رونوشت 🔗
WEBVTT
Kind: captions
Language: en-US
00:00:00.533 --> 00:00:01.797
[MUSIC]
00:00:01.797 --> 00:00:02.745
[BOOK PASSAGE]
00:00:02.745 --> 00:00:08.000
PETER LEE: “In US healthcare, quality ratings
are increasingly used to tie the improvement in
00:00:08.000 --> 00:00:13.120
patient health outcomes to the reimbursement
rates that healthcare providers can receive.
00:00:13.120 --> 00:00:18.560
The ability of GPT-4 to understand these
systems and give concrete advice … has a
00:00:18.560 --> 00:00:25.869
chance to make it easier for providers
to achieve success in both dimensions.”
00:00:25.869 --> 00:00:25.920
[END OF BOOK PASSAGE]
[THEME MUSIC]
00:00:25.920 --> 00:00:39.200
This is The AI Revolution in Medicine,
Revisited. I’m your host, Peter Lee.
00:00:39.200 --> 00:00:45.040
Shortly after OpenAI's GPT-4 was publicly
released, Carey Goldberg, Dr. Zak Kohane,
00:00:45.040 --> 00:00:50.480
and I published The AI Revolution in Medicine
to help educate the world of healthcare and
00:00:50.480 --> 00:00:56.160
medical research about the transformative impact
this new generative AI technology could have.
00:00:56.160 --> 00:01:02.880
But because we wrote the book when GPT-4 was
still a secret, we had to speculate. Now,
00:01:02.880 --> 00:01:07.440
two years later, what did we get
right, and what did we get wrong?
00:01:07.440 --> 00:01:12.080
In this series, we’ll talk to clinicians,
patients, hospital administrators,
00:01:12.080 --> 00:01:20.720
and others to understand the reality of AI
in the field and where we go from here.
00:01:20.720 --> 00:01:24.560
[THEME MUSIC FADES]
00:01:24.560 --> 00:01:30.560
The book passage I read at the top is from
Chapter 7, “The Ultimate Paperwork Shredder.”
00:01:30.560 --> 00:01:34.800
Public health officials and healthcare system
leaders influence the well-being and health
00:01:34.800 --> 00:01:41.040
of people at the population level. They help
shape people’s perceptions and responses to
00:01:41.040 --> 00:01:46.960
public health emergencies, as well as to chronic
disease. They help determine the type, quality,
00:01:46.960 --> 00:01:52.240
and availability of treatment. All this is
critical for maintaining good public health,
00:01:52.240 --> 00:01:57.440
as well as aligning better health and financial
outcomes. That, of course, is the main goal of
00:01:57.440 --> 00:02:04.480
the concept of value-based care. AI can definitely
have significant ramifications for achieving this.
00:02:04.480 --> 00:02:08.320
Joining us today to talk about how leaders
in public health and healthcare systems are
00:02:08.320 --> 00:02:16.480
thinking about and acting on this new generation
of AI is Dr. Umair Shah and Dr. Gianrico Farrugia.
00:02:16.480 --> 00:02:21.600
Dr. Umair Shah is a nationally recognized
health leader and innovator. He led one
00:02:21.600 --> 00:02:26.960
of America’s top-rated pandemic responses
as Washington State’s secretary of health,
00:02:26.960 --> 00:02:33.120
a position he held from 2020 to 2025. Umair
previously directed Harris County Public
00:02:33.120 --> 00:02:38.640
Health in Texas, overseeing large-scale emergency
response for the nation’s third-largest county,
00:02:38.640 --> 00:02:44.000
while building an emergency-care career
spanning 20-plus years. He now advises
00:02:44.000 --> 00:02:50.000
organizations on health innovation and strategy
as founder and principal of Rickshaw Health.
00:02:50.000 --> 00:02:55.840
Dr. Gianrico Farrugia is the president and
CEO of Mayo Clinic, the world's top-ranked
00:02:55.840 --> 00:03:00.480
hospital for seven consecutive years,
and a pioneer in technology-forward,
00:03:00.480 --> 00:03:06.640
platform-based healthcare. Under his leadership,
Mayo has built and deployed the Mayo Clinic
00:03:06.640 --> 00:03:11.920
Platform. The platform enables Mayo and its
partners to gain practical insights from a
00:03:11.920 --> 00:03:16.880
comprehensive repository of longitudinal
de-identified clinical data spanning four
00:03:16.880 --> 00:03:23.120
continents. Gianrico is also a Mayo Clinic
physician and professor and an author.
00:03:23.120 --> 00:03:28.320
Umair and Gianrico are CEO-level leaders
representing some of the best of the worlds
00:03:28.320 --> 00:03:33.181
of public health, healthcare delivery,
medical research, and medical education.
00:03:33.181 --> 00:03:43.520
[TRANSITION MUSIC]
Here is my interview with Dr. Umair Shah:
00:03:43.520 --> 00:03:46.708
LEE: Umair, it's really great to have you here.
00:03:46.708 --> 00:03:50.400
UMAIR SHAH: Peter, it's my pleasure. I've
been looking forward to this conversation,
00:03:50.400 --> 00:03:53.906
and I hope you are well today.
00:03:53.906 --> 00:03:56.560
LEE: [LAUGHS] I am doing extremely well.
00:03:57.280 --> 00:04:00.080
So, you know, what I'd like
to do in these conversations
00:04:00.080 --> 00:04:03.120
is first just to start, a little bit about you.
00:04:03.120 --> 00:04:04.480
SHAH: Sure.
00:04:04.480 --> 00:04:10.240
LEE: You served actually during a really
tumultuous time as the secretary of health
00:04:10.240 --> 00:04:16.880
in the State of Washington. But you recently
stepped away from that and you started
00:04:16.880 --> 00:04:21.680
your own firm, Rickshaw Health. So can
we start there? What's that all about?
00:04:21.680 --> 00:04:25.440
SHAH: Yeah, no, absolutely. First
of all, you know, I would say that
00:04:26.000 --> 00:04:29.200
the transition from Texas to
Washington could not have been
00:04:29.200 --> 00:04:33.120
more geopolitically different,
[LAUGHTER] as you can imagine.
00:04:33.120 --> 00:04:34.160
LEE: Sure.
00:04:34.160 --> 00:04:38.640
SHAH: You know, if you like
the red-blue paradigms,
00:04:39.200 --> 00:04:43.504
you couldn’t be more, you know, red
and you couldn’t be more blue, I think.
00:04:43.504 -->
Load full transcript
00:04:43.514
LEE: Yes.
00:04:43.514 --> 00:04:49.760
SHAH: But what happened is, back in November
this past year, as I saw some of the playout
00:04:49.760 --> 00:04:56.400
of continuation of this red-blue dynamic, I
made the decision to step down. And Jan. 15,
00:04:56.400 --> 00:05:00.720
I stepped down, as you mentioned,
and I spent some time really thinking
00:05:00.720 --> 00:05:03.920
about what I wanted to do next and was
looking at a number of opportunities.
00:05:03.920 --> 00:05:09.200
And then a moment in time, there were
some things happening in our—my wife
00:05:09.200 --> 00:05:12.560
and our family's personal lives that sort
of made me think that I wanted to focus
00:05:12.560 --> 00:05:16.960
a little bit more on family. And I felt the
universe was saying, “Stay still.” [LAUGHTER]
00:05:16.960 --> 00:05:22.400
And I launched Rickshaw Health and
the notion that, as you know, Peter,
00:05:22.400 --> 00:05:28.320
rickshaws are oftentimes known across the globe
as these modes of transport that reliably get
00:05:28.320 --> 00:05:34.640
you through ever-changing streets and traffic
patterns and all sorts of ecosystems that are
00:05:34.640 --> 00:05:40.240
evolving at all times. And they get you to the
other side and they get you also with a sense
00:05:40.240 --> 00:05:45.760
of exhilaration. Like when I took my boys to
Karachi, and we were—you know, they jumped in a
00:05:45.760 --> 00:05:49.440
rickshaw and the, you know, open air [LAUGHTER]
and they felt this incredible excitement.
00:05:49.440 --> 00:05:55.680
And so Rickshaw Health was speaking to the three
wheels of a rickshaw that symbolize the three
00:05:55.680 --> 00:06:00.880
children that we have and the real notion of how
do we bring balance and agility and performance
00:06:01.520 --> 00:06:08.560
to the forefront and then move in an ever—just
like streets—ever-changing healthcare environment
00:06:08.560 --> 00:06:13.840
that is constantly evolving, and we too must
evolve with it. And that's what Rickshaw Health
00:06:13.840 --> 00:06:20.240
is all about, is taking clients to that next level
of trying to navigate, especially at this time,
00:06:20.240 --> 00:06:25.800
a very, very different landscape than even
several months ago. So, excited about it.
00:06:25.800 --> 00:06:31.440
LEE: Yeah, absolutely. You know, you
made this transition from Texas to the
00:06:31.440 --> 00:06:35.920
State of Washington. And for people who
listen to this podcast and don't know,
00:06:35.920 --> 00:06:42.640
the particular part of Texas where you
were—Harris County—is really big, very,
00:06:42.640 --> 00:06:50.160
very important in that state. That's just
not, you know, the normal county in Texas.
00:06:50.160 --> 00:06:50.840
SHAH: Yeah. [LAUGHS]
00:06:50.840 --> 00:06:58.960
LEE: It's actually … it's actually known as
quite a forward-looking place, technologically.
00:06:58.960 --> 00:06:59.595
SHAH: That’s right.
00:06:59.595 --> 00:07:06.320
LEE: So what was, you know, the transition like,
then, going from, you know, possibly the most,
00:07:06.320 --> 00:07:14.280
sort of, maybe advanced county in the State of
Texas, a large place, to the State of Washington?
00:07:14.280 --> 00:07:18.640
SHAH: Yeah, you know, Harris County is the
third-largest county in the US. So it had
00:07:19.200 --> 00:07:24.960
close to five million. And now it's probably
… it's exceeded the five million people, and
00:07:24.960 --> 00:07:31.840
a very diverse, very forward-looking, as you
mentioned, technologically very, very much
00:07:31.840 --> 00:07:37.257
looking at what's the next horizon, and home to
Texas Medical Center [TMC] as well, which is …
00:07:37.257 --> 00:07:37.270
LEE: Right.
00:07:37.270 --> 00:07:40.160
SHAH: … the largest medical center. Of
course, it had to be Texas. So it can’t
00:07:40.160 --> 00:07:44.320
be the largest in the state or the country
[LAUGHTER]—the largest in the world, right.
00:07:44.320 --> 00:07:50.640
And TMC also had a number of
different initiatives related to
00:07:50.640 --> 00:07:56.160
startups and venture capital and VC. And so
they had launched something called TMCX. And
00:07:56.160 --> 00:08:00.640
that was a real opportunity—and I know you're
familiar with it—an opportunity to really look
00:08:00.640 --> 00:08:06.400
at how do you incubate all sorts of different
innovations and bringing private sector, public
00:08:06.400 --> 00:08:13.440
sector as well as healthcare delivery alongside
these startups to really look at the landscape.
00:08:13.440 --> 00:08:20.400
And so when I left Houston and came to
Washington, I realized that obviously,
00:08:20.400 --> 00:08:24.320
I was in the backyard … I mean, you know, you
all at Microsoft Research and the work that
00:08:24.320 --> 00:08:29.360
you're all doing is part of an ecosystem of
advanced innovation that's occurring in the
00:08:29.360 --> 00:08:34.080
Pacific Northwest that, you know, when we
see all the players that are here, all the,
00:08:34.080 --> 00:08:38.240
you know, ones that do so many different
things, but they're doing them with an
00:08:38.240 --> 00:08:44.720
eye towards technology, advancements,
and adoptions, it's been quite amazing.
00:08:44.720 --> 00:08:48.880
When I made that transition, it was really
about, you know, the vaccines and what was
00:08:48.880 --> 00:08:52.800
happening with, you know, with COVID
and fighting the—you know, remember,
00:08:52.800 --> 00:08:57.360
this was the state that had the first case in the
continental United States, had the first outbreak,
00:08:57.360 --> 00:09:01.440
and the first [lab-confirmed] death. And
fast-forward a few years later, we had
00:09:01.440 --> 00:09:08.360
the fifth-lowest death rate in the US. And that
was because we all came together to do so much.
