Tip:
Highlight text to annotate it
X
DR. BARRETT: Thank you. We're going to hear from Dr. Bates next.
Presentation – David Bates DR. BATES: Thanks so much to the committee
for the opportunity to present to you. And I'll
note that the FDA's mission is to protect the
public with respect to food and drugs, and I
believe to do that effectively, it's going to be
essential for it to think very carefully about this
new electronic world because things have really very dramatically changed in the last five
years, as Dr. Wilkins just underscored. And I think
this may require some paradigm shifts in the ways
that we think about labeling going forward.
From the electronic health record perspective, I want to note that drug-drug
interactions have had a highly disproportionate effect on the ability to get people to use
electronic health records and decision support in
particular, and sometimes far too many drug-drug interactions have been displayed, resulting
in providers being unwilling 1 to use systems
altogether. Within electronic health records, I think
the two most important things are when to interrupt
providers, and you heard some about that just now,
and then what messages providers see. And it will
be important to think about how the label interacts
with what providers see so that the management instructions are really especially important,
as was underscored earlier.
I also would like to suggest that the electronic health record is going to be the
way that providers will be able to navigate the
future in which they're thinking about how fast somebody
is metabolizing this one drug and how they're doing
things with another drug. Without that, I think as
Dr. Juurlink pointed out, providers really have no
hope. It's just too complicated to negotiate the
world. So what I'm going to talk about, I'm going
to start with clinical decision support in general.
Then I'll talk about drug-drug interactions in
particular. I'll talk about the 1 current state of
warnings around drug-drug interactions. I'll talk about how they're actually
implemented; give you a few recommendations about
drug-drug interactions, both in terms of content or
which drug-drug interactions should be displayed, but also about management, how they should
be delivered because that turns out to be a very
important thing as well. And then I'll wrap up.
We published a paper a number of years ago called, Ten Commandments for Effective Clinical
Decision Support. And it turns out that if you
want to make a difference in convincing providers to behave differently, you have to follow
a number of these tenets or you just won't get to where
you want to go.
The first is that speed is everything. Providers are really in a hurry. If you are
trying to take them through some big monograph, they
just will not go.
Second, you want to anticipate people's needs and deliver in real time. And with this,
with drugs, it should be possible 1 to know what
medications somebody's prescribing and bring the
information that a provider might want right to
them. That goes together with fitting into the
user's work flow. It turns out that little things, like where
you set the default, keep the prescription or
cancel the prescription, have a very big impact on
what providers do. Physicians resist stopping, so
if you tell them to stop, even if they're doing
something that's really a bad idea, they often won't do it.
On the other hand, if you say, well, instead of stopping, "We'd like you to, say, prescribe
a little different dose of this medication,"
they're much more willing to do that. That's human
nature, but it's important to consider that.
Simple interventions work best. You can ask providers to provide additional information
on occasion, but if you do that too much, things
won't work.
It's absolutely critical to monitor what the impact of the decision support is. 1 In many
of the health records today, the tools to do that
had not previously been built in to enable that. And
because of meaningful use, that is a required thing going forward. It will be essential
for organizations to look and see how providers
are responding to warnings and then for us to
make iterative changes.
Finally, these knowledge-based systems have to be managed and maintained, as has been
noted repeatedly. The state of the art here is
constantly changing, and if we don't keep up with
that, we won't get to where we want to go. Now, how do things work in the real world
with respect to drug-drug interactions broadly? Well, most institutions get their knowledge,
the databases, from one of several vendors. And
you'll be hearing from Karl next, which is really
terrific. It's not practical for most organizations
to maintain these databases because they're very
complex. However, the fundamental problem so far
has been that for drug-drug 1 interactions in
particular, far too many warnings have been given.
And in addition, the way that the alerts have been
delivered is often suboptimal. Over-alerting has really perverse effects.
It can make systems very hard for providers to use
them, and organizations may even turn off decision
support altogether, which is undesirable because a
lot of the benefits from electronic health records
do come from decision support. So finally, both
content and management have considerable room for
improvement. Now, it is clear that drug-drug interactions
do cause harm, and much of the data for that comes
from Dr. Juurlink. So one example is glyburide and
clotrimazole, resulting in hypoglycemia, again a
very big odds ratio. And you heard before about
the clarithromycin example. I think that that evidence is some of the
best evidence about how harmful these interactions can be. However, if you look at the flip side
of things, drug-drug interactions are responsible
for a relatively low proportion of adverse 1 drug
events overall. It's about 5 percent in most studies.
And yet in many systems, they're responsible for a
lot of the alerts. So I feel like this is a place where there's
big opportunity for improvement. They clearly cause harm. Particularly if we could start
to take into account more factors, I think we could
do a lot better. But right now we have this scattergun
approach. It is possible to do better with medication12
related rules. We went through in our system, which is a big integrated delivery system,
and identified a highly selected set of drug alerts
for the outpatient setting.
One thing that we did was to make most of those alerts non-interruptive. When a non18
interruptive alert appears, mostly the provider can look at it, but they don't have to do
anything different. Only 29 percent in this study were
interruptive, and of the interruptive alerts, 67 percent were accepted. The industry standard
around this is around 5 percent. 1 So this is
considerably different than has been reported in
many other studies. In addition, it's quite clear that tiering
is valuable. We did a study in which we took advantage of a natural experiment to look
at this. We studied two academic medical centers, which
were using exactly the same knowledge base, which
was nice.
Site A used three tiers. So in tier 1, you basically could not give the interacting drugs
together. Tier 2 strongly suggested that you do
something different; that might include, for example, monitoring more carefully. And tier
3 was non-interruptive. Site B had all the alerts
as interruptive, which is the way that things
are done in many electronic health records today.
What we found was that 100 percent of the most severe warnings were accepted at site
A versus 34 percent at the non-tiered site. So what
that means is that 66 percent of the time at the
non22 tiered site, people were running stop signs
and giving even drugs that can result 1 in cardiac
arrest together, for example. So not what you want
to see. And furthermore, the overall alert acceptance
was much higher at the tiered site, 29 percent versus 10 percent.
We've done some work to try and look at human factors and alarms, and worked with
some groups that have a lot of experience around
alarms and warnings from other industries, like nuclear
power. And these results were published in JAMIA
in 2011. There are a few principles. One is that
you need uniform alerting mechanisms and then standardized alarm responses. Second, alarm
philosophies should minimize the number of false
alerts that occur. Third, the placement of alerts has a big
effect on the likelihood that users will actually see the alerts. Visibility is critical. The
font size has to be big enough so that things are
readily legible. And the visual alerts need to be
prioritized. In addition, color should be used to
help cue the user about the level 1 of a specific alert, and the number of colors that you use
should be minimized. Often systems today don't do
that. In addition, to make visual alerts more
distinct, it's important to minimize the number of
visual features that are shared between alerts. Again, in many systems today, all the alerts
look exactly alike and you have to look at the
textual information to know what the message is. And
finally, the text-based information should be
succinct. We then took these principles and then
superimposed them on actual electronic records, and
looked to see what happened. And in this study we
looked at 51,000 drug-drug interaction alerts. Providers accepted only 1.4 percent of the
non-interruptive alerts. For the interruptive alerts, user acceptance
correlated positively with how often the alerts appeared; what the quality of the display
was -- the odds ratio there is 4.75, so pretty large; the alert level. In addition, alert
acceptance was higher in inpatients, 1 who tend to be
a little sicker, and also for drugs with dose3 dependent toxicity.
The textual information did influence the reaction, so providers were more likely to
modify their prescription if the message contained
detailed advice on how to manage the DDI. And
again, that has obvious implications with respect
to labeling. Here is just an example of a drug-drug
interaction, level 2. The patient here is getting
trimethoprim/sulfamethoxazole. There's a very succinct message, and the provider has to
then make a choice about what to do.
So how are organizations actually doing? Well, we worked with a group led by Jane Metzger
to study a number of hospitals around the country
and to see what they actually had in place with
respect to drug-drug interactions, among other things.
The way that this worked is we developed basically a computer-entry flight simulator.
People were given simulated patients, and then they
put in some orders that were errant 1 orders, and we
looked to see how often they were actually caught.
For drug-drug interactions, they were caught 52 percent of the time. So about half the
time, even important interactions just went right
by. In this study overall, there were
62 hospitals that voluntarily participated. Simulation overall detected only 53 percent
of the orders that would have been fatal, not a very
good performance. And they detected only between
10 and 82 percent of orders, which would have caused
serious adverse drug events. Notably, there was almost no relationship
with vendor. So this slide shows the relationship with vendor, and you can see that every vendor
had sites with very good performance; every vendor
had sites with poor performance. This, from my
perspective, argues for doing some post19 implementation testing because it's really
what the organizations put in place and not just
what vendor system they use.
In terms of which alerts, we made some suggestions about how to move forward. 1 Those
were, interrupt with only the most important warnings,
and then tier. The jury is still out regarding whether it's even useful to display the non5
interruptive warnings. Valuable to have regular review.
It's essential to track how providers are responding. As practices change, new information
becomes available. Sometimes you begin using drugs
together that were not okay to use together previously. And sharing regarding this would
help. We argued in this particular paper that this
would be a common good. Reference to the RAND work, which Dr. Wilkins mentioned before,
as a good start. This is the sort of thing that could
actually be international because the issues are
not really any different in other countries, and
every country is struggling with this. In terms of how to deliver, the key
recommendations are to follow the human factors principles. So you should tier. You should
have uniform display. Where you display suggestions
is important. Different levels 1 of warning should
appear different. You should use color wisely. And the textual information should be succinct.
I'm going to go through this very quickly because again, Dr. Wilkins talked about this.
We did the work that she described earlier. But
we basically, together with RAND, did this work
in which we identified 15 drug-drug interactions,
which should not be given together. Here are a few
examples. Some of the things that we did not include
as interactions were things like abatacept and
tumor necrosis factor inhibitors, which were felt
to be more therapeutic duplication than drug-drug interactions. And many of the people in this
room, I'll note, participated in that work, and
we're really grateful to them.
At the end of the day, as was mentioned, we ended up with 15 drug class pairs, which should
never be co-prescribed. We think they're candidates for hard stop alerts. We're not
sure that this is a complete list, but this represented,
we believe, the best available 1 consensus. I want to note that the less significant
drug-drug interactions are still very significant. They're much more prevalent. They probably
cause much more harm. Most of the warfarin interactions
fall into that category. But many of those tend to
depend on patient characteristics and drug dosages
and timing and concomitant conditions like hypokalemia, and our ability to deal with
all of that so far has been limited.
We recommended that to improve the sensitivity and specificity of these, we need
more investment and evidence review and generation,
and then methods to make drug-drug interactions
conditional on other patient data, which typically has not been done in most systems. But I'm
sure the panel will discuss that more later today.
With low priority drug-drug interactions, I
think that's also a helpful list, and I won't spend
more time on this. I do believe that a consortium to maintain this list will be helpful, and
I think this list is likely to be useful to organizations.
We're doing some work now to see 1 how much uptake
this actually gets. Another set of work which I wanted to
describe briefly relates to adherence to black box
warnings, and this is some work that we published in the Archives of Internal Medicine in 2006.
We identified all patients who had a 2002 black box warning. We found that when we did
this, 55 of the 95 warnings required clarification
to be computable. So another message to the FDA
is it'll be really helpful going forward is the black
box warnings are made computable from the beginning.
