Hochberg – Prone Positioning in ARDS

Dr. Chad Hochberg is an assistant professor in pulmonary and critical care medicine at Johns Hopkins University. In his lecture today, he will discuss his work on prone positioning in […]

Episode Transcript

(00:01):
It's my distinct pleasure to introduce, today for
our speaker, doctor Chad Hochberg.
Doctor Hochberg did his early training, I think,
at University of Chicago before he came to
Johns Hopkins where he did his training in
internal medicine and then stayed for his fellowship
in pulmonary critical care.
I was really fortunate to kind of cross
paths with Chad serendipitously,
saw him present kind of

(00:22):
a late in his fellowship grand rounds that
I thought was quite impressive, the work that
he had done and some of the training
he had gone through during his fellowship.
And and so I was delighted to have
Chad, come here and speak today about some
of his work in in prone positioning in
ARDS.
A lot of this work is really talking
about kind of the trends in prone positioning
and what have we learned and and why
are are people continuing to do or not

(00:42):
do prone position as you might expect. So,
Chad, I'm really happy to have you here
today. And without further ado, I'll let you
take it away.
Great. Thanks so much, Jacqueline. Yeah. It's a
real, pleasure to speak with you guys today
and certainly a long term fan of the
Maryland Critical Care Project, so excited to contribute.
I have no disclosures, financial conflicts, and interest

(01:04):
to disclose. I did have some funding that
supported,
some of this, early work and and now
ongoing work. And I have an intellectual disclosure
that I do believe, prone positioning reduces mortality,
but in the right, patients at the right
time.
So here's an overview of what what I'll
talk about today. I'm gonna start with the
premise that there are, evidence practice gaps in

(01:24):
critical care and give you a little bit
about why they may matter.
I'm then gonna talk about the adoption of
prone ventilation in COVID 19,
talk about some adaptations or how the therapy
was changed during COVID 19,
and then finally end on some reflections on
on what this might mean about how we
can optimize this therapy or how we can
learn from this example,

(01:44):
as we seek to optimize other therapies in
our in our ICUs.
So I wanted to start with a a
little bit of an overview of my research
journey because I think it speaks a little
bit of of how I got interested in
in this type of work.
This is an outline of, my fellowship where
I started my clinical year at Johns Hopkins
in 2018.

(02:05):
And really as a clinical fellow, I think
here is no different than anywhere else. The
goal is to learn best practices,
to learn the evidence and learn which patients,
to apply it to,
and and become very facile with it.
I then actually went back to my internal
medicine training program,
after my 1st year of fellowship, and I
was one of the chief residents, in the

(02:27):
Hopkins training program.
And there my responsibility is teaching best practices.
Now this was internal medicine, but,
I usually took the opportunity to teach more
about pulmonary if I could in critical care.
As you may recall, this was not a
typical year in 2019,
2020, and that COVID came to our shores,
in March of 2020.

(02:48):
And as you can imagine, as as one
of the leadership,
folks in the residency program, a lot of
my time was spent sort of re or
transitioning the residency.
A lot of my time was spent caring
for patients with,
at that time, a novel viral illness. And
I got really interested
in how to change practice and how we
apply evidence even in in the times of

(03:09):
great uncertainty,
to patients that were taken care of. So
that really launched me on my interest in
in research. And
I did a master's in health science or
clinical research,
at the Johns Hopkins Public School of Health.
And, really, there, the focus was thinking about
how evidence is generated, how it's evaluated.
And then I've moved into the work that

(03:29):
I did throughout my fellowship and then doing
now in my early time on faculty, which
is thinking about how we identify
both wanted and unwanted practice changes and practice
variation in our ICUs,
and then thinking about how not only can
we describe that, but how can we build
towards interventions that impact that.
Through this work, I've gotten interested in, using

(03:51):
the tools of implementation science. So I wanted
to give just a brief overview of of
how I conceptualize implementation science.
This is a figure adapted from the NHLBI
that,
conceptualized research along a spectrum.
So you start with basic and preclinical findings
in a lab, either studying
normal,
biologic processes or or pathophysiologic

(04:13):
processes,
discovering molecules and how they interact and, drug
development.
For those discoveries to have an impact in
clinical medicine, they have to make it into
clinical and population research.
We do trials to
study new drugs or trials studying new interventions.
And once we've proven that something's,
efficacious, so it has an effect for it

(04:35):
to actually be effective in the real world
and impact populations, it has to translate into
clinical practice.
And this is where implementation research really seeks
to to have its impact is to
act on this sort of second translational step
in the research trajectory.
And this matters for a couple reasons.

(04:57):
Certainly, if key discoveries don't make into practice,
it matters to an individual,
who's receiving care if they're not receiving an
indicated therapy.
It matters to society in terms of population
outcomes might not be optimized, if people aren't
receiving optimal care. In translational discoveries
that eventually make it into clinical practice. It
matters for funders who fund the basic
in translational discoveries that eventually make it into

(05:17):
clinical practice. If there's discoveries that are promising
that,
can't make that second translational step, that's one
way of considering, research waste.
Certainly, the ICU is a place where there,
are evidence practice gaps, and I'm gonna point
out a few of them as we move
into to my interest specifically in prone positioning.

