The CHIPS Trial: Bayesian Adaptive Trial in JAMA

WEBVTT
NOTE
This file was generated by Descript
00:00:00.029 --> 00:00:03.579
Judith: Welcome to Berry's In the
Interim podcast, where we explore the
00:00:03.579 --> 00:00:07.650
cutting edge of innovative clinical
trial design for the pharmaceutical and
00:00:07.650 --> 00:00:09.800
medical industries, and so much more.
00:00:10.289 --> 00:00:10.979
Let's dive in.
00:00:12.311 --> 00:00:12.711
Scott: All right.
00:00:12.711 --> 00:00:15.831
Welcome everybody back to In The Interim.
00:00:16.621 --> 00:00:17.901
your host, Scott Berry.
00:00:18.401 --> 00:00:24.241
And on today's episode, we are
going to dive into the creation, the
00:00:24.241 --> 00:00:29.521
design, the implementation, and the
results of a Bayesian adaptive trial.
00:00:29.741 --> 00:00:31.541
Really a pretty cool trial.
00:00:31.931 --> 00:00:34.151
Uh, the, the name is cool.
00:00:34.441 --> 00:00:39.371
Uh, this is the CHIPS trial, uh,
Chilled Platelet Study, and we'll
00:00:39.371 --> 00:00:41.441
talk more about what that means.
00:00:41.811 --> 00:00:47.741
The results of this trial were
recently published August 17th 2026.
00:00:48.051 --> 00:00:53.241
The results showed up online
in JAMA, Spinella et al.
00:00:54.231 --> 00:00:58.351
today I have three guests,
all co-authors on this paper.
00:00:59.311 --> 00:01:01.721
my three guests, uh, I'll start with Dr.
00:01:01.721 --> 00:01:02.831
Roger Lewis.
00:01:03.111 --> 00:01:07.741
Uh, Roger is a senior medical scientist
at Berry Consultants and a professor of
00:01:07.741 --> 00:01:13.441
emergency medicine at the David Geffen
School of Medicine at UCLA, and he's an
00:01:13.451 --> 00:01:18.681
elected member of the National Academy
of Medicine and a fellow of the ASA.
00:01:19.171 --> 00:01:20.651
Pretty cool combination.
00:01:21.321 --> 00:01:22.021
Uh, Dr.
00:01:22.021 --> 00:01:26.741
Anna McLaughlin is a senior statistical
scientist at Berry Consultants, and
00:01:26.741 --> 00:01:32.321
she's a director, and she specializes
in implementation of adaptive trials,
00:01:32.321 --> 00:01:36.781
complex trials, platform trials, and
that was her role on the CHIPS trial,
00:01:36.961 --> 00:01:40.291
and we'll get to talk to her about
implementing this adaptive design.
00:01:41.151 --> 00:01:42.241
And Dr.
00:01:42.241 --> 00:01:46.891
Nick Berry is a senior statistical
scientist at Berry Consultants, and
00:01:46.891 --> 00:01:52.851
Nick, uh, focuses on our Phac software,
software design, uh, the software team.
00:01:52.861 --> 00:01:59.081
Uh, he was involved in the design
of this trial, and that goes all
00:01:59.091 --> 00:02:01.611
the way back to, I think, 2019.
00:02:01.611 --> 00:02:05.701
And by the way, he picked up for Liz
Krejci, who was at Berry Consultants back
00:02:05.701 --> 00:02:07.521
then, and she started some of this work.
00:02:07.761 --> 00:02:10.071
So welcome, everybody, to In The Interim
00:02:11.782 --> 00:02:12.242
Nick Berry: Thanks
00:02:13.455 --> 00:02:14.565
Scott: So let's set up.
00:02:14.595 --> 00:02:20.165
We, we-- This is the CHIPS trial,
Roger, uh, Chilled Platelets Study.
00:02:20.535 --> 00:02:25.735
Uh, us what the trial is, what's the
clinical context, and what are the
00:02:25.745 --> 00:02:28.625
inter- inter- interventions in the trial?
00:02:30.579 --> 00:02:34.049
Roger Lewis: So the CHIPS trial
studied, um, platelet transfusion.
00:02:34.059 --> 00:02:37.299
So platelets are part of the
blood that help in clotting.
00:02:37.699 --> 00:02:41.869
They're commonly administered when
patients, um, are bleeding, need, um,
00:02:41.879 --> 00:02:44.709
blood products, uh, often during surgery.
00:02:45.089 --> 00:02:49.829
And one of the challenges with platelets
is that traditionally they're obtained
00:02:49.859 --> 00:02:55.439
from people and, and, uh, purified, and
then they're stored at room temperature.
00:02:56.049 --> 00:03:00.539
And because of the storage at room
temperature, they go bad very quickly.
00:03:00.859 --> 00:03:06.049
So maximum storage duration is
usually, uh, five days, sometimes
00:03:06.049 --> 00:03:09.289
up to seven days if you do
repeated testing for contamination.
00:03:09.669 --> 00:03:12.299
But after that period of time,
they can no longer be used.
00:03:12.639 --> 00:03:15.459
So because of that, platelets
are really only available at
00:03:15.469 --> 00:03:19.879
hospitals and blood banks that
have a lot of turnover in products.
00:03:19.879 --> 00:03:24.939
And so there's not a supply of platelets
available to patients that are treated at
00:03:24.939 --> 00:03:28.339
low-volume hospitals or in rural settings.
00:03:28.749 --> 00:03:32.569
And one of the ways to try to fix this
problem is to figure out a way we can
00:03:32.569 --> 00:03:38.789
store platelets for a longer period of
time so that they can be, um, kept in
00:03:38.819 --> 00:03:44.469
inventory at places, um, practically
that now can't currently, um, have them.
00:03:45.009 --> 00:03:50.239
So the CHIPS trial compared the
traditional room temperature platelets
00:03:51.119 --> 00:03:55.409
to platelets that were stored in
the cold, essentially refrigerator
00:03:55.409 --> 00:03:58.779
temperature, and potentially
stored for a longer period of time.
00:03:59.579 --> 00:04:06.429
the clinical model was, um, both children
and adults, including very small children,
00:04:06.899 --> 00:04:12.079
undergoing complex cardiac surgery because
that's a clinical setting in which we
00:04:12.079 --> 00:04:16.399
expect there to be substantial bleeding
and platelets are routinely administered
00:04:19.415 --> 00:04:22.785
Scott: Okay, so I, as, as somebody who
doesn't understand the sci- science
00:04:23.075 --> 00:04:29.315
of platelets, if, if you were to store
them for too long, what, what bad things
00:04:29.315 --> 00:04:31.415
happen if this is a bad thing to do?
00:04:31.745 --> 00:04:36.215
And is it different than warm
platelets go bad to some extent?
00:04:36.485 --> 00:04:41.605
If I were to store platelets for a
year cold and then use them and they
00:04:41.605 --> 00:04:43.815
were not functional, what would happen?
00:04:45.779 --> 00:04:47.669
Roger Lewis: So I think there's a
couple of qu- different questions there.
00:04:47.669 --> 00:04:52.169
So the most complication of storing
platelets too long is that they're
00:04:52.169 --> 00:04:56.419
con- contaminated by bacteria,
a very small number, and over a
00:04:56.419 --> 00:04:58.819
period of time, they multiply.
00:04:59.119 --> 00:05:02.089
And so when you transfuse those
platelets that are contaminated with
00:05:02.099 --> 00:05:05.459
bacteria, the patient develops an
infection, clinical sepsis, and that's
00:05:05.459 --> 00:05:07.209
a very bad thing when it happens.
00:05:07.769 --> 00:05:11.299
And it's just very difficult
to collect a blood product, you
00:05:11.299 --> 00:05:14.939
know, you pierce the skin, there's
bacteria on the skin, without ever
00:05:14.939 --> 00:05:16.959
getting any bacteria in at all.
00:05:17.269 --> 00:05:22.269
So just like, you know, milk or any
other product that goes bad, can-- even
00:05:22.269 --> 00:05:26.389
if there's a teeny bit of contamination,
you can store it for a while safely.
00:05:26.869 --> 00:05:30.429
But if you allow the bacteria
too much time to, to multiply,
00:05:30.429 --> 00:05:31.749
you can get into problems.
00:05:32.249 --> 00:05:36.119
The other issue with storing platelets
too long is they lose what we call their
00:05:36.189 --> 00:05:40.939
hemostatic efficacy, which is their
ability to help the blood clot in those
00:05:40.949 --> 00:05:43.049
settings in which clotting is important.
