Subscribe
Share
Share
Embed
In this episode of "In the Interim…", Dr. Scott Berry is joined by Dr. Will Meurer, professor of Emergency Medicine and Neurology at the University of Michigan, for an in-depth discussion of the ICECAP trial’s adaptive Bayesian design. The discussion breaks down the scientific rationale for hypothermia after cardiac arrest, critiques legacy studies, and explores the justification for including both shockable and non-shockable rhythm types. The episode provides a detailed account of ICECAP’s methodological strategies: a weighted mRS primary endpoint, Bayesian adaptive trial structure, response-adaptive randomization (governed by strict allocation guardrails), a unique Bayesian model for duration-response, and futility rules. The trial’s development is described in the context of the ADAPT-IT initiative, an FDA/NIH partnership, and the operational leadership of the MUSC Data Coordinating Center. Results are pending publication which will be highlighted in a future episode of “In the interim…”.
Key Highlights
Key Highlights
- Rationale for exploring duration of hypothermia after cardiac arrest with review of prior evidence.
- Enrollment of shockable and non-shockable populations to address clinical uncertainty.
- Primary endpoint: weighted mRS, independently developed for ICECAP.
- Bayesian adaptive design with response-adaptive randomization, interim analyses, and allocation guardrails.
- Management of missing data with multiple imputation from 30-day outcomes.
For more, visit us at https://www.berryconsultants.com/
Creators and Guests
What is In the Interim...?
A podcast on statistical science and clinical trials.
Explore the intricacies of Bayesian statistics and adaptive clinical trials. Uncover methods that push beyond conventional paradigms, ushering in data-driven insights that enhance trial outcomes while ensuring safety and efficacy. Join us as we dive into complex medical challenges and regulatory landscapes, offering innovative solutions tailored for pharma pioneers. Featuring expertise from industry leaders, each episode is crafted to provide clarity, foster debate, and challenge mainstream perspectives, ensuring you remain at the forefront of clinical trial excellence.
Judith: Welcome to Berry's In the
Interim podcast, where we explore the
cutting edge of innovative clinical
trial design for the pharmaceutical and
medical industries, and so much more.
Let's dive in.
Scott: All right.
Welcome everybody back to In the Interim.
Today, we are going to talk
about, uh, a really cool trial.
Uh, and, and I, uh...
By the way, I'm your host,
Scott Berry, and I have a guest
today, and I'll introduce the
guest as I introduce the topic.
The topic today is the ICECAP trial,
and ICECAP is, you know, if I, if
I put together of, of doing this
for 25 years, sort of my, my five
favorite, three favorite, 10 favorite,
whatever, ICECAP is on there.
This is a, a very, very cool trial.
It's a...
I, I think the trial design,
the scenario, very cool.
And we're gonna talk
about that trial design.
And, uh, joining me to talk about that
trial de- de- design is Will Moyer.
Will is a professor of emergency
medicine and professor of neurology
at the University of Michigan.
He's also a medical and scienti-
uh, statistical scientist
here at Berry Consultants.
So Will, welcome to In the Interim.
Will Meurer: Yeah, I'm
so, uh, happy to be here.
Scott: And, uh, uh, sometimes I ask
my guests, "Are you a Bayesian?"
And if, if you have video access to this,
there are some people that have only
audio access to this, I don't need to
ask that question as, as Will is wearing
a T-shirt with Bayes' theorem on it.
So he is declaring his
allegiances just from the start.
And ICECAP is a Bayesian trial.
Uh, the final analysis is Bayesian.
The trial is Bayesian.
But let's, let's get to the ICECAP trial.
The, the story of the design of this
trial is as fun as the design itself.
So let's talk about, uh, uh, we'll, we'll
talk about the development of the trial.
But to give people a sense,
what, what is the ICECAP trial?
What is the, the syndrome?
The, what are we treating in ICECAP?
Will Meurer: Yeah.
So, you know, when we take like
high school or middle school health
class, we know that, you know, if
your brain stops getting oxygen,
um, bad things happen to it.
Somet-- You know, people are lear--
you know, maybe learn, gosh, in,
in three, four minutes, your brain
is gonna be damaged irreversibly."
Um, so when people have problems with
their heart such that their heart suddenly
stops beating, like they're having a
heart attack, there's a blocked artery,
or for whatever reason, the electrical
information in the heart just goes
haywire, um, all of a sudden there's no
blood flowing up to the, up to the brain.
You know, the body goes down um,
hopefully there's somebody nearby
if that happens to you so that they
can start CPR and potentially get
An automa-- you know, an automated
external defibrillator and restart
the heart as, as quickly as possible.
Um, when I was a medical student
at the Cincinnati VA, I was just
by the ER one night, um, and, and
literally a guy, um, went into cardiac
arrest just as we were walking by.
And somebody got the, the old school
paddles, went up to the guy and,
and zapped him, and he just...
He woke up, screamed, like, like...
And he, he, he, he-- We got to him
before he developed any brain damage.
But in, in lots of cases, um, we
don't get the heart restarted that
quickly, and because of that loss of
blood flow, the brain is, is damaged.
And, we don't know, though, at the moment
that people get the heart restarted
how damaged that brain is gonna be.
All the cells that are inside
the brain, you know, we think of
neurons, there's blood vessel cells,
there's connective cells there.
These are all hurt by this,
this lack of oxygen, but they're
not all necessarily gonna die.
the idea of treating cardiac
arrest, you know, first you
have to deal with the heart.
If you don't get the heart restarted, you,
you know, it, it's very hard to, to live.
It's, it's really sort of game over.
