WEBVTT
00:00:04.799 --> 00:00:08.480
Hello friends and a very warm welcome to Christmas Christmas.
00:00:09.519 --> 00:00:12.320
I am ready to test into your content.
00:00:18.320 --> 00:00:21.120
India, the United Kingdom, and the United States.
00:00:41.759 --> 00:00:46.079
To help you manage your stress, find balance, and live a life of purpose.
00:00:46.240 --> 00:00:48.719
Please join us every Friday at 5 p.m.
00:00:48.880 --> 00:00:51.840
and let's start turning stress into strength together.
00:00:52.000 --> 00:00:54.479
Now let's dive into today's episode.
00:00:56.560 --> 00:00:57.759
Hello friends.
00:00:58.000 --> 00:01:00.640
Welcome to the Transforming Stress with Dr.
00:01:00.799 --> 00:01:02.320
Ash podcast.
00:01:02.560 --> 00:01:06.879
And today today's episode is really exciting.
00:01:07.280 --> 00:01:17.519
In the last couple of years, all of us have been hearing about AI, artificial intelligence, and it's very exciting.
00:01:18.079 --> 00:01:22.640
A lot of people say that it's the new fire.
00:01:22.879 --> 00:01:26.719
Equally, there are a lot of people who are worried about it.
00:01:34.640 --> 00:01:48.959
So to break this myth, a lot of myths, a lot of confusion, I wanted to get an expert, somebody who has been navigating this landscape for 40 years, 40 years, four decades.
00:01:49.359 --> 00:01:51.120
AI in healthcare.
00:01:51.439 --> 00:02:07.519
Daniel Byrne is a faculty of AI in healthcare, and he is the person who is going to help us take the help help us to understand what is the evidence and what is the myth.
00:02:07.760 --> 00:02:10.560
And please help me welcome Dan Byrne.
00:02:10.879 --> 00:02:15.199
Thank you, Dan, for joining us from John Hopkins.
00:02:15.680 --> 00:02:17.199
Well, thank you for inviting me.
00:02:17.439 --> 00:02:20.960
Yes, AI has become incredibly powerful.
00:02:21.120 --> 00:02:28.479
And I think almost everybody agrees that the area of society where we're going to see the real benefit of AI is in healthcare.
00:02:28.719 --> 00:02:37.439
But uh the big question is why is there almost no rigorous evidence that we're seeing AI is improving patient outcomes?
00:02:37.599 --> 00:02:44.479
There's a lot of talk, there's a lot of weak evaluations, but the big question is when is it going to help patients?
00:02:44.560 --> 00:02:46.639
And that's been the focus of my work.
00:02:47.039 --> 00:02:59.280
Thank you, Dan, for doing that very important work because medicine is a field where we cannot have any error, because an error means life.
00:02:59.520 --> 00:03:06.479
And whilst there is a lot of excitement, there are new tools, there are a lot lot much promise.
00:03:06.719 --> 00:03:11.840
And I know you are asking the question, what is the evidence?
00:03:12.000 --> 00:03:19.280
What is the evidence that this is going to improve improve the patient outcome?
00:03:19.599 --> 00:03:24.240
So then we have got listeners from 56 countries from all over the world.
00:03:24.400 --> 00:03:37.520
So they are not just doctors or healthcare professionals, we have got a huge wide audience who are very much interesting interested in learning this landscape because this is where the future is.
00:03:37.840 --> 00:03:43.919
And I know you've been in the data sciences uh for for four decades.
00:03:44.240 --> 00:03:57.599
Now, after spending like decades looking at data, looking at numbers, what can the numbers and data tell us even before humans understand that there is a problem?
00:03:58.000 --> 00:04:00.000
Yeah, so that's the important question.
00:04:00.240 --> 00:04:13.360
And what the real benefit of artificial intelligence and healthcare is that it can predict things much more accurately than humans, and it can predict things much sooner and upstream.
00:04:13.599 --> 00:04:16.639
So our current healthcare system is very reactive.
00:04:16.879 --> 00:04:19.360
We catch things when they're too late.
00:04:19.920 --> 00:04:25.920
And uh you humans are not that great at predicting a lot of complex things.
00:04:26.160 --> 00:04:30.959
Um and Daniel Kahneman won the Nobel Prize for his work in this area.
00:04:31.199 --> 00:04:37.600
And if anyone hasn't read his his book, Thinking Fast and Slow, it's a fascinating book to read.
00:04:37.920 --> 00:04:42.480
But uh he showed that humans are very, very confident in their predictions.
00:04:42.720 --> 00:04:44.480
They're just not very good at prediction.
00:04:45.199 --> 00:04:54.480
And in healthcare, there's so many areas where we ask doctors and nurses to predict things that are really not humanly possible to predict.
00:04:54.879 --> 00:04:56.560
So they'll do their best.
00:04:56.879 --> 00:05:02.079
But now we have these AI tools that can predict things much more accurately.
00:05:02.959 --> 00:05:04.480
I'll give you an example.
00:05:04.800 --> 00:05:10.639
So uh a physician asked us to help build a tool to predict postpartum hemorrhage.
00:05:12.000 --> 00:05:19.279
And if people don't know what that is, after a woman gives birth, she can bleed to death or bleed excessively.
00:05:19.439 --> 00:05:23.600
And you might think this doesn't happen much anymore, um, but it does happen.
00:05:23.680 --> 00:05:27.279
It's a leading cause of maternal morbidity and mortality.
00:05:28.079 --> 00:05:45.759
Um so we built a very accurate way to predict postpartum hemorrhage, and um then we also tested the doctors and nurses that do this every day and asked them, uh, do you think this woman is likely to have a postpartum hemorrhage?
00:05:46.399 --> 00:05:49.360
And we showed they weren't much better than flipping a coin.
00:05:50.000 --> 00:06:02.800
So we implemented this AI predictive model into the hospital computer and then we did something that almost nobody does that I think is the real key to the next level of AI in healthcare.
00:06:03.360 --> 00:06:08.639
And that's for one year we performed a pragmatic randomized control trial.
00:06:08.800 --> 00:06:20.079
And that means as every woman came who came through the hospital to deliver a baby, she was randomized either usual care or usual care plus this model.
00:06:20.639 --> 00:06:27.600
And if the probability of postpartum hemorrhage was high, then there's um a bundle uh to prevent it.
00:06:28.160 --> 00:06:30.560
So we're we're waiting to see at the end of this year.
00:06:30.639 --> 00:06:38.240
We'll uh look at the results and hopefully this model predicted and prevented postpartum hemorrhage.
00:06:38.639 --> 00:06:39.920
And I think that's the key.
00:06:40.079 --> 00:06:44.720
We need to take these AI tools and test them, make sure they're safe.
00:06:44.959 --> 00:06:46.160
It could be unsafe.
00:06:46.560 --> 00:06:47.839
And is it effective?
00:06:48.160 --> 00:06:50.160
And is there a return on investment?
00:06:50.480 --> 00:07:03.199
So the same thing can be done for cancer, catching it upstream, blood clots, readmissions, and in your area, predicting stress, burnout, and suicide, we catch these things too late.
00:07:03.439 --> 00:07:10.879
AI tools can catch it much earlier, but we can't just uh give it a free pass.
00:07:11.040 --> 00:07:16.079
We need to test in rigorous science is this safe and is it effective?
00:07:16.560 --> 00:07:20.879
Uh very uh very, very astute uh Dan.
00:07:21.279 --> 00:07:30.240
And I was reading about your burn test because you are so passionate about uh getting it right.
00:07:30.399 --> 00:07:39.839
And uh please correct me if I'm wrong here, that uh your colleagues in the corridor conversations used to say, well, does it pass the burn test or not?
00:07:40.240 --> 00:07:56.079
And it is it is whether this tool is able to go through that test of fire and see whether this is going to going to work in the real world or not.
00:07:56.240 --> 00:08:04.079
And as Dan just mentioned, the same thing, because um then let me try to simplify this and please correct me if I'm wrong.
00:08:04.240 --> 00:08:20.480
So, what I've understood, whether it is research, whether it is clinical medicine, whether it is stress management, it is it is all about it is all about pattern recognition.
00:08:21.040 --> 00:08:22.240
Pattern recognition.
00:08:22.480 --> 00:08:40.080
Data are continuously giving us patterns, and we see what these patterns are telling us, and how we can then gain insights and utilize these patterns.
00:08:40.399 --> 00:08:46.720
Right, and that's why AI ha can go beyond human knowledge and predict things better.
00:08:47.039 --> 00:08:58.320
So, for example, when humans struggle to predict postpartum hemorrhage, they could only put a few inputs in their brain to uh compute that probability.
00:08:58.559 --> 00:09:05.519
But our AI model uh includes 21 predictors uh to look at that pattern.
00:09:05.919 --> 00:09:14.639
So it's a it's a more complex um pattern than humans uh can can do in their head, and it does it fast and free.
00:09:14.799 --> 00:09:20.000
And and there are patterns also in a mammogram that radiologists can't see.
00:09:20.159 --> 00:09:30.320
There are patterns in um retinal scans that have amazed people at what uh the AI can can determine with a retinal scan, an EKG.
00:09:30.639 --> 00:09:38.399
There are there are amazing things in the patterns that AI detects that that humans never even imagined you could you could tell.
00:09:38.639 --> 00:09:53.200
And Daniel, that makes me the question with a lot of people and doctors and professionals are fearing, is AI going to take away their take away their jobs, going to replace, replace them.
00:09:53.679 --> 00:10:00.639
And I will tell I will tell you, I will share with you my experiment, which I did in the last six months.
00:10:00.960 --> 00:10:09.600
So what I did that whenever I'm I'm an internal medicine physician, so I'm on call and I'm seeing complicated patients.
00:10:09.919 --> 00:10:23.919
Now, if I'm in a situation that I need to take uh advice from a specialist like a cardiologist or a respiratory physician, and thankfully I've got the place I'm working, I've got access to the specialists.
00:10:24.240 --> 00:10:33.840
So I I see the patient and I know I need to speak to the to the cardiologist here, and these are the questions I'm going to ask.
00:10:34.080 --> 00:10:48.960
Now, before I do that, before I pick up the phone, in my mind I think suppose I am working in a remote area, or I am working in an island where there is no access to the specialist, what can I do?
00:10:49.200 --> 00:10:54.399
So I open the phone and go to open evidence six months.
00:10:54.720 --> 00:11:04.960
Um six months ago, I I anonymize all the data, and I would give that this is the situation I am, and I need some advice from the specialist.
00:11:05.120 --> 00:11:08.080
And the open evidence gives me the gives me the advice.
00:11:08.159 --> 00:11:23.600
And it makes because I am an internal medicine physician myself, I am able to make sense of what uh what the evidence or what the app is telling me, and then I speak to the specialist.