00:09:08.360 --> 00:09:15.760
LEE: Yeah,
well maybe that gets us into a question
that I ask a lot of our guests, which is, you
00:09:15.760 --> 00:09:24.480
know, and maybe let's, since we're on your time
as the secretary of health in Washington State,
00:09:24.480 --> 00:09:31.920
[start] with that job. I ask, how would you
explain to your mother what you do every day?
00:09:33.040 --> 00:09:37.600
SHAH: [LAUGHS] I laugh because that's been
such a fascinating conversation in public
00:09:37.600 --> 00:09:44.862
health because we have oftentimes been—it's
been really hard to describe what that is.
00:09:44.862 --> 00:09:44.874
LEE: Yeah.
00:09:44.874 --> 00:09:50.080
SHAH: And, you know, there are so many
metaphors and, you know, analogies that
00:09:50.880 --> 00:09:56.480
we've used. I've always wondered why we do
not have more television shows or sitcoms
00:09:56.480 --> 00:10:03.040
or dramas that are about the public health
workforce or the work that we do in the field,
00:10:03.040 --> 00:10:07.030
because you have, you know, all sorts
of healthcare delivery ones, right.
00:10:07.030 --> 00:10:07.040
LEE: Yup.
00:10:07.040 --> 00:10:14.320
SHAH: As a practicing physician for 20 years,
I realized that people knew what doctors did;
00:10:14.320 --> 00:10:18.080
they knew what nurses did, right. They
intimately touch the healthcare system.
00:10:18.080 --> 00:10:18.180
LEE: Yes.
00:10:18.180 --> 00:10:24.800
SHAH: They understood, you know, that an ambulance
picks you up at your home or somewhere else,
00:10:24.800 --> 00:10:29.440
transports you … gets you to the emergency
department. The emergency department, they do some
00:10:29.440 --> 00:10:34.960
things to you or within the four walls of that
ER, and then you're either admitted, sent home,
00:10:34.960 --> 00:10:41.680
and several days, weeks, whatever later, you
get home if you're admitted, and you start your,
00:10:41.680 --> 00:10:49.360
you know, post-hospital stay at home or your rehab
or what have you. And that all is known to people.
00:10:49.360 --> 00:10:52.240
But when you ask your mother, your
grandmother, or your, you know,
00:10:52.880 --> 00:10:56.560
your uncle, or your brother, your neighbor,
your coworker about what is public health,
00:10:56.560 --> 00:11:03.262
they have a very quizzical look
on their face of what that is.
00:11:03.262 --> 00:11:04.715
LEE: Right.
SHAH: And so what I've …
00:11:04.715 --> 00:11:09.600
LEE: You know, just one thing I've learned
is: it's not just all the people you
00:11:09.600 --> 00:11:14.056
mentioned. Even healthcare professionals
sometimes have that quizzical look.
00:11:14.056 --> 00:11:17.520
SHAH: Yeah, good point. That's right. Good point.
And a lot of it is because we don't get exposed
00:11:17.520 --> 00:11:22.400
to it or trained in it. You know, we think about
public health when we're in our training. And,
00:11:22.400 --> 00:11:26.640
you know, I'm sure you had a very similar piece
of this is that, you know, you see it as, oh,
00:11:26.640 --> 00:11:32.720
that's the health department that takes care of,
you know, STDs, or it takes care, you know, it
00:11:32.720 --> 00:11:38.940
does the immunizations, or, you know, maybe they
do some water quality, or maybe they do mosquitoes
00:11:38.940 --> 00:11:44.000
[mosquito control], and things like that. But the
reality is, we do all of those things and more.
00:11:45.440 --> 00:11:51.920
So my metaphor has been that we are
the offensive line of a football team,
00:11:51.920 --> 00:11:57.760
and the healthcare delivery is the
quarterback. So everybody focuses on,
00:11:57.760 --> 00:12:01.654
you know, from a few years back,
everybody knows Tom Brady, right.
00:12:01.654 --> 00:12:01.674
LEE: Yeah. [LAUGHS]
00:12:01.674 --> 00:12:05.760
SHAH: He won the Super Bowls, everybody
knows what … but if you asked people
00:12:05.760 --> 00:12:11.497
who was number 75 on the offensive
line of the New England Patriots …
00:12:11.497 --> 00:12:11.510
LEE: Right.
00:12:11.510 --> 00:12:14.880
SHAH: … or name your favorite football
team. And the answer would be: you would
00:12:14.880 --> 00:12:20.800
not be able to likely answer that question.
You would know Tom Brady, the quarterback,
00:12:20.800 --> 00:12:26.000
and that's healthcare delivery, the ER doc or
the hospitalist or the nurse or the, you know,
00:12:26.000 --> 00:12:30.640
the medical assistant, or the people that
are doing all the work in the field that
00:12:30.640 --> 00:12:36.080
are the ones that are more visible, but the
invisible workforce of the offensive line,
00:12:36.080 --> 00:12:40.240
that's who we don't know. And yet these are
the people that are blocking and sweating
00:12:40.240 --> 00:12:45.680
and doing all things to complement the work
and make sure the quarterback is successful.
00:12:45.680 --> 00:12:52.400
And here's where the metaphor breaks down, that
when Tom Brady wins the Super Bowl, we continue to
00:12:52.400 --> 00:12:57.840
invest in the offensive line because we recognize
the value of it and we want the quarterback to
00:12:57.840 --> 00:13:04.000
be successful the next season. But in public
health or in society, we do the exact opposite.
00:13:04.000 --> 00:13:07.280
When tuberculosis rates come down, we say, well,
00:13:07.280 --> 00:13:10.062
you know what? We've solved the
problem; we don't need it anymore.
00:13:10.062 --> 00:13:10.074
LEE: Right.
00:13:10.074 --> 00:13:14.240
SHAH: Or you have another, you know,
environmental issue that's no longer there,
00:13:14.240 --> 00:13:18.160
you say, “We don't need it anymore.”
And we disinvest from public health
00:13:18.160 --> 00:13:22.080
or that offensive line. And then you
start to see those rates go back up.
00:13:22.080 --> 00:13:29.440
And so my answer to Mom and Grandma and
Dad and Grandpa is we are critical to
00:13:29.440 --> 00:13:35.120
your health
because we touch you every
single day. And so please invest in us.
00:13:35.120 --> 00:13:40.000
LEE: Yeah. And, you know, I think I'm
going to want to get a little deeper on
00:13:40.000 --> 00:13:47.520
that in just a few minutes here, because, I think
especially during the pandemic, that issue of not
00:13:47.520 --> 00:13:52.720
understanding the importance of that offensive
lineman actually really came to the forefront.
00:13:52.720 --> 00:13:57.040
And so I'd like to get into that.
But the, kind of, second, kind of,
00:13:57.040 --> 00:14:04.080
standard thing I've been probing with people is
still just focusing on you and your background is
00:14:04.880 --> 00:14:10.320
what touchpoints or experiences
you've had with AI in the past.
00:14:10.320 --> 00:14:18.560
And not everyone has. Like, it maybe isn't too
surprising that doctors and healthcare developers,
00:14:18.560 --> 00:14:24.880
tech developers, have lots of contact
with AI, but would the top dog, you know,
00:14:24.880 --> 00:14:32.440
at a public health agency ever have had
significant contact with AI? What about you?
00:14:32.440 --> 00:14:38.000
SHAH: You know, it's interesting. Several years
ago, I was in the audience with the [then] FEMA
00:14:38.000 --> 00:14:43.680
director, [Rich Serino], who just did such an
incredible job. And I remember he made this
00:14:43.680 --> 00:14:48.640
comment at that time. And, Peter, this may have
been like … I don’t know—I'm dating myself—10,
00:14:48.640 --> 00:14:54.800
15, maybe even 20 years ago, and he said,
“Everybody in the audience, there's this,
00:14:54.800 --> 00:14:55.920
you know, app called Twitter.” And, you know,
00:15:01.760 --> 00:15:06.960
“How many people in the audience have ever sent
a tweet or know about this?” And I don't know,
00:15:06.960 --> 00:15:11.360
maybe—it was a public health audience—maybe
about 15% of the people raised their hands.
00:15:11.360 --> 00:15:14.560
He said, “I challenge you to
right now, pick up your phone,
00:15:14.560 --> 00:15:17.920
download the app, and go ahead
and send a tweet right now.”
00:15:17.920 --> 00:15:21.640
And I remember I sent my first tweet
at that time. And it was so thought
00:15:21.640 --> 00:15:26.320
provoking for me was that he was saying
you need to be engaged in social media,
00:15:26.960 --> 00:15:35.258
but the other 85% of the audience
had not even done that or had …
00:15:35.258 --> 00:15:35.270
LEE: Right.
00:15:35.270 --> 00:15:38.960
SHAH: … even understood the importance of social
media at that time. Or maybe they understood,
00:15:38.960 --> 00:15:42.160
but they had restrictions
on how to utilize, right.
00:15:42.160 --> 00:15:48.880
So that has stayed with me because that's
very much about this revolution of AI
00:15:48.880 --> 00:15:53.040
that I know that public health and
population health practitioners like
00:15:53.040 --> 00:15:57.280
myself who have been in the trenches
and understand the importance of it,
00:15:57.280 --> 00:16:00.640
they really believe in the importance
or think they know the importance.
00:16:00.640 --> 00:16:04.720
But NACCHO, the National Association
of County and City Health Officials,
00:16:04.720 --> 00:16:11.280
had done a survey of local health agencies.
And about two-thirds, if not three-quarters,
00:16:11.280 --> 00:16:19.658
of local health agencies reported that they had
an AI capacity that was low or lower than ideal.
00:16:19.658 --> 00:16:19.680
LEE: Hmm. Yeah. Yeah.
00:16:19.680 --> 00:16:26.880
SHAH: And that is very much where I come
from. When I was in public sector and at the
00:16:26.880 --> 00:16:33.680
state health agency, our transformation was
very much about how do we advance the work,
00:16:33.680 --> 00:16:36.800
and how do we utilize this in
a population health standpoint?
00:16:36.800 --> 00:16:42.880
And I was fortunate to have a chief
of innovation at Washington State
00:16:43.520 --> 00:16:48.400
Department of Health, Les Becker,
who understood the value of
00:16:48.960 --> 00:16:58.155
AI. And as you know, we did also
hold a AI science convening that …
00:16:58.155 --> 00:16:58.470
LEE: Yep. Yeah.
00:16:58.470 --> 00:17:02.160
SHAH: … your team was there with
University of Washington. And
00:17:02.160 --> 00:17:06.880
that was really an opportunity for us to
say that AI is here. It's not tomorrow.
00:17:06.880 --> 00:17:11.600
It's not next year. It's not the future.
It's already here. We need to embrace it.
00:17:11.600 --> 00:17:17.600
But here's the problem, Peter, far too few people
in our field understand just how to embrace it.
00:17:17.600 --> 00:17:17.709
LEE: Right.
00:17:17.709 --> 00:17:25.039
SHAH: So I have become a markedly more
champion of AI. One, since I read your
00:17:25.040 --> 00:17:28.960
book. So I think there’s that. So
thank you for writing it. But two,
00:17:29.520 --> 00:17:36.080
since I really recognize that when
I became a solo or a primary-few
00:17:36.080 --> 00:17:43.120
practitioner in my own realm, I needed to
force-amplify the work that I was doing.
00:17:43.120 --> 00:17:47.360
And when I look back, and I continue to stay in
touch with my colleagues in the field of public
00:17:47.360 --> 00:17:51.920
health, what they're also struggling with is
that you have an epidemiologist who's got a
00:17:51.920 --> 00:17:55.840
mound of information—data, statistics,
etc.—that they are going through,
00:17:55.840 --> 00:18:01.579
and they're doing everything in their
power to get that processed and analyzed.
00:18:01.579 --> 00:18:01.594
LEE: Yep. Yep.
00:18:01.594 --> 00:18:08.160
SHAH: AI can take 80% of that and do it.
