Latest / Cardio Pod: Heart of the Matter / Importance of Subgroups in Heart Failure Clinical Trials - Part 1

Transcript

Importance of Subgroups in Heart Failure Clinical Trials - Part 1 · Cardio Pod: Heart of the Matter

• 0:00
  Good morning, good afternoon, good evening.
• 0:09
  I'm Chris O'Connor, Principal Investigator of the Heart Failure Collaboratory.
• 0:13
  And we're here today with Heart of the Matter, the official podcast of the Heart Failure
• 0:17
  Collaboratory.
• 0:18
  And I'm here with three of my very favorite and distinguished guests today, Dr. Mariel
• 0:24
  Jessup, Dr. Javed Butler, and Dr. Mitch Sapka, welcome.
• 0:29
  And we're here today to talk about ways that we can improve therapeutic development.
• 0:35
  That's been the journey of the Heart Failure Collaboratory.
• 0:38
  And today's topic is subgroups.
• 0:41
  Subgroups is something that we've all had experiences with, all had challenges with,
• 0:46
  and yet we haven't really harmonized our way of approaching subgroups.
• 0:50
  And I'd just like to open up the discussion with Dr. Sapka and say, Mitch, what is a subgroup?
• 1:00
  And why do we talk about subgroups in clinical trial development?
• 1:03
  Thank you for letting me start here.
• 1:06
  I'm sure my colleagues are also going to have intellectual things to say here.
• 1:10
  And I think that we're constantly interested in a clinical trial, looking at which components
• 1:16
  or which subpopulations within the trial most benefited from treatment or most had the best
• 1:24
  benefit to risk profile.
• 1:26
  There are often not everyone in a clinical trial responds the same to an intervention.
• 1:30
  Not everyone in a clinical trial kind of responds the same to not having an intervention.
• 1:33
  And so we really want to best understand from a clinical standpoint, who is most likely
• 1:38
  to benefit from the overall population.
• 1:40
  And this matters both to a clinician from a clinician perspective and also from a regulatory
• 1:45
  perspective.
• 1:46
  And we want to know, are the patients that are going to receive this therapeutic intervention
• 1:51
  going to benefit and what is their risk to benefit ratio?
• 1:55
  And who can we best target this intervention and to get the best overall treatment effect
• 2:00
  for patients and make them feel better and live longer and happier and healthier lives?
• 2:05
  So really, Mitch, what you're saying is a subgroup is a category of participants that
• 2:12
  have measurable characteristics, something that you can measure such as demographics,
• 2:18
  biomarkers, genetic profiles, comorbidities.
• 2:23
  These are categories of potential subgroups that are of special interest to us that we
• 2:29
  are looking for perhaps signals of enhanced efficacy, signals of attenuated efficacy,
• 2:37
  perhaps issues of safety.
• 2:39
  Mariel, what subgroups do you think all of us in the heart failure space should be looking
• 2:45
  at when we talk about subgroups as trials?
• 2:49
  Well, I'm going to cheat and tell a story before I answer that question.
• 2:53
  And that is I've been looking at our heart failure path.
• 2:58
  And I came across the original VHEP trial.
• 3:02
  And you all know this story, but it's so worthwhile mentioning.
• 3:06
  There are only 640 patients put into the VHEP one.
• 3:11
  And if you look at the characteristics of the trial, there is no mention at all about
• 3:16
  race or ethnicity.
• 3:17
  Apparently, we only discovered that years later.
• 3:21
  But it turns out because it was done at the VA, there were 29% of Black patients that
• 3:27
  were entered into that trial, probably the highest number of Black patients in many,
• 3:32
  many, many heart failure trials.
• 3:34
  Well, it also turns out here was the very first vasodilator trial in chronic heart failure.
• 3:42
  If that many Black patients hadn't been put into that trial, the trial overall would have
• 3:48
  been negative because it showed when you look back that white patients did not respond to
• 3:54
  hydralazine, isosorb, iodine, nitrate.
• 3:57
  It was only the Black patients that did.
• 4:00
  Nevertheless, because the overall group was positive, it really for the first time gave
• 4:06
  hope to patients with heart failure and opened up the whole field of clinical trials.
• 4:12
  I think that's so ironic that for that first trial where there was actually diverse enrollment,
• 4:19
  it opened the door and then we closed again on bringing in the subgroup of Black Americans
• 4:27
  or Black patients anywhere.