00:11:24.000 --> 00:11:34.320
And I have been amazed that that the advice is so similar, and the other thing is that the AI has continued to improve.
00:11:34.399 --> 00:11:37.919
Now, unfortunately, open evidence is no longer working in Europe.
00:11:38.000 --> 00:11:44.399
So I after speaking to you, I've started using Gemini, which is also equally equally good.
00:11:44.639 --> 00:11:49.840
But the main thing here is to give AI the context.
00:11:50.080 --> 00:11:55.919
So I'm answering my own question that the human judgment has to be there.
00:11:57.039 --> 00:12:02.159
Deal who's dealing with the patient or situation in front of them.
00:12:02.639 --> 00:12:08.320
So we this is getting better and better as times goes.
00:12:08.480 --> 00:12:17.039
So I wanted to hear from you about um about what the landscape is looking and how do you foresee this shifting in the future?
00:12:17.360 --> 00:12:30.559
Yeah, I don't believe AI is going to replace doctors, but the doctors who learn about AI uh will replace doctors that refuse to learn it, just like it always happens with new technology.
00:12:30.720 --> 00:12:39.039
There are people that resist learning new technology and they're replaced by the people who are more open-minded and learn the new technology.
00:12:39.279 --> 00:12:53.840
Um and we really need a clinician computer symbiosis where we need the physician to do certain things, and we need to let the computer do certain things, and we need the best of both worlds there.
00:12:54.080 --> 00:12:57.039
And that's part of Kahneman's uh research too.
00:12:57.360 --> 00:13:07.840
So, for example, we built an AI tool to predict hospital readmissions, 30-day hospital readmissions, um, and it predicted very accurately.
00:13:08.159 --> 00:13:15.440
Now, some places also tried to have their AI tool decide what to do for the high-risk patients, and they failed.
00:13:16.000 --> 00:13:30.080
So you what you need to do is have the computer do what it's good at, predicting which patients are likely to be readmitted, and then have uh clinicians decide what does that patient need who's a high risk.
00:13:30.399 --> 00:13:32.720
And and then you get the best of both worlds.
00:13:33.120 --> 00:13:44.480
Now, you did point out that in some countries there's a lack of specialists, and we're gonna need to use AI because there just aren't enough uh doctors who are specialists.
00:13:44.559 --> 00:14:01.840
Uh one striking statistic, and you may know more about this than I do, but in India there's one pathologist every nine million people, and in a lot of countries, uh we even in the US, there's a lack of therapists and rheumatologists.
00:14:02.080 --> 00:14:06.720
So we are gonna have to use AI when there's a shortage of people.
00:14:06.799 --> 00:14:12.320
But uh but I think the symbiosis is really the key to using AI effectively.
00:14:12.720 --> 00:14:19.919
And how how do you think this landscape is going to change uh in the next five to ten years?
00:14:20.080 --> 00:14:24.480
Because you have seen how things have shifted in the last 40 years now.
00:14:24.720 --> 00:14:27.039
And how fast is the speed of change here?
00:14:27.440 --> 00:14:28.320
Well, it's interesting.
00:14:28.399 --> 00:14:34.240
So AI is moving fast, but healthcare is notoriously resistant to change.
00:14:34.480 --> 00:14:38.720
So I think what's gonna happen in the next five years is there's gonna be a shake out.
00:14:38.879 --> 00:14:41.039
There's a lot of people investing money in this.
00:14:41.120 --> 00:14:44.960
Uh uh,$100 billion is being invested in AI and healthcare.
00:14:45.279 --> 00:14:46.240
There's gonna be a shakeout.
00:14:47.120 --> 00:14:52.559
80% of these startups and and researchers and and experts are gonna fail.
00:14:52.960 --> 00:14:54.720
Twenty percent are gonna be successful.
00:14:54.960 --> 00:14:58.799
Twenty percent of hospitals are gonna be successful at implementing AI.
00:14:59.200 --> 00:15:07.120
They're gonna reduce complications, improve survival, uh, they're gonna have much better uh results, detect things earlier.
00:15:07.440 --> 00:15:16.799
And then patients are gonna flock to those 20% of hospitals that use AI right, and then the other 80% are gonna be forced to change.
00:15:17.039 --> 00:15:20.320
And what are the eighty what are the 20% gonna do differently?
00:15:20.639 --> 00:15:25.039
They're not just gonna embrace AI and invest in it.
00:15:25.440 --> 00:15:28.720
They're gonna really uh treat it like a new drug.
00:15:29.120 --> 00:15:32.399
And they're gonna say, is is this thing safe?
00:15:32.879 --> 00:15:36.639
They're gonna test that first, and then they're gonna test, is this effective?
00:15:37.360 --> 00:15:38.879
And they're gonna do this.
00:15:39.039 --> 00:15:50.480
Uh the burn test is really a pragmatic, randomized, controlled trial of AI improving patient outcomes, but it's done with a very high bar.
00:15:51.039 --> 00:15:56.639
And it's done in a way that if you get it right, you could go in the New England Journal of Medicine.
00:15:56.960 --> 00:16:00.639
So that means you have to pre-register it with clinicaltrials.gov.
00:16:01.120 --> 00:16:08.799
You have to do all of the rigorous steps to get the right answer, and that's really what the New England Journal of Medicine wants.
00:16:08.960 --> 00:16:11.279
They just want to know that they're not going to get burned.
00:16:11.360 --> 00:16:17.600
And if you say your AI tool improved patient outcomes, they're comfortable because you did all of the right steps there.
00:16:17.919 --> 00:16:18.559
Thank you, Dan.
00:16:18.720 --> 00:16:25.440
Then now we are going to shift to another area of very much interest to me.
00:16:26.240 --> 00:16:34.559
And uh, I know you mentioned about uh the book Thinking Fast and Slow, the Type 1 and Type 2 thinking.
00:16:34.639 --> 00:16:40.159
And I want to share that in the context of the physician and the healthcare burnout.
00:16:40.240 --> 00:16:58.080
You know, you know in the United States that uh in the last five to ten years there have been studies every every year which have said that there are 50 to 60 percent at least burnout rates in the physicians, and I think that all over the globe the stats are not very different.
00:16:58.240 --> 00:17:01.759
But why don't why do we see healthcare in other professions?
00:17:02.080 --> 00:17:07.200
According to the Gallup studies, every year the stats are very, very similar.
00:17:07.440 --> 00:17:29.440
Now, what happens in burnout, in that state of severe exhaustion, in the in the book Thinking Fast and Slow, what it talks about in the severe cognitive load, in the exhaustion, in the burnout, the thinking automatically shifts from the type 2 systematic thinking to the type 1 thinking.
00:17:29.599 --> 00:17:35.200
And the type 1 thinking has 10 to 15 percent of error rate.
00:17:35.359 --> 00:17:51.519
Now, if I tell you that you are going to fly, I'm flying you out to now give a talk, flying you from Boston to London, and the plane has a 10% risk of crashing and error rate, then you will not be able to sleep.
00:17:51.680 --> 00:17:53.599
You will not you're not able to sleep.
00:17:53.920 --> 00:18:11.519
Now, the aviation industry is the safest industry because what they have done at with the system's approach and all the work which has been put in the safety, the landscape has completely changed, and medicine has to learn from it.
00:18:11.599 --> 00:18:16.319
But of course, the human factors are so much in the in the industry.
00:18:16.640 --> 00:18:30.640
So we are talking about how AI can help us as a great tool to cognitively unload the physicians, unload the healthcare professionals.
00:18:30.799 --> 00:18:47.920
And I personally found it to be a great tool over the last year in um in um in doing in working working with uh as a as a Microsoft co-pilot or a chat for health or open evidence or or Gemini.
00:18:48.480 --> 00:18:53.359
So what are your what are your thoughts about uh the AI and the burnout?
00:18:53.759 --> 00:19:03.359
Yeah, and so you make a good point that the aviation industry has become remarkably successful at um at at being safe.
00:19:03.519 --> 00:19:07.759
And part of that is they've been using AI for a long time.
00:19:08.079 --> 00:19:16.079
And I I have a section in my book about how the aviation industry is safer because they apply AI.
00:19:16.400 --> 00:19:24.799
And I think we can learn from uh that and and apply AI to uh to do things safer in healthcare.
00:19:25.119 --> 00:19:29.119
And burnout is a huge problem in in healthcare and other industries.
00:19:29.359 --> 00:19:31.359
And AI can help with a lot of that.
00:19:31.839 --> 00:19:47.680
You know, uh w one area is with AI scribes so that physicians don't have to stay up late at night typing uh patient notes and we can just take some of the administrative work off of physicians and allow AI to do it.
00:19:48.160 --> 00:19:50.559
But again, it needs to be studied.
00:19:50.640 --> 00:19:56.079
We need to make sure that's safe and we need to make sure that that is actually uh helping them.
00:19:56.480 --> 00:20:05.200
And people often skip the opportunity to do a simple thing like a pragmatic RCT of the AI scribes.
00:20:05.359 --> 00:20:07.759
That's an easy thing to test out.
00:20:08.240 --> 00:20:19.279
The other way that I believe we need to test AI in healthcare is stop pretending like it's a one and done thing where we we just test it and we have one answer.
00:20:19.519 --> 00:20:22.720
We should be thinking about these things over the next ten years.
00:20:22.960 --> 00:20:47.759
We're gonna have burnout and stress, so we should think over the next ten years, there's gonna be various AI scribes and other tools, and we should do prag pragmatic uh RCTs over the next ten years and just keep asking different questions in an adaptive platform trial where we drop arms that are not working, and we add a new study arm that is working.
00:20:48.480 --> 00:20:58.640
And we keep asking questions in a factorial design, which just means that we can randomize in different ways to ask different questions.
00:20:58.880 --> 00:21:14.880
So we're really a learning healthcare system and we're continuously getting better, and and we apply continuous quality improvement to make uh the work uh safer and and just more enjoyable for physicians and nurses.
00:21:15.359 --> 00:21:20.319
AI can can help with so many administrative jobs in healthcare.
00:21:20.480 --> 00:21:24.400
Um for every doctor in the hospital, there are nine administrators.
00:21:25.039 --> 00:21:27.039
30% of healthcare is waste.
00:21:27.279 --> 00:21:37.119
AI can really help in a lot of these areas, and um but we we can't just be paying a lot of money for AI tools that maybe they don't work.
00:21:37.200 --> 00:21:39.680
We need to really evaluate if they do work.
00:21:40.000 --> 00:21:40.240
True.
00:21:40.319 --> 00:21:49.599
So the future, if it is the future if it is approached with caution and due diligence looks very promising.
00:21:50.079 --> 00:21:52.640
And I can I can go ahead.
00:21:52.960 --> 00:22:15.200
Yeah, no, I can say that from my own experience that the work over the last one year uh I have found a shift uh in both the cognitive load and all the learnings and all which which is which is going into it and uh I'm I feel I can sense that that it is going to go in a very very positive and a safe direction.
