it down on Epic because.
• 18:30
  Dr. Wachter: Um, in some ways, surprisingly, you say, how's this company sitting on Wisconsin farmland 10 miles outside of Madison that 10 years ago doesn't know the first thing about ai
• 18:40
  John: 1700 acres outside
• 18:43
  Dr. Wachter: 1700 acres, and, you know, and the quirkiest campus.
• 18:47
  Dr. Wachter: It doesn't know the first thing about AI 10 years ago.
• 18:49
  Dr. Wachter: How are they going to build these tools and compete with the Googles or all these very well-funded startups?
• 18:55
  Dr. Wachter: And the answer is they have.
• 18:58
  Dr. Wachter: The answer is the tools that they have been building over the last five years are not as good as the tools that a company built just to do this.
• 19:06
  Dr. Wachter: One thing might build, let's take scribes for example.
• 19:10
  Dr. Wachter: But generally are good enough so that if a purchaser, like a health system like mine is deciding, do I wait a year for Epic to come out with their tool?
• 19:20
  Dr. Wachter: But when they do, I know it's gonna be at least pretty good.
• 19:24
  Dr. Wachter: I know Epic's gonna be in business for five years.
• 19:26
  Dr. Wachter: I know integration will be a nothing burger.
• 19:28
  Dr. Wachter: All I'll have to do is turn on a switch or do I risk going with some startup tool.
• 19:33
  Dr. Wachter: The company might be out of business.
• 19:34
  Dr. Wachter: There's gonna be some integration headaches.
• 19:36
  Dr. Wachter: Might even be more costly over time.
• 19:39
  Dr. Wachter: Uh, the safe choice is to go with Epic.
• 19:41
  Dr. Wachter: And so far, I think if you look at them over, I was talking to an executive, one of these startups the other day and he said, damnit, they're good.
• 19:50
  Dr. Wachter: And, and I think that's, so right now I think they have this advantage of incumbency.
• 19:54
  Dr. Wachter: And as soon as you begin to say monopolies are bad.
• 19:58
  Dr. Wachter: Nobody wants, you know, to be dominated by this one company.
• 20:01
  Dr. Wachter: It's good.
• 20:01
  Dr. Wachter: It's bad for innovation.
• 20:03
  Dr. Wachter: You know, the counter example is people who live their life in Google Tools and really like the fact that their email integrates with their, you know, with their, their, their spreadsheet or their, uh,
• 20:15
  John: but, but
• 20:16
  Dr. Wachter: Google
• 20:16
  John: has settled with the depart.
• 20:18
  John: Google has settled with the Department of Justice.
• 20:20
  John: They haven't taken on, uh, the, the, the Republic of Epic yet.
• 20:24
  Dr. Wachter: Uh, well, they, they, they, they are, and they will.
• 20:27
  Dr. Wachter: I mean, Google has been building.
• 20:29
  Dr. Wachter: Um, a AI healthcare tools that you probably have seen.
• 20:34
  Dr. Wachter: One called aami, which is a tool built, a conversational agent built to essentially do primary and urgent care.
• 20:42
  Dr. Wachter: And at least in the early studies of it, it's pretty darn good.
• 20:46
  Dr. Wachter: So we'll see, we'll see where Google fits in.
• 20:48
  Dr. Wachter: Amazon just rolled out an AI tool for its, uh, one medical practices, and I haven't played with it yet, but a friend of mine is a one medical patient who was using it the other day to do essentially.
• 20:58
  Dr. Wachter: Chatbot about symptoms and she said it's pretty good.
• 21:01
  Dr. Wachter: So you're gonna see innovation coming from all over the place.
• 21:04
  Dr. Wachter: I think the big battle will be to beat Epic.
• 21:07
  Dr. Wachter: And I think Epic does have a pretty big incumbency advantage.
• 21:11
  John: Um, administrative tasks versus clinical transformation.
• 21:17
  John: Where is technology gonna be the biggest help in either reducing costs and, and making healthcare simpler on the administrative side or improving?
• 21:28
  John: You know, frankly, patient outcomes and clinical decisions.
• 21:33
  Dr. Wachter: I think at the end of the day, the home run is the latter.
