Latest / Becker’s Healthcare Digital Health + Health IT / Michael Hasselberg, PhD, Chief Digital Health Officer at the University of Rochester Medical Center

Transcript

Michael Hasselberg, PhD, Chief Digital Health Officer at the University of Rochester Medical Center · Becker’s Healthcare Digital Health + Health IT

0:00
Exciting things are happening at Becker's Healthcare. Stay
0:04
ahead of industry trends with the new Becker's
0:06
CFO plus Revenue Cycle podcast,
0:08
your go to source for insights from top
0:11
healthcare finance leaders. Tune in wherever you get
0:14
your podcasts.
0:15
And don't miss the tenth annual Health IT
0:18
plus Digital Health plus RCM conference, happening September
0:22
30 to 10/03/2025
0:24
in Chicago.
0:25
Join thousands of executives,
0:27
engage with industry leaders, and explore the future
0:30
of health care innovation.
0:32
Learn more about our upcoming events at beckershospitalreview.com.
0:36
See you there.
0:38
This is Laura Dierda with the Becker's Healthcare
0:40
Podcast. I'm thrilled today to be joined by
0:42
Michael Hasselberg, chief digital health officer at UR
0:46
Medicine and professor of psychiatry, clinical nursing, and
0:49
data science at the University of Rochester. Michael,
0:51
it's a pleasure to have you on the
0:52
podcast today.
0:54
Thank you for having me, Laura.
0:55
Looking forward to our discussion.
0:58
Absolutely. I am too. I I know it's,
0:59
you know, got a lot going on there
1:01
at UR Medicine, and and certainly something that,
1:04
we'd love to hear more about how you're
1:06
thinking about the future as well. But before
1:08
we dive in, can you take a minute
1:09
and tell us a little bit about UR
1:10
Medicine and what makes it unique?
1:13
Yeah. I'd love to. So those of you
1:15
who are not familiar with UR Medicine, we
1:17
are
1:19
a large academic health system in Western
1:24
Part of New York State. We are
1:27
the largest health system outside of New York
1:30
City in the in the state of New
1:32
York. So we have a very large
1:35
geographic reach. Essentially the whole Western half of
1:38
the state is,
1:40
our market
1:42
that we serve. So very diverse
1:45
patient population. Everything from the patients you would
1:49
see in in an inner city of a
1:51
mid sized city to
1:53
you go 30 miles outside of the city
1:55
limits and it's fairly rural.
1:58
What's really special
2:00
about,
2:01
UR Medicine is
2:04
our relationship with our university
2:06
which is
2:08
very very unique
2:10
for
2:11
academic health systems today.
2:13
We are one of the
2:15
few academic health systems in the country that
2:18
is fully integrated into our parent university. Our
2:21
parent university essentially owns the health system and
2:24
our CEO
2:26
and president of the health system reports right
2:28
up to the president of the university.
2:31
Our
2:33
quaternary hospital sits right on the university's campus.
2:37
And you know what makes that special for
2:40
us and we actually look at as a
2:42
differentiator
2:43
is we have access to some of the
2:46
most brilliant minds on our academic campus across
2:50
engineering,
2:51
across
2:52
computer science and data science,
2:56
even a world renowned school of music,
2:59
and business school and education school.
3:03
And we get to leverage that expertise and
3:06
apply it to healthcare. And so we have
3:08
a true
3:09
innovation team at the University of Rochester supported
3:12
by the university
3:15
to help move the strategic priorities of the
3:18
health system forward. Not a traditional
3:21
research shop where you know our faculty
3:24
the expectations
3:25
aren't to get as much NIH funding
3:28
as possible and publish papers. It's truly to
3:31
innovate
3:33
to support health system priorities. So under one
3:35
roof we have faculty
3:37
from all of our schools.
3:39
Again engineering and computer science and data science
3:42
and even the the School of Music under
3:44
the same roof as faculty of the medical
3:46
school, dental school, and nursing school. So having
3:49
that really eclectic group of,
3:52
minds allows us to come up with some
3:55
really exciting solutions to some of our pressing
3:57
problems in the health system side.