00:22:15.440 --> 00:24:03.440
You know last time we were speaking then about the uh and you gave the example of uh you gave the example of postpartum hemorrhage you spoke about SLE was it SLE uh Lupus you talk about the tool uh you were developing I would love for you to share about that because that is remarkable that amount and there are we know that so many times that there are missed diagnoses there are delayed diagnoses so can you speak to more from a chronic chronic health issues point of point of view sure so uh we we have about 10 different projects going on now testing ai but one of the ones uh that uh just passed the burn test is uh we had a physician who is a rheumatologist and she came to us and she said I need a way to predict which patients once they have a positive ANA test are going to develop an autoimmune disease like lupus and uh she said right now rheumatologists are just overwhelmed and they can't really predict this and the patients often uh go on a diagnostic odyssey and it takes on average seven years before patients get their diagnosis of lupus and other autoimmune diseases and seven years is a a long time and during that time the disease eats away at their body and so she asked can we use AI to predict it faster than seven years so we worked with her met with her every week and um and we developed um a way that once a patient has a positive ANA we take their information and we put it into an AI tool to compute probability of an autoimmune disease.
00:24:04.160 --> 00:24:17.359
And then we did what hardly anyone does we said for the next year we're gonna do a pragmatic randomized control trial and we don't take any anything away from anyone everybody continues to get usual care.
00:24:17.599 --> 00:24:22.640
So we randomized them to usual care or usual care so people calm down.
00:24:23.200 --> 00:24:33.359
And then in one arm on top of usual care we're gonna compute the probability of an autoimmune disease and if it's high we're gonna send them to a rheumatologist.
00:24:34.160 --> 00:24:43.279
And we just uh closed out the study and it it significantly reduced time to diagnosis down to 43 days.
00:24:44.000 --> 00:25:00.880
And we have the paper um the New England Journal of medicine AI Journal gave us very positive reviews and we're just wrapping up the reviews and hopefully in another month we'll get this published and uh that that's an example of how you can do this right.
00:25:01.039 --> 00:25:07.759
Because often people say well randomization is going to be too expensive take too long it's it's not possible.
00:25:08.079 --> 00:25:26.480
So we try to demonstrate with these projects um examples that people can copy and and you could think of um a hundred other areas of healthcare where it's reactive and you could use AI in a similar way but then evaluate it and uh and really prove it.
00:25:26.559 --> 00:25:29.359
Don't declare victory just because you have a model.
00:25:29.839 --> 00:25:41.759
True Dan and the same thing can be applied to any chronic physical health issue the same thing can be applied to any chronic mental health issue.
00:25:42.160 --> 00:25:50.079
So suicide stress burnout um there's lots of opportunities to catch these things earlier.
00:25:50.640 --> 00:26:02.720
Much much earlier the pattern recognition much earlier now I see uh on your desk behind a copy of my book The Boiling Frog sitting there.
00:26:03.039 --> 00:29:39.759
Yeah I really enjoyed that it's uh such such a great book I've I enjoy all the quotes and the and the illustrations and I was trying to think how could AI combine with your workshops and I was I was thinking um so let's say a medical center is interested in uh hiring you for some workshops and and helping with the doctors um first they could take um an AI tool or something like our stress checkup uh the nine items there and just have all the the physicians uh complete this and the ones that have a high probability of um stress or burnout uh could take your workshop but then you could also take this um to the next level and say let's do this for a random half and you could ask all the doctors would you be willing to participate in a pragmatic randomized controlled trial and the and the control group could be a delayed intervention so they could get it next year but that way you could test what's the impact of assessing uh uh burnout and then having an intervention and and that's uh that could be a landmark paper that really evaluates the uh the evidence thank you Dan so the boiling frog is uh a full a decade worth of work which I have put my efforts into and it's a systems based approach for stress management and what it means that what if I take hundred people through this intervention the everybody's answers are going to be different because you see everybody's physiology is different everybody's strengths are different their values are different their physiology in terms of their sleep and nutritional new aspects and and there are so their situations at work the culture they are working in there are so many data points into that so a one size fits all approach is not going to work that's why I created a systems based approach with the boiling frog now we are going to take it to a different level now with iFrog which is put another level of the AI integrated into this because as you were sharing with me earlier that it's all about pattern recognition stress is nothing but maladaptive thought patterns maladaptive stress responses of course maladaptive cultures maladaptive environments which are not right for the person but again it's all about the patterns the pattern recognition in an individual and understanding the patterns of workplaces workforces culture so the key to understand here is that it cannot be a one size fits all approach it has to be I think you're and and that's the way it is with so many of these things like readmissions the AI can predict who's at high risk but then we need a human to uh customize it postpartum hemorrhage the AI can tell who's at high risk but then we need a physician to decide what what exactly were we going to do for that patient.
00:29:39.920 --> 00:30:01.359
So almost all these require that um customized uh intervention the AI is not good at that part it's good at the prediction now Dan if I'm if I get the the if I get the questions you drafted when you were at Vanderbilt and you kindly send them to me would you mind if I just speak to them?
00:30:02.400 --> 00:30:20.799
No sure yeah when we were at Vanderbilt um burnout was an issue there so the dean asked us to um do survey of all the physicians and assess uh what what are some of the factors that are related to burnout and stress.
00:30:21.279 --> 00:30:33.599
And so we did this big study and um I analyzed the data and I tried to uh boil the the results down into something that would be useful for people.
00:30:34.240 --> 00:31:49.039
And um it it boiled down to nine things and I tried to do it in a positive way so that people could go through these questions and and see how many yeses do they get um so uh one question was I get at least eight hours of sleep on a typical night number two I work less than 55 hours in a typical week number three I can I take quiet time for myself four I feel satisfied with my workload five I chose the right career six I have an optimistic outlook on life seven I regularly give and receive affection eight I feel satisfied with my social slash love life and nine I organize my time effectively so these are all things that people can control and if they have a lot of these where they can check them as yes uh uh they're they're in good good shape um but our healthcare system um often has a culture that um that makes people think they don't even deserve to have enough sleep and they and they should work too much.
00:31:49.200 --> 00:32:41.359
So uh uh you know a lot of a lot of the culture in healthcare causes some of these problems very true coming back to this this stress check checkup questions um what I would like to ask you is that sometimes it might be very difficult for a no and yes answer sometimes it might be just in the middle so how how do how does one respond to that kind of a situation this would be you know a good project for somebody to take up and take it to the next level so they could take um how many hours of sleep do you get how and how many hours of work um so um adding the continuous inputs could probably significantly uh improve it but we wanted to keep it um pretty straightforward and uh this is what we were allowed to ask.
00:32:41.759 --> 00:33:38.400
So one of the things I've been doing like if you see that the the aura ring here or the Fitbit watch so it gives a lot of indicators for sleep really detailed indicators of sleep in terms of uh with how much is the total sleep efficiency restfulness REM sleep deep sleep uh and based on that it also creates something known as a sleep debt and also creates uh your readiness score for the day so sometimes if I'm in situations where I know you know some stress is going in the environment and I see that the sleep is starting to shift my my physiology is shifting so the boiling frog model helps us to pick that drift earlier.
00:33:38.640 --> 00:34:12.079
So if I see now that my readiness score is falling down or the sleep patterns are becoming more erratic and the the same thing can be applied for other physiological parameters one can then take the intervention you have but like have better sleep hygiene um work less take more breaks so that is where I feel that the personalized health data is also so helpful from a preventative point of view.
00:34:13.519 --> 00:34:40.960
Yeah I completely agree you know there's a lot of talk about personalized medicine and precision medicine but we really need AI to make that a reality and uh the combining AI and healthcare will allow us to uh truly use uh precision medicine um because it's not going to be possible to really implement precision medicine personalized medicine with without AI helping.
00:34:41.280 --> 00:34:49.360
So Dan where do you feel that uh if in 10 years from now what would be what is your what is your vision?
00:34:49.840 --> 00:35:08.639
Well um I believe over the next three to five years there's gonna be a shakeout in healthcare systems and the 20% are going to win and really use AI to improve patient outcomes and then the other 80% are going to have to quickly change.
00:35:09.119 --> 00:35:42.079
In 10 years uh I I believe that AI will significantly reduce uh medical errors reduce complications catch diseases uh um much sooner um you know the so much of our healthcare system is so reactive and we catch things like heart attacks multiple myeloma complications uh breast cancer AI can catch all of these upstream and then they're much more treatable and they're less expensive.
00:35:42.719 --> 00:35:45.280
And that will help physicians and nurses as well.
00:35:45.519 --> 00:35:59.599
Um you know it will be much less stressful to deal with things that are caught upstream and and have um AI helping um with them so so I think AI will reduce the burnout and stress in clinicians.
00:35:59.920 --> 00:36:23.599
I think it will get even more powerful at being a tool um uh but but there are still there's still uh challenges so for example most people don't realize that up to half the medical literature is wrong so just summarizing the medical literature when up to half of it can be wrong doesn't really give you the right answer.
00:36:24.079 --> 00:36:39.360
So the next uh iteration of these uh these tools has to critically interpret the medical literature just like you and I would and say well this this one has a conclusion that I believe but this one is a completely flawed study.
00:36:39.840 --> 00:36:43.119
Um so there's there's still a lot of work to be done.
00:36:43.599 --> 00:36:49.039
The the real work in AI is well first of all there's a lot of resistance to change.
00:36:49.360 --> 00:36:59.280
So we need healthcare leaders to decide I'm really gonna help my institution use AI to improve patient outcomes.
00:36:59.519 --> 00:37:29.760
And then they're gonna have to get involved in the weekly meetings to get past all of the obstacles and the bureaucracy and the politics so I think the 20% of hospitals that succeed will be the ones where the leaders show up at the weekly AI model meetings and and help that team really overcome the obstacles because without that it's just impossible to get the get the data, build the model, change things.
00:37:30.000 --> 00:37:43.039
So you really need the the CEO and the dean and the and the the executives in these meetings and uh and then they're gonna need some training so that they know how to lead.
00:37:43.119 --> 00:37:56.800
But and that's another problem we have people in our healthcare system who are leaders and paid a lot of money and they don't really understand AI and they don't really understand how to how to do their their job in this new world.
00:37:56.960 --> 00:38:03.519
So we we need some education and we can't expect them to go back get a PhD and they don't need to do that.
00:38:03.840 --> 00:38:08.400
But that's one of the reasons I wrote my book so that they could they could just read my book.
00:38:08.559 --> 00:38:19.760
And that's one of the reasons I teach these courses so they can they can just take our online course when they have some free time and um and get up to speed on what they do need to know.
00:38:19.920 --> 00:38:27.599
And we don't need to teach them how to program in Python but they do need some skills and they need to speak the language.
