Subscribe
Share
Share
Embed
In this episode of “In the Interim…,” Dr. Scott Berry talks with Dr. Roger Lewis, Dr. Anna McGlothlin, and Dr. Nick Berry — all co-authors on the CHIPS trial results recently published in JAMA (Spinella et al., August 2026) — about the design, implementation, and results. The conversation covers why room-temperature platelets, which can be stored only five to seven days, leave rural hospitals, low-volume centers, and military and disaster settings without a reliable supply, and how a Bayesian adaptive design was used to find the maximum safe cold-storage duration rather than testing a single fixed duration. Roger, Anna, and Nick walk through the monotonic dose-response model that governed escalation, an unplanned mid-trial complication when the FDA independently authorized fourteen-day cold storage, and how the Data Safety Monitoring Board reviewed results within days of each interim. The trial ultimately demonstrated non-inferiority of cold-stored platelets out to twenty-one days with a Bayesian probability greater than 99.9%, offering a path to expanding platelet access in settings where it was previously very challenging.
Key Highlights
Key Highlights
- Cold-stored platelets tested as a potential answer to platelet shortages in rural, low-volume, and military/disaster settings.
- Bayesian adaptive design used to find the maximum safe cold-storage duration, not just test a single fixed duration.
- A monotonic dose-response model constrained escalation to a pre-specified, safety-first ladder across four interim analyses.
- Mid-trial complication: an independent FDA decision allowing 14-day cold storage, absorbed into the design without unblinding.
- Non-inferiority demonstrated out to 21 days of cold storage, with a Bayesian probability greater than 99.9%.
- Interim data turned around by the unblinded implementation team in five business days, versus the six weeks often assumed for adaptive trials.
- Implications for platelet access in disaster, military, and low-volume hospital settings, and for how shelf-life-dependent products are tested going forward.
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.
your host, Scott Berry.
And on today's episode, we are
going to dive into the creation, the
design, the implementation, and the
results of a Bayesian adaptive trial.
Really a pretty cool trial.
Uh, the, the name is cool.
Uh, this is the CHIPS trial, uh,
Chilled Platelet Study, and we'll
talk more about what that means.
The results of this trial were
recently published August 17th 2026.
The results showed up online
in JAMA, Spinella et al.
today I have three guests,
all co-authors on this paper.
my three guests, uh, I'll start with Dr.
Roger Lewis.
Uh, Roger is a senior medical scientist
at Berry Consultants and a professor of
emergency medicine at the David Geffen
School of Medicine at UCLA, and he's an
elected member of the National Academy
of Medicine and a fellow of the ASA.
Pretty cool combination.
Uh, Dr.
Anna McLaughlin is a senior statistical
scientist at Berry Consultants, and
she's a director, and she specializes
in implementation of adaptive trials,
complex trials, platform trials, and
that was her role on the CHIPS trial,
and we'll get to talk to her about
implementing this adaptive design.
And Dr.
Nick Berry is a senior statistical
scientist at Berry Consultants, and
Nick, uh, focuses on our Phac software,
software design, uh, the software team.
Uh, he was involved in the design
of this trial, and that goes all
the way back to, I think, 2019.
And by the way, he picked up for Liz
Krejci, who was at Berry Consultants back
then, and she started some of this work.
So welcome, everybody, to In The Interim
Nick Berry: Thanks
Scott: So let's set up.
We, we-- This is the CHIPS trial,
Roger, uh, Chilled Platelets Study.
Uh, us what the trial is, what's the
clinical context, and what are the
inter- inter- interventions in the trial?
Roger Lewis: So the CHIPS trial
studied, um, platelet transfusion.
So platelets are part of the
blood that help in clotting.
They're commonly administered when
patients, um, are bleeding, need, um,
blood products, uh, often during surgery.
And one of the challenges with platelets
is that traditionally they're obtained
from people and, and, uh, purified, and
then they're stored at room temperature.
And because of the storage at room
temperature, they go bad very quickly.
So maximum storage duration is
usually, uh, five days, sometimes
up to seven days if you do
repeated testing for contamination.
But after that period of time,
they can no longer be used.
So because of that, platelets
are really only available at
hospitals and blood banks that
have a lot of turnover in products.
And so there's not a supply of platelets
available to patients that are treated at
low-volume hospitals or in rural settings.
And one of the ways to try to fix this
problem is to figure out a way we can
store platelets for a longer period of
time so that they can be, um, kept in
inventory at places, um, practically
that now can't currently, um, have them.