4:01
That's great to hear. And what a unique
4:03
opportunity to truly, truly be able to focus
4:05
on the patient care side and and being
4:08
innovative within health care delivery.
4:10
I'm curious. Looking in the last year or
4:12
so, what accomplishments are you most proud of?
4:14
Yeah. So I I the the accomplishment that
4:17
I'm most proud of, for our university is
4:20
maybe not the sexiest,
4:22
accomplishment,
4:23
especially, you know, when you think of,
4:25
all the great opportunities
4:27
and innovation in health care right now.
4:30
You know, our biggest accomplishment
4:32
was the setting up
4:34
of an artificial intelligence governance
4:37
across the health system
4:39
and actually
4:41
university wide. You know, we we felt that
4:44
having a governance that,
4:47
could help us think through, you know, how
4:50
do you deliver,
4:51
this really exciting technology in in a safe
4:55
and responsible way across all of our missions
4:58
at the university,
4:59
but also set it up in a way
5:01
that to allow us to be nimble and
5:04
to,
5:05
accelerate,
5:06
the advances using this technology
5:09
was not a small task to do.
5:11
And it took us, you know, easily six
5:15
six months to a year to really get
5:17
the governance structure
5:19
right.
5:20
But we were able to accomplish that. And,
5:23
you know, that that I think has really
5:25
set the foundation for us to continue to
5:28
be leaders
5:29
around the use of technology and specifically
5:32
artificial intelligence,
5:34
not only in health care, but in education
5:36
and and and our research missions also.
5:42
That's great to hear. And, you know, I
5:44
I know looking at the governance process is
5:46
not an easy one or a simple one
5:48
at all. So many health systems, right now
5:50
are still trying to figure it out and,
5:53
create the the right policies
5:55
and bringing the right people in order to
5:57
make those decisions. Could you go a little
5:59
bit deeper in there in terms of,
6:02
how do you did you develop the governance
6:04
process, especially around AI?
6:06
Who should be involved, and and how do
6:09
you, really then cascade the message once you've
6:12
got,
6:13
policies ready to go?
6:15
Yeah. So,
6:17
our governance
6:18
starts right up at the top of the
6:20
university. And so we have what we call
6:22
an an artificial intelligence
6:24
council
6:25
that reports right up to the president of
6:27
the university and the in the president's cabinets.
6:30
And
6:31
the artificial intelligence council
6:34
has lots of different
6:36
stakeholders on it including
6:39
chairs and representatives
6:41
from
6:42
the AI committees, which I'll talk about next.
6:45
But it also has stakeholders
6:47
from privacy and legal,
6:50
and compliance
6:51
from
6:52
IT
6:53
to ethics,
6:55
to,
6:56
some of our our our content experts, our
6:59
deep technical experts,
7:01
a and AI.
7:02
But we also have experts around thinking about
7:06
AI use in our community and and how
7:08
we can make it equitable
7:11
across our our our community. Below the AI
7:15
Council,
7:16
we have what we call domain committees and
7:18
we have five domain committees.
7:21
We have a clinical domain committee,
7:24
a research domain committee and education domain committee
7:28
and administrative
7:29
domain committee and a marketing and communications
7:32
domain committee. And those committees have
7:37
a diverse group of stakeholders that sit on
7:39
those committees and the purpose of the AI
7:41
Council is to really enable
7:44
the committees
7:45
at the domain level to
7:48
be autonomous
7:50
to to be able to identify what are
7:53
the specific artificial intelligence
7:56
drivers that align with the values of that
7:58
domain,
8:00
support those committees to develop their own guidelines,
8:04
procedures, and in some cases, like in our
8:07
clinical domain committee, actual policies,
8:10
to support
8:12
regulatory
8:13
and compliance and safety
8:16
concerns.