• 21:36
  Dr. Wachter: I think that that's where we have to go.
• 21:39
  Dr. Wachter: If all now the administrative stuff, the administrative bloat is real and very important.
• 21:44
  Dr. Wachter: You know, estimates that it's 30% of the healthcare dollar.
• 21:47
  Dr. Wachter: If somehow all of that friction can be decreased if a prior auth happens, because United can look at the patient's electronic record and see that everything's there and approve it on the spot rather than a fax back and forth, you know, that is, that is great.
• 22:04
  Dr. Wachter: If a patient can get an appointment without having to go through hell, that is great.
• 22:09
  Dr. Wachter: You know, if you cut that down from 30% to 15%, you've saved a lot of money and you've made a lot of, you've taken a lot of friction out of the system.
• 22:17
  Dr. Wachter: I don't know that you have markedly improved health or healthcare.
• 22:21
  Dr. Wachter: You've made the experience much better.
• 22:23
  Dr. Wachter: I think that all of the administrative stuff is important.
• 22:25
  Dr. Wachter: A, because it's, it is real money and it's real.
• 22:28
  Dr. Wachter: It's real friction, but B, but I think it's training wheels.
• 22:31
  Dr. Wachter: I think it is how we learn how to, how we get along with this, how.
• 22:34
  Dr. Wachter: Sepsis.
• 22:35
  Dr. Wachter: You know how we learn that this thing sounded good in, you know, as Yogi Berra said, in theory, there's no difference between theory and practice and practice.
• 22:42
  Dr. Wachter: There is how we learn the distinction between theory and practice and healthcare, but where this really makes a difference.
• 22:50
  Dr. Wachter: In the outcomes of patients and the cost of care is when the AI helps guide me as the physician to deliver the best care, the the correct care, the most cost effective care to the patient helps guide me to make the right diagnosis and helps the patient sometimes get care without seeing me if they don't need to.
• 23:08
  John: Let's, well, I think everybody wants to see you, Bob, but that's, oh, okay.
• 23:12
  John: But that's the nature of ev of, of, of patients.
• 23:14
  John: I think they want that human connection, but you really lay out some really terrifying but honest statistics.
• 23:21
  John: 800,000 to million people harmed or killed by medical errors every year.
• 23:27
  John: , That terrifying study.
• 23:28
  John: Another one met, a very large scale study of autopsies that, you know, 20 per, based on a scientific review of autopsies, 20% of diagnoses, some of which could have been life altering, are missed, um, in, uh, in our current system.
• 23:44
  John: So there's clearly a big opportunity, um, to improve.
• 23:49
  John: Um, clinical outcomes and you, you kind of anchor, the practice of medicine, um, back to the, the turn of the, of the 20th century when, when it, and, and sort of said that education hasn't moved forward and you kind of gently suggest that we aren't making as much progress on mand errors and improving clinical practice as we could.
• 24:12
  John: Okay.
• 24:12
  John: So if we wanna save more lives and improve.
• 24:15
  John: Patients' lives.
• 24:17
  John: What do we need to do in terms of leveraging technology to improve outcomes for patients?
• 24:23
  Dr. Wachter: Well, I, I could tell you that when I'm on the wards now, I am quite frequently using AI to help me be smarter, and I don't know that I was missing that many diagnoses before I did this.
• 24:39
  Dr. Wachter: But I think I'm smarter now than I, than I was, and that's not, that's not because my brain's getting better, any better at my age.
• 24:46
  Dr. Wachter: Um.
• 24:48
  Dr. Wachter: We have this practice we call a curbside consult, meaning I don't really need a full on cardiology or infectious disease consultant where they're gonna spend an hour with the patient.
• 24:56
  Dr. Wachter: But I just have a question that I'd love to have answered, and until a year or two ago, there was no good way for me to answer, get an answer to that question without calling my favorite infectious disease doctor.
• 25:09
  Dr. Wachter: You might say, well, how about, um, you know, medical textbooks like up to date?
• 25:13
  Dr. Wachter: Those are great when you have a very straightforward, what's the right dose of apixaban for a patient with pulmonary embolism?
