Tip:
Highlight text to annotate it
X
Male Speaker: Pass the gavel to Howard to lead the panel
discussion as the moderator.
Howard McLeod: Thank you, Dan. Brandy, did you receive the
slide?
Female Speaker: I do, and I'm loading them right now.
Howard McLeod: Thank you very much. So I was asked to be
the summarizer, but really it's not so much summarizer as it is stimulating some discussion.
And I was given one slide. And so I think that the quote from the famed American philosopher
Mary Violet Relling is really important, that you can't manipulate them if you don't know
how it works. And part of this is that I'm not sure we're discovering the right context.
And by looking at pharmacogenomic implementation from the way it's been done traditionally,
we're really looking at endpoints that offer the exact ones needed for implementation,
and we're not necessarily involving the right people in that effort. And so I think there's
an opportunity in eMERGE to really overcome some of this because the discovery will be
happening in the context of routine practice, not within the unusual situation of a clinical
trial. And especially in the oncology area, we know that the patients who are able to
go on trials are the rare, unusual patients, not the normal patients.
I think that the randomized trial versus the organically randomized data from the clinic
is certainly an important element, eMERGE has opportunity there, but I really worry
that -- and we can get the necessary sample sizes as Dan alluded, Dan Roden alluded, but
I'm not sure that we have the right methods for really optimizing data from the electronic
health record, and I don't mean the methods for extracting the data or deriving phenotypes,
but from a study design standpoint, yeah, this is a pretty nascent area and it needs
some focus.
Last thing, there's not a lot of evidence, so we can do interactive interventions into
the electronic health record. Now, Vanderbilt's done a little bit of that, but there's still
a lot to be learned about how do we take the eMERGE-like setting and turn it into an implementation
science setting in its fullness. And then lastly something from Dan Masys is guidance
for best evidence-based therapy selection is the sweet spot for pharmacogenomics. It's
something that can be done with these infrastructures, such as [unintelligible] or eMERGE, but there's
still a lot of work to be done to not only discover but to really push it forward into
practice.
So I'll stop at that -- at this point, and see if we can get some discussion started.
Male Speaker: Okay, the floor is open for questions or comment
here, and I'll lead with, I guess, a proactive question because it seems to me the dichotomy
between discovery and implementation makes it sound like implementation is just a done
deal, but operational health care is something you don't change. But, in fact, the other
major theme afoot in the national agenda is learning health care systems, and they continuously
change and they continuously learn. So how do -- so I'll ask Mary first, what's the intersection
of her set-up assertions and the idea of learning healthcare system?
Mary Relling: Well, I mean a learning healthcare system,
I guess you could say that there's aspects of -- that drive that learning that are based
on implementation projects, but I still think that those are largely coming from research,
right? So we just had some example of that with the system-based quote, "randomization,"
but that really is a research project to see whether outcomes are indeed better if genomic
medicine is used versus not. If you know that outcomes are better, if you know that a patient
has a completely inactivating variant in their TPMT status, you -- it would 100 percent unethical
to give them the normal dose of that drug, and to test yet again 30 years after we know
the answer, whether there's more or less toxicity, or better or worse outcomes in patients who
have their thiopurine dose based on TPMT or not. So, that could be a part of a learning
healthcare system, if the healthcare system is practicing really bad medicine. And I acknowledge
that there are many aspects of the healthcare system that are practicing horrible medicine,
but that doesn't make it right for us to try to capitalize on that lousy health care in
a kind of vulturistic way just to generate more data. It's something that's not ethical
to study.
Chris Chute: Chris Chute, Mayo Clinic. I don't disagree
with you, Mary, but I think the question of where's this boundary between research and
implementation. The whole elegance of a learning healthcare system is that, you know, you really
merge quality improvement and research, because they're the same methodology even though they're
done by different people, and apply the results not only to the literature but to the practice.
So, you know, we haven't discovered everything; I think we can safely agree upon that. And
in the context of, I guess, expanding the boundaries of our discovery, it's a more systematic
integration of research in a sense into practice so that we can harvest the experience of routine
clinical practice, which, for the most part, goes fallow in today's biomedical world. So
I actually see it more positively.