• 4:30
  So subgroups are critical primarily so that your individual patients can say, was someone
• 4:38
  like me in the trial?
• 4:41
  Is a patient that you're going to give this drug to, can they say, well, how many people
• 4:46
  like me were studied?
• 4:48
  Now that's impossible to answer in subgroups, but there needs to be at least women in a
• 4:53
  trial because, oh, by the way, they're 51% of the population.
• 4:59
  And we've done a really bad job of that too.
• 5:02
  So there needs to be some remote, at least passing resemblance of the percentages of
• 5:11
  patients in a clinical trial that is equivalent to the population where you want to use that
• 5:16
  drug.
• 5:17
  That's what I think subgroups need to be.
• 5:21
  That really great example and well-stated job, what Mariel is saying is that subgroups
• 5:30
  of interest, participants of interest, we should make sure we enroll a sufficient number
• 5:38
  of those patients in clinical trials so we can say something that's meaningful at the
• 5:42
  end of the trial.
• 5:43
  And I think the V-heft is a great example for a number of reasons, but are we ever powered
• 5:50
  well enough to talk about subgroups in clinical trials?
• 5:54
  Do trialists ever think about powering studies so large that they would have sufficient power
• 6:00
  in subgroups?
• 6:02
  It really depends on any given trial and what subgroup you're talking about.
• 6:06
  One way to think about the subgroup issue is that you do the trial that you do, there
• 6:11
  is only so much information you have when you go into a trial and then you learn a lot
• 6:15
  after the data comes in.
• 6:17
  So if you have a positive trial, it doesn't mean that everybody in that trial benefited.
• 6:23
  So you have this issue of risk benefit and maybe some groups it just does not matter
• 6:27
  and all you're doing is accruing risk without benefit.
• 6:31
  And the flip side is that a trial overall is negative, but there are certain subgroups
• 6:35
  that really benefited, but overall they were underrepresented and were in minority and
• 6:40
  now you're calling the trial negative.
• 6:42
  So you sort of do these analysis once the trial is over.
• 6:46
  Now as a trialist, I can tell you, you know, you love the subgroup analysis to be incredibly
• 6:51
  boring so that there is nothing there and the benefit is across all the groups and you
• 6:54
  use it in the largest amount of patient population.
• 6:58
  But when that doesn't occur and you do see the heterogeneity, then you're sort of always
• 7:02
  is stuck with this question, whether it's a chance finding or whether this is a real
• 7:06
  signal.
• 7:08
  And that comes to your point, Dr. O'Connor, that, you know, we don't power the subgroups
• 7:12
  unless and until there is something very specific that you're, you know, you expect an interaction
• 7:18
  with diabetes or with atrial fibrillation and you say, well, at least 30% of the patients
• 7:22
  should be this or 40% of the patients are that.
• 7:25
  But by and large, if you look at the demographics, if you look at some of the other laboratory
• 7:30
  or other clinical characteristics, we just let the trial roll and then however many percentage
• 7:34
  of patients with any given characteristics come, they come.
• 7:37
  And that's why we have sort of a poor man's way of dealing with that.
• 7:41
  So rather than having a p-value of less than 0.05 for interaction, we say, well, you know,
• 7:45
  if it's less than 0.1 or less than 0.2, we should really take it seriously.
• 7:49
  But the short answer to your question is that no, we really don't power for subgroups.
• 7:54
  And that's where this issue of populations of special interest becomes really important
• 8:00
  because you might consider powering for those groups.
• 8:03
  Well stated.
• 8:04
  I think, and back to VHF, I think one of the wonderful aspects of it, Mariel said, they
• 8:10
  enrolled the patient, but they didn't pre-specify that analysis in the black patient population.
• 8:16
  But that's where the signal was.
• 8:19
  And they did have the foresight after doing that analysis to do a prospective trial in
• 8:24
  that population, AHAF, and found a very strong single, as we all know.
• 8:30
  And what the regulators did is said, okay, you need two trials.
• 8:35
  We're going to give you VHF with the black population because the signal was there.
• 8:40
  And that's why subgroups are important.
• 8:42
  And they had AHAF.
• 8:43
  They had the two trials.
• 8:45
  They had the consistency.
• 8:47
  And they got the approval.