We studied 324,000 patients who were prescribed a medication. Of that, 10.4 percent
got a drug with a black box warning, so that's
not uncommon. Of the 1,107 who got a drug with
a drug-drug interaction warning, 36 percent
also got a contraindicated drug. So that comes up really
not infrequently. Overall, we found that the black box
warnings were often imprecise, and more precision would be valuable in making these things
computable. The violations appeared 1 frequently, and it would help a lot to have better assessment
of the actual level of risk in individual situations. Sometimes these were clinically
reasonable; other times they were probably not.
We also did a study more recently in which we looked at the marginal benefit of adding
black box warnings that we did not already include
in our clinical decision support system, and added
all the ones that were there that we had not included
previously. And we actually saw slightly higher nonadherence after doing this than before,
5.1 percent after, 4.8 percent before. The violations did decrease, though, for a
couple of very important categories, notably for
drug-drug interactions, and then also for drug
pregnancy checks. So overall, adding more of the
information that's in the black box warnings did
not improve adherence at all, but it did for a
couple of the really important subcategories. So to wrap up, I believe that checking for
drug-drug interactions can be highly beneficial, but I believe that there's a lot 1 of work
to do both around which alerts to display. I think that
having this consensus work is going to help greatly.
The RAND work is a good start. It doesn't take us through all the things that we want
to do. And Dan Malone, for example, is leading a
group to try and take us through some of the next steps
around that. We need best practices regarding both which
alerts, and sorting out how to share those would be
highly beneficial. We also need best practices regarding how to display them. Today drug-drug
interactions are a big problem in the clinical systems which don't follow best practices,
and that's many of the systems that are out there.
In addition, we need to leverage our systems to build the underlying evidence base, and
that has to be much more robust. I do personally think
we can use lots of the observational data. The
data, for example, from Canada have been very compelling
for me, and I think there'll be other opportunities to do that.
As we get broad electronic health record adoption, we should be able to have much bigger
data sets than we've had in the past, and it'll be
possible, for example, to link that with claims and
to see in much more detail what actually happens. So a few specific suggestions for the FDA
around this area. First, I would endorse the recommendations that Dr. Juurlink made before
about labeling. And I believe that it would be helpful
to include in the label both some very simple messages, but then also some more detailed,
because people want both things. But if you want to
make a difference, it's really important to get the
simple messages correct.
Regarding format and how to display this information, there aren't a lot of good data
that I could identify regarding which approaches
are best. But one of the nice things about information
technology is you can have your cake and eat it,
too, and it might be possible, for example, to both
have some forest plots and some tables and narrative and let people pick 1 what they
want to look at.
I'll note that data suggests that users only consult referential material about 2 percent
of the time. So it's an important role for the FDA
to get that right, I believe. On the other hand,
to make a difference, the short messages are important.
Finally, there are lots of complex situations which have come up today like multiple
drugs, interactions changing over time, and labeling clearly will need to evolve to address
that. That's a really tricky and complex matter. Thank you.
DR. BARRETT: We're going to hear from Dr. Matuszewski next.
Presentation – Karl Matuszewski DR. MATUSZEWSKI: First of all, I want to
thank the FDA for inviting me to present at this
committee meeting. I'm from First Databank. First
Databank is a drug knowledge database vendor, which
there are about five of those in the United States.
Three of them probably are responsible for about
to 90 percent of the use in 1 current clinical practice.
First Databank has been in existence for about 40 years. We have the bank in our name;
that was early on. We started the company with
pricing information, but it's nothing that Ben Bernanke
should get excited about. A subsidiary of Hearst
Corporation. Really, what First Databank does is it
provides the granularity for drug knowledge. So we
take a label, we take clinical evidence, and we put
it in relational tables, and that is then consumed
and used by EMR systems. It's used by pharmacy back benefit managers. It's used by insurance
companies. It's used in the ambulatory setting, in
the inpatient setting. So it's really providing what would be the knowledge that we hope drives
decision-making. You see what our goals are. Again, it's to
influence medication safety. So one of the vision
statements of First Databank, or FDB, is really to
have zero medication adverse events. That's our
vision. Now, we do that. We recently started
sponsoring some research in the area of clinical decision support; recently published an article
on sulfa antibiotic/non-antibiotic cross-reactivity.
So it's one of these things, even though the knowledge has been out there for a long time,
we still get calls to say, "Put back in those
cross9 allergies," even though there's no evidence
base to support them. A number of our staff belong
to a variety of national/international pharmacy
organizations that look at drug safety, and we very
much participate in those activities. So this is what I hope to cover today, a
quick overview of the complexity of clinical decision support and evidence review; talk
about a three-pronged approach we have in terms of
reducing alert fatigue; and finally, just touch upon
some patient parameters that would ideally increase
the specificity/sensitivity of the drug-drug
interaction alerts that are provided. So surveillance of evidence. When you look
at evidence, what we have is really 1 the sources of
evidence. So I would say the manufacturer labeling, the package insert, is very important.
A new drug comes out; often that's your only
source of information.
We have biomedical literature as it's constantly involving. We have clinical reviews.
We had MedWatches from the FDA. We have guidelines that are created by guideline specialty
organizations. Within all that evidence we have
the factors, other factors, that impact how that
works its way into clinical decision support. It
could be the simple constraints of an organization for time to take and implement some of the
knowledge. It's how it fits into the work flow.
It could be what the local practice patterns are.
We also have prescriber constraints, so how long has a prescriber been out in practice?
So I like to think that my highest knowledge level
was probably the day I took my pharmacy boards,
and it's just been a steady decrement since then.
(Laughter.) DR. MATUSZEWSKI: I 1 think I heard that
confirmed from one panel member. But it's probably
the same for physicians. We also have patient information. So the
more specific information that we can have about a
patient, the more likely we can provide the alert
that is appropriate for the prescriber at that
point in time. This is my evidence is in the eye of the
beholder. We have here a prosecutor. We have here
a defense attorney. I suspect if they looked at
the same pile of evidence, they would both equally
make strong cases for guilty or not guilty. We are faced, in the knowledge database
vendors space, with basically having to decide, is
the evidence sufficient for us to include in the
database, or is it inadequate in terms of it's not
quite ready for prime time? And we have to make
these decisions. So here you see -- this is not my staff.
It's not nine people. But I can tell you that often the decisions are not unanimous and
they're after great and lengthy deliberations. 1 And
at least our references, I think, are all still
available by Web link, so we don't make our sources
of judgment disappear after a while. This is the part of maintaining and the data
curation of a drug knowledge database. For the
drug-drug interaction space, we have three dedicated pharmacists who pretty much have
devoted their careers and their lives in the pursuit
of maintaining this database.
It is something that we -- in terms of the trigger events. So it's MedWatches. It's journal
publications. These are all part of our information capture system. And we have it
all computerized hen a label revision -- so we
have, I think, daily, probably about 10 label revisions
come in from CDER. Those are tracked, those are
dissected, and they go to the appropriate unit.
Besides drug-drug interactions, we also have dosing modules. We have side effects, indications
modules, and allergy modules. Then again, we have strict editorial
policies, timely review. So if a 1 new drug came out
today, that information would be incorporated in
our database tomorrow. So we have a weekly clinical data push to all the customers of
FDB. Some of our sources of information I also
mentioned, besides the biomedical literature, which
I think is really important, so if there's one
take-home message, the literature, if it can be
incorporated into the label, that's great because
that often is what defines current best clinical practice. We also are looking at some academic
metabolism and drug transport databases to get some
greater specificity in terms of some of the enzymatic pathways to improve our data.
Now, in terms of drug-drug interactions, FDB has been doing this since 1984 in a more
referential monograph type of information. This is
just the sleeve jacket from a hard copy of what we
have, 2,000 pages, 18 chapters based on major therapeutic areas. And of course, we consult
14 external advisory board members, often from
leading academic medical centers. And you can see
some of the sections and the information 1 that's
contained. So I like what I'm hearing in terms of the
greater granularity of information if somebody wants to dig into it; that info button and
being able to click into it. Dr. Bates mentioned
2 percent. I suspect that maybe in an academic teaching hospital, it's 2 percent, but that
in other venues it's probably much, much lower
in terms of having the time to go and read these
monographs and greater information. But it's available to individuals who use the knowledge
databases. So what exactly are we talking about in
terms of the severity levels? FDB has four severity levels, and the number 9 is the
miscellaneous, not really clinically significant. So the first three are of importance.
So severity level 1, that is the contraindicated drug pairs. And as the arrow
points, it's about 24 percent of the drug-drug interaction contraindications in our knowledge
base are what are contraindicated, don't use. And
this again comes from labeling, 1 from literature.
The majority are level 2s. Level 2s are the severe, or as I'm now leaning towards, the
series drug-drug interactions. These are the things
that you should avoid if you can, but often there
is no other therapeutic alternative, so these are
the ones you should use with great care.
Often these are filtered and the prescribing physician may not see these. And it's the
pharmacist who then deliberates -- is it worth the
phone call to the prescriber to offer him an
alternative, or should I just save some time and
just go ahead and override this? The severity level 3s are the moderate
interactions, and those are the ones that really
are -- keep an eye on this. Often in the inpatient environment, the patient is probably discharged
before, really, the effects of monitoring would
make this a safe choice. And there has to be that
transition then when the patient is continued in a
med rec standpoint; that if the drug-drug interaction adverse effect doesn't occur until
two or three weeks out, that that 1 indeed needs
to be followed up with the physician who's taking
care of the patient in the ambulatory environment.
Now, you can ask, with the override rate of drug-drug interaction, so is it alert fatigue
or can it be something a little bit more serious?
This is a recent paper from the Journal of Epilepsy
and Behavior, and I always used to think that -- I'm again a strong believer that there's
just way too much information out there for the
CPUs that we were born with to process all that
information. So here was a study that surveyed 500
neurologists -- these were all primarily board15 certified -- and asked them about four recent
MedWatches about anti-epileptic drugs and their
knowledge of those. As it turns out, 20 of them
did not recognize any of the four, and only 30 percent of those 500 neurologists recognized
all four of the warnings.
Now, to me this is just a sign that it's impossible for an individual, even with a
number of years of practice in their well-1 defined
specialty areas -- so these are drugs that they are
presumably prescribing quite a bit -- to keep up
with all the information that the FDA is looking at
in the biomedical literature. So this is a little bit about MedWatch
changes, so profile these for the last five years
with 2013 not yet being complete. You can see that
the pace is increasing. So FDA's been busy. And I
suspect that 2013 may be a banner year. Again, drug approvals are also going up in the last
three or four years compared to what they were in
the past.
So this is all information that a clinician out there ,whether it's in their narrow use
of the drugs they prescribe in their specialty or
a primary care physician who may see the whole
spectrum of drugs and indeed be dealing with drugs
that they've never prescribed initially and may not
really have in-depth knowledge about. So the phenomenon of alert management.
Again, I've seen the studies, 80 to 90 percent overrides in the drug-drug interaction 1 alerts.
So this is a three-pronged strategy that FDB
has undertaken in the last couple of years.