Mary Relling: I see it positively, I just see it as clinical
research not implementation. This is very --
Howard McLeod: Part of my reason for including the "Are the
right people involved in the effort" question was around what Chris just raised in that
a lot of the folks involved in quality improvement aren't involved in eMERGE or aren't involved
in a lot of these efforts, and certainly at my previous institution and my current institution,
I've been able to cause change to happen much faster by working with that group than by
sticking with the group that's more comfortable with a endless number of clinical trials.
Marc Williams: Yeah, this is Marc Williams, and I would also
add to the discussion, having worked in integrated healthcare delivery systems, and again, using
the methods of quality improvement coupled with research, I think there is a real sweet
spot there. But I think perhaps the takeaway for the questions that Howard has teed up
is if we think about the next phase of eMERGE, would a component of that be, what are the
appropriate trial methodologies, or would a proposal for an RFA have to include something
that says, you know, what is your pragmatic trial methodology to be able to study this,
because that's where the implementation research is really going. And as an example, I mean,
PCORI is highly emphasizing pragmatic clinical trials, which I think eMERGE is very well-positioned
to be able to leverage. So is that the question that we're really asking about inclusion here,
which is different trial methodologies.
Julie Johnson: So, this is Julie Johnson. I mean, as I listen
to this conversation, I think that the question was put forward as a dichotomy really, discovery
versus implementation. And in reality, there are three steps. There's, you know, the original
discovery of the genetic association, and unfortunately, most cases in pharmacogenetics
and otherwise aren't at the stage that TPMT is, where I think we still do need evidence
for whether the genetic association has clinical meaning. And so that's not discovery and that's
not implementation; I agree. It's some sort of clinical research, and so I think the question
is how do you do that best. And then there's the clinical implementation.
And so, you know, I think part of the problem is that the latter two things have been lumped
in some ways in, quote, "implementation," you know, so is implementation the stuff that
is truly ready and then you're testing uptake and, you know, attitudes about uptake and
that kind of thing, and then we need to come up with some term for that middle -- that
middle phase, so testing the relevance of implementation, testing, you know, or testing
the value, the clinical value of utilizing genetic information to guide care decisions.
And so, you know, it does seem like eMERGE is really obviously positioned for the -- you
say there's three things, the first and the latter; for the middle, depending on the trial
design, and I think like Marc said, if there's pragmatic design, then perhaps eMERGE is really
perfectly situated, and, I mean, I would tend to agree that that might be the better approach.
But if it's a more tradition clinical trial design, then I don't know that that makes
sense.
So I would argue that the question isn't maybe posed quite right because I think there's
really three phases, and we're talking about the first and the last when we say implementation
and discovery, and that middle piece, which is maybe where we need the most help, is kind
of missing.
Erwin Bottinger: This is Erwin from Mount Sinai. I think I
agree entirely with what Julie said, and certainly reflecting on the conversation, I think a
number of points were raised that clearly need to be considered very seriously. And
one of the major points from Howard on his slide is that do we have the right people
for implementation. And I would say within the constructure of eMERGE, probably we're
not -- we don't have the right level of expertise for implementation because you know what missing
in eMERGE, the great evil is that of the constituency of providers and experts in clinical care
and workflow. So if implementation is something going forward, then I think there needs to
be a clear expression that this stakeholder or these stakeholders will need to be a broad
inch of the fold that is capable of understanding clinical workflows that are essential in clinicians.
And currently, we don't have those at the table for true implementation.
I also think that, in agreement with Mary, that -- and truly -- that we are in well-positioned
in this sweet spot, and in between, and with some, you know, intelligent approaches, what
Marc pointed out, pragmatic trial design and others, we have some unique opportunities.