So the CHIPS trial compared the
traditional room temperature platelets
to platelets that were stored in
the cold, essentially refrigerator
temperature, and potentially
stored for a longer period of time.
the clinical model was, um, both children
and adults, including very small children,
undergoing complex cardiac surgery because
that's a clinical setting in which we
expect there to be substantial bleeding
and platelets are routinely administered
Scott: Okay, so I, as, as somebody who
doesn't understand the sci- science
of platelets, if, if you were to store
them for too long, what, what bad things
happen if this is a bad thing to do?
And is it different than warm
platelets go bad to some extent?
If I were to store platelets for a
year cold and then use them and they
were not functional, what would happen?
Roger Lewis: So I think there's a
couple of qu- different questions there.
So the most complication of storing
platelets too long is that they're
con- contaminated by bacteria,
a very small number, and over a
period of time, they multiply.
And so when you transfuse those
platelets that are contaminated with
bacteria, the patient develops an
infection, clinical sepsis, and that's
a very bad thing when it happens.
And it's just very difficult
to collect a blood product, you
know, you pierce the skin, there's
bacteria on the skin, without ever
getting any bacteria in at all.
So just like, you know, milk or any
other product that goes bad, can-- even
if there's a teeny bit of contamination,
you can store it for a while safely.
But if you allow the bacteria
too much time to, to multiply,
you can get into problems.
The other issue with storing platelets
too long is they lose what we call their
hemostatic efficacy, which is their
ability to help the blood clot in those
settings in which clotting is important.
Like for example, if a surgeon is
cutting you or you have a wound
Scott: Okay, and so, so a- and hence
that's sort of the, the, this trial
is about how long we can cool them.
So the, the two interventions are
room temperature stored platelets,
and that's up to the five to
seven days, up to seven days.
And the other interventions are cold
stored platelets, where the time you
store them may vary throughout the trial.
And so what is the goal of this trial?
Roger Lewis: Well, the goal is
to prevent excessive bleeding.
So as the, um, the patient undergoes
their operation and in the immediate
postoperative period, up to twenty-four
hours, um, after the surgery, the
patient is at risk for bleeding.
And bleeding has a lot of
important clinical implications.
You know, not only might you need blood
transfusions, um, if you're bleeding
excessively, but if their bleeding
cannot be controlled, sometimes patients
require reoperation, meaning actually
going back in, trying to find out where
the bleeding is coming from and, and
physically or surgically dealing with it.
So, um, the, the period of most
interest is in the first twenty-four
hours after surgery, and these patients
commonly have chest tubes in place,
which are tubes that drain any excess
bleeding from around the lungs.
And it is a very common model to test,
FDA, um, regulated products that are
intended to support clotting or hemostasis
to use the complex cardiac surgery model
and to look at things like the amount
of bleeding or the amount of chest
tube output, which is basically the
blood coming out of, out of the chest.
One of the features of this trial was
that it used a composite bleeding score,
is a five-level ordinal scale that
had been developed for these kinds of
applications, with one being the least
bleeding, five being substantial bleeding.
So it used a sort of holistic and, and
patient-centered measure of the overall
bleeding with the goal of demonstrating
that cold-stored platelets stored for
some maximum duration was non-inferior
to room temperature platelets,
and therefore could be safely and
appropriately deployed in settings where
platelets are not currently available
Scott: Okay, so, um, the, the sort
of objective then is to understand
the longer time they're stored cold
platelets, are they still non-inferior
to room, uh, uh, temperature platelets?
we're gonna use this homeostatic efficacy
score for, for judging that, th- this,
this bleeding score throughout the trial.
Okay.
So Nick, tell me about then
the ap- adaptive design.
We're, we're not gonna start
by randomizing patients to 21
days of cold stored platelets.
How does the desi- how does
the adaptive design work?
Nick Berry: Yeah, the idea is that,
uh, room temperature platelets
can be stored for seven days.
Um, so at the onset of the trial,
when we start randomizing patients,
we're going to allow cold store
platelets to be stored up to seven
days, um, before they're administered.
And so the, the arms start, you know,
the first patient in is randomized
to either room temperature platelets
or cold stored platelets that are
stored for less than seven days
As the trial runs, hopefully, the
idea is that we'll demonstrate that
seven days of storage on the cold
stored arm is non-inferior to room
stor-- room temperature platelets,
Scott: platelets.
Nick Berry: able to say, "Okay, seven
seems safe versus results are as good, if
not better, than the room temperature arm.
Let's increase the maximum allowed storage
duration on the cold stored platelet arm."
And so we des-devised a scheme in
advance, a sort of ladder that you
can-- that the, the trial could
climb, um, throughout the interims,
Scott: interim,
Nick Berry: we scheduled at two
hundred, four hundred, six hundred,
and eight hundred subjects, uh,
randomized, um, patients randomized.