The first strategy is again fine-tuning the content. So we have all these drug-drug
interactions embedded, and one of the steps that we
take is again taking a hard look, where can we
tease out to create less alerts, perhaps downgrading what might be a contraindication
in the package insert into something that is indeed
a severity level 2, providing the characteristics
match it. Here, for instance, we have drug
interactions that are based on strength breakouts. So lower doses of certain drugs are unlikely
to cause an interaction. And then we have 75
of those that have been broken out. We have route
breakouts. Often, topical formulations that are
not systemically absorbed. There's no reason for
that drug then to interact with another drug that
does have systemic effects. Then finally, taking a hard look at the
class effects that are mentioned 1 in some package
inserts when it may not be appropriate to include
the entire class. So for instance, clopidogrel and
proton pump inhibitors, in terms of that interaction, at least in the literature, we
believe there's a difference between lansoprazole
and pantoprazole and have indicated that as a
moderate interaction, something to monitor.
Here's an example again, a further example of fine-tuning content, so selected macrolides
interacting with selected statins. And we see that
again the strength breakout of atorvastatin at less
than 20 milligrams is a severity level 3, whereas
with the other statins and at higher doses of
atorvastatin, we give that a contraindication. So
this level of granularity allows us to decrease what are contraindicated pairs that the clinician
would normally override. The second prong of the strategy to decrease
alert fatigue is a product that was released about
three years ago for FDB customers. There's about
100 institutions currently using that. And that's
the allowance for local customization. 1 It was
mentioned in ONC as an option. And really what
we're finding is that when a severity level is
changed, that there are some institutions that
don't like it, complain about it, and some that
again just would love to get rid of it. So this idea that one size, one alert, fits
every possible scenario, every single institution, whether an institution has monitors for all
their patients or an institution is a small rural
hospital that basically has a minimal amount of
equipment, we feel is not appropriate and should
allow for some local customization. So here's an example, a mockup of a
screenshot. It's very small, but the circle that I
have shows some quick easy buttons. So the ONC
high priority list. If an institution says, that's
really where we want to start, they press that
button, and then the ONC list is imported into
their contraindicated severity level 1 drug interactions. If they want to exclude the
low priority interactions, again they press that
button and those 1 will be excluded.
We see this phenomenon of lists being generated as probably continuing. Whether
that's a good idea or bad idea, I'm not 100 percent
sold on it because I think the day-to-day curation
of that knowledge is extremely important. And when
I see pairs in lists that say QT prolonging agent
against QT prolonging agent, that drives me crazy
because even within those nuances, those are not all
contraindications. Again, there's a number of institutions that
have done extensive customizations to our severity
level rankings, either upgrading them or downgrading them or completely deleting them.
Here's an example of another custom severity level. So I mentioned that we have really,
in essence, three levels. So we have custom levels
of a 5. So these drug pairs, for instance, would
be invisible to cardiologists, who theoretically
are dealing with these drugs all the time, but
would be visible to all other specialty prescribers.
The custom severity level even could be site-specific, so whether 1 it's ambulatory
or whether it's inpatient for where the alerts
would be triggered. And this is a Web-based tool.
So when new data flows in from FDB, the levels
that have been changed at the local level are not
impacted. Now, we think, with this sort of local
customization, there is a potential to look at
what's called crowdsourcing of information. So
what do academic medical centers with teaching programs -- what sort of customizations are
they making? What sort of changes are community
hospitals making? What are ambulatory clinics perhaps making in terms of local customization?
This is something that we're looking to share with individuals who use AlertSpace,
and I think it again further guides us in terms
of fine18 tuning our alert content for all the other
users of knowledge database vendor drug interactions.
So we have this feedback loop. We have in the past had information that EMR vendors
have supplied to us. So here are four institutions.
Here are their patterns of overrides 1 that they're
seeing and where alerts are accepted. That again feeds back into allowing us to
fine-tune our content. So we get reports back like
this. We're able to identify the specific drug
pairs that are involved. So seeing those are involved, seeing how often the rates are
overridden, much like in Dr. Bates' institution, the ones that are routinely overridden and
of less serious nature, these are the ones that we
can look at our content to find again whether there
is a dose adjustment or some sort of route adjustment.
Then finally, I want to talk about individual patient parameters. So the things
to consider in any drug-drug interaction alert,
is this a new exposure or is this a continued
therapy? So if a patient's been on a pair that's been
fine, the disease is controlled, they've been on
there five years, and just because they're now being
admitted for the first time and the drugs are being
ordered, alerts are going off. And then of course
the clinician says, "Well, this is what the patient's been on five years. 1 I'm not going
to change anything."
We have laboratory parameters. So if potassium is going to go up, if an INR is
going to change, it may not happen that day. It may
be appropriate therapy in an inpatient environment,
but it's something that if you could bring in those
lab values as it is being prescribed, perhaps it's
an alert that can be delayed. Number of physicians ordering meds. So is
it the same physician ordering the med who will be
aware of the interaction? Or is it somebody doing
a consult who may not be familiar or may not even
be aware that the precipitant med is on board? Service location is something to look at,
again whether it's a clinic or if it's an intensive
care unit, where again monitoring is pretty heavy,
versus in an ambulatory environment, where the
patient may not be seen for another six months or so. And then we talked about comorbidities,
renal or hepatic deficits, and then the pharmacogenomics aspect of the patient.
Then again, for the 1 physician, what specialty are they? Are they a specialist
or a generalist? And their role, is it a hospitalist
with years of practice experience versus an intern
or a resident who really doesn't have that much
experience and seen that many cases? Finally, in terms of the drug, what's the
probability of the reaction, the percent occurrence, the incidence, and the severity
or the serious nature of the event. We talked about
some standard symbols, whether it's a go for it,
caution, or a stop. Then again, finally, in terms of
implementation, which again is very specific to
the institution. So the knowledge bases have an
incredible amount of granular data about the drugs,
but how the institution, how the EMR vendor, decides to program against it and implement
it is probably key in terms of that scatter plot
that you saw of the override rates.
So who's looking at the alert? Is it the prescriber? So again, in some institutions,
those are just the severity level 1s, the 1 contraindicated
pairs. Is it at the dispensing point? So again, the pharmacist often is the one who's looking
at that. Or perhaps it's at the EMR level, so
the nurse administering the drug who also is now
looking and seeing some sort of interaction. What other modules are simultaneously
implemented at an institution? Because there is
some overlap. So pharmacodynamically, for duplicate therapy, is this a drug interaction
or is this duplicate therapy? You could say it should
be one or the other, but often institutions may
not have both modules turned on. And maybe getting
alerts from both modules, and that may again lead
to alert fatigue. Or is it a side effect rather than being one of the other modules?
Then finally, drug disease, contraindications and precautions. And some
of the work we're finding in there is if the problem
list is not well-maintained and updated, then that
sort of module will lead to tremendous problems.
So EMR hygiene and maintenance.
Then the user interface 1 again was touched upon. It's if I've seen this once, maybe I
don't need to see it every single time. Maybe I
need to see it every five times just so that I'm reminded
of this drug interaction. Perhaps I've already approved this combination in the patient,
so if I'm just changing the dose, maybe I don't need
to see it again.
Is there some symbolic coding that can be used? Is there a way to bundle alerts and
prioritize alerts? So again, depending on the EMR
system, you may just get a long list, and perhaps
the more serious alerts are buried towards the
bottom because they're in some alpha order or by
module. But really, the ideal way would be to
present the alerts that are going to harm the
patient right up front, whether you color-code them
or emphasize them or bold them. That is of ultimate importance that a knowledge vendor
like First Databank has really minimal control
over. Screen size viewing. So at some point,
we're in the era of iPads and perhaps 1 smartphones being involved in the e-prescribing and the
clinical decision-making, so how much of that real
estate can we get on board to make sure that drug5
drug interaction pairs are ultimately looked at and
decided appropriately? Maybe audio alerts is
another option, though I know in many institutions the lack of sound is not a problem; that they're
various alarms. So what are some of the issues that staff
at First Databank have with package inserts?
And we talked about a couple of them. Labeling mismatches
between two drugs. So a newer drug comes out, lists out a number of drug interactions. And
then you go to an older package insert of one of
those listed and it won't have it. So we adjudicate
that. Then there's the case of label
inconsistencies, so even within one label, and
I'll share an example in my next slide. Imprecise label narrative. So if something
says, "Use these two drugs with caution," to me as
a pharmacist, you should use 1 every drug with
caution. That word has very little meaning. Outdated labels. So again, labels that are
just too old to be of any use at all. Then finally, broad class effect statements
within labeling, so this entire class interacts with this agent. Not very helpful because
some knowledge database vendors may just apply
it to the entire class. We try and slice it up as much
as possible as evidence allows.
So here's my Xenazine, tetrabenazine. It's used for Huntington's chorea. So here is a
label with various sections. So if I just read the
highlights section, "Do not prescribe," that to me
is a pretty strong statement. That's almost like a
severity level 1 contraindication. You read Section 5.11, and you've got,
"Should be avoided in combination" for QT prolonging agents. That to me sounds a little
bit like a severity level 2 interaction; avoid
it, maybe look for better therapeutic alternatives,
but it's not a "Do not prescribe."
Then you read Section 7.5 1 in that same PI, and what you read is, "Causes a small increase
in QTC prolongation," so 8 milliseconds. You
read that; it's the only thing you read, so that's
not too bad. That sounds like severity level 3,
maybe, just monitor and use cautiously. And there
I go, using that word "use cautiously" with every
drug. Then Section 7.6 says, "May be exaggerated
by concomitant use" of various other QT prolonging
agents. So when we look at a label like that,
there's a judgment that needs to be applied to
that. And in this case, it's a severity level 3
unless the other precipitant drug is a strong QT
prolonger. And those are the sort of judgments that have to be made every single day, every
single label.
Another recommendation I'd make to the FDA is, make sure the manufacturer has the label
on your site. So this is again Xenazine, and
"label not available." So I think you have the regulatory
might to make sure that if somebody's got a product
that's out there and dispensable, 1 the label should
be available to look up. In summary, it's not easy. It's not always
fun. But I think as David pointed out, that's why
there's probably only about five companies that are
doing this -- and you should not try this at home
unless you have vast, extensive resources and
pharmacy staff that you can apply to this every
single day; and that at least the three-pronged approach, while it is not guaranteed 100 percent
success, I think it's at least moving the bar, so
local customization, fine-tuning of content, and
then also adding more patient-specific parameters, which we hope to be able to do in the next
few years to decrease alerts.
Then finally, the evolving evidence database is again -- I don't think the label will ever
keep up with what's available in clinical practice.
And those are the things that my staff looks at
and incorporates into the knowledge base, as other
knowledge bases do also, and that I think is
important to providing the clinician with the best
evidence and information they have 1 for prescribing these drugs that interact safely. So thank
you. Clarifying Questions
DR. BARRETT: We're going to have some clarifying questions now. And again, I would
remind all of you to state your name before you
make your point. Marilyn? DR. MORRIS: Marilyn Morris. I wanted to
ask Dr. Wilkins a clarifying question. You talked
about certification of the various patient record
systems. And I was wondering, what does this mean
with regards to looking at DDI information in the
systems? DR. WILKINS: Sure. Thank you. So the
certification of the EHR product, there are functionalities that the EHR vendor has to
have to be certified by ONC as a meaningful use-certified
product. We have criteria that requires them to
have the ability to perform a drug-drug interaction alert or a drug allergy alert. We don't certify
how that's displayed or the content of those drug
classes or drug objects in those systems. So we leave that up to the 1 vendor themselves
to work in conjunction with the knowledge base to
have that information. We are simply saying that
for a provider to use a system that's meaningful use-certified, they should have the ability
to do this.
