Tip:
Highlight text to annotate it
X
Male Speaker: -- and pass the time to Chris.
Chris Chute: Thank you. If we could have the questions
come up. Basically, what I did is to clone the question slides that were in Josh's presentation
-- next slide, please -- and these are the questions that had emerged -- next slide -- and
these are more less his answers. So, I think he's explained these by -- the points to emphasize
are, how do we really make the process better and faster, and that synergizes very closely
with Dan's point of how do we make this a scalable environment? I mean, to what degree
can we engage academic medical centers, or, for that matter, medical centers, academic
or otherwise? To have a very low barrier of entry of sharing this -- it really comes down
to the big data sharing problem that NIH is looking at very carefully, as we all know,
but I think Josh did get on many of the issues. Many of you who know me will know that standards
and extensible methods are my middle name, and the whole notion of the EMR, which is
an evolving target, culturally and technically, in most of our organization is actually having
increasing capabilities. Next slide.
I tried to synthesize Dan's questions, so I -- this ultimate reductionist model that
-- forgive me. So, we've already talked about what would it take to scale, and I was very
intrigued with Dan's mention of the Lego pieces because, of course, this reflects the clinical
element of modeling world, and the whole clinical information modeling initiative that Stan
Huff is leading. And then the creation, if you will, of modular data elements that can
be aggregated into modular observations. So the notion of hypertension as a really computable
observation, as a component of more complex phenotypes, or any permutations as you said.
And then, of course, you went into the decision support problem, and you explicitly included
curly braces. I don't know if you noticed that.
[laughter]
At least in the decision support world the old joke was decision support is great as
long as you can solve the curly braces problem, which means a miracle occurs here. You map
your own local organization's information to whatever the author of that decision support
role had intended, and this was, of course, dumped through divine intervention.
So we have panelists that I don't want overlook in this discussion that were named and designated,
and those are -- if speakers -- if participants don't mind, we'll turn to the panelists first
for further discussion and then open it to general discussion. They're Gerardo Heiss,
Stan Huff, and Zack Gawaney [spelled phonetically]. So why don't we take them in that order. Gerardo,
are you on the line?
Gerardo Heiss: I am and just unmuted myself. Thank you. Fascinating,
thank you. I want to perhaps point to one -- many points, excellent points have been
raised. One that intrigued me in particular is the notion that at what point are we ready
to cross the -- that watershed between phenotype development toward implementation in the sense
of readiness? As these algorithms have been developed, they have done a superb job in
supporting physician ascertation [spelled phonetically] studies and discovery. The -- at
that point I think we have evidence that they do work, and the group has done a superb job.
As we approach the question of implementation, in fact, aren't [spelled phonetically] already
jumping into it, what are the costs of failing to appropriately optimize, say, predicted
value or something like that? Is that sufficient to support technical decision making? Is it
sufficient to support reporting to participants and so forth? In other words, what is the
cost of a less-than-perfect validity when it comes to these EHR-based, record-based
phenotypes that perform so well when it comes to ascertation studies, perhaps possibly reducing
a little bit the power, the a physical power at that level. But do they -- are they -- what
assurances do we have to have in place to move toward implementation and the next application
[unintelligible]? Things that we're approaching a very [unintelligible] a general sense and
that perhaps requires a little more thought in terms of specificity. That's all I have
to say.
Chris Chute:Thank you very much. Stan?
Stan Huff: Some of this probably goes back to the previous
discussion that I -- and I actually have more questions than I have conclusions or -- we're
working, and, you know, in terms of making a lot of these things actual practical and
workable, for instance, the commercial labs that we're ordering our genetic tests from
aren't set up to send back coded and structured data so that we can incorporate it in a computable
form in our record. What we typically get back electronically is something that's says,
"You printed a report." It's faxed to us as a written -- you know, as an image document
rather than even as a partial [inaudible]. And I don't --
Gerardo Heiss: There's a --
Chris Chute: I think he should continue.
Stan Huff: Okay.
Chris Chute: Stan? We'll ask others to mute until Stan
completes.
