(audience chattering)
- So I'm Sean Pokorney.
I'd like to welcome everybody
to research conference today.
We're very fortunate to be joined by John Rumsfeld today,
who's currently the Chief Innovation Officer
for the American College of Cardiology.
He's also former Director of Cardiology
for the VA Health System, prior to this job,
and he was on full time faculty
at University of Colorado as well
while he was overseeing the Cardiology Practices
at the VA.
So he's responsible for developing the long-term
innovation agenda to help the ACC maintain
its competitive advantage in the global marketplace,
investigating global and domestic market trends
in the delivery of cardiovascular care
and Health Information Technology,
and exploring new technologies
in a team-based approach to healthcare.
And so we were just talking a little bit
about the excitement of his job
and he's, in addition to this,
extraordinarily well-published and an expert
on big data.
And so we're really excited to have him
with us here today.
- Thank you.
(audience applauds)
All right, very kind.
Can you guys hear me okay?
All right.
There's always that awkward ocean between me and you.
So really a thrill for me to be here.
The last time I was here visiting Duke
as a visiting professor, Adrian Hernandez was a Fellow.
So it's been a little bit of time,
but I've had lifelong friends and colleagues here.
And so I'm talking about something today
which I think might be different
for a research conference.
But what I'm really hoping is that it's the start
of a conversation.
I know I've already started it with a lot of the Fellows,
but I'm hoping others of you,
please I'm asking you, if any of this sounds
like a new frontier, that where we need
to go in research and collaboration,
in this case between ACC and Duke,
I would love to follow up with you.
It's as much of a conversation starter as anything.
So I'm gonna talk about
The Digital Transformation of Healthcare,
more specifically, a lot of the challenges therein.
But I'm hoping to end with somewhat of a road map
or a way forward.
So let me start in the most obvious place,
which is here, which is headlines.
Okay, the hype that we're seeing around digital care.
This is not fake, this is totally real,
which is the Tricorder X Prize offered
$10 million to build a Star Trek inspired
health scanner.
This is what's happening in Silicon Valley,
this is what they're going after with a lot of money,
where they wanted to have a scanner to diagnose
all these diseases, and it was inspired
by the television show Star Trek.
We also see a lot of headlines like this.
Artificial intelligence, or AI, which is the most
common and hyped term right now in Silicon Valley
and in the tech world globally,
predicts heart attacks better than doctors.
So apparently they're way ahead of us
and we're doing this.
Now why is all this happening?
It's all happening because of what's happened
in the last 60 years with computers
and computer power.
So a little over 60 years ago,
in 1956, this is what it took
to ship a five megabyte hard drive.
This phone has 256 gigabytes.
That's where we've come in the 60 plus years.
And this is being driven by an exponential growth
in computing power which continues unabated.
Some people wondered if so-called Moore's Law
of the doubling of computer power every 18 months
would go away, but it's very unlikely to go away
and the exponential growth is likely to continue
because of the advent of so-called quantum computing
which is now no longer theoretic,
but both the Chinese, the Americans and others
are moving it forward very rapidly.
So we're gonna see this continue to grow.
The most obvious manifestation of this
is in everything else we do.
All the other sectors of our world,
the financial sector, the entertainment sector,
the transportation sectors of our economy
have undergone a so-called digital transformation,
which makes things more efficient,
makes it easier to access, gives us new power
we never had before in a more rapid way.
And so you have to ask yourselves
if we've had this computing power
and we have all this technology
and it's disrupted all those sectors,
how are we doing in healthcare?
Well, at least in delivery of healthcare,
we haven't seen anything change.
Or as Harlan Krumholz likes to say,
we're still practicing
in a department of motor vehicles model
of healthcare delivery.
Now, are we high tech in healthcare
and in cardiovascular disease in particular?
Darn right we are.
New drug, new defibrillator, new stent.
We immediately adopt them in practice.
But what I'm talking about here
is not high tech medical care.
I'm talking about health care delivery,
where you can't tell when,
except for that it's not in black and white,
you can't tell when this was taken.
By the way, you can't tell if this is a DMV
or an emergency room.
And you also can't tell if there's computers
behind those desks which might be true.
But many people would say the first step
in the digital transformation of healthcare
was the EHR which was a misstep.
I mean has that really made us more efficient and better?
Probably not, even if it's digital.
We also have this problem
and I'm not gonna spend too much time on it
'cause I get tired of it as I'm sure
all of you do, and that is hearing how expensive
and ineffective our healthcare system is,
but it is a persistent drum beat
and it doesn't mean it's not gonna force change.
Something has to happen to change the fact
that from a value perspective
we're not getting what we put into the system.
And it's not that we don't provide excellent care
to our patients and excellent medical care.
It is the delivery system that's the problem.
Just to keep it a little lighter,
we rank 169th in the world in health outcomes
between Croatia and Guam, and you know they're
not spending 18% of their gross domestic product.
I think this is actually compelling.
You know we're in trouble when Warren Buffett
refers to the U.S. Healthcare System
as a "tape worm" for the economy.
Now that's both funny but it's also not funny.
If Warren Buffett, he feels that our
U.S. healthcare delivery system is actually
a threat to us as an economy and as a leading nation
in the world, in other words,
healthcare could take us down
as an economic powerhouse.
Something is going to change,
it has to change.
And heart disease is right at the top.
We're right at the top of the cost,
we're right at the top of mostly aging population,
but also increase in risk factors.
Obesity epidemic worldwide, of course
and so forth, driving an increase in heart disease
both as a disease and in cost,
and then all this variation and stuff
we have to deal with.
So I think that there's a compelling need,
a compelling need for the digital transformation
of healthcare.
What is digital transformation?
