Latest / EHA Unplugged / Innovative Trial Design (EHA2024)

Transcript

Innovative Trial Design (EHA2024) · EHA Unplugged

• 0:06
  In this episode of EHA Unplugged, we will revisit a
• 0:09
  spotlight talk from EHA 2024. Doctor Yurian Firstlaus from the
• 0:14
  Erasmus Medical Center in Rotterdam discusses the need to
• 0:17
  use innovative clinical trial designs in the era of
• 0:20
  personalized medicine in hematology.
• 0:22
  My name is Yurian Firstlaus. I'm a hematologist at the
• 0:25
  Erasmus MCA in Rotterdam and I'm also highly involved in trial
• 0:30
  design and within the whole form and.
• 0:33
  Related to this topic, do some research so for those of you who
• 0:38
  are. Aware of clinical trial design.
• 0:41
  There are a couple of stages to consider.
• 0:43
  In the clinical trial development and the 1st.
• 0:47
  Part of trial. Development.
• 0:48
  Is the idea. Everything starts with an idea
• 0:50
  and a drug and. Opportunity to investigate.
• 0:55
  An expert do an experiment and usually use a control and when
• 0:59
  there is a design you need to be.
• 1:01
  Aware of the effect size of your.
• 1:04
  Drug and discuss. That with your statistician.
• 1:07
  And as a result of that. There usually.
• 1:09
  Comes a protocol and. Eventually you'll start.
• 1:13
  Recruiting patients and when you recruit patients, there might be
• 1:18
  room for interim analysis. When you.
• 1:20
  Have the time and the equipment to do that and at the end you do
• 1:25
  a final analysis which. Is usually set out at the
• 1:29
  beginning. At the design of a trial and at
• 1:32
  final analysis. You execute everything you.
• 1:34
  You thought of when you designed the trial?
• 1:38
  This is quite a. Fixed and sometimes.
• 1:40
  Rigid process. Where there's no room for
• 1:42
  changes during the trial and these.
• 1:45
  Are actually exactly. Those things that might be
• 1:47
  innovative. Which I will talk about today.
• 1:51
  So why do we need to innovate trial designs?
• 1:54
  We are all aware. Especially in hematology.
• 1:57
  Of the. ERA of personalized medicine
• 2:00
  there are multiple drugs becoming.
• 2:02
  Available specific to. Subtypes of patients who have.
• 2:06
  A target. Mutation, for instance, for whom
• 2:10
  those drugs are available but in the clinical trial.
• 2:13
  It's usually difficult to. Enroll.
• 2:15
  Large subsets of. Patients when there's only a
• 2:18
  few. Patients who have that.
• 2:19
  Actual mutation. So in that era, we need to think
• 2:24
  of other approaches to clinical trial design and also since
• 2:28
  outcomes of our patients are getting so much better, studies
• 2:33
  take a long time to enroll patients and as a result we need
• 2:38
  to wait a long time for the final analysis and studies
• 2:42
  become expensive because of that long follow up time.
• 2:45
  So these are, I believe. Reasons to innovate or think of
• 2:49
  innovation in clinical trial designs and there's also.
• 2:54
  So I talked about the targeted. Treatments which are also
• 2:56
  biomarker driven approaches and on the other hand, when talking
• 2:59
  about time and costs. And we have.
• 3:01
  The early endpoints which is a different.
• 3:05
  Topic I believe. But surrogate endpoints who
• 3:07
  might serve, for instance, for overall survival, which MRD is
• 3:11
  for instance being used, those early endpoints might.
• 3:15
  Actually facilitate a. Sooner end of your trial, but
• 3:20
  that's a different topic to discuss and I'm not going to
• 3:23
  focus on that today. There's also another methods
• 3:26
  which is more important I believe is early intervention.
• 3:30
  For a very efficacious or a futile trial.
• 3:33
  And which? Is getting very.
• 3:35
  Hot in the last. Years is the.
• 3:38
  Use of external control data for your clinical trial, which I
• 3:40
  will. Also address.
• 3:43
  So when we talk about personal, personalized.
• 3:45
  Medicine in clinical trial designs there.
• 3:47
  Are three key types of trial. Design one should consider and
• 3:52
  the first. Is the basket.
• 3:53
  Trial which? Is a type.
• 3:55
  Of trial which? Usually.
• 3:57
  Enrolls multiple diseases where there's one targeted
• 4:00
  intervention. Specific for usually.
• 4:03
  Mutation and that common target is is investigated in such a
• 4:09
  trial but. Such a trial could.
• 4:10
  Include solid oncology. Patients with also hematology
• 4:14
  patients. And it's really a mixed bag of.