But if you get the heart, heart
restarted, then the most common
way people die is because of brain
injury, so that they don't wake up.
So one of the things that we know is
that when the brain is injured, um,
temperatures like fevers really harmful.
So the, you know, the way I think
about this is there's a bunch of
these neurons and other, you know,
other cells inside the brain.
They've been hurt.
They're still functioning, but they're,
they're, they're maybe not happy about it.
They haven't decided, like, should
I, should I turn myself off?
Should I go to the-- you know, hit
the self-destruct button because
the situation is bad and I'm,
I'm not gonna be able to recover?
Or should I give this a little more time?"
Um, one way of thinking of this, it's,
you know, it's an oversimplification
of biology, is that when the
temperature goes up, more of those
cells decide, you know, they're
done, that they can't survive this.
And we see this not just in, in the
brain injury from cardiac arrest,
um, but we also see this after
stroke, traumatic brain injury.
Anytime there's a serious hit to the
brain, is, um, you know, this really
common situation that fever is bad.
And these are sick patients
in the intensive care unit.
They're on mechanical ventilators, so
there's plastic and, you know, coming
out of all sorts of body orifices.
So the- they're very-- people are very
prone to infection in this fragile state,
so these, these fevers can be quite bad.
so in the early two thousands, um,
well, even before that, like way back...
Scott: let's back up a little.
So these are patients
suffer A cardiac arrest.
There are other, there are other reasons
why somebody has lack of blood flow, and
so the whole therapy that we're gonna
investigate might be interesting in there.
But in this trial, we're looking
at they suffer cardiac arrest.
For some period of time,
there wasn't blood flow.
It gets restored, so these are
patients that have been restored.
They come into the
emergency room where Dr.
Moyer is, uh, in that.
And, and then the question
is, what do you do?
Uh, do we have therapies for them?
Is that right?
Did I get that right?
Will Meurer: Yeah.
No,
Scott: Okay.
Will Meurer: that's right.
Scott: Okay.
Will Meurer: and this is like a, a...
This is damage to the whole brain.
So if you've-- You know, a lot
of people are familiar with
stroke and stroke research.
In there, i-in stroke, you
have damage to one part of the
Scott: Yeah.
Will Meurer: that case, people
usually are awake, and you can measure
how big a stroke they have by how,
you know, how weak their arm is.
You know, this is part of, you know,
what I do in my, in my daily work, or
at least, you know, a few times a, a
month when I work on the stroke team.
So stroke, you can kinda quantify
the severity with this, you know,
with like an NIH Stroke Scale exam.
You can do a special type of imaging
test, which, which, which you, you've
all worked on, a CT perfusion that
can say, "Look, this, this is the
volume of the brain that's at risk."
Whereas in cardiac arrest, we
don't have a m- we don't have a
real true marker of the severity.
It's like the person looks comatose.
They're not responding.
Maybe they're looking around a little bit,
but they're not able to follow commands.
But we don't have a sense as to
how big a hit their brain took and
whether or not they can, can recover.
Um, so that's why we wanna try to
implement therapies that, that protect
the brain or, or can, you know, limit
the damage, um, you know, that would be
what can help, help improve the outcomes.
So that's, you know,
that's sort of the stage.
You know, the patient has
been resuscitated by usually
the paramedics in the field.
Um, sometimes a bystander brings
it, like an AED, and they g-
they get the heart restarted.
And then they're- they've come
to the emergency department.
And unlike that guy at the Cincinnati
VA who woke up screaming that his
chest hurt, these folks are, you
know, on a mechanical ventilator, you
know, and really, you know, not, not
able to speak or, or follow commands.
And that would be, be the
coma after cardiac arrest
that is, is the brain injury.
Scott: Okay.
And, and the heat you described that could
accelerate neuron death and it considered
to be a negative, that's the human body
creates this response of a fever, high
temperatures, and hence the therapy.
What therapy are we investigating in
this trial is potentially beneficial for
those patients that come in to see you?
Will Meurer: Yeah.
And, and it can go even a little further.
I mean, we've heard those stories of
people who fall through ice on a lake
and are are under the water in the lake
for a half hour, to some degree, that's,
that, that coldness of that lake is,
is much colder than we would do-- w-
than we did in, in the IceCap trial.
But that, that slows down all the body's
circulatory processes, can slow down
those cell death processes, which is
why we sometimes see people who have
these, these miraculous recoveries a-
even after being under water for you
know, mind-boggling amounts of time.
addition, for some types of cardiac
surgery they, they need to put...
where they need to stop the heart
to do work on, say, the aortic arch
and, you know, to, to, to fix things,
they put people on a bypass machine.
in certain cases there, they, they
know that if they cool the body down
quite a bit before that, You know,
they-- The fact that they have them
on a bypass machine helps keep blood
flowing through the brain, but there
still is a chance that you could injure
the brain in those types of procedures.
So there are some where they will
do, um, a hypothermic treatment
during these planned surgeries.
We We also know that in, in really,
really little kids who've just been born,
who maybe have a umbilical cord wrapped
around their neck or something, and they
don't get enough oxygen during the birth
process, they have some brain damage.
And those kids, you know, if you,
if you cool those kids also down to
about 33°F on a cooling blanket, you
can limit the amount of damage they
have and, you know, increase the
chances that they live a normal life
without any developmental problems.
So there's, there's these other sort
of human cases where we've we, we've,
seen hypothermia be protective.
A lot of people have done animal
research that is more mirrors
this cardiac arrest situation that
has shown that, that you can...
This is really one of the best
ways to protect the brain.
