David Mrazek:
Well, thank you very much for the invitation to come talk with you. I think one of the reasons that I was invited
was to tell you the story of what we’ve actually been doing at the Mayo Clinic related
to pharmacogenomic testing in the Department of Psychiatry. I have some relationships you should know
about. I’m primarily a full-time employee of the
Mayo Clinic.
However, the Mayo Clinic has a small equity
interest in a company that is named AssureRx Health. And AssureRx Health has license to intellectual
property that’s owned by the Mayo Clinic. I’m one of the inventors of this intellectual
property. And what AssureRx Health has done is built
a web-based electronic interface to support pharmacogenomic decision making in clinical
practice of psychiatry. My objectives today are pretty straightforward. I want to tell you about what has been happening
at the Mayo Clinic and what we are doing now. And I want to talk a little bit about this
question of clinical utility and put forward my own thoughts about how to move forward. Well, the story of psychiatric pharmacogenomic
testing really begins in about 2001. We started a series of clinical studies. And at that time we focused on cytochrome
P450 2D6. That was our only informative gene. And we found many associations between psychiatric
medication response and metabolic capacity. So in February of 2003, the clinical laboratory
of Mayo Clinic decided to make this test available. And we worked for a year to see how it settled
in our practice.
And, by that time, there was a sense that
other cytochrome P450 genes were also important. So, 2C19 arrived and 2C9 arrived. And in April 2003, the Mayo Medical Laboratories
made these tests available to clinicians everywhere, anywhere who had really contracted with Mayo
Medical Laboratories to be a reference lab. So that was a very important change. Now, as you’ve heard from many different
perspectives, implementation is a slow process. And it took a couple of years before the faculty
in the department, before the residents in the department became comfortable with this
new approach to adding information into the process of making a decision. But by 2006, the practice of ordering individual
genes had really begun. And I’ll show you a little bit of information
about that.
Now there’s another shift which occurred
in 2009, and that was the point where we stopped ordering individual genotypes and started
ordering an algorithmic profile of genes. And I’ll tell you more about that. Now I’m grateful to the organizers to have
done some polling ahead of this meeting. And, again, you’ll remember this particular
scenario. Consider CYP450 profile, 43-year-old Caucasian
woman is being considered for an SSRI. Should this variant be routinely used? Well, this did catch me a little off guard,
I must admit. First off, I was surprised that more people
didn’t know. Only 31 percent claimed not to know, as a
very definitive, intelligent, well-prepared group.
But there were 16 percent who had a strong
opinion it would not be routinely used. So what does that mean? Well, let’s deconstruct the question. Related questions — what is a CYP450 profile? I’d argue that there really is none today. You could create one, but that’s not what
we do. Important question, why is the woman receiving
an SSRI? Is she depressed? Does she have hot flashes? If we consider depression, does she have a
treatment-resistant depression or a new depression that has just begun? And then the question, what variance, singular,
is being considered? Well, we’ll see about that. And then what does “routinely” mean? Does that mean that every single patient has
the test, or does it only mean that patients where it’s clinically indicated would have
the test? Well, again, that’s a little ambiguous.
Well, this is what we do now. We have a algorithmic profile. And we don’t do 2D6 testing. We do these seven genes. And, of course, we don’t do one variant. We do more than 30 variants. And that’s an important concept, because
if you’re back in the mode of, well, one gene variant influenced clinical practice,
you’re not thinking about what we have available to us today to guide treatment. But let’s make it a little more interesting. What about just 2D6? Let’s go turn the clock back in time. It’s a different question than is a profile
useful, but is 2D6 information actionable? Well, Dr. Weinshilboum in 2003 put forward
some information in the New England Journal that would argue at least that there was strong
pharmacogenetic relationships. And this famous pharmacogenetic profile, looking
at different levels of metabolic capacity, which was really compiled 18 years ago, shows
a lot of difference in the area under the curve of nortriptyline depending on how many
active 2D6 drugs one has.