8:17
But also, you know, allow those domain committees
8:21
to inform the council about where we should
8:24
be making strategic investments,
8:26
at the university level,
8:28
within and within AI. And so at the
8:31
council level,
8:32
you know, it's it's it's that
8:34
group's job is to try to standardize where
8:37
we can across all of the domains on
8:40
our guiding principles,
8:42
and again guidance and and policies
8:46
to make sure that we have
8:48
safe,
8:49
and internally
8:51
improved,
8:52
AI tools that are equitable across our students,
8:55
staff, and and and faculty.
8:58
And, really serve
9:00
as a group,
9:01
sounding board for the the AI committees if
9:05
they run into,
9:06
issues that they need further strategic
9:10
guidance on. And then within that governance structure,
9:13
we have workflows,
9:15
identified
9:16
for
9:17
the submission of new AI tools that,
9:20
individuals within our university and health system would
9:24
like to
9:25
try or use.
9:27
And so we've we've built a kind of
9:28
a nimble workflow to get those tools vetted.
9:32
And then we've also, you know, developed,
9:35
a new contract addendum guidance for some of
9:39
our long standing vendors who are starting to
9:42
introduce
9:43
AI functionality into their platforms to make sure
9:47
that we have that partnership with our vendors
9:49
and they are,
9:51
you know, protecting our data,
9:53
to the standards that we,
9:55
hold our ourselves to at the university and
9:58
that they are also thinking about what are
10:00
the ethical
10:03
bias,
10:04
accuracy, reliability
10:06
implications
10:07
of these tools, especially on the generative AI
10:10
side.
10:13
That's helpful for understanding. Thank you, Michael, for
10:15
going through that with us and especially, like
10:17
you mentioned, getting into generative AI. I know
10:19
that's a whole new, can of worms opening
10:22
in some ways. So,
10:23
very much appreciated. Now I'm curious looking ahead,
10:27
where do you see some of the big
10:28
growth opportunities over the next twelve months or
10:30
so?
10:31
Yeah. Well, let's, Laura stay on generative
10:35
AI. You know,
10:36
we we as a as a health system,
10:38
you know, we were one of the
10:40
early,
10:42
users
10:42
of,
10:44
secured private instance of, GPT four, that specific
10:48
foundation model.
10:49
You know, we we also have other secure
10:53
private instances
10:54
of, public foundation models, and and we also
10:57
have a supercomputer where we have
11:00
a smaller,
11:02
open source,
11:04
model sitting on. And
11:06
for a university,
11:08
academic health system like ours,
11:11
it has never been easier for us to
11:14
develop our our own AI tools
11:16
rapidly
11:17
to solve our own problems by essentially,
11:21
doing some prompt engineering,
11:24
of these foundation models with our own data
11:26
or fine tuning these foundation models on our
11:29
own data.
11:30
And so, like, we're really
11:33
very narrowly focused in on where are those,
11:37
manual
11:38
administrative
11:40
tasks across the system
11:43
that if we were able to automate those,
11:47
using,
11:49
generative AI tools,
11:51
where what what would have the biggest, not
11:53
only return on investments,
11:55
but,
11:57
you know,
11:58
decreasing,
12:00
mitigation risks and and have the biggest impact
12:04
really on our employees,
12:06
our our clinicians,
12:08
and our staff? That's where we're
12:10
focusing
12:11
our efforts on and we're making a lot
12:13
of headway. We are cranking out tools in,
12:17
you know, days to weeks. Tools that typically
12:20
would have taken our engineering
12:22
and data science teams
12:24
six months to a year to build. So
12:26
it's it's been really exciting and quite powerful,
12:31
you know, our ability to to develop internally
12:36
by having
12:37
access to these these massive foundation models.
12:43
That's great to hear. And, you know, really
12:45
cool to have that opportunity to speed things
12:47
up and, like you mentioned, just,
12:50
do more and more as time goes on,
12:53
with degenerative AI. I know a lot of
12:55
opportunity there, but I can imagine also challenges.
12:58
What are some of the big, challenges that
12:59
you're anticipating
13:01
heading into the future as well?
