WEBVTT

00:00:00.166 --> 00:00:01.084
Sara, welcome.

00:00:01.584 --> 00:00:04.754
In your view, where are we
seeing digital approaches having

00:00:04.754 --> 00:00:06.965
the most impact in
health care delivery?

00:00:07.257 --> 00:00:11.302
So I work with a provider, and I
think there's certainly, you

00:00:11.302 --> 00:00:13.555
know, we actually
have 5A's of AI.

00:00:13.805 --> 00:00:17.225
We talk about assisting,
augmenting, automating,

00:00:17.225 --> 00:00:19.227
accelerating, and amplifying.

00:00:19.227 --> 00:00:22.689
And I think those all apply in
our scenarios.

00:00:23.314 --> 00:00:26.818
We've seen the most amount of, I
would say, impact on the

00:00:26.818 --> 00:00:29.946
clinician experience and on the
patient experience.

00:00:30.030 --> 00:00:32.824
And there's, of course, a lot of
overlap between the two, and

00:00:32.824 --> 00:00:34.034
they intersect quite a lot.

00:00:34.451 --> 00:00:35.702
A couple of examples.

00:00:36.494 --> 00:00:39.539
We are seeing a tremendous
amount of potential as it

00:00:39.539 --> 00:00:42.834
relates to digital to drive a
better consumer experience

00:00:42.834 --> 00:00:46.129
around what our Providence
promise is: which is know me,

00:00:46.129 --> 00:00:47.547
care for me, ease my way.

00:00:47.756 --> 00:00:50.508
So number one: know
who your users are.

00:00:50.508 --> 00:00:52.677
It's really critical to know who
your users are.

00:00:52.677 --> 00:00:55.847
And we've done a pretty poor job

00:00:55.847 --> 00:00:57.640
of that in health care historically.

00:00:57.640 --> 00:00:58.808
It's very episodic.

00:00:58.808 --> 00:01:01.644
It's very, like, you got sick,
you came in, that's all we know

00:01:01.644 --> 00:01:02.103
about you.

00:01:02.353 --> 00:01:05.774
But now, digital has the
potential, in particular through

00:01:05.774 --> 00:01:09.027
new identity approaches,
approaches that are much more

00:01:09.027 --> 00:01:12.655
cohesive across a full digital
ecosystem--not just limited to

00:01:12.655 --> 00:01:14.199
one specific organization.

00:01:14.783 --> 00:01:18.328
We can actually know people
across all the things that

00:01:18.328 --> 00:01:22.123
matter to them and then deliver
personalized experiences.

00:01:22.415 --> 00:01:26.002
So I think that is one really
tremendous area that we're going

00:01:26.002 --> 00:01:30.215
to see a lot of movement on in
the next couple of years.

00:01:30.715 --> 00:01:33.885
The other big areas are on
clinician experience.

00:01:33.885 --> 00:01:38.431
So I think these are still kind
of low-hanging fruit, a lot of

00:01:38.431 --> 00:01:42.519
basic blocking and tackling as
it relates to things like

00:01:42.519 --> 00:01:46.356
Ambient--support for
documentation--and charting for

00:01:46.356 --> 00:01:48.650
clinicians in basket management.

00:01:48.650 --> 00:01:51.736
If you're not familiar with the
concept of the in basket, it's

00:01:51.736 --> 00:01:53.571
basically a physicians
e-mail inbox.

00:01:53.571 --> 00:01:57.450
It's their work queue that's in
Epic, and they have to touch

00:01:57.450 --> 00:02:01.246
thousands and thousands of
messages, which is grinding away

00:02:01.246 --> 00:02:03.832
at physicians and APPs and

00:02:03.832 --> 00:02:05.250
making their lives really difficult.