And that epidemiologist can now turn to
00:18:08.160 --> 00:18:14.220
more of an overseer and a gatekeeper
and to really recognize the patterns …
00:18:14.220 --> 00:18:14.230
LEE: Yep.
00:18:14.230 --> 00:18:19.200
SHAH: … and let AI be able to do the,
you know, grunt work. And similarly,
00:18:19.200 --> 00:18:22.960
as you know, measles—with the outbreaks
that we've seen, especially in Texas but
00:18:22.960 --> 00:18:30.000
elsewhere—you've got an opportunity where our
communications people who are saying, “Look,
00:18:30.000 --> 00:18:35.680
we're about to have, or we know we're about to
announce that there's a measles outbreak in,
00:18:35.680 --> 00:18:39.120
you know, in our community or our
state or what have you—our region.”
00:18:39.120 --> 00:18:46.880
And they can have AI go through different
press briefings and/or press releases and say,
00:18:46.880 --> 00:18:51.823
“Give me the state of the art on how I should
communicate this message to the community.”
00:18:51.823 --> 00:18:51.834
LEE: Hmm.
00:18:51.834 --> 00:18:55.360
SHAH: And bam! You can do that.
And now you can oversee that work,
00:18:55.360 --> 00:19:00.480
as well. And then the third example is that
we are always looking at how do we find ways
00:19:00.480 --> 00:19:05.440
to have a deeper connection with those who
come to our, you know, our websites or come
00:19:05.440 --> 00:19:11.840
to our engagement tools—with bots and things
like that. AI can really accelerate that work,
00:19:11.840 --> 00:19:17.680
as well. So there's so many use cases that AI
has for population health or public health.
00:19:17.680 --> 00:19:17.732
LEE: Yeah.
00:19:17.732 --> 00:19:21.920
SHAH: But I think the challenge is that we just
don't have enough adoption because they're … one,
00:19:21.920 --> 00:19:27.600
we've had funding cuts, but two is
that there is this real hesitation on,
00:19:27.600 --> 00:19:31.680
what is it that we can do? And
I argue—the last thing I'll say
00:19:31.680 --> 00:19:37.760
about this, Peter—is that I argue that AI
is happening right now. The discussions,
00:19:37.760 --> 00:19:42.720
the technology advancements, the work, the
policy work, all that's happening right now.
00:19:42.720 --> 00:19:48.080
If public health practitioners are not at
the table, if they're not part of the …
00:19:48.080 --> 00:19:48.550
LEE: Right.
00:19:48.550 --> 00:19:53.040
SHAH: … “What does this look like? How does it
work in our field?” … guess what? It's going to
00:19:53.040 --> 00:20:00.160
be done to us and for us rather than with us. And
if we do not get with that and get to the table,
00:20:00.160 --> 00:20:05.000
then unfortunately it may not be exactly
what we want it to be at the end of the day.
00:20:05.000 --> 00:20:06.880
LEE: I find it really interesting that
00:20:06.880 --> 00:20:12.074
you are using the terms “public
health” and “population health” …
00:20:12.074 --> 00:20:12.480
SHAH: Yeah.
00:20:12.480 --> 00:20:19.360
LEE: … pretty much interchangeably here. And I
think that that's something that I think touches
00:20:19.360 --> 00:20:25.360
on an assumption that was both implicit
and explicit in the book that we wrote,
00:20:25.360 --> 00:20:29.760
which is: we were making some predictions that
00:20:29.760 --> 00:20:37.360
our ability to extract insights and knowledge
from population health data would be enhanced
00:20:37.360 --> 00:20:43.760
through the use of AI. And I think that
it looks to me like that has been more
00:20:43.760 --> 00:20:49.560
challenging and has come along more slowly
over the past two years. But what is your view?
00:20:49.560 --> 00:20:54.560
SHAH: Yeah, I think part of, and I think
you and I have had this conversation,
00:20:55.120 --> 00:21:01.760
you know, in bits and pieces. I think one of the
real challenges is that when even tech companies,
00:21:01.760 --> 00:21:07.200
and you can name all of them, when they
look at what they're doing in the AI space,
00:21:07.200 --> 00:21:10.440
they gravitate towards healthcare delivery.
00:21:10.440 --> 00:21:11.320
LEE: Yes.
00:21:11.320 --> 00:21:14.880
SHAH: Right? That's, it's …
00:21:14.880 --> 00:21:18.720
LEE: And in fact, it's not even
delivery. I think techies—I did this,
00:21:18.720 --> 00:21:22.520
too—tend to gravitate specifically to diagnosis.
00:21:22.520 --> 00:21:27.520
SHAH: Yes, that's right. That's right. You
know, I think that's a really good point. And,
00:21:27.520 --> 00:21:32.320
you know, when you look at sepsis or you look
at pneumonia or try to figure out ways that,
00:21:32.320 --> 00:21:36.720
you know, radiologists or x-rays or
CT scans can be read, it's, I mean,
00:21:36.720 --> 00:21:42.240
there are so many use cases that are within
the healthcare sector. And I think that gets
00:21:42.240 --> 00:21:49.680
back to this inequity that we have when we look
at population health or, you know, this broad,
00:21:49.680 --> 00:21:56.240
um, swath of land that is, oftentimes, left behind
or unexplored, and you have healthcare delivery.
00:21:56.240 --> 00:22:03.120
Now, healthcare delivery we know gets 95 cents or
96 cents of every dollar. So it makes sense why,
00:22:03.120 --> 00:22:09.520
right. But we also know that, at the end of
the day, we're looking at value-based outcomes,
00:22:09.520 --> 00:22:15.600
and you cannot be successful in the healthcare
delivery system unless we are truly looking at
00:22:15.600 --> 00:22:18.560
prevention and what's happening in
the community and the population.
00:22:18.560 --> 00:22:18.615
LEE: Right.
00:22:18.615 --> 00:22:23.040
SHAH: And that's why I use it interchangeably,
but I know that “public health” has got a very
00:22:23.040 --> 00:22:29.360
specific term, and “population health” is
a different set of ways of looking at the
00:22:29.360 --> 00:22:34.800
world. The reason that people try to shy
away from pop health in essence is that
00:22:34.800 --> 00:22:39.120
you could talk about population health as
being my population of patients in a clinic.
00:22:39.120 --> 00:22:44.000
It could be
my health systems population.
It could be an insurance company saying,
00:22:44.000 --> 00:22:48.160
these are the lives covered, right. So it
becomes, what is population? When we think of
00:22:48.160 --> 00:22:52.640
public health, we think of the entirety of the
population, right. In the State of Washington,
00:22:52.640 --> 00:22:57.600
eight million people. Harris County, five
million people. Or in the US, 300—whatever
00:22:57.600 --> 00:23:02.640
the number of millions of people that—we think
of the entire population. And what is it that
00:23:02.640 --> 00:23:07.680
actually impacts the health and well-being of
that population is really what that's about.
00:23:07.680 --> 00:23:13.840
Yet here's the challenge. When we then talk to
those of our partners and our colleagues in the
00:23:13.840 --> 00:23:19.120
tech field, there are two things happening. One
is, there's a motivation because of the amount of
00:23:19.120 --> 00:23:24.948
dollars that are in [the] healthcare sector. And
number two is, because it's more familiar, right.
00:23:24.948 --> 00:23:24.960
LEE: Right.
00:23:24.960 --> 00:23:30.880
SHAH: And so there are very few practitioners
similar to me that are out there, that are in
00:23:30.880 --> 00:23:34.560
the pop health who kind of know healthcare
delivery because they've also seen patients,
00:23:34.560 --> 00:23:39.040
but they're also—they worked at that federal,
state, local level, community level—they've,
00:23:39.040 --> 00:23:41.920
you know, they've done you know various
different kinds of environments.
00:23:41.920 --> 00:23:47.200
And they say, “Look, I've got a
perspective to really help a tech
00:23:47.200 --> 00:23:52.160
company or somebody see the rest of it,”
but you have to have both partners coming
00:23:52.160 --> 00:23:56.720
together to see that. And I think that's
one of the real challenges that we have.
00:23:56.720 --> 00:23:57.760
LEE: Yeah.
00:23:57.760 --> 00:24:02.613
And so now I'm going to want
to go into specific problems, …
00:24:02.613 --> 00:24:07.280
SHAH: Yeah. Sure.
LEE: … and maybe COVID is a good thing to
00:24:07.280 --> 00:24:12.880
focus on—the breadth of problems that had
to get solved in pandemic response and where
00:24:12.880 --> 00:24:18.080
the gaps between healthcare delivery
and public health were really exposed.
00:24:18.080 --> 00:24:24.960
And so the first problem that I remember
really keenly that just seemed so vexing
00:24:24.960 --> 00:24:32.629
was understanding where the PPE was,
the personal protective equipment ...
00:24:32.629 --> 00:24:33.271
SHAH: Hmm. Yeah.
00:24:33.271 --> 00:24:34.880
LEE: … and where it needed to be.
00:24:34.880 --> 00:24:35.600
SHAH: Yes.
00:24:35.600 --> 00:24:45.440
LEE: And so that turned out … you would
think just getting masks and gowns and
00:24:45.440 --> 00:24:50.800
gloves to the right places at the right times
or even understanding where they are so
that,
00:24:50.800 --> 00:24:56.320
you know … and being able to predict, you
know, what hospitals, what clinics are most
00:24:56.320 --> 00:25:03.360
likely to get a big influx of patients during
the height of the pandemic would be something
00:25:03.360 --> 00:25:10.160
that would be straightforward to solve, but that
turned out to be an extremely difficult problem.
00:25:10.160 --> 00:25:12.480
But how did it look from where you were sitting?
00:25:12.480 --> 00:25:17.640
Because you were sitting at the helm
having to deal with these problems.
00:25:17.640 --> 00:25:23.600
SHAH: Yeah, we were constantly chasing
data and information. And oftentimes,
00:25:23.600 --> 00:25:28.400
you know, because a lot of these
data systems in the public health
00:25:28.400 --> 00:25:33.920
sector have been underinvested in over the
decades, then, you know, you had our biggest
00:25:33.920 --> 00:25:39.760
emergency crisis of our time, and a lot of
public health agencies were either getting,
00:25:39.760 --> 00:25:45.760
you know, thrown a whole host of resources
or had to create things on the fly.
00:25:45.760 --> 00:25:49.040
And whether that was at Harris County
or in the State of Washington, I will
00:25:49.040 --> 00:25:53.200
tell you that what I saw was that,
you know, a lot of agencies across
00:25:53.200 --> 00:25:59.040
the country were still using fax machines,
you know, to get data that were coming in.
00:25:59.040 --> 00:26:05.520
And I remember actually—it's kind of a
funny story—there was a fax machine that
00:26:05.520 --> 00:26:14.880
was highlighted down in our agency in Texas. And
we actually had this fax machine, had mounds of,
00:26:14.880 --> 00:26:20.560
you know, data … sorry, papers that were next to
… faxes that were coming in and all these things.
00:26:20.560 --> 00:26:29.360
And you would have, you know, Mr. Peter
Lee listed as a patient. And then the next,
00:26:30.080 --> 00:26:36.240
you know, transmission would have Pete Lee. And
then the next transmission would have Peter Lee,
00:26:36.240 --> 00:26:44.058
but instead of L-E-E, it was L-E-A-H or something,
or L-I or something, right. And it was just …
00:26:44.058 --> 00:26:44.070
LEE: Right.
00:26:44.070 --> 00:26:46.480
SHAH: ... or you had a date of
birth missing, or you had, you know,
00:26:46.480 --> 00:26:51.440
an address that was off. And what we realized
is that over time, a lot of the data that were
00:26:51.440 --> 00:26:57.360
coming in were just incomplete data, and
being able to chase that was really hard.
00:26:57.360 --> 00:27:04.320
And so, you know, I think AI has
that potential to really organize it,
00:27:04.320 --> 00:27:13.280
and to stratify it, and to especially get
you to a point of at least cleaning it up.
00:27:13.280 --> 00:27:22.294
So I don't think it's just that AI … AI doesn't
just save time; it saves lives. Truly used …
00:27:22.294 --> 00:27:24.240
LEE: Hmm.