(05:37):
So I think this is probably near and
dear to, the heart of many folks on
the call.
This is a slide showing the evidence in
2,000 arising from, the low tidal volume ventilation
trial, the ARMA trial from the acute respiratory
distress syndrome network,
showing that, tidal volumes of 6 to 8
milliliters per kilogram of predicted body weight compared
to traditional tidal volumes led to a mortality

(05:59):
benefit.
Then fast forward, though, more than a decade
later, so this is practice in 2014,
and this is data from,
the LungSafe study, which was a large international
observational study looking at ARDS practices.
And what this figure shows is the cumulative
frequency of, tidal volumes on days 1 and
2, of ARDS.
And it may not be projecting very well,

(06:20):
but this sort
of shaded out portion here,
right at 8 cc's going up to about
60%.
This indicates that 60% of patients were receiving
tidal volumes of less than 8 cc's per
kg or predicted body weight. On the other
hand, that means about 40% are still receiving
tidal volumes above that range, which is above
what was studied in ARMA, in the intervention

(06:42):
arm.
And if you look at tidal volumes that
are sort of now more,
commonly targeted in our patients early in ARDS,
6 cc's or less, less than 20% of
patients are receiving them.
What about with prone positioning?
Well, here's a a brief snapshot of some
of the landmark evidence. This is the 20
13, PROCIVA trial done in, franchise CUs that

(07:04):
showed that early prolonged protein, so 16 hours
a day
or longer
in patients with moderate to severe ARDS
associated with the really marked mortality benefit, 17%
absolute decrease in mortality.
And, certainly, you can see with the survival
curve with the prone group in red, the
supine group in blue,
a nice separation there.

(07:25):
So this is 2013.
How did that impact practice?
Well, there was less time between,
Perceva and
some of these studies than,
the ARMA trial and the study I pointed
out before.
But these are examples of several,
either large international, like LungSafe or Apranet,
or a large national. The Dwan study is

(07:45):
a Canadian study. Sage was US practice
showing the percent of patients that were prone
with ARDS,
and less than 15% was that was the
highest in in the APRA NET study, but
every other study is less than that. So
showing a really significant
gap between the Perceva evidence
and, practice.
There's a couple reasons positive in the literature

(08:07):
about why protein adoption
historically was so low.
One potential is that there's underdiagnosis
of ARDS. So if clinicians aren't,
recognizing the syndrome and naming the syndrome, they
may be less likely to then,
use therapies that are shown to be a
benefit in that syndrome.

(08:27):
There's a preference for other adjuncts.
This,
snippet here is from a critical care perspective
by doctors Lee and senior author Kavanaugh,
unproven and expensive before proven and cheap. And
what this described was the phenomenon that was
quite common in the pre COVID era of
folks that went on to be supported
by ECMO,
actually still infrequently receive prone positioning trials before

(08:51):
going on to ECMO.
At that time, the evidence for ECMO was
certainly less robust than the evidence for prone
positioning.
So this was a certain example of people
using a therapy that was maybe more
novel or exciting to them, but certainly more
expensive, certainly more high risk, and with less
robust evidence.

(09:13):
There's a perception of proning as a labor
and resource intensive therapy.
Surveys from Massachusetts ICUs, for example, have showed
that,
many hospitals sort of say that, yeah, it's
something we'd like to offer, but we don't
have, you know, the special bed to do
proning, or we don't have the staff to
do proning. So many many hospitals and ICUs

(09:34):
had seen it as something that was not
achievable in their,
context.
Then lastly, I think many clinicians consider it
as a salvage therapy, something only to be
used in the in the patients with the
most refractory of hypoxemia.
So it was in this context of thinking
about evidence to practice gaps, that COVID 19,
came along.

(09:55):
In in COVID 19, my own experience and
certainly experience that I'm sure everyone on this
call and clinicians worldwide,
it really gave us a concentrated experience of
caring for ARDS patients. And,
in my own work and and clinical work,
I'd certainly noticed how this changed the the
the culture in the unit that I was
working in, how it changed the types of

(10:16):
therapies that we were using or at least
pulling out more quickly.
And this influenced some of my early, research
questions about, evidence based practices and COVID 19,
One being how much and how quickly is
the use of evidence based prone positioning change?
If it has changed, what factors have allowed
this greater pruning uptake? And then thinking again

(10:36):
down the road is is how can we
learn from this? How can we use this
experience to further optimize this therapy?
So address these questions, I, took an approach
of using,
big data for for perhaps lack of a
a better word,
And I'm just gonna sort of outline some
of the approaches that I'm using have used
in my work and using my work going

(10:57):
forward that I think are important for monitoring
implementation.
So we have a
initiative here called the Precision Medicine Analytics Platform,
which is essentially
a way of having a centralized relational database
that can integrate multiple streams of data so
that researchers
can
use it for their work. So one stream

(11:19):
of data and probably the most, prevalent in
this source right now is EPIC electronic medical
record data.
But we also have imaging data feeding into
this and data from our ICU monitors. And
there's also option to bring in other wearable
monitors that people may be using in the
inventory setting.
Through this platform, we've had, 2 resources, which
I've used in my work and will be

(11:40):
part of the studies that I described next.
One is the Johns Hopkins crown registry, which
is a registry of all COVID encounters across
the Johns Hopkins health system.
And the next is a resource that I'm,
building with, my, the PI, doctor, Hager, who's
our MICU director, and, Jack Awashnia,
another of our faculty members. And this is