Stan Huff:And another part of that, again just sort of practical plumbing kind of things,
are improving our test ordering systems so that we get appropriate family history and
historical data and clinical data available to the laboratories when they do their testing
so that you get better testing. And then another thing I think is interesting to think about
is -- and, you know, there is a motivation and there have been pre-agreement by this
group about sharing of data, and when I get back to my own sort of state and institution,
it's clear that this kind of sharing of data is going to improve research and also going
to improve patient care. What's not clear or seems to be a problem in that sense is
that even though that's true, the local politics of competition between healthcare organizations
and other things aren't conducive to people actually sharing data, even though it's, you
know, at some level, everybody understands this best. And I wonder again what people
have thought or seen, or ways to create better incentives for people to actually do the data
sharing within a community.
Chris Chute: And let's move to Zack if we can. Zack, are
you there? Can we go to the last slide? Last slide, yes. I -- this is my simplistic summary
and points of emphasis, if we're -- no, we can't, Zack.
[laughter]
I just saw your message pop up. But what degree of clinical normalization is needed, and will
meaningful use standards help us -- oh, there you are, Zack. Let me just finish this, and
then we'll let you continue.
And then finally, culturally, within an organization, if we're going to achieve the kinds of scalability
and low cost for phenotyping, organizations and medical centers -- and this is starting
to happen to accountable care organizations, thank heavens -- need to think of clinical
data as a primary resource, whereas right now, it's almost a byproduct of the care process.
And concerns about comparability consistency, or worse, any mode of reusability, are second
or third tier in the context of many organizations. But, Zack, carry on.
Zack Gawaney: Yes, well, first of all, amen to what you
just said, brother.
[laughter]
That's exactly right. And it actually dovetails exactly with what I was going to say. So,
I think the point that was just made about the laboratory vendors is valid. But it's
dwarfed by the problem of the electronic health records system vendors are not particularly
exercised to either represent the family history or genomic data in the electronic health record
systems, and I will not name the individuals in our group who reported that their vendors,
their commercial vendors, were now a couple years behind what they had promised they would
be -- where they would be when they first sold them these systems, but to genomics.
But, here's my real point. My point is -- it comes back to us, and it comes back to Chris's
point. If our clinical leadership does not make this genomic and genetic decision support
part of their primary criteria for selecting electronic health record systems, we will
be waiting many, many years. And no matter what we do here in eMERGE, we'll be working
at the margins if our clinical leadership does not make that selection feature a primary
selection metric in choosing a vendor. And I think we just have to recognize that. Without
that, we will be working at the margins.
Chris Chute: All right. Thank you, Zack. So, now we have
time -- we still have time, don't we -- for general discussion. Any comments from panelists,
or presenters, or anyone else?
Marc Williams:Yeah, this is Marc Williams. A comment and a question. A comment to relate
to clinical decision support, and I think looking at Dan's sort of five versions of
this, I just want to let the group know that there are actually some ongoing discussions
in a couple of those areas. The Infobutton project that you've already heard about, which
would be a way to represent educational information related to genomics, is interacting with the
open Infobutton standards group and others to develop hopefully a generalizable solution
that can represent genetic and genomic data. So that's number one.
Number two is that the clinical decision support consortium which had been funded through AHRQ,
whose funding ended last year, has been resurrected in version 2.0 with Blackford Middleton, the
original PI leading this effort. Blackford is now at Vanderbilt, and they have been very
interested and have reached out to the electronic health record integration group about using
some of our eMERGE PGx use cases in clinical decision support that we're building as exemplars
for their work. So we do have some nascent efforts in that area.
The question that I have comes back to the Lego model, and I'm wondering if anyone has,
or if it would be possible to, define a set of, you know, basic phenotypes that we would
agree could be assembled to answer some substantial proportion of clinical questions. If that
is possible to do, then that could very clearly lay out prioritization of phenotypes that
we would want to be able to do, that would then allow the reassembly of the phenotypes
in this modular way to answer a ton of clinical questions.
Chris Chute: Okay, Peggy, you want to answer that or [unintelligible].
Male Speaker: Yeah, with respect to the --
[laughter]
-- modular phenotypes, actually Luke Rasmussen has a paper under review apropos of what you
were saying, applying software design patterns to phenotype design patterns and showing that
there are a finite number of repeatable patterns that compose our phenotypes. So, that suggests
that the modularization is actually quite tractable. The --
Chris Chute: And just to --
Male Speaker: One other thing I wanted --
Chris Chute: Oh, I'm sorry.