Yeah.
SHAH: … that's, I think, where we're talking here.
00:27:24.240 --> 00:27:27.600
And so when you have PPE and things of that
nature, as you talked about, here in the State
00:27:27.600 --> 00:27:31.920
of Washington or what we were trying to do to get
vaccines out or everything we're doing to try to
00:27:31.920 --> 00:27:37.680
get communication messages to the public. And we
did a fantastic job of that, although not ideal.
00:27:37.680 --> 00:27:40.320
I mean, there are so many things that I
could point to that we could have done
00:27:40.320 --> 00:27:44.640
better—all of us in the field of public
health and healthcare delivery alike.
00:27:44.640 --> 00:27:50.720
I will tell you that the one thing that stays
with me is that if we had those tools then,
00:27:50.720 --> 00:27:58.240
and we had them in place then, and we had
invested in them at that time in advance of,
00:27:58.240 --> 00:28:06.960
I think there was a real opportunity for us to be
able to move ahead and even be better at how we
00:28:07.760 --> 00:28:13.840
affected the health outcomes of the very
populations that we were trying to get to.
00:28:13.840 --> 00:28:22.480
And I think it's [that] AI allows us to shift
from reactive to proactive systems, catching
00:28:22.480 --> 00:28:30.080
health issues before they escalate and allow
us to really communicate with empathy at scale.
00:28:30.080 --> 00:28:30.133
LEE: Right.
00:28:30.133 --> 00:28:35.040
SHAH: And when we can do those things,
whether it's opioids or whether it's,
00:28:35.040 --> 00:28:38.720
you know, something that's happening
related to an infectious disease,
00:28:38.720 --> 00:28:43.680
or, you know, even this, the new agenda with
Make America Healthy Again—which by the way,
00:28:43.680 --> 00:28:46.990
as you know, we had a Be Well,
WA ... Be Well, Washington …
00:28:46.990 --> 00:28:47.120
LEE: Right. Yes.
00:28:47.120 --> 00:28:50.160
SHAH: … very much that was about,
you know, looking at, you know,
00:28:50.160 --> 00:28:52.640
physical health and nutritional
health and emotional well-being
00:28:52.640 --> 00:28:58.800
and social connectedness—that there is a real
opportunity for us to address the very drivers of
00:28:58.800 --> 00:29:03.920
ill health. And when we can do that, and
AI can help us accelerate that, I think we
00:29:03.920 --> 00:29:10.560
truly have the ability to drive down costs and
increase the value that's returned to all of us.
00:29:10.560 --> 00:29:18.800
LEE: What is your assessment of public health
agencies’ readiness to use technology like
00:29:18.800 --> 00:29:23.200
AI? Because if there's one thing AI
is good at, it's predicting things.
00:29:23.200 --> 00:29:27.360
Are they [public health agencies] in a
better position to predict things now?
00:29:27.360 --> 00:29:29.600
SHAH: You know, I think it's a tale of two cities.
00:29:29.600 --> 00:29:34.800
I think on the one hand, we're better because
we have the tools. On the other hand, we've
00:29:34.800 --> 00:29:41.040
lost the capacity to be able to utilize those
tools. So, you know, it's a plus and a minus.
00:29:41.840 --> 00:29:46.000
Many, many years ago, there was the
buzzword of what we called syndromic
00:29:46.000 --> 00:29:48.480
surveillance. And, Peter, you know this term well.
00:29:48.480 --> 00:29:54.640
It was like you would have, you know, a
whole host of accumulation of data points in,
00:29:54.640 --> 00:29:56.694
let's say, a hospital setting
or an emergency department …
00:29:56.694 --> 00:29:56.710
LEE: Yup. Yup.
00:29:56.710 --> 00:30:02.800
SHAH: … where, you know, you’d have runny nose,
you'd have cough, you'd have a fever, and you
00:30:02.800 --> 00:30:06.080
would take that, what was happening and people
presenting to the emergency department, with
00:30:06.080 --> 00:30:11.020
what was happening in the area pharmacies where
people were going to get Kleenexes and tissues …
00:30:11.020 --> 00:30:13.040
LEE: Yep.
SHAH: … and buying over-the-counter, you know,
00:30:13.760 --> 00:30:16.480
medication, and things of
that nature, Tylenol, etc.
00:30:17.200 --> 00:30:20.640
And you would say … you would put those two
things together, and you would come up with
00:30:20.640 --> 00:30:25.760
a quote-unquote “syndrome,” and you would say
our ability to say there was an alert to that
00:30:25.760 --> 00:30:32.160
syndrome allows us to say something uh-oh is
going on in the community, and we got many, many
00:30:32.160 --> 00:30:38.133
advancements related to wastewater surveillance
over the last several years as you know …
00:30:38.133 --> 00:30:41.520
LEE: Yep. Yep. Well, also, wasn't patient
number one in the United States discovered
00:30:41.520 --> 00:30:45.840
also because of the Seattle Flu Study, or at
least that sort of syndromic surveillance.
00:30:45.840 --> 00:30:46.551
SHAH: That's right.
00:30:46.551 --> 00:30:48.400
LEE: They weren't even looking for COVID. They
00:30:48.400 --> 00:30:49.500
were just taking, you know,
snot samples from people.
00:30:49.500 --> 00:30:50.480
SHAH: That's right. That's right. That's right.
00:30:52.640 --> 00:30:56.000
And so that's the kind of
thing that you, you know,
00:30:56.000 --> 00:31:06.560
we underappreciate. Is you have to have a
smart, intelligent, agile practitioner, right.
00:31:06.560 --> 00:31:15.760
So if I think about down in Dallas when Ebola was,
you know, the gentleman who was, you know, the
00:31:15.760 --> 00:31:22.560
index case for Ebola was sent out of the emergency
department and came back several days later.
00:31:22.560 --> 00:31:26.320
And it was the nurse who picked up
this time because the practitioner,
00:31:26.320 --> 00:31:30.240
the provider, the healthcare provider,
the doc missed it. And I wouldn't want to
00:31:30.240 --> 00:31:34.080
say in a negative way. It was just, like,
not obvious. You aren't
thinking of Ebola
00:31:34.080 --> 00:31:39.360
in the middle of Texas. And it was the nurse
who picked up: there's something wrong here.
00:31:39.360 --> 00:31:44.058
And what AI has the ability to do
is to pick up those symptoms ...
00:31:44.058 --> 00:31:44.070
LEE: Yeah.
00:31:44.070 --> 00:31:47.440
SHAH: … or those patterns and be
able to recognize the importance
00:31:47.440 --> 00:31:53.600
of those and be able to then alert
the practitioner. So what I … we
00:31:53.600 --> 00:31:59.510
call it artificial intelligence—it
almost becomes artificial wisdom.
00:31:59.510 --> 00:32:04.640
LEE: Hmm. Yeah, interesting. So that actually
reminds me of my next question, which is another
00:32:04.640 --> 00:32:10.320
thing that I watched you and public health
officials do is try to play “what if” games.
00:32:10.320 --> 00:32:15.200
So, for example, I think one decision
you were involved in had to do with,
00:32:15.840 --> 00:32:21.920
you know, what would be the impact if
we put a ban on large gatherings like
00:32:21.920 --> 00:32:31.680
concerts or movie theaters or imposed an 8 PM
curfew on restaurants, and you were trying to
00:32:31.680 --> 00:32:38.080
play “what if” games. Like, what would be the
impact on the spread of the pandemic there?
00:32:38.960 --> 00:32:43.920
So now, again, today with AI, would that aspect of
00:32:43.920 --> 00:32:47.800
what you did play out differently
than it did during the pandemic?
00:32:47.800 --> 00:32:57.360
SHAH: As you know, COVID was the most studied
condition on the planet at one point. And it was,
00:32:57.360 --> 00:33:01.200
you know, things that usually we
would learn over years or months,
00:33:01.200 --> 00:33:04.000
we were learning in weeks or days or hours.
00:33:04.000 --> 00:33:07.280
And I remember in Houston, I would
say something in the morning,
00:33:07.280 --> 00:33:10.960
and I would always try to give the caveat,
“This is the best information we know right
00:33:10.960 --> 00:33:14.960
now,” because it kept changing, whether it
was around masks or whether it was around,
00:33:14.960 --> 00:33:17.520
you know, the way the virus was
operating, whether it was around ...
00:33:17.520 --> 00:33:23.040
I remember even … I was just watching
something recently where I was asked
00:33:23.040 --> 00:33:27.440
to comment about whether spiders
could transmit COVID-19. You know,
00:33:27.440 --> 00:33:31.920
just questions that were just evolving,
evolving, evolving. And the information
00:33:31.920 --> 00:33:35.600
was evolving. By morning, you would say
something. By evening, it would change.
00:33:35.600 --> 00:33:41.360
And why I say that is that it would have been
great in the pandemic if we could have said,
00:33:41.360 --> 00:33:45.600
if you could give us all the information that's
happening across the globe, synthesize that
00:33:45.600 --> 00:33:52.240
information, and be able to help us forecast the
right decisions that we should be making and help
00:33:52.240 --> 00:33:58.400
us model that information so we could decide: if
you did a curfew, or if you did, you know, a mask,
00:33:58.400 --> 00:34:04.703
or if you could, you know, change something else
related to policy—what are the impacts of it?
00:34:04.703 --> 00:34:06.800
LEE: Yeah.
SHAH: What we found constantly in public health
00:34:06.800 --> 00:34:13.040
was that we were weighing decisions in
incomplete data, incomplete information.
00:34:13.040 --> 00:34:17.360
So great now that everybody can armchair
quarterback looking back three, five years
00:34:17.360 --> 00:34:21.040
ago and say, “I would have done it this way,”
or “I would have done it that way.” Gosh,
00:34:21.040 --> 00:34:25.600
I would have as well. But guess what—we
didn't have that information at that time.
00:34:25.600 --> 00:34:30.400
And so you had to make the best
decisions you could with incomplete data.
00:34:30.400 --> 00:34:36.000
But what AI has the potential
to do is to help complete the
00:34:36.000 --> 00:34:38.947
incomplete data. Now, it's not going to get 100%.
00:34:38.947 --> 00:34:38.960
LEE: Right.
00:34:38.960 --> 00:34:43.040
SHAH: And I think, Peter, you know, the one
thing we've got to be really mindful [of] is
00:34:43.040 --> 00:34:46.240
phantom information, or information
where it sort of makes up things,
00:34:46.240 --> 00:34:51.280
or may somehow get you incomplete
information, or skews it a certain way.
00:34:51.920 --> 00:34:55.501
This is why we can't take
the person out of it yet.
00:34:55.501 --> 00:34:57.200
LEE: Right.
SHAH: Now, maybe one day we can.
00:34:57.200 --> 00:35:00.880
I'm not one of those Pollyanna-ish that
people will never be replaced. I actually
00:35:00.880 --> 00:35:06.560
believe that those people who are skilled
with AI and the tools will eventually have
00:35:06.560 --> 00:35:13.440
a competitive advantage over those who
are not. Just like if I had a physician
00:35:13.440 --> 00:35:18.480
who knows how to use their smartphone or
knows how to use a word processor or knows
00:35:18.480 --> 00:35:23.733
how to do a PowerPoint presentation is going
to replace the ones that use scantrons ...
00:35:23.733 --> 00:35:23.750
LEE: Yeah. Yeah.
00:35:23.750 --> 00:35:28.640
SHAH: … or the ones that write it on pieces of
paper—that eventually it makes it more efficient
00:35:28.640 --> 00:35:34.560
and effective, but we're not there yet. But I
think that the potential is absolutely there.
00:35:34.560 --> 00:35:40.560
LEE: So I have one more question. And you
can, kind of, tell I'm trying to expand
00:35:40.560 --> 00:35:45.120
people's understanding of just the incredible
breadth of what goes on in public health,
00:35:45.120 --> 00:35:47.440
you know, all of these sorts of different issues.
00:35:48.080 --> 00:35:53.680
And again, just sticking to COVID,
but this is a much broader issue.
00:35:53.680 --> 00:36:02.954
Another thing you had to cope with were
significant rise of misinformation …
00:36:02.954 --> 00:36:03.520
SHAH: Yes.