00:02:05.250 --> 00:02:09.254
So actually understanding the
context of messages, both in the

00:02:09.254 --> 00:02:12.757
clinical context--sort of more
generally as to what an

00:02:12.757 --> 00:02:16.845
appropriate action would be--but
also that specific patient and

00:02:16.845 --> 00:02:20.098
then supporting them in driving
a specific action.

00:02:20.098 --> 00:02:23.726
So similar to the consumer
personalization side, but much

00:02:23.726 --> 00:02:26.062
more focused on
physicians and APPs.

00:02:26.062 --> 00:02:29.107
And I think all of those
things, you'll notice similar

00:02:29.107 --> 00:02:29.858
themes, right?

00:02:30.400 --> 00:02:35.280
Know what the context is,
deliver insights, and then be

00:02:35.280 --> 00:02:37.031
able to take action.

00:02:37.282 --> 00:02:40.618
And those common themes are
going to become more and more

00:02:40.618 --> 00:02:43.413
widespread with the improvements
in technology.

00:02:44.080 --> 00:02:48.042
Last thing I'll say is a bit of
a caution and thing that I worry

00:02:48.042 --> 00:02:51.713
about is that over the last
eight years or so, we've seen a

00:02:51.713 --> 00:02:53.882
massive proliferation
of solutions.

00:02:54.340 --> 00:02:58.428
And I know we'll talk a little
bit about the challenges, but

00:02:58.428 --> 00:03:02.515
that we can't repeat that same
pattern because providers are

00:03:02.515 --> 00:03:03.850
already overwhelmed.

00:03:04.142 --> 00:03:07.270
It's just we should learn from
our mistakes.

00:03:07.270 --> 00:03:10.857
What do you see as the main
stumbling box to realizing those

00:03:10.857 --> 00:03:13.484
potentials for both
patients and clinicians?

00:03:13.985 --> 00:03:18.948
The difficulty in separating out
signal from noise,

00:03:18.948 --> 00:03:19.699
The difficulty in separating out
signal from noise,

00:03:19.699 --> 00:03:22.744
especially with GenAI solutions

00:03:22.744 --> 00:03:26.122
or LLM--large-language
model--powered solutions

00:03:26.122 --> 00:03:26.998
is very real.

00:03:27.373 --> 00:03:30.585
And the reason for that
proliferation of solutions and

00:03:30.585 --> 00:03:33.254
the difficulty of
differentiating signal from

00:03:33.254 --> 00:03:35.757
noise is because it's actually
gotten really easy

00:03:35.757 --> 00:03:37.467
to build applications.

00:03:37.842 --> 00:03:41.721
It's like I could do it, and I'm
not a technologist, right?

00:03:42.639 --> 00:03:46.184
There are so many tools out
there for building applications,

00:03:46.184 --> 00:03:49.103
but then how do they integrate
into the workflow?

00:03:49.103 --> 00:03:51.397
How do they
integrate into the

00:03:51.397 --> 00:03:53.441
underlying technical infrastructure?

00:03:53.566 --> 00:03:56.277
How do you monitor them on an
ongoing basis to make sure

00:03:56.277 --> 00:03:59.239
they're doing the thing that you
want them to do and that you

00:03:59.239 --> 00:04:00.448
think that they're doing?

00:04:00.448 --> 00:04:01.574
Are they actually working?

00:04:02.533 --> 00:04:05.036
And by the way, do they share
the data?

00:04:05.036 --> 00:04:07.747
Do you have the visibility to
even be able to do monitoring in

00:04:07.747 --> 00:04:08.456
the first place?

00:04:08.915 --> 00:04:11.834
So those are some of the big
challenges that we're seeing:

00:04:11.834 --> 00:04:12.919
massive proliferation.

00:04:12.961 --> 00:04:18.800
And so it's just a massive sea
change in how easy it is to

00:04:18.800 --> 00:04:20.426
build something.

00:04:20.593 --> 00:04:22.011
But then what's the value?

00:04:22.011 --> 00:04:22.971
What are you getting for it?