• 25:20
  Dr. Wachter: It has no idea what I'm talking about.
• 25:22
  Dr. Wachter: If I say, this is an 82-year-old man with CLL who comes in with a fever, a white count of creatinine of 3.2 and abnormal LFTs, I can say exactly that sentence into open evidence or GPT or Gemini, and it will give me a cogent answer that is.
• 25:36
  Dr. Wachter: On par with what I would get from my favorite neighborhood specialists.
• 25:40
  Dr. Wachter: So I think where you're going to see real, uh, improvements in medical practice and diagnosis and in in treatment choices is the fact that the AI now can, can essentially scale the knowledge of specialists.
• 25:55
  Dr. Wachter: So it makes me as a generalist.
• 25:57
  Dr. Wachter: Have access to specialty knowledge in my pocket that it makes a nurse practitioner potentially as good as a generalist in some ways, and I don't think we're quite there yet.
• 26:07
  Dr. Wachter: It may make patients and their families able to do things that they previously required a, a physician to do.
• 26:12
  Dr. Wachter: So now this gets much better.
• 26:15
  Dr. Wachter: Right now I have to pull my phone outta the pocket and say to open evidence, this is an 82-year-old man, et cetera, et cetera, in the future.
• 26:22
  Dr. Wachter: Of course, that's silly that that function should be completely embedded in the electronic health record.
• 26:27
  Dr. Wachter: So then I have to tell, I don't have to tell it anything.
• 26:29
  Dr. Wachter: It is, it knows everything about the patient and you then can envision a world where it not only knows that for, but for an outpatient, it's following the data that's coming from their Apple watch or from their oral ring.
• 26:39
  Dr. Wachter: It's integrating that.
• 26:41
  Dr. Wachter: It means that I can track how a patient's blood pressure or asthma doing without them having to come from an office visit.
• 26:47
  Dr. Wachter: All of that points to a world that's much more efficient and, and much more, much less error prone than the one we have today.
• 26:53
  John: Yeah.
• 26:53
  John: The only gen gentle ad I would you get dug into is how AI and technology may help on the upskilling of patients and non non docs, because I do think there's enormous leverage there, but you also add in a fair amount of cautionary references to technology and flying a plane.
• 27:12
  Dr. Wachter: Mm-hmm.
• 27:13
  John: What's, what's happened with, do you want to talk a little bit about what did you learn from studying how technology has both made pilots and, and flying safer, but also potentially raised some questions about deferring to the technology or the, or the mind, the mind of the machine more than the mind of the pilot.
• 27:34
  Dr. Wachter: Yeah, I, I mean tech, first of all, I would say it on whole, aviation has become safer because of technology.
• 27:42
  Dr. Wachter: There's, there's, I think, little empirical question about that, but there's a number of cautionary notes from technology.
• 27:48
  Dr. Wachter: There are cautionary notes from, from, uh, from driverless cars as well.
• 27:52
  Dr. Wachter: And they relate to a number of phenomenon.
• 27:55
  Dr. Wachter: One is, is probably the main one being de-skilling that as the technology gets better, the humans who used to be very good at a task now become dependent on the technology and, and become less good at the task.
• 28:09
  Dr. Wachter: And we have seen a few really tragic aviation accidents over the last several years.
• 28:14
  Dr. Wachter: Where the technology went on the fritz, the humans were then tasked with flying the plane.
• 28:21
  Dr. Wachter: And as Captain Sullenberger, uh, at one point said to me, uh, referring to the crash of, uh, uh, of an Air France flight off Brazil, a number of years ago, Sully said, uh, they were flying a plane.
• 28:33
  Dr. Wachter: They were unfamiliar.
• 28:34
  Dr. Wachter: Meaning a plane where they didn't have their digital wing wingman and they made all the wrong choices.
• 28:40
  Dr. Wachter: So we've seen examples of that.
• 28:42
  Dr. Wachter: We've seen examples of people overtrusting in the technology.
• 28:46
  Dr. Wachter: A number of interesting cases.