00:39:11.920 --> 00:39:49.599
There has to be a cultural change to embrace the AI and understand that though there are limitations and which are getting better and better but we don't compare we don't compare AI to perfect what we compare what we compare is that look I'm working in this situation I do not have an access to a specialist and here we have got an AI who is getting better and better and it's getting remarkable and in my experience with the experiments I did I had a very very positive result.
00:39:50.239 --> 00:43:58.880
And I'm as you said that this is getting better and better as the years are going on we are seeing that with Claude we are seeing that with Gemini we're seeing that with chat for health open evidence and so many tools out there so the future really looks uh uh very promising but and then I did not want to deliberately go into the details uh a lot many technical details of LLMs LLM evaluations and the things uh which we learn in the John Hopkins course I just finished actually this weekend and it is such uh synchronous to have a conversation with you now it's it was a great course I would highly recommend anybody who's in healthcare uh to do the John Hopkins healthcare ai course I learned so much so many things it was a personalized um personalized uh uh I would say interactive course uh where one could learn so much in just a three months uh time so I'm I'm really I I feel I've really grown doing with the course but of course uh this is something which you have more and more hands-on experience you keep uh getting better and better so then Dan thank you so much for uh designing such a user friendly and a student friendly friendly course now one thing I'm before we finish I uh was listening to your previous podcast and I found it is very important to discuss that in especially on transforming stress and uh the opposite of stress is joy fulfillment and happiness and you were talking about the American American dream but because we have got a global audience here we know people don't want to get burned out people don't want to be unhappy in their lives they want fulfillment and where do you see the AI's role there in the in the coming um in in the coming decade how can AI make people lead more fulfilled lives yeah um and that's a great question well first of all thanks for uh taking our course we we have a new course coming out on August 1st it's kind of a part two of that course that has a little more hands-on work for people that want to know how to how to make a model and we have some synthetic data sets um you know I think ai can help people improve their lives in uh in so many ways uh we it's not perfect um but uh as you said our healthcare system is not perfect either so nothing nothing is but nothing is perfect and a perfection perfection is an illusion if you see a magnet then I don't have my magnet just now um if you see a magnet a magnet has a south pole and a north pole so any system whether it is a healthcare system or any kind of a system with will have their strengths and weaknesses and if we want to take ourselves as a system we will have our own strengths and weaknesses the key here is to understand what our strengths are how do we use our strengths and understand the patterns of our weaknesses and how do we manage them so I feel the answer is self-awareness how we can with self-awareness understand that and and continue to improvise yeah and I think AI can make people's lives um uh significantly better in so many ways um but we we do have to be careful especially with children and and sensitive topics um but it can function um uh as a as a therapist sometimes some people don't have access to a therapist and um or they might not feel comfortable talking to a therapist and there's a shortage of therapists so you can help out there um but I'd like to see some some rigorous studies to make sure that's safe.
00:43:59.199 --> 00:44:09.119
It's great for Everyday things, you know, things break in your house and you don't know how to fix it, and you just ask Gemini or ChatGPT and it saves you know half an hour.
00:44:09.280 --> 00:44:16.320
And uh every day with uh your to-do list, you can think like, could AI really help me here?
00:44:16.800 --> 00:44:23.679
Um, you know, in in medical research, it's enormously helpful and it's gotten very, very sophisticated.
00:44:23.840 --> 00:44:33.920
Where you know it used to catch a spelling mistake or a grammatical error, and now it it's uh like the smartest person you ever met giving you advice.
00:44:34.320 --> 00:44:45.440
So AI, I I'd encourage people to experiment with Gemini and Chat GPT and open evidence and uh just you know, ask you questions throughout the day.
00:44:45.599 --> 00:45:01.199
And uh the other way to handle this is on my phone and on my computer, I set up a folder and I put about six different AI tools in there and I have them compete with one another and see like which one impresses me today with this this job.
00:45:01.519 --> 00:45:18.880
And before I send out an email or or an important letter, I'll cut and paste it and you know ask it to improve it, and then I'll take that version and paste it into another AI tool, and uh it's become remarkably uh uh sophisticated.
00:45:19.039 --> 00:45:29.199
It's it's really astonishing the level of of intelligence that these tools are are now providing, and they're they're just getting better.
00:45:29.440 --> 00:45:40.639
But for healthcare, like you said at the beginning, um you know, the the the saying of move fast and break things, that doesn't work in healthcare because that's move fast and kill people.
00:45:40.880 --> 00:45:43.119
And in healthcare we we can't do that.
00:45:43.280 --> 00:45:49.119
We have to we have to use these tools, but we have to we have to use them like like a brand new drug.
00:45:49.519 --> 00:46:05.199
And and I've tried to demonstrate with my work and my teaching that it's possible to do it the right way, and the people that make excuses and say, Oh, randomization is gonna be unethical, take too long, it's too expensive.
00:46:05.440 --> 00:46:06.719
We've proven them wrong.
00:46:07.280 --> 00:46:13.119
And not everyone's gonna listen to us, but the 20% that are going down the successful path will.
00:46:13.519 --> 00:46:19.760
The 80% that are going down the wrong path, they'll have short-term success and then it will implode.
00:46:20.400 --> 00:46:28.159
So, you know, everybody gets uh a decision at the fork in the road, and uh I can only encourage people to go down the right fork.
00:46:28.559 --> 00:46:28.960
Thank you, Dan.
00:46:29.679 --> 00:46:35.920
I I feel that uh the future is promising with caution, like with anything.
00:46:36.559 --> 00:46:44.639
When computers came twenty twenty twenty-five years ago, we see it's been a game changer, and the same thing with the AI.
00:46:44.719 --> 00:46:56.880
I think there will be much more quantum shift with uh with with AI, and uh it will be definitely a better definitely a better better world.
00:46:57.119 --> 00:47:02.559
So Dan, we are coming to the end of our our talk, and I'm sure we are going to meet again.
00:47:02.800 --> 00:47:16.559
Would you have any final message for the audience and anything you would like to say your from your experience and your learnings uh to quick share?
00:47:17.039 --> 00:47:26.400
Well, I believe AI has become very powerful, and I believe it's really gonna help our day-to-day lives, and it's really gonna transform healthcare.
00:47:26.719 --> 00:47:38.559
But I think we do need to uh uh keep raising the bar and uh and rigorously evaluate it and make sure it's safe and uh and make sure we do this the right way.
00:47:38.960 --> 00:47:47.119
We we need to we need to have a meter to know is AI actually making things better compared to the way we used to do things.
00:47:47.440 --> 00:48:00.639
And uh and the people that do that, I think it will it it will reduce medical errors and complications, catch things upstream, and make life better for doctors and nurses and patients.
00:48:02.400 --> 00:48:02.880
Thank you, Dan.
00:48:03.039 --> 00:48:29.280
And Dan, thank you so much for your support with the frog, uh with the boiling frog, and also your insights that how AI can be integrated into this, and we take it even to a higher taken it even to a higher level, that not only the stress management is contextualized to the person in front of us, but also it's a dynamic thing.
00:48:29.519 --> 00:48:35.599
So, what where the person is today, he might not be at the same place uh next year.
00:48:35.760 --> 00:48:44.400
You know, life is happening to all of us, whether in personal life or professional environments and different kinds of shifts are happening.
00:48:44.639 --> 00:48:52.800
So the boiling frog catches these drifts before they become much more serious.
00:48:53.039 --> 00:49:08.000
And the new version, the I frog version, is is have will have a layer, will have a layer, and I'm gonna integrate what all I learned at the John Hopkins course into this to make this really better and better.
00:49:08.239 --> 00:49:27.599
So Dan, thank you so much for uh joining us at Transforming Space and also for the amazing work you have done in this area for the last 40 years, and we know that your work is going to help millions of people all around the world.
00:49:27.679 --> 00:49:29.119
There's no doubt about it.
00:49:29.280 --> 00:49:46.320
The way the landscape is shifting, because here we are taking the globe into consideration where there is no access to healthcare, where there is absolutely minimal access to the specialist, I can see where it is going.
00:49:46.559 --> 00:49:50.719
So it's really been uh a joy and a privilege to learn with you.
00:49:50.800 --> 00:49:55.360
And I look forward to our communications and collaborations in the future.
00:49:55.840 --> 00:50:09.519
And thank you for inviting me, and thank you for writing those great books and providing that uh great program, uh Burnout and Stress is such a big issue in healthcare, and you're one of the leaders at fixing that problem.
00:50:09.679 --> 00:50:11.679
So thank you for all you've done.
00:50:12.079 --> 00:50:12.800
Thank you, Dan.
00:50:12.960 --> 00:50:28.159
And as you said, that if AI is going to be used by physicians and professionals, and we don't want the burnt-out professionals to use it, because if they if they use it incorrectly, that's what the result is going to be.
00:50:28.400 --> 00:51:01.679
So they if they embrace it in the right spirit, it will help them to cognitively offload and have a better mental well-being, and those positions then will it's going to be a positive spiral of hope, of fulfillment, of achievement, and day-to-day practice, your day-to-day uh your day-to-day experience of and it's it's uh, and we are going a little longer, and it's not just about patient safety.
00:51:01.920 --> 00:51:08.719
In any, I would say, in healthcare and any industry, it is safety is one of the biggest parameters.
00:51:08.960 --> 00:51:16.000
But more than that, that there is a patient or a client experience.
00:51:16.800 --> 00:51:23.440
So the person who is coming to you, you would at least expect them to be smiling, expect them to be courteous.
00:51:24.159 --> 00:51:32.239
And if somebody is burnt out, they might not have that empathy, that care, that compassion to give.
00:51:32.480 --> 00:51:33.840
Yeah, yeah.
00:51:33.920 --> 00:51:43.519
So I think it's a great uh direction we are taking it, and thank you for being a pioneer in this field, and I will look forward to our collaboration.
00:51:44.719 --> 00:51:48.400
If you enjoyed today's episode, we'd greatly appreciate it.
00:51:48.480 --> 00:51:56.159
If you could leave a five-star review, a like, or subscribe on Apple, Spotify, or wherever you listen to your podcast.
00:51:56.480 --> 00:52:02.079
Your support helps us reach more people, looking to transform their stress into new content.
00:52:02.400 --> 00:52:05.840
We'd love to hear your thoughts, so don't forget to comment and share.
00:52:06.079 --> 00:52:12.239
For more tips and updates, please be sure to check out our social media links in the description box below.
00:52:12.480 --> 00:52:19.039
We can't wait to have you with us next time as we continue this journey towards turning stress into resilience.
00:52:19.280 --> 00:52:24.559
Remember, it's not the stress itself, but how we rise above it that defines our strength.
00:52:24.639 --> 00:52:29.760
So stay resilient and keep thriving, and we will see you next time.
00:52:38.079 --> 00:52:39.199
All right, well, thank you.
00:52:39.280 --> 00:52:40.159
I enjoyed it.
00:52:40.400 --> 00:52:41.280
Thank you.
