Latest / Becker’s Healthcare -- Pediatric Leadership Podcast / Scaling AI Innovation at Seattle Children's with Dr. Zafar Chaudry

Transcript

Scaling AI Innovation at Seattle Children's with Dr. Zafar Chaudry · Becker’s Healthcare -- Pediatric Leadership Podcast

0:01
This is where health care leadership comes together.
0:04
Becker's sixteenth annual meeting brings more than 3,500
0:07
hospital and health system executives and nearly 800
0:10
speakers to Chicago
0:12
April.
0:14
This year's event includes keynote conversations with Dallas
0:17
Cowboys legend Troy Aikman and former president George
0:20
w Bush. For the agenda and event details,
0:22
visit beckershospitalreview.com
0:25
and click on the events tab in the
0:26
upper right. We're looking forward to hosting you
0:29
in Chicago.
0:31
This is Laura Dirda with the Becker's Healthcare
0:34
podcast. I'm thrilled today to be joined by
0:36
doctor Zafar Chowdhury, senior vice president and chief
0:38
digital officer as well as chief AI and
0:40
information officer at Seattle Children's.
0:43
Doctor Chowdhury, thank you so much for being
0:44
here today. Thanks for having me. My pleasure.
0:48
Absolutely. Well, I know we're gonna talk a
0:50
lot about technology and AI and how you're
0:52
looking at the future. But before we dive
0:54
into that, can you tell us a little
0:55
bit more about yourself and Seattle Children's?
0:58
Yeah. So
0:59
I've been running the IT shop here at
1:01
Seattle Children's. It's been about eight and a
1:04
half years,
1:05
but I've been in the health care IT
1:07
industry.
1:09
This is year 40, and I actually started
1:11
my journey
1:12
in health,
1:14
as a physician. So I'm
1:16
one of those folks that has defected to
1:18
the dark side over many years and now
1:21
fully embedded
1:22
on the technology side.
1:25
Absolutely. That's, fantastic to hear. And, you know,
1:27
I know you've got a lot going on,
1:30
especially when it comes to AI and and
1:32
some really advanced things that you're doing there
1:33
at Seattle Children's. And so I'm excited to
1:35
dig in. Can you tell us a little
1:37
bit more about some of the things you've
1:38
been doing in the last year? What's an
1:40
initiative you're most proud of? What did you
1:42
do, and what were the results?
1:44
Yeah. So I think, you know, AI,
1:46
obviously,
1:48
2026,
1:48
huge buzz around AI. We've been doing AI
1:52
for quite a few years now, and it
1:54
just
1:55
seems to get better with the more data
1:58
you feed into the engine.
2:00
So about eighteen months ago, we started a
2:02
refresh of
2:04
our data warehouse. So the question was,
2:07
how do you get the right information into
2:09
the right place? Because that's the information you're
2:11
going to need
2:13
to build your AI models
2:15
so that they're accurate and are beneficial to
2:18
an organization.
2:19
So we moved from an older platform, which
2:22
was IBM Netezza on prem
2:24
to a full Google Cloud platform
2:27
for all of our data analytics.
2:29
And
2:30
our data lake is fed with a 171
2:34
systems
2:35
at present, and that's where all of our
2:37
data resides. And then from that data,
2:41
we've been working with stakeholders. So I truly
2:43
believe that the use cases for AI are
2:46
focused
2:47
on what do the end users need, not
2:49
necessarily what I think they need. And so
2:52
one of our most successful programs, we use
2:56
that Google platform. We use Gemini
2:59
AI. We use Vertex AI.
3:01
So, of course, we did
3:04
the training mandatory for everyone in AI in
3:07
our organization.
3:08
We set up a AI
3:10
user policy.
3:11
We have an AI review board that reviews
3:14
all the intake
3:16
for use cases that
3:18
our stakeholders are asking for.
3:20
And then out of that, what came
3:23
about was our first agentic AI model, which
3:27
we call Pathway Assistant. And what we did
3:29
is we took
3:31
many, many, many years' worth of knowledge
3:34
around clinical pathways
3:36
to treat
3:37
kids
3:38
that we had developed.