• 28:48
  Dr. Wachter: One, one amazing simulation study where, uh, uh, clinical psychologists put a number of pilots into a simulated cockpit and then gave them a signal that one of the, that there was a fire on the plane.
• 29:00
  Dr. Wachter: And the, uh, all of the pilots, uh, turned off the engines, which was absolutely the wrong thing to do because there was a lot of other digital evidence that there was no fire.
• 29:09
  Dr. Wachter: And when they debriefed the pilots on what they saw, and all of them said, well, we saw that signal and we saw a number of other signals.
• 29:18
  Dr. Wachter: Saying the plane was in fire, and in fact there were no other signals.
• 29:21
  Dr. Wachter: So it what she called, uh, phantom memory or you know, that, that they, their ba their brains basically played tricks on them.
• 29:28
  Dr. Wachter: So this is tr harder that looks, and I think it's gonna be very important to think a lot and study the human AI interface because here's the situation we're gonna find ourselves in for the foreseeable future in healthcare, John.
• 29:39
  Dr. Wachter: We are going to have now AI that is good enough to be useful, but not good enough to be entirely trusted to act autonomously.
• 29:48
  Dr. Wachter: And so the system that we will set up, whether it's me reviewing my scribed draft note, uh, or me being the final arbiter of the diagnosis, even if the AI is weighed in.
• 29:59
  Dr. Wachter: The system will set up will, it will be one where the human will be in the loop and the problem is a, I will de-skill over time.
• 30:07
  Dr. Wachter: I, I no longer know how to read a map and, and I will probably no longer be as good at diagnosis.
• 30:12
  Dr. Wachter: B, we are not good at looking over the shoulder of a technology that we've learned to generally trust.
• 30:18
  Dr. Wachter: And if you wonder about that, just look at the a hundred Tesla accidents where the drivers were supposed to be paying attention, but of course they're human and therefore they weren't.
• 30:28
  Dr. Wachter: And so these are trying to get that interface that dyad between the human and a generally reliable but not perfect.
• 30:35
  Dr. Wachter: AI may turn out to be the dominant issue that we face for the next several years until the AI is perfect, when it will begin to, uh, be put in positions of, of much more autonomy.
• 30:45
  John: But I, I think that one of the things that's interesting about your book is that you start with the doctors.
• 30:52
  John: Humans in the loop are kind of doomed.
• 30:54
  John: Then at the end, you actually, in a number of different points, talk about the importance of pattern recognition and illness scripts for actually interpreting the, the, the complex humanity of illness.
• 31:08
  John: And maybe you could speak to that, that, that challenge of what an illness script is and why it's so important in, in, in, in, in, in interpretation.
• 31:18
  John: And the most critical thing that a doc does, I think is, is, is, is diagnosis.
• 31:23
  Dr. Wachter: Yeah, we, we, it's a term of art In medicine, we call it an illness script.
• 31:27
  Dr. Wachter: It basically is the pattern, the overall pattern of symptoms, signs, sort of look and feel.
• 31:35
  Dr. Wachter: That we have learned from both the textbooks we read in medical school, but I could say more importantly, the years of experience that we have.
• 31:42
  Dr. Wachter: Uh, in some ways the, the, the sort of quintessential manifestation of the illness script is what we call the eyeball test, which is I can go into a patient's room and look at a two patients that are exactly the same on paper.
• 31:56
  Dr. Wachter: Their blood pressure's the same, their heart rate's the same, their temperature's the same.
• 31:59
  Dr. Wachter: And I have learned over the years to be able to say that one on the left really sick.
• 32:04
  Dr. Wachter: I'm very worried.
• 32:04
  Dr. Wachter: My spidey sense is going off.
• 32:06
  Dr. Wachter: That one on the right, I'm not that worried about there's something.
• 32:09
  Dr. Wachter: About the eyeball test.
• 32:10
  Dr. Wachter: Now there are, there are technology optimists, and I remember interviewing Vinod Cosa on this who say, well, if you can do it, the tech can do it.
• 32:19
  Dr. Wachter: If you're looking at the sweat on their brow or the dilation of their pupils, that's all measurable.
• 32:24
  Dr. Wachter: Maybe, uh, I guess we will see, but for now, there does seem to be value.