A couple things.
First of all the DX on the screen is not diagnosis.
I know we're all a lot of clinical background here.
This is the big DX that Silicon Valley refers to.
They call it digital transformation.
And what the most obvious example,
although it's a narrow example,
is the smartphone, where the idea is
that the best technology is silent technology
that makes something easier.
That is something comes in and replaces
all these other functions you were supposed to use.
That's the ideal of digital transformation.
We just have not yet seen it
in our healthcare delivery system.
Now here's the next paradox.
Silicon Valley thinks it's already happened.
Now when I go to clinic on Thursday,
when I go back today and I go to clinic
on Thursday at the VA, I'm pretty much seeing patients
exactly the way I did for the last 20 plus years.
But boy, you wouldn't know it from the headlines
that come out of Silicon Valley.
A digital revolution in healthcare is speeding up.
Telemedicine, predictive diagnostics,
wearable sensors and new apps will transform
the managed health.
Why data analytics, remote care
and interconnectivity are prepared to transform
medical care.
I still believe all that to be true.
And these are headlines from 2017.
The only problem is I can find almost
they exact same headlines from 2010, 2011,
2012, 2013, 2014, et cetera.
It really hasn't changed
all the promise and excitement.
But what has changed is this,
is that the money going into it is mind boggling.
Over $7 billion, with a B, $7 billion
went into digital healthcare startup companies
in 2017 alone.
$7 billion startup companies only,
and that doesn't even touch, and this is bearing,
oh, this is the manifest of that.
What has that $7 billion bought us?
It has bought us all these companies.
And, by the way, behind each of these groupings
are dozens and dozens and dozens of other companies,
and the second you make this slide,
it's out of date because they're being acquired,
they're folding, they're failing, they're whatever.
So there is just a massive hype of tons of companies
trying to sell you everything
from artificial intelligence analytics
to digital health, to everything else.
It's absolutely overwhelming,
and like I said, it still doesn't really change
how I practice health care on Thursdays at the VA.
But I have buried the lead a little bit.
I mentioned them earlier.
Some of those startups will succeed, potentially.
Not one digital health startup, however,
has made it to initial public offering.
So it hasn't succeeded yet.
But what is happening is they get acquired
by these people, and these are likely
to be the drivers in the digital transformation
of health care.
And it begs the question of what our role is.
They acquire those companies.
By the way, just as an aside, why do they
acquire those companies?
Not necessarily for their technology,
they acquire them for the people.
They're trying to get the data scientists
and the smartest people inside,
and then they're trying to drive this forward.
The other thing that's kind of, I think it's missed a lot,
is what's happening in the companies
we in healthcare know very well.
And these are just meant to be examples.
So if any of you (laughs)
I can add other logos to the slide
if anybody wants.
Don't overlook the fact that the companies we've known
and worked with forever, especially in the
medical device space, and so forth,
are skating as fast as possible towards this AI stuff
and the digital health stuff.
Have you noticed they've started stopping referring
to their stuff as products
and increasingly use the word solutions and platforms.
When you hear solution or platform,
you know they are tying it to a digital health platform
and/or they're using artificial intelligence.
They're going all in on this
and the FDA has said, and we've talked to them directly,
they're already wondering what they're gonna do,
all the next generation of medical devices
that we implant are gonna have AI embedded in them.
To do what?
To predict what?
And what are we supposed to do with that information?
And the FDA is saying, how are we supposed
to evaluate that?
So if you think this isn't happening,
it's happening and it's happening fast.
And where do we position ourselves?
This is causing a lot of discomfort.
We're seeing a lot of stuff like this.
The rise of artificial intelligence
and the uncertain future for physicians.
And I don't think it's just physicians,
it's clinicians in general.
And then of course you have some people
like Vinod Khosla, billionaire,
who runs a venture capital in healthcare,
who is basically saying, do we need doctors
or algorithms, I think maybe AI will just
replace clinicians altogether.
Potentially researchers too, by the way,
I come back to that.
Because, we'll just have this.
And it actually gets so far out there
it's actually hard to tell what is,
like so many things in our world these days,
real and fake news,
that actually when I saw this headline
which isn't real, I actually paused
and thought, wait a minute.
Amazon warehouses stocked with 20,000 doctors
in preparation for healthcare launch.
Also amusing but also speaks
to this point of, if we don't take
a leadership position and know what we're trying
to do in this whole thing,
we may become commoditized in this.
There's a pretty high chance we could become
commoditized if it just happens to us,
if it's just done to us
like the electronic health record
was just done to us.
So we have one huge thing in our advantage here,
and that is this, that despite all those companies
and all that money and all that enthusiasm,
the digital transformation of healthcare delivery
hasn't happened, and one of the big reasons why
is right here.
Is that the hype is crazy, what they promise.
The actual delivery or evidence
that these things can improve how we deliver care,
much less improve the outcomes of our patients
and do so in a way that creates value and efficiency,
almost purely lacking.
And so when we go back to Silicon Valley
and all these companies, if I've learned nothing else
in the last two and a half years, is there's an awful lot
of technology in search of a problem.
And they're very overt about it, we have this amazing tech,
and you say, well, what does it do, what does it do,
they're like, well no, that's for the doctors,
or the nurses to figure out what it's supposed to do.
They are technology solutions in search of a problem.
That's not going to work.
The second thing is that evidence,
I already said this about the hype and evidence,
there just isn't much evidence
about health technology, or how to evaluate these
to show they do what they do.
And the second part of it, I just wanna say,
and can be an enemy.
You can't underestimate that most of these companies
have venture capital backing.
