2015 Public Health Grand Rounds 10/16

PUBLIC HEALTH GRAND ROUNDS Linking Research to Community Health Improvement

(audience member speaking)
(applause)
- Thank you very much
for coming and inviting me
to this pretty impressive venue.
I have to admit, this is the first time
I've had a sign language interpretor,
I'm mic'd up, I'm being video recorded,
there's like more than four people listening to me talk,
and so thank you yeah because I actually told,
you know I asked ahead of time,
is it okay for me to present preliminary research
and he said yes and so what I'm going
to be presenting today is preliminary
but at the same time
I am really excited about it,
it's funded by this Early Independence award
which Elaine just recently got
and so super excited about kind of
this research agenda.
I hear there are quite a few people here
that do VA work and are interested
in kind of physician behavior,
or kind of information technology,
preventive care and this kind of intersects
with all of that, so.
Let's get out of here.
Okay so today I'm going to be talking
about kind of the general research agenda
of you know the benefit and burden electronic reminders
for optimizing patient care,
and spending a little bit of time
you know talking about the research I've done so far
which is kind of smaller than that of course,
but you know then broadening out again
to what the general research agenda is.
So, so to motivate this you know,
I don't think I need to tell
most of you guys in this room
that health IT is becoming increasingly kind of
you know discussed and important
from a policy stand point
and also from a clinician stand point.
You know we're seeing health IT you know
in our daily care of patients increasingly
so it's something that is ramping up.
We're you know, the Quest is spending
about 35 billion in annual investments
in health IT.
We've got things like meaningful use
kind of in policy discussions
like it's incentivized in the Affordable Care Act.
A lot people kind of talk about health IT.
This is one of the things
that you know both Republicans
and Democrats agree you know
there should be more health IT.
And you know it's increasingly viewed
as you know as a tool kind of
to get doctors to provide in a quality health care
and kind of from the public health stand point
as a tool for doctors
to kind of provide you know care
that is not necessarily right in front of your face
like needing to be addressed right now,
but preventative care.
Things that are good you know best practices
for taking care of chronic diseases
like diabetes and hypertension
and hyperlipidemia.
But from a kind of research stand point,
we know surprisingly little
about what health IT actually does
to clinical care,
and so that's kind of the challenge here
because how we redesign the system
that optimizes clinical care
if we know very little about it
and so far what we know
is actually quite conflicting.
You know on the effective health IT on outcomes
and part of this is the fundamental difficulty
to kind of consider exactly what health IT is.
So health IT can mean
a number of different things.
There are like a number of bells and whistles
and even if you're comparing something
that kind of on paper does the same thing,
how do you compare like the Google website
with a Yahoo kind of website.
Like you know that's kind of like comparing Epic
with Cerner, there's so many things
that differ between these two health IT platforms,
how are you even going to do a study
that can you know compares all of these things
that are pretty much,
they've got more differences between the systems
than you've actually got systems.
So in practice,
you know so what I'm wanting to do
is to see if there's,
you know there's some research,
you know opportunities particularly in the VA
to kind of help open this black box
and to kind of hold a lot of stuff constant
in health IT but look at important kind of things
that might vary such as electronic reminders.
So and you know,
and part of this is to kind of open up this black box
and look at from a behavioral kind of stand point
because I think that's really kind of
this argument for whether health IT's good or bad
has this interaction with how physicians
respond to health IT from a behavioral stand point.
So on the one hand,
if physicians kind of were machines themselves
and didn't like need information presented to them,
they wouldn't need health IT
so one argument is that health IT could be good
because it provides useful information
to clinicians who might otherwise
be you know too busy to kind of think
about those things.
They might otherwise overlook
you know the fact that this patient
is due for you know a diabetic eye exam,
or a diabetic foot exam.
But on the other hand,
because of the very fact
that they do have,
they do have limited cognary processing,
they have busy days
and they have to deal
with a lot of information coming their way.
