Dr. Eric Cheng, Chief Medical Informatics Officer at UCLA Health Sciences

In this episode, Dr. Eric Cheng, Chief Medical Informatics Officer at UCLA Health Sciences, shares his unique journey from neurology to informatics, dives into the intersection of AI and data quality, and explores how UCLA is reimagining growth, regulation, and patient-centered documentation in healthcare.

Episode Transcript

(00:00):
This is Rosie Talago with the Becker's Healthcare
podcast.
I'm thrilled today to be joined by doctor
Eric Chang, chief medical informatics officer of UCLA
Health Sciences
and professor in the department of neurology at
the David
Geffen School of Medicine at UCLA.
Doctor Chang, it's a pleasure to have you
on the podcast today.
Yeah. Thanks for the invitation.

(00:21):
We have a great conversation in store, diving
into some of the most exciting trends in
health care, the innovations happening at UCLA Health,
and how leaders like today's guests are thinking
about growth in the year ahead. Before we
get into it, doctor Chang, can you please
introduce yourself and tell us a bit about
your background?
Oh, yeah. Sure. So what first, again, thanks

(00:41):
for the invitation. I've really enjoyed listening to
the podcast of some of my friends and
former coworkers. I'll just call out Mike Pfeffer
at Stanford and Clara Lynn at Seattle Children's
who have done your podcast.
I've interviewed some applicants for our clinical informatics
fellowship program, and many describe
a, like, childhood hobby in computing or an

(01:02):
engineering background,
but I don't have that. So I'll describe
my background in a bit more detail just
to show there's another path to my role.
So
I developed an interest in outcomes research during
my neurology residency.
UCLA is a national leader in health services
research, so I did a fellowship in that
area. And I spent over a decade as

(01:24):
a grant supported
health services researcher
measuring quality of care
for persons of neurologic conditions.
And I also volunteer for my specialty society
in developing
quality measures for regulatory programs.
I had
ventured into informatics
even though I really didn't know,

(01:45):
what that word meant.
So I'll give a brief,
example. So we were analyzing the VA national
database and whether blood pressures were
well controlled the year after a stroke,
and we showed that a stroke patient's
antihypertensive
regimen
frequently wasn't intensified
even though the recorded blood pressure in the

(02:06):
clinic was high.
But then we
reviewed some charts, and we read that many
physicians would remeasure
a high blood pressure. They would document that
value, which is typically a lower value, as
free text in their notes.
So at that time, I call that an
administrative
database error, but that's actually informatics.
To fully understand data on the back end,

(02:28):
you need to understand how it's entered on
the front end. So using informatics vocabulary,
there wasn't a workflow for physicians who enter
blood pressure values in the same data field
where the automated blood pressure machines were transmitting
it. So when UCLA purchased an EHR,
they sought some physicians who could help configure
it. I did my clinical work at the

(02:50):
VA, and we had implemented an EHR for
years.
UCLA also sought someone who could help interpret
the federal
regulations about meaningful use, and I was pretty
familiar with government regulatory programs based on my
volunteer work. So
I guess they thought it was a good
fit. It's been said, like, many times that
life doesn't
always make sense looking forward, and these aren't

(03:12):
the career choices I would recommend to someone
who would like to be a CMIO.
But looking backwards,
yeah, I can see how these experiences
led to my position and my way of
thinking.
Absolutely. That's so interesting, and it's great to
see another path, like you mentioned, opposed to
that more exclusively or assumed
engineering technical side of things.

(03:34):
And I found it very interesting that you
said you veered into informatics without truly knowing
exactly what that word meant.
But it's sounds like you bring a very
valuable combination to the table, especially
in today's health care environment.
Yeah. It's it's a so there's,
again, not a single path. And

(03:55):
maybe one of the reasons why I was
elected to be one of the first physician
informaticists
was that I was a little different from
the other physician informaticists who did have a
different background,
maybe more business or may maybe more computer
science related.
Yeah. So with your
new or fresh perspective

(04:15):
perspective in mind, I'd love to hear how
you're viewing the current
landscape. So what are the top three trends
that you're following in health care today?
Okay. Well well,
answer number one has got to be artificial
intelligence, of course. Right? There's so much that
people have said, and I follow,
you know, the developments of both excitement and

(04:37):
skepticism.
I'll I'll just make this brief comment.
Based on my health services research days, I
know that
there are
methodologic
flaws, especially like self selection bias, that can
make it difficult to interpret observational studies.
So when you really need confidence in an
answer,