• 4:16
  Different types of diseases. On the other hand, we have also
• 4:21
  umbrella trials. Which is usually for a single
• 4:24
  disease, but within. That single disease when there
• 4:27
  are multiple targeted interventions available, those
• 4:30
  targeted interventions are investigated in such a trial and
• 4:34
  in in these. Types of trials.
• 4:35
  There could even be a randomization.
• 4:38
  Both of these trials basket. Trials and umbrella trials.
• 4:42
  Are early phase trials usually? Phase one or phase two clinical
• 4:45
  trials? Mostly phase two clinical trials
• 4:49
  during the COVID-19 pandemic, platform trials have become
• 4:53
  very. Popular where there's.
• 4:55
  A continuously running standard of care arm where there.
• 4:59
  Are always. Patients receiving that standard
• 5:01
  of care and when there, when there are becoming
• 5:04
  interventional treatments, alternative treatments
• 5:06
  available, patients are at that point could be randomized to
• 5:10
  that intervention and there are in the platform trial.
• 5:14
  Usually a couple. Of interim analysis where at
• 5:18
  that time an intervention is reassessed and when there's a
• 5:22
  clear sign of. Efficacy or futility?
• 5:25
  That drug, that intervention is dropped.
• 5:29
  These trials are attractive because if you.
• 5:31
  If you design it properly, the the standard of care arm could
• 5:35
  run for. Multiple years.
• 5:36
  And when available, different types of.
• 5:39
  Drugs or. Experiments could be.
• 5:40
  Added. During that trial, and this
• 5:42
  serves as a sort of master. Protocol.
• 5:45
  Which you could. Always rely.
• 5:47
  On when adding the new treatments.
• 5:50
  These types of trials are a bit more.
• 5:52
  Difficult to design when you talk to your statistician, but I
• 5:57
  really encourage you if you. Have these type of.
• 5:59
  Treatments within your disease. Type.
• 6:01
  To consider these. Trials because the.
• 6:04
  Standard of continuously running, standard of care arm is
• 6:06
  very attractive. The next thing I want to address
• 6:12
  is, like I mentioned, the early intervention for and very.
• 6:15
  Efficacious or futile trial? No one wants to do a.
• 6:19
  Trial who? At the end.
• 6:20
  Is is either. Very negative.
• 6:23
  Or. Very positive and always comes
• 6:26
  the question if you could have seen that coming and these type
• 6:30
  of. Trials are called.
• 6:31
  Adaptive trial designs and the example I want to highlight here
• 6:36
  is based on the whole phone data and this is an example in
• 6:39
  multiple myeloma. Where the trial?
• 6:42
  Randomized 2 two treatment arms, a control treatment arm versus
• 6:46
  an intervention arm, and at the design of the.
• 6:49
  Study it was. Estimated that 668 patients
• 6:54
  needed. To be enrolled.
• 6:56
  And at final analysis. The result?
• 6:58
  Was quite disappointing because there was no difference between
• 7:03
  the experimental and the control arm.
• 7:05
  We questioned whether we could have seen this coming and
• 7:07
  whether. If we could.
• 7:09
  Introduce interim analysis. Maybe have?
• 7:12
  Changed the course in this. Trial retrospectively I.
• 7:15
  Will show you 2 methods now. The first one is the group
• 7:18
  sequential design. Which actually allows you.
• 7:20
  To introduce interim analysis and AT interim analysis.
• 7:25
  You could. You could.
• 7:26
  Decide the based on the the boundaries.
• 7:28
  I will show you whether a trial. Should be continued.
• 7:33
  And in this example, we introduced 2 interim analysis.
• 7:36
  The first interim analysis was when there were still patients
• 7:39
  being enrolled and at the second interim analysis.
• 7:43
  We observed. That all patients were already
• 7:45
  included. But at 2:00.
• 7:46
  Thirds of events we had still some follow up time to monitor
• 7:52
  and if you for instance could consider at the first interim
• 7:54
  analysis to stop the trial. You would actually.
• 7:57
  Save patients from a potential futile or.
• 8:00
  Efficacious treatment and. At the second interim analysis,
• 8:04
  you could primarily say follow up time.
• 8:07
  So if we. Again retrospectively.
• 8:10
  Apply this to the. To this hopeful multiple myeloma
• 8:13
  trial, you'll see interim analysis one, interim analysis
• 8:16
  two, we were not able. To firmly say that this.
• 8:19
  Trial was going to be a negative trial and we should stop the
• 8:22
  trial. And a key concept of the group
• 8:24
  sequential design is that when more data are being observed,
• 8:28
  the bounds of efficacy and futility.
• 8:31
  Are getting. Closer, In other words, your
• 8:33
  estimate of of the data is getting more precise and
• 8:37
  therefore you're more. Precisely to say.