00:05:43.059 --> 00:05:46.059
Like for example, if a surgeon is
cutting you or you have a wound
00:05:48.057 --> 00:05:52.677
Scott: Okay, and so, so a- and hence
that's sort of the, the, this trial
00:05:52.947 --> 00:05:55.687
is about how long we can cool them.
00:05:55.687 --> 00:06:00.607
So the, the two interventions are
room temperature stored platelets,
00:06:00.807 --> 00:06:04.477
and that's up to the five to
seven days, up to seven days.
00:06:04.857 --> 00:06:12.317
And the other interventions are cold
stored platelets, where the time you
00:06:12.317 --> 00:06:15.167
store them may vary throughout the trial.
00:06:15.237 --> 00:06:17.457
And so what is the goal of this trial?
00:06:19.261 --> 00:06:21.831
Roger Lewis: Well, the goal is
to prevent excessive bleeding.
00:06:22.321 --> 00:06:27.511
So as the, um, the patient undergoes
their operation and in the immediate
00:06:27.511 --> 00:06:31.741
postoperative period, up to twenty-four
hours, um, after the surgery, the
00:06:31.741 --> 00:06:33.541
patient is at risk for bleeding.
00:06:33.871 --> 00:06:37.871
And bleeding has a lot of
important clinical implications.
00:06:38.181 --> 00:06:43.661
You know, not only might you need blood
transfusions, um, if you're bleeding
00:06:43.661 --> 00:06:48.711
excessively, but if their bleeding
cannot be controlled, sometimes patients
00:06:48.711 --> 00:06:52.521
require reoperation, meaning actually
going back in, trying to find out where
00:06:52.521 --> 00:06:56.521
the bleeding is coming from and, and
physically or surgically dealing with it.
00:06:57.251 --> 00:07:03.651
So, um, the, the period of most
interest is in the first twenty-four
00:07:03.651 --> 00:07:08.921
hours after surgery, and these patients
commonly have chest tubes in place,
00:07:08.941 --> 00:07:13.261
which are tubes that drain any excess
bleeding from around the lungs.
00:07:14.061 --> 00:07:21.041
And it is a very common model to test,
FDA, um, regulated products that are
00:07:21.041 --> 00:07:27.821
intended to support clotting or hemostasis
to use the complex cardiac surgery model
00:07:27.881 --> 00:07:30.781
and to look at things like the amount
of bleeding or the amount of chest
00:07:30.781 --> 00:07:34.181
tube output, which is basically the
blood coming out of, out of the chest.
00:07:34.881 --> 00:07:39.331
One of the features of this trial was
that it used a composite bleeding score,
00:07:40.051 --> 00:07:44.761
is a five-level ordinal scale that
had been developed for these kinds of
00:07:44.761 --> 00:07:49.881
applications, with one being the least
bleeding, five being substantial bleeding.
00:07:50.011 --> 00:07:55.021
So it used a sort of holistic and, and
patient-centered measure of the overall
00:07:55.031 --> 00:07:59.051
bleeding with the goal of demonstrating
that cold-stored platelets stored for
00:07:59.051 --> 00:08:04.191
some maximum duration was non-inferior
to room temperature platelets,
00:08:04.381 --> 00:08:08.221
and therefore could be safely and
appropriately deployed in settings where
00:08:08.221 --> 00:08:09.751
platelets are not currently available
00:08:11.069 --> 00:08:16.169
Scott: Okay, so, um, the, the sort
of objective then is to understand
00:08:16.449 --> 00:08:22.299
the longer time they're stored cold
platelets, are they still non-inferior
00:08:22.669 --> 00:08:27.499
to room, uh, uh, temperature platelets?
00:08:28.569 --> 00:08:34.959
we're gonna use this homeostatic efficacy
score for, for judging that, th- this,
00:08:35.019 --> 00:08:37.509
this bleeding score throughout the trial.
00:08:37.569 --> 00:08:38.079
Okay.
00:08:38.359 --> 00:08:42.079
So Nick, tell me about then
the ap- adaptive design.
00:08:42.389 --> 00:08:47.329
We're, we're not gonna start
by randomizing patients to 21
00:08:47.339 --> 00:08:50.129
days of cold stored platelets.
00:08:50.289 --> 00:08:51.349
How does the desi- how does
the adaptive design work?
00:08:54.714 --> 00:08:57.394
Nick Berry: Yeah, the idea is that,
uh, room temperature platelets
00:08:57.394 --> 00:08:58.814
can be stored for seven days.
00:08:58.874 --> 00:09:02.404
Um, so at the onset of the trial,
when we start randomizing patients,
00:09:02.424 --> 00:09:08.074
we're going to allow cold store
platelets to be stored up to seven
00:09:08.074 --> 00:09:10.454
days, um, before they're administered.
00:09:10.454 --> 00:09:13.934
And so the, the arms start, you know,
the first patient in is randomized
00:09:13.944 --> 00:09:18.534
to either room temperature platelets
or cold stored platelets that are
00:09:18.534 --> 00:09:20.574
stored for less than seven days
00:09:23.078 --> 00:09:29.508
As the trial runs, hopefully, the
idea is that we'll demonstrate that
00:09:29.628 --> 00:09:34.948
seven days of storage on the cold
stored arm is non-inferior to room
00:09:34.948 --> 00:09:36.518
stor-- room temperature platelets,
00:09:36.642 --> 00:09:37.119
Scott: platelets.
00:09:37.268 --> 00:09:42.198
Nick Berry: able to say, "Okay, seven
seems safe versus results are as good, if
00:09:42.198 --> 00:09:43.578
not better, than the room temperature arm.
00:09:43.588 --> 00:09:48.688
Let's increase the maximum allowed storage
duration on the cold stored platelet arm."
00:09:49.538 --> 00:09:53.898
And so we des-devised a scheme in
advance, a sort of ladder that you
00:09:53.898 --> 00:09:57.328
can-- that the, the trial could
climb, um, throughout the interims,
00:09:57.399 --> 00:09:57.699
Scott: interim,
00:09:57.818 --> 00:10:00.598
Nick Berry: we scheduled at two
hundred, four hundred, six hundred,
00:10:00.598 --> 00:10:06.428
and eight hundred subjects, uh,
randomized, um, patients randomized.
00:10:07.518 --> 00:10:14.178
And at each of those interim analyses,
the cold stored platelet arm's maximum
00:10:14.178 --> 00:10:18.628
allowable duration was allowed to increase
based on the results of the model, um,
00:10:18.638 --> 00:10:20.218
based on the data collected at that point.
00:10:20.218 --> 00:10:25.378
And so this, the, the, the original
idea was that we could construct
00:10:25.378 --> 00:10:27.828
a model that would say, "Yeah,
here's where you should go.
00:10:27.878 --> 00:10:33.028
This is the, the maximum allowable
duration that the model says is safe."
00:10:33.068 --> 00:10:38.628
And it became clear as we put that
together that there was a balance in
00:10:38.638 --> 00:10:46.008
being aggressive in increasing that
maximum duration and not putting
00:10:46.008 --> 00:10:47.718
patients at undue risk, essentially.
00:10:47.718 --> 00:10:53.718
So the, the way we, we structured
this was we had a model to model the
00:10:53.718 --> 00:10:58.548
cold stored arm with the duration
response included in that estimation.
00:10:58.908 --> 00:11:03.258
So, you know, when a, a
patient is randomized to the
00:11:03.258 --> 00:11:04.808
cold stored arm, maybe their
00:11:04.922 --> 00:11:05.277
Scott: get
00:11:05.418 --> 00:11:07.788
Nick Berry: I'm gonna simplify this for
a second, maybe they get platelets stored
00:11:07.788 --> 00:11:10.518
for twelve days on the cold stored arm.
00:11:11.928 --> 00:11:15.618
Each patient has a hemostatic
efficacy score, one of those one to
00:11:15.618 --> 00:11:18.438
fives Roger mentioned, along with
the duration, and you can imagine
00:11:18.438 --> 00:11:22.678
a curve over the cold storage time
a-and, and we fit a model to that.
00:11:23.048 --> 00:11:28.038
It ended up being what we'll
call a piecewise linear model.
00:11:28.118 --> 00:11:31.548
Um, you know, it can wiggle
when it needs to, um, but it,
00:11:31.548 --> 00:11:32.898
it liked to be a straight line.
00:11:32.898 --> 00:11:36.248
So we tried to limit the, the
flexibility of the model while
00:11:36.248 --> 00:11:42.958
still allowing a possible, um, curve
in it as, as durations increased.
00:11:43.068 --> 00:11:48.808
Um, and we compared to basically just
a, a simple model on the warm arm.