And part of that, we think, is because
when, you know, we talk about these
cells deciding to die, there's a
lot of ways they can decide to die.
And you could think, "Oh, I can get a
drug that blocks one of those ways."
But then the, then the, you know,
the cell just goes around, you know,
basically goes around the corner and
takes a different pathway to dying.
Um, dissimilar from cancer, where it
may be like- Right this is the specific
thing, and if I block that or, you
know, a, a sort of monogenetic, you
know, neurologic disease where you're
like, "If I can stop the body from,
from making that one bad protein, I
can, I can stop the disease here."
But there's lots of ways
that the brain dies.
So the th- th- the thought is
that hypothermia maybe slows
down multiple mechanisms.
And, you know, that's, that's
borne out in the animal data, but
to some degree, we don't really
know how to implement it people.
Like, who are the people we should cool?
You don't wanna cool people who aren't
very injured because then you're
exposing them to more time in the ICU
and on sedatives and all this stuff.
on the other hand, don't wanna...
If you can protect somebody, you
wanna cool them long enough so that
You're giving that brain a chance to,
to get as much recovery as it can.
so they, they did some studies back
in two thousand that were, were
relatively small studies called the
Haka study and the Bernard study,
but they did have pretty eye-popping
results of improving mortality.
Um, they didn't use modern devices.
They more or less used ice packs
and, and cooling, cooling, cooling
blankets, fans, things like that.
But they also, in their control groups,
didn't do much to control fever.
So that was one of the criticisms of those
studies in that the effect they saw from
hypothermia more this protection from
being low, was it the fact that being a
little low was a good way to not go high?
So that, that left-- even though the
guidelines, um, that came to be after,
after two thousand and two when those
studies came out were favorable and
saying that we should cool people for
roughly twelve to twenty-four hours,
there, there were these unanswered
questions as to would this really work
in, in a, you know, bigger population
th- and also a population where the
ICU was, was more diligently treating
fever in a, in a control group.
Scott: Yeah.
Okay.
So, so walking into this, the
design of this trial and, and the,
the design process of this trial
was, it was extended, getting
funding and various things.
Uh, it's thought that hypothermic cooling
of patients that come in after cardiac
arrest, restoration of, of, uh, of the
heart, of, of blood flow, they come
in, does cooling them, it-- uh, it's
thought that cooling is beneficial.
There still remains a number of questions.
Conclusive evidence of
benefit is also not there.
So going into this trial, it is somewhat
standard of care at your University
of Michigan Hospital that if somebody
suffers from this, they will be cooled.
So what is it that ICECAP
decides to explore?
I imagine it could be temperature.
I could imagine how fast you get
them there, whether you cool them
at all, whether you just prevent
this, how long you cool them.
What is it that ICECAP
is going to explore?
Will Meurer: Yeah.
I think one other thing that, that maybe I
Scott: Hmm.
Will Meurer: out about those early studies
is one thing that they did was they
focused on sort of the patients who they
thought had the best chance of recovery,
and these were people they thought had
a cardiac arrest from cardiac causes.
And that meant when, when EMS shows up,
they're in this ventricular fibrillation
or v-ventricular tachycardia-type
rhythm, like my guy at the VA.
Um, they get shocked, and their,
their heart gets restored.
So that's sort of one mechanism
why the, why the heart starts or
h- why the heart stops And that's
thought to be favorable because you
basically you're like a normal person.
You don't know you have
any medical problems.
And all of a sudden, a clot
flicks off to a big vessel in
your coronary and bam, you drop.
So your body isn't very sick from
other diseases, maybe a little bit
of hypertension here and there.
But, you know, getting that person,
you know, back and going is one thing.
The other type of cardiac arrest
where somebody's heart stops is
what we call non-shockable arrest.
And that comes from else
going on with the patient.
Like they have a bad respiratory problem
and they get less and less oxygen.
And all of a sudden, there's not
enough oxygen in their bloodstream
to keep the heart moving or they
have a blood clot heaven forbid.
And this is something that we will find
came up more commonly than we thought.
They take an overdose
and they stop breathing.
And when they stop breathing, the body
is more slowly being deprived of oxygen.
Whereas in that initial case of the sudden
drop, you have a normal oxygen level.
So even though your heart has stopped,
there was a fair bit of oxygen and
stuff that was pushed up to the brain.
Whereas if you're not breathing for
a while and a while and a while,
there's even more sort of pre-injury.
So when I said we don't really
know much about severity,
we did know that one thing.
And the one limitation of the
trials was those 2002 trials, they
didn't think those people with the
latter type, the non-shockable,
had a lot of prospect for recovery.
So they really only studied
those who had shockable rhythms.
So another thing that another big
question we had is the guidelines
were vague the non-shockable people.
They said, please cool the people who had
shockable rhythms, but the people with
non-shockable rhythms, we're not sure.
There's no reason to think that
the brain wouldn't potentially be
protected from hypothermia based
on these two different mechanisms.
But there was a worry that maybe these
people had such a bad prognosis that if
you include too many people like that
in your trials, you're not going to have
enough good outcomes to see a difference.
Scott: Okay, so the, the, the reason
for the cardiac arrest, shock, um,
shockable and non-shockable is a question.
Um, and then largely hypothermia.
This trial-- By the way, I, I,
I'm not sure we, we laid this out.
When you and I are speaking right
now, this trial is over, um,
that we, we know the results.
We are not gonna talk about the results.
The results are coming out.
Uh, believe they've been accepted
into JAMA, but, um, that we're
going to have another podcast of
the results, and we're gonna talk
about the results and what they mean.