Health IT could actually worsen care
if it provides too much distracting information.
So either too much information,
or information that's not particularly relevant
for you know your patient in front of you,
and I was just you know talking you know before hand
like with another clinician,
you know most guidelines
are really written in broad and generic ways
that the chances you're going to have
like you know an electronic reminder
sent to you for a patient
that it might not be applicable for you
is actually quite high.
So how do you like design
a system given that?
And so,
so you have these two kind of two things,
you know on the one hand providing
more useful information
and on the other hand having
the risk for information overload
or alert for the use of the concepts
that have been talked about quite a bit
in the popular press,
and a lot of doctors have kind of brought attention to this
but we don't really have much evidence on this
and so the idea is,
can we take a closer look
at this trade off in order
to optimize the use of health IT
for clinical care.
So as an outline for the rest of the time here,
so I'm going to basically kind of define
you know this research problem
a little bit more.
Talk a little bit about the state of the literature,
and then describe the institutional setting
that I'm going to study this.
In particular that, in particular the VA
kind of institutional setting which has its
you know pretty significant health IT platform
that is unique in some ways,
and I'm going to argue it's unique in a good way
from a research perspective.
The second thing I'm going to to do
is then kind of go into
this natural experiment of the VA
which is you know what I argued
for to you know for BM2
and I used to kind of fund this research.
Basically at the VA there's a kind
of uniform platform of health IT.
So there's not like one side usesEpic
and the other side uses Cerner.
Everything uses the same health IT platform
but there is kind of variation
in meaningful measures such as electronic reminders
and I'll spend a little bit of time
describing exactly what electronic reminders are
to the non clinicians here.
Then kind of in the middle of the talk,
I'll describe some of the preliminary work
I have so far,
which is research in progress
looking in a clinical setting of diabetes
and using what the VA calls our health factors
and this is like a very VA specific term
which are pieces of information
that are generated by electronic reminders.
And then finally,
you know I want to end by talking about future work.
Where this could go,
you know in the VA setting
and hopefully in other settings.
So you know a review of the literature.
Again like as I was saying,
there's really very little known
you know about like the specific affect
of health IT on patient outcomes,
but there's a lot of work,
kind of more generally about you know
human cognitive limitations.
Most of this is lab based.
And you know it goes back a long ways
such as Miller's famous seven digits
like we you know,
humans can't really memorize more than seven digits,
like it's seven plus or minus two
and so these things are kind of,
you know have been documented
time and time and again in the lab,
but the question is,
how do we use those you know lab based findings
and apply them to the real world
where kind of by design,
we're throwing you know information
that's more than seven of course,
but configured in different ways.
What's the optimal way
to kind of organize that information.
All these you know practical questions
that these lab based kind of studies
don't really you know give us
as much guidance as we need.
And so second,
even though there's been a huge policy attention
to health IT, there's been no consensus
on the overall, you know whether health IT
is good or bad actually.
So there have been case studies
about how health IT has improved outcomes.
Left outcomes unchanged
were actually worse than the outcomes.
You know maybe part of this is publication bias
like after the guy has written
the paper that you know health IT
improves the outcomes.
There's always an example
of some hospital out there
that implemented health IT
and actually you know increased
you know adverse outcomes.
So, so you know so there's,
and you can kind of see
you know given that there are these two
kind of things that you're trading off against,
there certainly should be
the possibility that health IT
could worsen the outcomes.
And these have been case studies.
Now moving on to like natural experiments
which is what kind of you know people
in my discipline, like in economics do,
so you know is there a way
to quantify an effect of health IT
using some kind of natural experiment
such as one system implemented health IT
at one point or a number of system
implemented health IT at one point
and controlling that against systems
that didn't implement health IT at those points in time.
Now the fundamental problem here
is what I kind of eluded to earlier.
How do you describe or compare systems of health IT?
Especially if they vary in more ways
than there are systems of health IT.