00:11:49.458 --> 00:11:54.568
So this is just a basic, uh, normal,
normal posterior where, um, the, the,
00:11:54.628 --> 00:11:57.988
and then, you know, the cold arm versus
the warm arm, we'd look at where it
00:11:57.988 --> 00:12:04.658
crossed to be more than one point worse
than the, the warm stored platelet arm.
00:12:06.398 --> 00:12:07.758
Uh, there's a lot to mention here.
00:12:07.778 --> 00:12:08.078
Yeah.
00:12:08.167 --> 00:12:10.557
Scott: so, so that, that's a
Bayesian model, why we're calling
00:12:10.567 --> 00:12:11.717
it Bayesian adaptive design.
00:12:12.017 --> 00:12:16.127
But so you've got this Bayesian
model that's monotonic, that's an
00:12:16.127 --> 00:12:20.897
important part of this, and you have
this piecewise exponential so that
00:12:20.927 --> 00:12:27.997
the longer it's stored, the worse the
average bleeding score i- has to be
00:12:28.596 --> 00:12:29.906
Nick Berry: Yeah, that's necessarily true.
00:12:29.906 --> 00:12:31.416
This is a monotonic model, like you said.
00:12:31.416 --> 00:12:35.646
So, uh, fifteen days stored cold
has to have a worse estimate
00:12:35.646 --> 00:12:37.036
than seven days stored cold.
00:12:37.436 --> 00:12:42.526
Um, it can get pretty flat, uh, which,
you know, is part of the goal was to
00:12:42.526 --> 00:12:46.796
both let it curve but also be straight
and flat, uh, as the data recommended.
00:12:47.215 --> 00:12:47.405
Scott: There are a
00:12:47.596 --> 00:12:47.756
Nick Berry: Yep.
00:12:47.815 --> 00:12:51.835
Scott: similarities to this to a
phase one oncology dose escalation
00:12:51.835 --> 00:12:55.685
trial where we're starting slow,
we're escalating as long as it's safe.
00:12:55.685 --> 00:13:00.115
We've got a Bayesian logistic regression
model that's allowing us to, to escalate.
00:13:00.125 --> 00:13:01.915
You're doing this, the same thing here.
00:13:02.215 --> 00:13:06.785
You spent a huge amount of time with
different models, their ability to
00:13:06.795 --> 00:13:10.035
escalate, their ability to get the
right answer under a lot of different
00:13:10.035 --> 00:13:16.105
scenarios, simulating cases where nine
days was the last non-inferior one.
00:13:16.115 --> 00:13:17.265
Do we get a good answer?
00:13:17.265 --> 00:13:19.925
What if, what if all
doses are non-inferior?
00:13:19.925 --> 00:13:21.465
What if no doses are safe?
00:13:21.715 --> 00:13:27.645
So you did a huge range of potential
truths to make sure the design didn't put
00:13:27.645 --> 00:13:34.025
patients at risk, got the right answer,
uh, across the range of possible truths
00:13:35.316 --> 00:13:35.796
Nick Berry: Yeah.
00:13:35.796 --> 00:13:36.096
Yeah.
00:13:36.236 --> 00:13:38.286
So you mentioned the
phase one dose escalation.
00:13:38.336 --> 00:13:43.656
Uh, one of the, I guess the harder parts
that took some, some iteration and some
00:13:43.656 --> 00:13:51.086
work to get right was like in a da-- uh,
phase one escalation study, how good you
00:13:51.086 --> 00:13:56.326
are at estimating kind of at the upper
limit of your range really dictates how
00:13:56.326 --> 00:13:58.256
good you are at driving that escalation.
00:13:58.266 --> 00:14:05.356
And if you're estimating cold stored
platelet efficacy for durations
00:14:05.476 --> 00:14:09.186
around five really well, but you're
not estimating up by twelve and
00:14:09.186 --> 00:14:13.066
fourteen if that's near your max, you
don't know anything about the max.
00:14:13.066 --> 00:14:15.206
It's really hard to push
the envelope and escalate.
00:14:15.206 --> 00:14:15.646
And so
00:14:15.697 --> 00:14:16.049
Scott: most
00:14:16.066 --> 00:14:22.336
Nick Berry: both the model had to,
you know, drive that maximum allowable
00:14:22.336 --> 00:14:27.746
duration to large values, and we
had to, to try to estimate that.
00:14:27.746 --> 00:14:34.126
But we also had to-- Well, I say we, um,
the study team really had to devise a way
00:14:34.446 --> 00:14:42.136
to actually collect patients that-- or
administer platelets near the max, and I
00:14:42.146 --> 00:14:47.306
think that was a huge driving, uh, kind
of b- I'll say behind the scenes for us.
00:14:47.306 --> 00:14:50.356
I'm sure it was, uh, in front of the
scenes for many people, but that was
00:14:50.356 --> 00:14:52.906
a really, really important part of
the operating characteristics that
00:14:52.906 --> 00:14:55.866
we saw when we were simulating the
study, is that the study worked really
00:14:55.866 --> 00:14:58.996
well if we could collect patient or
administer platelets near the max.
00:14:59.266 --> 00:15:02.796
It did not work very well if we
couldn't actually store platelets near
00:15:02.796 --> 00:15:04.246
the, the maximum allowable duration.
00:15:04.246 --> 00:15:07.906
And so, um, yeah, that was a,
a interesting characteristic of
00:15:07.906 --> 00:15:08.856
the simulations in this case.
00:15:09.463 --> 00:15:12.013
Scott: I think this is
unique, unique to this trial.
00:15:12.013 --> 00:15:16.773
And I, um, just to make clear,
you can't say, "Give me, uh,
00:15:17.163 --> 00:15:19.293
14 days cold platelet," Roger.
00:15:19.673 --> 00:15:23.113
So when a patient's assigned to cold
platelets, you might have a range.
00:15:23.113 --> 00:15:26.623
So this is a little bit different
than a traditional assignment trial
00:15:28.563 --> 00:15:29.823
Roger Lewis: Yeah, that's,
that's exactly right.
00:15:29.823 --> 00:15:34.153
And I think I wanna make a distinction
between the maximum allowable
00:15:34.163 --> 00:15:37.543
duration, which is what was stepped
up in the ladder that, that Nick
00:15:37.563 --> 00:15:40.683
mentioned, what patients receive.
00:15:41.543 --> 00:15:46.443
patients receive-- it's as if you
were doing dose escalation, and as
00:15:46.443 --> 00:15:50.263
you do dose escalation, you set the
maximum dose, but then someone randomly
00:15:50.263 --> 00:15:52.133
decides to give any dose under that.
00:15:52.573 --> 00:15:55.443
That's not as efficient an
experimental design as when you
00:15:55.443 --> 00:15:57.553
can actually manipulate the dose.
00:15:57.933 --> 00:16:03.433
And so the study team worked incredibly
hard behind the scenes to try to
00:16:03.433 --> 00:16:08.393
administer platelets closest to the
maximum duration because that gave the
00:16:08.393 --> 00:16:10.763
model the most information to learn from.
00:16:11.233 --> 00:16:15.683
But I will say if that hadn't been
done successfully, the model would
00:16:15.683 --> 00:16:19.613
have actually known that and known that
it didn't have very much information
00:16:19.613 --> 00:16:23.393
near the upper end of the allowable
storage duration, and so it just
00:16:23.403 --> 00:16:25.763
wouldn't have allowed escalation.
00:16:26.123 --> 00:16:29.303
And I'm not sure if Nick mentioned
that the, the model only allowed
00:16:29.303 --> 00:16:32.113
a certain amount of each time.
00:16:32.453 --> 00:16:38.323
Even if the current estimate was
that cold stored platelets were after
00:16:38.323 --> 00:16:41.983
twenty-one days of storage, if you were
at ten days of storage, it would only
00:16:41.983 --> 00:16:43.913
allow you to escalate a little bit.
00:16:44.123 --> 00:16:49.843
So it sort of carefully explored
upward in the maximum duration space.
00:16:50.433 --> 00:16:54.063
And then as you mentioned, all
the work Nick did to look at
00:16:54.185 --> 00:16:54.235
Scott: to look
00:16:54.313 --> 00:16:59.623
Roger Lewis: accuracy of finding the
maximum storage duration at which
00:16:59.623 --> 00:17:01.883
non-inferiority first failed, we
00:17:02.009 --> 00:17:03.439
Scott: We tried very hard
to make sure that when we
00:17:03.673 --> 00:17:04.043
Roger Lewis: we
00:17:04.399 --> 00:17:06.749
Scott: didn't get the right answer,
we didn't get the right answer in a
00:17:06.813 --> 00:17:06.933
Roger Lewis: the
00:17:07.209 --> 00:17:08.469
Scott: conservative direction and
00:17:08.643 --> 00:17:08.773
Roger Lewis: to
00:17:08.789 --> 00:17:09.249
Scott: try to
00:17:09.433 --> 00:17:09.793
Roger Lewis: put, put patients at risk
00:17:13.665 --> 00:17:15.885
Scott: Okay, so you've
got an adaptive design.