And at each of those interim analyses,
the cold stored platelet arm's maximum
allowable duration was allowed to increase
based on the results of the model, um,
based on the data collected at that point.
And so this, the, the, the original
idea was that we could construct
a model that would say, "Yeah,
here's where you should go.
This is the, the maximum allowable
duration that the model says is safe."
And it became clear as we put that
together that there was a balance in
being aggressive in increasing that
maximum duration and not putting
patients at undue risk, essentially.
So the, the way we, we structured
this was we had a model to model the
cold stored arm with the duration
response included in that estimation.
So, you know, when a, a
patient is randomized to the
cold stored arm, maybe their
Scott: get
Nick Berry: I'm gonna simplify this for
a second, maybe they get platelets stored
for twelve days on the cold stored arm.
Each patient has a hemostatic
efficacy score, one of those one to
fives Roger mentioned, along with
the duration, and you can imagine
a curve over the cold storage time
a-and, and we fit a model to that.
It ended up being what we'll
call a piecewise linear model.
Um, you know, it can wiggle
when it needs to, um, but it,
it liked to be a straight line.
So we tried to limit the, the
flexibility of the model while
still allowing a possible, um, curve
in it as, as durations increased.
Um, and we compared to basically just
a, a simple model on the warm arm.
So this is just a basic, uh, normal,
normal posterior where, um, the, the,
and then, you know, the cold arm versus
the warm arm, we'd look at where it
crossed to be more than one point worse
than the, the warm stored platelet arm.
Uh, there's a lot to mention here.
Yeah.
Scott: so, so that, that's a
Bayesian model, why we're calling
it Bayesian adaptive design.
But so you've got this Bayesian
model that's monotonic, that's an
important part of this, and you have
this piecewise exponential so that
the longer it's stored, the worse the
average bleeding score i- has to be
Nick Berry: Yeah, that's necessarily true.
This is a monotonic model, like you said.
So, uh, fifteen days stored cold
has to have a worse estimate
than seven days stored cold.
Um, it can get pretty flat, uh, which,
you know, is part of the goal was to
both let it curve but also be straight
and flat, uh, as the data recommended.
Scott: There are a
Nick Berry: Yep.
Scott: similarities to this to a
phase one oncology dose escalation
trial where we're starting slow,
we're escalating as long as it's safe.
We've got a Bayesian logistic regression
model that's allowing us to, to escalate.
You're doing this, the same thing here.
You spent a huge amount of time with
different models, their ability to
escalate, their ability to get the
right answer under a lot of different
scenarios, simulating cases where nine
days was the last non-inferior one.
Do we get a good answer?
What if, what if all
doses are non-inferior?
What if no doses are safe?
So you did a huge range of potential
truths to make sure the design didn't put
patients at risk, got the right answer,
uh, across the range of possible truths
Nick Berry: Yeah.
Yeah.
So you mentioned the
phase one dose escalation.
Uh, one of the, I guess the harder parts
that took some, some iteration and some
work to get right was like in a da-- uh,
phase one escalation study, how good you
are at estimating kind of at the upper
limit of your range really dictates how
good you are at driving that escalation.
And if you're estimating cold stored
platelet efficacy for durations
around five really well, but you're
not estimating up by twelve and
fourteen if that's near your max, you
don't know anything about the max.
It's really hard to push
the envelope and escalate.
And so
Scott: most
Nick Berry: both the model had to,
you know, drive that maximum allowable
duration to large values, and we
had to, to try to estimate that.
But we also had to-- Well, I say we, um,
the study team really had to devise a way
to actually collect patients that-- or
administer platelets near the max, and I
think that was a huge driving, uh, kind
of b- I'll say behind the scenes for us.
I'm sure it was, uh, in front of the
scenes for many people, but that was
a really, really important part of
the operating characteristics that
we saw when we were simulating the
study, is that the study worked really
well if we could collect patient or
administer platelets near the max.
It did not work very well if we
couldn't actually store platelets near
the, the maximum allowable duration.
And so, um, yeah, that was a,
a interesting characteristic of
the simulations in this case.
Scott: I think this is
unique, unique to this trial.
And I, um, just to make clear,
you can't say, "Give me, uh,
14 days cold platelet," Roger.
So when a patient's assigned to cold
platelets, you might have a range.
So this is a little bit different
than a traditional assignment trial
Roger Lewis: Yeah, that's,
that's exactly right.
And I think I wanna make a distinction
between the maximum allowable
duration, which is what was stepped
up in the ladder that, that Nick
mentioned, what patients receive.
patients receive-- it's as if you
were doing dose escalation, and as
you do dose escalation, you set the
maximum dose, but then someone randomly
decides to give any dose under that.