So the way that people have primarily
kind of addressed this is to either
look at individual instances
of health IT adoption
such as like there was an early study
at the Brigham which kind of looked
at the Brigham specific health IT system
that was adopted and compare that
against a system that didn't adopt health IT.
So that's very kind of specific
to that bundle of things
in the Brigham health IT system.
Another way that people have kind of gotten at this
is to kind of just you know lump together
all systems, all health IT systems
and look whether a system
has adopted any health IT system.
And so that's essentially kind of
you know estimating an average affect
across different types of health IT system,
and a lot, systems, and a lot of these papers
have found no affect
on outcomes if you're you know averaging,
if you're averaging different types
of health IT systems.
So now the question is:
given that there's an average zero affect,
but there are examples
of systems that do better,
and systems that do worse,
how can we optimally configure
a health IT system so that we're on
the right side of that kind of zero?
So now kind of taking a step back
and kind of looking at this
from a clinician's point of view,
not a researchers point of view.
You know as a doctor.
So you're sitting there,
oh looks like we have a little
teamwork going on here,
so if you're a doctor,
you kind of, primarily what you do
is an informational job.
And so you're you know a patient
kind of shows up you know with a set of complaints,
you're processing information
to figure out what that patient has
as well as what kind of you know things
to recommend to that patient
and you know as economics as early as Arro,
you know basically this has been recognized
as a fundamental kind of feature
of the medical industry.
It's an informational job
and so it's not kind of surprising
that you would have health IT kind of
you know be provide like an informational role
that interfaces with clinicians.
But if you're sitting there in that office,
like basically the fact
that you have cognitive limitations
doesn't change whether you're dealing
with health IT or not.
So if you didn't have health IT,
you would have to process
a lot of information kind of on your own.
With health IT that could either
that could either make your job easier by kind of removing
stuff that you need to process,
but it could also add in more stuff
that you need to process.
And increasingly what we're seeing
is kind of the latter,
where we're in you know these systems
where quality measures,
like, and there's definitely a drive
to kind of measure quality objectively.
So there's a bunch
of kind of interventions kind of done
in the name of quality improvement
which are kind of pinging doctors,
telling them you know you should think of this,
you know you should think
of prescribing an ACE inhibitor
for this you know CHF case.
You know congestive heart failure.
We should think of adding
this blood pressure medication
and these quality improvement initiatives
are kind of adding additional things
that they're reminding doctors of
and you know one can argue
that these are good things
because you're kind of pushing doctors
to you know provide better quality care,
but there is this underlying
kind of fundamentally behavioral limit
on how much doctors can process
and you might be running up against that.
So you know finally,
this is the point that I've already made
which is just that you know,
by construction you've got physicians
who have a limited cognitive bandwidth
and you've got to work within that.
By having health IT you're not kind of
getting rid of that constraint.
You could be worsening that constraint.
So now I'm going to be talking
a little bit about health IT at the VA.
So the VA has had one of
the oldest and kind of most widely used health IT platforms
for years, so they have a platform
called VistA which is an acronym
for this longer name here.
Which is you know implemented
kind of in its stable form in 1994
and you can kind of read
the you know there's a gigantic Wikipedia article
on VistA and there's a lot of stuff
written about VistA because it is one
of the first kind of health IT systems
to kind of really do
a lot of these, you know,
have all these functionalities
that we're looking for
in the modern health IT system.
And you know a lot of people attribute
like better quality of care to VistA
but basically because the VA's so large,
VistA you know, VistA is an important health IT system
in its own right just because a lot
of physicians in the US have trained at a VA
so over 60% of US physicians have used VistA at some point,
and because the VA is so large
compared, it's the nation's largest
healthcare delivery system,
if you look at you know all the systems
that have health IT in place,
VA hospitals comprise nearly half
of kind of US hospitals that have
a health IT system in place.