00:17:16.165 --> 00:17:21.575
Um, you, you have these escalation
rules, a governor on escalation,
00:17:21.575 --> 00:17:23.805
a Bayesian model driving this.
00:17:23.815 --> 00:17:27.065
That's the primary analysis at
the end is this Bayesian model.
00:17:27.555 --> 00:17:31.825
Uh, you've gone to the FDA, so
the FDA's reviewed this design
00:17:34.228 --> 00:17:35.248
Nick Berry: Yep, they did.
00:17:35.438 --> 00:17:38.098
And can we-- we, we should probably,
as we talk about FDA, we should talk
00:17:38.108 --> 00:17:42.388
about, uh, type one error and the
null case because what is the null in
00:17:42.388 --> 00:17:46.218
this trial is an interesting question
and something we discussed with FDA.
00:17:46.818 --> 00:17:51.438
Um, we simulated a lot of
different null scenarios, right?
00:17:51.468 --> 00:17:56.548
Uh, and, and the primary analysis, as in
was it a type one error in this case was,
00:17:58.568 --> 00:18:04.318
was seven days cold store duration deemed
non-inferior to warm platelets, um,
00:18:04.328 --> 00:18:06.068
was the, I guess, the primary analysis.
00:18:06.088 --> 00:18:10.438
And then after that, we would say, "Okay,
what's the maximum cold storage duration
00:18:10.438 --> 00:18:12.898
that's non-inferior to warm stored
platelet, to room temperature platelet?"
00:18:13.558 --> 00:18:18.088
Um, and so we, we simulated a
bunch of different reasonable nulls
00:18:18.195 --> 00:18:18.565
Scott: Yeah, so
00:18:18.598 --> 00:18:19.098
Nick Berry: in that case
00:18:19.165 --> 00:18:19.815
Scott: a null.
00:18:19.835 --> 00:18:26.415
Presumably a null is where cold
store platelets are inferior.
00:18:28.084 --> 00:18:32.294
Nick Berry: They crossed to in-
inferiority at exactly seven days.
00:18:32.847 --> 00:18:33.037
Scott: Yeah.
00:18:33.574 --> 00:18:33.704
Nick Berry: yeah
00:18:34.207 --> 00:18:37.117
Scott: Okay, so a- and so that,
that was a case particularly
00:18:37.217 --> 00:18:38.877
s- worrisome I, I imagine.
00:18:39.307 --> 00:18:43.777
But then at the same time, suppose
it crosses at 11 days, and up
00:18:43.777 --> 00:18:46.257
to 11 days is non-inferior.
00:18:47.077 --> 00:18:53.417
you looking at the rate at which you say
it's higher than 11 and making a mistake
00:18:53.417 --> 00:18:55.887
on that and trying to control that?
00:18:57.044 --> 00:18:57.474
Nick Berry: Yes.
00:18:57.724 --> 00:18:58.804
Yeah, that's also part of this.
00:18:58.814 --> 00:19:03.314
So I, I've never seen this before in an
adaptive design, but I think our power
00:19:03.314 --> 00:19:08.124
was one in every single scenario we
reported in the, the protocol because,
00:19:09.284 --> 00:19:14.214
you know, the, the, the scenarios we
simulated, the seven-day, uh, comparison
00:19:14.214 --> 00:19:15.474
wasn't actually the interesting part.
00:19:15.474 --> 00:19:16.304
It was what happened later.
00:19:16.304 --> 00:19:20.264
And so we reported a lot of things
like were we within three days?
00:19:20.274 --> 00:19:23.254
How often did we estimate something
bigger than the true crossing?
00:19:23.354 --> 00:19:24.454
That rate was really low
00:19:24.605 --> 00:19:24.715
Scott: Yeah.
00:19:24.764 --> 00:19:25.284
Nick Berry: at the end of it.
00:19:25.324 --> 00:19:29.764
Um, well below what you'd check for
like a type one error type of level.
00:19:29.795 --> 00:19:29.825
Scott: Okay.
00:19:29.854 --> 00:19:33.114
Nick Berry: Um, you know, one
percent, less than one percent,
00:19:33.144 --> 00:19:33.944
uh, if I remember right.
00:19:34.014 --> 00:19:37.284
How many patients were administered
platelets that were truly inferior,
00:19:37.314 --> 00:19:41.334
things like that, where we were trying to
govern how aggressive we were with, with
00:19:41.334 --> 00:19:45.574
administering platelets to, you know,
at high durations early in the study.
00:19:46.197 --> 00:19:46.307
Scott: Yeah.
00:19:46.767 --> 00:19:49.807
That, that's, uh, that whole
discussion is so fantastic in what,
00:19:49.877 --> 00:19:51.317
what actually happened in the trial.
00:19:51.677 --> 00:19:55.147
Um, the developments of the
trial, which we'll get to.
00:19:55.627 --> 00:20:00.497
Okay, so you turn this over, and
now we need to implement this trial.
00:20:00.747 --> 00:20:03.947
We need to do these multiple
interims every 200 patients.
00:20:04.317 --> 00:20:08.267
Uh, we need to be running these
models, uh, doing escalation.
00:20:08.687 --> 00:20:12.047
Anna, this gets turned over and
you're on the implementation type.
00:20:12.047 --> 00:20:18.117
Take us a little bit through the poss- the
process of preparing to execute this trial
00:20:19.754 --> 00:20:20.114
Anna McGlothlin: Sure.
00:20:20.134 --> 00:20:24.954
So as with any adaptive trial, the
key to making it run smoothly is in
00:20:24.954 --> 00:20:26.604
that preparation phase, like you said.
00:20:27.414 --> 00:20:31.064
so we had a great team that we worked
with at the data coordinating center,
00:20:31.094 --> 00:20:33.264
which was University of Utah here.
00:20:33.624 --> 00:20:38.644
Um, so early on, we had discussions
with them about the data that we would
00:20:38.644 --> 00:20:42.754
need in order to run the analysis,
kind of spelled out, here's these key
00:20:42.754 --> 00:20:45.864
variables that we need to run the interim.
00:20:46.264 --> 00:20:50.704
So, you know, we, we don't need the full
database, we just need a, a, a subset
00:20:50.814 --> 00:20:55.404
o-of key variables, and those are the
ones that they focused on cleaning and,
00:20:55.434 --> 00:20:57.634
and, um, getting put together for us.
00:20:58.414 --> 00:21:02.394
when we hit that interim trigger
at, at 200 patients, 400 patients,
00:21:02.574 --> 00:21:06.034
they were very efficient at being
able to hand that data over to us.
00:21:06.324 --> 00:21:09.924
I think, if I remember correctly, it was
about one or two days from the time we
00:21:09.924 --> 00:21:14.914
hit the trigger until the data arrived,
um, for us to start our analysis for that.
00:21:15.274 --> 00:21:18.064
Um, so that was really amazing
that they could do that for us.
00:21:18.504 --> 00:21:22.604
Um, and then on our side, you know,
ahead of time before we got to the first
00:21:22.614 --> 00:21:29.054
interim, um, we spent time putting all
of our programs together, um, getting
00:21:29.084 --> 00:21:34.554
our report shell in place, thinking about
how we were gonna present the data to
00:21:34.614 --> 00:21:36.334
the Data and Safety Monitoring Board.
00:21:36.664 --> 00:21:42.414
Nick put together some really cool plots,
um, to summarize, the data and how the
00:21:42.414 --> 00:21:46.444
model was performing so that, uh, the
committee had the information they needed
00:21:46.454 --> 00:21:50.844
in a really succinct way to be able
to monitor that as the trial went on.
00:21:51.964 --> 00:21:55.874
you know, a couple things we were looking
at really carefully, as was mentioned,
00:21:55.874 --> 00:22:01.114
this is a non-inferiority trial, um, so
we wanted to keep an eye on things like,
00:22:01.454 --> 00:22:04.904
uh, are patients getting the treatment
they were actually randomized to?
00:22:04.994 --> 00:22:09.784
Or if a patient was randomized to cold
platelets, are they actually getting
00:22:09.984 --> 00:22:13.164
room temperature platelets or are
they getting some kind of a mixture?
00:22:13.494 --> 00:22:16.644
Um, so that was something
we, we kept an eye on.
00:22:16.644 --> 00:22:19.694
We wanted to make sure we summarized
that information for DSMB.