That's not as efficient an
experimental design as when you
can actually manipulate the dose.
And so the study team worked incredibly
hard behind the scenes to try to
administer platelets closest to the
maximum duration because that gave the
model the most information to learn from.
But I will say if that hadn't been
done successfully, the model would
have actually known that and known that
it didn't have very much information
near the upper end of the allowable
storage duration, and so it just
wouldn't have allowed escalation.
And I'm not sure if Nick mentioned
that the, the model only allowed
a certain amount of each time.
Even if the current estimate was
that cold stored platelets were after
twenty-one days of storage, if you were
at ten days of storage, it would only
allow you to escalate a little bit.
So it sort of carefully explored
upward in the maximum duration space.
And then as you mentioned, all
the work Nick did to look at
Scott: to look
Roger Lewis: accuracy of finding the
maximum storage duration at which
non-inferiority first failed, we
Scott: We tried very hard
to make sure that when we
Roger Lewis: we
Scott: didn't get the right answer,
we didn't get the right answer in a
Roger Lewis: the
Scott: conservative direction and
Roger Lewis: to
Scott: try to
Roger Lewis: put, put patients at risk
Scott: Okay, so you've
got an adaptive design.
Um, you, you have these escalation
rules, a governor on escalation,
a Bayesian model driving this.
That's the primary analysis at
the end is this Bayesian model.
Uh, you've gone to the FDA, so
the FDA's reviewed this design
Nick Berry: Yep, they did.
And can we-- we, we should probably,
as we talk about FDA, we should talk
about, uh, type one error and the
null case because what is the null in
this trial is an interesting question
and something we discussed with FDA.
Um, we simulated a lot of
different null scenarios, right?
Uh, and, and the primary analysis, as in
was it a type one error in this case was,
was seven days cold store duration deemed
non-inferior to warm platelets, um,
was the, I guess, the primary analysis.
And then after that, we would say, "Okay,
what's the maximum cold storage duration
that's non-inferior to warm stored
platelet, to room temperature platelet?"
Um, and so we, we simulated a
bunch of different reasonable nulls
Scott: Yeah, so
Nick Berry: in that case
Scott: a null.
Presumably a null is where cold
store platelets are inferior.
Nick Berry: They crossed to in-
inferiority at exactly seven days.
Scott: Yeah.
Nick Berry: yeah
Scott: Okay, so a- and so that,
that was a case particularly
s- worrisome I, I imagine.
But then at the same time, suppose
it crosses at 11 days, and up
to 11 days is non-inferior.
you looking at the rate at which you say
it's higher than 11 and making a mistake
on that and trying to control that?
Nick Berry: Yes.
Yeah, that's also part of this.
So I, I've never seen this before in an
adaptive design, but I think our power
was one in every single scenario we
reported in the, the protocol because,
you know, the, the, the scenarios we
simulated, the seven-day, uh, comparison
wasn't actually the interesting part.
It was what happened later.
And so we reported a lot of things
like were we within three days?
How often did we estimate something
bigger than the true crossing?
That rate was really low
Scott: Yeah.
Nick Berry: at the end of it.
Um, well below what you'd check for
like a type one error type of level.
Scott: Okay.
Nick Berry: Um, you know, one
percent, less than one percent,
uh, if I remember right.
How many patients were administered
platelets that were truly inferior,
things like that, where we were trying to
govern how aggressive we were with, with
administering platelets to, you know,
at high durations early in the study.
Scott: Yeah.
That, that's, uh, that whole
discussion is so fantastic in what,
what actually happened in the trial.
Um, the developments of the
trial, which we'll get to.
Okay, so you turn this over, and
now we need to implement this trial.
We need to do these multiple
interims every 200 patients.
Uh, we need to be running these
models, uh, doing escalation.
Anna, this gets turned over and
you're on the implementation type.
Take us a little bit through the poss- the
process of preparing to execute this trial
Anna McGlothlin: Sure.
So as with any adaptive trial, the
key to making it run smoothly is in
that preparation phase, like you said.
so we had a great team that we worked
with at the data coordinating center,
which was University of Utah here.
Um, so early on, we had discussions
with them about the data that we would
need in order to run the analysis,
kind of spelled out, here's these key
variables that we need to run the interim.
So, you know, we, we don't need the full
database, we just need a, a, a subset
o-of key variables, and those are the
ones that they focused on cleaning and,
and, um, getting put together for us.
when we hit that interim trigger
at, at 200 patients, 400 patients,
they were very efficient at being
able to hand that data over to us.
I think, if I remember correctly, it was
about one or two days from the time we
hit the trigger until the data arrived,
um, for us to start our analysis for that.
Um, so that was really amazing
that they could do that for us.