• 8:48
  Mitch, in the world of subgroups, have we used it to tailor the regulatory labeling?
• 8:57
  I can think of perhaps we did a little bit of narrowing of the label when we looked at
• 9:03
  succubitrol valsartan across the spectrum of EFs.
• 9:07
  It kind of says, based on the paragon showing that that very high EF, there wasn't much
• 9:13
  of a signal that they...
• 9:14
  Correct me if I'm wrong, but the label kind of says across the spectrum of EFs that have
• 9:20
  a mildly reduced ejection fraction or something like that.
• 9:25
  That's exactly right.
• 9:26
  I think that is the one where it is the most recent and most obvious where the label says,
• 9:31
  and I'm forgetting the exact terminology, but something to the effect of the effects
• 9:36
  is most obvious in those patients with left ventricular ejection fraction below normal.
• 9:42
  Which attempts to describe this differential treatment effect or this effect modification
• 9:49
  associated with left ventricular ejection fraction.
• 9:51
  As the ejection fraction increases, there is a less discernible treatment effect until
• 9:56
  you get it up into the above normal ejection fraction where it's really undetectable.
• 10:01
  And I think that now that we have this data and a lot of other different therapeutics
• 10:07
  that we use in this space, this sort of effect is probably visible in a bunch of different
• 10:11
  therapeutics but has not made it into the label because it was not part of the label
• 10:15
  application.
• 10:16
  But this is certainly something that I think is going to rear its head more frequently
• 10:20
  now that we have bigger data sets across broader ranges of these predictor variables such as
• 10:26
  ejection fraction.
• 10:29
  We're faced with the dilemma that when we go back and praise 1M-lodipine and heart failure
• 10:34
  reduced ejection fraction, highly statistically significant interaction on the stratification.
• 10:40
  That subgroup, we really thought something was going on, but it was in the non- ischemics.
• 10:46
  We thought it really was going to be, the action would be in ischemics.
• 10:50
  Went to the FDA, they said not good enough for standalone, do another trial.
• 10:56
  Same sites, same protocol, same baseline characteristics, same geography, absolutely neutral result.
• 11:07
  Javed, I know you're getting all the tough questions.
• 11:14
  Why did that happen?
• 11:15
  Yeah, well, so that's the whole issue about chance finding, noise, multiple testing and
• 11:23
  getting a false positive result, and that's why I think that we need to confirm our findings.
• 11:29
  So obviously, you know, if it is only one study, but your p-value is 0.000125 or something
• 11:35
  like that, then you can be pretty darn certain that if you repeat the experiment, your p-value
• 11:39
  is going to be less than 0.05.
• 11:42
  But if you're at, you know, 0.02, I mean, the chances that you're going to hit less
• 11:47
  than 0.05 is not that high, and that's why the two experiment rule exists.
• 11:52
  And I think it's a wise thing to do, because on one hand, we definitely don't want eligible
• 11:57
  patients to be deprived of therapy, but newer therapies are costly and almost all therapies
• 12:03
  have side effects.
• 12:04
  So we just want to make sure that when we prescribe the therapies, we at least have
• 12:08
  some reasonable comfort that the patient will benefit from it.
• 12:11
  So I think that the first study in which the non-ischemic signal was found was a chance
• 12:16
  finding and when a fully powered study was done, we could not replicate the results.
• 12:22
  If a trial is positive though, Mario, we look at Galactic HF, we look at the Paterson study,
• 12:30
  and then you see a subgroup, you say, okay, well, it was sort of marginally positive,
• 12:35
  but it was a large study in the case of Galactic HF.
• 12:39
  And then you see amplification of signal in a subgroup.
• 12:43
  It's a large subgroup, the LOEF, thousands of patients, I think close to a thousand deaths
• 12:49
  or events.
• 12:51
  That signal's magnified.
• 12:52
  Why couldn't we go forward with one trial and given the trial's positive and narrow
• 12:58
  it to the subgroup that has a very strong signal?
• 13:02
  Mariel?
• 13:03
  We can, just that that's very costly and most companies that make these therapeutics are
• 13:12
  not interested in that.
• 13:13
  They want to be able to say, this is the widest group of patients that are drugged.
• 13:19
  But I think it goes back to the fundamental problem that the Heart Philly Collaboratory
• 13:24
  is also trying to address, which is we really don't characterize, or the word that people
• 13:30
  are using is phenotype, our patients going into trials very well at all.