Male Speaker: -- to comment on was with respect to the phenotypes
and clinical application, one issue that we've run into a lot in eMERGE-I and eMERGE-II with
respect to phenotypes is that the definite yeses and the definite noes are often a minority
of the total patients, so Dan's comment about when do you collect more information, I think
when we go to clinical applicability will be critically important, and figuring out
how to structure those questions and how to build hooks into our EHRs so that our decision
support, instead of being "do this" becomes "ask this," will be, I think, a huge step
forward.
Chris Chute: Peggy?
Peggy Peissig: Hello?
Male Speaker: If you think about the number of phenotypes,
there are certain phenotypes that have been reused, and the entire data set has been labeled
with like diabetes -- type II diabetes [inaudible], so there are some examples of that already
in play that you could imagine being extended for sure.
Reed Pyeritz: This is Reed Pyeritz, can you hear me?
Chris Chute: Yes.
Reed Pyeritz: Yeah, so there's a paper in this month's Genetics
in Medicine that speaks to both phenotype ascertainment and decision support. It's by
Scott Grosse and his group, looking at insurance data to identify individuals in large health
systems that have individual features of a pleiotropic condition, hereditary hemorrhagic
pionjectasia [spelled phonetically]. And by looking at individuals who have more than
one of the phenotypic features but without an underlying diagnosis suggests this condition
is markedly underdiagnosed, which may be a good clinical point, but it suggests a decision
support tool that could be embedded, number one, but also another way of getting at where
phenotypes in all of our databases.
Chris Chute:Peggy? Anyone else? We still have about three minutes.
Male Speaker: Following up on Reed's comment on a little
thread that's going on in a chat room, I think eMERGE is, I hate to use the word "uniquely,"
but eMERGE and resources that couple very large EMRs to extensive DNA and genotypes
have the potential to discover pleiotropy of the type that you described, Reed. And
I think that is an opportunity for us --
Male Speaker:I need to fix this.
Male Speaker: Somebody else is talking over me.
Male Speaker: And -- so I would just reiterate the idea
that starting to think about what variants that have minor allele frequencies of .1 or
.5 percent do in a general population is something that we are well suited for and ought to be
strongly considered [inaudible].
Gerardo Heiss: I have a question for Zack. I think if the
selection of medical records vendors has already taken place, visiting [spelled phonetically]
that is going to be difficult. and I know you've spoken at times about these apps, which
I don't know all the technical details, but at Mayo we have an EMR like that which is
basically a g-centricity [spelled phonetically] on which our -- applied these apps and we
found it very useful, and I wondered whether you could comment on that being a solution
to this rather difficult problem. Or others in the room that can --
Chris Chute: You're muted again, Zack.
Zack Gawaney: Can you hear me? Can you hear me?
Male Speaker: Yes, we can now.
Male Speaker: Yes, we can.
Zack Gawaney: So, there's a two-part answer. I still believe
that the absolution that I talk about is a solution and I -- we can go long on that some
other time, although we did have a webinar recently where we told you more about that.
Nonetheless, I actually believe the fundamental solution [inaudible], and here is a .9 for
eMERGE-III. And that is, I think it would be [inaudible] interesting [inaudible] eMERGE
participants and clinical leadership [inaudible] understand how their leadership position can
be improved by talking to their vendors as a group, and -- because I think otherwise
they don't really understand the lost opportunity. And many of them are embracing this notion
of the precision medicine that's driven by molecular characteristics, and they don't
understand that they're not driving their fundamental tools of decision support, a.k.a.
the vendors, of electronic health records systems to actually -- to that point.
So I would argue strongly in favor of a meeting where we could actually try to address this
issue not with the vendors, but with our clinical leadership.