Um, and then on our side, you know,
ahead of time before we got to the first
interim, um, we spent time putting all
of our programs together, um, getting
our report shell in place, thinking about
how we were gonna present the data to
the Data and Safety Monitoring Board.
Nick put together some really cool plots,
um, to summarize, the data and how the
model was performing so that, uh, the
committee had the information they needed
in a really succinct way to be able
to monitor that as the trial went on.
you know, a couple things we were looking
at really carefully, as was mentioned,
this is a non-inferiority trial, um, so
we wanted to keep an eye on things like,
uh, are patients getting the treatment
they were actually randomized to?
Or if a patient was randomized to cold
platelets, are they actually getting
room temperature platelets or are
they getting some kind of a mixture?
Um, so that was something
we, we kept an eye on.
We wanted to make sure we summarized
that information for DSMB.
uh, and, and, you know, our
analysis was done very quickly.
So by the time we received the data until
we turned around a report to the DSMB
was about five business days, um, to do
that process and have it in their hands.
at the same time, whenever we completed
our analysis, we would prepare a memo to
go to an unblinded person that, a-as Roger
said, it was a big logistical challenge
to make sure that the, the storage,
um, the supply, um, of platelets to the
sites, um, really maximized that duration.
They had the right set
of platelets available.
So we would communicate that to someone
who was managing that logistical supply.
and that was all kind
of happening in parallel
Scott: Okay.
So, so the trial starts and Roger,
there's, uh, the interesting
discussions with the, the, the FDA.
There's an unplanned adaptation,
if you will, in the trial, in
the whole escalation, a bit
of a wrench into the trial.
What happened?
Roger Lewis: So the, the FDA was all of
us trying to address this issue of the
shortage of platelets, and they were
well aware that the, uh, limited storage
duration of form-- of room temperature
platelets was a problem in lots of
settings, um, disaster settings, military
settings, and just low volume hospitals.
And, um, FDA made a, a decision to
allow, in specific circumstances,
the administration of cold stored
platelets stored up to fourteen
days, um, uh, upon review.
My understanding, I'm certainly not
an expert in this regulatory area,
was that the FDA made the decision
that blood banks could apply to them
to get a variance to provide cold
stored platelets up to fourteen days.
So the trial was, um, following the
pre-specified adaptive design, then
suddenly, um, the FDA essentially said
up to fourteen was definitely okay.
Uh, this was, um, this external
information was certainly not informed
by any knowledge of what was going
on within the CHIPS trial, so it's
completely external information.
But it did a couple of things.
First of all, it established a
new standard of, of care, it made
the, the sort of clinically and
scientifically important question,
what about longer than fourteen days?
also would have made it difficult to
restrict the storage to less than fourteen
days in the cold stored arm because we
wouldn't have been doing, um, having as
much flexibility as the FDA was allowing.
that external information was reviewed
by the, um, the steering committee or
executive committee for the trial, uh,
the principal investigators, um, Phil
Spinella at, at Pittsburgh, who managed
to hold this whole thing together.
Um, and without knowledge of the
interim data, the decision, um, was
made in conjunction with the DSMB to
immediately move the maximum duration in
the cold stored arm up to the fourteen
days to meet the FDA's, um, new rule.
And what that did is it allowed us to
restart the algorithm with fourteen as
opposed to, uh, the storage duration that
had been allowed up to that point, um,
which was somewhat less than fourteen
Scott: Did this happen w- um, while
enrollment was going on, right?
This was not before.
Roger Lewis: It, it happened
while enrollment was going on
Scott: So I maybe Iâ¦
Anna, do you remember this happening?
And where, where was the limit?
Was it 10 when this triggered?
Anna McGlothlin: So this happened
after our first interim analysis.
So, you know, when the trial started, the
maximum storage duration was seven days.
After the first interim
analysis, the model indicated
that we could go up to 12 days.
So that's what was currently
happening at the time that the
guidance came out and said 14 days.
so at the time, um, once we received
that decision from the study team that
they were going to, you know, essentially
approve going up to 14 days, that's
what was then implemented going forward.
So we had kind of this blip that we
went from 12 to 14, between interims
Scott: Okay.
And, and Nick, you're
unblinded during this trial.
Even though you were involved in
the design, you were unblinded
and helped make the, the model go?
Nick Berry: Yep.
Did the design and transitioned
over to implementation
Scott: Yep, yep.
And I think Kurt Vielhaber jumped in
as a blinded statistician to advise.
Uh, and I think he was the last
blinded human on Earth of not
knowing the results of the trial.
Uh, he and I, I think, were
both the last ones blinded.
Okay, so what happens in the trial then?