(12:01):
a critical care research database,
that's actively accruing patients. So it starts in
2017, and it continues to update each week
with new admissions.
And it seeks to capture every patient that's
admitted to either an intensive care unit or
an intermediate care unit across our, 6 hospital
system.
Just to give you a snapshot of the
scope of the data, through September, we have

(12:23):
about 88,000
admissions and close to 70,000 unique patients.
There's a lot of advantages to working with
this kind of data with EMR,
data.
It can be,
quite granular. It doesn't have to be, but
there's the ability to describe things at a
a very fine level.
And in terms of monitoring implementation, it can

(12:45):
it can you can update these data streams
in near real time. And so,
rather than do a multicenter study years later
and and show that practice
has not cut up,
over time, you can actually look month to
month and see how your your ICUs are
doing in terms of providing this type of
care you'd like to provide.

(13:05):
There's certainly disadvantages, though. The data
in this registry from the EMR registries are
are generally collected for routine clinical care, so
you have to think really carefully of how
a data point ends up in the registry
that a clinician decided to measure something.
It was entered. It was generally entered by
a human, not always. Some of it's electronically

(13:26):
entered.
But you have to be wary of the
sources of error and, use algorithmic
approaches to try to minimize those sources of
bias.
And then, ultimately, we'd love to expand this
work, to multiple systems,
which I think is quite possible that there's
there's some coordination,
that's needed for,
either data sharing or use of, common data

(13:46):
formats, which I think is a big wave
of the future, to do these multisystem studies.
So we did use our EMR data to
track, pruning use.
I apologize if the clinical flow sheet is
is triggering anyone just coming off service, perhaps
like myself.
But this is a a snapshot of the
comprehensive flu sheet that I see when I'm
taking care of patients in the ICU,

(14:08):
in our EPIC medical record. I know Maryland
uses the same record.
This is an example of probably a pretty
typical patient early in the COVID pandemic. They
are, prone, as you can see in the
patient reposition,
indicator row.
They're on high doses of Fentanyl and a
Benzodiazepine
as well as a paralytic

(14:29):
drip.
And so in one of our studies, we
had 512 patients. To give you a sense
of the scope of the data, we were
able to extract around 90,000
positions that are all recorded during mechanical ventilation.
And by ordering these temporarily, so by time,
you can define transition periods where someone goes
from a position that's supine to prone, and

(14:52):
then back again.
We were not confident that this was gonna
work. It is an approach to define,
prone positioning.
So we did pretty extensive validation in our
early work and found that amongst these
512 patients,
it very accurately classified pronate yes or no.
510 out of the 512 patients were,

(15:14):
correctly classified.
And even the the one one of the
patients who was not correctly classified,
they were prone for a renal biopsy, and
that's how they ended up getting triggered as
as prone in our, system, but they weren't
really pruned as part of a therapeutic approach
for ARDS.
So with that background, I'm gonna tell you
about the first study we did, which is

(15:35):
a it was a retrospective observational study of
pruning practice across the Johns Hopkins health system.
And we included 5 of our 6 high
hospitals, so the Johns Hopkins Hospital,
Johns Hopkins Bayview,
Sibley Memorial Suburban in Howard County. These are
the 5 hospitals that take care of adult
patients, and there's another children's hospital in Florida

(15:56):
that we have data for, but we were,
interested in adult patients.
And we compared 2 cohorts or 2 groups.
So one,
with COVID 19 ARDS in the 1st year
of the pandemic,
and then the second, group that we called
historic,
ARDS. These were patients with ARDS
pre pandemic, 2018, 2019,
and we sourced them from a cohort of

(16:17):
patients with, an ICD 10 diagnosis of pneumonia.
We also wanted to make sure that we
are in fact measuring ARDS. So in addition
to some algorithmic processing where we,
broke the cohort down to patients that appeared
to meet oxygenation criteria for ARDS.
We had clinicians review,
each of these charts and review that Berlin

(16:37):
criteria for ARDS were met.
Some key inclusion criteria,
were around oxygenation
and timing. And, really, our goal here was
to try to to find a cohort that
looked,
like the types of patients that would be
in the Perceva criteria
to really try to get at, which patients
are,
receiving sort of gold standard evidence, which is

(16:59):
the types of patients that perceive a trial
actually receiving the intervention from the PROCIVA trial,
which is pruning.
So we looked at moderate to severe ARDS
with these,
indices,
and they had to meet this early in
mechanical ventilation.
Some key exclusion criteria, we excluded people that
were intubated outside hospitals because we couldn't,
determine their their time on ventilation,

(17:20):
because of that fact. And we wanted to
exclude patients on chronic mechanical ventilation.
The best way we found of doing that
in our data source was just excluding those
with tracheostomy
as the first airway.
Certainly, it's possible that a tracheostomy is the
first airway for some patients if they,
have an airway obstructing
problem, but,

(17:41):
very unlikely. So that's how we excluded for
this group.
And then our primary outcome was looking at,
proning within 48 hours of meeting early criteria.
Again, trying to get a really a similar
type of population intervention that would have been,
shown in the PROCIVA trial.
So here's a brief look at the the
population characteristics. We ended up with a 123

(18:03):
patients in this historic ARDS group and,
390 or so in the COVID 19 group.
The patients were in their sixties.
In terms of gender, race, and BMI,
some of the risk factors early in the
COVID pandemic for severe COVID ARDS are, on
demonstration here. There were
more males in the COVID 19 group.