(04:58):
there isn't a substitute for randomization.
So I'll describe a project here for Ambient
Scribes, very hot topic. We randomized 250
physicians to
two vendors and a control arm and then
crossed them over at set intervals.
So then we
got some insights on how they were the

(05:18):
vendors were similar and different from each other
and when they did or did not differ
from the control group.
The overhead to conduct a randomized trial
is too high to run for every AI
implementation.
But when a project has a, you know,
potential seven digit price tag, it is worth

(05:39):
slowing down the implementation
just a bit so they can study it
properly.
So I definitely encourage others to do that
as well.
The second,
I think, thing I'm following
is
related to AI.
So I'm cheating a little bit, but I'll
call it,
systematic
data collection.

(05:59):
So there's a well known
epidemiological,
like, study that shows that about twenty percent
of early mortality
is attributable to health care,
and the rest is attributed
to factors
we don't systematically
collect, such as patient behaviors, social drivers of
health, the built environments, and genetics.

(06:21):
So I'll tell another story from my research
days. We were wrapping up a trial. We
had some difficulty
getting the final survey results from some participants.
So I asked a statistician,
can't we run some fancy
imputation
imputation methods to overcome this?
And he looked at me and said,

(06:42):
you know, imputation
is certainly better than no imputation,
but there is no substitute for primary data
collection.
So likewise right now, if we don't collect
the right
predictors,
then the outcomes will be inaccurate due to
omitted variable bias, and AI can't overcome that.
What AI does is generate better predictions for

(07:04):
the data that we already have. And to
fully leverage it, we just need
accurate and more types of data. So I
view that as complementary
or kind of almost more fundamental to get
AI to to work at its best.
And then the third topic, I'll mention a
topic that maybe
no one else has mentioned in your podcast.

(07:27):
But in 2025,
it's gotta be government regulations. Right? It's it's,
it's the world we live in. CMIOs have
always needed to monitor this space, you know,
meaningful use, the twenty first Century Cures Act,
there's TEFCA,
so all the quality measures.
But it's not just at the federal level,
but it's at the state level as well.

(07:48):
So I'll give an example. So the federal
twenty first Century Cures Act states that results
need to be made available to patients
without delay,
but our state government passed a law stating
that if you have radiology results that show
new or recurrent malignancy,
that should be verbally disclosed instead of through

(08:08):
a portal.
So we developed a process for the radiologist
to manually tag such scans so they are
held from the portal for a short time
so that the ordering provider
has the opportunity to reach out to the
patients.
But because that process is inconsistent,
we just supplemented it with a homegrown machine
learning algorithm

(08:29):
that analyzes the text of the radiologist report
to determine
whether it meets a criteria
mentioned in state law and then holds it
back from the patient portal.
Now did I need a state law to
tell me that informing patients of a new
diagnosis of cancer through the portal
is
now potentially problematic?
No. But

(08:50):
I couldn't champion and prioritize
the developments of a machine learning
algorithm
if I didn't have the weight of the
state law behind me. So it is
important to follow, like, what's going on in
government and
possible, you know, really leverage it to drive
change.
Absolutely.

(09:10):
I particularly liked your example of the connection
between that systematic data collection
and artificial intelligence and machine learning
tools such as those.
I feel like a lot of times it's
easy to think of this artificial intelligence as
this higher
higher thing that's all knowing and super smart,

(09:31):
which it is.
But we gotta pay attention to the data
that we are feeding it because at the
end of the day, like you mentioned,
if we're omitting
variables,
that's bias, and it's very important for the
results that it produces.
Yeah.
People, yeah, others have commented
it as well in a maybe a slightly

(09:52):
different way. They point to the quality of
the data. If we
if the notes are somehow inaccurate, then certainly
the predictions are inaccurate. But in addition to
the quality of data, I think the types
of data makes, should be expanded.
Absolutely.
So zooming in a bit now, UCLA

(10:12):
is known for being on the leading edge
of innovation.
So I'm curious to know what are you
most excited about right now at UCLA?
So I've listened to this podcast, and a
lot of people mention,
burnout reduction, and I agree.
And people point to several different factors, typically

(10:34):
patient message volume and pajama time and EHR
usability.
And I agree with that too.
But I wanna highlight maybe a personal interest
of mine. It doesn't have a financial,
ROI, but I still think it's just as
important.
I think one of the reasons why it's
so
cognitively taxing