Again, there are also adverse effects. A patient died taking doxepin. Again, unknown metabolic capacity. Originally, it was thought to be a cardiac
death. It was only at autopsy that it was a toxicology
problem. You can argue doxepin’s a pretty dangerous
drug. What about a safe drug like fluoxetine? Well, again, it’s a rare event, but in the
case of this trial, it was a needless death, poor metabolizer. It was not recognized. And what about actionability? Would you give a patient who had very poor
2D6 metabolic capacity Prozac, Paxil or Effexor? Well, I would not. Would you give them norpramin or tofranil? No, I don’t think so. Would you give this patient haldol or risperdal? Nope. Would you give this patient Strattera? I don’t think so. Would you give this patient codeine? Well, I hope not. Those are actions. What about the reverse? Would you give this patient Celexa, Lexapro
or Luvox? Well, yes, unless there was another contraindication. Would you give this patient Cimbalta or Pristiq? Yes, unless there was another indication. Would you give this patient Zyprexa or Geodon? Yes.
Would you give this patient Concerta? So there are options to choose. Now, I have given all of these drugs to patients
with poor 2D6 metabolic capacity because I didn’t know any better before 2000, but
I try not to these days. Now, what is this? This is a component of the algorithmic pharmacogenomic
report. I’m just going to show you two. I’m going to take a second to try to explain
to you that in addition to stating the genotype, in addition to stating the phenotype that
we would impute, there’s guidance to the clinician. And medication in that first group are antidepressants. And you see there are many. There are 18 in that first set of rectangles. And this is patient that has a very common
genotype, not a poor metabolizer, but someone with intermediate metabolism of 2D6 and 2Z19. And, again, the three categories are use as
directed, use with caution, and use with caution and with more frequent monitoring. So these medications can be used regardless
of the genotype, but the focus of the clinician should differ based on the genotype.
And, again, there are many choices for this
individual, some probably better than others. But now the picture changes when you have
a patient — there’s another patient that has impairment, but this impairment is really
inactivity of both of the key enzymes. And there you see there are many more medications
that are likely to be associated with side effects in this patient. And once you have this information, it’s
quite unlikely to start prescribing a medication that would be in that red rectangle. Well, again, the Pharmacogenomic Research
Network Group has been thinking about the issue, how do we facilitate clinical implementation
of pharmacogenomics.
And you heard quite a bit about that from
Howard. But one aspect of that is we tried to think
through what is the method by which one can establish the right amount of evidence. So this is a commentary that Dr. Lerman and
I wrote this summer. And it argues for considering a broadening
of the base of evidence. So, yes, randomized control trials are a wonderful
well-established gold standard for some things, but given the nature of questions that are
being addressed related to pharmacogenomic testing, related to specific drugs and a wide
range of different gene variants, we would argue that the development of pragmatic clinical
trials that are actually embedded within clinical treatment settings is a reasonable alternative
to provide additional information that is more feasibly obtainable. Now, again, there’s also other methods of
finding out information that will help guide a decision about adoption. And well said that if there’s something
related to safety, that really forecloses the process.
So, when there was a focus on a risk for Stevens-Johnson
with Carbamazepine, there were no randomized control trials. It was much more an issue of let’s be sure
we know which patients are at risk. Now we’ve tried to do retrospective studies. And during the era before the algorithm, we
reviewed 2,390 patients who were seen on just the consultation service. And, again, these are patients referred for
all sorts of reasons. And 19 percent were tested with at least one
genotype, 58 percent of those with serious treatment resistance to antidepressant treatment
were tested. We have two pragmatic clinical trials that
have been conducted, one at the Hamm Clinic, a proof of principle trial.
And there, to our surprise, despite the underpowered
nature of this preliminary trial, there was a difference in actual decrease in depressive
symptoms in patients whose clinicians had the information about their genotype from
the algorithm before starting treatment versus those who did not. And then in a somewhat larger sample with
a very similar design, again, in La Crosse, Wisconsin, and not a particularly sophisticated
but a very lovely place — and again, the finding was demonstrated that not only did
clinicians and patients find this to be helpful, but there was actually an outcome difference. Again, these are early days and these, you
know studies need to be replicated. They need to be replicated in much larger
samples.