13:03
Yeah. You know, I think,
13:05
one of the biggest challenges
13:07
is understanding
13:08
what the regulatory
13:11
landscape may be around the use of of
13:15
of specifically generative AI on on the health
13:18
care side. And, you know, we know that
13:20
there is a lot of changes happening at
13:23
the federal level,
13:25
in this space and trying to understand that.
13:27
But when push comes to shove,
13:29
you know, as a as a health care
13:32
institution,
13:33
you know, we
13:34
our primary business is providing high quality care
13:37
to our patients.
13:39
And so, like, that is that's the the
13:42
essence of what we do. And so it
13:44
doesn't really matter, you know,
13:46
what
13:47
regulations come out or maybe lack of regulations
13:50
that may come out. We still have to
13:53
hold ourselves accountable
13:55
of
13:56
of making sure that these technologies
14:00
are safe
14:01
and trustworthy,
14:03
when,
14:04
delivered within clinical care settings. And we need
14:07
to make sure
14:08
that we can create assurances
14:10
for our clinicians using these tools that they
14:13
can trust,
14:15
the the the output,
14:17
that that we're seeing
14:19
from from the AI. And so, you know,
14:22
I think,
14:24
you know, really trying to
14:26
get a handle on that and and really
14:29
trying to
14:30
figure out how do we create,
14:34
pre deployment validation opportunities
14:36
of of these generative tools and
14:39
post deployment
14:40
auditing and monitoring,
14:43
systems,
14:44
to watch what these tools are doing.
14:46
You know, that that is a a space
14:48
that I I see a lot of opportunity,
14:51
but again, a lot of challenge just given
14:54
there's so much
14:56
black box or
14:58
things we don't know about the these these
15:01
large foundation models. And so, you know, that
15:04
has been the,
15:05
I think,
15:07
biggest thing that we've been
15:09
running into and thinking about. Again, never been
15:11
easier to build a point solution. Like, we
15:14
we we can we can do that really,
15:16
really quickly.
15:17
But to be able to go from pilot
15:20
to
15:21
full scale across the health system in a
15:24
safe and trustworthy way,
15:26
that has been a challenge and difficult. And,
15:29
again, it's
15:30
because of
15:32
not having that
15:34
clarity or mechanisms,
15:37
in place to to, again, monitor and audit
15:40
these tools at scale.
15:44
That's fascinating to hear. You know? And, certainly,
15:47
I can imagine,
15:48
a a challenge that others others in your
15:50
position are are seeing in health systems as
15:52
well. And I'm curious, you know, especially when
15:55
you think about, the taking pilots or some
15:58
of the small scale things that you're doing
16:00
and then seeing successes, rolling them out, and
16:03
and,
16:04
defining challenges
16:05
within that. I guess, could you share with
16:08
us any examples of needing to
16:10
kind of put the brakes on a project
16:12
or,
16:14
something that was like a a a near
16:15
miss that really kinda highlighted what you're talking
16:18
about here?
16:20
You know, not not per se,
16:24
a near miss,
16:25
yet, at least, because we are very,
16:30
thoughtful of leaving our tools that we develop
16:33
turned on outside of production for, you know,
16:36
some some cases, you know, six months after
16:40
we've built the tool and kind of watching
16:42
the tool outside of production and comparing to
16:45
what we're seeing from,
16:47
our humans,
16:49
managing
16:50
that
16:51
task,
16:53
within production.
16:55
What we have seen, however, you know, with
16:58
things of,
17:00
you know, these
17:01
these foundation models changing so quickly and getting
17:04
updated,
17:05
so quickly. We have seen
17:08
things that,
17:10
when we
17:11
deployed a tool,
17:13
again, in in outside of production, we've been
17:16
watching it. It, you know, maybe did really,
17:18
really well at,
17:22
triaging a patient message to the the right
17:26
endpoint,
17:27
or the right clinician or staff member that
17:30
that message should go to.