When they only have a little bit of time
to develop a technology and get it to market,
a vast majority of them have little to zero interest
in actually studying the impact
of their technology on patients or health care,
because what they're really hoping
is to get attention and get acquired.
They're looking for an exit,
a financial exit.
And they're under pressure from VC funding,
so you have to understand that a lot of these
are not going to be interested even
in the kind of evaluation we want.
And then last, but most important,
clinical insights and integration are largely absent.
I'm picking on Silicon Valley, but it's really true
for tech around the world.
They just fundamentally don't understand
how we care for patients, how we interact with patients,
our goal, even as care evolves, maybe away from
hospital and clinic-based and longitudinal care
and so forth, which I hope it continues to do.
They still don't understand what we're trying to do.
They lack this insight and it's really a problem
for their success in the long run.
So let me go back to the headlines I started with,
just to give a couple of examples.
So you'll notice down here in the low right,
this award for the tricorder prize,
they got 35 companies went in and 10 got to the finals.
They actually produced a product.
They gave the award, okay, last year.
But notice, and it's subtle,
but notice what the headline is.
The Qualcomm Tricorder X Prize has its winner,
but, notice the word but not and,
but work on tricorders will continue.
You know what happened is they had no trouble
getting a bunch of sensors into this thing
and get physiologic parameters, heart rate,
pulse ox, blood pressure, whatever.
They had no clue how to turn that into information,
it was just a bunch of data.
The goal for this was that this thing
was supposed to magically diagnose up to 10 diseases.
No, couldn't do it.
They couldn't figure out how to take data
and turn it into information,
probably for all the reasons that I mentioned.
And on this one, and this is a tough part
of the talk for me, because I wanna give credit
to the researchers here, because they're actually
studying whether or not artificial intelligence matters.
I want to give them all the credit in the world.
And by the way, they can't control the headlines.
But I do wanna point out that AI predicts heart attacks
better than doctors, wasn't really predicting
heart attacks and had nothing to do with doctors.
Otherwise, it was a highly accurate headline
from NBC News.
This is the actual study that it came from,
in PLOS One.
Can machine learning improve cardiovascular risk prediction
using routine clinical data.
And what I wanna point out is
all they were doing was taking that ASCVD risk score,
you know, like six variables.
That's a parsimonious risk score.
We would expect that if you added
a bunch more variables to that,
you might get a little better.
And look at what the actual results were.
Now C index, or AUC is not the be-all and end-all.
There are other ways to evaluate predictive models,
but it is what they primarily reported.
And I just wanna point out
that just that ASCVD risk score, the six variables,
did pretty good.
It had an AUC or a C index of 0.723,
and when they put in hundreds of other valuables
from machine learning from the EHR,
it went all the way up to 0.764.
Now as researchers, clinicians
and all of your backgrounds, is that important?
Is that a clinically important difference in prediction?
And by the way, it has nothing to do
with actually making a care decision.
The other thing, and I found this
pretty disappointing, including from the researchers,
which is they also did plain old logistic regression.
And they put more variables in the logistic regression
and they did just as well as the machine learning.
So this does not tell me that machine learning and AI
is this magical thing.
I'm glad they did the study,
but the result, the conclusion here
should not be that the machine learning
significantly increases risk prediction,
increasing the number of patients identified
who benefit from preventive therapy
while avoiding unnecessary treatment of others.
To me that's a conclusion way beyond.
I'm glad they did the study, but it's way beyond
the results of the actual study.
And it just screams that we need
to be doing meaningful research here,
which asks us the question and then we are objective
about where machine learning is or isn't going to be
an advantage to us.
And by the way, even where the prediction is better,
then we need to put it in and show how it changes
how we make care decisions,
integrated into care.
We've got a long way to go here.
I wanna point out this is starting to come out.
There's starting to be backlash.
We're starting to hear the venture capitals,
they're starting to get cold feet
about digital health in general and so forth.
And we're seeing more things like this,
which is going in the opposite direction
of what we want.
Eric Topol, Scripps wired for health study,
randomized trial had no clinical economic benefit
for digital health monitoring.
Poor healthcare apps could cost money,
well that's not supposed to be.
It's supposed to be helping us save money,
et cetera, et cetera.
So at the end of the day,
and I promise I'm getting past the challenges
to the potential road forward is
if digital health is the future,
the future is not here yet.
I think this is actually good news.
So despite all the hype and all the stuff,
we're actually at the beginning.
We're in the introduction to the book.
We're not even in chapter one or two,
which means we can have a role in this
and guide it forward.
I do believe this is true.
My friend, Ashish Atreja,
is the Chief Innovation Officer at Mt. Sinai,
likes to say that, in the future,
digital medicine will just be called medicine.
Okay, but there's no way we're not going to evolve
and start to adopt technologies where it makes sense.
But the question is, can we guide the evidence base
to have it there.
So with that in mind, that's what ACC is trying to do.
Now we're trying to do it, not alone,
in fact ACC will just fail miserably
if they try to do it themselves.
This is a idea to facilitate and work collaboratively
towards this, but we have tried
to put our name out there in collaboration
with stakeholders across healthcare.
And by the way, we wrote this.
We had patients, consumer groups, payers,
government, hospital and health systems,
the tech industry itself, both startups
and established companies as well as ACC involved in this.
And we did publish a few months ago in November,
A Roadmap for Innovation, how to pursue healthcare
transformation in the era of digital health, big data,
and precision health.
I'm glad it did actually get more attention
than I thought it would in the tech world,
probably getting coverage in Mobile Health News
is more important than getting coverage in
most of the scientific journals that we read,
and it did lead to an amazing amount of,
I mean ridiculous amount of incoming interest,
I was glad to hear, from tech companies
large and small saying, well are you really willing
to partner and tell us what problems to solve
and how to evaluate these?