(10:55):
to use an EHR
is because the notes are just too hard
to read.
They're long. They're repetitive.
It mixes both past and current data.
The information that's useful to the author is
frequently not relevant to the reader. But but
I get it. You know, the clinic has
such time pressure.
It's faster to start from a template or

(11:16):
your prior notes than a blank page.
Yet despite its long length, it doesn't always
capture the essence of the patient.
Like many medical students who chose neurology,
you know, I was drawn by Oliver Sacks,
narratives of patients.
So when I'm reading a note now, I'm
not expecting literature, you know, but it is

(11:39):
hard to hear the patient's voice in many
notes. So if the Ambient scribe
can capture the essence of the conversation
while the upcoming, like, summarization tools can bring
in the relevant parts of the note, that
frees the physician to focus on the activities
we were trained to do, and people cite
that as a reason for better satisfaction.
But on top of that,

(12:00):
it could lead to a better experience for
the reader as well. We could achieve both
a
detailed rich note yet relatively
concise.
It's so nice to read a well written
note, and I like to find a way
to make that easier.
That's a great point. I do hear a
lot about how it can streamline the process

(12:21):
for
the providers and the nurses and the doctors,
but that's a very interesting perspective to bring
up for the reader and the patient bringing
the patient's voice into the notes too. Very
interesting.
Yeah. There there's a tradition. Neurology at these
long form,
back to Luria and Sachs of, like, describing

(12:42):
kind of how how a patient is, you
know, thinking
and doing. And,
yeah, it's maybe impractical in everyday use, but
I do I do enjoy reading things like
that.
Absolutely.
And finally, looking ahead, growth is always on
everyone's mind, especially in this industry that's adapting

(13:02):
to new challenges and opportunities
so quickly. So I'm wondering how you are
thinking about growth over the next twelve months.
So
I'll think of it from the perspective of
the organization.
UCLA has always
seems to be expanding their outpatient footprints,
but they've recently

(13:23):
expanded in other ways as well.
I've been at UCLA now for thirty years,
and
we purchased new hospitals for the first time
in the in the past year since I've,
since I've been here.
We just implemented our EHR there. We're gonna
open a new psychiatric hospital next year.
We're implementing our EHR in our university student

(13:46):
health center this summer.
We just rolled out our
managed care medical
Medicare Advantage health plan.
So we're expanding
into
lots of different areas other than opening up
outpatient clinics.
So I think that's probably
a main kind of focus of activity. But

(14:07):
even if we weren't expanding,
there's a lot of work to keep up
with, you know, developments from our EHR vendor,
from other third party vendors, and from our
own community as well. And I I guess
I wanna mention that. An academic medical center
is in the position to build some of
their own tools, especially when it comes to
AI.
And sometimes that's led by

(14:29):
the IT organization itself, but sometimes it's led
by our own faculty.
I'll say it's a lot. I just wanna
add that no one can do this alone.
There are great people
under me, next to me, above me, whom
I interact in
everyday basis, and that is one of the
most enjoyable aspects of my job. There's kind
of a lot of variety of the things

(14:50):
that I'm exposed to.
That's wonderful. And that's very cool to see
that you've been there for so many years,
but you just now are expanding in new
ways you haven't seen,
aside from the outpatient that you've been doing.
But seeing this expansion and growth in new
areas is very exciting and
also very exciting to think of that possibility

(15:11):
or the capability of an academic medical center
to have faculty led
new tools,
be in that position to build some of
their own tools is very, very exciting.
Yeah. I think because of that, we are
in the process of hiring a chief health
AI AI officer to help

(15:32):
kind of what's the best way to put
it? Maybe organize some of those activities.
You know, faculty, one of their strengths is
that they are,
you know, fiercely independent, but it it would
help to have some way to connect them
all together. And I think that's that'll be
one of the
functions of this, new,
person.

(15:53):
Absolutely. Like you said, you have great people
all around you. No one can do it
alone. So bringing all those great powerful minds
together to do it together.
Yeah.
Well, thank you very much. That is all
the time we have for today, but I
wanna thank you, doctor Chang, for sharing such
wonderful insights and experiences
gained from your experience at UCLA.

(16:15):
It's been really fascinating to hear about all
the wonderful things that are underway, and we
look forward to collaborating with you again soon.
Thank you. So look forward to that.