Male Speaker: I'd put that maybe on an NIH "to do" list
and their possibility wishlist. So I agree strongly, Zack, that nothing that could be
more productive I think in --
Female Speaker:Could you repeat the critical issues because we are having trouble hearing
you --
Male Speaker: You were breaking up significantly. From what
I gather, Zack is suggesting that we assemble the clinical leaders of major academic and
perhaps non-academic medical centers to meet with vendors so that the value proposition
of treating clinical data as a primary resource can be articulated and can be shared and impressed
upon the vendors. And we would be in a very different world if we weren't shadow boxing
with our electronic medical records as many of us are, trying to sort of squeeze data
out of that [spelled phonetically].
Chris Chute: Okay. Well, that's been a very --
Zack Gawaney: Chris, can you hear me better now?
Chris Chute: Yes, we can.
Zack Gawaney: Okay, so it's actually [inaudible] you trying
to channel me, and you almost had it right. What I was suggesting was not that we meet
with the electronic -- our clinical leaders should meet with electronic health records
vendors. I was suggesting that we, the leadership of eMERGE, clinical leaders [inaudible] institutions
and explain to them why and what they should be demanding of electronic health record vendors.
I don't think that you can understand that right now.
Male Speaker: [inaudible]
Zack Gawaney: I'm suggesting a joint meeting between us
and our clinical leaders.
Chris Chute: Yes, that's an important distinction.
Teri Manolio: And explain what they should be demanding
of medical records. Yes.
Male Speaker: Of the records, yes.
Dan Roden: Teri, this is Dan Roden. Teri, this is Dan
Roden. This sounds like an opportunity for the Genomic Medicine Working Group.
Teri Manolio: Yeah, that's right.
Zack Gawaney: So, yes, so all I can say is I know for a
fact that when you actually talk to the leadership of these clinical -- large clinical academic
centers, they understand that this is a priority in terms of where they hope to go, and they
don't understand how they have not made this a priority for their vendors. They haven't
closed that synapse. And I think by having them meet with us, their own leaders in the
same space, collectively in a group work job [spelled phonetically], we could make that
abundantly clear and turn that into a collective action item for them.
Male Speaker:I agree.
Female Speaker: I think if you do meet with them, I think
that's a great suggestion, and it'd be really important to also ask them what kind of issues
are facing them first for implementation? What do they want to implement? So we're hearing
from Group Health they're not that interested in a lot of the pharmacogenetics that isn't
really evidence-based right now. They are much more invested in polypenetrance things
[spelled phonetically] for implementation because they see that as something that's
absolute right now. So I think it's really important to understand what the health systems
want from genetics and in what order. What are they --
Zack Gawaney: So that's great. So instead of making it into
a browbeating session, we turn it into a "what's important in this space," and how do we use
our electronic health record systems to operationalize that, but we don't make it a technical discussion,
we make it how do we [inaudible] priorities into decisions at the C-level suite. And I
think if we had the right participants, it would be a high-profile meeting. It would
be very interesting for all of us, and I think it really would move our agenda forward. And
I think if we are interested in this, we can have an email discussion about who should
be invited.
Male Speaker:Can I make a comment here? One thing you haven't discussed which I think
also is important in this context is the economics, because, for our medical center, I think it's
probably true for many, genetics and genomics is looked upon as a money loser. And so I
think their incentivization on this topic is going to depend on painting an image for
them about how in the future this is going to have financial and -- positive financial
impacts.
Zack Gawaney: Well, I think it's an important message for
us to hear by direction. If there is really no economic value, which I doubt is true,
then we should hear it.
Male Speaker: Well, I just tell you things on our own experience
of our medical center. I think of many others so far, you know, supporting genetics and
is basically -- has been a money loser, primarily because there are procedures involved. We're
trying to work on this in some fashion to, you know, to at least you know -- we refer
for procedures but we don't get the income from that. So, I think there [inaudible] many
medical centers this could potentially be an issue. Things may change as more -- as
insurance, but that's why, you know, the insurance is important in this context also, because
as they cover more and more genetic services, things could change. But there are not too
many models out there where genetics is really a money-making business.
Chris Chute: Okay, so what we're going to do is, in order
to stay on the agenda, we actually have a clear recommendation now to this meeting about
convening a forum to discuss these issues. So, we'll close this segment on phenotypes
and move on now to the EMR and genomic discovery, and for this topic, the eMERGE presenter is
Marylyn Ritchie. Marylyn, you have the floor.