'Cause that's what they don't know how to do,
and that's what we're trying to do.
We're still at the beginning.
This is probably too complex for the size of screen
to where you guys are.
Let me just point out a couple things.
But I just wanna, we laid out a roadmap of steps.
We're embracing that this is not easy.
But I do wanna point out that we have
a strong emphasis on what problems
we're trying to solve, which is a good place
to start rather than the tech forward thing.
That we're trying to think about workflow integration,
how does this integrate into clinical care,
even as we evolve our care models.
That we're willing to partner, or actually that
if we don't partner this is not gonna work,
and we are partnering with the tech world,
which I'll come back to.
That we need to develop a research and evaluations
network, I'm gonna come back to this
but I'm increasingly thinking this is the key to this.
That we have to figure out a way to evaluate these.
And then of course there has to be the payment model,
alignment, and this isn't a policy talk
but I'm gonna come back to that in just a second.
Generically what are we trying to do?
And I know you've heard it a lot,
but I keep trying to think of a better way
to say it, which is, okay right now
we mostly deliver care.
I like to say we deliver care cross-sectionally.
I either see my patients in the hospital
or I see them in clinic and then I say
I'll see you in either three, six, nine or 12 months,
as those are for some reason the only choices I have.
And then you come back and I see another cross section
and then another cross section.
And of course I'm missing, and this idea that when,
although I love it from a social standpoint
with my own patients, but when they come in
and I say, hey, how ya been?
(laughs)
It's ridiculous.
I have no idea what their actual health status is
over time, and how can we measure that meaningfully
in a way that matters?
So I do think we're after-- this is from our friends
at Phillips, but I couldn't find a better one.
I do think we are conceptually after this idea
of way more emphasis on when people are doing well,
why do we keep bringing them back in and re-testing them
and changing things?
That doesn't make any sense.
(audience member sneezes)
Obviously we'd like to have a focus, bless you,
to the left on prevention, and then bolster digital tech
on how we do diagnosis and treatment.
So all of this is supported by monitoring informatics
and connected care, which is the digital transformation.
None of this is gonna work without
the payment model changes.
And I'm not, again, not a health policy person,
but there's a lot of questions about this.
Well it seemed like the kind of,
like CMS backed away from the bundles,
we're still in fee-for-service,
and so what are we doing?
And I just wanted to say that I,
as best I can tell, just by surveillance
and talking to payers, government and everything,
we're still skating that way.
I still firmly believe.
It's hard to know the exact timeline
that we're skating that way.
Certainly Seema Verma has renewed
the commitment to value-based models
and there is at least one or more cardiologists
in the CMS payment group now
who have made it pretty clear they're gonna
re-invigorate the bundles and go back at it.
And the private payers are skating this way.
Even in the fee-for-service system,
I don't know how many of you saw FDA,
CMS announced new codes for reimbursement
for remote monitoring for the first time.
This does open the door.
There could be a payment model for it
in addition to chronic care monitoring.
It's a different code.
It's an additional code to get paid
for remote monitoring of our patients.
And then of course, the big disruptions,
which is, you see things like CVS and Aetna
coming together and you've seen other big things
where these large payers may not only
be bringing in a payer to be
with a health system delivery,
they may actually invent their own delivery system
and bypass the payment system altogether.
It would be very disruptive
but it may open the door to,
what do you think they're gonna be interested in
if they do that?
They're gonna be interested in the long-term
health management of the patients,
not bringing them back in to clinic
or bringing them back into the hospital.
When they're doing well, leaving them alone
and saying, great, you're doing great with your health.
So it could be a game changer.
And then that could shift the rest of the system
to stay out.
So what are we talking about with digital transformation?
I'm gonna go down a whole step here from the concept
and then get back to research before ending my time.
So Kamal Jethvani at Partners in Boston
has put out, I think, a nice way of thinking
about digital transformation in buckets.
He calls them phases and I took the word phases out
'cause I don't they're necessarily in series phases,
I think they're all going in parallel, okay,
there's four of them, but I think they require
different levels of evidence
and I think they're gonna happen at a different speed,
and so that's why I like separating them.
One is just digital tools,
and digital tools in my view
are just replacing something we already do now,
like in the hospital, like during hospitalization,
predicting risk and outcome.
A new digital tool.
I saw a nice study about a new smartphone app
that, through the camera, does a better job
of the Allen's test than we do.
Like superior, it just came out of randomized trial.
Probably that's a digital tool we should adopt
sooner than later, but it's in the hospital.
This is in-hospital or in-clinic digital tools.
The second phase of digital transformation
of healthcare using technology will be virtual care.
We're getting there.
You're seeing a lot about telehealth, telemedicine.
You're seeing some cost-effectiveness studies
that are finally positive (chuckles) in this regard.
This is replacing what you do now
to improve access, patient experience and engagement.
And we are seeing the payment models
slowly but surely move that way.
We still have some state line issues
and other things to overcome.
The things that get the most attention though,
rightfully so, I think are remote monitoring.
So this is data we didn't have before
to inform care management.
Mobile apps, biosensors, voice interactions,
a big one, including health assessment
just from hearing people speak over time.
Video, other things like this.
This is, I think, a major area
for digital transformation, one where we need
intensive evaluation for evidence.
And then the one that everybody talks about.
It's hard to get away from the phrase AI
or artificial intelligence.
So-called AI-driven care, this is where you're using
data science in evolving models of artificial intelligence
to predict, interpret, potentially provide
care recommendations and even
to the point of digital therapeutics,
that is where the AI itself is making
the recommendation for what decision's to be made,
including direct to patients.
And this is where you get out on that crazy, uh oh,
how far is this gonna go.