00:36:03.520 --> 00:36:14.080
LEE: … and maybe going along with that, very,
very significant inequities in outcomes in the
00:36:14.080 --> 00:36:20.720
COVID response. And when you think about AI
there, I think you can argue it both ways,
00:36:20.720 --> 00:36:25.840
that it both exacerbates the problem but also
gives you new tools to mitigate the problems.
00:36:25.840 --> 00:36:27.360
What is your view?
00:36:27.360 --> 00:36:31.280
SHAH: I think you … I don't even have to
say it … I think you hit on it, is that,
00:36:31.280 --> 00:36:34.880
you know, it really is two sides of one coin.
00:36:34.880 --> 00:36:39.200
On the one hand, it has the power of really
advancing and allowing us to move forward in
00:36:39.200 --> 00:36:45.040
a way that incredibly accelerates and
accentuates, but on the other hand,
00:36:45.040 --> 00:36:50.560
in the case of inequities, right? So if you
have inequitable information data that's
00:36:50.560 --> 00:36:54.880
already out in the literature or already out
in the, you know, media, or what have you,
00:36:54.880 --> 00:37:01.440
about a certain population or people or certain
kinds of ideas or thoughts, etc., then AI will
00:37:01.440 --> 00:37:05.920
tend to accumulate that. You're going to take that
information, thinking that's the best out there,
00:37:05.920 --> 00:37:13.920
but it may have missed out on information and now
you go with it. And that's a potential problem.
00:37:13.920 --> 00:37:20.400
And I think it's the same thing on information
is that when we have people that are able to
00:37:21.360 --> 00:37:28.080
classify or misclassify information, I think it
really becomes hard because it can accelerate
00:37:28.080 --> 00:37:35.920
the inequities of trust or inequities of trusted
sources of information. It can also close the gap.
00:37:35.920 --> 00:37:40.320
So I think, you know, it's really
up to us and this responsible AI
00:37:40.320 --> 00:37:44.320
to really think about how we can go about
doing this in a way that's going to allow
00:37:44.320 --> 00:37:50.160
us to further the advancements but also be
careful of those, you know, those kind of
00:37:50.160 --> 00:37:56.640
places where we're going to step into that are
not going to be well received or successful.
00:37:56.640 --> 00:37:59.440
You know, the one thing that's
really fascinating about this
00:37:59.440 --> 00:38:03.640
whole conversation is that this is why
we've got to be at the table, Peter.
00:38:03.640 --> 00:38:04.545
LEE: Yep. Yep.
00:38:04.545 --> 00:38:08.720
SHAH: Because if we're not at the
table, you know, what's the, you know,
00:38:08.720 --> 00:38:15.040
or if tech companies that are out there doing
this work and aren't even seeing a field of
00:38:15.040 --> 00:38:20.480
practitioners that are actually wrestling with
the same problems but just cannot actually get
00:38:20.480 --> 00:38:26.000
to the solutions, we're just going to
continue to accentuate the problems.
00:38:26.000 --> 00:38:30.720
And that's why I'm a firm proponent
of: we've got to be at the table.
00:38:30.720 --> 00:38:35.680
And so even when we've seen in, and this
is going to be a little controversial,
00:38:35.680 --> 00:38:40.720
but governmental spaces where, you
know, policymakers have said, “Look,
00:38:42.640 --> 00:38:47.920
we are not going to let you do certain
things,” or they say to public health
00:38:47.920 --> 00:38:52.960
practitioners or even healthcare delivery
practitioners in certain spaces, “You cannot
00:38:53.520 --> 00:38:57.280
even play with this. You cannot have it
on your phones. You can't do any ... ”
00:38:57.280 --> 00:39:02.960
You know, what I really believe it does is that
it takes [an] almost like we put our head in the
00:39:02.960 --> 00:39:07.840
sand type of approach rather than saying,
“What is it that we can do to help improve
00:39:07.840 --> 00:39:13.200
AI and make it work for all of us?” What
we're doing is we're essentially saying,
00:39:13.200 --> 00:39:20.080
“We're going to let the tech companies and all
the other developers come up with the solutions,
00:39:20.080 --> 00:39:23.520
but it's not going to be informed by
the people in the field.” And that's
00:39:23.520 --> 00:39:27.840
dangerous. We have to do both.
We have to be working together.
00:39:27.840 --> 00:39:33.280
LEE: Umair, that's really so well said,
and I think a great way to wrap things
00:39:33.280 --> 00:39:37.200
up. I've certainly learned a lot from
this conversation. So thank you again.
00:39:37.200 --> 00:39:41.680
SHAH: It's been a pleasure to be with you this
morning. Thank you so much for the time. And
00:39:41.680 --> 00:39:51.898
I'm looking forward to further conversations.
00:39:51.898 --> 00:39:51.920
[TRANSITION MUSIC]
00:39:51.920 --> 00:39:57.520
I live in the State of Washington and because of
that, I've been able to watch Umair in action as
00:39:57.520 --> 00:40:03.520
our state's former secretary of health. And
some of that action was pretty intense to
00:40:03.520 --> 00:40:10.480
say the least because his tenure as secretary of
health spanned the period of the COVID pandemic.
00:40:10.480 --> 00:40:15.200
Now, as a dyed-in-the-wool techie, I
have to admit that at the beginning,
00:40:15.200 --> 00:40:20.880
I don't think I really understood the scope
and importance of the field of public health.
00:40:20.880 --> 00:40:27.680
But as the conversation with Umair showed, it's
really important and it is arguably both an
00:40:27.680 --> 00:40:31.760
underfunded and underappreciated
part of our healthcare system.
00:40:31.760 --> 00:40:36.800
Now, public health is also very much
an area that's ripe for advancement
00:40:36.800
--> 00:40:42.480
and transformation through AI. As Umair
explained in our discussion, the core of
00:40:42.480 --> 00:40:47.920
public health is the idea of population health,
the idea of extracting new health insights from
00:40:47.920 --> 00:40:55.520
signals from population-scale data. And already
we're starting to see AI making a difference.
00:40:55.520 --> 00:41:00.320
Now here's my interview
with Dr. Gianrico Farrugia.
00:41:06.880 --> 00:41:09.661
LEE: Gianrico, it's really
great to have you here today.
00:41:09.661 --> 00:41:13.240
GIANRICO FARRUGIA: Peter, thanks for having
me. Thanks for making me part of your podcast.
00:41:13.240 --> 00:41:18.880
LEE: You know, what I'd like to do
in these conversations is, you know,
00:41:18.880 --> 00:41:22.560
we'll definitely want to talk about the overall
healthcare system, the state of healthcare,
00:41:22.560 --> 00:41:29.520
and what AI could or might do to help or even
hurt all of that. But I always like to start
00:41:30.720 --> 00:41:37.840
with a sharper focus just on you specifically.
And my first question always is, you know,
00:41:37.840 --> 00:41:46.880
I think people imagine what a hospital or
a health system president and CEO does,
00:41:46.880 --> 00:41:53.720
but not really. And so how would you explain
to your mother what you do every day?
00:41:53.720 --> 00:41:58.080
FARRUGIA: So, Peter, my mother’s
88 years old. She lives in Malta,
00:41:58.080 --> 00:42:00.131
and she’s visiting at the moment, …
00:42:00.131 --> 00:42:01.680
LEE: Oh, wow.
FARRUGIA: … which is kind of nice, really.
00:42:01.680 --> 00:42:03.240
LEE: Wow, that is amazing.
00:42:03.240 --> 00:42:08.640
FARRUGIA: I'm proud that she's still proud of
me. So she does ask. I'll tell her the scope
00:42:08.640 --> 00:42:14.720
of Mayo Clinic. We serve patients across the
globe. We have about 83,000 staff members that
00:42:14.720 --> 00:42:19.760
work with us, and we're very proud of the work
we do in research, education, and the practice.
00:42:19.760 --> 00:42:26.400
Mayo Clinic is built to serve people with serious
disease. So what I tell my mother is that here we
00:42:26.400 --> 00:42:31.360
are. We're a healthcare organization that knows
what it needs to do: keep patients as the North
00:42:31.360 --> 00:42:37.120
Star. The needs of the patient come first.
We have 83,000 people who want to do that,
00:42:37.120 --> 00:42:45.040
several thousand physicians and scientists. My
job is to look slightly ahead and then share
00:42:45.040 --> 00:42:50.960
what I'm seeing and then, sort of, smooth the way
for others to make sure Mayo remains true to its
00:42:50.960 --> 00:42:55.200
mission but also true to the fact that at the
moment, we are in a category of one. We need to
00:42:55.200 --> 00:43:00.880
remain there not just from an ego standpoint, but
really from a “do good to the world” standpoint.
00:43:00.880 --> 00:43:05.680
At that point, invariably my mother will tell
me that I'm working too hard. [LAUGHTER] And
00:43:05.680 --> 00:43:10.800
then of course, I change the subject, and I ask
her what she cooked today because my mother,
00:43:10.800 --> 00:43:16.560
who’s 88, cooks for the whole family in Malta,
and there are usually four generations eating
00:43:16.560 --> 00:43:22.720
around the table. So I tell her what she does
for the family is what I do for the Mayo family.
00:43:22.720 --> 00:43:28.240
LEE: Wow, that's a great way to put it.
And it sounds like you actually have a
00:43:28.240 --> 00:43:33.000
good chance to have some good genes
if she's still that active at age 88.
00:43:33.000 --> 00:43:37.040
FARRUGIA: I think I chose a little
more stressful job that may limit
00:43:37.040 --> 00:43:42.880
[that]. I will tell you very briefly is that one
of the AI algorithms we have estimates biological
00:43:42.880 --> 00:43:49.106
age from an electrocardiogram. My biological
age jumped by 3.7 years when I became CEO.
00:43:49.106 --> 00:43:49.480
LEE: [LAUGHS] Oh no.
00:43:49.480 --> 00:43:52.800
FARRUGIA: I'm hoping it will
reverse on the other side.
00:43:52.800 --> 00:44:01.440
LEE: To stick with you just for one more moment
here, second question I ask is about your origin
00:44:01.440 --> 00:44:09.280
story with respect to AI. And typically, for
most people, there is AI before ChatGPT and
00:44:09.280 --> 00:44:18.960
generative AI and then after the generative AI
revolution. So can you share a little bit about
00:44:19.680 --> 00:44:24.720
this? Because it must be the case that you've
been thinking about this a long time since
00:44:24.720 --> 00:44:29.560
you've really led Mayo Clinic to
be so tech forward in this way.
00:44:29.560 --> 00:44:34.560
FARRUGIA: Well, I've been, as you said,
a physician for way too long. I got my MD
00:44:34.560 --> 00:44:41.040
degree in ’87. So that sort of dates me.
But it also means that I saw a lot of the
00:44:41.040 --> 00:44:45.680
promise for AI that never seemed to pan out for
decades and decades and decades like you did.
00:44:45.680 --> 00:44:46.280
LEE: Yeah.
00:44:46.280 --> 00:44:52.320
FARRUGIA: Around 10 years ago, Mayo could sense
that there was something different, that something
00:44:52.320 --> 00:44:57.520
was changing, that we actually—at that time,
predictive AI—could make a big difference. And I
00:44:57.520 --> 00:45:04.000
think that's the moment where I and others jumped
in and said Mayo Clinic needs to be involved.
00:45:04.000 --> 00:45:09.040
And then about six years ago, when—six and
a half years ago—I became CEO, it was clear
00:45:09.040 --> 00:45:14.960
that there was the right confluence
of data, knowledge, tech expertise,
00:45:14.960 --> 00:45:21.520
that we could deal with what was increasingly
bothering me, which is that we knew what was
00:45:21.520 --> 00:45:27.760
coming from a technology standpoint and
we
knew the current healthcare system could not
00:45:27.760 --> 00:45:32.560
deliver on what patients need and want within
that current system. And so the answer is,
00:45:32.560 --> 00:45:38.720
how could a place like Mayo Clinic with our
reputation not jump in and say there has to
00:45:38.720 --> 00:45:43.440
be a better way of doing things? I've always said
that it is impossible for me to understand that
00:45:43.440 --> 00:45:48.539
every single government employee is incompetent.
