ICTR Short Video #8 Methods for using multiple treatments within the same subject


Hello. I’m Melanie Quintana, a
senior statistical scientist at Berry Consultants. At Berry Consultants,
we design a wide range of clinical trials,
most of which involve an individual
subject being randomized to a single treatment
or placebo and being followed over a period of time. However, there are
situations in clinical trials where we want to allow
individual subjects to receive multiple treatments. And that’s what we’re
going to talk about today in this short video– why we do this, when we do this,
and some key considerations that we want to think
about when we’re doing it. In particular, I’m going to
talk about two types of designs. The first is the
one you probably think about most when you
think about multiple treatments within the same subject. And that’s a randomized
controlled crossover design. The second is often
called treatment switching to differentiate it from a
randomized controlled crossover design. And it’s when an
individual is allowed to switch, or crossover, based
on a period of progression or non-response. But let’s start with randomized
controlled crossover, a very simple what’s
often called AB/BA design. In these types of designs,
each individual patient receives each treatment
that’s being studied. And the order in which they
receive those treatments is normally randomized. Why we do this is so that
every participant can serve as their own control, we
can reduce variability, and we can increase power. Within these types
of studies, a period in which you receive the
treatment is called a period. And each period tends to be
prespecified in length, so, say, six months
for each treatment. And those periods tend to be
the same for each treatment. There could be two
periods, or 20, 30, up to 100 periods of
different types of treatments given to a single individual. Between periods, we tend to have
a washout period so that we can mitigate any carryover effects. These are effects of
the first treatment that was given carrying
over into the second period. Some key considerations
that we want to think about in the randomized
controlled crossover design are, in fact, this
carryover effect. We want to make sure that
the washout period itself is substantial enough
to limit any effect that that treatment might
have in the second period. Another thing that we want
to think critically about is, is there a period
or an order effect? So if you give a
treatment in period one, would we expect the effect
of that to be different if it was given in period two? Does it matter the order that
the treatments were given? Oftentimes what we do to
mitigate issues with this is we want to ensure that
we have a disease that’s chronic and stable so
that the disease itself is constant over the entire trial. The final thing that
we want to think about is the timing of
the outcome itself. Oftentimes in a randomized
controlled crossover study, we’re thinking about short-term
outcomes that are completely contained to each period. So, for instance,
change from baseline over the six-month period. Under these ideal situations, so
no carryover effect, no period effect, a nice short
outcome that we can observe at each period, we
can see a substantial increase in power between
a crossover design and, say, a parallel design. Here in this plot, I’m
showing the sample size needed for 80% power on the y-axis. And on the x-axis, I have
the treatment effect itself that we’re assuming. Here, these are standardized
so that a 0.5 means 0.5 standardized unit different
between treatment and control. If you look at the 0.5
standardized unit difference, you can see that we
need a little bit less than 50 subjects for
a crossover design compared to over 125 subjects
for a parallel design. So crossover designs
in this ideal situation allow us to get away
with less than half of the same participants. Now, there can be some
negatives to a crossover design. One is that every patient
receives every treatment. So that means if we’re thinking
about a placebo controlled trial, that every
patient will be receiving a placebo for
some period of time. And this may really not be ideal
to subjects entering the trial and may limit the subjects
who want to enter. Another issue is that
these types of trials could take longer in terms of
the duration of the full trial, because each treatment
period is sequential instead of run in parallel. These are done many
times in practice. This is just one example
of a press release from Axsome Therapeutics. This was released in December. They ran a crossover design. They had two two-week
periods, and in between that, a
one-week washout. And this was a design to
look at their efficacy of their drug in narcolepsy. And what they saw was a
statistically significant reduction in cataplexy
events, or attacks, in the treatment period
compared to the placebo period. So now that we’ve talked about
the randomized controlled crossover design, I want to
switch gears a little bit to talk about another type
of design often called treatment switching. And these designs come up