00:22:21.094 --> 00:22:24.974
uh, and, and, you know, our
analysis was done very quickly.
00:22:25.024 --> 00:22:30.464
So by the time we received the data until
we turned around a report to the DSMB
00:22:30.834 --> 00:22:35.204
was about five business days, um, to do
that process and have it in their hands.
00:22:36.154 --> 00:22:40.744
at the same time, whenever we completed
our analysis, we would prepare a memo to
00:22:40.744 --> 00:22:44.984
go to an unblinded person that, a-as Roger
said, it was a big logistical challenge
00:22:44.984 --> 00:22:50.894
to make sure that the, the storage,
um, the supply, um, of platelets to the
00:22:50.894 --> 00:22:53.764
sites, um, really maximized that duration.
00:22:53.764 --> 00:22:55.774
They had the right set
of platelets available.
00:22:55.774 --> 00:23:00.114
So we would communicate that to someone
who was managing that logistical supply.
00:23:00.874 --> 00:23:02.464
and that was all kind
of happening in parallel
00:23:07.575 --> 00:23:07.855
Scott: Okay.
00:23:08.145 --> 00:23:12.205
So, so the trial starts and Roger,
there's, uh, the interesting
00:23:12.205 --> 00:23:14.485
discussions with the, the, the FDA.
00:23:14.735 --> 00:23:18.785
There's an unplanned adaptation,
if you will, in the trial, in
00:23:18.785 --> 00:23:22.075
the whole escalation, a bit
of a wrench into the trial.
00:23:22.075 --> 00:23:22.815
What happened?
00:23:25.651 --> 00:23:30.141
Roger Lewis: So the, the FDA was all of
us trying to address this issue of the
00:23:30.141 --> 00:23:35.151
shortage of platelets, and they were
well aware that the, uh, limited storage
00:23:35.151 --> 00:23:39.041
duration of form-- of room temperature
platelets was a problem in lots of
00:23:39.041 --> 00:23:44.231
settings, um, disaster settings, military
settings, and just low volume hospitals.
00:23:45.001 --> 00:23:53.131
And, um, FDA made a, a decision to
allow, in specific circumstances,
00:23:53.151 --> 00:23:57.051
the administration of cold stored
platelets stored up to fourteen
00:23:57.061 --> 00:24:00.231
days, um, uh, upon review.
00:24:00.261 --> 00:24:03.301
My understanding, I'm certainly not
an expert in this regulatory area,
00:24:03.581 --> 00:24:07.731
was that the FDA made the decision
that blood banks could apply to them
00:24:08.081 --> 00:24:11.941
to get a variance to provide cold
stored platelets up to fourteen days.
00:24:12.421 --> 00:24:19.441
So the trial was, um, following the
pre-specified adaptive design, then
00:24:19.451 --> 00:24:24.631
suddenly, um, the FDA essentially said
up to fourteen was definitely okay.
00:24:25.551 --> 00:24:29.991
Uh, this was, um, this external
information was certainly not informed
00:24:30.061 --> 00:24:33.541
by any knowledge of what was going
on within the CHIPS trial, so it's
00:24:33.551 --> 00:24:35.511
completely external information.
00:24:36.091 --> 00:24:37.561
But it did a couple of things.
00:24:37.581 --> 00:24:42.311
First of all, it established a
new standard of, of care, it made
00:24:42.311 --> 00:24:45.741
the, the sort of clinically and
scientifically important question,
00:24:46.161 --> 00:24:48.481
what about longer than fourteen days?
00:24:49.161 --> 00:24:55.111
also would have made it difficult to
restrict the storage to less than fourteen
00:24:55.111 --> 00:25:00.481
days in the cold stored arm because we
wouldn't have been doing, um, having as
00:25:00.481 --> 00:25:02.661
much flexibility as the FDA was allowing.
00:25:04.171 --> 00:25:10.461
that external information was reviewed
by the, um, the steering committee or
00:25:10.461 --> 00:25:15.021
executive committee for the trial, uh,
the principal investigators, um, Phil
00:25:15.021 --> 00:25:19.471
Spinella at, at Pittsburgh, who managed
to hold this whole thing together.
00:25:19.951 --> 00:25:25.011
Um, and without knowledge of the
interim data, the decision, um, was
00:25:25.011 --> 00:25:31.521
made in conjunction with the DSMB to
immediately move the maximum duration in
00:25:31.521 --> 00:25:38.671
the cold stored arm up to the fourteen
days to meet the FDA's, um, new rule.
00:25:39.111 --> 00:25:46.731
And what that did is it allowed us to
restart the algorithm with fourteen as
00:25:46.731 --> 00:25:51.341
opposed to, uh, the storage duration that
had been allowed up to that point, um,
00:25:51.371 --> 00:25:53.031
which was somewhat less than fourteen
00:25:54.927 --> 00:25:58.667
Scott: Did this happen w- um, while
enrollment was going on, right?
00:25:58.667 --> 00:26:00.167
This was not before.
00:26:00.243 --> 00:26:02.263
Roger Lewis: It, it happened
while enrollment was going on
00:26:02.607 --> 00:26:03.997
Scott: So I maybe I…
00:26:04.037 --> 00:26:05.907
Anna, do you remember this happening?
00:26:05.907 --> 00:26:07.937
And where, where was the limit?
00:26:07.937 --> 00:26:10.057
Was it 10 when this triggered?
00:26:11.072 --> 00:26:14.652
Anna McGlothlin: So this happened
after our first interim analysis.
00:26:14.842 --> 00:26:19.362
So, you know, when the trial started, the
maximum storage duration was seven days.
00:26:20.192 --> 00:26:23.922
After the first interim
analysis, the model indicated
00:26:23.922 --> 00:26:26.002
that we could go up to 12 days.
00:26:26.492 --> 00:26:29.982
So that's what was currently
happening at the time that the
00:26:29.982 --> 00:26:32.102
guidance came out and said 14 days.
00:26:33.042 --> 00:26:39.542
so at the time, um, once we received
that decision from the study team that
00:26:39.572 --> 00:26:43.132
they were going to, you know, essentially
approve going up to 14 days, that's
00:26:43.132 --> 00:26:45.082
what was then implemented going forward.
00:26:45.092 --> 00:26:51.692
So we had kind of this blip that we
went from 12 to 14, between interims
00:26:54.607 --> 00:26:54.847
Scott: Okay.
00:26:54.847 --> 00:26:57.667
And, and Nick, you're
unblinded during this trial.
00:26:57.667 --> 00:27:00.197
Even though you were involved in
the design, you were unblinded
00:27:00.197 --> 00:27:02.287
and helped make the, the model go?
00:27:04.434 --> 00:27:04.714
Nick Berry: Yep.
00:27:04.734 --> 00:27:06.944
Did the design and transitioned
over to implementation
00:27:07.321 --> 00:27:07.951
Scott: Yep, yep.
00:27:07.991 --> 00:27:12.371
And I think Kurt Vielhaber jumped in
as a blinded statistician to advise.
00:27:12.701 --> 00:27:17.261
Uh, and I think he was the last
blinded human on Earth of not
00:27:17.361 --> 00:27:18.831
knowing the results of the trial.
00:27:19.191 --> 00:27:21.541
Uh, he and I, I think, were
both the last ones blinded.
00:27:21.581 --> 00:27:24.151
Okay, so what happens in the trial then?
00:27:24.161 --> 00:27:27.861
During the course of these
adaptations, what are the data saying?
00:27:27.871 --> 00:27:29.371
Are, are you being restricted?
00:27:29.371 --> 00:27:30.481
Are you escalating?
00:27:30.761 --> 00:27:34.411
Generally, the- we'll get to the final
results in the trial, but generally,
00:27:34.411 --> 00:27:35.751
what's happening at the interims?
00:27:38.748 --> 00:27:39.778
Nick Berry: Um, yeah.
00:27:40.768 --> 00:27:44.498
Like Anna said, the at fir-- so I
don't think we explicitly mentioned
00:27:44.498 --> 00:27:49.508
this, but the, the cap was that
escalation could only happen five, uh,
00:27:49.508 --> 00:27:51.308
by a five-day increase at an interim.
00:27:51.308 --> 00:27:53.728
So even if the model recommended
more, it was capped at five.
00:27:53.838 --> 00:27:58.598
So at the first interim, up to-- the
first interim analysis up to seven
00:27:58.598 --> 00:28:05.788
days, uh, cold storage duration, the
responses on the cold platelets arm
00:28:05.848 --> 00:28:09.638
looked very similar to the mean on
the room temperature platelets arm.