One was raised and was mentioned recently in the conversations that we've had with our
external advisors, so this opportunity that we have is to recall by genotype, and we should
think about this, what kind of fantastic opportunity this is for us where we have phenotypes across
the electronic heath record, we have genotypes for tens of thousands of individuals, and
if there are burning questions or scenarios for which evidence needs to be generated in
a specific case to move something over the border from research to generate the evidence
that would allow us to formulate an implementation strategy, I think that's what we can do, is
this kind of approach. And so perhaps that's one way of thinking about that sweet spot.
Male Speaker: So this is Justin [spelled phonetically],
and I want to reemphasize what was said. One of my favorite quotes is from Paul Clayton,
who said that, "You implement every system three times. You implement it once to find
out if it can be built at all, you implement it the second time to figure out how you should
build it, and then you actually build the one you use."
When we talk about implementation science, a lot of that focuses on the uptake of the
intervention by the target party. But, in fact, in eMERGE-II, we're already at the first
step of implementation, you know, can you even build a genomic decision support system
that will bolt onto an EHR. So most of the questions of implementation science are the
ones we will address when we build the systems the second time, which is to figure out how
we should build them and integrate them into the workflow. So I think we need to think
about, as they were saying, implementation as multiple things, and a lot of what we're
doing right now is just can we make the technology jump through this hoop at all. And I think
an appropriate target for eMERGE-III is, okay, if we can make the technology jump, how should
we make it jump to optimize the clinical practice?
Male Speaker: This is Zach [spelled phonetically]. Can you
hear me?
Male Speaker: Yes, we can.
Male Speaker: Okay, great. It's very unusual for me not
to be able to jump right in, so I'll do my best now.
So, Dan Roden made some comments which triggered in my mind some, I think, important scientific
questions for eMERGE-III. So Dan mentioned that some rare variants were at a higher frequency
in the African population. And then on the chat part of this meeting software, he also
responded to everybody that there seems to be about 50,000, perhaps, African Americans
in the cohort. And I think that's important because, to say it in a three-part question,
a) does Dan think those rare variants that are more common in the Africans are actually
the causal variants, and therefore actually cause a pharmacological change in those individuals,
if they're subjected to the same drugs? In the context of, b), recent very nice papers
showing for a number of heart diseases, such as hypertrophic cardiomyopathy, rare variants
that were supposedly causal and first ascertained in European populations have prevalences of
30 percent in Africans, where it's clearly not the case there's 30 percent HCM in that
population. Which leads me to the scientific agenda for eMERGE-III, is I really think we
can start addressing, and we should address because very few others have and we're in
a unique position to be able to do so, because of the electronic health records system and
health center derived population to be able to start understanding the degree to which
some of these variants are genuinely causal variants or are incidental findings that actually
could result in over treatment or mistreatment of individuals in underrepresented minorities.
And to that, I'll leave to Dan Roden and others to respond.
Dan Roden: So I'll take the opportunity to answer Zach
because he asked me specifically. So the answer to whether 2C9*6 is causal or not, I suspect
it is because we know something about its function, and that it's a reduced function
allele, and therefore it makes biological sense as well as -- as well as statistical
sense. But I think that the more generic question of variant of uncertain significance, especially
in the rare variant spaces, is one that anyone who takes care of patients with hypertrophic
cardiomyopathy, or in my case, you know, cardiac channelopathies, struggles with every time
you see a patient. And I think that we're going to have to come to an understanding
that there are diseases that are caused by rare variants, that are diseases -- that are
phenotypes that are likely to be modulated by rare variants, and then there are rare
variants whose role in pathophysiology has been dramatically overstated by initial studies.
And I agree with Zach. If we're ever going to make headway on a variant that is 1 in
1,000, how are you going to figure out what it does to a phenotype? You can either do
an in-vitro evaluation of function, and sometimes even those are misleading, or you can ask
the question, does it associated with some kind of phenotype. And to do that, you have
to have very large numbers. And we're one of the places -- eMERGE is one of the places
that can do this. I see that people are talking about Biobank U.K., about Kaiser, about the
VA, and I think that those are large resources with which eMERGE ought to consider collaborating.
