Back to the Top
Hello everybody,
Hope someone has a suggestion i would be very grateful.
I am in preparation for a BE study, standard crossover, the drug has a
very long half life, and I would like to have washout around 4 weeks,
but even then there is a possibility for a predose concentrations in
second period. Can someone instruct me how can I adjust concentrations
of subjects that are going to have this predose detectable values. I
have been running into published studies where this adjusting was
performed but I never had the opportunity to see how is this performed
what is the method for this kind of adjustment.
The increase of washout period to completely avoid this scenario is
not an option, due to many reasons!
Thx in advance,
Lidija
[I have a distant memory of a method/reference in an older mid level
stats book that took carryover into account. Another approach might be
to simultaneously model both doses, as a multiple dose regimen with
different ka and F but then stats may be more difficult - db]
Back to the Top
The following message was posted to: PharmPK
Lidija,
For long elimination half-life drugs the FDA accepts parallel study
designs. Have you consider this option?
In addition, blood sampling is required for at least 3 days (enough to
ensure that gastrointestinal transit is completed) and truncated areas
up to 72 h from dosing for drugs that have low intra-subjects
variability in distribution and clearance can be used in place of AUCt
or AUCinf.
Best
Stefano
Stefano Persiani, PhD
Director,
Translational Sciences and Pharmacokinetics
Rottapharm spa
Via Valosa di Sopra, 9
20052 Monza (MI)
ITALY
e-mail: stefano.persiani.aaa.rottapharm.com
Back to the Top
The following message was posted to: PharmPK
Hi Lidija,
What regulatory body are you submitting to? In general, baseline
subtraction is only permissable for endogenous substances. It is not
advisable to adjust the Period 2 concentrations based on the pre-dose
values. Usually in the protocol you would indicate that if the pre-
dose value for a subject exceeds 5% of their Cmax, you would exclude
their data from the statistical analysis. For situations where a pre-
dose is anticipated in all periods because of background amounts of
the drug in the blood, you would take 3 pre-dose samples, take their
average, and subtract that number from the entire curve of that
subject. Any negative concentrations resulting from baseline
subtraction would be set to zero.
If you know ahead of time that the washout will be insufficient, why
wouldn't you either plan a longer washout, or consider running a
parallel study? It is preferable to design bioequivalence studies to
succeed, rather than to knowing ahead of time they may fail and plan
questionable methods in order to salvage the data. An insufficient
washout could introduce period effects, and in any event will lower
the power of your study.
Hope this helps,
David Dubins
--
David Dubins, Ph.D., B.Eng.
Global Bioequivalence Consulting
Assistant Professor, Leslie Dan Faculty of Pharmacy
University of Toronto
Back to the Top
Hi Koren,
Why dont you plan a parallel design??!!
Cordially,
Dr. Renu Jain
Research Associate
Clinical Research & Regulatory Affairs
Torrent Research Centre,
Village: Bhat, Dist: Gandhinagar - 382 428
Back to the Top
The following message was posted to: PharmPK
Dear David,
Thank you for your answer, it is very helpful.
This study that I am planning is pilot study and the only reason that we
are doing it is to get ISCV, I don't have much time for this study so I
need to shorten washout period.
For the pivotal study we still don't have clear strategy, so maybe it
will be parallel and maybe crossover and more data we have the better
decision we are going to make. This is for EU submission and I think
that EU regulatory authorities don't have this 5% of Cmax as border for
taking the subject in or out of stat analysis, do you have some other
information?
Can you just clear something up, 3 pre-dose samples?
Is this 3 pre dose samples taken for every subject, or is this 3
analysis done on one pre dose sample?
Have you ever done something like that?
Regards,
Lidija
Back to the Top
The following message was posted to: PharmPK
Hi Lidija,
As pilot studies are small, the reliability of the ISCV estimate is
limited. In the absence of any other data, it's better than nothing,
but the most valuable information to come out of a pilot isn't the
ISCV at all, it's the actual Test/Reference ratio. Why trust a
variability estimate from 6-12 subjects? You could run a larger pilot
(16-18 subjects), which will give you a better idea. I have used the
5% of Cmax rule with EU submissions, as long as it's in the protocol,
it's an accepted method. Did your literature search turn up any
estimates of the drug variability? Is it low or high?
To answer your question, three pre-dose concentrations (at different
time points) are planned for each subject (e.g. -2 hours, -1 hour, 0
hours). The 3 pre-dose samples are analyzed with the rest of the
curve.. The average of a given subject's 3 pre-dose concentrations is
subtracted from the rest of that particular subject's curve. However,
the assumption of using this method is that over the sampling
schedule, the baseline levels won't change (which isn't even valid for
endogenous substances). This assumption certainly isn't valid for
adjusting for carry-over, because the "baseline" in this case will be
diminishing throughout Period 2.
You can be comfortable in the fact that for a pivotal study, if the
drug isn't endogenous, baseline correction will not be accepted.
Perhaps it would make an interesting secondary analysis, but not a
primary one. Now, given that this is a pilot study, you can use any
method you want to analyze the data and not worry about the
acceptability of baseline normalization from a regulatory point of
view. If your hands are tied and you must run a 2-period pilot, I can
dream up a completely unconventional and inappropriate way to adjust
your data. Since I don't know the identity of the drug, my next
question would be, are the pharmacokinetics linear with respect to dose?