I would point out, and we'll come back to it
in a minute, that this is the one
that needs the most careful evidence evaluation.
So if you look across, to summarize the phases
or stages or what that takes for digital transformation,
from digital tools, virtual care, remote monitoring,
AI-driven care.
I'm just gonna, and this is just an end of one opinion.
I'm gonna say that the left side of the screen
is gonna accelerate and go very fast.
Why? Because when we have a digital tool,
we already have a criterion to compare it to,
like the thing I said with the smart phone
compared to the Allen's test.
It's very straightforward for us to assess
whether or not these digital tools,
A, work and make us more efficient,
or something we wanna adopt in care,
plus it's in our environment of the hospital or clinic.
I think virtual care is more or less a slam dunk here.
Telling me we're not going to have more virtual
longitudinal care and interaction, it's not.
The technology's pretty straightforward.
Yes we have to align the payment,
but I think it's pretty straightforward
and we're seeing some really great clinics coming.
Stanford has one, the congenital heart disease clinic
that Ami Bhatt has developed at MGH.
If anybody wants to be connected to her
and how to set up a telehealth clinic, phenomenal.
Young, early career cardiologist, who I think she's
done a phenomenal job.
Where I think we need to put all of our effort
and focus as clinicians, researchers
and the health system as far as requiring evidence
is all on the right here in remote monitoring
and AI-driven care.
I think this idea of just buying the hype
it's not gonna work, it's a huge mistake,
and I think it can actually harm patients.
Because if you're gonna use AI-driven care
on genetic screening for markers
that are not validated and then put defibrillators in them,
we're hurting them, and it's already happened.
There are already case studies of this happening.
This is where we need to weigh in
if digital transformation's gonna be successful.
So even in that AI-driven care,
on the right side where I had the red,
I do wanna be clear that it's still a spectrum.
I do think some things will come faster than others.
And I think the thing that's gonna come faster
than we think is image interpretation.
We are already seeing the big companies,
Siemens, Phillips, GE.
Whether you know it or not,
they're in their new releases of their software,
they are using machine learning and AI
to increasingly improve the views
and the interpretation of algorithms.
And I know from some of the publications
that are coming out, I know these investigators.
Publications lag where you are,
and there are cardiologists in this AI space
and the imaging interpretation space
that are further along than we realize
in interpreting images.
And yes, I am talking about things like
AI pre-reading echos, and then we'll just over-read them
and who says how many.
This idea that we scan through
and look at all the images and then,
I think that's gonna go away sooner than we think.
It's hard to say exactly how soon,
but I think image interpretation's gonna happen fast.
It's the prediction, and what we do
with predictive models and AI, that's the real challenge.
So earlier I picked on a study a little bit.
I wasn't trying to pick on the investigators
that did the study, but how it was covered anyway.
Let me give you a more recent one.
January 2018, Google, University of California,
San Francisco, Stanford and University of Chicago.
Love to see the academic and tech world collaboration.
I wanna point out that they did deep learning,
which is just another way to say AI,
machine learning, they're subtle differences.
Electronic health records, just from two hospitals.
They found 216,000 patients.
Want to guess how many data elements
that translates into?
The electronic health record?
Two hospitals, 216,000 patients.
I can tell you I was off by a magnitude,
and in fact I read it wrong the first time.
Yeah, that would be 46 billion data points.
I thought it was 46 million when I read the study
and then I realized there was a whole other comma.
(laughs) Three things.
46 billion data points.
It does show the power you can get
from machine learning and AI approaches,
that it can actually requires and handle that much data.
And look at these risk prediction models,
just real quick, we won't spend too much time on this.
But they did it at different time points.
Admission, 24 hours into admission,
point of discharge.
They were able to predict in-hospital mortality
with an AUC of 0.93, 30 day readmission
at the point of discharge 0.75,
prolonged length of stay 24 hours in,
and then even predicting what the
primary discharge diagnosis would be,
which the only comparative standard
is about 50-60% accurate.
They outperformed every existing model that we have
for this stuff.
Very impressive.
But I still wanna point out that we've had
risk prediction from the beginning of medicine.
How much do we use it and how much
does it change care decisions.
This is a new frontier of having powerful prediction,
but we still have a long way to go
to show it actually changes care,
makes us more efficient and improves outcomes,
and one of the first people to weight in on this
in Twitter was Dr. Califf, who said about this study,
"Nice advance in applying quantitative methods
"to EHR data, but be careful.
"Is medicine mesmerized by machine learning?"
That is, are we just so wowed
by the predictive accuracy, but what do we do with it.
And what he's referring here to
is an article by a biostatistician,
who used to be at Duke, Frank Harrell,
many of you will know.
I highly recommend for anybody vaguely interested
in this thing, Frank Harrell's statistical thinking blog,
or just follow him on Twitter.
Frank Harrell's a leading biostatistician
in this country and he is not buying
machine learning and AI in healthcare.
Simply not buying it.
He thinks it has fundamental flaws
in the way it approaches data.
"Cause what does machine learning and AI do?
Machine learning and AI is basically
very powerful and iterative, 'cause it learns,
but it's pattern recognition.
It classifies people into groups
and compares them.
It's not prediction.
And his point is, Frank Harrell's,
is that classifiers are far from optimal
in medical decision support.
And he's really worried that this isn't going
to inform us as well as we think it is.
And I love this analogy he draws out, a lot,
which is, a poker player wins because she is able
to estimate the probability she will win
with her current hand, not because she recalls
how often she's had such a hand when she won in the past.
And that's the difference between prediction
and classification.
So will he be right or not, I still think it's very
powerful and impressive but we have to prove
how it's clinically useful.