3:40
And we actually
3:41
freely give those pathways away to hospitals that
3:44
want to use them around the world.
3:46
And we took those pathways,
3:48
six, seven thousand pages of PDFs
3:52
validated by clinicians over the years,
3:54
and we fed it into an agent, which
3:57
we call Pathway Assistant.
3:59
And that agent was able to amalgamate that
4:02
information
4:03
and then create an AgenTiKa
4:05
agent that allows people
4:08
to query it and question it and have
4:10
a dialogue with it. So if you're a
4:13
a new physician and you're trying to learn
4:15
how to practice medicine in pediatrics,
4:18
you can tell the agent
4:20
about your patient, the age, the weight, the
4:23
symptoms,
4:24
what's going on with the patient. And the
4:26
agent will look at all the information it
4:29
has access to and start guiding that physician
4:32
through the care pathway
4:34
for that patient. It will tell
4:36
the physician
4:37
what to do next, what tests to run,
4:41
ask for the results of the test, then
4:43
step through the diagnosis, the differential,
4:46
the treatment plan.
4:48
That helps
4:49
newly trained physicians learn how to practice.
4:52
For
4:53
seasoned physicians,
4:55
it's a validation
4:56
tool. It makes sure that they
4:58
are doing the right thing for the right
5:01
diagnosis, the right condition. And that's been very,
5:04
very accurate
5:06
and very, very successful.
5:08
And that that that then becomes your friend,
5:10
you know, that phone a friend thing, and
5:13
you're able to query it and get the
5:15
answers that you need. So that's one example
5:18
of an agentic AI that we built. Another
5:20
example,
5:22
is a translation
5:23
assistant.
5:24
So we have a patient population that is
5:27
varied.
5:28
We tested
5:30
English to Spanish translation
5:32
of instructions,
5:34
discharge instructions, etcetera, post op instructions.
5:37
That's been hugely successful.
5:40
We did find, though, that
5:42
AI agents translating to other languages
5:46
like Ukrainian
5:47
and Somali word as accurate, and so we
5:50
didn't pursue
5:52
other languages. So we're doing English to Spanish
5:54
and Spanish to English.
5:56
That's another example,
5:58
of a use case that has been valuable
6:01
not only to our clinical staff but also
6:04
to our patient population. So those are some
6:07
of the things we've been working on. There's
6:09
many more. Of course, there's more in the
6:11
pipeline
6:12
than we can probably afford to do or
6:15
our resource to do in one year, but
6:17
we're stepping through multiple
6:20
use cases. The third one I'll throw out
6:23
there is Ambient AI.
6:25
We've agreed we've
6:27
set up a partnership with the bridge, and
6:28
we have launched the Ambient AI transcription
6:33
for all of our physicians. We bought an
6:34
enterprise license, and that's currently being rolled out
6:38
with huge success,
6:40
great feedback from our clinicians in the test
6:42
phase.
6:43
Many tell us that they would never allow
6:46
us to take this tool away from them.
6:47
So that's another win that we've had in
6:50
the AI space.
6:52
That's amazing to hear. And I think especially
6:54
really cool to understand how you're using AI,
6:57
looking at AI agents,
6:58
looking at some of the diagnoses and validations,
7:01
as well as then getting into, the the,
7:06
translation as well. And, of course, you know,
7:08
ambient technology is something that has become so
7:11
much more prevalent, especially in the last twelve
7:13
months, within many organizations. And echoing those same
7:16
sentiments that clinicians are are finding it,
7:19
extremely beneficial in in time savings and and,
7:23
also for their relationships with patients and satisfaction
7:26
with their jobs overall. So I think that's
7:28
amazing to hear. And, you know,
7:30
my thing I'm curious about is for any
7:32
one of these use cases, whether it's Ambient
7:34
or some of the agents and more, how
7:37
do you make sure that they are continuing
7:39
to work at the level that they're supposed
7:41
to? How do you measure that success and
7:43
then, continue monitoring them, to make sure that
7:46
there aren't unwanted changes as time goes on.