00:28:09.698 --> 00:28:12.738
So the, the-- They were very
similar, and the model said,
00:28:12.748 --> 00:28:14.068
"I'm extrapolating forward.
00:28:14.068 --> 00:28:18.578
I think up to some, you know,
larger number, they're going
00:28:18.578 --> 00:28:20.558
to be, uh, non-inferior.
00:28:20.578 --> 00:28:24.328
Like I can't-- I don't see any reason that
they would be inferior at any duration
00:28:24.328 --> 00:28:26.098
close to where I've collected data.
00:28:26.258 --> 00:28:31.258
You should escalate," and we escalated
by five, the maximum allowable, uh, step.
00:28:32.658 --> 00:28:35.928
Then we had this fun jump from
twelve to fourteen in between
00:28:35.928 --> 00:28:36.998
the first and second interim.
00:28:37.178 --> 00:28:41.788
Um, and so we're, we're at fourteen
max storage duration, and the next
00:28:41.788 --> 00:28:50.118
two interims we again collected data
that wasn't pushing the cold stored
00:28:50.118 --> 00:28:56.158
platelet arm's, uh, duration response
model into inferior territory.
00:28:56.158 --> 00:29:01.498
It was, it was very similar to what
we were seeing, um, on the warm
00:29:01.498 --> 00:29:06.218
stored platelet arm, and so at the,
uh, second and third interim, we
00:29:06.248 --> 00:29:08.548
increased by the maximum allowable jump.
00:29:08.548 --> 00:29:11.928
We increased by five, then we
increased up to twenty-one days.
00:29:12.288 --> 00:29:15.138
Um, we went to, I guess, twelve
to seventeen and seventeen to
00:29:15.138 --> 00:29:16.678
twenty-one, which was the max.
00:29:16.708 --> 00:29:18.758
We were, you know, that was
the max we were gonna explore.
00:29:19.228 --> 00:29:20.798
Um, so when we got to twenty-one
00:29:21.880 --> 00:29:23.020
Anna McGlothlin: I was just
gonna jump in, Nick, because
00:29:23.020 --> 00:29:25.020
we went from 14, so we didn't
00:29:25.182 --> 00:29:26.062
Nick Berry: Well, 14 to ni-
00:29:26.100 --> 00:29:26.320
Anna McGlothlin: Yeah,
00:29:26.322 --> 00:29:26.632
Nick Berry: yep.
00:29:26.632 --> 00:29:27.642
14 to
00:29:27.730 --> 00:29:27.740
Anna McGlothlin: go
00:29:27.952 --> 00:29:30.632
Nick Berry: 19 to 21,
and then we stayed at 21.
00:29:31.122 --> 00:29:36.192
Um, and so we increased by the
most that the trial's pre-specified
00:29:36.192 --> 00:29:39.962
rules were allowed to increase
by at every interim analysis
00:29:41.327 --> 00:29:45.127
Scott: So Anna, you get to be
involved a lot of times on this team
00:29:45.127 --> 00:29:47.187
that is unblinded to the results.
00:29:47.437 --> 00:29:49.317
You're talking to the DSMB.
00:29:49.327 --> 00:29:52.557
Now you've got a complex
Bayesian model that's driving
00:29:52.557 --> 00:29:54.477
decisions, patient safety there.
00:29:55.317 --> 00:30:00.887
try to design these so that they're,
the model is providing good results
00:30:00.887 --> 00:30:04.707
and the triggers make sense, and
it's, it's, but it's run by algorithm.
00:30:05.047 --> 00:30:11.207
Was, how, how was the DSMB receptive
to this modeling and the decisions
00:30:11.517 --> 00:30:13.247
during the course of this trial?
00:30:16.288 --> 00:30:16.538
Anna McGlothlin: Yeah.
00:30:16.538 --> 00:30:20.528
So, um, I, I think this
all went very smoothly.
00:30:20.598 --> 00:30:25.598
I think the DSMB was, um, you know,
on board with the design of the trial.
00:30:25.998 --> 00:30:29.958
Um, so each interim analysis, you know,
they were able to convene very quickly.
00:30:30.208 --> 00:30:33.388
Um, it's one of the, the challenges
sometimes within adaptive design is
00:30:33.408 --> 00:30:37.828
getting everybody together, um, to
review the data and, and that, uh,
00:30:37.898 --> 00:30:39.538
was able to be done very quickly here.
00:30:40.908 --> 00:30:44.788
you know, we had a report that we would
send them after every analysis that
00:30:44.788 --> 00:30:49.858
would summarize the data, that would
summarize the model and the adaptive
00:30:49.908 --> 00:30:51.838
decision that was pre-specified.
00:30:52.998 --> 00:30:57.398
um, you know, Nick had this really
nice dashboard that he put together
00:30:57.638 --> 00:31:01.448
that would be in the front of every
report that gave kind of the, the
00:31:01.498 --> 00:31:06.458
overall, um, you know, 10,000 foot view
o-of what was happening in the trial.
00:31:06.458 --> 00:31:11.038
And that became of the, the way that we
communicated the results to the DSMB.
00:31:11.368 --> 00:31:13.968
Um, it was really this one page
with, with several figures on it
00:31:13.968 --> 00:31:16.918
that they could look at and get all
of the information in, in one place.
00:31:17.298 --> 00:31:20.458
And then if they needed additional
details, those were in the report.
00:31:20.768 --> 00:31:24.958
Um, but really focusing on here's,
here's the key information that
00:31:24.958 --> 00:31:28.748
they needed to, to make their
recommendations back to the study team.
00:31:31.477 --> 00:31:35.927
Scott: Okay, so the trial goes to the
maximum sample size, 1,000 patients, and
00:31:35.937 --> 00:31:43.317
I think this was two to one randomized,
so, uh, uh, at the end it was 660 patients
00:31:43.317 --> 00:31:48.357
on cold platelets, 340 patients on
room temperature, um, uh, within this.
00:31:49.347 --> 00:31:51.347
what's the final result of the trial?
00:31:51.367 --> 00:31:53.967
What's the primary analysis
of the trial, Roger?
00:31:57.009 --> 00:32:02.469
Roger Lewis: Well, the primary result is
that the, um, the model demonstrates, uh,
00:32:02.469 --> 00:32:07.559
non-inferiority at all, all chosen, uh,
storage durations up to twenty-one days.
00:32:08.299 --> 00:32:15.109
in fact, the, the raw data looks very flat
with respect to the, the efficacy score.
00:32:15.429 --> 00:32:19.899
So the model fit actually looks a
little more pessimistic to the data
00:32:20.279 --> 00:32:25.469
because we in-- uh, incorporated
the prior information that, that
00:32:25.479 --> 00:32:29.359
platelets were not wine, they could
not get better with age and storage.
00:32:30.319 --> 00:32:34.769
Um, and the model appropriately
incorporates that, that information
00:32:34.769 --> 00:32:37.989
or, or bias, if you will, if
you will, into its estimates.
00:32:38.259 --> 00:32:44.129
So the model strongly, um, uh, supports
the use of them again, um, if you
00:32:44.159 --> 00:32:45.729
agree with that non-inferiority margin.
00:32:45.999 --> 00:32:48.849
Now, in fact, the data would
have been positive with a much
00:32:48.859 --> 00:32:51.119
smaller non-inferiority margin.
00:32:51.809 --> 00:32:54.899
There has been some discussion about
whether that non-inferiority margin is
00:32:54.899 --> 00:33:00.149
generous, and I think it, it was generous,
but it reflected the fact that in reality
00:33:00.149 --> 00:33:05.019
we were d- testing a product that would
be used in places where there was no
00:33:05.019 --> 00:33:10.259
alternative as opposed to room temperature
platelets, um, as, as the alternative.
00:33:10.969 --> 00:33:11.449
you mentioned
00:33:11.495 --> 00:33:14.014
the 2:1 ratio.
00:33:14.074 --> 00:33:14.334
Another
00:33:14.349 --> 00:33:19.369
complexity of this trial is that
platelets are not a single product.
00:33:19.859 --> 00:33:23.769
They are, um, collected using what
are called different platforms.
00:33:23.799 --> 00:33:28.579
They're sometimes, um, put into
different, um, solutions or, or,
00:33:28.579 --> 00:33:30.569
um, to suspend the platelets.
00:33:30.739 --> 00:33:34.099
They may be treated in various
ways, such as pathogen reduction.
00:33:34.309 --> 00:33:37.639
So there's lots of different
flavors and subtypes of platelets.