DR. BARRETT: Dr. Ruth? DR. DAY: Ruth Day. I'd like to thank the
speakers, the recent speakers, about providing evidence about how these databases are used
in everyday life. It's very important and quite
impressive. I do have a question for the FDB
presentation. We know that your database is used
by many clients, many different institutions. If
you could just briefly give us an idea of how many
patients or patients per year are benefitting from
this use, and then go on to talk a little bit about
customization. The liability implications for local
customization are really frightening in some ways
and challenging in other ways, I presume. But do
you keep tabs on how the different 1 clients do
customize the database and what problems have occurred so that then you could step back
and provide guidance about things that can and
should not be customized?
DR. MATUSZEWSKI: So for your first question, how many patients are supported
with the FDB drug knowledge, I can't give you an exact
number. I can tell you that there's thousands of
customers in a variety of different uses, everything from pricing analysis to use in
clinical decision support. A number of hospitals and
health systems, even retail pharmacies that serve
millions of patients every single year. But I can't
give you an exact number.
In terms of the AlertSpace modification, that's a relatively new product, so it's been
out three years. About a hundred customers are
using it now and making modifications. And yes,
FDB does have records of those modifications.
I can tell you that institutions who use AlertSpace use it very gingerly. And gingerly
is they don't make wholesale changes 1 because
before something like AlertSpace allowed local
customization, they just basically used severity levels for crude adjustments and just said,
we're going to either turn everything off, which
doesn't help you at all, at least for Leapfrog, or
we're just going to turn off level 2s and 3s, or
we're just going to provide level 2s and 3s to
pharmacists for review and not for prescribing purposes.
So the legal liability, I would say that most institutions are not making wholesale
changes, but are also taking any changes they make
through their P&T committees or med exec committees,
and being very careful about when they change
a severity level 1 contraindication that FDB
has indicated to downgrade.
Now, a number of institutions have actually upgraded. So things that have been considered
severity level 2 based on evidence, they may have
had a problem with before, some med errors, they've
upgraded them for their entire staff. In terms of completely eliminating 1 alerts,
whether it was a 1, 2, or 3, that again is based on
some of the data I've seen, not done very often.
But occasionally, for the nuisance alerts that they
perceive their institution has been done. I think if you asked me that question in
about another year or so, we'd have much more data.
DR. DAY: And do they ever add any drugs? DR. MATUSZEWSKI: At this point, there have
been some requests. So something that's not identified in any of our monitoring of the
literature or not identified in the label. We are
looking to add that functionality probably in early
2014 because that's a whole nother level of use,
where nobody can really pinpoint but they say
that's a problem at our institution. DR. BARRETT: Dr. Horn?
DR. HORN: I'll just make a comment on the customization. I applaud the vendors for their
ability to add that. We in our institution started
customizing in 2006 our DDI database, and at that
time, we had about 8,000 drug pairs that were in
the highest severity category. I 1 wish I had Karl's
database; it would only have been 1600. I would
have been done much quicker. We went through every one of those drug
pairs and reviewed the literature on every one of
them and reassigned categories. There are now
16,000 drug pairs in our highest severity list that
we get from our vendor. So it's not a trivial process to do this.
We have done it for about 12 other institutions, helped them through that process.
And what we find fundamentally is that you reduce
the number of alerts that are firing, obviously, because you downgrade the highest ones to
something less.
But also we find that the number of irritating alerts, for lack of a better word,
is markedly reduced, and the practitioners recognize
that. They're not getting alerts for silly things
any more. And that's exactly what we want. We
want the alerts to fire that we believe have risk
for patient harm. The legal question is one 1 that gets bantered
around a lot. There was recently a symposium held
on that various issue. I don't have the reference in my head, but I'd be happy to share it with
you later, if you'd like to look at it. It's really
wonderful. Their bottom line was, there's really not
a big deal here if it's done in a prospective, knowledgeable manner as opposed to, oh, let's
just shut them off, which is a real big risk. But
if it's done with knowledge and with forethought,
you're probably reducing your risk because there's
a huge risk if you ignore an alert that's in the
system and it causes harm. In fact, the only case that I've been called
on, a medical-legal one, was exactly for that -- regarding the customization stuff;
was a situation where they shut an alert off and
then it caused harm. But if you have specifically
modified an alert and done that with forethought, you're
probably not going to have much legal risk. You
can't eliminate the risk. You've always got the
risk. But I don't think you're 1 increasing your
legal risk at all. DR. DAY: But the risk transfers to the
customizer, I hear, not the original vendor. Is
that correct? DR. HORN: Yes. But the original vendor has
no risk, either. It's like we have no risk in our
book because of the learned intermediary rules. So
if the providers all were reliable, there would be
no books. There would be no software. There would
be nothing. At the end of the day, the risk is the
physicians. And then they'll go after the institution because those are the deep pockets.
So if your institution has a policy to evaluate
the interaction, look at the evidence, and make
a decision based on that, that should hold up
quite well in a court as opposed to, well, we were
just tired of getting a lot of alerts so we shut
80 percent of them off. That's not going to look
very good to a jury in any case. Then we also have in place a system where
we have a monthly review committee 1 that does
nothing but look at interactions in our database because
we are continually getting updates from the vendor.
So we have to continually look at the new interaction alerts that are coming in as well
as the data because we do the same thing. As
Karl pointed out, we raise and lower alert rates,
or severity levels, based on data.
DR. BARRETT: Again, just a reminder, please state your name when you speak in the mike.
Dave Flockhart? Okay. Dr. Zineh?
DR. ZINEH: Two questions for clarification, one for Dr. Bates, the other for the speakers.
You mentioned a recommendation to make boxed warnings
computable. What does that mean? DR. BATES: Just that when a boxed warning
is released, it will be helpful to consumers of
them if they are framed in such a way that you can
actually put them into an algorithm. Often the
warnings include words that are vague. Caution is
an example. A caution is not a computable term.
So we're looking for things 1 like, "If the ALT is above a certain level, then do X."
DR. ZINEH: Thank you. The other question is for all speakers.
There is a question before the advisory committee on a framework to assess literature, drug
interactions from literature. That framework doesn't necessarily talk about evidence, and
it's probably beyond the scope of the conversation
here, but it's implicit. That would be the next
step. So my question is, these knowledge bases,
are there rubrics for putting things into the
system and assigning severity? Are those publicly available, transparent, et cetera, or are
those part of the proprietary nature of these platforms?
DR. MATUSZEWSKI: There is no rubric. There is no formula. So if in the span of six months,
there are five case reports, or if there is a
series of 10 cases that identifies a significant interaction, that would be then judged on
its merits on the strength and the quality of
the study publication, whether that indeed gets incorporated
into 1 the database. With drug-drug interactions, you're not
going to see randomized, controlled trials. And
often early reports, if they're serious in nature
and the mechanism is well explained, that in itself
in a couple of case reports may cause a severity level to change or for an interaction to be
added in our database for the first time.
We even have referenced animal studies, but rarely would an animal study be of sufficient
quality evidence to include in the database in
terms of a new drug interaction. So the answer is,
there is no secret, magic formula. DR. BATES: I feel like the existing
evidence frameworks don't necessarily translate that well to this particular domain, and so
developing something new would be a real contribution. I think that the group that
Dr. Malone has brought together has talked about
doing that. Dr. Horn may have been involved in efforts
like that as well. DR. BARRETT: Dr. Venitz?
DR. VENITZ: Jurgen Venitz. 1 Let me ask a follow-up question. How important is, in terms
of evidentiary assessment, the knowledge of a
mechanism? In other words, would you accept case
reports, whatever, without any mechanism and incorporate that in your database?
DR. MATUSZEWSKI: "It depends" is too flip of an answer. If it was a strong study and
the mechanism was applicable to other drugs and
was now uncovered, I would say that there would be
a reasonable chance that it would be included
in terms of assigning a severity level.
I think one of the new evidence sources that we're looking at again is a drug metabolism
and drug transport database and using that to
refine our contents. So the more of that that's
available, either from the labeling or from the
literature, I think improves our ability to appropriately categorize a drug-drug interaction
in terms of severity.
DR. VENITZ: But in the extreme case, if you had no evidence of any mechanism, but you
have either uncontrolled studies 1 or case reports
suggesting that there's an interaction? DR. MATUSZEWSKI: Then I would say if the
adverse effects from that interaction were serious
and of a high enough frequency, that probably would
be included without having the mechanism. DR. VENITZ: Thank you.
DR. BARRETT: Dr. Au? DR. AU: Jessie Au. My question is actually
for the entire morning, what I heard. I heard prediction that you use in FDA to make your
projection. I heard quantitative versus qualitative. And I also heard from several
speakers now that the DDI situation is becoming more and more complex.
So looking ahead and looking backward, most of the DDI that we have so far are based on
PK interactions, whereas the situations are easier
to handle from a quantitative standpoint because
the drug level goes up, goes down. You can project.
However, we are now in this beginning or already in the middle of the molecular medicine
era where we're dealing with molecular 1 targets,
which the plasma level really doesn't say much about
that. It doesn't help us to understand the mechanism. And I would like to use this one
example and then ask my question. So just to give an example how fast things
are coming at us, in the last seven years FDA
approved seven drugs, molecular targeted drugs for
renal cell cancer. It's coming so fast. And then
if you look back a few years, eGFR inhibitors, of
course, has been out there for a while now, and of
course if one drug works, adding two drugs that
work must be better. However, a trial was done with 200-some
patients, where it gave eGFR inhibitors and combined it with standard cytotoxics to non-small
cell cancer patients. So instead of dying on
average in nine months, they now died on average in
six months when they got this extra drug. Okay.
So now finally we know we can't just combine drugs.
So now my question here is, based on something like this, what do you do? Do you
now start to predict that you should 1 not combine
eGFR inhibitors with standard cytotoxics? And at
what level do you get this information out?
Because it's not really quantitative. There's no way to quantify that except we
do know, evidence-wise, patients are now dying faster
because when we did a trial, we didn't know better.
How do you handle that information in the FDA or in
FDB? Do you actually get this information out
there so patients know that they shouldn't be
getting things if we don't know how they work? Since we're in the molecular medicine era.
DR. MATUSZEWSKI: I'll go first, and then FDA can give the final answer. So it's almost
the beta blocker. If I give one beta blocker and
I think I'm going to get twice as much effect
by giving a second beta blocker, that's duplicate
therapy. No? Not quite the same? DR. AU: No, because the signaling pathway
is more complicated than just beta blocker. Beta
blocker, you have a finite target. When you talk
about signaling, you have P transcription, post
transcription, post translation. 1 You interact at
so many levels. And if you do the equation and do
the math, you can get antagonism sometimes. You
can get synergism sometimes. But that part of the research is still in
the infancy. So you don't even have the guidelines to give out advice. But you do see the outcome
of it; 200-some patients are now dying faster.
DR. MATUSZEWSKI: So in that case we probably would not include it in the database.
If you have drugs that are given for the same
indication, then that might again trigger some sort
of alert; is a second drug necessary? But in terms of the molecular pathways
having detrimental effects, until that either appears in the label or in a publication that
would make that something to be contraindicated,
we wouldn't be picking up on that.