And then of course in Silicon Valley in the tech world
they think the EHR is truth, and that that's
all the data.
It's absolutely true, and they're totally missing
that it's only as good as the underlying data.
That's why the imaging is gonna go further faster,
because the underlying data in imaging is good,
high quality data.
Whereas all this risk prediction is still based
on observational data with underlying
data quality problems, inherent bias
and this correlation causation problem.
So we have a ways to go,
and it opens the door for research.
I hope it's an evolution in thought,
but I've been thinking more and more
that as we look at digital health
and health analytics and health technology in general,
that the way that gets translated
into the actual meaningful digital transformation
of health care has got to be through
health technology evidence generation.
That we have to figure out how to do this
and it can't be this long cycle,
academic seven plus years, phase I, phase II, phase III.
It's not gonna work.
But I do think we need evidence
and I think we have to figure out
how to collaborate on this.
ACC is gonna need to collaborate
with academic research organizations
like Duke and others who have started a little bit,
and I'll show you that in a minute,
to realize the gains of digital transformation.
And I know this is my geeky scientific thing
for the day, but it's like the Bernoulli Equation here.
What I'm gonna say is, remember in the
Bernoulli Equation as you go through that
actually the flow speeds up.
And I think right now all the technology on the left
isn't getting into our care.
It's just not working.
And I think the way to speed up
the digital transformation is we have
to put together health technology evidence,
and how we're going to generate that.
Another way to say it is that
you may also like, like Amazon and Netflix
and a new product or a new movie,
it's just not good enough.
That's not good enough for healthcare.
The stakes are higher.
We rightfully require evidence,
and, importantly, assessment of effects
when it goes into care.
Let me get to the end here
with a couple of examples.
So at ACC we are, I'm not just conceptualizing this,
we're trying to do this.
I am very open to additional partnerships by the way.
We are launching an Institute for Computational Health
with Yale, with the Data Science Math Group,
on machine learning and AI to ask actual seminal questions
about where does AI machine learning
actually gain us, where can we apply it
and gain information.
And then how do we deploy that in care,
rather than just assuming the hype of AI.
We have launched a company on remote monitoring
and heart failure.
So it's a tough area, but we are doing it.
Understand we're designing it together,
co-creating from scratch, including the
clinical enterprise and using our own practices
through the Pinnacle Registry to kick the tires
and give feedback how we want
to get this information back into the practice.
And just more generically we're trying to use
the cardiovascular practices from across the country
to work with all these start-ups
or other organizations where we're actually partnering
to either co-create for the very first time
the digital health solutions.
So in other words, not just taking them as a customer
but can we actually develop them and then test them
for evidence in practice.
So far so good, but we're right at the beginning of this.
Even though we're now getting deeper and deeper
into the tech world and all those logos,
I just wanna say that the commitment
is that we will never forget,
that no matter how fancy the digital health tech is,
that the care delivery matters.
And I think that's the secret to this
is remembering that.
We can remember the sphygmomanometer,
everybody remember that?
So the sphygmomanometer was sort of personalized data
in its day.
So before that we knew high blood pressure
predicted stroke and death.
We didn't have it.
All of a sudden we had a point of care tool
with real data that told us
if an individual patient had high blood pressure
and we could treat it.
But no matter how fancy the tech is,
and it's digital and home monitor using trends,
we still debate how low to treat people's blood pressure,
we still debate which medications to use,
and we still need to work with our patients
on adherence and how to take those meds.
It just reminds me that the tech and the data itself
can enable healthcare transformation,
it doesn't cause it.
And I think it's a fundamental thing missing
in Silicon Valley.
One last concept I wanna get across,
'cause I'm increasingly thinking about this
is we need to combine artificial intelligence
and clinical intelligence.
And what we're not after is artificial intelligence
to make decisions.
I think it's simply not going to happen,
this generalized AI thing.
I think what we want is augmented intelligence,
that it gives us information we can use
that makes us better at what we do.
And this isn't my concept, Bob Harrington,
former DCRI Director, now at Stanford, as you know,
wrote this, I love this thing.
They wrote this in January, in JAMA
with Abraham Verghese and Nigam Shah.
What this computer needs is a physician.
And talking about this idea of augmented intelligence
to make us better and sort of getting this back
on track, this digital transformation.
At the end of the day, when my daughter,
if she decides, she says she wants
to be a doctor, we'll see.
If she is, I do believe she will practice
in a digitally transformed way,
and I do believe that clinicians
who use digital medicine and digital tools
will be superior to clinicians without them,
but only if we build the evidence base.
And I'll end with one hopefully amusing anecdote
about this idea of AI or tech replacing clinicians.
I'm gonna be on a panel at Stanford
with Dr. Harrington and others,
and they want it to be provocative.
So I'm on with a leading AI data scientist from Stanford
to plan this session.
And the people organizing the conference said,
what if we did the panel topic of,
"Will AI replace Cardiologists?"
And I was gonna weigh in and say,
that's fine with me, 'cause you could probably tell
I'd be fine to tell you about this.
But actually before I could say a word,
the data scientist jumped in and said, very strongly,
"No, because it's a stupid question.
"It's never gonna happen," and this guy's right
at the front of AI in healthcare.
"It's never gonna happen and it gets
"in the way of talking about how AI can help cardiologists."
And I thought, wow, that's amazing.
And just as I was processing that,
he said, under his breath, "I'm not so sure
"about radiologists though."
(audience laughs)
Thank you for your time.
(audience applauds)
Oh good, enough time, anybody want questions,
thoughts, anyone?
- [Woman] Thank you for the very interesting talk.
My question is like if we are going so much
towards technology, in the future or like
not a very long future, but what do you think
will change in how we train our medical students?
