57 - Data Analysis in Early Trials (S4E12)

All right, welcome back to the deep dive. Today,
we're taking a look at, well, I'd say a crucial
step in drug development data analysis in those
super early clinical trials, phase one and two,
where a potential new treatment, it gets tested
in humans for the very first time. Exactly. These
early studies, they're all about gathering that
initial information on safety, figuring out,
you know, is this drug safe for people to take?
And do we see any early signs, any hints at all
that, you know, it might actually work? And just
so everyone's on the same page, what we want
to do today, our goal is to really dig into those
fundamental statistical methods, the nuts and
bolts, you know, how researchers interpret the
initial signals, the early data coming out of
these trials. Right, because it's not always
straightforward. We'll also be discussing how
researchers navigate those tricky decision -making
processes based on those initial findings. And
of course, we can't forget about variability,
those inherent differences in how people respond
to a drug. We'll be exploring how that plays
a role in the analysis. And of course, the FDA,
the ICH, all those regulatory guidelines will
come into play. And speaking of sources, we've
got a lot to work with today, you know, actual
drug discovery case studies, the basic principles
of medicinal chemistry, you know, like how drugs
are designed, some overviews of the whole drug
development process. Then we've got stuff on
preclinical testing, clinical trial design, and
then of course, you know, we have those essential
regulatory guidelines. Yeah, real smorgasbord.
And all of these different perspectives, they're
going to give us a pretty comprehensive picture
of the challenges and considerations in analyzing
data from those early human studies. So to kick
things off, let's really break down what data
analysis actually looks like in these initial
phase one trials. What's the main aim? What are
we trying to achieve? Well, in phase one, the
data analysis, it's laser focused on safety and
tolerability. We're talking about meticulously
collecting information on any adverse events,
any side effects the participants experience.
Right. So we're watching very carefully for any
potential red flags. Exactly. But at the same
time, we're also gathering those crucial early
insights into pharmacokinetics. Basically, we
want to understand how the drug moves through
the body, how it's absorbed, distributed, metabolized,
and finally eliminated. the whole journey. So
it's like we're creating a map of the drugs adventure
through the body. But phase one trials are usually
pretty small, right? Limited number of participants,
often healthy volunteers. How does that impact
the way we can analyze the data from a statistical
point of view? Yeah, that's a really important
point. With smaller numbers, we have what we
call lower statistical power. That basically
means it's harder to detect those subtle. you
know, maybe less obvious safety signals, simply
because we have fewer data points to work with.
So our interpretations at this stage, they have
to be very cautious, you know, taking into account
those limitations. So it's all about nuance,
really, considering the context. Absolutely.
OK, so let's dive a little deeper into the safety
data itself. How are these adverse events actually
collected and analyzed? What are researchers
really looking for in all that information? So
when an adverse event happens, it's documented
very carefully. We're talking about recording
what happened, how severe it was, you know, mild,
moderate, severe, how long it lasted, and whether
it could potentially be related to the drug.
And then we use descriptive statistics, a fancy
way of saying we're looking at things like, you
know, how often different types of adverse events
occur. Right. So, like, what percentage of people
experience a particular side effect? Exactly.
But a really critical goal here is to identify
those dose -limiting toxicities, those DLTs.
These are the serious adverse events that happen
at a specific dose and they basically prevent
us from safely increasing the dose any further.
So those warning signs telling us, hey, this
is where it starts to get risky. Yeah, exactly.
Analyzing when and at what dose these DLTs pop
up, that's fundamental to figuring out what's
called the maximum tolerated dose, the MTD. It's
that delicate balance, right? We want to find
that sweet spot where we're getting the potential
benefit of the drug without pushing people into
dangerous territory. It's all about finding that
fine line between benefit and risk. Now, you
mentioned pharmacokinetic data. Earlier, you
know that whole journey of the drug through the
body. What's the analysis process for that? I
remember hearing about something called plasma
concentration curves Right. So pharmacokinetic
data. It's all about tracking the drugs concentration
in the body over time So imagine, you know, someone
gets a single dose of a drug intravenously like
a rapid injection what we call a bolus dose We
can then measure how much drug is present in
their blood plasma at different time points,
like 30 minutes later, an hour later, two hours
later. So we're basically taking snapshots of
the drug's levels at various points along its
journey. Exactly. And if we plot those measurements
on a graph, we get what's called a plasma concentration
versus time curve. You can think of it as visually
mapping out how the drug travels through the
body. That's like a visual representation of
that journey. Right. And then we use these things
called pharmacokinetic models, these mathematical
equations that help us make sense of those curves.