• 13:36
  So we always have to scramble and go backwards and say, ah, it was the patients with a LOEF
• 13:42
  or women with a higher EF or people with an anti-pro BMP that was elevated or people that
• 13:47
  didn't have an anti-pro.
• 13:49
  I mean, we always go back and come up with an explanation, as you've already said.
• 13:54
  But what would happen if we took a few characteristic and decided on each patient that went into
• 14:03
  the trial, what phenotype do they exist?
• 14:06
  Then we've got a little better opportunity to understand as we go forward.
• 14:12
  I mean, we spend millions and millions and millions of dollars on these trials.
• 14:17
  And at the end, we have so many questions.
• 14:20
  Now, will that approach answer all the questions?
• 14:24
  No, but we'll be smarter at the end.
• 14:28
  We always just feel like, ah, that was a negative trial, or, hold up, that was a positive trial.
• 14:34
  Now let's do it in this group of people, or let's do it in that group of people.
• 14:38
  It would be nice to know a lot about the patients at the end of the trial.
• 14:44
  Well, you're exactly right.
• 14:46
  I mean, that is one important lesson, is that we should do deeper interrogation of largely
• 14:53
  positive neutral negative trials to understand subgroups of potential benefit.
• 14:59
  They understand the phenotypes better.
• 15:02
  But it's just a challenge for these sponsors.
• 15:06
  Javed, you're leading an investigation of verisigua.
• 15:11
  Okay, it's a positive trial.
• 15:14
  It actually is a proved drug.
• 15:16
  And now you're leading a trial called Victor in which they're going perhaps into a sweet
• 15:22
  spot of enhanced efficacy, which was in that more stabilized, a little bit lower, natriuretic
• 15:29
  peptide population.
• 15:31
  And if everything falls the way that Victoria showed, this should be a win.
• 15:38
  But could we have averted that and just said, the sweet spot is the sweet spot?
• 15:45
  Yeah.
• 15:46
  So, I mean, if you look at the development of therapeutics for all of the agents that
• 15:50
  we use today, ACE inhibitors, beta blockers, MRAs, RNAs, SGLD2 inhibitors, all of these
• 15:54
  drugs have been sort of tested across the spectrum.
• 15:58
  Sometimes we start at the more difficult end of the spectrum.
• 16:02
  Like ACE inhibitors and MRAs, we started with consensus and RALs, and those trials were
• 16:06
  positive and we came to the less sick patient populations.
• 16:09
  With beta blockers and SGLD2 inhibitors, we did the opposite.
• 16:12
  We started at the less sick end of the spectrum first, and then we went to the sicker patient.
• 16:17
  So you want to cover sort of all the spectrum of patients.
• 16:20
  And I think if you haven't studied any specific given patient population, then where do you
• 16:25
  draw the line on adjacency?
• 16:28
  If you do a trial in HFREF, do you say extend it to mildly reduced, all heart failure, hospitalized
• 16:34
  heart failure?
• 16:35
  So this issue of adjacency and expansion of indication is really tough.
• 16:40
  And I think that in as much as it is costly and time consuming to do multiple trials,
• 16:45
  I think that probably is the best thing to do because we have not always been successful.
• 16:51
  You know, ACE inhibitor, beta blocker tends to work in post MI patients.
• 16:56
  RNA trial was negative.
• 16:57
  It did not.
• 16:58
  RNA trial in severe heart failure in the live trial really did not show much benefit.
• 17:03
  So I think that doing trials in populations that have not been studied, rather than extending
• 17:10
  the indication on the basis of logic, which may be true, but untested, is a little bit
• 17:15
  of a slip you saw in my opinion.
• 17:19
  If Mitch had done the Victoria trial with vitality and expanded and said, okay, we're
• 17:26
  going to study a new drug across the entire spectrum of EF, but we're going to stratify
• 17:34
  F and you get a nice signal in R and you get nothing in P and you get this sort of maybe
• 17:42
  gets diluted out a borderline P value.
• 17:45
  Maybe it's for the overall thing.
• 17:46
  It's oh five.
• 17:47
  It's oh six.
• 17:48
  It's oh four.
• 17:50
  Would the would the agency be willing to approve that subgroup in one problem?
• 18:00
  So maybe this is what's happening.