- (laughs) Boy is that a good question.
I don't think anybody's asked me that before.
The short answer, and I'm guessing
it's why you asked the question,
is we better start changing how we train
our medical students right now.
I don't know if you've seen in the,
they have a lot of pros and cons,
but in the national health system
in the UK they have fundamentally started doing this.
They've recognized the rise
of the medical entrepreneur, for example,
that a lot of Millennials,
and what's behind Millennials, Gen Z,
thank you, I guess that'll be my children.
Yeah, they're extremely interested in this stuff,
and, in fact, comfortable with it and promoting it.
Yes, even if they wanna go into healthcare
and into medical school, but they also
wanna be entrepreneurial, they want to embrace tech.
Even if they wanna go into academics,
they want a hybrid of academics and tech focus,
even in practice they wanna be the front tech,
and they've started to put into their medical training,
courses on entrepreneurship, technology,
some, if you want, informatics and computer science.
I don't know the degree, so does every medical student
need all of that?
I guess I could make a pretty strong argument
that most medical schools in this country
still in the first year of medical school
teach biochemistry.
And my question is, is that necessary
in medical training?
There's also a big move towards thinking about
how do you get, people are gonna do
this augmented intelligence
and at the humanistic side of healthcare,
we need to be more humanistic,
EQ over IQ, this kind of stuff,
there's also that move.
But I think there's every reason to think
that we should be teaching, informing
the next generation about entrepreneurship,
digital tech and so forth, and including
in research training.
This shouldn't just be clinical training,
because look at this open space.
I hope it came across clear.
Wide open space for us to get in here
and say here's how we should be evaluating technology,
and almost no one leading in this space.
It's pretty open.
Thank you for that question.
Anyone else?
Okay, oh, yeah, sure.
- So what is that magical partnership
between tech and healthcare like?
- I like to use the word magical
because it implies things that don't exist
and so it's pretty new.
I do wanna say that there are a few places
that have started to do it.
Doesn't mean we're doing it right.
But what we've started to do,
okay, so I'll start with ACC,
'cause you should always self,
be as critical eye on yourself,
and then let me tell you about
a couple other places real quick.
We have started to form,
for the very first time,
actual partnerships.
I mean, we're not paying them
and they're not paying us.
If they're willing to partner,
if they get this, they aren't just trying
to sell us something,
I just screen all those out that do.
But if they actually get this,
and they're willing to co-create, even from scratch,
what problem were you trying to solve,
how will you use it?
They're the technology and entrepreneurs,
but we're the clinical enterprise
and we also have the practices and the care integration.
If they're willing to partner,
I'm willing to consider actual development
of the tech solutions, try them in the clinics,
and we're willing to go to market
with the brand and revenue share and all that.
So we're willing to actually, for the first time,
co-create and go to market to do that.
It's a new area, it is tech that we're working with.
I'm sorry, it is the industry.
It's a different kinda industry, mostly start-ups.
But if we're not gonna solve this,
I don't know what else we're gonna do.
So we're actually trying to build
the digital health tools we have.
I mentioned the one company we started,
but that's just one where we launched it.
We have 10 other active partnerships going
in various parts, AI, digital health,
across different conditions.
You know a lot of these are gonna fail,
'cause this is innovation and there'll be
a high fail rate.
This has been a huge shift for the ACC Board.
Several other professional societies
have asked me to come talk to their boards
about what the ACC is doing
and I will tell you they're very well-known
medical associations with acronyms you would recognize.
Every single one of them has told me no way.
We don't have the risk.
Our board just can't do this.
So we're definitely out on a limb
so we may fail spectacularly,
but I hope we'll do it with style. (laughs)
(audience laughs)
But I'm emboldened by the shifts I'm seeing
in academic research.
Academics and research are kind of bad words
in Silicon Valley 'cause they think
they're gonna get caught up
in some grant process or whatever.
But all of a sudden this is shifting
and shifting fast.
I'll just give a few examples.
Scripps in San Diego, all of a sudden,
they have a Qualcomm-funded Fellow
to learn entrepreneurship and tech evaluation
and they have multiple junior faculty
in cardiology who are studying digital health
for the first time.
And they're looking for national mentorship
and stuff but they're starting to create
an academic bridge.
You definitely see this at, a lot of these
are in California, UCSF.
The system itself has been ignoring,
the cardiology group has built a very nice
digital health platform for evaluation of
digital health tools.
They've run multiple studies.
And for the first time, the health system
is saying, well wait a minute,
maybe we could actually evaluate
the digital health tool
and it'll help us figure out
which ones we should actually put into the system.
Wow, what an idea.
Actually marrying the research enterprise
with the system.
Stanford has launched it's Center for Digital Health
thanks to Bob Harrington.
It's still young but you've probably seen
they're working directly with Apple
on the Apple Heart Study and how to figure out
detection of AFib.
You know that's being driven by
the researchers and clinicians at Stanford,
advising Apple, I think that's pretty interesting.
So those are the examples back there.
WashU is starting to try to figure out this
under the direction of Tom Maddox,
and then I'm hoping that Duke is doing this.
In meetings yesterday, including with Adrian and others,
hearing what's happening
with the commitment of the health system
for the first time and the potential learning
health system to do just this.
I know it's new and nascent
and barely being announced here,
but I'm hopeful that this is maybe the next place
that steps into this fray,
including potentially training of DCRI,
research Fellows and so forth.
And if ACC can help support that
and facilitate and get national mentorship
for these people, we're there.
Okay, that's too long of an answer for your question.
(audience chattering)
- Well you talk about how digital technology
can add value in health, and of course
our payment programs.