And from there, we can estimate those key parameters,
like how long the drug stays in different parts
of the body or how quickly it's eliminated. So
we're not just seeing the drug's movement. We're
actually understanding the mechanics of that
journey. Exactly. And in relation to those curves,
you mentioned MEC and MTC. So MEC stands for
minimum effective concentration. And it's basically
the lowest concentration of the drug needed to
actually have a therapeutic effect. Right. So
we need to hit at least that level for the drug
to do its job. Exactly. And MTC, on the other
hand, stands for minimum toxic concentration.
That's the level above which we start to see
those unwanted side effects. So in drug development,
You know, we want to keep that drug concentration
within what we call a therapeutic window. It
needs to be of the MEC, so it's effective, but
below the MTC to avoid toxicity. So like a Goldilocks
zone for drug levels. Not too low, not too high,
just right. Exactly. These pharmacokinetic parameters
are crucial for figuring out that perfect zone.
OK, so we're gathering all this data on safety
and pharmacokinetics in phase one. How does this
information actually guide the decisions about
moving forward with the trial? especially when
it comes to deciding whether to increase the
dose. Yeah, that's a really important part of
the process. So dose escalation in phase one,
it's a very carefully controlled process. We
usually have these predefined rules outlined
in the study protocol, and those rules help us
make those decisions based on the safety data.
For example, if we see too many DLTs at a particular
dose, we know we can't go higher. We might pause
the trial or even reduce the dose to ensure patient
safety. It's like that cautious climb up a ladder.
We take one step at a time checking for stability
before moving higher. Exactly. And there are
some early statistical models that we use to
help with these decisions. We have something
called up and down designs, which guide us in
finding that maximum tolerated dose efficiently,
even with limited data. Another one is the continual
reassessment method or CRM. It's more complex,
but it allows us to incorporate more information
as we go along. So it's like those models. They
act as our guide as we carefully navigate this
dose escalation process. Now you mentioned earlier
about variability in this early human data. How
much of an impact does that have on the analysis
and how do researchers take that into account?
Right, variability in phase one data, it can
be pretty significant. I mean, people, they metabolize
drugs differently. They have different body weights,
different underlying health conditions. So statistical
analyses, they need to consider those individual
differences. We can't just focus on the average
response. We also need to look at the range and
spread of the data to get a more realistic picture.
So it's about appreciating those individual responses,
not just looking at the big group average. Exactly.
Even with a small sample size, we want to understand
how the drug might be. behave across a more diverse
population. Okay, so phase one, we're figuring
out safety, we're understanding how the drug
behaves in the body. Now, let's switch gears
to phase two. What's the main goal of data analysis
at this stage? So in phase two, safety is still
paramount, of course, but the focus really expands
to include efficacy. We want to know, does the
drug actually work? Does it show signs of having
a beneficial effect in patients who have the
specific condition we're targeting? So it's like
now we're really starting to ask, can this drug
deliver on its promise? Exactly. And this is
where analyzing what we call efficacy endpoints,
those measures of how well the treatment works,
becomes really central to the whole data analysis
process. And phase two usually involves a larger
group of participants than phase one, right?
What kind of impact does that have on the analysis?
Yeah, phase two trials typically involve a bigger
group. which gives us more statistical power.
We basically have a better chance of detecting
a real treatment effect, you know, if the drug
is actually doing something. And it also allows
for more robust, more reliable analyses of those
efficacy endpoints. So a bigger sample size helps
us see those effects more clearly, even if they're
subtle. Precisely. Okay, let's talk about those
efficacy endpoints. What kind of data are we
actually collecting and how is it analyzed? Well...
The types of endpoints, they can vary quite a
bit depending on the condition being studied.
So for instance, in cancer trials, we might look
at tumor shrinkage, you know, a continuous variable
as the primary endpoint. In hypertension, we
might look at changes in blood pressure. Sometimes
those endpoints are binary, like did the patient
respond to treatment or not? A yes or no type
of outcome. So different diseases, different
ways of measuring how well the treatment is working.