We wanna talk about the design, so we're
setting up the design part of this.
So what is it y- So you, you are going
to look at two different kinds of
patients, shockable and non-shockable.
What is it-- What are the interventions
you're assigning to a patient?
Will Meurer: I think one of the
things is, we arrived at this after
having these meetings as part of
this interesting and, and really cool
regulatory science initiative that
was sponsored by the US Food and Drug
Administration and the, the US, uh, and
the Office of the Director at the NIH.
Um, we did this project called
ADAPTT, which is, which is where we
s- we, we started to work together.
And, um, in the, the process, the,
the sort of straw man design for
something like ICECAP was to look
at a, a, a few different cooling
durations like twelve, twenty-four, and
forty-eight in shockable rhythms only.
we had more feedback from a variety
of stakeholders who were at these
meetings, it became clear that,
that that was probably too narrow.
There was so much uncertainty in
the non-shockable people, but a,
a great unmet need that we should
work to try to study both of those
populations at the same time.
But as we emerged in terms of, you
know, how we were planning the trial
and what decisions we're making, we, um,
to make sure that if there was a signal
the shockable rhythms, that we could
sort of find that even if things were
different in the non-shockable rhythms.
Or similarly, if because the
non-shockables typically have
more brain damage, maybe they
need to be cooled longer.
So th-those were, those were some of
the things that, that came through.
We really wanted-- We didn't wanna
study how cold you made them.
We, we focused on thirty-three degrees
or roughly ninety Fahrenheit from the
animal models, that was, that was sort of
the best-- that was a sort of sweet spot.
Some people have looked at other, you
know, like thirty-four or thirty-six.
And we, we, we based on our read of
the, the human and animal literature,
looking at depth wasn't so interesting.
duration we felt was most
interesting like I said, in the
neonates, it was about 72 hours.
There was a very encouraging study in
pediatrics, which didn't quite achieve
statistical significance called BABKA,
but 48 hours of cooling in, you know,
kids who are 18 and under, most of whom
were actually two and under, 48 hours
of cooling was, was quite promising
in terms of improving survival.
So there were those sorts of numbers.
But then in that one of those
first Australian studies, they got
people cooled really quick, and
they only cooled them for 12 hours,
and they also saw great results.
So we had this sort of
range of uncertainty.
So duration was important.
But another thing that was
crucially important to us
was the therapeutic window.
we really felt that based on the
preclinical data, that it was
important to get people cooled quickly.
Some people come in cold, you know,
obviously, you've spent a lot of
your life up in Minnesota, easier
to be-- easier for a patient to get
cold up in Minnesota in the winter
than it is in Austin in the summer.
So there is some of that, but a
lot of it is really, if the body's
brain is injured, it, turns off your
temperature center to some degree.
So your body is sort of even in
a relatively warm environment,
body is going to get cold.
So people cold early something we
looked at as very important to reduce
variability in the trial, right?
Because if we're like, we're
testing these different durations,
but some people are getting cooled
starting 12 hours after injury.
Some people it's two hours after injury,
some people it's six hours after injury.
We felt that that could--
that heterogeneity could,
could really sink our trial.
So we did have a focus on with the
hopes that, that sites would get
really engaged and get devices on
people quickly to have, um, cooled
quickly as a condition of the trial.
Because of that, and because the
guidelines at the time it were-- was
designed were cool everyone, we couldn't
have what one would consider a classic
control group, um, in terms of let's
just turn-- let's just put a device
on and set it to 37 degrees, you know,
ninety-eight point six, and just make
sure they don't get fever, you know.
So that-- we couldn't do that because
we requi-- we wanted people to be
cooled before they were even to be
eligible for the trial so that, that
sites-- since sites were cooling
everybody as a condition of being in
the study, at least anybody with a
coma or that had to be their usual
practice to get into the study, that
meant we could require patients to be
cooled to a certain amount before they
could be randomized to a duration.
Scott: Okay.
So we are in the ICECAP trial, we're
gonna randomize people to a duration,
and a critical part of all of this is
that we can't do no duration, no cooling.
Uh, you would slow down people that are
gonna be cooled by trying to get consent
or a waiver of consent or all of that.
Standard of care is to cool them.
So everybody is cooled in the
trial, but you're going to vary
the length of time they're cooled
in the trial, um, uh, within this.
And initially, uh, a- and, and going
back to this-- By the way, the ADAPT-IT
project that you mentioned was, was a
fabulous project, and I think we had six
different trials, and the goal of that
project was to explore innovative designs
being brought to clinical trials in
the emergency room setting and to study
barriers to bringing in innovative design.
And, uh, interestingly, I think the number
one barrier was statisticians, but that,
that's a, a topic for a different day.
Um, uh, and this-- there, there was,
there was research, and we, we had,
uh, IRB approval to study the people
building the designs as to, to what
were barriers and what were going
on So I remember the initial design
that I simulated for this trial was
randomizing patients to twelve hours,
twenty-four hours, and forty-eight hours.
And initially, it wasn't the
two populations, so we'll
kind of ignore that part.
And, uh, we had a dichotomous outcome
of their ninety-day MRS status, and we
simulated those trials and showed a number
of those results, and a lot of times
it didn't really answer the questions.
There was a great deal of, "Wow,
boy, I wish we had more on the
smaller ones in this case."
Is even somewhat of an optimal curve
that it kind of goes up at twenty-four
and kind of comes down, left huge
uncertainty whether this was flat,
whether what the shape of this curve was.
So there was a desire to have more
durations, focusing patients where the
best benefit was being, being seen.