Every physician is greedy. Something's wrong here.
00:45:48.539 --> 00:45:48.550
LEE: Yeah.
00:45:48.550 --> 00:45:51.600
FARRUGIA: And that wrong was
the architecture was wrong. And
00:45:51.600 --> 00:45:54.720
we knew that we could incorporate
AI and make it better. So for me,
00:45:54.720 --> 00:46:00.720
that journey was one of wait, wait, wait. 10
years ago, begin to jump in. Six years ago,
00:46:00.720 --> 00:46:05.840
really jump in with our platform. And then, of
course, in November 2022, things changed again.
00:46:05.840 --> 00:46:11.680
LEE: Yeah. When did this idea of a data platform,
00:46:11.680 --> 00:46:15.549
what you now call the Mayo Clinic Platform—by
the way, I refer to this as MCP, …
00:46:15.549 --> 00:46:15.582
FARRUGIA: Yeah, I know. [LAUGHS]
00:46:15.582 --> 00:46:18.640
LEE: [LAUGHS] … which I always
smirk a little bit because,
00:46:18.640 --> 00:46:24.000
of course, for those of us in computer
science research, the AI research,
00:46:24.000 --> 00:46:32.960
MCP has also become quite a hot topic
because of the model context protocol
00:46:32.960 --> 00:46:42.000
version of this. But for Mayo's MCP, when did
that become a serious, defined initiative?
00:46:42.000 --> 00:46:49.760
FARRUGIA: So around the end of ’18,
2018, beginning of 2019. At that point,
00:46:49.760 --> 00:46:54.560
we knew that we were going to do something
differently. We came up with a strategic plan,
00:46:55.280 --> 00:46:59.680
as I took on the job, that we needed to cure
more patients. There’s just not enough cures
00:46:59.680 --> 00:47:05.520
in the world. There's too much suffering. And that
we had all these chronic diseases that people have
00:47:05.520 --> 00:47:10.080
accepted are chronic, but really the only reason
that disease is chronic is you haven't cured it.
00:47:10.080 --> 00:47:13.520
And physicians have been afraid
to talk about cure because,
00:47:13.520 --> 00:47:16.779
of course, eventually everybody passes away.
00:47:16.779 --> 00:47:16.790
LEE: Yeah.
00:47:16.790 --> 00:47:23.120
FARRUGIA: But I really pushed hard to say, no,
it's OK to talk about cure. It's OK to aspire
00:47:23.120 --> 00:47:27.520
to cure. The second was connect—connecting
people with data to create new knowledge.
00:47:27.520 --> 00:47:33.280
And that's where it became clear that data
were not currently in a format that were
00:47:33.280 --> 00:47:36.880
particularly useful. By the
way, you'll
hear me talk about data in the singular
00:47:36.880 --> 00:47:42.800
and the plural. I'm old school. I talk about
data as plural, but I know that most younger
00:47:42.800 --> 00:47:46.400
people now use data singular. [LAUGHTER]
And I apologize if I'll go through that.
00:47:46.400 --> 00:47:52.240
And then the third was transform. Let's use
Mayo's resources to transform healthcare for
00:47:52.240 --> 00:47:58.800
ourselves and for others. And that's the concept
of, if we are able to use data in a different way,
00:47:58.800 --> 00:48:03.440
let's create a different architecture. And
that architecture had to be very closely
00:48:03.440 --> 00:48:09.520
linked to using artificial intelligence in
order to create better outcomes for patients.
00:48:09.520 --> 00:48:16.800
So patients can live not only longer lives but
healthier lives. And that's the genesis of MCP,
00:48:16.800 --> 00:48:23.520
Mayo Clinic Platform, so I'll timestamp
that as end of 2018, beginning of 2019.
00:48:23.520 --> 00:48:30.320
LEE: So I'm really wanting to delve in in
this episode, in this conversation, you know,
00:48:30.320 --> 00:48:39.760
[into the] mindset of a health system or hospital
CEO. And so you're obviously thinking about,
00:48:39.760 --> 00:48:50.160
I guess, machine learning and predictive analytics
and so on. What were the, kind of, like … in 2018,
00:48:50.160 --> 00:48:56.000
what were the outcomes that you were dreaming
about from this? So if you had this thing, you
00:48:56.000 --> 00:49:02.720
know, what were the things that you were hoping to
be able to show or, kind of, produce as results?
00:49:02.720 --> 00:49:08.400
FARRUGIA: So first of all, I think all of us who
work at Mayo Clinic, and this tends to be a bit
00:49:08.400 --> 00:49:14.640
sugary, but it's true, strongly feel that we have
a responsibility to leave the place better than
00:49:14.640 --> 00:49:20.560
when we started. And so the Mayo brothers,
when they started, did two really important
00:49:20.560 --> 00:49:26.800
things. The first was that they created the first
integrated healthcare system. And the second,
00:49:26.800 --> 00:49:32.480
they created the first unified record. And that
record was, of course, paper at that point.
00:49:32.480 --> 00:49:38.560
Part of that is to say, OK, what does it look
like now versus how can we improve what we
00:49:38.560 --> 00:49:43.440
have if … it'd be blasphemy to say, let's
think of ourselves as the Mayo brothers,
00:49:43.440 --> 00:49:48.800
but let's think of ourselves as reasonably
smart people at Mayo Clinic, really lucky
00:49:48.800 --> 00:49:53.600
to be surrounded by very smart people with
resources. What will we do? And so we said
00:49:53.600 --> 00:50:00.960
let's not aim for the low-hanging fruit. Let's
aim to get at whatever you want to call it, the
00:50:00.960 --> 00:50:07.600
intractable knot, the hardest problem, and that is
clinical care. Let's improve clinical care. Yes,
00:50:07.600 --> 00:50:14.800
we can deal with burnout. Yes, we can deal with
administrative burden. But let's not focus on
00:50:14.800 --> 00:50:23.200
that. Let's really create an architecture that
allows us to tackle better clinical outcomes.
00:50:23.200 --> 00:50:29.360
And by starting there, then everything
flows from that. That it's not really
00:50:29.360 --> 00:50:34.880
worth doing unless at the end of the day,
people are experiencing better health.
00:50:34.880 --> 00:50:43.040
LEE: And so I know a very good colleague and
friend of mine, John Halamka, you ended up hiring.
00:50:43.040 --> 00:50:48.480
I thought he was a very interesting choice because
he is, of course, in terms of technology, quite
00:50:48.480 --> 00:50:56.320
deep and very expert, but he's, I think, first
and foremost, a doctor. And so I assume you must
00:50:56.320 --> 00:51:03.280
have had to decide what type of person you would
bring in and what kinds of people you would bring
00:51:03.280 --> 00:51:07.720
in to try to create such a thing. What was your
thinking around the choice of someone like John?
00:51:07.720 --> 00:51:16.240
FARRUGIA: It was one of the harder decisions.
First of all, [I'm] a physician myself. We tend
00:51:16.240 --> 00:51:23.760
to want to maintain some control. And so now I
am the CEO, [LAUGHTER] and I have to give this
00:51:23.760 --> 00:51:33.200
baby to somebody else. That's very hard. Second
is Mayo Clinic is really good because it is flat,
00:51:33.200 --> 00:51:38.160
and we run a lot by committee. But it also
means that, therefore, you have to work
00:51:38.160 --> 00:51:44.640
really hard at change, and you cannot change by
fiat. You have to change by convincing people.
00:51:44.640 --> 00:51:50.560
So I just … I've always made the point that
the right change agent is a servant leader
00:51:50.560 --> 00:51:55.680
because that's how change becomes embedded. But
it also means you've got to have that personality,
00:51:55.680 --> 00:51:59.840
the Mayo personality. And it became clear
when we interviewed [that] there were some
00:51:59.840 --> 00:52:09.040
people that were really hardcore tech; others
that were passionate about social issues. But
00:52:09.040 --> 00:52:17.040
John really fit that of being, as you said,
deep in IT but also himself very aligned with
00:52:17.040 --> 00:52:21.280
the Mayo Clinic values. It's as if he was a
Mayo Clinic physician even though he wasn't.
00:52:21.280 --> 00:52:28.880
And that came together, and I felt, we felt,
that as we were hiring, that we could do it.
00:52:28.880 --> 00:52:34.400
And then we did something interesting. We
paired John with a … we created the role of
00:52:34.400 --> 00:52:39.440
a chief medical officer for the platform, which
was a longstanding Mayo Clinic physician. And so
00:52:39.440 --> 00:52:44.160
we brought them together
so we could get the past
and the present and the future working together.
00:52:44.160 --> 00:52:49.680
LEE: So I'm going to ask you about what has
come out of this. But before that, let's get
00:52:49.680 --> 00:52:58.000
back to this origin story. So now, all of that
is being set up starting around 2018. But then,
00:52:58.000 --> 00:53:06.320
you know, in 2022, there is generative AI. Now
you were already experimenting with transformers,
00:53:06.960 --> 00:53:13.120
starting with BERT out of Google there. So maybe
that's a couple of years earlier. But still,
00:53:13.120 --> 00:53:16.880
there has to come a point where
things are feeling very disrupted.
00:53:16.880 --> 00:53:26.480
FARRUGIA: Yeah, so, you know, it really wasn't.
It, to me, was a relief because it gave this … we
00:53:26.480 --> 00:53:34.960
were feeling pretty good about what we're
doing. We were feeling a little impatient,
00:53:34.960 --> 00:53:39.920
but, in true Mayo fashion, were willing
to, sort of, do everything, take its time,
00:53:39.920 --> 00:53:44.000
take it to the right committees, get
the right approvals, and get it done.
00:53:44.000 --> 00:53:50.960
And so when generative AI came, for us, it's like,
I wouldn't say we told you so, but it's like, ah,
00:53:50.960 --> 00:53:54.560
there you go. Here's another tool. This is
what we've been talking about. Now we can do
00:53:54.560 --> 00:54:00.800
it even better. Now we can move even faster.
Now we can do more for our patients. It truly
00:54:00.800 --> 00:54:05.814
never was disruptive. It truly immediately
became enabling, which is strange, right, …
00:54:05.814 --> 00:54:05.826
LEE: Yeah.
00:54:05.826 --> 00:54:11.280
FARRUGIA: … because something as disruptive as
that instantly became enabling at Mayo Clinic. And
00:54:11.280 --> 00:54:19.440
I'll take … as I think about it with you and take
a moment to think and reflect on it, I think there
00:54:19.440 --> 00:54:25.040
were a couple of decisions we made earlier on that
really helped us. We made the decision against the
00:54:25.040 --> 00:54:31.200
advice of any consulting firm to completely
decentralize AI at Mayo Clinic six years ago.
00:54:31.200 --> 00:54:36.640
And we told our clinical department, you need to
own this. You need to hire basic scientists in
00:54:36.640 --> 00:54:41.360
AI. We'll help you by creating the infrastructure.
We'll help you by doing all the rest. We'll have
00:54:41.360 --> 00:54:45.840
the compute. We'll have the partners. You need
to do this on your own. You need to treat this
00:54:45.840 --> 00:54:51.040
the same way as if a new radiological technique
happened or a new surgical technique happened.
00:54:51.040 --> 00:54:57.680
And so there was a lot of expertise already
present in a very diffused way that then we
00:54:57.680 --> 00:55:05.440
were able to layer on generative AI onto that.
And we found a very willingness to embrace it.
00:55:05.440 --> 00:55:08.800
In fact, I would argue initially a
bit too willing because as you know,
00:55:08.800 --> 00:55:13.360
we haven't quite figured out what's legitimate
use, what's not use. We all learned together.
00:55:13.360 --> 00:55:14.560
LEE: Right. Yep. Yep.
00:55:14.560 --> 00:55:20.040
FARRUGIA: But it was mostly energy, which
is really interesting. It was mostly energy.
00:55:20.040 --> 00:55:25.280
LEE: Wow. And, you know, it's
an amazing thing to hear because
00:55:27.040 --> 00:55:33.520
one common theme that we hear is that the initial
reaction is oftentimes one of skepticism. In fact,
00:55:33.520 --> 00:55:37.760
I've been very open that even I
initially had some skepticism.