But I think the direction is becoming clear. And my conclusion is that for us to really
move forward with implementation, we have to make the reports more understandable and
friendly and useful to clinicians. Again, we had experience with reports that
were genotypes and phenotypes. They weren’t perceived to be particularly
helpful. What I think now we need to do is to make
information available to clinicians that they can actually use. So I think my time just ended, so I’ll stop. [applause] Female Speaker:
Thank you. I think we have time just for a question or
two. Yeah. Howard:
Hello. It struck me that most of the decisions you’re
making were already being made for the exact same pathway in the context of drug interactions. So how do you look at the same pathway from
those two lenses? David Mrazek:
Howard, that’s a terrific point, and I think it actually does provide some insight into
the rate of adoption that psychiatrists were able to demonstrate here.
Because you’re entirely correct. We were thinking about, you know, the impairment
of pathways due to drug-drug interactions. And, of course, the correct response is to
put both of those features together. You need to know the genotype, and you need
to know the drug interaction. And you can create a phenocopy of a poor metabolizer
by simply inhibiting that particular enzyme. So, my response is that it’s additive and
perhaps synergistic as you have a more complete perspective of what are the factors affecting
drug response. Female Speaker:
Jonathan, did you have a question? Jonathan:
Yeah, just — and this sort of gets at a topic that was brought up in the last session about
should genetic professionals, you know, genetic counselors and medical geneticists be involved
in these discussions. And my feeling, and I was sort of chatting
with Howard about this, is, you know, thinking, say a few years from now, whenever the sort
of genetic information is already there and resident in some place in the medical record,
these types of things almost don’t need any genetic counselor involvement.
This is decision support matrices. This is embedded in the medical record that
is just-in-time information for physicians that are doing prescribing with information
there. So, I think this is a good example of how,
you know, once the genetic information is already in their medical record, it becomes
available when needed at times in the future when drugs are being prescribed. David Mrazek:
I think that’s right. Female Speaker:
Okay, one more. Male Speaker:
So, you have mentioned standards of evidence for the use of these tests. And looking at it from the clinical pathology
perspective, therapeutic drug monitoring is available. It’s not as readily available for these
drugs as it is — as functional testing, for example, for warfarin. But I’m wondering whether you’ve considered
incorporating measurement of drug levels, which is the hypothesized or posited method
of a relationship between the genetic metabolic — the change in — the genetic change in
metabolic enzymes and the outcome, that fundamental link of drug levels in the same patients in
whom your performing studies could be done.
And I’m wondering whether you’ve considered
doing it. Thanks. David Mrazek:
Yeah, it’s a good question. Basically you focused on what is the potential
synergistic role between therapeutic drug monitoring the level — monitoring the level
of medication in the patient versus genetic testing. And I do think that we would benefit in this
country by doing more therapeutic drug monitoring. And there are many reasons why. Certainly, it’s reassuring to see a correlation
between genotype and blood level. But that, again, may not add much value, but
there are other circumstances. For example, the patient has a good metabolic
profile and has a bad response to the medication, you raise some hypotheses.
It may be a target gene variant, or it may
be something as simple as not taking the medication. And therapeutic drug monitoring is a very
good way to identify that. And we see this fairly regularly due to serum
levels. Female Speaker:
Okay. One more quick question. Male Speaker:
In relation to your trials, which I understand are pragmatic, have you had problems persuading
clinicians to randomize at the point of care and problems with IRB approval of such work? David Mrazek:
Oh, okay, the question is related to our pragmatic clinical trials and what are the issue related
to the IRB and the issues related to potential for randomization. What we chose to do — well, first, the IRB
has been overseeing hundreds of genetic studies at the Mayo Clinic. So, the IRB is very knowledgeable about the
risk related to genetic testing.
So we did not have a problem with a genetic
testing component. The way the studies were both designed, there
are two phases. So there is no randomization. Phase one is a period of treatment in the
standard practice, so without genetic testing, and monitoring the outcome of those patients. Phase two, the guided phases, clinicians are
given the profile that you saw at the onset of treatment and can use that information
in any way they choose to care for the patient.
One interesting observation that I did not
anticipate, and it relates to other topics that were discussed at this meeting, is how
valuable is genetic testing to the patient. And we have had some NIH trials where we have
worked very hard to recruit 700 patients and have to approach two for every one we get
and in some studies three for every one that signs up. In these two studies, because everyone eventually
got a genomic profile, we had one person that we approached who did not — one patient,
one subject who did not decide to participate in this particular trial. Female Speaker:
Great, thank you. David Mrazek:
Good, thank you..