We want to encourage value in the delivery system,
and I know you're not the policy guy,
but if you have thoughts about whether
accountable care organizations
are an appropriate model
to encourage this type of innovation,
or what role can payers play?
- So, I think the payers have been
a obvious barrier, because the fact is
they talk about innovation and ACO models
and alternative payment models,
but they mostly are still sitting there
with actuarial tables and estimating,
they really aren't committing yet
to innovative payment models,
which would say, all right, fine.
So for your heart failure patients
will go at risk, and first of all,
we're not very good at the at risk part either,
which is the second thing,
so it's us and them.
But we haven't come together to say,
so all right, so take care of
your population of heart failure patients.
We'll pay you this much.
It would suddenly be in our effort,
it would suddenly be directly in our best benefit
and greatly the best benefit of our patients
that if they're home and healthy and doing well
we leave them alone.
Why do we keep reminding them they're sick
and bring them in and do stuff, I don't know.
Because right now the payment model
isn't set up.
So if we could get to that, that's great.
The payers, I think, are behind on that,
to your point.
I will say, and this isn't meant to be
too glass half full, cause this thing
is going way too slow.
The most common new position being created,
at private payers in the United States right now is
innovation payment model,
or chief of innovation payment models.
Blue Shield of California has one,
he's married to a cardiologist at Stanford,
and he is actually talking about this stuff.
He's admitting it's hard
to get the Blues and others to buy off on this.
But at least they are appointing people
to think about doing this.
I told ya I think there's some movement
in CMS back towards the bundles,
and back towards alternative payment models.
So it's like kinda the slow painful steps,
but they all are going in the same direction.
And what I don't know,
and maybe you have an opinion on it,
is what if somebody actually bites
and does this and shows it works.
So maybe the Blue Shield of California,
Ed Jen is his name.
If Ed puts together a payment model
that could do that and gets practices to buy off on that,
can it work?
I think a lot of the reason
the hospitals health systems,
which you know are all getting bigger and bigger
and going together, I think the reason
they haven't pushed the payers harder
on doing this is 'cause we're not sure
how to estimate at risk.
So we're really good at risk in the hospital,
we're really not good at the,
and which of these digital tools do you trust
to monitor the patients at home
and know they're doing okay.
So we're gonna have to have a skate
on the digital technology with some evidence,
the payers being willing, and then
our health system saying okay,
we're ready to go at risk.
And that's the only thing,
that they just don't know when that happens,
but I hope you think it's a realistic assessment
of the situation as of today.
Doesn't mean it won't happen, but, when.
- [Man] So with the digital equipment
that are currently available,
what are your recommendations for practicing clinicians
and what they should incorporate into their practice
and how to balance what's available
versus what's reimbursable.
- Yeah, okay, those are two important questions.
Okay, you know there's no magical answer to this,
so I'll use a story.
My wife is a hospital administrator,
ER doc by background, CMO, now COO of a hospital.
So you might imagine she keeps asking me,
'cause she runs a cardiovascular service line,
and for her system too,
they keep asking.
All right, so we keep getting pitched
by a lot of these start-ups and stuff.
We do this amazing artificial intelligence,
we do this amazing digital health thing,
I will tell you for the most part,
I'm just telling her to ignore them.
Because I don't think they have any evidence
and I don't think they can deliver.
You're just adding to the cost of care.
I will tell you that the system nonetheless
gets wowed by some of these technologies
and buys them, and then tells the individual hospitals
in the system or the clinicians,
here use this AI tool.
And all of those have failed in her system,
and Harlan told me Yale just did this,
and the same thing, it failed.
Shocking, right?
That the system administrators would buy something
that's not been proven and then it doesn't work.
So, I know it sounds crazy
with all the hype and the stuff,
but I'm sorta preaching patience
and saying I don't know, because none of them
have any evidence to tell me which ones actually work,
so why don't we go build the evidence.
They can't wait forever.
I would say the answer to that is
that a few places,
oh and this was to an earlier,
I lost the thought in an earlier reply,
so I'm glad it came back in my head,
which is, I would pay close attention
to health systems that also have their own payment model.
Geisinger, UPMC, two examples.
Geisinger and UPMC already have those incentives aligned
that if they could figure out
which of these digital tools work.
And look what's happening in their innovation section.
The UPMC innovation, Rasu, I can't remember his last name,
Shrestha, Rasu Shrestha runs it.
It was a small little innovation thing
five years ago.
He has 150 people.
You know why?
Because UPMC is actually looking to them
to do exactly what I'm talking about.
Okay fine.
Either from scratch build a digital solution,
problem, or try to solve working at the tech companies,
or take their rapid cycle,
let's test them in our system,
and if they work we'll take them.
Because we do care about the home monitoring
because we have the aligning incentives.
I'd say the system leaders so far in this
are the Geisingers and UPMC.
Some people will say Kaiser.
I think it's been more of a mixed bag
in Kaiser so far, but those three.
And then, otherwise wait and lets get
the evidence out there.
All right, well thank you for your time.
Appreciate it.
(audience applauds)
(upbeat music)
For more infomation >> Digital Inking a Graphic Novel - Duration: 10:33. 
For more infomation >> Awesome Free Intro Templates - The Best Digital Hi Tech Logo/Name Reveals - Duration: 1:08. 
For more infomation >> Edición Digital Houston 05/01/18 - Duration: 37:05. 
For more infomation >> Spectrum's change to all-digital could impact your TV channels - Duration: 2:25. 
For more infomation >> News 8 Digital Update – Tuesday evening - Duration: 0:48. 
For more infomation >> Video: Scammers target consumers using digital devices - Duration: 1:51. 
Không có nhận xét nào:
Đăng nhận xét