(18:25):
Unfortunate,
marked increase in in nonwhite race sort of
recapitulating what we know about, the disparity in
terms of who was impacted with severe COVID,
and, heavier patients as well.
The ratios were pretty similar and both in
the severe range. In terms of the median,
SOWFA scores were on the high side. This
is just the nonrespiratory SOWFA score. So the
respiratory SOWFA scores here would have all been

(18:47):
4.
And then mortality was right in line, with
what we'd expect for moderate to severe ARDS,
in the the 40% range.
And here's what we found. So
to orient you to this figure, the x
axis is study quarter, so 3 month period
starting in 2018,

(19:07):
and going through 2021.
And on the y axis is the percent
of patients that receive early pronates, so proned
within 48 hours.
You can see in the pre COVID period,
kinda right in line with historic evidence, less
than 20% of patients, more like, you know,
10 to 15
are receiving proning in any
one quarter and
both statistically. And you can see sort of

(19:29):
here visually, there was no discernible
trend over time. It's not that proning was
increasing over time, and then we saw a
big increase during COVID.
It was really flat.
And then right at the beginning of the
COVID pandemic, so q one 2020, this is
actually March of 2020,
60% of patients that meet criteria for proning
with COVID 19 ARDS,

(19:49):
received proning, and that appeared to be sustained,
over the 1st year.
So a really, you know, clear
change, in practice, and the the change happened
suddenly. It wasn't, something that was happening over
time.
We also were interested in looking in some
prespecified subgroups to try to get at,
where is there still variation. And, you know,

(20:11):
to be honest, we didn't find that much
variation across our 5 hospital health system.
This figure shows the absolute percentage increase in
proning from pre pandemic to post pandemic
in a couple different subgroups. So moderate to
severe AS were both, statistically similar.
Academic and community hospitals, this was a finding
that we were, surprised about. We thought that

(20:33):
maybe our academic hospitals would have
staffing that would allow for a a greater
increase in proning, but it was really the
same.
That being said, our medical ICUs that,
perhaps a more preexisting experience taking care of
patients with ARDS
had a larger percent increase in prone positioning
than our nonmedical ICUs.
There were not statistically significant differences in terms

(20:56):
of interaction between the BMI categories.
But I'll just note that the the BMI
of 30 to 50, so overweight but not
severely,
obese patients
had the greatest absolute increase in proning. This
is notable given that, body has been
at least positive as a barrier to using
proning in the past.

(21:17):
One of the benefits again of, I think,
using an EMR,
registry approach to monitoring implementation and tracking practices
is that a few years later, just last
year, we were interested in,
understanding, really, is this a practice that we
saw this big increase and now our ICUs
are very adept at using proning and everyone's
gonna be,
using proning going forward, or,

(21:38):
what kind of trends have there been in
practice?
So we looked then
at every year, in patients with COVID 19
ARDS,
2020 through 2020 2. You can see, as
the pandemic has waned, cases have dropped off,
fortunately.
There's a little bit of a a slight
distribution towards higher age in 2022.
The Sofa scores slightly shift, to a slightly

(22:00):
higher distribution as well in the later years.
Severe ARDS is pretty similar across all years.
And then a finding which we we found
here and has been shown actually in a
number of different health systems
is that severe ARDS later in the pandemic
in 2021,
2022,
do appear to have, higher mortality.

(22:23):
And what we found was that sustainment,
is a problem. So this figure is plotting,
6 month periods by year on the x
axis and then prone positioning,
in the y axis within 48 hours.
And you can see in 2020, 2021,
proning is being used at a at a
high level and is sustained at around that

(22:44):
60% of eligible patients is all eligible patients.
And then in 2022, we see a marked
drop off,
to the point where from July to December
in 2022,
we're back, at less than 20%. So this
is the historic
rate of proning both in our health system
and, very consistent what's been shown in the
literature elsewhere.

(23:06):
We, did perform adjusted analysis that we can
only adjust for things that we measured.
So I think there probably are unmeasured things
that are,
accounting for some of this difference, but things
like age, BMI,
trials and comorbidities,
severity of hypoxemia,
and PEEP,
this this marked drop off persisted about a

(23:27):
40% decline adjusted rate ratio,
2022 versus 2020.
So let me transition a little bit to
the work that we've done to really try
to understand the why. Because it's one thing
to demonstrate,
changes,
but to then sort of move to the
next step, which is to actually impact changes

(23:47):
going forward in terms of how ICUs practice,
you have to understand how practices change.
And a real motivating question as I went
into this work was
great COVID change practice, but COVID also,
you know, aimed and killed,
thousands and thousands of people. So it'd be
nice to be able to change our cultures
without

(24:08):
the shock of a worldwide pandemic,
to understand the why we used a a
qualitative research approach.
And here, I just have a clip of
a a title for an article that I
found very helpful as I, went on to
this research journey,
talking about how qualitative research can contribute to
research in intensive care unit.
Qualitative research
generally is well suited to

(24:31):
measuring sort of complex phenomena or why people
do what they do.
Quantitative measures are good for measuring what people
do, but the why is is harder to
get at with some of those quantitative data.
So one of the first studies we did
was a a study
of the facilitators of, prone positioning during COVID
19. So we did semi structured interviews, which

(24:52):
is a a method where we have an
interview guide that sort of outlines some of
the main questions that we wanna ask people,
and we try to make the questions very
open ended,
so that, information can emerge that we may
be surprised by if it's present.
And we interviewed 40 ICU clinicians from 3
academic ICUs. This is our Johns Hopkins Hospital

(25:12):
and and Bayview Medical Center.
And I interviewed,
clinicians that would have been involved in proning,
so our physicians,
our advanced practice providers, our nurses, respiratory therapists.
And at Hopkins Hospital, we did have physical
therapists that were part of a PRIM team,
at least during the height of the pandemic,
and and we interviewed them as well.