And so as you described, the next
generation of this trial was, first
of all, changing the endpoint to
analyzing the whole scale, and
we'll get to how we did that.
But we opened up six hours, twelve
hours, eighteen hours, twenty-four hours,
all the way up, I think, originally
to eighty-four and ninety-six hours.
There were twelve different durations
that a patient could be randomized
in version number two of this
trial that we started to simulate.
Will Meurer: And yeah, and then I think
this was one of the really great things
about the Adaptive project was that by
sort of illustrating design first, you
could show people like Oh, I don't know.
At the end of that trial, I'm gonna
be sad or have this anticipated
regret as we, we talk about it.
Or, or doing the pre-mortem, right?
If we do that twelve, twenty-four,
forty-eight trial and it's, it's, it's
sort of like, "Oh, we should have gone
out to seventy-two," or, "Oh, maybe,
maybe six would've been better."
Or so, so, so that-- You know, I
think that's an important concept.
I think the thing that also is, you know,
I think really fascinating is that, you
know, by seeing the way the trial could
come out, then you get people together
and they're like, "Gosh, we need to, we
need to do something a little different."
And the other thing, though, is
you do have to put this-- And, and
sometimes I think, again, when you
say like statisticians are the, are
the barrier, you, you're also working
with, hey, the guidelines say cool
everybody for twenty-four hours.
So to some degree you could say, yeah,
let's look from ninety-six to zero.
from a statistical perspective, you know,
the most efficient way to, to do that
is you're gonna start putting people
on both ends to sort of see like, oh,
is this, i-is this possibly flat or, or
maybe there's gonna be an upslope in the
middle, but, but, but, you know, kind
of go into the ends and, and fill in.
But since those are far from, you
know, both those are far from what was
guideline recommended at the time, the
human constraint on the model was like,
we do have to do it this way because
it, it's not ethical to all of a sudden
cool people for ninety-six hours.
Um, you know, people who are in
ICUs, you're gonna-- You know, if
they start to wake up, you're gonna
want them to let, let them wake up.
So, so the, the...
You know, if it's only six hours of
cooling, the guidelines say twenty-four.
Patients are gonna be like,
"Why are you putting my family
member on such a small duration?"
So, so we did have this constraint
because there was, you know, real
recommendations for practice that we...
You know, I think in retrospect
sometimes people say, "Well,
like, why didn't you just evenly
distribute them across ten arms?"
And it's like, wasn't...
You know, based on what we knew about the
science at the time we were coming up with
the design and starting the trial, you
know, that wouldn't have been ethical.
So that, so that's I think, you
know, important, and I think
that was one of the things.
You know, you would show us a d-
Scott: Yep.
Will Meurer: a little too,
too long on this neutral case.
Like, why is it, why is it
putting people over there?"
And then you'd say, because it
wants to know, and that's how it's
gonna find that slope of that line."
And say, "Well, like, let's, let's,
let's, let's see if we can, you know,
nudge that backwards so that it's...
It needs to have stronger evidence that
there's an upslope, you know, between, you
know, durations that we've used before,
twelve to forty-eight, before it starts
looking at those longer durations."
So I think that was so
crucially important.
It took a lot of, you know,
time and energy to go back
and, and think of those things.
But, you know, it's one of those
things, though, when people look
back at the design, sometimes they're
like, "Hey, why did they do that?"
But once you know the results- Like,
the reason we were doing the trial
was we didn't know the results.
We didn't know what this
was gonna look like.
Um, so we wanted it to be flexible to a
variety of different truths and be able
to get us, uh, an answer that could be
very useful to us under a variety of these
scenarios, whether it was like up sloping
late, up sloping early, flat, going down.
And I think that was something that,
that when people say, "Well, what--
why'd you do all this complicated stuff?"
You know, this is something that's
come up in adaptive designs.
I think like the ASTON trial of the
neutrophil inhibitor for stroke,
I think is, is sometimes cited.
It was a really innovative design.
They, they went through so many different
dose tiers in their design, and I
thought it was like an amazing trial.
But then critics later are like, they
could have learned that with three doses
and, you know, a third of the patients."
But didn't know that in advance, right?
Like, if they had done three
doses, maybe they would have tested
five-- you know, three different
ones after that initial trial.
As opposed to they got to a very
convincing answer that in, in
the population they were testing,
their agent, you know, didn't,
didn't help those stroke patients.
So, so I think sometimes when we
have adaptive designs that, that
kind of go more into that, neutral
space, people are, people are like...
Or even adaptive designs that show
something positive, people are like, "Oh,
well, it would've been much easier to just
assign them to those two groups," right?
But it's like,
Scott: Right, right,
Will Meurer: that when we started.
Scott: right.
Right.
Right, right.
Okay.
So i- through the simulations, a- as
you describe this, quickly, by the
way, 84 and 96 hours dropped out,
largely for operational things, that
people are in the, for five days in
the, the, the emergency room, and
it would've been awkward to do the
trial where they go to the ICU as...
So we dropped those off, and we-- I,
I remember showing a simulation where
initially we simulate to those same
three arms, 12, 24, and 48, and then
we do response adaptive randomization.
We allocate patients where they're most
likely to be benefit across that curve.
If the low doses are doing the best, we
put more there, refining that later ones.
And I showed an example, a single
trial where the curve went up, it
was flat, it went down a little bit,
but then it started to go back up.
I was using a smoothing spline
across the, the 12 durations.
And it started to assign
patients at 72 hours.
And the clinicians all said, "Nope, we
would never do that," that that, that
dose response curve is not believable.