So because VistA was kind of,
you know implemented in the 1990s,
like it has kind of an idea
of interacting with doctors
that's a little bit older
than kind of our current idea
of what are the, you know the state of the art,
how you interact with doctors,
and one of these primitive ideas
of interacting with doctors
is just kind of sending doctors reminders.
So it's a very kind of,
it's a very easy way to program.
You basically, you can have somebody
who doesn't know how to,
you know write complex programs,
but just simply define cohorts of patients,
like if you have a female of a certain age
then that should be a patient
that you would ping the doctor about
doing a breast you know cancer screen.
Doing a mammography.
And so it's like something you can just
kind of program;
like if the patient meets certain criteria,
you ping the doctor.
If the doctor does certain things
or the patient meets certain other criteria,
that reminder is what they call resolved.
So there are people in the cohort of the reminder
and people kind of in the resolution
who've had that reminder resolved.
And so even though it's a relatively primitive way
of communicating with doctors,
by and large, both in and out of the VA,
this is kind of the main way
that health IT systems,
one of the you know most common ways
that health IT systems kind of provide
information to doctors.
So it seems, it's you know,
compared to like you know any advanced reasoning
that you might have to kind of take into consideration
as a doctor, this is like
a really kind of relatively dumb way
of you know asking doctors to think the things,
and so you might,
you might you know understand why sometimes
like it could be over done
and have a lot of patients
where they don't need that thing
to be done on them,
but you're sending them a reminder just in case.
So here's some examples of electronic reminders
in the VA system.
And outside of the VA as well.
So for example, a reminder to screen for smoking.
This might kind of apply to any patient you know
so, all, in basically the reminder logic,
if you're any patient at all
you should ask you know about whether they smoke.
And then you have other reminders
that you know have you know for example.
A reminder to counsel against smoking.
You would need that patient to be a smoker
so you need to have asked
the first question about whether they smoked first
in order to kind of you know,
have this other reminder fire off.
And then what I'm going to be studying today
is like if you have a patient
who has a diagnosis,
kind of in this category of diabetic diagnosis codes,
you're going to send reminders
on controlling for,
controlling their diabetic you know,
their blood sugar, as well as things like screening.
So that yearly foot and eye exams.
So these are just an example,
a few examples, like in practice
there are hundreds,
if not close to a thousand different reminders
kind of floating around in the VA.
And so a little bit more kind of description
of exactly how this looks like in the VA.
So again as I was saying,
there's a fixed health IT platform,
but there's a lot of variation
in whether a specific VA location
at a specific time has a certain reminder in place.
So there are reminders
that are implemented VA system wide.
These are national reminders.
They're relatively few
and some examples of these
are hypertension assessments,
screening for traumatic brain injury,
and so forth.
There is also a set of reminders
that are implemented at the regional level.
And there's 21 different regions
or VISNs in the VA
and here is some examples of this.
But the vast majority of reminders
are kind of implemented at the local station level.
So an example of this would be
Palo Alto healthcare system is a station.
San Francisco healthcare system
is another station.
Seattle and LA are both also a system.
And I'll show you some examples
where you know this variation
can kind of grow over time
because if you talk to some of the people
who are in charge of programming these reminders,
they might you know hire a new person
and that person becomes,
takes some time to become good
at programming reminders,
and when you see that happen,
like you see, like you basically see
an uptake in the number
of you know reminder related information at that site.
So for reasons like that,
you have a natural experiment;
where you know different sites are exposed
to different levels of electronic reminders over time.
So this is just a picture
of the cross sectional variation
you know in the VA.
There's a 20 full variation
in reminder related informations
which I'll take about later.
They're called health factors,
across VA health care systems,
and they're not correlated with,
they're not related to the number of patients
at that site or the number of visits at that side,
so you'll see relatively small sites
with a lot of reminders
and relatively big sites
with very few reminders,
and some of this,
you know is essentially kind of you know
you could see these increases
at certain points in time you know
and you know grounded
by that anecdote I just told you
where you know.