(25:33):
Our interview guide, and then subsequent analysis was
structured with an implementation framework,
or implementation science tool called the consolidated framework
for implementation research. This is commonly used, and
it essentially helps researchers and
readers of implementation science work just really conceptualize
the the main features that can impact implementation.

(25:55):
So whether it's individuals working in a unit,
whether it's the units themselves, whether it's the
hospitals, larger factors, and it helps just make
sure you're hitting on on all those different
levels.
So what we found,
were several themes that emerged as as the
most important things in terms of why pruning
had changed in our setting.

(26:18):
The first was shifting individual and ICU attitudes
towards pruning.
So here's a quote from one of our
participants noting that, they now see proning as
a potentially high benefit, low risk intervention.
So instead of asking, should we prone, they're
really asking, is there a reason I shouldn't?
So a shift in,
the way they're approaching therapy.
And don't have a quote to share on

(26:39):
this here, but another notable shift was just
in team approaches to proning,
particularly
nursing and respiratory therapist teams,
became very enthusiastic,
about proning. It's not that they had not
believed that it worked before, but I think
they were much more willing, to do it
and to do it quickly.
And so, nurses talked about that. Physicians talked

(27:02):
about sort of mentioning that they would like
to have the patient permanent and then having
it done,
you know, before they were on to the
next patient, which was very different than the
pre pandemic era.
There had to be the availability of basic
but important equipment.
And I just wanna emphasize this was simple
equipment. This was not rodent prone beds

(27:22):
or, any type of, fancy or expensive equipment.
This is just making sure that the basic
stuff that
allows nurses to take the excellent care of
patients they do, was available. So things like
bandages to protect bony provinces,
pressure redistributing pills or what are called Zflow
pillows here,
to prevent, pressure ulcers on the face.

(27:44):
Just having that available and and stock in
ICUs helped, our nursing teams particular feel really
comfortable,
doing this for high volumes of patients.
Our hospitals certainly did commit,
resources to pruning. So
we had an educational video and a flyer
that were prune specific and put on a
central COVID 19 resource portal and advertised. So

(28:06):
these weren't
expensive resources, but someone had to or teams
really had to have the foresight to to
put that out there, and it signaled to
teams that this was potentially an important
therapy to offer.
And then, of course, we did have a
an actual prone team.
This was initially
nurses, but pretty quickly transitioned to outpatient physical

(28:26):
therapists that were not currently working. And they,
in the height of the pandemic, when there
were a lot of patients pruned, they were
helping with the repositioning, the supining, the pruning.
And although it was only there for a
short period of time, it was just another
signal from the hospital. It's and this is
a therapy that is probably important for your
ICUs to be, offering.

(28:48):
Then lastly, staff led training was really important.
And by staff led,
I mean, at the bedside, just in time
training, particularly for nursing. And of all the
teams that help carry out pruning,
having a nursing team that is very, adept
at doing this and comfortable at doing this
is, one of the most important things.

(29:08):
And,
our nurses over and over told us that,
you know, really, our our senior or more
experienced nurses helped lead this change, because they
had done this before, and they're able to
bring their junior colleagues, along with them.
What also arose from some of our qualitative
work,
was this concept that proposition was adapted from

(29:30):
historic evidence and practice.
And I like to share this quote because
this is one of the quotes as I
was doing the analysis for this qualitative study.
Got me interested in the idea of
therapeutic adaptation, which I'll talk about next.
But the participant essentially said, you know, this
this isn't the PERCEIVA trial. It's not what
we're doing
in in this era, and it got me

(29:51):
thinking about how therapies change over time.
Adaptation is
a concept that's described in implementation science that
that could be defined as a process of
changing a therapy to align it with clinical
and local context.
Now I would posit that if if we
think of critical care therapies,
all therapies, but as our interest is in
critical care, particularly thinking about critical care interventions

(30:14):
or therapies,
they almost
always get adapted. It occurs constantly. Therapies are
studied in tightly controlled RCTs,
with patients that may not be fully generalizable
to one's practice,
or scenarios that may not be fully generalizable
to one's practice.
And so when you then bring in a
therapy to real world practice, it it necessarily

(30:34):
changes.
And that's not a problem per se. It's
just something that we need to be aware
of, and I think it's something that is
probably underutilized in terms of harnessing the the
power of of how things are changing.
You think about it. Some adaptations can really
alter the intervention being delivered where it ends
up being quite different than, what's studied.