Right. And the analysis methods, they also differ.
For those continuous endpoints, we often compare
the average change in the group getting the drug
versus a control group, you know, folks who might
be getting a placebo or the standard treatment.
And with those binary endpoints, we might compare
the proportion of responders in each group. And
this brings us to the whole idea of a control
group, right? Why is having that comparison group
so critical for analyzing efficacy? Absolutely.
The control group is the key. Without it, we
can't be sure if the changes we're seeing are
because of the drug or because of something else
entirely like, you know, the disease just naturally
getting better or the placebo effect. Randomly
assigning people to either the treatment group
or the control group, that helps us minimize
bias and make a fair comparison. So it's like
we're leveling the playing field at the start,
making sure the groups are as similar as possible,
except for whether they get the drug or not.
Exactly. And then we use these things called
inferential statistics, which help us decide
whether those differences between the groups
are likely real or just due to random chance.
Right. So we want to be confident that the drug
is actually making a difference. Exactly. We're
looking for statistical significance, which basically
means the results are unlikely to have happened
by chance alone. Now, I've also read about these
things called response rates, especially in cancer
trials. Can you explain how those are analyzed?
Yeah. So in phase two oncology trials, the response
rate is a big one. It's the percentage of patients
whose tumors shrink by a certain amount after
treatment. And to analyze those response rates,
we calculate the observed response rate in the
group getting the drug and often create a confidence
interval around that estimate. It basically gives
us a range of values where the true response
rate in a larger population likely falls. So
it's like we're trying to understand how reliable
our estimate is. Right. It gives us a sense of
the precision of those findings. Now phase two
trials, they can run for a while, and I understand
there are times when researchers can actually
peek at the data before the trial is officially
over. These are called interim analyses, right?
Yeah. What's the point of doing that? Yes, exactly.
Interim analyses are like these planned check
-ins along the way. And the main reason we do
them is to look at the data early on and make
informed decisions about the future of the trial.
So for instance, if the early data are really
promising, we're seeing strong signs of efficacy,
we might even stop the trial early. So we don't
want to keep people on a placebo or an inferior
treatment if we already have good evidence that
the new drug is working well. Exactly. On the
flip side, If the early data show that the drug
isn't working at all, we might stop the trial
for futility. We don't want to continue giving
a treatment that's unlikely to benefit anyone.
So those interim analyses, they're like those
critical decision points that can change the
course of the trial. Right. They're important
for both ethical and practical reasons. I'm guessing
those decisions to stop or continue the trial,
they're not just made randomly, right? There
must be some kind of guidelines in place. Oh,
absolutely. We have these predefined statistical
decision rules that are laid out in the trial
protocol before the trial even begins. These
rules clearly define what criteria would trigger
a decision to continue, to stop for futility,
or to stop early for those really promising or
concerning results. Sticking to these rules is
crucial you know, to maintain the integrity of
the trial and to make sure our decisions are
based on solid data. So it's like having a clear
roadmap that keeps everyone on the same page
and helps us sure we're making sound judgments.
Exactly. Now we talked about variability earlier
in the context of phase one. Is that still something
that researchers need to consider when they're
analyzing the data from phase two? Absolutely.
Even though we have a larger sample size in phase
two, you know, people are still different. They
can respond to the same drug in different ways
due to their individual characteristics, you
know, their genetics, their overall health, and
so on. Right. So those individual differences
are still in play. Exactly. And we need to account
for that variability when we're interpreting
the phase two data. Some statistical models actually
include ways to estimate and adjust for those
patient to patient differences. So it's like
we're acknowledging that not everyone is going
to fit neatly into the average. Precisely. We've
discussed the types of data and some of the statistical
concepts. Are there specific statistical models
that are commonly used in these early stage clinical
trials? Yeah, there are some go -to models, although
we usually try to keep things relatively simple
at this stage. So for continuous endpoints, like
measuring a change in blood pressure, we might
use a t -test. That helps us see if the average
change in the treatment group is truly different
from the average change in the control group.
Or we might use something called ANOVA if we
have more than two groups to compare. Right,
so it's like... choosing the right tool for the
job based on the type of data we have. Exactly.