00:04:24.180 --> 00:04:27.183
Finally, what are some of the
trends, the innovations that

00:04:27.183 --> 00:04:29.519
you're most excited about for
the year ahead?

00:04:30.186 --> 00:04:32.063
I'm very excited
about a few things.

00:04:32.272 --> 00:04:35.733
One, and it sounds like probably
the least exciting thing, but I

00:04:35.733 --> 00:04:38.111
love infrastructure,
and I think it has so

00:04:38.111 --> 00:04:39.487
much potential to transform.

00:04:39.487 --> 00:04:43.491
health care  . . . is everything
that's being done in monitoring.

00:04:43.866 --> 00:04:49.289
What do you need to do to test
and validate a model from an AI

00:04:49.289 --> 00:04:52.333
context before you
ever deploy it?

00:04:52.625 --> 00:04:55.586
And at what point is it a
snapshot in time?

00:04:56.254 --> 00:04:59.674
The model, the second you
release that model out into the

00:04:59.674 --> 00:05:01.175
wild it starts to change.

00:05:01.301 --> 00:05:05.138
And so how we think about
monitoring from a predeployment

00:05:05.138 --> 00:05:09.225
perspective and then from a
postdeployment perspective, it's

00:05:09.225 --> 00:05:11.978
. . . there's things called
model drift.

00:05:12.729 --> 00:05:17.066
You need to ensure, you know,
all of these models, the

00:05:17.066 --> 00:05:21.654
outcomes, the outputs are what
we call probabilistic,

00:05:21.654 --> 00:05:22.780
not deterministic.

00:05:22.780 --> 00:05:26.117
So you don't get a single output
for the input that is always

00:05:26.117 --> 00:05:27.243
going to be the same.

00:05:27.452 --> 00:05:30.413
There's virtually an infinite
number of possibilities

00:05:30.413 --> 00:05:31.080
within a range.

00:05:31.497 --> 00:05:35.168
And so how you do that work is,
I think, very interesting is

00:05:35.168 --> 00:05:38.713
very interesting technical
problem, how you do governance

00:05:38.713 --> 00:05:42.425
around it from a data, you know,
training set perspective, a

00:05:42.425 --> 00:05:43.801
provenance perspective.

00:05:44.552 --> 00:05:48.848
I think all of that is extremely
exciting, even though it's the

00:05:48.848 --> 00:05:51.351
the sort of nerdy
technical stuff.

00:05:51.559 --> 00:05:55.605
But what it really like the
output of that is if we don't do

00:05:55.605 --> 00:05:59.609
that stuff, we're not going to
responsibly deploy AI, and we

00:05:59.609 --> 00:06:03.863
will be held to account with our
patients, with our clinicians,

00:06:03.863 --> 00:06:09.452
in the court of public opinion,
even at the state and

00:06:09.452 --> 00:06:11.662
federal regulatory level, even
with some themes around

00:06:11.662 --> 00:06:13.956
deregulation at
the federal level.

00:06:13.956 --> 00:06:16.626
So I think we have to do that,
and I'm very excited about it.

00:06:16.876 --> 00:06:21.714
And then ultimately, I do think
that this is a major inflection

00:06:21.714 --> 00:06:26.302
point as it relates to being
able to achieve the quad aim in

00:06:26.302 --> 00:06:27.220
health care.

00:06:27.845 --> 00:06:30.598
Bringing down costs has been
really hard, right?

00:06:30.640 --> 00:06:34.560
It's because complexity has
increased so much. And we can

00:06:34.560 --> 00:06:39.816
actually take complexity out and
reimagine the way that health

00:06:39.816 --> 00:06:41.442
care is administered,

00:06:41.442 --> 00:06:42.944
in addition to the
way it's delivered.

00:06:42.944 --> 00:06:46.364
We often talk about, oh, you
know, you know, these sort of

00:06:46.364 --> 00:06:49.951
things that influence clinical
delivery, but I think that the

00:06:49.951 --> 00:06:53.329
other stuff actually is really
powerful too--just how you

00:06:53.329 --> 00:06:57.041
administer a system that is so
complex and make it really easy?