(30:55):
You can then imagine that has effects
at least potentially on patient outcomes, and it
has effects on implementation of how things are
used in our ICUs.
This is understudied certainly in implementation science and
and definitely in critical care,
and I think is a potential
powerful source of really understanding areas for future

(31:16):
therapeutic optimization.
And so
sort of being interested in this idea of
adaptation,
we were interested in how pruning use had
changed in our setting and then how we
could harness those changes to understand if there's
a way to further optimize that therapy.
And one of the ways or one of
the things we focused on was this concept
of, extended prone positioning. That is leaving someone

(31:39):
in the prone position for much longer than
at least what's been historically done in RCTs,
which is around, you know, 12, 16 hours.
And in clinical practice, historically,
16 hours with daily supination was the the
most commonly used
strategy.
Here's a quote from one of our participants
saying, you know, protein is not a rotisserie.

(32:00):
They're not up for flipping back and forth.
They think it adds to the labor of
the team and
not necessarily consistent with the physiology. So sort
of advocating for a a at least consideration
of a of a prolonged approach.
So I'll say that extended duration pruning is
happening,
both at our center and others.

(32:21):
This chart is a, a density chart just
showing the distribution of the time of the
first proning session or the duration of the
first proning session
across our 5 hospitals.
And it's a little bit of a of
a busy figure, but I I just wanna
point out a couple things. One is that
there's there is variation,
between,
hospitals and within hospitals.

(32:42):
This big spike here is Howard, one of
our community hospitals, Howard County General,
showing that they're probably for
for the vast majority of their patients are
really in this 12 to 16 hour range,
right around 20 hours is the peak here.
And then hidden in the back, but a
much broader distribution,
is
JHH, one of our academic hospitals, Johns Hopkins

(33:04):
Hospital here in this purple and hidden behind
it, Bayview. And you can see that their
peaks in terms of the the most common,
duration is is much closer to 30 hours
or even 48 hours.
And you'll notice that this graph gets cut
off at a 120 plus, so there were
certainly patients that were still prone to 5
days in in continuing.

(33:26):
There's some early data showing, an association of
benefit with this type of extended or standard
duration pruning sessions from a paper,
done at Harvard by, Oken and colleagues,
published in CHEST.
And we were,
interested in in sort of furthering this work.
Why might extended duration pruning be a good
idea?

(33:47):
Well, we think of pruning as, a strategy
to enhance lung protective ventilation
and therefore reduce
ventilator,
induced lung injury.
And
I guess a concept that you can think
of with extended proning is if a little
bit is good, maybe more is better.
So doing it early when the lung is
vulnerable and giving a larger dose. And a
larger dose, there's a couple ways to accomplish

(34:09):
it. 1 would just be by leaving the
patient pruned longer.
One would be by continuing pruning sessions further
out through the course,
but that's a little harder to conceptualize as
patients may be improving and, no longer,
are indicated to be pruned.
There's also an implementation argument for, potentially extended
duration pruning. This is the less strain in

(34:30):
ICU resources of the staff. If they're not
having to do daily supinations and pronations,
staff are are freed up to do and
concentrate and focus on other things for the
patients they're taking care of.
But there's certainly drawbacks to think of.
One I I worry about a lot is
delayed de escalation,
in patients that are otherwise improving. So if

(34:51):
you think of the patients that that the
patient that's pruned,
maybe they have sort of a middling
ratio. It's, you know, in the 1 twenties
to 140, 150.
And so people say, well, look. I'm gonna
supinate them. They're still gonna meet pruning criteria.
Why why even go through that dance?
Well, maybe we're surprised. Maybe we supinate them.
They're improving, and now you can stop the,

(35:12):
deep sedation if that's what you were using
or paralysis if they were on neuromuscular blockade,
and can start letting them wake up and
do some spontaneous breathing and and move towards
ventilator liberation.
Additionally, the longer time you're pruned, the more
exposure you have to the adverse effects of
pruning. So pressure wounds,
airway complications,

(35:34):
feeding intolerance,
all reasons to to be wary of of
just doing extended duration pruning.
So we are interested in examining this in
our setting,
and we took a target trial, emulation approach.
So I'm just gonna take a moment to
talk about, what this approach is and why
one would want to use it,

(35:54):
and how we did it. So a target
trial emulation approach is a established approach of
using observational data to generate some causal estimates.
So often we're told and all throughout our
training that
association does not equal causation.
This is a way of being explicit that
what we're interested in in examining

(36:15):
is the causal associate the causal
impact of something.
And a target trial emulation is established for
use when either a randomized controlled trial is
unethical, so it will never be done,
infeasible,
also meaning that it will never be done,
or not forthcoming in a in a timely
manner. And I'll say that, to my knowledge,

(36:36):
there's not a a trial
underway last time I checked in, in extended
duration,
prone positioning, yet people are using it. So
it's a a timely clinical question.
When you're approaching a target trial emulation,
it it's actually
not so complex statistically. It's more of a
conceptual way of going through to try to

(36:57):
avoid the common errors that can arise in
observational data analyses.
And the the key
concept behind it is specifying the randomized controlled
trial that you would like to do that
won't be done, and then doing your best
to emulate it using observational data.
And by carefully emulating by specifying and then

(37:18):
carefully emulating it,
you're very clear on the timing of,
of eligibility and the timing of exposure and
the timing of outcome,
which is something that can commonly
actually get quite confusing
and
source of bias and observational data.
And you also
separate a little bit the design of the
study from