00:33:38.079 --> 00:33:39.189
So There was definitely
00:33:39.189 --> 00:33:45.749
a desire here to get as much data on
the different subtypes or flavors of
00:33:45.759 --> 00:33:50.099
platelets, if you will, so that, um,
at least descriptively, people could
00:33:50.099 --> 00:33:55.509
evaluate whether the, um, the success
of the cold-stored platelets looked,
00:33:55.519 --> 00:34:00.819
you know, uniform, uh, to the extent we
can tell, um, over the different types
00:34:02.703 --> 00:34:07.783
Scott: So the, the final Bayesian
probability was greater than 99.9%
00:34:07.793 --> 00:34:08.963
of non-inferiority.
00:34:08.963 --> 00:34:14.843
I think the top of the 95% confidence
interval for 21 days was .23,
00:34:15.253 --> 00:34:19.233
and the non-inferiority margin
was one, so, so well below that.
00:34:19.453 --> 00:34:22.913
And as you described, it was almost like
the model was slightly pessimistic because
00:34:22.913 --> 00:34:25.623
it had to be monotonic, uh, very flat.
00:34:25.953 --> 00:34:29.593
Now how about other parts to
this, um, uh, this endpoint?
00:34:29.603 --> 00:34:35.653
You talked about the, the, uh,
potential of infections with too
00:34:35.653 --> 00:34:41.033
long, or part of the primary analysis
was this chest tube measurements.
00:34:41.293 --> 00:34:49.553
Did everything else line up, that cold
platelets was similar to room temperature?
00:34:50.931 --> 00:34:53.131
Roger Lewis: So I think if you
use the term similar, I think
00:34:53.131 --> 00:34:56.121
everything did line up, but there's
definitely some complexity to it.
00:34:56.491 --> 00:35:01.231
The, uh, first of all, twenty-four
hour chest tube drainage is a
00:35:01.231 --> 00:35:06.021
commonly used regulatory endpoint
for, for hemostatic therapies.
00:35:06.321 --> 00:35:11.561
And so that was a secondary endpoint,
um, pre-specified secondary endpoint.
00:35:11.661 --> 00:35:16.491
Um, and the chest tube drainages were
similar, although there was numerically
00:35:16.491 --> 00:35:20.461
a slight increase in chest tube
drainage in the room-- in the, uh, cold
00:35:20.461 --> 00:35:22.431
stored platelet arm as, as I recall.
00:35:23.341 --> 00:35:29.731
But, um, there was also a finding that,
um, there was some additional use of
00:35:29.731 --> 00:35:36.061
blood products, plasma, red cells, um,
in the patients in the cold stored arm.
00:35:36.061 --> 00:35:40.381
So even though the amount of bleeding
clinically looked the same, there was a
00:35:40.381 --> 00:35:45.621
little more use of other blood products
in the, in the active or cold stored arm.
00:35:46.561 --> 00:35:48.021
there's a couple of possibilities.
00:35:48.021 --> 00:35:50.741
Obviously, one possibility is
that the cold stored platelets
00:35:50.741 --> 00:35:52.701
don't work quite as well.
00:35:53.161 --> 00:35:58.011
Um, and although they were, you know,
clearly met the non-inferiority criteria,
00:35:58.011 --> 00:36:00.361
they, they were a little less e-effective.
00:36:01.151 --> 00:36:05.251
Another possibility has to do with
the pragmatic nature of the trial.
00:36:06.111 --> 00:36:10.871
the trial looked at the, um, total
amount of blood products through
00:36:10.871 --> 00:36:12.291
that first twenty-four hours.
00:36:12.291 --> 00:36:16.541
And during that twenty-four hours,
people measure the platelet count,
00:36:16.551 --> 00:36:19.761
the number of platelets through
cubic millimeter in the blood.
00:36:20.151 --> 00:36:24.991
And sometimes our decision to transfuse
platelets or to transfuse other blood
00:36:24.991 --> 00:36:29.491
products are influenced a little
bit by just that numerical count.
00:36:30.381 --> 00:36:36.911
is reason to believe, based on work done
decades ago, and verified within this
00:36:36.921 --> 00:36:42.341
trial, that when you give the same number
of cold stored platelets versus the same
00:36:42.341 --> 00:36:46.331
number of room temperature platelets,
the platelet count doesn't go up as
00:36:46.331 --> 00:36:48.201
much with the cold stored platelets.
00:36:48.701 --> 00:36:53.311
And that probably, um, reflects not
that they aren't working as well,
00:36:54.031 --> 00:36:57.541
that they're more activated and they
actually do their job faster and then
00:36:57.541 --> 00:36:58.981
are no longer in the bloodstream.
00:36:58.991 --> 00:37:01.091
So it may actually be
that they're, they're
00:37:01.283 --> 00:37:01.383
Scott: There is no work
00:37:01.681 --> 00:37:03.941
Roger Lewis: faster at
the site of the bleeding.
00:37:04.291 --> 00:37:07.631
But I think there's the possibility
with the pragmatic design and the
00:37:07.641 --> 00:37:11.171
fact that, that those platelet counts
were available to clinicians, that
00:37:11.171 --> 00:37:15.331
their decision to transfuse was
at least partially influenced by
00:37:15.331 --> 00:37:16.951
knowledge of these platelet counts.
00:37:17.141 --> 00:37:20.811
Because traditionally, we think of low
platelet counts as being associated
00:37:20.811 --> 00:37:24.201
with more risk of bleeding, so we're
more aggressive about transfusing
00:37:24.971 --> 00:37:29.181
Scott: I, by the way, we should mention
this was funded by the US Department of
00:37:29.181 --> 00:37:32.191
Defense, uh, who presumably interested.
00:37:32.191 --> 00:37:36.781
So coming to the question of the
implications of this result, Roger,
00:37:36.971 --> 00:37:42.301
what are the, the, the implications
of the result of this paper?
00:37:44.627 --> 00:37:47.777
Roger Lewis: So I think there's both
medical implications, and I think there's
00:37:47.777 --> 00:37:51.117
actually some research implications.
00:37:51.127 --> 00:37:56.677
So the medical implications, um, that
cold stored platelets are a viable
00:37:56.687 --> 00:38:03.237
opport-- um, option for the treatment
of active bleeding in patients, um,
00:38:04.207 --> 00:38:08.097
probably in general, but certainly in
settings in which, uh, an inventory
00:38:08.097 --> 00:38:10.597
of room temperature platelets
would be difficult to maintain.
00:38:10.967 --> 00:38:15.937
So as I mentioned, rural hospitals, low
volume hospitals, um, disaster settings,
00:38:15.957 --> 00:38:18.337
military settings, um, and, and the like.
00:38:19.577 --> 00:38:25.427
I think the other thing about this, um,
trial that's important, and this is where
00:38:25.427 --> 00:38:33.107
the, the design is critical, if you think
about it, we often test products by just
00:38:33.117 --> 00:38:36.487
throwing them all together, regardless
of how long they've been stored.
00:38:36.487 --> 00:38:42.647
So for a, a blood component, or any
other product that has a, a shelf life,
00:38:43.467 --> 00:38:51.147
we simply compare some mixture of the,
of ages some other standard of care,
00:38:51.147 --> 00:38:53.127
and we get a single yes, no answer.
00:38:53.277 --> 00:38:56.967
And we're really left relatively
naive to whether there is an effect
00:38:56.967 --> 00:39:00.507
of, of shelf life storage duration,
whatever it is, on efficacy.
00:39:01.427 --> 00:39:08.147
is a trial in which the design
specifically, um, structured and
00:39:08.147 --> 00:39:14.517
then extensively evaluated on the
accuracy of getting the duration right.
00:39:15.427 --> 00:39:21.557
And as we all know, um, in dose finding
studies, many commonly used dose
00:39:21.557 --> 00:39:25.247
finding, um, designs actually aren't
very good at getting the dose right.
00:39:25.247 --> 00:39:29.587
They have error rates of, you know,
twenty, thirty percent or, or, or higher.
00:39:29.977 --> 00:39:35.257
Here we objectively evaluated how good
we were at getting the storage duration
00:39:35.257 --> 00:39:39.257
right and making sure when we got it
wrong, we did so in a conservative
00:39:39.257 --> 00:39:41.087
way to minimize risk to patients.
00:39:41.377 --> 00:39:48.977
So I think this design, um, really
represents a model for how we ought to
00:39:49.017 --> 00:39:53.487
evaluate therapies, both to demonstrate
their efficacy and to figure out how
00:39:53.487 --> 00:39:56.967
to get the most value out of them
by being able to use them as long as
00:39:56.967 --> 00:39:59.117
they're, um, the right thing to use
00:40:00.245 --> 00:40:03.135
Scott: Yeah, I mean, you could
imagine this, this result comes
00:40:03.135 --> 00:40:07.875
out in three separate trials where
you do 10 to 14 hours-- for 10
00:40:09.375 --> 00:40:11.705
to 14 days, you run the trial.