There's the-- It's, it's not conceivable
that the curve across durations would
start to decline, that patients do
worse as you cool them more, and all
of a sudden it starts to go up again.
And that was a, that was a,
a, a really important insight.
Also, uh, on the, on the other side,
largely it was that this curve is gonna be
an inverted U, that if this is beneficial,
or if it's not beneficial, it fits on
this curve, that there's gonna be a
point where cooling more is beneficial.
It may then be flat for some length
of time, and then it's gonna go down.
And we know at the extremes
those would be the cases.
If you cool somebody for 30 days,
they're not going to do well, and,
and we know that it's gonna get worse.
Now, so that became the analysis
model of this trial, that it's
an inverted U dose response.
It has a-- it can have a flat area.
The important part is all of the, the,
the 10 doses could be on the upslope.
So we modeled it so the doses
are somewhere on that inverted U.
They could all be on the upslope,
and 72 hours is the best.
They could all be on the
downslope, and cooling is bad.
The more you cool them,
the worse they get.
They could all be flat, that from six
to 72, it doesn't matter, it's flat.
Or it could go up and be flat,
it could be flat and go down.
That's the model that drives the trial
And what's really nice about that,
so there's a Bayesian model of this
inverted U th- that allows the doses
to be ordered somewhere on that scale.
And what we're looking for is the smallest
dose that achieves the highest level,
and that's defined by this, the, as this
optimal dose that's defined by the curve.
Because we know if it's getting flat,
there's-- you don't wanna go bigger.
There's no reason to go bigger.
It's not gonna go flat and then up.
It can't in this shape.
So we use that in response-adaptive
randomization to allocate patients,
and that was the next version of this.
And we added in, you, you addressed
these gu- guardrails that we didn't
wanna start with people on six
hours, and we didn't wanna start
with people on 60 hours or 72 hours.
If the evidence was building that the
dose response was continually increasing
through up to 60, and the algorithm
thought it was likely that was the best,
we would open that arm with guardrails,
and we would open six with guardrails if
it appeared that six might be the most
effective dose or this optimal dose.
Like, if it looks really flat, we
could open six, only with adaptive
conditions in the trial, is how
we started to migrate the design.
Then...
Yeah, yeah, go ahead.
Will Meurer: no.
No, no, you can, you keep
Scott: I was gonna say, then it became
we want to investigate both non-shockable
and shockable, and the concern is
that this optimal place is different
for shockable and non-shockable.
As you set up, it might be, depending
on the etiology for why the person
went down, you might wanna cool
longer or shorter within that.
So we ended up moving the design to
having the same duration response
model modeled separately in the two
populations and having different
response adaptive randomization
probabilities by rhythm type.
Will Meurer: Yeah, the thing
I was gonna say is, you know,
that's another thing, right?
People, you know, would say,
"We don't think this duration
response can go up and down."
But somebody later might, like, do
some animal research and say, "Oh,
gosh, biologically, you need to cool
during this reperfusion period, and
then maybe you need to cool later
during this, you know, period where
the brain is getting swollen."
But again, that's not one of the arms
we had, where it's like cool from
six to twelve, pause, and then if you
haven't woken up, cool again from,
you know, seventy-two to ninety-six.
One of the things that, that I sort of
brought up, I think, back when we were
having these planning meetings is, gosh,
wouldn't it be nice if there was some
way that we could tailor the treatment?
You know, this is, this sort of
Goldilocks approach to cooling.
Like, if we had some way to measure
the brain and, and, oh gosh, it's, it's
crying out to us as we rewarm somebody.
Let's, oh, oh, let's, let's
like, then make them cool longer.
And again, that was a lovely idea, but
we as a scientific community at the time
said, "Hey, Will, that's cute, but we
don't know enough about how this works
to do something that complicated that,
that might take a billion patients."
Um, but we have earnest uncertainty in
these, you know, in these durations that
we're testing and whether there's, know,
I think, you know, as, as we'll get
to next, while we don't have a control
group, if everything's on the upslope,
and it's a pretty good upslope, that's
a, it's a pretty good way to connect the
dots with a, with a no cooling group.
Um, so, so I think have to make...
You know, we did th-think that there
could be individual designs that, that
maybe could be better for patients,
but we didn't have any of them off the
shelf, and didn't have a good way to,
to sort of think about how we could
implement something like that, that
sort of let me cool them until their
brain tells me It's ready to wake
Scott: We weren't ready.
Yeah
Will Meurer: we, you know, one of the
other constraints that we all have in,
in this space that is also something
that's sort of really interesting is
there is a lot of pressure in, in the US
healthcare system in particular kind of
get bored with providing critical care
to somebody who might not wake up say,
"Yeah, you know, Uncle Joe's lived a good
life, but sure if he's gonna wake up."
And then having people wanna do
care limitations and remove people
from technologic support which,
which, which commonly happens.
But in a trial like ICECAP,
that's a huge problem, right?
Because these people aren't being
given an opportunity to wake up.
of the potential mechanisms...
You know, I talk about
biological mechanisms of cooling.
One of the potential mechanisms of
cooling, at least in the clinical
trial s-sense of potentially improving
outcomes, is the longer you cool--
You can't, you can't pull the plug
on somebody while they're cold.
You have to, you know...
So, so if you have a longer time
period, you're basically, you know,
sort of creating this sort of reverse
shot clock of, you gotta wait at
least this long before you, you take
a shot at letting this person wake up.
And in some of the trials that
have been done, have been waking
up, you know, twenty-two to
thirty days after the injury.