DR. ZINEH: I can try to address that. I think it's a little out of scope. The answer
to -- I don't know enough about this example. But
if this is observed in drug development, that's clearly handled by not approving 1 the combination.
If this emerges just like any other post3 approval issue in terms of diminished efficacy
in a subgroup, enhanced risk in a particular subgroup,
that's handled through a variety of ways. It
includes updated labeling, risk mitigation strategies, drug communications. In worst
case scenarios, if we find out something that was
untoward in terms of the risk/benefit analysis, drugs get pulled off the market.
So without knowing the specifics of your example, I would say there are a variety of
ways to handle unexpected risk/benefit balances
in subpopulations after drug approval.
DR. BARRETT: Maybe just to come back to the labeling issue, though. I saw in, Dr. Matuszewski,
your pie chart here when you list the different sources of evidence. And of course, the labeling
is only one part of this. But I'm curious. Do you keep track of the
extent to which the database matches the labeling? Or is that something that is at all part of
this? Certainly you allow some flexibilities 1 at
the end user level. But from the standpoint of the
label as it weights the clinical evidence portion
of this pie chart, do you keep track of that at all?
DR. MATUSZEWSKI: All the drug-drug interactions are then further detailed in
a monograph. So all those pairs have monographs.
So there would be reference to whether the interaction
is based on the PI or other literature. In terms of weighting with again a new drug
on the market that has interactions, there often is
PI is the only source. So you really don't have
any weighting. As a drug's been used and on the
market for a number of years, that's when the
product information may become out of date, where
again information from the published literature would override what might be in the product
information. So is there a weighting system? I would say
there isn't. But again, the manufacturer's labeling is a very important thing that we
look at every single time.
DR. BARRETT: But you're 1 not keeping track of when you go outside of the labeling? That's
DR. MATUSZEWSKI: Oh, we are. In terms of the references for a recommendation of a specific
severity level, that would be included in the
monograph. If you're asking me what percentage of
the time -- DR. BARRETT: Yes. Yes.
DR. MATUSZEWSKI: -- that probably requires some extensive research, which perhaps if
I get a student or fellow in the next couple of months,
I might be able to look at it.
DR. BARRETT: Good enough. Our last question will go to -- Dr. Horn?
I'm sorry. DR. HORN: I was just going to
comment -- this John Horn -- on the question that
was asked about the case studies. And these are a
huge problem for all of us to try and make sense
out of this literature. And some years ago, we
developed something called the DIPS, which is a
Drug Interaction Probability 1 Scale, which was
designed to take where Naranjo started with the ADR
scale and make that applicable to drug interactions; in other words, not just one
drug but two drugs; and then whether those caused the
ADR. That's really what we use now in our
evaluation. And one of the parameters of that scale is mechanism because if you don't have
biologic plausibility, I don't care how good your
study is, it's nonsense. And there's plenty of
that in the literature. So we're pretty cynical about case reports
because usually they're not well done and it's a
huge problem. But I think that case reports are a
lot like other things; they're a good trigger, and
the hair on the back of your neck goes up, and then
you remember to watch for more information. DR. BARRETT: Dr. Malone?
DR. MALONE: So, Karl and David, one of the things that both of you raised -- and Tricia,
this applies to the ONC as well, and certainly
to the FDA; I'd like to hear comments across all
of you -- with regard to the 1 use of the term
contraindicated, we see that term used, especially with drug interactions.
I'm wondering if we could have a little bit of a discussion amongst you about what sort
of criteria you would use or do use to imply
that because many times, people imply or assume
that contraindicated means there was never, ever
a situation where one would want to use these
medications together, and therefore it would be
inappropriate to use the medications together. As we've done some of our work with the
conference series that we've alluded to earlier today, we're struggling with that concept.
So I'm sure you guys have all struggled with it,
too. But I'm interested to hear your perspectives on
the use of that term, especially as it applies to
the drug interactions.
DR. BATES: This is an important area, and I
guess what I would say is it would be very valuable
to really, across the industry, have some agreement
about what we mean by perform terms. This really came out for 1 me when we did
a study in which we compared -- we basically
looked at terms that radiologists used in radiographs
to say whether something was present and when
it was absent. And we looked at a number of reports.
We found all the terms that they used. Then we
had them rank them in terms of probability.
It turned out that amongst the radiologists, there was almost no agreement as to what any
of those terms meant. And there was even less
agreement when you compared things to what the
primary care providers who are the consumers of the
reports meant. So unless we agree on what we mean, I think
it's a big issue. And in domain after domain in
medicine, after you develop some terms and everybody agreed about what they meant, you're
better off. That happened in sepsis, for example. When we use the term contraindicated, we
mean that the two drugs should never be given together. But unless everybody else agrees
about that, too, I think we're not where we want
to be. And it would be very helpful 1 to have just
a few terms and then get some agreement about those.
DR. WILKINS: I think it's a great question, and we should get some consensus. I would
say, from ONC's perspective, we're looking at this
from clinical decision support. How do we support
clinicians to make these decisions? How do we
provide them the right information for them to do
their jobs effectively? In the work that we do with electronic
quality measures, we allow for exceptions and
exclusions in different scenarios. And so we have
the goals for these measures and what the outcomes
should be, but we know that in practice, things aren't always cut and dried and that we, from
our perspective, aren't in a position to say what
that should be.
I think that we would approach it -- as we continue to work in the drug-drug interaction
ream, we will continue to approach it from that
angle and have ways that these systems can acknowledge,
if we're doing this for certification, exclusions
and instances where the benefit outweighs 1 the
risk and we allowed providers to do their jobs without
them being restricted in that way.
I think that what would help us, though, is if the knowledge base community gets better
consensus on the severity ratings and how these
drugs are categorized, that we don't have to put
things back on clinicians to have to readjust severity ratings on their end.
I think that it would be useful for us to have more of that discussion, though. So I
agree. But we would not -- I shouldn't say we wouldn't
not; we are more interested in supporting the
decisions that clinicians have to make in their
context with whatever parameters they have to deal
with as opposed to looking for hard and fast rules.
So I would say that we would take a similar approach as we have with clinical quality
measures and allowing for exclusions, and allowing
physicians to document how and when those take
place. DR. MATUSZEWSKI: I might just say that if
contraindicated as a section or a 1 statement appears
in the package insert, that is a major signal for a
drug knowledge database to say it's contraindicated. That means, don't give it.
Then if you put "should not prescribe together"
in a black box warning, that's also a pretty strong
signal that that's contraindicated. Now, after those statements are made, can
you look at breakouts in terms of dose intensity? Can you look out for route distinctions? That's
where we would try and fine-tune the content if
evidence was available to make that breakout. But again, our definition of contraindicated
is, you should not give these together. And unfortunately, the amount of overrides suggest
that that may not be true.
DR. BARRETT: Final question to Dr. Muzzio. DR. MUZZIO: Yes. A very good question. So
when you evaluate literature to decide to include
something or not to include it, do you pay any
attention to who funded the work, the corporate relationships of the investigator? I mean,
not that I want to doubt anybody, 1 but just out
of curiosity.
DR. MATUSZEWSKI: In any good study review, the source of funding would be probably something
that one would look at. Unfortunately, I think in
a lot of the case reports, these are not things that are necessarily funded by industry, not
likely to have bias implicit in their results, and
if anything else, are independent, this is a
problem at an academic level, report. So this is not
about effectiveness or off-label use. This is really
about negative things. So I would say yes. I don't have a list
with me, but there's probably over a hundred journals that are looked at in terms of drug-drug
interaction information, case reports, or case
series, and the source of funding would be something evaluated. We don't necessarily
document that, but that would be considered. But I
don't think that's a major source of contention
at this point.
DR. BARRETT: Kellie, did you -- DR. MATUSZEWSKI: We would 1 love to see more
funded drug-drug interaction study. DR. BARRETT: Did you have a comment?
DR. REYNOLDS: I was just going to respond to the contraindication question.
DR. BARRETT: Please. DR. REYNOLDS: Our intent when we indicate
two drugs are contraindicated, there are no situations where risk/benefit indicates the
drugs can be given together. It needs to be based
on some kind of evidence. Usually it's not based
on a drug interaction study. Usually it's based
on mechanism or extrapolation from another drug
interaction study. But that is our intent, where
it's more difficult is where in other sections of
the label we say "Avoid" or "Should not use." That's a little more wiggle room there. But
it's not the same as contraindication.
DR. BARRETT: Thank you. We will break for lunch now. We will
reconvene in this room in one hour, at about 11:55.
Please take any personal belongings you may want at
this time. The room will be secured 1 by FDA staff
during the break. Panel members, please remember that you should not discuss the meeting topic
during lunch among yourselves. Thank you. (Whereupon, at 11:52 a.m., a luncheon recess
was taken.) A F T E R N O 1 O N S E S S I O N
(12:56 p.m.) Questions to the Committee and Discussion
DR. BARRETT: Could everyone come in and take their seats, please? We're going to get
started here. We will now proceed with the questions to
the committee and the panel discussions. I would
like to remind the public observers at this meeting
that while the meeting is open for public observation, public attendees may not participate
except at the specific request of the panel. I'd also like to recognize the FDA press
person, Stephen King. Are you here? No? All right. Well, he's in the house writing "The
Shining, Part 2." No. So we're going to go through, and I will
read the questions, and then we will go around the
horn and get some feedback from the committee members. Again, just please, after we go through
once, then we'll go and have additional discussion. Please discuss the following with regard to
the format of drug interaction 1 study results presentation in prescription drug labeling:
a) The level of detail on study designs and study results;
b) The advantages and disadvantages of presenting the drug interaction study results
in a forest plot versus a table versus a narrative.
So would anyone like to begin? Or Jack, are you okay if we start down in order? Just your
initial thoughts on those questions. DR. COOK: Okay. The level on detail on
study design and study results? Well, the design,
I submit, is probably of minimal use in the label.
I would expect that that be detailed at the FDA,
and they can deem whether the results are appropriate or not. So I wouldn't spend a
lot of label space on something like that.
I wouldn't say as far as the advantages or disadvantages of producing forest plots, tables,
or narratives. Certainly the forest plot, we've
started to use those more and more, not only in our
labels but internally, to present a large amount of
data. And it puts it into 1 relative context. One thing I do like about it is you can make
sure that your recommendations are consistent for
at least consistent PK changes as far as dose adaptation. I think that's a little easier
to do than in a table.
So I'll leave it at those opening remarks, and you can go left.
DR. BARRETT: Jim? DR. KEIRNS: Yes. I have the same comment
that Jack does about design. I think we probably don't have enough real estate in the label
to have design, so we just have to trust the judgment
of the people that put it in the label that actually
it was a good study or it wouldn't be there at all.
In terms of the data presentation, like Jack, we're using forest plots a lot. We
particularly started using it about three years ago
when we saw a publication from OCP scientists. And
in the example that Dr. Reynolds showed on mirabegron, I was intimately involved in that,
and we were quite pleased with the way that worked
out. Now, one thing that was kind 1 of interesting
about it was that our proposed labeling for Europe
was exactly the same as the U.S. But then at the
late stage, during label negotiations, the reviewers in Europe said, "Oh, well, we don't
understand this plot. Please replace it all with
text." So if you go look at the European label for
mirabegron, it looks kind of old-fashioned for the
presentation of DDI, whereas the U.S. label is what
Dr. Reynolds summarized. DR. BARRETT: Maybe just as we're going
around here, if you would like to comment from
FDA's perspective, just let me know because some of
this, I think, maybe you want to make comment to as
we go through the initial comments. Kathleen?