• 8:39
  If your trial is going to be still promising or whether your
• 8:42
  trial is going to be a negative trial.
• 8:44
  At the final analysis, we did not observe any difference.
• 8:47
  Another approach we introduced. Here is a sample size. 3
• 8:51
  estimation, design and what? Is really attractive here is
• 8:55
  that when you. Are conducting your.
• 8:57
  Trial. You're reassessing your power,
• 9:01
  and the power is important when designing a trial because you
• 9:04
  want to be sure you want, you want to have some.
• 9:06
  Probability that your. Trial is actually going to be
• 9:09
  showing a real effect and when the.
• 9:12
  Power is very high. Above 90% usually.
• 9:16
  That's a very favourable result and.
• 9:17
  That's a scenario where you could consider.
• 9:19
  Stopping your trial for efficacy.
• 9:22
  On the other hand, when your. Trial is at.
• 9:24
  Below 30%. Which is a probability of 30.
• 9:28
  Percent that you're going to. Observe a real.
• 9:30
  Difference. That's considered a futile
• 9:32
  trial. And then usually a trial is
• 9:34
  stopped and the promising zone is the interesting zone where
• 9:38
  and that's. Usually set at.
• 9:39
  Between 30 to 90%. Where the sample.
• 9:42
  Size is being re. Estimated.
• 9:44
  So during your trial, you. Have the feeling that this trial
• 9:47
  might have some interest, but at that time you're not sure
• 9:52
  definitely. Not sure whether.
• 9:53
  Your trial is going to show that eventually and then you could
• 9:56
  enroll additional patients. You could re estimate your
• 9:59
  sample size and in this scenario that could have been resolved in
• 10:03
  1/3 of additional patients in your trial.
• 10:07
  When we calculated the conditional power, and we did
• 10:10
  that at the second interim analysis, we found that this.
• 10:13
  Study actually had. 17% conditional power, meaning that
• 10:18
  this trial was very unlikely. To show efficacy.
• 10:21
  And actually was futile and therefore in this scenario
• 10:25
  retrospectively again. We could have decided that this
• 10:29
  trials should have stopped. Which in this case because it
• 10:32
  was after already. 2/3 of. Events, meaning the trial was
• 10:37
  already completed for inclusion. But it could have saved.
• 10:40
  Follow up time. These are the.
• 10:44
  Adaptive trial designs. We also recently.
• 10:48
  Used a Bayesian type of analysis to.
• 10:50
  Investigate whether we could. Improve decision making in a
• 10:53
  clinical trial by using a Bayesian methodology.
• 10:58
  The basin methodology. Is has three key concepts
• 11:03
  consisting of a prior. Which is the belief what you
• 11:07
  have before you start on a. Trial.
• 11:09
  And this is, for instance, a complete remission rate you
• 11:11
  expect for a leukemia patient. Which is usually.
• 11:15
  After induction treatment, about 80 percent, 85%.
• 11:18
  And when you observe additional. Data in In a very positive
• 11:21
  trial, you might. Observe a very.
• 11:22
  High remission rate, for instance.
• 11:24
  Updated likelihood. Your guess what you observe is
• 11:27
  is better, so your updated. Belief is.
• 11:30
  In between, that's the posterior.
• 11:32
  And that's a key concept of Bayesian analysis, which.
• 11:36
  I will show. You how that performed in our
• 11:38
  AML analysis and this is again maybe a disappointing hopeful
• 11:43
  example. But it helps to use these.
• 11:45
  Examples. To investigate what what could
• 11:49
  have changed? And if we looked in this.
• 11:51
  Trial this was. The whole from 132 trial we
• 11:54
  randomized almost 800 patients between a controlled treatment
• 11:59
  and experimental treatments and we observed clearly no
• 12:03
  difference, however, at the design of the study.
• 12:07
  The expected hazard ratio for benefit.
• 12:09
  Was .76. While the observed.
• 12:12
  Hazard ratio was .99. And again, we questioned what
• 12:16
  we've seen at this coming when we introduced interim analysis.
• 12:19
  So that's what we did. We introduced 4 interim analysis
• 12:22
  and when using that group, sequential design.
• 12:24
  We actually observed that. At the third interim analysis,
• 12:28
  the fertility bound was crossed. So at the third interim analysis
• 12:31
  after 450 patients we using. This type of.
• 12:36
  Adaptive designs might have considered to stop the.
• 12:39
  Trial because efficacy. Was never going to be shown.
• 12:44
  However, if we. Use a Bayesian analysis and one
• 12:48
  of the strengths of Bayesian analysis.
• 12:50
  Is that you're also able. To reinforce your control arm by
• 12:53
  adding additional patients. And we use the previous.