It's easy to say that, and it's actually, operationally, hard to do. Each one of them
has their own access models. Each one of them has their own datasets that are -- that are
bigger in some ways and smaller in other ways compared to eMERGE. I know a little bit about
Biobank U.K., and they are having trouble getting the kind of detailed electronic health
records that we are used to within eMERGE, so they may have 500,000 samples, but they
have that drawback. They have the great advantage that they have very detailed phenotypes for
some particular diseases. So I think each of them bring something to the table, and
eMERGE, as large as it is, ought to be a player in this space. And I think I've said enough.
*** Weinshilboum: Dan, this is *** Weinshilboum, and I couldn't
agree with you more, which will shock you that I'm saying that, but, as a matter of
fact, what you said, and I think what eMERGE has shown, has profound implications for clinical
trial design. I think the eMERGE program is beginning to tell us that the way we've been
doing the studies with 2,000 patients in one arm, standard therapy and 2,000 in the other
arm, standard therapy plus another drug, may not be the way to go forward. Do you have
any comments with regard to the implications of your own comments for clinical trial design?
Dan Roden: No. [laughs] I think I'll let other people
talk; I don't want to monopolize the air time. But I think that -- I will just say a generic
comment -- that implementation and trial design has to happen after we do discovery. Or coincident
with -- and I made the point that, you know, Dan Masys made the point about a learning
healthcare system, and that was sort of what I was trying to do on the last slide. I think
that the implementation side and the discovery side go hand in hand: as one dataset grows,
the other dataset, by its nature, gets richer.
Michael Gaziano: So this is Mike, just two quick comments about
the VA's Million Veteran Program, we do have approaching about 50,000 African Americans,
so I do agree, Dan, that there is plenty of reason to figure out how best to collaborate,
and we are working behind the scenes on figuring out how to create datasets that can be accessed
from outside, and also be identified, to overcome some of our collaborative barriers. And the
second is that we've got a big initiative on implementing trials within the system,
and I really think that we do need to focus some attention on the implementation of trials
that are more broad in terms of their enrollment, but that use the EHR backbone, rather than
every time we do a trial, creating a whole new electronic backbone to support the trial
activity. And we've got a trial in that point-of-care mode that's under review next month that will
be randomizing people at the time they pick up their drugs. It happens to be a hypertension
trial of Corzalidone [spelled phonetically] compared to hydrochlorothiazide, and we've
got some regulatory issues to get around. But I think that, you know, eMERGE with its
expertise in implementation could help play a big role in figuring out how to do very
large trials at a very low cost utilizing the electronic health record and making trials
go from $100 million to trials to, you know, $5 million or $10 million trials.
John Harley: This is John Harley in Cincinnati. Maybe one
of the themes of eMERGE-III would be to assess benefits. We have the electronic medical record
to go to and we would have this enormous database, and so if the issue in clinical application
is cancer, then we would be in a really powerful position to actually reach some kind of resolution
or make progress about what judgments to make about [inaudible].
Female Speaker: Yes, I think because, I mean, we've been excited
about hemochromatosis project [inaudible] to a variety of things, and that's a really
interesting question, and I would lengthen that to penetrance and which variants are
and are not pathogenic. And I think a huge amount of the work in genomic medicine now
is understanding which variants are pathogenic and which are not. And we have a lot of data
that we could put to that question and really understand that in eMERGE should be really
exciting.
Male Speaker: Howard, I want to really echo that point because
both penetrance and heritability can be derived from this dataset with current technologies,
and can really add some rich context to almost every phenotype that is just totally missing
right now. I mean, we are chasing things that really we shouldn't and vice versa, and that
would be a huge service to the community, in addition to all the cool methods and findings
that come from it.
Male Speaker: I agree, and I think this is the sweet spot.
Male Speaker: There was just a comment earlier that if there's
a rare gene frequency of 10 percent or less, we really need to enroll 1,000 people with
that genotype in there. We are not going to be able to do that for the number of potential
genetic variants. We have to at some point go ahead with implementation and then use
different study designs to study the rare variant genotypes. We cannot randomize every
one of them and think that we can get prospective studies of everyone. Certainly some we'll
need to, but by and large, we can't hold back on patients until we have every SNP sorted
out.