00:00:04.799 --> 00:00:08.480
Hello friends and a very warm welcome to Christmas Christmas.
00:00:09.519 --> 00:00:12.320
I am ready to test into your content.
00:00:18.320 --> 00:00:21.120
India, the United Kingdom, and the United States.
00:00:41.759 --> 00:00:46.079
To help you manage your stress, find balance, and live a life of purpose.
00:00:46.240 --> 00:00:48.719
Please join us every Friday at 5 p.m.
00:00:48.880 --> 00:00:51.840
and let's start turning stress into strength together.
00:00:52.000 --> 00:00:54.479
Now let's dive into today's episode.
00:00:56.560 --> 00:00:57.759
Hello friends.
00:00:58.000 --> 00:01:00.640
Welcome to the Transforming Stress with Dr.
00:01:00.799 --> 00:01:02.320
Ash podcast.
00:01:02.560 --> 00:01:06.879
And today today's episode is really exciting.
00:01:07.280 --> 00:01:17.519
In the last couple of years, all of us have been hearing about AI, artificial intelligence, and it's very exciting.
00:01:18.079 --> 00:01:22.640
A lot of people say that it's the new fire.
00:01:22.879 --> 00:01:26.719
Equally, there are a lot of people who are worried about it.
00:01:34.640 --> 00:01:48.959
So to break this myth, a lot of myths, a lot of confusion, I wanted to get an expert, somebody who has been navigating this landscape for 40 years, 40 years, four decades.
00:01:49.359 --> 00:01:51.120
AI in healthcare.
00:01:51.439 --> 00:02:07.519
Daniel Byrne is a faculty of AI in healthcare, and he is the person who is going to help us take the help help us to understand what is the evidence and what is the myth.
00:02:07.760 --> 00:02:10.560
And please help me welcome Dan Byrne.
00:02:10.879 --> 00:02:15.199
Thank you, Dan, for joining us from John Hopkins.
00:02:15.680 --> 00:02:17.199
Well, thank you for inviting me.
00:02:17.439 --> 00:02:20.960
Yes, AI has become incredibly powerful.
00:02:21.120 --> 00:02:28.479
And I think almost everybody agrees that the area of society where we're going to see the real benefit of AI is in healthcare.
00:02:28.719 --> 00:02:37.439
But uh the big question is why is there almost no rigorous evidence that we're seeing AI is improving patient outcomes?
00:02:37.599 --> 00:02:44.479
There's a lot of talk, there's a lot of weak evaluations, but the big question is when is it going to help patients?
00:02:44.560 --> 00:02:46.639
And that's been the focus of my work.
00:02:47.039 --> 00:02:59.280
Thank you, Dan, for doing that very important work because medicine is a field where we cannot have any error, because an error means life.
00:02:59.520 --> 00:03:06.479
And whilst there is a lot of excitement, there are new tools, there are a lot lot much promise.
00:03:06.719 --> 00:03:11.840
And I know you are asking the question, what is the evidence?
00:03:12.000 --> 00:03:19.280
What is the evidence that this is going to improve improve the patient outcome?
00:03:19.599 --> 00:03:24.240
So then we have got listeners from 56 countries from all over the world.
00:03:24.400 --> 00:03:37.520
So they are not just doctors or healthcare professionals, we have got a huge wide audience who are very much interesting interested in learning this landscape because this is where the future is.
00:03:37.840 --> 00:03:43.919
And I know you've been in the data sciences uh for for four decades.
00:03:44.240 --> 00:03:57.599
Now, after spending like decades looking at data, looking at numbers, what can the numbers and data tell us even before humans understand that there is a problem?
00:03:58.000 --> 00:04:00.000
Yeah, so that's the important question.
00:04:00.240 --> 00:04:13.360
And what the real benefit of artificial intelligence and healthcare is that it can predict things much more accurately than humans, and it can predict things much sooner and upstream.
00:04:13.599 --> 00:04:16.639
So our current healthcare system is very reactive.
00:04:16.879 --> 00:04:19.360
We catch things when they're too late.
00:04:19.920 --> 00:04:25.920
And uh you humans are not that great at predicting a lot of complex things.
00:04:26.160 --> 00:04:30.959
Um and Daniel Kahneman won the Nobel Prize for his work in this area.
00:04:31.199 --> 00:04:37.600
And if anyone hasn't read his his book, Thinking Fast and Slow, it's a fascinating book to read.
00:04:37.920 --> 00:04:42.480
But uh he showed that humans are very, very confident in their predictions.
00:04:42.720 --> 00:04:44.480
They're just not very good at prediction.
00:04:45.199 --> 00:04:54.480
And in healthcare, there's so many areas where we ask doctors and nurses to predict things that are really not humanly possible to predict.
00:04:54.879 --> 00:04:56.560
So they'll do their best.
00:04:56.879 --> 00:05:02.079
But now we have these AI tools that can predict things much more accurately.
00:05:02.959 --> 00:05:04.480
I'll give you an example.
00:05:04.800 --> 00:05:10.639
So uh a physician asked us to help build a tool to predict postpartum hemorrhage.
00:05:12.000 --> 00:05:19.279
And if people don't know what that is, after a woman gives birth, she can bleed to death or bleed excessively.
00:05:19.439 --> 00:05:23.600
And you might think this doesn't happen much anymore, um, but it does happen.
00:05:23.680 --> 00:05:27.279
It's a leading cause of maternal morbidity and mortality.
00:05:28.079 --> 00:05:45.759
Um so we built a very accurate way to predict postpartum hemorrhage, and um then we also tested the doctors and nurses that do this every day and asked them, uh, do you think this woman is likely to have a postpartum hemorrhage?
00:05:46.399 --> 00:05:49.360
And we showed they weren't much better than flipping a coin.
00:05:50.000 --> 00:06:02.800
So we implemented this AI predictive model into the hospital computer and then we did something that almost nobody does that I think is the real key to the next level of AI in healthcare.
00:06:03.360 --> 00:06:08.639
And that's for one year we performed a pragmatic randomized control trial.
00:06:08.800 --> 00:06:20.079
And that means as every woman came who came through the hospital to deliver a baby, she was randomized either usual care or usual care plus this model.
00:06:20.639 --> 00:06:27.600
And if the probability of postpartum hemorrhage was high, then there's um a bundle uh to prevent it.
00:06:28.160 --> 00:06:30.560
So we're we're waiting to see at the end of this year.
00:06:30.639 --> 00:06:38.240
We'll uh look at the results and hopefully this model predicted and prevented postpartum hemorrhage.
00:06:38.639 --> 00:06:39.920
And I think that's the key.
00:06:40.079 --> 00:06:44.720
We need to take these AI tools and test them, make sure they're safe.
00:06:44.959 --> 00:06:46.160
It could be unsafe.
00:06:46.560 --> 00:06:47.839
And is it effective?
00:06:48.160 --> 00:06:50.160
And is there a return on investment?
00:06:50.480 --> 00:07:03.199
So the same thing can be done for cancer, catching it upstream, blood clots, readmissions, and in your area, predicting stress, burnout, and suicide, we catch these things too late.
00:07:03.439 --> 00:07:10.879
AI tools can catch it much earlier, but we can't just uh give it a free pass.
00:07:11.040 --> 00:07:16.079
We need to test in rigorous science is this safe and is it effective?
00:07:16.560 --> 00:07:20.879
Uh very uh very, very astute uh Dan.
00:07:21.279 --> 00:07:30.240
And I was reading about your burn test because you are so passionate about uh getting it right.
00:07:30.399 --> 00:07:39.839
And uh please correct me if I'm wrong here, that uh your colleagues in the corridor conversations used to say, well, does it pass the burn test or not?
00:07:40.240 --> 00:07:56.079
And it is it is whether this tool is able to go through that test of fire and see whether this is going to going to work in the real world or not.
00:07:56.240 --> 00:08:04.079
And as Dan just mentioned, the same thing, because um then let me try to simplify this and please correct me if I'm wrong.
00:08:04.240 --> 00:08:20.480
So, what I've understood, whether it is research, whether it is clinical medicine, whether it is stress management, it is it is all about it is all about pattern recognition.
00:08:21.040 --> 00:08:22.240
Pattern recognition.
00:08:22.480 --> 00:08:40.080
Data are continuously giving us patterns, and we see what these patterns are telling us, and how we can then gain insights and utilize these patterns.
00:08:40.399 --> 00:08:46.720
Right, and that's why AI ha can go beyond human knowledge and predict things better.
00:08:47.039 --> 00:08:58.320
So, for example, when humans struggle to predict postpartum hemorrhage, they could only put a few inputs in their brain to uh compute that probability.
00:08:58.559 --> 00:09:05.519
But our AI model uh includes 21 predictors uh to look at that pattern.
00:09:05.919 --> 00:09:14.639
So it's a it's a more complex um pattern than humans uh can can do in their head, and it does it fast and free.
00:09:14.799 --> 00:09:20.000
And and there are patterns also in a mammogram that radiologists can't see.
00:09:20.159 --> 00:09:30.320
There are patterns in um retinal scans that have amazed people at what uh the AI can can determine with a retinal scan, an EKG.
00:09:30.639 --> 00:09:38.399
There are there are amazing things in the patterns that AI detects that that humans never even imagined you could you could tell.
00:09:38.639 --> 00:09:53.200
And Daniel, that makes me the question with a lot of people and doctors and professionals are fearing, is AI going to take away their take away their jobs, going to replace, replace them.
00:09:53.679 --> 00:10:00.639
And I will tell I will tell you, I will share with you my experiment, which I did in the last six months.
00:10:00.960 --> 00:10:09.600
So what I did that whenever I'm I'm an internal medicine physician, so I'm on call and I'm seeing complicated patients.
00:10:09.919 --> 00:10:23.919
Now, if I'm in a situation that I need to take uh advice from a specialist like a cardiologist or a respiratory physician, and thankfully I've got the place I'm working, I've got access to the specialists.
00:10:24.240 --> 00:10:33.840
So I I see the patient and I know I need to speak to the to the cardiologist here, and these are the questions I'm going to ask.
00:10:34.080 --> 00:10:48.960
Now, before I do that, before I pick up the phone, in my mind I think suppose I am working in a remote area, or I am working in an island where there is no access to the specialist, what can I do?
00:10:49.200 --> 00:10:54.399
So I open the phone and go to open evidence six months.
00:10:54.720 --> 00:11:04.960
Um six months ago, I I anonymize all the data, and I would give that this is the situation I am, and I need some advice from the specialist.
00:11:05.120 --> 00:11:08.080
And the open evidence gives me the gives me the advice.
00:11:08.159 --> 00:11:23.600
And it makes because I am an internal medicine physician myself, I am able to make sense of what uh what the evidence or what the app is telling me, and then I speak to the specialist.
00:11:24.000 --> 00:11:34.320
And I have been amazed that that the advice is so similar, and the other thing is that the AI has continued to improve.