• 12:57
  Hopeful trial also including AML patients.
• 13:01
  We actually showed. That using that Bayesian
• 13:06
  approach after the first interim analysis.
• 13:08
  So after 150 patients there was a 1.2%.
• 13:12
  Probability of being. Successful of being able to show
• 13:16
  that .76 expected hazard. Ratio we had.
• 13:20
  At the beginning of the trial. So using a Bayesian analysis and
• 13:24
  reinforcing your control arm, we were.
• 13:28
  Able to show that that trial was already.
• 13:30
  Very early, going to be a negative trial.
• 13:34
  So these are. Examples of methods to introduce
• 13:37
  interim analysis which might really.
• 13:39
  Help you. During a trial.
• 13:42
  To to be sure that you're. Able to demonstrate the beliefs
• 13:46
  you had when you designed the trial.
• 13:50
  External control data, I mentioned that at the beginning,
• 13:53
  are increasingly being used, and we also investigated thinking of
• 13:57
  the Bayesian analysis. Where you're able to.
• 13:59
  Reinforce your control arm if you're.
• 14:02
  Able to. Use external control.
• 14:04
  Data whether that helps to compare to.
• 14:06
  Reinforce your control arm. But the question is, are
• 14:10
  external control? Data really helpful.
• 14:13
  Are those really comparable? To your clinical trial.
• 14:16
  So we again I used an approach. Where we.
• 14:20
  Took the control arm of a hopeful trial and we.
• 14:23
  Sought for. External control data to
• 14:26
  reinforce the control arm and we identified 2 potential
• 14:30
  alternative. Data sources.
• 14:32
  The 1st is the. Harmony and Alliance data.
• 14:34
  Source. And the other is a.
• 14:36
  Truly real world data set. Of patients gathered in the in
• 14:40
  the Netherlands by the Dutch Cancer.
• 14:42
  Society, who also received the same.
• 14:44
  Intensive AML induction treatments and we questioned
• 14:48
  whether the control patients of the clinical.
• 14:51
  Trial were actually similar to the.
• 14:53
  External real worlds control data arms and when we looked for
• 15:00
  baseline characteristics, we clearly, and that's not
• 15:03
  surprising, observed differences.
• 15:06
  Missing data was much more common in the real world data
• 15:08
  source, but also within the harmony.
• 15:10
  Data source and. When we performed the matching
• 15:13
  methods, we. Observed this is overall
• 15:15
  survival that. Outcome for the whole phone
• 15:18
  control. Arm was actually.
• 15:20
  Similar similar compared to the Harmony data.
• 15:23
  But I have to acknowledge that the harmony data is basic.
• 15:26
  Is basically a data set. Who consists?
• 15:28
  Of the majority of clinical trial patients.
• 15:30
  And in this case, in our analysis, we found that about 60
• 15:35
  to 70% were actually enrolled in a clinical trial.
• 15:39
  However, the real. Data from the Dutch Cancer.
• 15:41
  Society. Those patients.
• 15:44
  Clearly performed worse. 10% worse.
• 15:46
  Survival Overall survival at two years.
• 15:50
  When compared to hope and controls and we really believe.
• 15:53
  That this is due to. To various reasons, but for
• 15:56
  instance the missing performance status.
• 15:58
  How? Fit is your patient, and is it
• 15:59
  really a patient who could also be?
• 16:02
  Included in a clinical trial? Is it comorbidity?
• 16:05
  Data all all types of. Data that are not available in a
• 16:09
  real. World data set so we are really
• 16:11
  hesitant to firmly say that real real world data could actually.
• 16:15
  Supplement the data in a clinical trial at this.
• 16:18
  Stage so I hope to have show. You that innovative trial
• 16:24
  design. Is the future.
• 16:26
  And with the personalized medicine becoming available now
• 16:30
  we need to. Adapt our trials according to
• 16:32
  the. Needs to the accelerating needs
• 16:34
  within the clinical trial development.
• 16:37
  To deliver the right drug. For the right patient at the
• 16:40
  right time, adaptive designs I've I've shown you could help
• 16:45
  you. For potential and early trial
• 16:46
  intervention. External data might reinforce
• 16:50
  your control arm and important in that.
• 16:52
  Perspective, but. You need to define and identify
• 16:57
  the right and the proper external control.
• 16:59
  Data set. And with all this, and
• 17:01
  especially if you want to adapt your trial during the conduct of
• 17:04
  your trial, you need a solid trial infrastructure.
• 17:08
  Where you have. Immediate data available and who
• 17:11
  can support the complex trial that you might design?
• 17:17
  And with that, I would like to conclude.
• 17:18
  Thank you.