DR. NEVILLE: I think levels of evidence, maybe not detailed, per se, but levels of
evidence for study design would be helpful. And while
I appreciate these guys' comments, we struggle
with having practitioners understand what Cmax
and AUC is, never mind 1 a forest plot.
If you're trying to get the average practitioner to understand what drug-drug
interactions matter and what don't, I don't think
a forest plot will accomplish that; perhaps in
addition to tables and narratives, maybe. But as a
standalone, I think that that would absolutely not
achieve the goals that we're looking for. DR. BARRETT: Kathleen, let me follow up,
though. As a caregiver, what do you want to see in
there? What do you see is the biggest benefit? DR. NEVILLE: In the materials that were
given in preparation for this, I found that tables
and the narrative the most helpful. I think a more
concise, like you said, high yield introductory paragraph is very helpful.
I've found that tables with what happens to the -- whether it's a victim or a perpetrator,
and then potential. I know FDA can't dictate care,
but potential implications for dosing is very
helpful to the practitioner, especially my biases
in the upcoming years.
As trainees get less and 1 less pharmacology, they are going to understand the implications
of DDIs less and less. So they're getting less
statistics, too, so the simplest language. And
like was mentioned earlier in some of the talks,
perhaps not referring to AUC but to exposure, things like that that make it easier for the
average practitioner who doesn't understand the
level that we do clinical pharmacology, would help
them understand the implications of drug-drug interactions.
DR. MORRIS: Well, I found Kathleen's comments very interesting. So I agree. I think
the information that is presented should be very
simple and straightforward. And maybe one of the
most important aspects is changes in dosing regimens if something is completely
contraindicated. You have to say, what should you do with
this interaction? This is what, I think, the physicians are looking for. Does the dose
have to be decreased? What should it be decreased
to? So that sort of information is 1 very important.
With regards to study design, I agree that I
don't think that needs to be there. But study results, I'd like to see maybe a link to
information, so those individuals that want to look
at this in more detail, what exactly did this study
show? And these would maybe be for specialists or
be for residents that really wanted to understand the interaction. So having a link to that
information would be valuable. With regards to forest plots, I really like
the forest plot, so it shows the way I was looking
at it. But what I thought was -- and certainly you
could use, instead of AUC, exposure. That would be
one way of doing it. I liked where you have really the stippled
lines, the variability that is within the normal
range. And you can see if you're outside, if
you're higher or lower. But then also, what should
you do? So you have this interaction, so you reduce the dosage to 20 milligrams daily.
So again, giving clear information on what you
do with this type of interaction. Contraindicated.
1 Do not administer together. Something like this.
But again, some interactions are very complex, and I think those interactions you
can't really describe by a forest plot. And that's
where I think you really need to get into at least
a table to describe in a bit more detail the
interactions. So that's how I felt it would be most
valuable to practitioners. DR. MILLER: A couple of thoughts. One
observation I made, the drug interaction information and results come in a lot of different
places in product labeling. I find that very confusing to follow through all the different
areas. If this is all a function of drug
interaction, maybe there should be a concentrated effort to place as much of that information
in that particular area, so you don't have to hunt
and peck around for information. That's one issue.
As far as how you present the information from a literacy perspective, 1 using pictures,
diagrams, to supplement some narrative information is an important aspect of that. But perhaps
rather than having lengthy narratives, maybe a figure
with some callout boxes that highlight key results
or something like that may be a useful strategy.
Then the last point I'd like to make is that whatever result is presented, there has to
be actionable information aligned with it. And
the reality is, I'm looking at some of the examples
from earlier slides, and it just simply says, well,
this increases the plasma concentration. So the prescriber -- how do you interpret
that? How do you make that judgment about what
action do I take now because of that? Or should I
just be aware that that does that? So those are my
three points. DR. MALONE: Well, I'd like to thank the FDA
for assembling this committee. I've been working on some of these issues for over 10 years
with respect to how to evaluate the evidence and
putting it into meaningful clinical decision support
to clinicians to improve patient safety. 1 So
a lot of these issues are fairly close to home for
where I do my research and the types of projects I'm
involved in. I guess importantly, at this point in time
I'm the principal investigator on a funded study
from the Agency for Healthcare Research and Quality
that has three different working groups that are
addressing various issues with drug interactions and clinical decision support.
So I'm going to interject some of my comments with what I'm hearing from these
working groups, some of the issues that they're struggling
with, although my comments are specific to myself,
not necessarily reflective of the entire working group.
But with respect to the level of study detail and study design results, the FDA label
is a relatively static document. There's no page
restrictions on how long that label is. There's no
width or size restrictions. We see slim jims. We
see huge poster-sized labels, et cetera. So that the notion 1 that we need to be
restrictive in how much information we provide in
the label to me is kind of silly in that when we
try to evaluate the evidence that's included in the
label relative to the strength of evidence we're
seeing from other sources, many times these studies
that are done premarketing never get published and
they're black box phenomena. And I don't mean black box warnings; black box in terms of
we don't know what happened. We don't know what type
of study it was.
So I think the level of detail needs to be dramatically ramped up. And I may be alone
on this, but if there was a reasonable expectation
why we would want to keep this information hidden,
I could understand it. But I see no reason for
that, that we really do need to know the study design.
We don't know need to know all the details, just like we don't see all the details about
randomized clinical trials in the package insert.
But we need to have some basic information about
the approach of the study. With respect to the study 1 results, most
individuals who are looking at this information are
usually not the practitioners. They're synthesizing it to a practitioner at some
level, whether it be at the drug knowledge database
level or some other intermediary that's going to
take that information and synthesize it into, hey,
listen, I don't think you should give these together, or, that's fine, I don't see a problem
here. What happens is when you are using terms
that are subjective in nature -- we recommend, not
recommended, may reduce the dose -- those general
terms really become difficult to be actionable to
the clinician. So the more detail that we have
with regard to study results, I think it's key.
With respect to that, the notion of a narrative -- narratives are less meaningful,
I think, than having the data in the tables
and/or forest plots. The advantage of forest plots,
relatively quickly interpreted. The advantage of a
table if the data's there is that it's detailed enough so that you can do secondary 1 analyses
in the long run if you feel like you have the information
across multiple studies. But because many of these studies never get
published, having that raw source of information I
think is critical for people trying to evaluate this to take it to the next level, meaning
what should the clinician do.
So I would avoid narrative statements. I would argue to include as much information
as you possibly can, especially given that the label
has largely become not the primary source of
information for the busy, active practitioner. It's the people who are working in the drug
information centers, the drug knowledge database vendors, these other trained intermediaries
that are taking this information and synthesizing
it. Thank you.
DR. BARRETT: Again, please state your name before you start.
DR. MALONE: So that was by Dan Malone from the University of Arizona. Sorry.
DR. BARRETT: 1 Thank you. DR. HORN: John Horn. Yes. Thanks again
for inviting me. It's been very stimulating and
entertaining. And before I talk about those two
specific things, I would just like to make two
simple pleas. One is, I know we're not supposed to
talk about where in the label the information is,
but I can't stop myself. (Laughter.)
DR. HORN: I have no objection at all to having it spread out. But the only thing I
would ask is that all of the drug interaction information
be put in the section labeled drug interactions. I
go nuts having to go through the label having to
find all of the drug interaction information. Okay? A simple thing. Please? Thank you.
The second one, everybody down this line has said, sometimes we need detail and sometimes
we don't. I have a really simple suggestion.
Put links in between the label and the review
of the NDA because that's what I have to do. I have
to go back to the NDA reviews to look at the data
because I don't believe what you guys 1 write most
of the time. No offense, but I want to know what
did the study involve?
Now, I'm odd because I like that stuff, and other people don't. So you don't have to clutter
the label with it, but put a link in. It would save me the time of going up online, downloading
those huge documents, and flipping through 40,000 pages in order to find the one I want.
Easy to do. It would make our lives much, much
easier. As far as the actual questions we're
supposed to be answering, the level of detail in
study design, as I said, if you can link it, that
would be great. I think there's some minimal amount of detail that needs to be in the labeling
so that anybody can get it. And that's why I
really don't like the forest plots. The problem with those is, you can't see the trees because
of the forest.
(Laughter.) DR. HORN: You look at a forest plot, and
first of all, I have a lot of trouble figuring out
where that thing comes down on the 1 line. I want to
have to get a ruler out because the bigger they
get, the harder it is to figure out what they are.
There's no dosing data. There's no duration data
on those plots. Totally useless to me. I can't make anything out of that. I don't know what
those numbers mean.
I'm sure the statisticians knowledge exactly what they mean, but have you guys actually
gone out and asked a bunch of practitioners to describe
what's on those things? Because I'll bet there isn't five practitioners in the world that
knows what's on those things. They are very difficult
to interpret compared to a table. Everybody can
read a table.
With regard to the tables, again a couple of really simple things. Please alphabetize
the listings in the tables. You've got lists of
tables that are 40 drugs long, and I'm looking for
one drug. Why do I have to go through 40 of them
to find it? It should be alpha. Right? Simple.
Do it. The tables usually contain information.
They've got the dosing of both 1 drugs. They have
the duration, usually, of both drugs involved. That's really the information I'm after.
Then we get to the outcomes, Cmax, AUC. I personally like that because the only way
I can decide whether this is likely to be a problem
or not is to know what the AUC change is. I'm
looking for that.
What I don't particularly care for is the statistical presentation of that data. First
of all, there's no easy way to look at it and
ask the simple question, is this statistically significant
or not? Now, of course you can figure out what the
confidence interval is. You can figure it out.
But we're talking about practitioners, like I used
to be. I don't know how to do that with a confidence interval.
I would much prefer, instead of a confidence interval, to see the range of outcomes because
we know there is a huge inter-patient variability
in the outcomes of interactions. And that's a
very important piece for me when I'm making a decision
about whether I want to do 1 something about a
particular interaction. If I know that even though the average
changes 40 percent, if there are people that are
having a 200 percent change, that's important for
me because some drugs a 200 percent change may not
be very important, but for others that might be
really important. So knowing what the range of response is,
is much more useful to me than confidence interval.
I don't care about confidence interval. It really
doesn't help me. And I know a lot of people out
there who are less sophisticated than you all would
be able to deal with, I think, just the simple range numbers much -- if you want to put the
confidence interval in, that's fine. Try and be consistent in your labeling.
Just looking at the examples, we've got LS mean
ratio. I think LS means least squares, but I'm not
sure about that. I suspect if I gave that to my
students, nobody would know what that meant. Change in mean ratio. Why ratio estimate?
That's 1 minus the change in 1 percent. We don't
have to clutter the label with that. I can do that
math. That's not hard. I'm not sure what difference that is compared to the others.
So those are some really simple things that would clean it up. And obviously, my preference
is tables, not the -- the forest plots to me
are just -- make me nuts. I don't like those at
all. I just can't get enough information out of
it to make any sense out of it. It's just not helpful.