(37:38):
the analysis of the study.
And so it's, it can be more rigorous
from that standpoint.
So I'm gonna walk through now the hypothetical
target trial that we would like to do
and then how we've emulated in our observational
data. And I'll say this is working in
progress. We have manuscript,
in preparation now.
We hope to have this out,

(37:58):
soonish.
So for our hypothetical target trial,
I would want to randomize
adults with moderate to severe. For this trial,
we focused on COVID 19 ARDS as a
practical consideration,
for patients on mechanical ventilation.
We wanted patients to meet price procedure
and to do so early. So,

(38:20):
they're prone within 72 hours mechanical ventilation, really
trying to to get that early severe
ARDS
phenotype into our trial. So how do we
emulate this observational data?
Relatively straightforward, we had COVID 19 ARES that
we pragmatically defined as a a PDAF ratio
of less than 150.
F r two greater than 0.6 while on

(38:41):
the ventilator within the first, 72 hours.
The intervention
for our trial is that we were gonna
randomize just the 1st pruning session.
And a couple reasons that we focus just
on the 1st pruning session.
One is there's a a biologic rationale.
We know from animal studies and,

(39:02):
nicely done physiologic studies in humans that pruning
later in the course seems to have
less of the beneficial at least measured
physiologic effects than pruning early in the course.
Once the lung is more fibrotic, more stiff,
we we think there's potentially
less benefit.
So we wanted to target people whose lung
injury is
both vulnerable to worsening and potentially,

(39:24):
somewhat reversible.
So that's why the first pruning session was
randomized, and then there's a feasibility consideration.
It's much easier to get clinicians to adhere
to an upfront
strategy rather than something that you're gonna say
now you're sort of you're stuck to this,
and it's hard to to keep people prone
for for days days at a time.
So with this randomization, the first pruning session,

(39:46):
the two strategies we would study are proning
to continue 24 hours or greater. So
patient's pruned. You leave them pruned at least
through 24 hours, and then clinicians can decide
when to supinate.
Or a standard duration
pruning
where patients are supinated between hour 16 and
24. So that's much more in line with
what's been studied in in, say, the PRECIVA

(40:06):
trial, which was 16 hours pruned.
The average in the study was was 17
hours. Patients were supined and then prone again
within 4 hours if they still met criteria.
And here's where we had to use some
statistical tools to
in our emulation.
So one,

(40:27):
we had the intervention assigned as observed in
data. So we just took the 1st pruning
session,
and we assigned it as was this was
a standard duration pruning session or extended duration
pruning session. But then, of course, there's selection
bias. Right? Clinicians are deciding who's getting extended,
who's getting standard. So how do we deal
with that in the observational design?
We gotta do something to simulate randomization.

(40:49):
What randomization does is it balances confounders.
And so we used a technique called inverse
probability of treatment weighting. It's a propensity score
method where
you
estimate the probability that someone receives the treatment
they receive based on measured,
known confounders,
and then you weight the data by the
inverse of that probability,

(41:10):
essentially creating a a pseudo population,
of people that got,
the the counterfactual
treatment.
And then our outcome for this study would
be all cause
mortality by day 90, same in our observational,
emulation.
So here's the early,
preliminary findings from the study. This is the

(41:31):
study population. We have extended pruning 234
patients,
standard proning, just 80 patients, a lot less
in our system did receive this therapy.
Again, age in the sixties consistent with what
we saw with our prior studies,
predominance of,
males versus females.
Here, the ratio was a little more, severe.

(41:52):
We're a little more restrictive about requiring,
when exactly we
And you can see that
70 this is the interquartile range. So 75%
of patients are
p to f's in the mid nineties or
less.
And then highlighted in blue are where there's
some serious imbalances between the groups that you'd

(42:13):
have to be very careful of when you're,
analyzing this.
So 87% of patients at academic hospital,
were proned,
in their sorry, in the extended proning group,
and then 26%
in the standard group. So a major imbalance
there and also an imbalance in some other
practice aspects that may be related to this

(42:34):
this confounding by site.
So timing from mechanical ventilation from admission, excuse
me, to when someone is put on the
ventilator
about 20 hours longer in the standard group.
So this is the unweighted cohort. We then
applied our waiting strategy.
And after you apply the waiting,
to sort of assure
yourself that you've

(42:54):
done your best to try to balance things
that you know are confounders,
you can examine that. So this is a
plot that they're called love plots.
And what they show is on the the
y axis are absolute mean differences,
which means that for continuous variables, it's the
standardized difference, which is the mean divided by
the the difference in means divided by the

(43:15):
standard deviation
between the 2,
cohorts, the 2 groups.
And then for categorical variables, it's a simple
difference in proportion.
And what we're showing here is that the
unweighted cohort is this, orangish,
color, and then the weighted cohort is the
teal color.
This horizontal dotted line at point 1,

(43:37):
represents
a established
or accepted
threshold below which balance is considered adequate,
so point 1. And you can see that
a lot of imbalance
in,
the unweighted cohort. And then after weighting, all
the that we had in the model
are balanced.
Now these are just known confounders. There's certainly

(44:00):
still always concern for unmeasured
confounding,
but at least this is reassuring,
that we've been able to balance some of
the things that we know are are different.
Just to give you a sense of how
this work is working, like I said, in
the in the unweighted cohort, 87%
of the extended group are pruned at the
academic hospital and 26%

(44:20):
at this at the,
of the standard group. And so you can
see that difference of 50% or so, and
now that difference is is much lower and
sort of controlled for, with this approach.
So what do we find?
This is the primary outcome of, 90 day,
mortality, and this is actually all location mortality

(44:40):
because we were able to link to death
national death records.
So it's both in hospital and even patients
that left the hospital and died, elsewhere.
And what we see are the extended group
in blue, this top line, and the standard
group in orange, this bottom line. And in
the unweighted cohort, this is just the data
as they are, certainly a large difference.