• 32:31
  Dr. Wachter: In an experienced, thoughtful human physician, uh, looking at what the algorithm says and periodically saying, Hmm, you know, in this patient, I don't think so.
• 32:41
  Dr. Wachter: I think something different is going on.
• 32:43
  Dr. Wachter: How good we really are at that, whether we stay as good at that, if we de-skill over time, if we become more dependent on the computer, whether society is going to want to pay for a doctor from what might be a pretty small incremental increase in.
• 32:57
  Dr. Wachter: Diagnostic accuracy, you might say, well, I mean, you just said John, thank you.
• 33:01
  Dr. Wachter: Everybody wants to see me.
• 33:02
  Dr. Wachter: But what if not seeing me, uh, costs you a thousand dollars less on your insurance every year?
• 33:09
  Dr. Wachter: Uh, are you really willing to pay to see me?
• 33:11
  Dr. Wachter: You might be, but many people may not be.
• 33:13
  Dr. Wachter: So there are a lot of complex questions we have to work our way through.
• 33:17
  John: Um, you, you also, I mean, I think in January of this year, UCSF published a study about.
• 33:26
  John: Leveraging AI to diagnose complex lung diagnosis in the ICU and it's pretty remarkable.
• 33:32
  John: I mean, do you wanna, do you wanna speak, I dunno that you recall that particular study, but it's pretty, pretty remarkable.
• 33:38
  Dr. Wachter: Well, I, I, I don't recall that study, but, uh, and it's, it's potentially upsetting to me because my, uh, my.
• 33:44
  Dr. Wachter: Daughter and son-in-law are both pulmonary fellows at UCSF, so,
• 33:47
  John: well, it's, I'm hoping they have jobs.
• 33:48
  John: It's basic, basic, basically, you guys used the, the AI to identify genetic variations that immediate, that perfectly diagnosed lung challenges in the IC, which were, were harder to do prior.
• 34:01
  John: That there are incremental by diagnosis.
• 34:05
  John: Um, um, insights that AI is creating even under your, on, on, above and below.
• 34:12
  John: You probably at UCSF right now.
• 34:15
  John: Yeah, of course.
• 34:15
  John: And it's happening everywhere.
• 34:16
  John: We're pretty, pretty optimistic.
• 34:19
  John: At the same time, I'm kind of scared about this de-skilling thing, Bob.
• 34:22
  John: How do we, how do we, how do we, you know, I, I'm, I'm sort of always looked up to you guys as docs.
• 34:27
  John: What do you mean you're getting de-skilled?
• 34:29
  John: That, that's scary.
• 34:30
  Dr. Wachter: There was a, there was a really.
• 34:32
  Dr. Wachter: It's really pretty concerning study that came out about a year ago, uh, out of, uh, out of Europe where.
• 34:37
  Dr. Wachter: Um, colonoscopist, uh, gastroenterologists do colonoscopy.
• 34:42
  Dr. Wachter: Uh, we're given for three months access to an AI colonoscopy tool.
• 34:46
  Dr. Wachter: And what the tool does is the colonoscopist, the colonoscopy, uh, is, is, is up one's rear, uh, looking around.
• 34:56
  Dr. Wachter: And the AI puts a little green box around all of the lesions that look suspicious, and it's quite accurate and quite good.
• 35:02
  Dr. Wachter: And the colonoscopy doctors liked having it, then they turned it off.
• 35:06
  Dr. Wachter: And looked at their performance over the subsequent months, and it turned out that their performance in identifying, uh, uh, uh, uh, precancerous lesions had degraded significantly from what it was before.
• 35:19
  Dr. Wachter: These weren't newbies.
• 35:19
  Dr. Wachter: These were docs who had an average of 10 years of experience doing colonoscopy.
• 35:24
  Dr. Wachter: And in just a three month exposure to this AI tool, they got worse.
• 35:29
  Dr. Wachter: And this comes up in my mind, every time I put a paragraph into Claude and say, can you make this better?
• 35:34
  Dr. Wachter: It was, there's a part of me that says, oh, you know, I'm learning from ai, but no, probably I'm not, probably, I'm getting a little bit dumber as I become more dependent on the tool, uh, to do, to do whatever function we're using it for.