I'll stop there. Thank you. DR. FLOCKHART: I'm a forest plot fan, but
I'll come back to that. I'm Dave Flockhart from
Indiana University. I think in terms of the first question, the
level of detail, there were two things here, and I
think we're walking between the two. One is, I
think, we're not creating the label for a bunch of
academic researchers. We're creating the label for
practitioners. I think it's perfectly legitimate to link
it, to link it to the NDA or to link it to good
academic research. But Dr. Bates 1 and others made
very clear that 2 percent or much, much less of the
people are actually going to be the people digging
in. I think, as a physician who practices with
patients who have lots of drug interactions -- I
see a very biased group of people, I think. But I
think more pharmacists see them. But I think at
the patient level is something I really think is
very important. We were given a series of tools this morning
in the four excellent talks before this that allow
us to prioritize interactions. And there's a lot
of data on this. Whether it be the word contraindications, whether it be specifically
some serious discussion at the FDA about what goes
on the highlights section, or whether it be Dr.
Bates' 15 or 17 really bad interactions, I think
a really good thing for everybody would be a binary
decision, drugs for which drug interactions might
matter and drugs for which they might not. Just
yes or no as a first thing, and something you could
translate eventually into some 1 kind of symbol or
something that would really be patient. Now, that allows you to actually have to get
cortical and think about what that level of risk
would be. Some level of scientifically guided decision would have to be made about what
goes into the interaction group and what goes into the
not. Without getting into that debate, I think
my only thing about it would be the general perception
that we horribly, horribly, horribly overestimate
these interactions at the moment because of the
way the drug interaction databases that are commercially
available have practiced over the last 20 years.
It is clear from the whole morning we have vast surfeits of information, and we have
huge alert fatigue. And that is a public health
risk, that itself. So every time you add more
information, I think you have to think you're adding a public health risk if you add more
information. So I'm for proposing a binary decision.
Also, I think then if one walks through it, the
next step is to alert people to 1 who are the people
that you're focusing on at most risk? Because even
with bad interactions, the pharmacokinetics often
doesn't mean anything. I was informed in my own training by the
legendary Dr. Abernethy here, who pointed out to me
many, many years ago that there's a cimetidine/ benzodiazepine interaction that is purely
kinetic and not dynamic. And he measured both, and
this is probably the first study that really killed
that point.
But it's been made many, many times, even in
the worst interactions that we have -- terfenadine/ketoconazole -- if everybody
had died, half the population wouldn't be here. So
there's a very, very small number of people who
suffered in that particular context. So it's very useful, I think, to put right
at the top of the label that you care about, what
are the risks? Hypokalemia. Which interacting drugs? Right up front. What are the risks
that increase the risk of a person experiencing
that interaction? And then, of course, 1 as was
pointed out, what do to. Those three things -- do
you care, who do you care most about, and what
to do; those three things.
Now, specifically to forest plots. I'm a fan of pictures as opposed to -- I think a
picture is worth a thousand words. I think most tables
aren't read by people, that's the problem. And
they're better, from a scientist point of view, but
they don't get read. They're fine to link to. But I think you
need something for practitioners, and I think forest plots are up there as one of the best
ways of presenting it. I have problems with forest
plots, too, like John does. I think log scales should be banned because
your average medical student, never mind anybody else, can't appreciate the value of that.
I think it's got to be really clear what the error
bars are, and I'm a fan of the range as well, putting
on the range on there rather than some estimate
of the error that's clear to a statistician but not
to a practitioner.
But a range, I totally agree with John about that. A range is clear. And a range also deals
with one really important thing, and that is that
very often you're looking at the mean of something that is not normally distributed. It's not
a nice, normal distribution.
There are people who don't experience the interaction at all, and there are people who
experience a bad interaction, and there are people
who even have the interaction experienced in the
opposite direction. So I think having a range is a way of
communicating that without implying, by putting down there a mean with a standard deviation
or a standard error, that it's normally distributed,
I think that can be deceptive.
So to summarize, I think the scale should be
clear. The size of the interaction should be
clear. And it shouldn't be presented in a way that
is deceptive in terms of the error. But I do think
a picture like a forest plot is something valuable,
and they communicate very quickly 1 that there's a
big difference to an arm. One last point about it. A pharmacokinetic
change on a forest plot to me is pretty useless. It's got to be some respectable clinical outcome
derived hopefully -- and this is a fantasy, really -- derived hopefully from some kind
of randomized thing. But the problem with that
is we'd all go bankrupt if all these things were
randomized, controlled trials. So to Dr. Abernethy's point early on, I
think we should entertain a discussion about what
other data, beyond tightly controlled, what observational data might be included in that.
And I think one could usefully come up with a
series of criteria of what are valuable observational
data and what are not.
There is such a thing as a really, really good case study that's very carefully conducted;
not all case studies are simple, quick observations. So I think having a discussion
about that would be something valuable.
I'll stop there 1 and shut up. DR. POLLI: James Polli. My major comment
is very similar to what David was talking about
when he first started talking. So he saw them, and
his first comments were -- he talked about how the
label has several stakeholders. I think that point was made very clear
from this morning, sometimes in a painful way.
Dr. Juurlink talked about how most practitioners, most physicians, prescribers, don't use labels
frequently. Meanwhile, the gentleman from First
Databank says it's the most important information for what they do.
I guess during the course of the morning I was mostly thinking about prescribers, pharmacists,
and actually also patients. And I guess my major
comment would be, I have a hard time thinking about
this question because it seems like a single label
that's black and white PDF, found at dailymed.gov, it's clearly not working for all stakeholders,
it seems.
So to me it would be great to have labels for different stakeholders. And 1 you do have,
at least for some drugs, a label for patients,
which we maybe didn't talk too much about.
As far as study design, I think most stakeholders probably -- I agree with Jack,
probably not very interested. If it's not done
well, then don't include it. Study results, I
agree with Michael in terms of actionable. Seems
to be extraordinarily important for several different stakeholders.
As far as representation, I think I had the same experience as Kathleen. I like graphs,
but as I was reading through the materials, I said,
I think I like tables and simple text a little
bit more than I had originally thought of. I like
the idea of a binary decision tree, at least for
some stakeholders.
MS. CABALLERO: Rose Caballero, representing consumers. I'm from San Antonio, Texas, and
I've been sitting here listening to all these
discussions. And when you consider that all the
physicians/clinicians are confused by the results,
I'm sitting here thinking, where 1 do you think the
patients are? Certainly, if they were to look at
one of those reports, how do you think they're going to interpret it?
So what my hope is that as you look at finding ways of making the reports easier
for physicians and more physician-friendly, that
that will trickle down to making consumer reports
simple. KIS, Keep It Simple for patients to be
able to look at and see and question themselves and
ask their physician, "Is this something that I
should be using? How is it going to benefit me?
Is there any concern for me? Is there any risk?"
So you mentioned the forest, you'd get lost in it. Well, imagine what the patient's going
through with all those trees. So my concern from
the consumer aspect is, I would very much like to
see reports on it because as a consumer, usually the only report that the consumer can get
-- yes, there is dot gov, but there's not too many
consumers that realistically go to that site to
look up reports. Pharmacists from certain 1 pharmacy chains
do give out a written report, attach it to the
prescription and give it to the patient. I would
venture to say very few take the time to read it to
see what the medication is for, what they should
look for, for side effects. There's consultation that is available.
Sometimes they'll take the pharmacist up on it if
they want to have information because they'll usually say, "Oh, no. My doctor already told
me how to use it." And what the doctor may have
told them is, "Take it three times a day before
meals," but they just heard "three times a day."
So education-wise, there's still a lot missing for the consumer. I can tell you that.
So my hope and what I'm looking for is that there'll
be more information made available for a consumer. Thank you.
DR. BARRETT: I think you're really hearing an issue of audience, and that comes through
a lot of these discussions from the standpoint of
who reads the labels. Hearing just the diversity,
we haven't even gotten halfway around 1 the room,
and it's just amazing.
I'm going to add to that because in terms of
the level of detail on the study design and results, I would be really reluctant not to
have that in there. And maybe some of this is from
my work in pediatrics, but I know I could say
when you're looking for any information at all,
and particularly if you're going to pull in data
from the literature now, some of those studies
will not fall in the category of very well-defined,
or perhaps not in a large number of subjects.
But you're still going to put it in there because it's the best information we have.
But it will be quasi-dynamic, and as new information
becomes generated, it will be replaced. But I
would still like to know the details for who this
study was in, at least the duration of therapy, the
number of patients. Are these critically ill patients or are they healthy volunteers?
That, I think, is important, maybe not from the standpoint of a rapid-fire assessment
from the standpoint of the caregiver, 1 but anyone
who's judging the validity of that information at
any point in time needs to know that detail, especially
when you see your changed control when new information becomes available. Other parts
on the label, particularly the clinical trials, are
described in more detail. So I think there's some
level of consistency by keeping that in there. In terms of the actual results, on this side
I think again you're looking for interpretation. The results are there. There is lots of numeric
data. This is always the compromise in terms of
being able to recapitulate it in the label, that
it's informative but understandable. Again, recognizing that, we would love this
to be simple, but it's typically not. There's still a lot of information that you're capturing
that we have uncertainty about. And I think the
great role that the FDA does in collaboration with
the sponsor is to do your best job at summarizing this in reasonable detail. But I think it's
just a QC check, that it's vetted against the caregiver
that they can interpret this and 1 make sense of it,
that you are delivering a message. So on that topic, I guess what I would say
is I would prefer some narrative interpreting these
results specific to dosing. And I agree. I think
the clinicians don't think in terms of even concentrations.
You may have this evolving section of the label that describes the pharmacokinetics.
So you've seen it in the beginning part of the
label, so I don't think we need to -- you want that
consistency across this. So if you're going to
describe pharmacokinetic metrics, then it's not
unreasonable to use that later in terms of judging
the results of a DDI study. But having said that,
the caregiver is past that. They want that quick
information, and talk to me about dose. The other thing that we haven't mentioned
yet is the therapeutic window. The problem I have
with the forest plot is not that it's not a quick
assessment, but it doesn't speak in the context of
a therapeutic window. For some drugs, 1 one tree, so one
presentation, may be fine. But you really need to
look at that in the context of, what's the expected
variability in the exposure for that drug? Should
I be concerned or not about it? So it's an issue of providing, I think, the
narrative that interprets the data. That is really
the key in my mind. DR. VENITZ: This Jurgen Venitz. It's still
me. And I think the question comes down to, who is
your primary target audience for the label that we
are talking about? And my personal opinion is it
is not the practitioner. It is not the patient. Because I think they get their information
from secondhand, the curating databases that we
were talking about earlier today that involve more
than just label. And they condense it in a way
that makes it usable to the practitioner.
So I don't think your target is a practitioner. I do think your target is the
kind of people sitting around this table that try
to make sense of it and find out or figure out
how to make 1 it palatable.
I'll give you an example and little anecdote. A couple of years ago I did a lecture
to a bunch of specialists, medical specialists,
on drug-drug interactions. We talked primarily
about metabolic drug interactions, and by the time
that I was done, they all enjoyed it; at least, apparently
they did. One person approached me, and he told me
that he finally understood why all those new drugs
were tested in combination with ketoconazole, a
drug that he had never used and he never anticipated using. But he now understood that
ketoconazole was not really used as an anti-fungal. It was used as a prototypical 3A4 inhibitor.