7:50
So the success with these use cases has
7:52
to come from the partnership
7:54
between IT
7:55
and the clinical teams. So
7:57
we may be the folks that engineer the
8:00
back end of these tools and these agents,
8:04
but on the front end, there are many
8:06
doctors involved. So for our pathway assistant,
8:09
we had 50 physicians work on this project,
8:12
each of which dedicated about one hundred hours
8:15
of their time.
8:16
So that was in the build, design,
8:19
test,
8:20
post test phase.
8:23
And the same group of physicians
8:25
predominantly in my informatics team
8:28
are always
8:29
testing and checking. If new pathways are added,
8:33
those pathways are validated,
8:35
tested, torture tested, and then released into the
8:39
agent. So it's a it's a continuous PDSA
8:42
cycle.
8:43
But you don't do that with IT people.
8:45
You do that with the people that use
8:47
the tool on a day in, day out
8:49
basis.
8:50
And the people
8:51
are volunteering their time because they see the
8:54
value of the tool for them in their
8:56
day to day work. And so that's the
8:59
partnership you really have to build.
9:01
Because if it's only driven by IT because
9:04
it's the next shiny toy, you're not going
9:06
to get
9:07
adoption and you're certainly not going to get
9:09
the most important thing in AI, which is
9:12
sustainability
9:12
of solutions.
9:14
Because it's easy to create something shiny and
9:16
then people
9:18
won't use it after a few weeks. It's
9:20
hard to get them to continually use it,
9:23
provide feedback.
9:25
If there's technical tweaks, the team can do
9:27
any technical tweaks, but it is a PDSA
9:30
cycle.
9:32
Got it. That makes a lot of sense.
9:33
And, you know, it is really helpful context
9:35
for how these different,
9:36
teams and departments work together in order to
9:39
make sure that the technology is functioning in
9:41
a the most beneficial way possible.
9:44
Now looking ahead into 2026, what are some
9:46
of the big priorities and headwinds that you're
9:48
most focused on?
9:50
So in 2026,
9:51
of course, we've got to make sure that
9:54
the costs
9:56
that we have
9:57
in providing
9:58
the support services and IT
10:01
are optimized and continue to remain optimized.
10:04
We're going to continue to do some application
10:07
rationalization
10:08
because
10:09
we have a lot of different systems, and
10:12
we need to validate whether those systems are
10:14
still off use in 2026.
10:18
We have consolidated
10:20
and outsourced
10:22
most recently all of our infrastructure.
10:25
So all of our servers, storage,
10:27
backup,
10:28
disaster recovery
10:30
for all of our systems is now a
10:32
service contract
10:33
that we don't run that's outsourced to Rackspace.
10:37
And that's been very successful because that helps
10:39
us deal with our technology
10:41
debt
10:42
in the future and also allows us to
10:44
have service level based,
10:47
infrastructure, which is important if you're trying to
10:49
grow in the AI space because the demands
10:52
on infrastructure won't go down. They will continue
10:54
to go up.
10:56
And at the same time, I like to,
10:59
almost every year, take a look at my
11:01
organizational structure in IT
11:04
and make sure that that is still optimized
11:06
and fits the needs
11:08
of the customers I serve, which are the
11:10
end users,
11:11
the clinicians, the patients, the parents, caregivers.
11:15
We'll also continue to build tools,
11:19
in our patient facing app.
11:21
We have a patient and parent advisory group.
11:23
We meet with them monthly,
11:26
and they tell us the things that we
11:28
should build, that they need, that to help
11:30
them navigate our health system. So we'll continue
11:32
to innovate
11:33
in the digital space
11:36
at the front end and that will also
11:38
continue this year. And the other thing we'll
11:41
do is we'll continue to track any regulatory
11:44
changes,
11:45
federal changes that impact us.
11:48
The thing that will always keep us up
11:50
at night moving through 2026
11:52
will be cybersecurity.