Then, for binary endpoints, those yes or no outcomes,
we often use logistic regression. That helps
us figure out the odds of a particular outcome
happening in different groups, like the odds
of responding to treatment. So we're not just
looking at whether there's a difference, but
also trying to quantify how likely that difference
is. Precisely. And sometimes, our endpoint might
be time to an event, like how long does it take
for a patient's disease to progress? In those
cases, we might use survival analysis techniques.
We can create these Kaplan -Meier curves, which
visually show how the time to event differs between
groups. And we can use Cox proportional hazards
models, which help us statistically compare those
time to event differences, while account for
other factors that might be influencing the results.
So a whole range of models to help us make sense
of different types of data. Right. And the specific
model we choose really depends on the research
question we're asking and the characteristics
of our data. OK. So we run these analyses and
we get our numbers. But how do researchers actually
interpret those results? What does it all mean
in a practical sense? Right. Interpretation.
It goes beyond just looking at whether a result
is statistically significant, whether it's unlikely
to have happened by chance. We also need to consider
what's called clinical significance. So is it
meaningful for actual patients? Exactly. It's
not just about numbers, it's about impact. Even
if a result is statistically significant, the
effect size, meaning how much better the drug
performed compared to the control, that has to
be big enough to actually make a real difference
in people's lives. Right. A small improvement
might be statistically significant, but not really
worth it from a patient's perspective. Exactly.
And we also have to look at the safety data alongside
the efficacy data. A drug that shows a statistically
significant benefit but has really bad side effects,
well, that might not be a good candidate for
further development. So it's like a balancing
act, weighing the potential benefits against
the potential risks. Precisely. And this is where
clinical judgment really comes in. Now, we've
been focusing on the scientific aspects, but
what about the regulatory side of things? Agencies
like the FDA and ICH view all this data analysis
in these early trials. So regulatory agencies,
their primary concern in these early phases is
safety, first and foremost. They want to make
sure that trial participants are protected and
that the risks they're exposed to are justifiable.
So they want to see a very thorough and rigorous
analysis of all the safety data. They have very
specific guidelines that dictate how we need
to collect, analyze and report that data. And
what about in terms of efficacy? Do they have
expectations for phase two data? Yeah, for a
drug to be considered truly effective by the
FDA in those later stages, you know, for it to
get approved, we usually need to show statistically
significant and clinically meaningful results
on those pre -specified primary endpoints that
we outlined in the trial protocol. So it's all
about setting those goals up front and then demonstrating
that the drug can hit those targets. Exactly.
And there are some key guidelines that they refer
to. For instance, there's ICH E6, which provides
comprehensive guidance on good clinical practice,
or GCP, which covers all aspects of clinical
trial conduct, including making sure the data
we collect is high quality and reliable. And
then there's ICH E9, which specifically focuses
on statistical principles for clinical trials.
So it's all about making sure that we're using
sound statistical methods throughout the trial,
even in these early stages. Right, so those guidelines
provide a framework for conducting those analyses
in a rigorous and trustworthy way. Exactly. And
what about the investigational new drug application,
the IND? Remember that being a big step in the
drug development process. How does the data analysis
from those early trials play into that? The IND
is basically the sponsor's request to the FDA
to allow them to move into those later stage
trials. And the data from phase one and two,
that forms a critical part of that application.
The FDA reviews all that data very carefully
to assess the drug safety profile and to decide
if there's enough evidence of potential benefit
to justify further investigation. A strong, well
-analyzed early stage data set that's really
key for getting the green light to move forward.
So it's like passing a crucial checkpoint on
that long road to drug approval. Precisely. Wow,
this has been a really fascinating deep dive.
It's clear that data analysis in those early
stage trials, it's so much more than just crunching
numbers. It's a complex process with a lot of
careful consideration and interpretation involved.
Absolutely. It's about bringing together statistical
rigor, clinical judgment, and a deep understanding
of those regulatory requirements. And it's all
in service of making sure we're developing new
therapies in a way that's safe, ethical, and
ultimately beneficial to patients. So as we wrap
up, here's something for everyone listening to
We've seen how much uncertainty there is in these
early phases. You know, we're working with limited
data trying to piece together those early signals.
So how do researchers and regulators, how do
they balance that need for solid evidence with
the very real urgency of developing treatments
for diseases where, you know, there might not
be any good options right now? It's a challenging
question, and it's one that really shapes the
whole landscape of drug development. Thanks for
joining us.