Female Speaker: Yeah, so, I mean, the idea is not to do a
giant trial of every variant. The idea is that we need very few people with the same
variant to say that if they all have breast cancer by, you know, 89, then it's probably
a pathogenic variant, and if none of them do, if three people don't have breast cancer
by 89, then it's probably, you know, not one of the pathogenic genes because we're talking
about genes that we know what the gene does, we want to know what certain variants in that
gene do. The most important genotypes are the ones that we know cause disease. We want
to know for each variant in that gene does it cause disease or not, and actually you
don't have to throw a lot of people at each single variant to get that information.
Male Speaker: I think there can be a lot of different phenotypes
associated with different variants --
Female Speaker: Sure, sure, and I think that's an interesting
question, but I think that the idea for individual rare-ish variants, that you don't need a ton
of data.
Male Speaker: Right.
Male Speaker: Okay, so we're just a little bit behind schedule,
but it's been a very rich discussion, and I think that it's exactly squarely in the
sights of what eMERGE should be doing and taking advantage of opportunities that only
this network has. So we'll --
Female Speaker: Can we just ask if there are burning things
that people--
Male Speaker: Okay, is anything burning out there? [laughs]
Okay, well --
Male Speaker: So let me get started -- Neil Risch either
is not on the phone, or hasn't had the chance to say anything, but has been active in the
chat part of the box, and for those of you haven't looked at it, you should, but he makes
the point that there will be, between Kaiser, VA, MVP, and UK Biobank, let alone eMERGE,
there will be, you know, 800,000 or a million GWAS subjects soon, and so, is there a role
for further GWAS in this space. I will say that I don't think U.K. Biobank has GWAS data
yet. They're going to have the Biobank ChIP, and that's going to be late 2015, but be that
as it may, these resources are getting very, very large. And so is Neal on the phone?
Neil Risch: Yeah, can you hear me? I was muted before.
Male Speaker: Yes, we can hear you.
Neil Risch: Yeah, so, as some of you know, at Kaiser we
had a grand opportunity where we did GWAS on -- we have about 104,000 individuals, GWAS,
a multi-ethnical cohort. It's an adult cohort, and of course we have very extensive electronic
health records connected to all these peoples. It's longitudinal over 20-plus years. Mike
Gaziano is on the phone; he can talk about what the VA Million Veterans Program is doing,
but of course they have extensive electronic health record data also. And the U.K. Biobank,
I think they've already genotypes over 100,000 individuals. They are going to be genotyping
500,000 total over the next, I think, about two years. Currently, they don't link -- they
cannot link to electronic health data in the British healthcare system, but there are certain
things they link to. They're also doing a lot of their own phenotyping. They are doing
imaging on 100,000 individuals, they are doing lab tasks on everybody, and I think they have
extensive plans for doing more phenotyping.
So I guess my point just being that in terms of the balance between -- which was the topic
here -- the balance between discovery versus implementation, it seems to be, going forward,
if I had to characterize what I would see as a big strength in terms of eMERGE is that
you have a lot of different health delivery systems linked to EHRs, you know, in this
network, and the question is how -- you know, you can address which others cannot how this
is going to be rolled out in different settings, how genomic medicine is going to be rolled
out in these different settings, because then the size of the system really doesn't matter.
It's really the variation across the systems that matters.
So, actually, this is the way discussion has been going anyway. A lot of the latter discussion
has been about implementation, which seemed -- to me, seemed to be appropriate.
Male Speaker: Okay, good, excellent. So what -- so we are
a little bit behind, but, again, an excellent discussion. And what we'll do is attempt to
do a 10-minute break, even though that's never been reported in the history of science.
[laughter]
And so please leave your desktops open and your phones muted. Don't go offline so that
you have to log back in. And we will plan to begin again -- the presentations right
-- well, we'll do it right at 30 minutes past the hour, whatever the hour might be in your
time zone.
[laughter]
So we'll talk to you then.