00:40:11.705 --> 00:40:12.315
That's great.
00:40:12.315 --> 00:40:16.125
And then you run 14 to 17, and
then you run a whole nother trial.
00:40:16.125 --> 00:40:17.595
So yeah, it's brilliant.
00:40:17.655 --> 00:40:18.635
I, I love that.
00:40:19.445 --> 00:40:21.345
Nick, you were simulating this.
00:40:21.345 --> 00:40:23.515
You created a modeling of this.
00:40:23.515 --> 00:40:25.085
You simulated scenarios.
00:40:25.085 --> 00:40:28.425
You certainly simulated scenarios
where it was really flat, and
00:40:28.425 --> 00:40:29.985
that's what it turned out to be.
00:40:30.845 --> 00:40:35.015
Uh, whenever we design trials and it
runs, maybe you have some regrets,
00:40:35.055 --> 00:40:36.605
could've done this, could've done that.
00:40:36.605 --> 00:40:38.235
Any of that in this trial?
00:40:39.914 --> 00:40:46.344
Nick Berry: I, I mean, it's terrifying
when a trial runs and you're fitting
00:40:46.554 --> 00:40:51.464
a, a model, a complicated model
that, you know, you designed and,
00:40:51.464 --> 00:40:57.914
and created, and especially when
that model has constraints on it that
00:40:58.254 --> 00:41:01.234
make it inflexible by definition.
00:41:01.234 --> 00:41:06.514
Because if platelets had looked
better by some random chance or
00:41:06.514 --> 00:41:10.254
something, the model that we fit was
not going to look like the raw data.
00:41:10.524 --> 00:41:14.514
And that's an intimidating thing for, you
know, me to stand in front of DSMB and
00:41:14.514 --> 00:41:17.294
say, "Here's the data, here's the model.
00:41:17.884 --> 00:41:21.974
I know they don't look the same, but
well, plate- platelets aren't wine."
00:41:22.034 --> 00:41:22.924
Is that compelling?
00:41:22.924 --> 00:41:24.264
Like, do you, do you
believe that, you know?
00:41:24.584 --> 00:41:25.404
And so you have to…
00:41:25.484 --> 00:41:26.634
A- and I'm, I'm fine.
00:41:26.634 --> 00:41:27.894
I l- I don't mind.
00:41:28.004 --> 00:41:29.654
I believed in the model, of course.
00:41:29.664 --> 00:41:34.834
You know, we fit it 10 million times
probably to different data sets
00:41:34.834 --> 00:41:36.254
throughout the course of developing this.
00:41:36.254 --> 00:41:40.494
So, um, you know, it, we had, we
had pressure tested it, but it's
00:41:40.494 --> 00:41:44.344
still terrifying to, to actually
look at the data and, and see it.
00:41:45.814 --> 00:41:46.984
Um, let's see.
00:41:49.244 --> 00:41:50.684
I, I had just started.
00:41:50.824 --> 00:41:55.464
You know, I, I, I had just joined Berry
Consultants and, uh, you know, I, I
00:41:55.464 --> 00:41:59.174
was, I was guided pretty heavily with a
heavy hand on, on where this should go.
00:41:59.184 --> 00:42:03.964
But, um, you know, it was a big
adventure in operating characteristics
00:42:03.964 --> 00:42:07.064
and simulating these nulls,
and this is a realistic null.
00:42:07.074 --> 00:42:08.664
This one's pathological.
00:42:08.664 --> 00:42:13.904
And, um, so I, I don't think there's
regrets, but that's easy to say when
00:42:14.224 --> 00:42:15.484
things look like they did, right?
00:42:15.534 --> 00:42:16.704
Um, the, the…
00:42:17.394 --> 00:42:20.814
Everything worked out great, so the
model was perfect for the, for the job
00:42:21.341 --> 00:42:21.601
Scott: Yeah.
00:42:22.061 --> 00:42:22.301
Yeah.
00:42:22.691 --> 00:42:26.931
And in terms of implementation, Anna,
it sounds like this ran very smoothly.
00:42:27.311 --> 00:42:28.421
Uh, all good?
00:42:28.421 --> 00:42:32.341
Any, any, uh, things you'd
change in this or lessons?
00:42:33.504 --> 00:42:36.374
Anna McGlothlin: Yeah, I think this is
a really nice success story that we can
00:42:36.374 --> 00:42:39.294
point to that trials like this can be run.
00:42:39.384 --> 00:42:42.664
You know, this was a fairly
complicated design with the
00:42:42.664 --> 00:42:45.794
escalation of the different storage
durations, with the modeling,
00:42:46.214 --> 00:42:48.534
and it, it all ran very smoothly.
00:42:48.564 --> 00:42:54.054
So I think with the right preparation and,
and, um, you know, setting things, uh,
00:42:54.114 --> 00:42:57.754
in place for the communication channels
and all of that, this can absolutely be
00:42:57.754 --> 00:42:59.764
done, even a complicated design like this
00:43:01.113 --> 00:43:01.133
Scott: Yeah.
00:43:01.583 --> 00:43:05.873
And, and I reiterate something you
said that we're told very frequently
00:43:05.873 --> 00:43:10.153
when we wanna do an adaptive design
that each interim is six weeks,
00:43:10.153 --> 00:43:11.723
and this is gonna be problematic.
00:43:11.723 --> 00:43:14.953
That, you know, data was delivered
in a day, and the results were
00:43:14.953 --> 00:43:18.643
delivered in five business
days, and the trial moved on.
00:43:18.703 --> 00:43:21.033
Uh, fantastic, fantastic effort.
00:43:21.343 --> 00:43:26.293
Uh, Roger, uh, any last words
on the trial, next steps?
00:43:27.587 --> 00:43:29.927
Roger Lewis: Well, I just think
we, we need to acknowledge
00:43:29.957 --> 00:43:31.777
the, the partners that we had.
00:43:32.267 --> 00:43:37.957
Anna mentioned the, the Data Coordinating
Center at Utah, the investigator team, um,
00:43:38.137 --> 00:43:43.317
which was, uh, you know, b-based at many
institutions but that led by Phil Spinella
00:43:43.457 --> 00:43:48.207
and his colleagues, and the folks behind
the scene that made this all happen.
00:43:49.227 --> 00:43:55.527
the, the, the work to get the cold
stored platelet inventory, get it
00:43:55.527 --> 00:43:59.677
shipped, make sure there's enough of it
at each of the sites so that patients
00:43:59.677 --> 00:44:05.617
can get almost always the, the, the
arm to which they're randomized, was
00:44:05.617 --> 00:44:08.337
really a logistical tou-tour de force.
00:44:08.797 --> 00:44:14.247
And so I think the, the combination of
the folks on the ground were completely
00:44:14.267 --> 00:44:19.327
committed to the success, people willing
to do things differently for the data
00:44:19.337 --> 00:44:24.727
availability, as you mentioned, and then
the presentation from the implementation
00:44:24.737 --> 00:44:30.447
team that gave us just snapshot that
you could look at and just immediately
00:44:30.447 --> 00:44:33.747
know what was going on with the trial
and what the right thing to do was,
00:44:34.017 --> 00:44:36.687
is all the pieces sort of, um…
00:44:37.107 --> 00:44:41.597
Each one of the pieces was executed
so well, and it answered the question.
00:44:41.597 --> 00:44:45.107
My only regret is that we
didn't, um, build the design to
00:44:45.147 --> 00:44:46.817
randomize up to twenty-eight days.
00:44:47.297 --> 00:44:52.647
I sort of feel like we were, we, we
were on a roll and, and it wouldn't
00:44:52.647 --> 00:44:55.857
have been hard to get an extra week
and see if there was actually any
00:44:55.857 --> 00:44:58.497
upturn, um, towards inferiority
00:45:00.453 --> 00:45:00.553
Scott: Right.
00:45:00.553 --> 00:45:00.983
Nice.
00:45:00.983 --> 00:45:04.023
Well, we'll, we'll move that
to, to next step certainly.
00:45:04.023 --> 00:45:07.623
And, and number of really cool
questions within trauma, within
00:45:07.623 --> 00:45:10.343
bleeding, uh, exciting trials for that.
00:45:10.753 --> 00:45:14.103
So thank you Roger,
Anna, Nick, for joining.
00:45:14.423 --> 00:45:20.013
Fantastic, fantastic trial, fantastic
result, fantastic clinical impact
00:45:22.285 --> 00:45:24.655
And appreciate everybody for listening.
00:45:24.765 --> 00:45:27.815
Uh, until next time, we'll
be here in the interim