00:06:57.583 --> 00:07:01.421
And I think that AI might allow
us to see new things as well:

00:07:01.421 --> 00:07:07.593
new patterns, new ways of
delivering without, you know

00:07:07.593 --> 00:07:09.846
one of the challenges have of
technology has always been,

00:07:09.846 --> 00:07:11.222
well, you need
standardization first.

00:07:11.681 --> 00:07:15.935
I think that we can see new ways
of delivering care in a hospital

00:07:15.935 --> 00:07:21.190
with and it just administration
of care as well with GenAI.

00:07:21.190 --> 00:07:22.859
Sara, can you give me an example

00:07:22.859 --> 00:07:24.318
of how this benefits
patient outcome?

00:07:24.986 --> 00:07:25.945
Yes, absolutely.

00:07:25.945 --> 00:07:29.282
So, think about you as a user.

00:07:29.282 --> 00:07:31.117
You're different from
me as a user.

00:07:31.117 --> 00:07:33.453
We are, we live in different
parts of the world.

00:07:33.619 --> 00:07:35.997
We have different families.

00:07:35.997 --> 00:07:40.001
We have different clinical needs
and probably different social

00:07:40.001 --> 00:07:43.880
situations as well, like you
might have, you might live in an

00:07:43.880 --> 00:07:46.799
area with public transportation,
I might not.

00:07:47.425 --> 00:07:51.637
And so when I talked
about knowing the user, for

00:07:51.637 --> 00:07:56.017
instance, think about if you
logged into your experience, it

00:07:56.017 --> 00:07:59.604
was tailored to you, your
clinical provider, your

00:07:59.604 --> 00:08:01.314
physician is at the top.

00:08:01.481 --> 00:08:04.525
The actions that you
specifically need to take. Let's

00:08:04.525 --> 00:08:08.029
say you need to go in and get
immunized, and I need to go and

00:08:08.029 --> 00:08:10.406
do a mammogram. We’re
different people.

00:08:10.698 --> 00:08:14.494
And that one thing that you have
to do is tailored specifically

00:08:14.494 --> 00:08:18.664
to you and delivered
specifically to you versus what

00:08:18.664 --> 00:08:21.417
I need to do is tailor specific
to me and deliver specifically

00:08:21.417 --> 00:08:23.503
to me and in the way
that we want.

00:08:23.503 --> 00:08:27.131
So you might want to access via
mobile, I might want to access

00:08:27.131 --> 00:08:29.759
via web, or I want someone to
text it to me.

00:08:30.218 --> 00:08:35.348
And so knowing those individual
things drives better outcomes.

00:08:35.556 --> 00:08:38.935
And what we know today is that
when we have personalized

00:08:38.935 --> 00:08:43.397
insights that are delivered to
our patients that it's just the

00:08:43.397 --> 00:08:44.690
next thing you have to do.

00:08:44.690 --> 00:08:47.276
It's not the long list, the 25
things you have to do.

00:08:47.276 --> 00:08:50.530
It's just the one thing that we
drive, what's in very sort of

00:08:50.530 --> 00:08:52.657
plain terms is called
better conversion.

00:08:52.990 --> 00:08:56.494
We know that folks will engage
and when they engage their

00:08:56.494 --> 00:08:57.745
outcomes are better.

00:08:58.204 --> 00:09:02.041
So that is a really plain
example of people getting the

00:09:02.041 --> 00:09:07.213
for instance, the preventive
care that they need just because

00:09:07.213 --> 00:09:11.092
it was made easy for them to
understand and presented to them

00:09:11.092 --> 00:09:11.676
where they are.

00:09:12.093 --> 00:09:13.553
Sara, thank you so much.

00:09:13.719 --> 00:09:14.095
Thank you.