00:55:38.480 --> 00:55:42.720
Was that not present in your mind or on
your team's mind at all at the beginning?
00:55:42.720 --> 00:55:49.120
FARRUGIA: So you're asking a physician if they
are skeptical about something. [LAUGHTER] Yeah. I
00:55:49.120 --> 00:55:56.800
wonder what the answer to that is. Absolutely. The
first hallucination, the first wrong reference.
00:55:56.800 --> 00:56:00.374
Can you imagine if you write the grant and
the wrong reference comes. As you know, …
00:56:00.374 --> 00:56:00.386
LEE: Right.
00:56:00.386 --> 00:56:06.880
FARRUGIA: … earlier on when some references
were being made up. So massive amounts of
00:56:06.880 --> 00:56:12.000
skepticism. But the energy was such there
that the people [who] were skeptical were
00:56:12.000 --> 00:56:17.120
also at the same time saying, “Let's do
a RAG [retrieval augmented generation] to
00:56:17.120 --> 00:56:22.560
clean up those references. Let's create …” We
were experimenting with discharge summaries,
00:56:22.560 --> 00:56:29.680
but let's use AI to police AI, and let's
see what's going on. So there was more
00:56:29.680 --> 00:56:35.920
massive skepticism, but the energy was
pushing that skepticism into a positive
00:56:35.920 --> 00:56:40.640
versus into a negative frame. Now,
I say that summarizing in hindsight.
00:56:40.640 --> 00:56:41.320
LEE: Yeah.
00:56:41.320 --> 00:56:46.160
FARRUGIA: Day to day, much more
complicated than that. But overall,
00:56:46.160 --> 00:56:50.320
if you just … and remember, I
had been at the World Economic
00:56:50.320 --> 00:56:54.960
Forum many years ago and had said,
healthcare needs to run towards AI.
00:56:54.960 --> 00:56:55.040
LEE: Yes.
00:56:55.040 --> 00:57:00.000
FARRUGIA: If healthcare was perfect, we would
wait. Healthcare is not perfect by any means,
00:57:00.000 --> 00:57:04.960
therefore let's run and embrace AI. And,
sort of, that mentality was part of who we
00:57:04.960 --> 00:57:08.240
were because at the same time, we
were also saying the other thing,
00:57:08.240 --> 00:57:12.720
that we need to be the ones to lead validation.
We need to be the ones that set the rules. We
00:57:12.720 --> 00:57:15.680
need to be participating in the creation
of CHAI [Coalition for Health AI]. We
00:57:15.680 --> 00:57:18.800
need to be participating as the
[National] Academy of Medicine.
00:57:18.800 --> 00:57:25.120
So people did feel that Mayo was being
fairly responsible about it, but that
00:57:25.120 --> 00:57:31.360
urge to, the needs of the patient come first,
was the driver that kept people wanting to say,
00:57:31.360 --> 00:57:37.680
“Not ready yet, but let's make it ready.” And
we now have 320 algorithms in the practice,
00:57:37.680 --> 00:57:45.200
and they run and we constantly are looking and
seeing what else we can do to improve. But as
00:57:45.200 --> 00:57:49.760
you well know, things evolve and change.
And we're also looking and seeing which
00:57:49.760 --> 00:57:53.440
ones work and which ones don't and which ones
we have to work together on to make better.
00:57:53.440 --> 00:57:58.400
LEE: Yeah, you know, of course
Mayo has such a, you know,
00:57:58.400 --> 00:58:03.760
such a reputation and is so influential,
but in the world of healthcare broadly,
00:58:03.760 --> 00:58:12.000
let's just focus on the United States to
start. How common is this experience? You know,
00:58:12.000 --> 00:58:18.240
so if you are at a meeting with fellow
CEOs of hospitals and health systems,
00:58:18.240 --> 00:58:24.480
what is the attitude and what is the, kind of
… how common is the approach to all of this?
00:58:24.480 --> 00:58:29.200
FARRUGIA: I think it's more common now, but
going back a few years, I think it's fair to
00:58:29.200 --> 00:58:37.280
say that it was scary for people to know how
it's going to change things. Healthcare runs
00:58:37.280 --> 00:58:46.480
on very narrow margins. It's very expensive. So
your expenses and your revenue are both massive,
00:58:46.480 --> 00:58:52.800
and they are very close to each other. So anything
that changes that balance is really scary.
00:58:52.800 --> 00:58:58.080
Because it's not like you have the opportunity
to erode into a margin or get it right the
00:58:58.080 --> 00:59:04.000
second time. So I think that is what drove
a lot of the initial hesitancy. Was, one,
00:59:04.000 --> 00:59:10.560
is lack of knowledge and, two, understanding that
you didn't have a lot of room to make a mistake.
00:59:10.560 --> 00:59:19.520
LEE: On the economics of this, when you
are embarking on what I suspect is a very
00:59:19.520 --> 00:59:26.720
expensive initiative like Mayo Clinic Platform,
how on earth do you justify that early on?
00:59:26.720 --> 00:59:32.400
FARRUGIA: So again, I'm trying hard to
try and remember how things were versus
00:59:32.400 --> 00:59:40.160
how I think about them now. [LAUGHTER] It goes
back to our history. Mayo has always invested
00:59:40.160 --> 00:59:45.040
in what it thinks is the right thing that is
coming. And that's how we've stayed where we
00:59:45.040 --> 00:59:53.440
are. So the investment really was having an open
discussion: is this worth it for our patients?
00:59:53.440 --> 00:59:59.280
And once that discussion was over,
then the board was saying, go, go, go.
00:59:59.280 --> 01:00:05.520
Now we are lucky in that we have the
size that we're able to hire and absorb.
01:00:05.520 --> 01:00:11.200
We're lucky in that the people [who] came
before us have been financially astute,
01:00:11.200 --> 01:00:20.160
and one of our values is stewardship. And we're
lucky that we had a lot of patients at Mayo Clinic
01:00:20.160 --> 01:00:29.760
who were able to listen, be inspired by, and be
willing to help support. And so that gave us the
01:00:29.760 --> 01:00:35.600
ability to build what we're doing not only into
the long-range plan but actually into the yearly
01:00:35.600 --> 01:00:40.400
plan. And so we built it into the yearly plan.
We set up a center for digital health. We set
01:00:40.400 --> 01:00:48.080
up the platform. And then we set up the budgets
to be able to do that. And the budgets came from
01:00:48.080 --> 01:00:56.080
assets we've had, assets that we would get as
the year came by, and then from philanthropy.
01:00:56.080 --> 01:01:02.240
We also had a really powerful calling
card. And that's one advantage I had,
01:01:02.240 --> 01:01:07.920
and that's … and I'd been very open when I was
speaking to other CEOs that would use it is that
01:01:07.920 --> 01:01:17.840
right at that very beginning, really, really in
2019, our cardiologists, both the researchers and
01:01:17.840 --> 01:01:25.280
the clinicians, had come together and had used
electrocardiograms to create an AI algorithm.
01:01:25.280 --> 01:01:30.320
The first one was for diagnosing from an
electrocardiogram, which is very cheap, very easy
01:01:30.320 --> 01:01:37.760
to do, left ventricular dysfunction. That's how
hard the left part of the heart contracts. If it
01:01:37.760 --> 01:01:44.480
doesn't do well, you get heart failure. And they
were able to show that that algorithm was already
01:01:44.480 --> 01:01:52.080
making nurses better than the physician without
the algorithm. And after that went on to show
01:01:52.080 --> 01:01:56.960
that you could do it from a single strip, really
with an area under the curve for that single strip
01:01:57.840 --> 01:02:04.139
on a watch, that was as good as mammograms or
pap smears. And so we already had that proof.
01:02:04.139 --> 01:02:04.150
LEE: Yeah.
01:02:04.150 --> 01:02:09.920
FARRUGIA: That quickly then came into Mayo.
We put it into it so that any patient now
01:02:09.920 --> 01:02:14.720
can benefit from it. And now there are, I
think, 14 algorithms just from that same one.
01:02:14.720 --> 01:02:14.851
LEE: Yeah.
01:02:14.851 --> 01:02:20.880
FARRUGIA: So we had a proof of concept thanks
to those really far-seeing cardiologists that
01:02:20.880 --> 01:02:26.480
enabled things to happen a little faster
and also, as I talked to other CEOs,
01:02:26.480 --> 01:02:31.920
enabled me to say, “This actually
works. This is the path forward.”
01:02:31.920 --> 01:02:37.600
I have recently been vocal about also
saying, we are at a point now where I
01:02:37.600 --> 01:02:44.160
believe that for some medical conditions, it
is not right to not use AI to help treat them.
01:02:44.160 --> 01:02:51.280
LEE: Wow, that's so interesting. So I think I
want to get into another topic here, which is
01:02:52.480 --> 01:03:00.800
when you think about the use of AI and data,
what are some of the results that maybe are
01:03:00.800 --> 01:03:06.320
top of mind for you or you think are particularly
important? And if you don't mind, I'd like to see
01:03:06.320 --> 01:03:14.000
if we can think about this not only in terms
of results in terms of patient outcomes but
01:03:14.000 --> 01:03:19.840
in your other activities, core activities, like
research, in the education mission, and then even
01:03:19.840 --> 01:03:25.240
in the broader impacts on the healthcare system.
But maybe we start with on patient outcomes.
01:03:25.240 --> 01:03:27.120
FARRUGIA: Yeah, they're all linked, right.
01:03:27.120 --> 01:03:27.190
LEE: Yes.
01:03:27.190 --> 01:03:31.280
FARRUGIA: They're part of the same ecosystem. We
think of ourselves as three shields— research,
01:03:31.280 --> 01:03:36.240
education, and the practice—and
that one goes into the other. So,
01:03:36.240 --> 01:03:41.200
as I said, we have about 320 AI
algorithms from the practice.
01:03:41.200 --> 01:03:46.720
Some run on every patient; some run on some
patients. And we have good evidence for what
01:03:46.720 --> 01:03:52.320
they do. So some specific examples, and then
I'll get into the transformer part of this.
01:03:53.520 --> 01:03:58.960
We have a program called CEDAR [Clinical Detection
and Response (tool)], and like most other
01:03:58.960 --> 01:04:06.880
people, I like acronyms for things. [LAUGHTER] But
what it is, in our hospitals with patient consent,
01:04:06.880 --> 01:04:12.720
we monitor vitals. We monitor in the patient
room—not in the ICU [intensive care unit], in
01:04:12.720 --> 01:04:18.960
the patient room. We monitor all sorts of
things. But there's a camera in the room,
01:04:18.960 --> 01:04:25.760
and we have a team of intensivists—nurses
and physicians—who do not have any patient
01:04:25.760 --> 01:04:31.760
responsibilities but are just monitoring the
algorithms, and when the algorithms are predicting
01:04:31.760 --> 01:04:37.040
decompensation, they're able to get into
the room. And what we've shown, for example,
01:04:37.040 --> 01:04:42.240
with that algorithm, is we've shown we've
decreased length of stay in the hospital,
01:04:42.240 --> 01:04:47.680
decreased transfers into the intensive
care units, and interestingly,
01:04:47.680 --> 01:04:52.320
decreased mortality and morbidity, which
is not easy to show. I talked about the
01:04:52.320 --> 01:04:58.480
electrocardiogram as a good example. Of course,
everybody knows about the radiology things.
01:04:58.480 --> 01:05:03.920
We've created … taken part of this and
said, if we can do this in the hospital,
01:05:03.920 --> 01:05:09.280
why cannot we do it in patients' homes?
So being very active in looking after
01:05:09.280 --> 01:05:15.200
patients that would come to the ED, emergency
room, would normally be admitted, and we say,
01:05:15.200 --> 01:05:21.200
no, here are the things we can give you. Go
home if you want to, and we will safely look
01:05:21.200 --> 01:05:26.800
after you at your home. And we recently have
been, looking at the last two years of data,
01:05:26.800 --> 01:05:31.040
been able to show that we're also successfully
able to give intravenous chemotherapy in
01:05:31.040 --> 01:05:37.120
patients' homes because we can monitor;
we can do all the things that we can do.