00:11:34.399 --> 00:11:37.919
Now, unfortunately, open evidence is no longer working in Europe.
00:11:38.000 --> 00:11:44.399
So I after speaking to you, I've started using Gemini, which is also equally equally good.
00:11:44.639 --> 00:11:49.840
But the main thing here is to give AI the context.
00:11:50.080 --> 00:11:55.919
So I'm answering my own question that the human judgment has to be there.
00:11:57.039 --> 00:12:02.159
Deal who's dealing with the patient or situation in front of them.
00:12:02.639 --> 00:12:08.320
So we this is getting better and better as times goes.
00:12:08.480 --> 00:12:17.039
So I wanted to hear from you about um about what the landscape is looking and how do you foresee this shifting in the future?
00:12:17.360 --> 00:12:30.559
Yeah, I don't believe AI is going to replace doctors, but the doctors who learn about AI uh will replace doctors that refuse to learn it, just like it always happens with new technology.
00:12:30.720 --> 00:12:39.039
There are people that resist learning new technology and they're replaced by the people who are more open-minded and learn the new technology.
00:12:39.279 --> 00:12:53.840
Um and we really need a clinician computer symbiosis where we need the physician to do certain things, and we need to let the computer do certain things, and we need the best of both worlds there.
00:12:54.080 --> 00:12:57.039
And that's part of Kahneman's uh research too.
00:12:57.360 --> 00:13:07.840
So, for example, we built an AI tool to predict hospital readmissions, 30-day hospital readmissions, um, and it predicted very accurately.
00:13:08.159 --> 00:13:15.440
Now, some places also tried to have their AI tool decide what to do for the high-risk patients, and they failed.
00:13:16.000 --> 00:13:30.080
So you what you need to do is have the computer do what it's good at, predicting which patients are likely to be readmitted, and then have uh clinicians decide what does that patient need who's a high risk.
00:13:30.399 --> 00:13:32.720
And and then you get the best of both worlds.
00:13:33.120 --> 00:13:44.480
Now, you did point out that in some countries there's a lack of specialists, and we're gonna need to use AI because there just aren't enough uh doctors who are specialists.
00:13:44.559 --> 00:14:01.840
Uh one striking statistic, and you may know more about this than I do, but in India there's one pathologist every nine million people, and in a lot of countries, uh we even in the US, there's a lack of therapists and rheumatologists.
00:14:02.080 --> 00:14:06.720
So we are gonna have to use AI when there's a shortage of people.
00:14:06.799 --> 00:14:12.320
But uh but I think the symbiosis is really the key to using AI effectively.
00:14:12.720 --> 00:14:19.919
And how how do you think this landscape is going to change uh in the next five to ten years?
00:14:20.080 --> 00:14:24.480
Because you have seen how things have shifted in the last 40 years now.
00:14:24.720 --> 00:14:27.039
And how fast is the speed of change here?
00:14:27.440 --> 00:14:28.320
Well, it's interesting.
00:14:28.399 --> 00:14:34.240
So AI is moving fast, but healthcare is notoriously resistant to change.
00:14:34.480 --> 00:14:38.720
So I think what's gonna happen in the next five years is there's gonna be a shake out.
00:14:38.879 --> 00:14:41.039
There's a lot of people investing money in this.
00:14:41.120 --> 00:14:44.960
Uh uh,$100 billion is being invested in AI and healthcare.
00:14:45.279 --> 00:14:46.240
There's gonna be a shakeout.
00:14:47.120 --> 00:14:52.559
80% of these startups and and researchers and and experts are gonna fail.
00:14:52.960 --> 00:14:54.720
Twenty percent are gonna be successful.
00:14:54.960 --> 00:14:58.799
Twenty percent of hospitals are gonna be successful at implementing AI.
00:14:59.200 --> 00:15:07.120
They're gonna reduce complications, improve survival, uh, they're gonna have much better uh results, detect things earlier.
00:15:07.440 --> 00:15:16.799
And then patients are gonna flock to those 20% of hospitals that use AI right, and then the other 80% are gonna be forced to change.
00:15:17.039 --> 00:15:20.320
And what are the eighty what are the 20% gonna do differently?
00:15:20.639 --> 00:15:25.039
They're not just gonna embrace AI and invest in it.
00:15:25.440 --> 00:15:28.720
They're gonna really uh treat it like a new drug.
00:15:29.120 --> 00:15:32.399
And they're gonna say, is is this thing safe?
00:15:32.879 --> 00:15:36.639
They're gonna test that first, and then they're gonna test, is this effective?
00:15:37.360 --> 00:15:38.879
And they're gonna do this.
00:15:39.039 --> 00:15:50.480
Uh the burn test is really a pragmatic, randomized, controlled trial of AI improving patient outcomes, but it's done with a very high bar.
00:15:51.039 --> 00:15:56.639
And it's done in a way that if you get it right, you could go in the New England Journal of Medicine.
00:15:56.960 --> 00:16:00.639
So that means you have to pre-register it with clinicaltrials.gov.
00:16:01.120 --> 00:16:08.799
You have to do all of the rigorous steps to get the right answer, and that's really what the New England Journal of Medicine wants.
00:16:08.960 --> 00:16:11.279
They just want to know that they're not going to get burned.
00:16:11.360 --> 00:16:17.600
And if you say your AI tool improved patient outcomes, they're comfortable because you did all of the right steps there.
00:16:17.919 --> 00:16:18.559
Thank you, Dan.
00:16:18.720 --> 00:16:25.440
Then now we are going to shift to another area of very much interest to me.
00:16:26.240 --> 00:16:34.559
And uh, I know you mentioned about uh the book Thinking Fast and Slow, the Type 1 and Type 2 thinking.
00:16:34.639 --> 00:16:40.159
And I want to share that in the context of the physician and the healthcare burnout.
00:16:40.240 --> 00:16:58.080
You know, you know in the United States that uh in the last five to ten years there have been studies every every year which have said that there are 50 to 60 percent at least burnout rates in the physicians, and I think that all over the globe the stats are not very different.
00:16:58.240 --> 00:17:01.759
But why don't why do we see healthcare in other professions?
00:17:02.080 --> 00:17:07.200
According to the Gallup studies, every year the stats are very, very similar.
00:17:07.440 --> 00:17:29.440
Now, what happens in burnout, in that state of severe exhaustion, in the in the book Thinking Fast and Slow, what it talks about in the severe cognitive load, in the exhaustion, in the burnout, the thinking automatically shifts from the type 2 systematic thinking to the type 1 thinking.
00:17:29.599 --> 00:17:35.200
And the type 1 thinking has 10 to 15 percent of error rate.
00:17:35.359 --> 00:17:51.519
Now, if I tell you that you are going to fly, I'm flying you out to now give a talk, flying you from Boston to London, and the plane has a 10% risk of crashing and error rate, then you will not be able to sleep.
00:17:51.680 --> 00:17:53.599
You will not you're not able to sleep.
00:17:53.920 --> 00:18:11.519
Now, the aviation industry is the safest industry because what they have done at with the system's approach and all the work which has been put in the safety, the landscape has completely changed, and medicine has to learn from it.
00:18:11.599 --> 00:18:16.319
But of course, the human factors are so much in the in the industry.
00:18:16.640 --> 00:18:30.640
So we are talking about how AI can help us as a great tool to cognitively unload the physicians, unload the healthcare professionals.
00:18:30.799 --> 00:18:47.920
And I personally found it to be a great tool over the last year in um in um in doing in working working with uh as a as a Microsoft co-pilot or a chat for health or open evidence or or Gemini.
00:18:48.480 --> 00:18:53.359
So what are your what are your thoughts about uh the AI and the burnout?
00:18:53.759 --> 00:19:03.359
Yeah, and so you make a good point that the aviation industry has become remarkably successful at um at at being safe.
00:19:03.519 --> 00:19:07.759
And part of that is they've been using AI for a long time.
00:19:08.079 --> 00:19:16.079
And I I have a section in my book about how the aviation industry is safer because they apply AI.
00:19:16.400 --> 00:19:24.799
And I think we can learn from uh that and and apply AI to uh to do things safer in healthcare.
00:19:25.119 --> 00:19:29.119
And burnout is a huge problem in in healthcare and other industries.
00:19:29.359 --> 00:19:31.359
And AI can help with a lot of that.
00:19:31.839 --> 00:19:47.680
You know, uh w one area is with AI scribes so that physicians don't have to stay up late at night typing uh patient notes and we can just take some of the administrative work off of physicians and allow AI to do it.
00:19:48.160 --> 00:19:50.559
But again, it needs to be studied.
00:19:50.640 --> 00:19:56.079
We need to make sure that's safe and we need to make sure that that is actually uh helping them.
00:19:56.480 --> 00:20:05.200
And people often skip the opportunity to do a simple thing like a pragmatic RCT of the AI scribes.
00:20:05.359 --> 00:20:07.759
That's an easy thing to test out.
00:20:08.240 --> 00:20:19.279
The other way that I believe we need to test AI in healthcare is stop pretending like it's a one and done thing where we we just test it and we have one answer.
00:20:19.519 --> 00:20:22.720
We should be thinking about these things over the next ten years.
00:20:22.960 --> 00:20:47.759
We're gonna have burnout and stress, so we should think over the next ten years, there's gonna be various AI scribes and other tools, and we should do prag pragmatic uh RCTs over the next ten years and just keep asking different questions in an adaptive platform trial where we drop arms that are not working, and we add a new study arm that is working.
00:20:48.480 --> 00:20:58.640
And we keep asking questions in a factorial design, which just means that we can randomize in different ways to ask different questions.
00:20:58.880 --> 00:21:14.880
So we're really a learning healthcare system and we're continuously getting better, and and we apply continuous quality improvement to make uh the work uh safer and and just more enjoyable for physicians and nurses.
00:21:15.359 --> 00:21:20.319
AI can can help with so many administrative jobs in healthcare.
00:21:20.480 --> 00:21:24.400
Um for every doctor in the hospital, there are nine administrators.
00:21:25.039 --> 00:21:27.039
30% of healthcare is waste.
00:21:27.279 --> 00:21:37.119
AI can really help in a lot of these areas, and um but we we can't just be paying a lot of money for AI tools that maybe they don't work.
00:21:37.200 --> 00:21:39.680
We need to really evaluate if they do work.
00:21:40.000 --> 00:21:40.240
True.
00:21:40.319 --> 00:21:49.599
So the future, if it is the future if it is approached with caution and due diligence looks very promising.
00:21:50.079 --> 00:21:52.640
And I can I can go ahead.
00:21:52.960 --> 00:22:15.200
Yeah, no, I can say that from my own experience that the work over the last one year uh I have found a shift uh in both the cognitive load and all the learnings and all which which is which is going into it and uh I'm I feel I can sense that that it is going to go in a very very positive and a safe direction.