17:32
But, you know, as there's updates on that,
17:35
on the foundation model, we're finding that, you
17:38
know, some messages are getting caught in the
17:41
filters of the foundation model that, you know,
17:44
initially weren't getting, caught. And so,
17:47
you know, we're having to go in and
17:48
try to understand why is it getting caught
17:50
in the filter and changing some of our
17:52
prompts to make sure it gets to the
17:55
to to the right place.
17:57
You know, we're really trying to be sensitive
18:00
around
18:03
what problems we're trying to solve
18:05
that
18:06
if if the if the model,
18:09
or tool was to get it wrong,
18:12
and and and, you know, that if there
18:15
was a near miss or an actual miss,
18:17
the the risk of harm is is gonna
18:20
be much lower. And so most of the
18:23
tools that we've developed thus far,
18:26
there still
18:28
absolutely
18:29
has to be a human in the loop.
18:30
And when I say a human in the
18:31
loop, it's one of our clinicians
18:33
or one of our staff members and employees,
18:38
that validates,
18:40
the appropriateness
18:41
of of the output. And so having that
18:43
extra layer
18:45
in, I think, has,
18:48
created that safety net so we haven't had
18:50
that miss yet. I I I don't think,
18:52
you know, we're ready,
18:55
culturally,
18:56
or from a societal standpoint
18:58
yet. Or I'm not even sure if the
19:01
technology is ready yet to
19:03
be directly
19:04
interfacing with with our patients without having that,
19:09
clinician in the loop yet.
19:13
Got it. Absolutely. I think that's such helpful
19:15
perspective. It it really truly illustrates,
19:18
some of the the things that we've been
19:19
going through today. So,
19:21
I I appreciate it. Before we wrap up
19:23
here, I'm curious. What is the number one
19:25
thing that you're doing right now to set
19:27
set up your medicine for long term success?
19:30
Yeah. That's an easy question. You know, for
19:32
us, you know, it's
19:34
your medicine
19:36
is not gonna figure this out all by
19:38
ourselves.
19:39
Industry is not gonna figure this out all
19:42
by themselves.
19:43
The feds and states are not, from the
19:46
government level, are not gonna figure them figure
19:48
this out all by themselves.
19:50
It's truly gonna be through partnerships.
19:52
And so we have
19:54
really thought out of of, you know, what
19:57
are the the
19:59
the the right groups of consortiums,
20:02
of industry,
20:04
private,
20:05
public,
20:07
groups that we can join and have a
20:09
seat at the table
20:10
to kinda talk through some of these
20:13
difficult challenges and learn from others about how
20:17
they're
20:18
overcoming these challenges or the way that they're
20:20
thinking about that. And so we've made a
20:23
conscious effort and we have protected our time
20:26
of a lot of our leaders in our
20:28
organizations
20:29
to participate
20:31
in,
20:32
again, you know, these
20:34
these consortiums that are forming around AI and
20:37
health care and you know, really trying to
20:39
get out,
20:40
and travel to other health systems
20:43
and meet with our industry partners
20:46
and really understand how they're trying to tackle
20:48
these problems. Because I actually
20:51
believe that the future,
20:53
I mean, where we're really gonna be transformative
20:55
in health care in this country,
20:57
is is is gonna be more, you know,
20:59
on this sharing of ideas and almost on,
21:02
like, an open source concept of of us
21:06
across industry, across providers,
21:09
across payers,
21:11
across government,
21:13
you know, giving away some of our our
21:16
learnings and sharing it with others so others
21:18
can build on top of that. And that's
21:20
where I think we're really going to be
21:23
able to leverage technology like generative AI to
21:26
transform health care for the better of our
21:28
our patients
21:31
and our clinicians. And so again for us,
21:34
you know,
21:36
long term success is gonna be through partnerships
21:39
and making sure we find those those right
21:42
groups that have a similar why and vision
21:45
that we have.
21:49
I love that. Michael, thank you so much
21:50
for joining us on the podcast today. This
21:52
has been just a really fun conversation, and
21:54
I look forward to connecting with you again
21:56
soon.
21:57
Thank you, Laura, for having me.