During the course of these
adaptations, what are the data saying?
Are, are you being restricted?
Are you escalating?
Generally, the- we'll get to the final
results in the trial, but generally,
what's happening at the interims?
Nick Berry: Um, yeah.
Like Anna said, the at fir-- so I
don't think we explicitly mentioned
this, but the, the cap was that
escalation could only happen five, uh,
by a five-day increase at an interim.
So even if the model recommended
more, it was capped at five.
So at the first interim, up to-- the
first interim analysis up to seven
days, uh, cold storage duration, the
responses on the cold platelets arm
looked very similar to the mean on
the room temperature platelets arm.
So the, the-- They were very
similar, and the model said,
"I'm extrapolating forward.
I think up to some, you know,
larger number, they're going
to be, uh, non-inferior.
Like I can't-- I don't see any reason that
they would be inferior at any duration
close to where I've collected data.
You should escalate," and we escalated
by five, the maximum allowable, uh, step.
Then we had this fun jump from
twelve to fourteen in between
the first and second interim.
Um, and so we're, we're at fourteen
max storage duration, and the next
two interims we again collected data
that wasn't pushing the cold stored
platelet arm's, uh, duration response
model into inferior territory.
It was, it was very similar to what
we were seeing, um, on the warm
stored platelet arm, and so at the,
uh, second and third interim, we
increased by the maximum allowable jump.
We increased by five, then we
increased up to twenty-one days.
Um, we went to, I guess, twelve
to seventeen and seventeen to
twenty-one, which was the max.
We were, you know, that was
the max we were gonna explore.
Um, so when we got to twenty-one
Anna McGlothlin: I was just
gonna jump in, Nick, because
we went from 14, so we didn't
Nick Berry: Well, 14 to ni-
Anna McGlothlin: Yeah,
Nick Berry: yep.
14 to
Anna McGlothlin: go
Nick Berry: 19 to 21,
and then we stayed at 21.
Um, and so we increased by the
most that the trial's pre-specified
rules were allowed to increase
by at every interim analysis
Scott: So Anna, you get to be
involved a lot of times on this team
that is unblinded to the results.
You're talking to the DSMB.
Now you've got a complex
Bayesian model that's driving
decisions, patient safety there.
try to design these so that they're,
the model is providing good results
and the triggers make sense, and
it's, it's, but it's run by algorithm.
Was, how, how was the DSMB receptive
to this modeling and the decisions
during the course of this trial?
Anna McGlothlin: Yeah.
So, um, I, I think this
all went very smoothly.
I think the DSMB was, um, you know,
on board with the design of the trial.
Um, so each interim analysis, you know,
they were able to convene very quickly.
Um, it's one of the, the challenges
sometimes within adaptive design is
getting everybody together, um, to
review the data and, and that, uh,
was able to be done very quickly here.
you know, we had a report that we would
send them after every analysis that
would summarize the data, that would
summarize the model and the adaptive
decision that was pre-specified.
um, you know, Nick had this really
nice dashboard that he put together
that would be in the front of every
report that gave kind of the, the
overall, um, you know, 10,000 foot view
o-of what was happening in the trial.
And that became of the, the way that we
communicated the results to the DSMB.
Um, it was really this one page
with, with several figures on it
that they could look at and get all
of the information in, in one place.
And then if they needed additional
details, those were in the report.
Um, but really focusing on here's,
here's the key information that
they needed to, to make their
recommendations back to the study team.
Scott: Okay, so the trial goes to the
maximum sample size, 1,000 patients, and
I think this was two to one randomized,
so, uh, uh, at the end it was 660 patients
on cold platelets, 340 patients on
room temperature, um, uh, within this.
what's the final result of the trial?
What's the primary analysis
of the trial, Roger?
Roger Lewis: Well, the primary result is
that the, um, the model demonstrates, uh,
non-inferiority at all, all chosen, uh,
storage durations up to twenty-one days.
in fact, the, the raw data looks very flat
with respect to the, the efficacy score.
So the model fit actually looks a
little more pessimistic to the data
because we in-- uh, incorporated
the prior information that, that
platelets were not wine, they could
not get better with age and storage.
Um, and the model appropriately
incorporates that, that information
or, or bias, if you will, if
you will, into its estimates.
So the model strongly, um, uh, supports
the use of them again, um, if you
agree with that non-inferiority margin.
Now, in fact, the data would
have been positive with a much
smaller non-inferiority margin.
There has been some discussion about
whether that non-inferiority margin is
generous, and I think it, it was generous,
but it reflected the fact that in reality
we were d- testing a product that would
be used in places where there was no
alternative as opposed to room temperature
platelets, um, as, as the alternative.
you mentioned
the 2:1 ratio.