So it's that level of sophistication, pun intended, that you're going to have to deal
with. So it's not just a matter of whether we're
using forest plots or geometric mean ratios. There's
a much more fundamental lack of understanding
in the practitioners that you have to assume. That's
why they need to use databases.
So in my mind, there 1 should be detailed information on drug-drug interactions. And
I would make the argument if you are a practitioner
and you read the reproductive sections of a label,
I'm not sure whether they would understand that either.
So I'm not picking on drug-drug interaction. I'm just saying the labels have evolved to
something that goes beyond an instruction manual
for a primary care physician to figure out how to
give the drug. They use other sources to do that.
So as far as the specific information is concerned, I'm a fan for tables because they
are more informative. They also get me away from
this comparative aspect that the forest plot has.
I like to look at not only mean ratios, I do
like ranges. And I think you heard that comment
before because the 90 percent confidence interval
just tells me how confident am I that the mean
actually falls into that particular range. It doesn't
tell me anything about the range of inhibition,
if that's what is concerned. So I would like
to see the range expressed rather than the 90 percent
confidence interval on the exposure 1 metrics. In addition to that, lots of times the
half-life is not mentioned, which sometimes helps
me figure out whether the drug is really affecting absorption versus elimination. So that's something
I think on a case-by-case basis. But in addition to the exposure changes, I
do think you should discuss briefly, maybe in a
narrative or maybe in a comments section, what the
presumed mechanism is as well as what the potential
consequences are clinically. So I would put all the high level
information that a practitioner might need in the
highlights section. That's really stuff that they
ought to know and ought to understand. I think it's also important -- we didn't
discuss that in any detail other than during some
of the presentations earlier on today -- when I
teach this material to my students, I tell them,
"There are two things that you need to know. You
need to know the odds and you need to know the
stakes. Then you can gamble. Otherwise, you gamble but you're not rationally 1 gambling."
All right?
So you need to know what the stakes are. In other words, are you worried about lack of
efficacy or loss of efficacy? Or are you worried about
toxicity for whatever interaction of whatever special population you're looking at? And
that's stuff that should be up front in the highlights
section so the practitioner understands, this is
what I'm gambling with. And then the highlights section tells them enough to rationally gamble,
and if they need to know more, whether they need
to use those databases.
Yes. I think that's it for right now. DR. AU: I'm Jessie Au. So I've been a
pharmacist. I've been an academic scientist generating the type of data that you see.
Now I'm a drug developer. In all three roles, I care
about drug-drug interaction because you can imagine,
if my new drug had an interaction that would
kill a patient, that's the end of my drug.
However, it's really the fourth role that I
see that I would like to offer 1 my opinion on, and
that's the end user, is the patient or as the
mother of patients. More and more now, we don't
even go to pharmacy. We just get our drugs through
the mail. So I get a package insert. Then I say,
"Shoot, I can't read it. I don't have my reading glasses. I don't know where it is." And there's
a long list of things.
So I think that end user, other than this lady here, is really not being represented
in this particular meeting. And I think if you look
at the reality of healthcare delivery nowadays,
everybody's in a rush. I cannot tell you how many
times my physician misprescribes drugs for me -- wrong dose, wrong drug. Happens all
the time, because they didn't have time. Pharmacists,
they don't have time. Technicians hand out the
drug. So ultimately, you're really looking at the
patients. And I think now we are talking about -- even the Baby Boomers are now in
their 60s. So yes, they are becoming more and more
technology-savvy. I think we have 1 to find a way to
communicate with patients so they can take care of
themselves. Right? You cannot rely on the healthcare delivery system to work perfectly.
As a scientist, however, I do like high level of information. So on your question
number 1, I say, yes, give all the details on your
study design, study results. I think it should be
there. However, I think the communication to patients
can be done a different way, maybe not so much
information on the one page that they get from the
pharmacy. Also bear in mind that those names, those
chemical names, are very intimidating. And I say
that as a PhD in chemistry. Right? So I have a
problem with all those names. However, a patient always know what disease
they have. They know the hypertension, the type 2
diabetes. They know all that. So if you can at
least say you have these other conditions, make
sure that you check on this website for more information relating to a drug that you may
be taking that may have a drug-drug 1 interaction.
So I think that would be a good way to communicate
to patients.
In the package, however, it should be simple. It has to be, like you say, the high
level. Contraindicated, you may end in death, that's a black box warning. They should know
all those things. But they have to have a way
to get the information when they need to. So that's
the first question.
The second one about the forest plot, I like pictures. So a forest plot to me is really
easy to read. There's another plot called waterfall
plot; you're not even talking about that here. I'm
used to reading plots like that, and it's very
easy. I take one look and I know what the data means,
and obviously, because that's my work.
However, I think plots are easier to get to. You have all this explanation on the side.
I really like the forest plots. So I think the
table will get lost. The forest plot will not.
DR. MUZZIO: Fernando Muzzio, Rutgers University. So I'm neither a prescribing 1
physician nor a pharmacist, so I'm going to give you
a perspective from the point of view of perhaps
an engineer, and somebody who teaches experimental
design, and somebody who has an 85-year-old mother
and a 96-year-old father-in-law. So let me start with the last because as
fate will have it, both of these people happen to
be in the hospital right now for separate reasons.
And both of them in the last 30 days were given the
wrong medication. I think it's a fact we all know
that older people are the people most likely to be
taking multiple medications. Now, in the case of the people in my family,
none of the doctors that see them actually know
what it is that they are taking because they go to
three different doctors. The doctors don't talk to
each other. These people are both memory-impaired, so they cannot recite the six or seven or
eight things they are taking. And they don't get
all their things from the same corner pharmacy.
Yes? So there is no place right now where all
that information is except in the 1 mind of my wife
and my sister. They are the two that actually keep
track, neither of which is a doctor. Right? My
wife is a pharmacist. But in both of these cases, we figured out
they were getting the wrong medication and there
were interactions because somebody in the family
took the time to actually read the labels and found
that, oh, my God, they shouldn't be taking this if
they are taking that. So yes, you might think that you're only
writing this for the doctor, but in fact, I think
these kind of situations call for a lay person being able to, at least on a very basic level,
ask the right question. Okay?
So moving on now to on a more scientific basis, I don't understand question b at all.
From the perspective of somebody who's actually
written a lot of papers, some of the papers I write
also have multiple audiences.
They go to the PhD student, who's really going to read it closely; to the professor,
who's only going to look at the abstract; 1 to the
person in industry, who's only going to look at the
pictures, maybe. What's wrong with that? We are
talking to multiple audiences. Why don't we use multiple ways of conveying
the information? Some people capture the information better in a picture, some people
get it out of a table, and some people actually want
to read every word.
I actually really, really like the suggestion about maintaining a website with
all the appendices and all the other stuff that the
statistical geeks like me are actually going to
want to know. When I teach experimental design, I
teach to my students, but it's ridiculous to look
at whether a variable is statistically significant or not if you didn't look at the design because
you can look at only main effects or you can look
at interactions. And guess what? Your conclusions
about what's significant will change. So if you
don't know the design, you know? So I hope that there are ways in which we
can use modern tools to convey information 1 to make
the information available in different formats to
the different audiences that might need it for
different reasons. DR. PAU: Thank you. Alice Pau from NIH. I
guess I'll give a background. I use the package insert probably every single day as two purposes.
I'm a clinical pharmacist; I do take care of
patients in our clinic, and get asked questions almost every day about drug interactions.
I don't memorize all these drug interactions, and in many cases are drugs
that I'm not familiar with, so I have to go and look
it up. And that's where I find discrepancies between
different labels that don't have the information in
all of them. Secondly, my other role is to write
treatment guidelines for ***, which, as we heard
over and over again, that there are multiple drug20
drug interactions. So I go to the labels to look
up the information so that I can translate that
into our guidelines. So for that purposes, there 1 are two things
that I think are important. Dr. Horn said, and I
totally agree with, please, please put everything about drug interaction into one section. Many
times on a daily basis when I look at these, I
missed one section or another because I have to go
from one place to another to a third place, and
sometimes I missed some information that could be
crucial. It could be very easy to have a section just
called drug interaction and have all the information in there, and particularly important,
to try to translate the clinical or the PK data
into recommendation because you have one place that
give you the data, and then you have to go to
another place to look and see what the recommendation truly is.
The other thing that's also difficult when I
look at these tables is that if you have a drug
interaction study that is done that is going to
look at interactions of the two drugs and have PK
data on both drugs, why not put them in the same
table of drug A, drug B, this is 1 the end result of
drug A and this is the end result of drug B?
Right now we have to go to two separate tables, the first one to say, this is what
it does to the sponsor's drug, and then you to go
a second table to say, this is what happened to the
other drug, when they can be put in the same table.
When we make decision, we make decision together to decide on what to do and not separately
have to go two different tables. And if you look
at the clinicians on a daily basis, they might not
have time, and oftentimes what will happen is that
they go to just one table, expect that that information is there, and then stop right
there and not go to the second one.
So I would really recommend putting everything into one place all at one time,
including both the data. And I like data because I
need them. I also like the data to know that is
this a study that is a single-dose study versus multiple-dose study. How many patients? Is
it healthy volunteer versus this being used in
patients? Because there might 1 be a difference. The second thing that I think that I have
not seen in any of the labels are relating to what
is the role of therapeutic drug monitoring. There
is no mention -- and there are many drugs that have
commercially available drug concentrations that can
be monitored, and we use them all the time in my
clinical practice. If I'm using rifampin or rifabutin with a
drug, I always would monitor the drug level to make
sure that I'm getting the right drug level. There's no mention whatsoever if there is
a role of therapeutic drug monitoring. I put in my
guidelines I recommended for the clinician to do
it. But it is not anywhere in the label to be
seen. The third thing I wanted to mention is that
most of the information in the label relate to, as
we mentioned before, studies that were done by the
sponsor. I want to give an example, atazanavir with PPI or atazanavir with antacid.
The current label has very difficult to interpret information about 1 how to take
them together at the same time. You have space
it by 2 hours before, 12 hours before, whatever.
In fact, when the label was going to be put out,
I was given the language to review, and I drew a
line of a 24-hour line and see how I'm going to teach
my patient how to take the medicine. And I was
totally confused. And I don't know how a pharmacist or a doctor can teach the patient,
what does it mean by taking this drug 12 hours
before that drug, not to take it 2 hours later?
Since then, there have been multiple drug interaction studies with atazanavir and PPI
that were done by individual investigators using
different strategies -- different time, different doses -- and come up with different results.
None of those got into the label.
So the question is, for the consumers, if there are other results, other ways of taking
these medications that might be easier for them
to do, and those information are not available for
them, how can they get that information outside
of it? So I think those type of information 1 is
very important.
Lastly, about what information are not necessary -- well, I talk about these. So
for me, as far as the forest plot versus table, I
like table better than forest plot. I mean, I
actually -- reading the material, and I share it
with multiple of the clinicians in our clinic and
ask them, do they like the forest plot? They say
no. This doesn't give me the information. The main reason is part of it is, especially
if you are talking about the forest plot of the
multiple different other drugs that has different therapeutic windows, there are different
significance in terms of the interactions, it's
very difficult to interpret what that really means.
I guess we are more used to numbers, and that's the reason why I like the table much
better. And the table being able to give us the information
about the study design also helped me as well.