11:54
So, we will continue to keep
11:56
our eye on that
11:58
and also
11:59
continue to optimize tools in that space. There
12:02
is also use of AI
12:04
around cybersecurity
12:05
to protect our organization,
12:07
and we'll continue to
12:09
educate and train our end users in the
12:12
cybersecurity
12:13
risks and pitfalls
12:15
that are very easy to fall into.
12:19
That's helpful to understand. You know, I I
12:21
know so many different areas that,
12:24
you touched on are are so critical, but,
12:26
it it was cool to hear you talking
12:29
about some of the different,
12:31
things that you're doing in order to meet
12:33
the needs of of patients,
12:35
especially,
12:36
you know, with some of the the feedback
12:38
you're getting in the advisory group.
12:40
Is there anything you've been able to build
12:42
because of their feedback that, you know, has
12:44
been particularly successful? Any success stories you can
12:47
share with us? Yeah. So I think one
12:49
of the really interesting ones was
12:53
around surgical notification. So the the use case
12:57
was that when,
12:59
a patient goes into surgery,
13:01
the parent usually
13:04
accompanies their child
13:06
and then waits for their child whilst the
13:08
child is in surgery. And as you know,
13:11
there are different types of surgeries. Some are
13:14
short, some are extremely long. But when you
13:16
are a worried patient,
13:19
you don't necessarily know if you should leave
13:22
the surgical suite because you don't know when
13:24
your child is coming out of surgery
13:27
or if there are any updates about your
13:29
child in surgery.
13:30
So the feedback we had from our patients
13:33
was the parents of our patients was that
13:36
I can't leave the Surgical Suite and sometimes
13:39
my kid's in surgery for four, five, six,
13:42
seven hours
13:43
and I'm afraid to leave
13:44
because something might happen and I wouldn't know
13:47
because they wouldn't be able to find me.
13:49
And what that means for me is I
13:50
don't even have the opportunity to go and
13:52
get a coffee or even get something to
13:55
eat.
13:55
So could you build us something
13:57
that could keep us informed and then allow
14:00
us to not be stuck in one place?
14:03
Within our digital front door app, we added
14:06
a component
14:07
that actually
14:09
keeps that parent or caregiver
14:12
updated
14:12
on where their child is in in what
14:16
stage in the surgical suite.
14:18
And, it's code based. So we issue them
14:20
a code. They can see on their app
14:22
or even on the the big board in
14:25
surgery, and it will tell them whether the
14:27
child has been
14:28
taken in,
14:29
is in intake, is in surgery,
14:32
is in post op, is in recovery,
14:35
where they are exactly,
14:36
which then frees the parent or caregiver
14:40
to go out and, yeah, get that small
14:42
sandwich to eat so that they can continue
14:45
to
14:47
not be anxious,
14:48
and then come back in time to
14:51
welcome their child back from surgery. So that's
14:53
a really small thing,
14:55
but it's a big thing to someone who
14:56
is
14:57
super worried about their child and doesn't wanna
15:00
move from one location.
15:03
Absolutely. That's such a powerful example of the
15:05
ways that, you know, technology,
15:07
can provide
15:09
a big difference within the the care journey
15:11
and, you know, truly, truly,
15:14
amazing, I can imagine, for the families to
15:16
be able to have, you know, that peace
15:18
of mind when a child is going through
15:20
surgery, which is huge.
15:22
Now I'm curious. What do you think the
15:24
hardest thing you'll have to do in the
15:26
coming year will be?
15:28
I think the hardest thing for us in
15:30
'26
15:31
will be
15:33
deciphering the impact on our health system around
15:37
the big bill
15:38
that was approved.
15:39
And then for me in IT, trying to
15:42
figure out what that means financially to my
15:45
budget.
15:46
Because I have to right size my budget,
15:49
balance my books, and still continue to provide
15:52
the services
15:53
that our stakeholders and our patients and our
15:56
parents need.