Now, that's not to say that every single
patient who has a cardiac arrest should
be given twenty-two to thirty days to
wake up, but, but many of them aren't,
aren't giving, being given more than two.
And that's, again, that, that, you
know, certain European countries,
people are much more willing to
give it a, give it a week or two.
Um, again, not trying to make this a,
a sort of value judgment on different
societies, but it's an observation
that, that people do handle these
things a little differently in places.
Um, and, and there's also a
lot of clinicians who are like,
"I'm sure they won't wake up."
But if you tell that to everybody at day
two, and sometimes it takes people five
to twenty days to wake up, you're gonna
be right, none of your patients wake up.
But some of them would have
if you had given them longer.
So, so we also have
Scott: And
Will Meurer: of, you know, social
element of the trial and, and working
to do, you know, another intervention.
You say, "Well, why did you,
you, you pick duration?"
'Cause that was most interesting.
We tried to limit variability
on this care limitation by
having clinical standardization.
Say, you know, "Please talk
to us if you're doing this."
You have to have a defined protocol to
prognosticate whether you're doing an EEG,
you're doing brain imaging, you're doing
blood tests to make sure that if you, if
you wanna tell a family that this person
isn't gonna wake up, you do have some
additional evidence that sort of backs
that up, that it's, it's, A, it's pretty
unlikely that they're gonna wake up.
So that was another aspect of the trial.
It was a feature of some concurrent
European trials that, that went
on, that they did, they did
that in a standardized way.
So that's another thing, but from a study
design perspective people don't complete
their duration of cooling because of
early care limitations, it introduces,
um, you know, noise into our trial.
You know, our trial has less ability to
pick up a treatment effect when those
observations aren't contributing really
to the, the treatment effect estimate.
Scott: we, we, for, for
especially the statisticians out
there, we do intend to treat.
So if a patient is, is g- is assigned
forty-eight hours and dies at twelve
hours, they're in the forty-eight-hour
arm, and their outcome is that.
If they somehow come off
of it at twenty-four hours,
they're forty-eight hours.
So it's intent to treat,
uh, from the arms of that.
Um, I, I, yeah, I think it's an or-- in a
really important part in the interactions
of this, I remember we, we went to
the NIH presenting to get this funded.
I think there was discussion about...
Sorry, at that time, we went
with members of the FDA.
The FDA really wanted this trial to run.
Uh, Bram Zuckerman, in particular,
really wanted this trial to run.
Now, medical devices out there have
been approved that they cool a patient.
That's not tied to clinical benefit.
It doesn't say it's good to cool people,
but you've demonstrated you can cool
somebody to thirty-three degrees And
these are being used, and so there's
not a great incentive for devices to
run this trial, to fund this trial,
so they're not gonna fund this trial.
There might even be a disincentive
to, to, to fund this trial.
Um, the FDA wants this trial to run,
and so we went to the NIH, and the
FDA went, uh, on behalf of the design
that we'd really like to see this
trial run, which was sort of a unique
experience, uh, uh, in, in the whole
interactions of getting this trial run.
And eventually this trial
was, uh, funded by the NIH.
Will Meurer: Yeah.
I, I think the FDA even a little
more excited so to, so to speak
about it, because really the
Scott: Yep.
Will Meurer: So the use of them to--
You, you can set a different target,
but really the idea was that they were
to keep people at normal temperatures
or pay-- or potentially warm people
up who have environmental hypothermia.
So to some degree-- Well, and, you know,
every time I've given a presentation
about ICECAP, I, I put in that
disclosure, "I will discuss off-label
uses of these devices," because, you
know, the, the way that their label was.
So the FDA felt that it was, uh,
important from a regulatory science
perspective to learn more about
the safety of these devices when
being used outside of their label.
Um, so, so that was-- That, that gave
them additional enthusiasm for our design.
And you're right, it was really
unique, I think, to have FDA I think
FDA does a lot of great things.
I think they can be great scientific
partners, and I think in this case,
this is a great example of that.
Like, in that they were interested in
doing what's right for patients, but,
but learning in an unbiased way through
a clinical trial where all the adverse
events are being adjudicated, people
are-- the outcome assessors are blinded
to the duration of this cooling device
that they were subjected to and so forth.
So, so a lot of real positives.
I mean, it was, it was, it, a
delight working with them on it.
And, you know, I, I was the
sponsor of the IDE, so I've had,
had great interactions with them.
They're really, really
thoughtful scientists there.
Scott: Okay, so let's, let's summarize
the design because, uh, uh, uh, our
listeners are gonna tune into this, and
then they're gonna tune into the results,
and we'll do a podcast of the results.
So this is a design of eighteen
hundred patients combined across
rhythm types, where initially the
allocation is twelve, twenty-four, and
forty-eight hours, equal randomization.
After two hundred patients,
an interim analysis is done.
The Bayesian model is fit
for both ris-- rhythm types.
We do multiple imputation of patients
with thirty days MRS status projecting
to their ninety-day status We have 90-day
MRS as the primary endpoint, and we use
a weighted scale of the MRS for, for
analyzing that, not a dichotomization
of that, so a weighted scale.
At the interim analysis,
updated randomizations are made
to the different durations.
That is done every fifty patients in the
trial that these-- the RAR is updated
throughout the course of the trial.
There's a possibility of opening up,
uh, them on the right or the left.
So the, the ones at the beginning,
twelve to forty-eight are open and
can be, and it can be allocated
to right away, and anything in
between that by every six hours.
And certain conditions
could open up six hours.
We, of course, we talked about opening
up zero, but that was not a possibility.
So it could open up six.
We talked about that in the design stage.