(45:01):
So you would like to be the type
patient that received
extended proning, but this is observational,
unadjusted
data.
And this is why we do things, as
carefully as we can is once we weight,
the data
and try to do our best to estimate
a causal effect,
recognizing limitations,

(45:21):
we see no difference.
So now you see the curves are are
much more similar to one another. Actually, the
standard curve is now the the top curve,
for this patient. So a word of caution
about, retrospective observational effect sizes,
in research.
And then our secondary outcomes were time to

(45:42):
being off a ventilator and alive
at day 90
and time to ICU discharge also at day
90.
And, again, we see the same type of
pattern where in the unweighted group,
You'd much rather be the type of patient
that receives extended patient extended proning.
They get off the ventilator faster. They get
out of the ICU faster, and this accounts

(46:04):
for,
the competing risk of death. But once you
weight the cohort and control for that confounding,
there's there's no differences.
And then the last piece of data before
we wrap up is, this is just the,
these results in regression
form.
These are the weighted results or our best
estimate of the causal,

(46:25):
effect in terms of time to mortality at
day 90,
extended for standard pruning.
Hazard ratio is 1.
No finding. No difference. I will note the
confidence intervals are wide, so there's there's some
imprecision here.
The time to ventilator liberation accounting for the
competing risk of death extended versus standard. The
subdistribution
hazard ratio,

(46:46):
because it's a competing risk analysis, is, 1.44,
which actually favors
the extended group.
Here, you're essentially
saying the the risk of getting off the
ventilator is higher in the extended group, but
not statistically significant wide confidence intervals,
same with with time to ICU discharge.

(47:06):
So our conclusion is that there's no significant
differences in 90 day all location mortality, time
ventilator to ventilator liberation, or ICU discharge.
Although I'd say I think further studies,
needed. This is our best attempt at coming
up with the causal estimate, but it's not
gold standard causal data.
There's,
I think, still reasons to to consider extended

(47:28):
proning from an implementation standpoint, being able to
use it more easily in ICUs.
And there may be patients who are, you
know, you think are really certain to be
reproned. And, I think people still may make
the argument that I'm just gonna leave this
patient pruned a little longer.
So how are we gonna fill the gap
and and really understand if this is beneficial
or not or if it's harmful? I think

(47:50):
careful physiologic studies
certainly have a role here. Things like electric
impedance tomography, lung imaging that helped us to
understand how the lung state is changing later
on in these sessions.
Perhaps someone's prone for a very long period
of time. You now have,
a degree of anterior atelectasis
and post and dorsal hyperinflation,

(48:10):
that is it doesn't appear to be beneficial
anymore. And so really trying to get to
that optimal prony duration.
Ideally, this would be something that we could
study in a randomized controlled trial,
but I think there's there I would have
a lot of concerns about the feasibility
of this study, and I'm not sure it's
something that that practically could ever get done.

(48:30):
So I'm gonna end with just 3, takeaway
points and future directions. And I should have
said from the beginning, I'm happy to take
any questions,
if they come up.
So my first takeaway point is that while
implementation
gaps exist in critical care,
change is possible. And,
we saw changes
during the COVID pandemic,

(48:51):
some of which were beneficial for our ICUs
in terms of use of evidence based therapy.
There's others that probably are not beneficial, but
culture can change, practice can change. So we
wanna be intentional about how we do it.
So some of my future work for for
an ongoing,
k 23
grant is
developing implementation interventions that both support adoption,

(49:12):
but also try to hit that sustainability piece.
So it's one thing to sort of do
something and get people to to do the
therapy, but if it doesn't make it into
sustained practice, it doesn't have the
possibility of helping patients in the future.
2 is that health informatics and EMR data
really can provide a powerful source of of

(49:32):
near real time information
on practice shifts.
So if the prior paradigm was,
doing large observational studies that get published 4
years after the practice is done, describing that
there's a practice that's not being done. This
will allow us to actually near real time
sort of describe how ICUs
across,

(49:53):
our our systems are are doing with providing
care.
To harness this health informatics,
we're gonna need sort of cooperation across health
systems with data sharing,
common data formats,
that we're working on now and hope hopefully
can expand this outside of single systems to

(50:13):
to multiple systems and get to some powerful
data.
And then finally, therapies and interventions are sort
of, by definition, changed by the context they
end up being used in. They're adapted. And
one such example of this adaptation was the
much longer
protein duration
sessions that were seen.
We attempted to learn from the crowd and
understand if this was something that was

(50:41):
one study. It's just one study. There's limitations
to every study, and, of of course, I
think, further research is needed.
And I'll just end with a thank you,
to University of Maryland Critical Care Medicine for
the invitation. Doctor Levine, thank you so much
for having me. Big thank to my mentorship
team,
funders, and, happy to take any questions.