15:57
And that's not always easy in a climate
16:00
when
16:01
costs are actually going up. You know, as
16:04
you know,
16:05
we've been talking about AI today, but those
16:08
tools cost money.
16:09
Infrastructure
16:10
costs money.
16:11
Labor and skill sets cost money, and those
16:14
costs aren't coming down. So
16:17
this year, we'll have to be very, very
16:19
careful around
16:21
continuing to innovate, but also continuing
16:24
to do more for less and be fiscally
16:27
responsible because we need to do that so
16:30
that we can keep our lights on and
16:31
doors open.
16:33
That's such a good point. And I think
16:35
especially when you look at some of the
16:37
financial pressures, I know every dollar in health
16:39
care is precious and,
16:41
investing in technology, while critical, especially when you're
16:44
trying to make sure you've got the right
16:46
experience
16:46
in quality and efficiencies,
16:49
you know, certainly requires an amount of,
16:52
value that it has to bring back into
16:53
the system. And so when you're making these
16:56
decisions,
16:58
and having that fiscal responsibility
17:00
in mind,
17:01
what do you see as being rising to
17:03
the top, I guess, for your,
17:05
the next year? And and how do you
17:07
work with different companies and vendors in your
17:10
team as well, to make sure that they're
17:12
making the right decisions,
17:14
given those limited resources?
17:17
Yeah. So I think the most important thing
17:18
for any health IT shop is to validate
17:22
everything that they are actually running on a
17:24
daily basis. You know, we have,
17:27
over a thousand different applications that we support
17:30
and run at Seattle Children's. And the question
17:32
I always ask my team
17:34
is, are we sure those are the right
17:37
tools
17:38
and there are the right software applications that
17:40
we should continue to use?
17:42
And that's not a question we answer. We
17:44
have to go and ask the stakeholders who
17:46
use those tools.
17:48
If there are tools that really aren't being
17:50
used or can be archived, can be deprecated,
17:54
can be consolidated
17:56
into newer tools that we've purchased,
17:58
then we need to undertake that exercise.
18:01
Rationalization
18:03
year on year is always key, and you
18:06
do that rationalization
18:07
in partnership
18:08
with the actual users
18:11
of those systems.
18:12
Also in parallel,
18:15
we're going to have many, many
18:17
tools and subscriptions
18:18
come up this year as we do every
18:21
year. And we will we have a vendor
18:23
management team,
18:25
in IT. We work with our supply chain,
18:28
and we always
18:31
fight back. So when a renewal comes through,
18:33
we look at the renewal. We see how
18:35
much it has gone up. We then go
18:37
back and forth and fight the good fight
18:39
to make sure that
18:41
the increase isn't too taxing
18:43
on our budget budgetary model. And so that's
18:47
I would say sometimes that's the fun part
18:49
of going back and forth and arguing till
18:51
you get to something that's amicable
18:53
for both sides because
18:55
vendor relationships are partnerships.
18:57
So we do that consistently, and I have
18:59
a team that pushes back on renewals and
19:02
make sure that we're renewing at the right
19:04
rate so that we can actually afford it.
19:07
And then
19:09
going through the health system and making sure
19:11
that we've inventoried all of our assets,
19:14
Asset management can sometimes be a problem.
19:17
Make sure that we have,
19:19
taken in all the equipment that needs to
19:21
be replaced and making sure that
19:24
any equipment that's end of life is either
19:27
deprecated or replaced.
19:29
And if you continually do that, then that
19:31
you'll find tools,
19:33
pieces of hardware
19:35
that you don't need. And if you can
19:37
decommission those pieces of hardware, that also ends
19:40
up saving money because you don't renew
19:43
on maintenance and support.
19:45
So then you have to do a good
19:46
job around your IT asset management portfolio.
19:49
Otherwise, you'll have
19:51
massive amounts of hardware sprawl that becomes unmanageable.
19:54
So the team will also spend the year
19:57
going through all of those lists and making
20:00
sure that,
20:01
firstly, we can find everything, and secondly,
20:03
making sure everything should stay in place or
20:06
be deprecated.