Another
complexity of this trial is that
platelets are not a single product.
They are, um, collected using what
are called different platforms.
They're sometimes, um, put into
different, um, solutions or, or,
um, to suspend the platelets.
They may be treated in various
ways, such as pathogen reduction.
So there's lots of different
flavors and subtypes of platelets.
So There was definitely
a desire here to get as much data on
the different subtypes or flavors of
platelets, if you will, so that, um,
at least descriptively, people could
evaluate whether the, um, the success
of the cold-stored platelets looked,
you know, uniform, uh, to the extent we
can tell, um, over the different types
Scott: So the, the final Bayesian
probability was greater than 99.9%
of non-inferiority.
I think the top of the 95% confidence
interval for 21 days was .23,
and the non-inferiority margin
was one, so, so well below that.
And as you described, it was almost like
the model was slightly pessimistic because
it had to be monotonic, uh, very flat.
Now how about other parts to
this, um, uh, this endpoint?
You talked about the, the, uh,
potential of infections with too
long, or part of the primary analysis
was this chest tube measurements.
Did everything else line up, that cold
platelets was similar to room temperature?
Roger Lewis: So I think if you
use the term similar, I think
everything did line up, but there's
definitely some complexity to it.
The, uh, first of all, twenty-four
hour chest tube drainage is a
commonly used regulatory endpoint
for, for hemostatic therapies.
And so that was a secondary endpoint,
um, pre-specified secondary endpoint.
Um, and the chest tube drainages were
similar, although there was numerically
a slight increase in chest tube
drainage in the room-- in the, uh, cold
stored platelet arm as, as I recall.
But, um, there was also a finding that,
um, there was some additional use of
blood products, plasma, red cells, um,
in the patients in the cold stored arm.
So even though the amount of bleeding
clinically looked the same, there was a
little more use of other blood products
in the, in the active or cold stored arm.
there's a couple of possibilities.
Obviously, one possibility is
that the cold stored platelets
don't work quite as well.
Um, and although they were, you know,
clearly met the non-inferiority criteria,
they, they were a little less e-effective.
Another possibility has to do with
the pragmatic nature of the trial.
the trial looked at the, um, total
amount of blood products through
that first twenty-four hours.
And during that twenty-four hours,
people measure the platelet count,
the number of platelets through
cubic millimeter in the blood.
And sometimes our decision to transfuse
platelets or to transfuse other blood
products are influenced a little
bit by just that numerical count.
is reason to believe, based on work done
decades ago, and verified within this
trial, that when you give the same number
of cold stored platelets versus the same
number of room temperature platelets,
the platelet count doesn't go up as
much with the cold stored platelets.
And that probably, um, reflects not
that they aren't working as well,
that they're more activated and they
actually do their job faster and then
are no longer in the bloodstream.
So it may actually be
that they're, they're
Scott: There is no work
Roger Lewis: faster at
the site of the bleeding.
But I think there's the possibility
with the pragmatic design and the
fact that, that those platelet counts
were available to clinicians, that
their decision to transfuse was
at least partially influenced by
knowledge of these platelet counts.
Because traditionally, we think of low
platelet counts as being associated
with more risk of bleeding, so we're
more aggressive about transfusing
Scott: I, by the way, we should mention
this was funded by the US Department of
Defense, uh, who presumably interested.
So coming to the question of the
implications of this result, Roger,
what are the, the, the implications
of the result of this paper?
Roger Lewis: So I think there's both
medical implications, and I think there's
actually some research implications.
So the medical implications, um, that
cold stored platelets are a viable
opport-- um, option for the treatment
of active bleeding in patients, um,
probably in general, but certainly in
settings in which, uh, an inventory
of room temperature platelets
would be difficult to maintain.
So as I mentioned, rural hospitals, low
volume hospitals, um, disaster settings,
military settings, um, and, and the like.
I think the other thing about this, um,
trial that's important, and this is where
the, the design is critical, if you think
about it, we often test products by just
throwing them all together, regardless
of how long they've been stored.
So for a, a blood component, or any
other product that has a, a shelf life,
we simply compare some mixture of the,
of ages some other standard of care,
and we get a single yes, no answer.
And we're really left relatively
naive to whether there is an effect
of, of shelf life storage duration,
whatever it is, on efficacy.
is a trial in which the design
specifically, um, structured and
then extensively evaluated on the
accuracy of getting the duration right.
And as we all know, um, in dose finding
studies, many commonly used dose
finding, um, designs actually aren't
very good at getting the dose right.
They have error rates of, you know,
twenty, thirty percent or, or, or higher.
Here we objectively evaluated how good
we were at getting the storage duration
right and making sure when we got it
wrong, we did so in a conservative
way to minimize risk to patients.