01:05:37.120 --> 01:05:42.720
Now, with generative AI, that gave us
many other opportunities. One biggest
01:05:42.720 --> 01:05:46.720
opportunity for me has always been
digital pathology. When we see how
01:05:46.720 --> 01:05:50.375
pathology’s currently run with a
glass slide, not much has changed …
01:05:50.375 --> 01:05:52.459
LEE: Yeah.
FARRUGIA: … in many, many, many years, right.
01:05:52.459 --> 01:05:52.470
LEE: Yeah.
01:05:52.470 --> 01:05:59.280
FARRUGIA: And so really we have made a massive
push to digitize pathology not just for us but
01:05:59.280 --> 01:06:04.160
for others. But talking about ourselves, we
started by saying, it has to be very cheap to
01:06:04.160 --> 01:06:10.320
digitize. So we worked and created a company
with partners called Pramana that allows us
01:06:10.320 --> 01:06:16.320
to digitize slides relatively cheaply using
AI algorithms that can take away the dirt,
01:06:16.320 --> 01:06:22.720
the fingerprint. And so we end up with 21
million of our slides digitized, and that
01:06:22.720 --> 01:06:29.600
gives you now a massive opportunity. Worked with
another company called Aignostics to create a,
01:06:29.600 --> 01:06:37.579
what we call, Atlas, which is an LLM
that allows us to then build upon it.
01:06:37.579 --> 01:06:37.590
LEE: Yeah.
01:06:37.590 --> 01:06:45.280
FARRUGIA: And we, a hundred and, I think,
120 years ago, invented frozen sections
01:06:45.280 --> 01:06:50.160
at Mayo Clinic. So what that is, is that
while the patient's still on the table,
01:06:50.160 --> 01:06:55.120
you can take a piece of tissue, look at it,
and tell the surgeon the margins of what
01:06:55.120 --> 01:07:01.040
you're trying to resect are clear or not. But
as a result of that, because you have to hurry,
01:07:01.040 --> 01:07:05.520
you get no information as a surgeon about,
is it an invasive cancer, is it noninvasive
01:07:05.520 --> 01:07:11.360
cancer, or other things. So we've just found
a way to digitize our frozen
section practice
01:07:11.360 --> 01:07:17.600
and will completely go across the enterprise
with AI-enabled digitized frozen sections,
01:07:17.600 --> 01:07:22.640
which then enables us to then do it for
anybody across the globe if we need to.
01:07:22.640 --> 01:07:29.680
And then in the genomic space, we're working
to create a true exomic transformer that is
01:07:29.680 --> 01:07:36.720
short range. And we originally started doing
it to see if we can test it against the fact
01:07:36.720 --> 01:07:41.254
that 40% of people with rheumatoid arthritis
don't respond to the first-line therapy, …
01:07:41.254 --> 01:07:41.266
LEE: Right.
01:07:41.266 --> 01:07:44.080
FARRUGIA: … but you have to wait six
months to find out. And we found that
01:07:44.080 --> 01:07:48.320
we can actually do that. But it
has much greater uses, of course.
01:07:48.320 --> 01:07:51.760
And then we're working with you—I don't
know how much you want to get into this,
01:07:51.760 --> 01:07:56.560
Peter, or [if] you want to talk about it
yourself—MAIRA-2, which is really exciting,
01:07:56.560 --> 01:08:02.080
about how taking a simple problem—can
you create a transformer that is able
01:08:02.080 --> 01:08:07.440
to detect if lines on the chest are in the
right place, breathing tube is in the right
01:08:07.440 --> 01:08:11.760
place?—and then do it in a way that then
can be used for many, many other things.
01:08:11.760 --> 01:08:15.494
And then, Peter, because you asked
about education and research, …
01:08:15.494 --> 01:08:15.506
LEE: Yeah.
01:08:15.506 --> 01:08:19.360
FARRUGIA: … imagine what this does now
to the education system, right. And so
01:08:20.320 --> 01:08:25.440
we've got to train our physicians differently.
We now have an AI curriculum for all our medical
01:08:25.440 --> 01:08:31.280
students. We offer masters and PhDs in AI. We
think it's essential for the people who want
01:08:31.280 --> 01:08:37.440
to be able to truly become experts, the same
way I became an expert in my area of research.
01:08:37.440 --> 01:08:43.120
And then from a research standpoint, when you
think about all the registries that exist in
01:08:43.120 --> 01:08:51.200
people's labs, all the spatial genomics, all
the epigenomics, all the omics that exist. And
01:08:51.200 --> 01:08:58.400
if you are able to coalesce them into one big,
what we call, an atlas, how that could really
01:08:58.400 --> 01:09:04.320
spur research at a scale that we haven't thought
of before. And so that is our aim at the moment.
01:09:04.320 --> 01:09:10.480
From a research standpoint, we are, with Vijay
Shah, who's our dean of research, is to say,
01:09:10.480 --> 01:09:19.040
let's make the effort of making sure all
the data are available to be able to use
01:09:19.040 --> 01:09:23.360
and enable for us to take advantage
of AI. And that is not easy because,
01:09:23.360 --> 01:09:27.360
of course, people have collected the
data. They tend to want to embrace it.
01:09:27.360 --> 01:09:27.400
LEE: Yep.
01:09:27.400 --> 01:09:31.200
FARRUGIA: So there have to be the
right incentives, the right privacy,
01:09:31.200 --> 01:09:36.240
and the right ways of doing it. And
we think we're on the way there,
01:09:36.240 --> 01:09:39.320
and we’re already seeing some
advantages from doing it this way.
01:09:39.320 --> 01:09:44.080
LEE: So we're running short on time. And so
I always like to end with one or two more
01:09:44.080 --> 01:09:49.120
provocative questions. And, you know, it's
tempting to ask you the provocative question
01:09:49.120 --> 01:09:54.320
of whether you think AI will ever replace human
doctors, but I don't want to go there with you.
01:09:54.320 --> 01:10:01.120
In fact, as I thought about our discussion, I was
reflecting. We were at a conference together once,
01:10:01.120 --> 01:10:07.280
and I was on stage in a fireside chat. And then,
you know, after the fireside chat, there were
01:10:07.280 --> 01:10:13.840
audience questions, and I don't remember any of
the questions from the audience except yours.
01:10:14.480 --> 01:10:19.200
And just to remind you, you know, I think when
I was on stage, we were talking about a lot
01:10:19.200 --> 01:10:26.320
of practical uses of AI to, let's say,
reduce administrative burdens and so on
01:10:26.320 --> 01:10:33.920
in healthcare. But you got up and you, I won't
say you scolded me, but you more or less said,
01:10:33.920 --> 01:10:41.280
is it the right idea to use AI to optimize
today's somewhat broken healthcare system,
01:10:41.280 --> 01:10:48.400
or should we be thinking more boldly about,
you know, a more fundamental transformation?
01:10:48.400 --> 01:10:57.520
And so what I thought I would try to close with
here is to hear what was really behind that
01:10:57.520 --> 01:11:03.160
question. You know, what were you trying to get
me to think about when you asked that question?
01:11:03.160 --> 01:11:10.240
FARRUGIA: So first of all, darn
your great memory [LAUGHTER].
01:11:10.240 --> 01:11:13.595
Belated apologies … I probably should have ...
01:11:13.595 --> 01:11:16.720
LEE: It was by far the best and
most sophisticated and, I think,
01:11:16.720 --> 01:11:20.160
thought-provoking question of all of
the ones that came out of the audience.
01:11:20.160 --> 01:11:24.800
FARRUGIA: What I was trying to get to is actually
trying to clarify it in my own head and then in
01:11:24.800 --> 01:11:31.440
the head of others is that we do not need to have
a linear path to get to where we want to get to.
01:11:31.440 --> 01:11:37.280
And we seemed to be on a linear path, which is,
let's try and reduce administrative burden. Let's
01:11:37.280 --> 01:11:46.080
try and truly be a companion to a physician or
other provider. Let's make their problems better,
01:11:46.080 --> 01:11:50.720
make them feel better about
providing
healthcare. And then in the next step,
01:11:50.720 --> 01:11:55.920
we keep going until we get to, now we can call it
agentic AI, whatever we want to talk about. And
01:11:55.920 --> 01:12:04.000
my view was, no, is that let's start with
that aim, the last aim, and do the others
01:12:04.000 --> 01:12:08.880
because the others will come automatically
if you're working on that harder problem.
01:12:08.880 --> 01:12:13.520
Because one, to get to that harder problem,
you'll find all the other solutions.
01:12:13.520 --> 01:12:19.040
I was just trying to push that here's
this wonderful tool that's been given
01:12:19.040 --> 01:12:26.240
to us. Let's take advantage of it as quickly
as we can. I think we had gotten a little too
01:12:27.360 --> 01:12:35.040
sensitized to need to say the right things.
“Careful, be very careful” versus saying,
01:12:35.040 --> 01:12:40.160
“Massive opportunity. Do it right, and
healthcare will be much better. Go for it.”
01:12:40.160 --> 01:12:46.960
LEE: Well, I think I understand better now
where the vision, insight, and frankly,
01:12:46.960 --> 01:12:53.360
courage to take on something as
ambitious and transformational
01:12:53.360 --> 01:12:58.080
as the Mayo Clinic Platform and
really all of your leadership
01:12:58.640 --> 01:13:05.920
in your tenure as the president and CEO of Mayo
Clinic. I think I understand it much better now.
01:13:07.120 --> 01:13:12.640
Gianrico, it's just always such a privilege to
interact with you and now to have a chance to work
01:13:12.640 --> 01:13:17.240
with you more closely. So thank you for everything
that you do and thank you for joining us today.
01:13:17.240 --> 01:13:19.200
FARRUGIA: Thank you for making it so easy,
01:13:19.200 --> 01:13:28.000
and thanks for giving us this
opportunity to do good for the world.
01:13:33.440 --> 01:13:33.675
[TRANSITION MUSIC]
01:13:33.675 --> 01:13:38.080
LEE: Gianrico leads what is arguably the
crown jewel of the world's healthcare systems,
01:13:38.080 --> 01:13:43.600
and so I feel it's such a privilege to be able
to talk and sometimes even brainstorm with him.
01:13:43.600 --> 01:13:50.320
Our conversation, I think, exposed just how
tech forward Gianrico is as he charts the
01:13:50.320 --> 01:13:57.120
strategies for healthcare delivery well into the
future. And as I've interacted with many others,
01:13:57.120 --> 01:14:03.600
what I've learned is that this is a common trait
among major health system CEOs. Roughly speaking,
01:14:03.600 --> 01:14:06.960
like we've seen in previous episodes
where doctors and med students are
01:14:06.960 --> 01:14:14.000
polymath clinician-technologists, the same thing
is true of health system CEOs and other leaders.
01:14:14.640 --> 01:14:20.400
AI in the mind of a health system CEO today
is not only a technology that can transform
01:14:20.400 --> 01:14:25.840
diagnosis and treatment, but it's also
something that can have a huge impact
01:14:25.840 --> 01:14:30.560
on the business of healthcare delivery, the
connection of healthcare to medical research,
01:14:30.560 --> 01:14:34.480
and the journeys that patients go
through as they seek better health.
01:14:35.200 --> 01:14:39.600
These two conversations show that virtually
all leaders in health and medicine are
01:14:39.600 --> 01:14:45.120
confronting head-on the opportunities,
challenges, and the reality of AI,
01:14:45.120 --> 01:14:54.640
and they see a future that is potentially
very different than what we have today.
01:14:54.640 --> 01:15:00.480
[THEME MUSIC]
01:15:00.480 --> 01:15:06.000
I'd like to thank Umair and Gianrico again for
their time and insights. And to our listeners,
01:15:06.000 --> 01:15:11.600
thank you for joining us. We hope you'll tune in
to our final episode of the series. My coauthors,
01:15:11.600 --> 01:15:17.200
Carey and Zak, will be back to examine the
takeaways from our most recent conversations.
01:15:17.200 --> 01:15:21.120
Until next time.
01:15:21.120 --> 01:15:22.005
[MUSIC FADES]