when an individual switches, or crosses over, to another
treatment after a period of non-response or progression. Oftentimes, these are one-way
switching, so a switching from, say, placebo
to active treatment. And these are done to
address ethical issues and increase study participation
so that everybody knows that if there is a
period of non-response or they’re performing
badly on placebo, they will receive
an active therapy. Oftentimes, these are also
done to actually learn what looks best both in
first-line and second-line therapy. So instead of there just
being a one-way switch, where a subject goes from
placebo to active therapy, there could be randomization
post-progression to learn what all possible second-line
therapies work best. Within treatment
switching, you can start to see that things get
a little bit more complicated than the nice randomized
controlled crossover design. For one, the time of the
periods could be very different. And they depend on when
a subject progresses. So for each subject, they
could be in period one, say, for one month, or another
subject may come in and never go into period two– stay in
period one the entire time. Another issue is that
a subject switches based on an intermediate
outcome, not the final outcome that we’re most interested in. So we are not able to
observe the final outcome and have it be contained
within each treatment period. So all of the same
considerations from randomized controlled
crossover studies still hold with
treatment switching. But we also want to account
for when we switched and what treatments we switched
to in the primary analysis, because switching on
this intermediate outcome before we observe the
final outcome may actually dilute the overall
treatment effect that we are able to observe. Imagine if everyone on
placebo switches over to the active treatment
after a month, then those two
groups are probably going to end up
looking pretty similar. So we need to be very
careful and adjust for different effects
of the switching. We need to be able to adjust
for any potential carryover effect from the first-line
treatment to the second-line. We need to be able to
understand if a treatment is different in the first line
versus the second line. Is that effect different? And then we need to
take into consideration the amount of time that each
patient is in each period. Now, treatment switching is
done many times in oncology. This is just one example of
when it’s done in practice. This is the Precision Promise,
Pancreatic Cancer Platform Trial. And this is a trial that’s
in– a perpetual trial that’s interested in looking
at many therapies over the course
of time and which ones help pancreatic cancer. In this trial, we have
adaptive randomization to receive a first-line therapy. And then if an
individual progresses, there’s adaptive
re-randomization to the best performing
second-line therapy. This trial is
meant to understand which therapies work
best in first line, as well as which work
best in second line, and what are the sequence
of therapies that work best. And the statistical
analyses themselves allow differential effects
in the first and second-line therapies. So now that we’ve talked about
two main types of designs with multiple
treatments, I just want to acknowledge that there
are many more designs that involve multiple treatments. One that’s gaining popularity
is a dynamic treatment regimen, or what people are
calling a smart design. And this also involves
re-randomization to subsequent treatments based
on intermediate outcomes. And also, we’re
interested in learning what is the best and most
ideal sequence of therapies. And finally, where
this is coming up is also in perpetual,
or platform, trials. So platform trials
themselves are meant to be these freestanding
arenas where drugs come in and are
tested, and then they come out to see which ones
are the most effective within a certain disease. And what we’re seeing happen in
these platform trials is that after your initial
period of enrollment and serving to test
for a single regimen, a subject may
still after they’re finished with that period be
able to re-enroll in the trial and have a repeat randomization. And we need to take into
consideration this repeat randomization. In summary, I hope
you’ve seen today that there are many valid
reasons for allowing an individual subject to
receive multiple treatments over the course of
a clinical trial– for increase in power,
for ethical reasons, and also to purely understand
what sequence of treatments work best. Across the board
when we’re dealing with a trial of
this nature, we need to make sure we clearly
document and report the multiple treatments
that each subject receives so that the statistical analyses
themselves can carefully consider the impact that
these multiple treatments have on the final endpoint. Thank you all for watching
this short video today. I hope you learned a
little bit something more about multiple treatments
within the same subject. And please go to
our website to watch some more of these videos.

Leave a Reply

Your email address will not be published. Required fields are marked *