So I think this design, um, really
represents a model for how we ought to
evaluate therapies, both to demonstrate
their efficacy and to figure out how
to get the most value out of them
by being able to use them as long as
they're, um, the right thing to use
Scott: Yeah, I mean, you could
imagine this, this result comes
out in three separate trials where
you do 10 to 14 hours-- for 10
to 14 days, you run the trial.
That's great.
And then you run 14 to 17, and
then you run a whole nother trial.
So yeah, it's brilliant.
I, I love that.
Nick, you were simulating this.
You created a modeling of this.
You simulated scenarios.
You certainly simulated scenarios
where it was really flat, and
that's what it turned out to be.
Uh, whenever we design trials and it
runs, maybe you have some regrets,
could've done this, could've done that.
Any of that in this trial?
Nick Berry: I, I mean, it's terrifying
when a trial runs and you're fitting
a, a model, a complicated model
that, you know, you designed and,
and created, and especially when
that model has constraints on it that
make it inflexible by definition.
Because if platelets had looked
better by some random chance or
something, the model that we fit was
not going to look like the raw data.
And that's an intimidating thing for, you
know, me to stand in front of DSMB and
say, "Here's the data, here's the model.
I know they don't look the same, but
well, plate- platelets aren't wine."
Is that compelling?
Like, do you, do you
believe that, you know?
And so you have toâ¦
A- and I'm, I'm fine.
I l- I don't mind.
I believed in the model, of course.
You know, we fit it 10 million times
probably to different data sets
throughout the course of developing this.
So, um, you know, it, we had, we
had pressure tested it, but it's
still terrifying to, to actually
look at the data and, and see it.
Um, let's see.
I, I had just started.
You know, I, I, I had just joined Berry
Consultants and, uh, you know, I, I
was, I was guided pretty heavily with a
heavy hand on, on where this should go.
But, um, you know, it was a big
adventure in operating characteristics
and simulating these nulls,
and this is a realistic null.
This one's pathological.
And, um, so I, I don't think there's
regrets, but that's easy to say when
things look like they did, right?
Um, the, theâ¦
Everything worked out great, so the
model was perfect for the, for the job
Scott: Yeah.
Yeah.
And in terms of implementation, Anna,
it sounds like this ran very smoothly.
Uh, all good?
Any, any, uh, things you'd
change in this or lessons?
Anna McGlothlin: Yeah, I think this is
a really nice success story that we can
point to that trials like this can be run.
You know, this was a fairly
complicated design with the
escalation of the different storage
durations, with the modeling,
and it, it all ran very smoothly.
So I think with the right preparation and,
and, um, you know, setting things, uh,
in place for the communication channels
and all of that, this can absolutely be
done, even a complicated design like this
Scott: Yeah.
And, and I reiterate something you
said that we're told very frequently
when we wanna do an adaptive design
that each interim is six weeks,
and this is gonna be problematic.
That, you know, data was delivered
in a day, and the results were
delivered in five business
days, and the trial moved on.
Uh, fantastic, fantastic effort.
Uh, Roger, uh, any last words
on the trial, next steps?
Roger Lewis: Well, I just think
we, we need to acknowledge
the, the partners that we had.
Anna mentioned the, the Data Coordinating
Center at Utah, the investigator team, um,
which was, uh, you know, b-based at many
institutions but that led by Phil Spinella
and his colleagues, and the folks behind
the scene that made this all happen.
the, the, the work to get the cold
stored platelet inventory, get it
shipped, make sure there's enough of it
at each of the sites so that patients
can get almost always the, the, the
arm to which they're randomized, was
really a logistical tou-tour de force.
And so I think the, the combination of
the folks on the ground were completely
committed to the success, people willing
to do things differently for the data
availability, as you mentioned, and then
the presentation from the implementation
team that gave us just snapshot that
you could look at and just immediately
know what was going on with the trial
and what the right thing to do was,
is all the pieces sort of, umâ¦
Each one of the pieces was executed
so well, and it answered the question.
My only regret is that we
didn't, um, build the design to
randomize up to twenty-eight days.
I sort of feel like we were, we, we
were on a roll and, and it wouldn't
have been hard to get an extra week
and see if there was actually any
upturn, um, towards inferiority
Scott: Right.
Nice.
Well, we'll, we'll move that
to, to next step certainly.
And, and number of really cool
questions within trauma, within
bleeding, uh, exciting trials for that.
So thank you Roger,
Anna, Nick, for joining.
Fantastic, fantastic trial, fantastic
result, fantastic clinical impact
And appreciate everybody for listening.
Uh, until next time, we'll
be here in the interim