And then it could open up sixty
or seventy-two on the other end
if this appears to be a continual
increasing, um, uh, duration response.
There is no early success.
There's no stopping because we've
determined the right answer.
There is futility if there's, uh,
comes to be at least a fifty percent
probability that six hours is the best
thing to do, we'll stop a rhythm type.
And we, we call it futility.
It's largely that there's reasonable
evidence that there isn't an
increasing duration response curve.
The two primary analyses are this, are
what is the optimal duration to cool that
comes from the Bayesian model, and is
there an increasing duration response?
If there's an increasing duration
response, if twelve is better than
six and eighteen is better than
twelve, it gives evidence without a
zero arm that cooling is beneficial.
And the FDA understood the non-non-zero
arm, and they understood this as a,
as a signal that cooling is beneficial
to patients if there's an increasing
relationship with length of cooling.
Within that is part of the
primary analysis, uh, in the
trial and the two rhythm types.
Have, have I, have I summarized
the design appropriately?
Will Meurer: Yeah, no, but we, we
maybe take a brief little tangent
into the, uh, weighted mRS.
And I think, you know, others, you
know, other, others have probably
seen this in, in the stroke world.
Scott: Yep.
Will Meurer: for stroke, uh, we developed
this weighting scale for, for ICECAP,
um, a little earlier than the strokes--
the stroke one, m- m-- you know, emerged.
In the stroke one, they, they talked to
clinicians and, and patients and families
to develop a series of we-- a series
of weights for the, the mRS states.
And, um, in, in, in pedia or in
adult ICECAP, we, we set these sort
of based on, on cl-- on what the
clinicians thought with the goal of
accomplishing two things Can we
find people who more or less wake
up, you know, their mRS is, is, is,
you know, sort of better than four?
And amongst those who wake up, can
we distinguish amongst the people
who have really excellent outcomes?
I think when we commonly dichotomize,
it's like you get full credit at a 0 full
credit at a 1 full credit at a 2 and all
of a sudden at a 3 you get no credit.
It's the same as being dead.
And that, in this disease, we
felt was, was the possibility of
showing important improvements.
Because someone with an mRS of three,
if they had a really severe stroke
or severe stroke, had a ve- really
severe cardiac arrest, at n- at, at
three months, that might be a very good
outcome, and they actually may have
a, a potential for future trajectory.
And we're doing a study called
POST-ICECAP um, that we designed
to actually live after ICECAP.
initially, it was following patients
after their three-month follow-up, but
now it's following people at three,
six, nine, and twelve months to look at
that trajectory of recovery over time.
And, and there are certainly patients
for whom an mRS of three is, is, is
a r- is a really, is a win in terms
of their severity of, of, of injury.
So, so we did design this weighted scale.
It's very similar to the utility-weighted
scale that is used, that was used in
DAWN and DEFUSE and other stroke trials.
Um, it's subtly different, but it,
it, it's-- the, the concept is the
same, even though we arrived at it
for a scientific reason as opposed
to inducing it from sort of Clinician
and patient's, uh, preferences.
Scott: just for disclosure of what
it is, the MRS of zero, which is
the best outcome, you get 10 points.
A one, you get nine points.
A two, you get eight points.
A three, you get six points.
So the two to three difference
was considered greater.
And then below three for four, five,
and six, you get zero points, was,
was, is the, the, the scale used, the
weighting of that used in the trial.
Okay, and, and we should, uh,
highlight that MUSC and Sharon
Yates did, uh, ran this trial.
Uh, she's the-- they're the data
coordinating center, did a tremendous
job o-on this, um, in the trial.
So a complex mult-- many interim analyses
done during the trial, and they did a
fantastic job, um, uh, running the trial.
But we can't...
Will Meurer: thing, like sometimes
people are like, "Oh gosh, turning
around an interim analysis to change your
Scott: And...
Will Meurer: a data freeze."
And, and you know, they were-- they, they
would, they would turn it around in a day
Scott: Yeah.
Will Meurer: you know, a
Scott: Yep.
Will Meurer: really.
And it was-- They, they know it
was coming, but, you know, it's
Scott: Yep.
Will Meurer: protocol choice we made
to say we are gonna do this every
Scott: Yep.
Will Meurer: Um, know, we did, we did
also design a pediatric trial that has
many features that are similar to adult
ICECAP, but some that are different.
Um, and one of them is that because
there was more equipoise in that
trial, it does have a no cooling group.
It has a zero-hour duration.
Goes all the way out to ninety-six
'cause of that data from
neonates where seventy-two to
ninety-six worked well in them.
But one thing is we did say that
the interim analysis would be
every, approximately every ten
weeks to just make it a little bit
more predictable, um, and nice.
But, but, but Dr.
Yates and the team at MUSC,
were phenomenal and, um, we were
always keeping our eye on that.
And but, but yeah, but it would, it
would change, you know, as an emergency
physician and somebody who works in
the ICU, and I take stroke call, I'm
like, "Oh, I gotta do something in
the middle of the night on a Tuesday."
I'm used to that.
Generally, and again, this is not a slight
on academic biostatisticians, that's
not typically part of their workflow.
Um, but, but they, they, they
got everything going so well, and
it was, it was just a delight.
Scott: Yeah.
Yeah.
Yep.
Yep.
Okay.
We, we need to stop now because
we'd start telling results.
So look for ICE CAP results, uh,
coming when we do another interim
analysis, which we love to do here.
So thank you, Will, and I look
to, to see you very shortly.
Uh, and thanks everybody, and,
uh, thanks for joining us.
Until next time, we are
here in the interim.