• 18:02
  Maybe this is the barrier.
• 18:04
  We have to separate and do two trials longer, more money.
• 18:08
  Totally against what we're talking about in the HFC.
• 18:12
  Because I think you might say that's a hard one.
• 18:17
  Well, so I have two answers for you.
• 18:19
  The first is that there are clear FDA guidance on what constitutes substantial evidence for
• 18:27
  a single clinical trial is actually just released by the agency and the level of evidence that you
• 18:48
  need to really get away from this two clinical trial metric.
• 18:52
  So you know, it's not just enough to pre-specify a secondary outcome or secondary endpoint or really
• 19:08
  even to pre-specify what your you know, how you're going to perform sub-analyses.
• 19:17
  It's really to say my primary hypothesis will spend my statistical shots on goal in this
• 19:23
  way and it will look at these two sub-populations separately.
• 19:29
  You can in fact design a trial that would be similar to doing two trials where you enroll
• 19:32
  the entire spectrum of say ejection fraction and take two shots on goal both with smaller
• 19:38
  alphas that would be similar to taking one shot on goal for each clinical trial.
• 19:44
  And then you know, that would be essentially the same output.
• 19:48
  And so I think that that is probably the way that these things can could go forward and
• 19:54
  could get managed in terms of taking an appropriate and statistically robust pathway to doing these
• 20:01
  sorts of analyses.
• 20:02
  And I think that pre-specifying not only the shot on goal, but also the directionality,
• 20:09
  not just that you are going to stratify based on ejection fraction, but that you actually
• 20:14
  predict say that the drug will be more effective in a low ejection fraction population because
• 20:19
  we can always imagine that if you choose a low enough ejection fraction population, you
• 20:23
  may actually argue post hoc that oh, it's actually those patients were too sick that
• 20:28
  I wouldn't expect them to respond.
• 20:30
  So you need to say I expect that we will see a benefit, a better benefit in this population
• 20:35
  than this one, and that needs to be built into your statistical testing plan.
• 20:40
  Really well said.
• 20:41
  So as we wrap up here, for the audience, the lessons, you're designing a trial, you got
• 20:49
  all these characteristics subgroups that you want to look at at the end of the trial.
• 20:58
  How do we prior to rise that?
• 21:00
  What are the mechanisms, Javadat, of how we can prior to rise subgroups, subgroup analysis?
• 21:07
  I think Mitch alluded to it a little bit in the alpha spend and weight, but how do we
• 21:12
  do it?
• 21:13
  Yes, so it depends on how you're feeling about any given particular subgroup.
• 21:19
  So for instance, when the delivered trial was coming out, they really felt that they
• 21:24
  need to study the less than 60% EF as a separate group.
• 21:28
  So that was part of the primary endpoint and they split the alpha.
• 21:32
  That's one way of doing it.
• 21:34
  That's a special circumstances.
• 21:35
  If you really feel strongly about some subgroups, you can make it a secondary endpoint with
• 21:40
  hierarchical alpha protection.
• 21:42
  That's another approach to do it.
• 21:44
  But I think the simplest thing is to just really think through the study, the mechanism
• 21:48
  of action, the patient population, re-specify the analysis, and perhaps give some order
• 21:54
  and some hierarchy to your subgroups as well.
• 21:57
  And if you feel that something is important enough that you hypothesize that this group
• 22:03
  will benefit more, just make it into an alpha protected secondary endpoint because then
• 22:08
  you can claim label if the primary endpoint is met.
• 22:12
  The issue comes up is that when you do it as a secondary exploratory analysis, not re-specified,
• 22:19
  not alpha protected, there is just no way to know whether or not you have had a chance
• 22:25
  finding.
• 22:26
  And at that point, it becomes a very responsible thing to do to repeat the experiment and make
• 22:31
  sure that this is not noise, but this is a real signal.
• 22:35
  Well, thank you, Javed, and thank you, Mitch and Mariel, for great conversation on subgroups
• 22:42
  today.
• 22:43
  And there'll be more of this conversation.
• 22:45
  We know this is a very important topic for the collaboratory and for therapeutic development.
• 22:51
  But for today, a heart of the matter, what counts in HFC and therapeutic development.
• 22:57
  I want to thank you, and let's continue to charge Javed to make trials better.
• 23:02
  Thank you.