00:22:15.440 --> 00:24:03.440
You know last time we were speaking then about the uh and you gave the example of uh you gave the example of postpartum hemorrhage you spoke about SLE was it SLE uh Lupus you talk about the tool uh you were developing I would love for you to share about that because that is remarkable that amount and there are we know that so many times that there are missed diagnoses there are delayed diagnoses so can you speak to more from a chronic chronic health issues point of point of view sure so uh we we have about 10 different projects going on now testing ai but one of the ones uh that uh just passed the burn test is uh we had a physician who is a rheumatologist and she came to us and she said I need a way to predict which patients once they have a positive ANA test are going to develop an autoimmune disease like lupus and uh she said right now rheumatologists are just overwhelmed and they can't really predict this and the patients often uh go on a diagnostic odyssey and it takes on average seven years before patients get their diagnosis of lupus and other autoimmune diseases and seven years is a a long time and during that time the disease eats away at their body and so she asked can we use AI to predict it faster than seven years so we worked with her met with her every week and um and we developed um a way that once a patient has a positive ANA we take their information and we put it into an AI tool to compute probability of an autoimmune disease.
00:24:04.160 --> 00:24:17.359
And then we did what hardly anyone does we said for the next year we're gonna do a pragmatic randomized control trial and we don't take any anything away from anyone everybody continues to get usual care.
00:24:17.599 --> 00:24:22.640
So we randomized them to usual care or usual care so people calm down.
00:24:23.200 --> 00:24:33.359
And then in one arm on top of usual care we're gonna compute the probability of an autoimmune disease and if it's high we're gonna send them to a rheumatologist.
00:24:34.160 --> 00:24:43.279
And we just uh closed out the study and it it significantly reduced time to diagnosis down to 43 days.
00:24:44.000 --> 00:25:00.880
And we have the paper um the New England Journal of medicine AI Journal gave us very positive reviews and we're just wrapping up the reviews and hopefully in another month we'll get this published and uh that that's an example of how you can do this right.
00:25:01.039 --> 00:25:07.759
Because often people say well randomization is going to be too expensive take too long it's it's not possible.
00:25:08.079 --> 00:25:26.480
So we try to demonstrate with these projects um examples that people can copy and and you could think of um a hundred other areas of healthcare where it's reactive and you could use AI in a similar way but then evaluate it and uh and really prove it.
00:25:26.559 --> 00:25:29.359
Don't declare victory just because you have a model.
00:25:29.839 --> 00:25:41.759
True Dan and the same thing can be applied to any chronic physical health issue the same thing can be applied to any chronic mental health issue.
00:25:42.160 --> 00:25:50.079
So suicide stress burnout um there's lots of opportunities to catch these things earlier.
00:25:50.640 --> 00:26:02.720
Much much earlier the pattern recognition much earlier now I see uh on your desk behind a copy of my book The Boiling Frog sitting there.
00:26:03.039 --> 00:29:39.759
Yeah I really enjoyed that it's uh such such a great book I've I enjoy all the quotes and the and the illustrations and I was trying to think how could AI combine with your workshops and I was I was thinking um so let's say a medical center is interested in uh hiring you for some workshops and and helping with the doctors um first they could take um an AI tool or something like our stress checkup uh the nine items there and just have all the the physicians uh complete this and the ones that have a high probability of um stress or burnout uh could take your workshop but then you could also take this um to the next level and say let's do this for a random half and you could ask all the doctors would you be willing to participate in a pragmatic randomized controlled trial and the and the control group could be a delayed intervention so they could get it next year but that way you could test what's the impact of assessing uh uh burnout and then having an intervention and and that's uh that could be a landmark paper that really evaluates the uh the evidence thank you Dan so the boiling frog is uh a full a decade worth of work which I have put my efforts into and it's a systems based approach for stress management and what it means that what if I take hundred people through this intervention the everybody's answers are going to be different because you see everybody's physiology is different everybody's strengths are different their values are different their physiology in terms of their sleep and nutritional new aspects and and there are so their situations at work the culture they are working in there are so many data points into that so a one size fits all approach is not going to work that's why I created a systems based approach with the boiling frog now we are going to take it to a different level now with iFrog which is put another level of the AI integrated into this because as you were sharing with me earlier that it's all about pattern recognition stress is nothing but maladaptive thought patterns maladaptive stress responses of course maladaptive cultures maladaptive environments which are not right for the person but again it's all about the patterns the pattern recognition in an individual and understanding the patterns of workplaces workforces culture so the key to understand here is that it cannot be a one size fits all approach it has to be I think you're and and that's the way it is with so many of these things like readmissions the AI can predict who's at high risk but then we need a human to uh customize it postpartum hemorrhage the AI can tell who's at high risk but then we need a physician to decide what what exactly were we going to do for that patient.
00:29:39.920 --> 00:30:01.359
So almost all these require that um customized uh intervention the AI is not good at that part it's good at the prediction now Dan if I'm if I get the the if I get the questions you drafted when you were at Vanderbilt and you kindly send them to me would you mind if I just speak to them?
00:30:02.400 --> 00:30:20.799
No sure yeah when we were at Vanderbilt um burnout was an issue there so the dean asked us to um do survey of all the physicians and assess uh what what are some of the factors that are related to burnout and stress.
00:30:21.279 --> 00:30:33.599
And so we did this big study and um I analyzed the data and I tried to uh boil the the results down into something that would be useful for people.
00:30:34.240 --> 00:31:49.039
And um it it boiled down to nine things and I tried to do it in a positive way so that people could go through these questions and and see how many yeses do they get um so uh one question was I get at least eight hours of sleep on a typical night number two I work less than 55 hours in a typical week number three I can I take quiet time for myself four I feel satisfied with my workload five I chose the right career six I have an optimistic outlook on life seven I regularly give and receive affection eight I feel satisfied with my social slash love life and nine I organize my time effectively so these are all things that people can control and if they have a lot of these where they can check them as yes uh uh they're they're in good good shape um but our healthcare system um often has a culture that um that makes people think they don't even deserve to have enough sleep and they and they should work too much.
00:31:49.200 --> 00:32:41.359
So uh uh you know a lot of a lot of the culture in healthcare causes some of these problems very true coming back to this this stress check checkup questions um what I would like to ask you is that sometimes it might be very difficult for a no and yes answer sometimes it might be just in the middle so how how do how does one respond to that kind of a situation this would be you know a good project for somebody to take up and take it to the next level so they could take um how many hours of sleep do you get how and how many hours of work um so um adding the continuous inputs could probably significantly uh improve it but we wanted to keep it um pretty straightforward and uh this is what we were allowed to ask.
00:32:41.759 --> 00:33:38.400
So one of the things I've been doing like if you see that the the aura ring here or the Fitbit watch so it gives a lot of indicators for sleep really detailed indicators of sleep in terms of uh with how much is the total sleep efficiency restfulness REM sleep deep sleep uh and based on that it also creates something known as a sleep debt and also creates uh your readiness score for the day so sometimes if I'm in situations where I know you know some stress is going in the environment and I see that the sleep is starting to shift my my physiology is shifting so the boiling frog model helps us to pick that drift earlier.
00:33:38.640 --> 00:34:12.079
So if I see now that my readiness score is falling down or the sleep patterns are becoming more erratic and the the same thing can be applied for other physiological parameters one can then take the intervention you have but like have better sleep hygiene um work less take more breaks so that is where I feel that the personalized health data is also so helpful from a preventative point of view.
00:34:13.519 --> 00:34:40.960
Yeah I completely agree you know there's a lot of talk about personalized medicine and precision medicine but we really need AI to make that a reality and uh the combining AI and healthcare will allow us to uh truly use uh precision medicine um because it's not going to be possible to really implement precision medicine personalized medicine with without AI helping.
00:34:41.280 --> 00:34:49.360
So Dan where do you feel that uh if in 10 years from now what would be what is your what is your vision?
00:34:49.840 --> 00:35:08.639
Well um I believe over the next three to five years there's gonna be a shakeout in healthcare systems and the 20% are going to win and really use AI to improve patient outcomes and then the other 80% are going to have to quickly change.
00:35:09.119 --> 00:35:42.079
In 10 years uh I I believe that AI will significantly reduce uh medical errors reduce complications catch diseases uh um much sooner um you know the so much of our healthcare system is so reactive and we catch things like heart attacks multiple myeloma complications uh breast cancer AI can catch all of these upstream and then they're much more treatable and they're less expensive.
00:35:42.719 --> 00:35:45.280
And that will help physicians and nurses as well.
00:35:45.519 --> 00:35:59.599
Um you know it will be much less stressful to deal with things that are caught upstream and and have um AI helping um with them so so I think AI will reduce the burnout and stress in clinicians.
00:35:59.920 --> 00:36:23.599
I think it will get even more powerful at being a tool um uh but but there are still there's still uh challenges so for example most people don't realize that up to half the medical literature is wrong so just summarizing the medical literature when up to half of it can be wrong doesn't really give you the right answer.
00:36:24.079 --> 00:36:39.360
So the next uh iteration of these uh these tools has to critically interpret the medical literature just like you and I would and say well this this one has a conclusion that I believe but this one is a completely flawed study.
00:36:39.840 --> 00:36:43.119
Um so there's there's still a lot of work to be done.
00:36:43.599 --> 00:36:49.039
The the real work in AI is well first of all there's a lot of resistance to change.
00:36:49.360 --> 00:36:59.280
So we need healthcare leaders to decide I'm really gonna help my institution use AI to improve patient outcomes.
00:36:59.519 --> 00:37:29.760
And then they're gonna have to get involved in the weekly meetings to get past all of the obstacles and the bureaucracy and the politics so I think the 20% of hospitals that succeed will be the ones where the leaders show up at the weekly AI model meetings and and help that team really overcome the obstacles because without that it's just impossible to get the get the data, build the model, change things.
00:37:30.000 --> 00:37:43.039
So you really need the the CEO and the dean and the and the the executives in these meetings and uh and then they're gonna need some training so that they know how to lead.
00:37:43.119 --> 00:37:56.800
But and that's another problem we have people in our healthcare system who are leaders and paid a lot of money and they don't really understand AI and they don't really understand how to how to do their their job in this new world.
00:37:56.960 --> 00:38:03.519
So we we need some education and we can't expect them to go back get a PhD and they don't need to do that.
00:38:03.840 --> 00:38:08.400
But that's one of the reasons I wrote my book so that they could they could just read my book.
00:38:08.559 --> 00:38:19.760
And that's one of the reasons I teach these courses so they can they can just take our online course when they have some free time and um and get up to speed on what they do need to know.
00:38:19.920 --> 00:38:27.599
And we don't need to teach them how to program in Python but they do need some skills and they need to speak the language.