20:08
That's helpful to understand. Thank you for laying
20:10
out that process so clearly, and I think
20:11
it goes back to something else you mentioned
20:13
at the beginning of our conversation looking at,
20:15
you know, decommissioning those things that aren't in
20:19
use anymore in in managing that technical debt
20:21
in a different way.
20:22
So, especially, when you think about that technical
20:25
debt, I mean, how does your team approach,
20:27
those conversations or really efficiently
20:30
managing some of those things and so you
20:31
can cut back on on what, isn't in
20:34
use?
20:35
Yeah. So I think the interesting thing about
20:38
working in health and health IT is
20:41
a lot of the work that we do
20:42
isn't technical. It's very much communication.
20:46
So once you've
20:47
identified
20:48
assets,
20:50
the next step is to go and validate
20:52
the asset with the person
20:54
who has the asset or who we think
20:56
has the asset
20:57
and then validate whether that use case that
21:00
they have the asset for is still a
21:02
valid use case or not. Now we can
21:04
take the example of laptops.
21:06
How many times do you find out that
21:08
the user
21:09
has been issued more than one laptop in
21:11
your organization?
21:13
And then if they have, go and validate
21:15
whether they truly have two or three
21:18
or whether or not two of them are
21:20
no longer functioning and they never gave it
21:22
back to us for decommissioning.
21:24
So it's it's it's sort of
21:26
a detective work
21:28
in addition to lots of communication.
21:31
And my job is to communicate to the
21:34
health system
21:35
about all these different initiatives that we have
21:38
and ask
21:39
my user community for help because we can't
21:43
do it alone. So sometimes we have amnesty
21:46
programs. Hey. If you have any of our
21:48
equipment that you didn't give back, let us
21:50
know. And people come forward and say, yeah.
21:52
I forgot to tell you this wasn't working.
21:55
I'll hand it back, and then we can
21:56
decommission it. So
21:58
my job is to communicate at system level
22:00
the different initiatives we have. Then the team
22:03
on the ground will go and interact
22:06
round in our different facilities,
22:10
look at things, find things,
22:12
either refresh things or
22:14
deprecate them altogether. And that's a that's a
22:17
continuous
22:18
cycle that we will undertake
22:20
during the year. Because the more you can
22:23
communicate
22:24
your challenges to your end users, the more
22:27
helpful they can actually be in
22:30
partnering with you to
22:33
to deal with the challenges ahead.
22:36
I love that. Thank you again for digging
22:38
a bit deeper there. Now before we wrap
22:40
up, I wanted to ask you about growth.
22:42
Where do you see some of the best
22:43
opportunities for organizational growth in the next couple
22:45
of years?
22:47
Well, certainly for us in pediatrics,
22:49
we are seeing growth,
22:51
in terms of patient demand
22:53
for our services.
22:54
So we will definitely be expanding
22:57
across the region
22:58
that we serve in terms of
23:01
access to clinics,
23:03
access to out of hours services
23:05
that our
23:06
parents and kids need,
23:08
and, you know,
23:10
more streamlined ways in getting
23:13
the appointments
23:14
and set secondary and tertiary care
23:18
that parents need
23:19
for their kiddos in the Pacific Northwest. So
23:22
it's a multipronged approach.
23:24
Some growth, some new clinics, some new ambulatory
23:28
centers,
23:28
and we will continue to do that through
23:30
'26
23:31
so that we can meet the growing capacity
23:34
that we have here in the Pacific Northwest.
23:37
I love that. Thank you so much, doctor
23:39
Chowdhury, for everything that, you have talked through
23:42
with us today. This has been such a
23:44
fascinating and valuable conversation. And, you know, I'm
23:46
excited to see you as well at our
23:48
annual meeting. I know you'll be,
23:50
there speaking on a panel and just, you
23:52
know, always excited to have you as part
23:54
of what we're doing here at Becker. So
23:56
thank you. Well, thank you. Thank you for
23:58
spending some time with me. Always a pleasure
24:00
pleasure to chat with you, Laura.