• 35:45
  Dr. Wachter: So it is a major concern.
• 35:47
  Dr. Wachter: I think it's a huge, uh, issue for us in medical education.
• 35:51
  Dr. Wachter: I think what it says to me at least is we've gotta be darn careful not to degrade the educational experience in areas of diagnosis and clinical reasoning.
• 36:03
  Dr. Wachter: Maybe at some point those become less important because the AI is so good.
• 36:06
  Dr. Wachter: You don't need the doctor in the loop anymore.
• 36:08
  Dr. Wachter: But for the foreseeable future, as I say, I think we're gonna be a world where the AI is helpful but not perfect, and we have to do everything we can in the world of education to be sure that we don't allow those skills to degrade.
• 36:19
  Dr. Wachter: How to do that I think is, is a trillion dollar question.
• 36:22
  Dr. Wachter: I think it's a very tough one, but I think the mistake now would be to say, well, the AI is so good that, why do we need to teach doctors diagnostic reasoning anymore?
• 36:31
  Dr. Wachter: The AI could do it for them.
• 36:32
  Dr. Wachter: That would be a death spiral.
• 36:33
  Dr. Wachter: And I think we've gotta push back on it really hard.
• 36:36
  John: So, uh, quick lightning round and then we will wrap.
• 36:39
  John: Sure.
• 36:39
  John: Most overhyped AI tool in healthcare.
• 36:45
  Dr. Wachter: It probably comes from your world more than my world.
• 36:47
  Dr. Wachter: I'm told by the folks in drug development that, uh, you know, the idea that it's gonna shorten the period from, you know, I have an idea to here's a new approved drug, you know, from 10 years to a year.
• 36:58
  Dr. Wachter: Uh, I think that has been overhyped by, uh, from what I see in medicine, uh, right now.
• 37:05
  Dr. Wachter: Right now, clinical decision support is still being over-hyped.
• 37:09
  Dr. Wachter: And I'd say maybe one last and this I've, I've gone outta lightning, sorry.
• 37:12
  Dr. Wachter: But precision medicine, the idea of integrating genomics into this and making better predictions, still not quite ready for prime time and anything other than oncology.
• 37:21
  Dr. Wachter: I think we'll get there, but still pretty early.
• 37:24
  John: Most under hyped AI technology or approach in healthcare today.
• 37:30
  Dr. Wachter: Full integration of decision support into the electronic health record.
• 37:35
  Dr. Wachter: So that it just happens automatically.
• 37:38
  Dr. Wachter: I don't have to tell it anything.
• 37:39
  Dr. Wachter: It knows everything about you as a patient, and it's seamlessly in an ambient way, is suggesting diagnoses, watching you during the hospitalization and saying, well, you the patient.
• 37:49
  Dr. Wachter: You've said the patient had pneumonia, but the patient still has a fever in a white count.
• 37:52
  Dr. Wachter: On day four, we've looked at the last a hundred thousand patients who had pneumonia and the uh, 95% of them were better.
• 37:58
  Dr. Wachter: You should rethink your diagnosis.
• 38:00
  Dr. Wachter: Just that sort of real life.
• 38:02
  Dr. Wachter: Uh, in line decision support coming directly from your electronic tools.
• 38:07
  John: Amazing.
• 38:08
  John: Well, Dr. Bob, uh, an Amma, a wonderful book, a, a very balanced book, and an artfully crafted book that gives us a lot, lot to work on and to think about as we, as we, as we deal in real time with a, an AI transformation that you point is, is, is moving in a rapid clip and we ought to be paying attention to it.
• 38:27
  John: Thanks for joining, uh, this morning at Care Talk, and if you liked what you heard or you didn't, we'd love you to subscribe on your favorites.
• 38:34
  John: Service.
• 38:35
  John: I'm John Driscoll, the chairman of the Yukon Health System.
• 38:37
  John: Thank you, Bob for joining us this morning.
• 38:39
  Dr. Wachter: Thank you, John.
• 38:40
  Dr. Wachter: It was a real pleasure.