00:39:11.920 --> 00:39:49.599
There has to be a cultural change to embrace the AI and understand that though there are limitations and which are getting better and better but we don't compare we don't compare AI to perfect what we compare what we compare is that look I'm working in this situation I do not have an access to a specialist and here we have got an AI who is getting better and better and it's getting remarkable and in my experience with the experiments I did I had a very very positive result.
00:39:50.239 --> 00:43:58.880
And I'm as you said that this is getting better and better as the years are going on we are seeing that with Claude we are seeing that with Gemini we're seeing that with chat for health open evidence and so many tools out there so the future really looks uh uh very promising but and then I did not want to deliberately go into the details uh a lot many technical details of LLMs LLM evaluations and the things uh which we learn in the John Hopkins course I just finished actually this weekend and it is such uh synchronous to have a conversation with you now it's it was a great course I would highly recommend anybody who's in healthcare uh to do the John Hopkins healthcare ai course I learned so much so many things it was a personalized um personalized uh uh I would say interactive course uh where one could learn so much in just a three months uh time so I'm I'm really I I feel I've really grown doing with the course but of course uh this is something which you have more and more hands-on experience you keep uh getting better and better so then Dan thank you so much for uh designing such a user friendly and a student friendly friendly course now one thing I'm before we finish I uh was listening to your previous podcast and I found it is very important to discuss that in especially on transforming stress and uh the opposite of stress is joy fulfillment and happiness and you were talking about the American American dream but because we have got a global audience here we know people don't want to get burned out people don't want to be unhappy in their lives they want fulfillment and where do you see the AI's role there in the in the coming um in in the coming decade how can AI make people lead more fulfilled lives yeah um and that's a great question well first of all thanks for uh taking our course we we have a new course coming out on August 1st it's kind of a part two of that course that has a little more hands-on work for people that want to know how to how to make a model and we have some synthetic data sets um you know I think ai can help people improve their lives in uh in so many ways uh we it's not perfect um but uh as you said our healthcare system is not perfect either so nothing nothing is but nothing is perfect and a perfection perfection is an illusion if you see a magnet then I don't have my magnet just now um if you see a magnet a magnet has a south pole and a north pole so any system whether it is a healthcare system or any kind of a system with will have their strengths and weaknesses and if we want to take ourselves as a system we will have our own strengths and weaknesses the key here is to understand what our strengths are how do we use our strengths and understand the patterns of our weaknesses and how do we manage them so I feel the answer is self-awareness how we can with self-awareness understand that and and continue to improvise yeah and I think AI can make people's lives um uh significantly better in so many ways um but we we do have to be careful especially with children and and sensitive topics um but it can function um uh as a as a therapist sometimes some people don't have access to a therapist and um or they might not feel comfortable talking to a therapist and there's a shortage of therapists so you can help out there um but I'd like to see some some rigorous studies to make sure that's safe.
00:43:59.199 --> 00:44:09.119
It's great for Everyday things, you know, things break in your house and you don't know how to fix it, and you just ask Gemini or ChatGPT and it saves you know half an hour.
00:44:09.280 --> 00:44:16.320
And uh every day with uh your to-do list, you can think like, could AI really help me here?
00:44:16.800 --> 00:44:23.679
Um, you know, in in medical research, it's enormously helpful and it's gotten very, very sophisticated.
00:44:23.840 --> 00:44:33.920
Where you know it used to catch a spelling mistake or a grammatical error, and now it it's uh like the smartest person you ever met giving you advice.
00:44:34.320 --> 00:44:45.440
So AI, I I'd encourage people to experiment with Gemini and Chat GPT and open evidence and uh just you know, ask you questions throughout the day.
00:44:45.599 --> 00:45:01.199
And uh the other way to handle this is on my phone and on my computer, I set up a folder and I put about six different AI tools in there and I have them compete with one another and see like which one impresses me today with this this job.
00:45:01.519 --> 00:45:18.880
And before I send out an email or or an important letter, I'll cut and paste it and you know ask it to improve it, and then I'll take that version and paste it into another AI tool, and uh it's become remarkably uh uh sophisticated.
00:45:19.039 --> 00:45:29.199
It's it's really astonishing the level of of intelligence that these tools are are now providing, and they're they're just getting better.
00:45:29.440 --> 00:45:40.639
But for healthcare, like you said at the beginning, um you know, the the the saying of move fast and break things, that doesn't work in healthcare because that's move fast and kill people.
00:45:40.880 --> 00:45:43.119
And in healthcare we we can't do that.
00:45:43.280 --> 00:45:49.119
We have to we have to use these tools, but we have to we have to use them like like a brand new drug.
00:45:49.519 --> 00:46:05.199
And and I've tried to demonstrate with my work and my teaching that it's possible to do it the right way, and the people that make excuses and say, Oh, randomization is gonna be unethical, take too long, it's too expensive.
00:46:05.440 --> 00:46:06.719
We've proven them wrong.
00:46:07.280 --> 00:46:13.119
And not everyone's gonna listen to us, but the 20% that are going down the successful path will.
00:46:13.519 --> 00:46:19.760
The 80% that are going down the wrong path, they'll have short-term success and then it will implode.
00:46:20.400 --> 00:46:28.159
So, you know, everybody gets uh a decision at the fork in the road, and uh I can only encourage people to go down the right fork.
00:46:28.559 --> 00:46:28.960
Thank you, Dan.
00:46:29.679 --> 00:46:35.920
I I feel that uh the future is promising with caution, like with anything.
00:46:36.559 --> 00:46:44.639
When computers came twenty twenty twenty-five years ago, we see it's been a game changer, and the same thing with the AI.
00:46:44.719 --> 00:46:56.880
I think there will be much more quantum shift with uh with with AI, and uh it will be definitely a better definitely a better better world.
00:46:57.119 --> 00:47:02.559
So Dan, we are coming to the end of our our talk, and I'm sure we are going to meet again.
00:47:02.800 --> 00:47:16.559
Would you have any final message for the audience and anything you would like to say your from your experience and your learnings uh to quick share?
00:47:17.039 --> 00:47:26.400
Well, I believe AI has become very powerful, and I believe it's really gonna help our day-to-day lives, and it's really gonna transform healthcare.
00:47:26.719 --> 00:47:38.559
But I think we do need to uh uh keep raising the bar and uh and rigorously evaluate it and make sure it's safe and uh and make sure we do this the right way.
00:47:38.960 --> 00:47:47.119
We we need to we need to have a meter to know is AI actually making things better compared to the way we used to do things.
00:47:47.440 --> 00:48:00.639
And uh and the people that do that, I think it will it it will reduce medical errors and complications, catch things upstream, and make life better for doctors and nurses and patients.
00:48:02.400 --> 00:48:02.880
Thank you, Dan.
00:48:03.039 --> 00:48:29.280
And Dan, thank you so much for your support with the frog, uh with the boiling frog, and also your insights that how AI can be integrated into this, and we take it even to a higher taken it even to a higher level, that not only the stress management is contextualized to the person in front of us, but also it's a dynamic thing.
00:48:29.519 --> 00:48:35.599
So, what where the person is today, he might not be at the same place uh next year.
00:48:35.760 --> 00:48:44.400
You know, life is happening to all of us, whether in personal life or professional environments and different kinds of shifts are happening.
00:48:44.639 --> 00:48:52.800
So the boiling frog catches these drifts before they become much more serious.
00:48:53.039 --> 00:49:08.000
And the new version, the I frog version, is is have will have a layer, will have a layer, and I'm gonna integrate what all I learned at the John Hopkins course into this to make this really better and better.
00:49:08.239 --> 00:49:27.599
So Dan, thank you so much for uh joining us at Transforming Space and also for the amazing work you have done in this area for the last 40 years, and we know that your work is going to help millions of people all around the world.
00:49:27.679 --> 00:49:29.119
There's no doubt about it.
00:49:29.280 --> 00:49:46.320
The way the landscape is shifting, because here we are taking the globe into consideration where there is no access to healthcare, where there is absolutely minimal access to the specialist, I can see where it is going.
00:49:46.559 --> 00:49:50.719
So it's really been uh a joy and a privilege to learn with you.
00:49:50.800 --> 00:49:55.360
And I look forward to our communications and collaborations in the future.
00:49:55.840 --> 00:50:09.519
And thank you for inviting me, and thank you for writing those great books and providing that uh great program, uh Burnout and Stress is such a big issue in healthcare, and you're one of the leaders at fixing that problem.
00:50:09.679 --> 00:50:11.679
So thank you for all you've done.
00:50:12.079 --> 00:50:12.800
Thank you, Dan.
00:50:12.960 --> 00:50:28.159
And as you said, that if AI is going to be used by physicians and professionals, and we don't want the burnt-out professionals to use it, because if they if they use it incorrectly, that's what the result is going to be.
00:50:28.400 --> 00:51:01.679
So they if they embrace it in the right spirit, it will help them to cognitively offload and have a better mental well-being, and those positions then will it's going to be a positive spiral of hope, of fulfillment, of achievement, and day-to-day practice, your day-to-day uh your day-to-day experience of and it's it's uh, and we are going a little longer, and it's not just about patient safety.
00:51:01.920 --> 00:51:08.719
In any, I would say, in healthcare and any industry, it is safety is one of the biggest parameters.
00:51:08.960 --> 00:51:16.000
But more than that, that there is a patient or a client experience.
00:51:16.800 --> 00:51:23.440
So the person who is coming to you, you would at least expect them to be smiling, expect them to be courteous.
00:51:24.159 --> 00:51:32.239
And if somebody is burnt out, they might not have that empathy, that care, that compassion to give.
00:51:32.480 --> 00:51:33.840
Yeah, yeah.
00:51:33.920 --> 00:51:43.519
So I think it's a great uh direction we are taking it, and thank you for being a pioneer in this field, and I will look forward to our collaboration.
00:51:44.719 --> 00:51:48.400
If you enjoyed today's episode, we'd greatly appreciate it.
00:51:48.480 --> 00:51:56.159
If you could leave a five-star review, a like, or subscribe on Apple, Spotify, or wherever you listen to your podcast.
00:51:56.480 --> 00:52:02.079
Your support helps us reach more people, looking to transform their stress into new content.
00:52:02.400 --> 00:52:05.840
We'd love to hear your thoughts, so don't forget to comment and share.
00:52:06.079 --> 00:52:12.239
For more tips and updates, please be sure to check out our social media links in the description box below.
00:52:12.480 --> 00:52:19.039
We can't wait to have you with us next time as we continue this journey towards turning stress into resilience.
00:52:19.280 --> 00:52:24.559
Remember, it's not the stress itself, but how we rise above it that defines our strength.
00:52:24.639 --> 00:52:29.760
So stay resilient and keep thriving, and we will see you next time.
00:52:38.079 --> 00:52:39.199
All right, well, thank you.
00:52:39.280 --> 00:52:40.159
I enjoyed it.
00:52:40.400 --> 00:52:41.280
Thank you.