Here is one possibility:
Unless flip-flop kinetics are expected, the half-life of a drug in a
subject shouldn't *really* be expected to change by orders of
magnitude depending on the formulation. In addition, we can try making
the assumption that the residual Period 1 drug we measure at pre-dose
in Period will decay the same rate that we estimated in Period 1. This
is also assuming you've sampled out far enough to cover the
elimination phase (truncated AUC won't cut it here, you should cover
at least 80% of your curve). If we make this assumption, you can use
the Subject's Period 1 half-life to baseline subtract their Period 2
data, and you won't need 3 Pre-dose samples to do this (in fact, it
would work better with just one). What you can do is model the "end of
the curve" using the same assumption (and leap of faith) that is used
to calculate AUCinf: monoexponential decay (Clast/lambda). The
baseline "function" for Period 2 would look like this for each subject:
(1) C_baseline(t) = Cpd_P2 * exp(-lambda_P1*t)
C_baseline(t): The theoretical concentration of a monoexponential
decay baseline, starting from the pre-dose collection time at Period 2
(Cpd_P2) and a Period 1 terminal elimination rate constant of lambda_P1.
To be really good about this, since we are using the Pre-dose and it
isn't actually collected at zero hours, we should adjust for it.
Record *exactly* when pre-dose is collected (let's call it "tpd" - to
bad this isn't a Canadian submission. Sorry, bad regulatory joke.)
So then, each concentration in Period 2 is adjusted with the
theoretical baseline in (1) at the actual time Period 2 concentrations
are collected. Don't forget to account for blood sampling deviations.
(2) Ci_adj(ti) = Ci - Cbaseline(t+tpd) = Ci - Cpd_P2 * exp(-
lambda_P1*(ti+tpd))
Ci_adj : the baseline-subtracted Period 2 concentrations, calculated
by starting with the actual concentration (each Ci) collected at ti,
and subtracting equation (1) calculated at that timepoint.
This wouldn't be acceptable from a regulatory perspective at all, but
then this is a pilot study. It's also not wonderful science because it
will fail if the following (and probably other) criteria are not
satisfied:
- PK are not linear (i.e. if the drug levels are kicked up too high in
Period 2: auto-induction or auto-inhibition? Saturated metabolism or
elimination? Lots of ways this can go wrong.)
- Flip-flop kinetics
- The lambda could not be estimated in P1
- The lambda in P1 is much different than P2 - the very nature of
intra-individual variability?
- The decay after P1 was not actually monoexponential (multi-
compartmental elimination will massacre this whole idea)
- Murphy's law? It always seems to show up.
I thought about adjusting the P2 baselines using the lambda estimated
in Period 2, but then I thought that is sort of like a surgeon
operating on themselves. Any baseline correction you decide to go with
will have these drawbacks, which is why planning a proper washout is
preferable, not to sound like a broken record. Whatever method you
decide to go with, suture-self! (Sorry, another bad joke).
Hope this helps,
David Dubins
--
David Dubins, Ph.D., B.Eng.
Global Bioequivalence Consulting
Assistant Professor, Leslie Dan Faculty of Pharmacy
University of Toronto
Back to the Top
Dear Dave,
Here are some additional comments,
> As pilot studies are small, the reliability of the ISCV estimate is
> limited. In the absence of any other data, it's better than nothing,
> but the most valuable information to come out of a pilot isn't the
> ISCV at all, it's the actual Test/Reference ratio. Why trust a
> variability estimate from 6-12 subjects? You could run a larger pilot
> (16-18 subjects), which will give you a better idea.
Yes, this is going to be a large pilot study with around 20 subjects.
I know that variability and C.I. is not something you can trust when
studies are small, but I am a bit confused regarding Test/Reference
ratio, isn't this also largely influenced by the number of subjects
especially with highly variable drugs.
For example: T/R ratio around 80%, variability around 70% and all that
gathered on a population of 20 subjects. Is this something reliable?
Or do to this high variability and small study population even T/R
ratio is in question? Maybe the study population just doesn't
adequately represent general population and in that way doesn't give
us reliable T/R ratio.
> I have used the 5% of Cmax rule with EU submissions, as long as
it's in the protocol, it's an accepted method.
Thank you for sharing your experience!!!!
> Did your literature search turn up any
> estimates of the drug variability? Is it low or high?
Well this is something confusing, in the literature it is stated that
INTER- and INTRA-subject variability is high, but we did really small
study with just one arm and INTER-subject variability turned out not
so high even <30%, so now is the question is INTRA-subject variability
as high as claimed in the literature. I was reading some discussions
regarding approximation of INTRA-subject when INTER-subject
variability is known but I didn't find anything useful. As a see it
this INTER-subject is in most cases higher than INTRA-subject. Do you
have any other thoughts?
Thank you for proposed calculations but I am afraid that it will not
be applicable for this drug as it has bi-exponential elimination.
Regards,
Lidija
Back to the Top
The following message was posted to: PharmPK
Three animals per group. Some studies include a sample 1 week before
dose
and another just before dosing. Three analyses on the same sample is
excessive unless you are doing that for all samples and need it to
generate
a reasonable CV. You are not shortening the washout period unless you
are
enhancing the clearance of drug. What you may be doing is sacrificing
the
washout interval in the interest of time. That may lead to difficulty
in
interpretation of the PK of the subsequent drug.
Back to the Top
The following message was posted to: PharmPK
Hi Lidija,
>For example: T/R ratio around 80%, variability around 70% and all that
>gathered on a population of 20 subjects. Is this something reliable?
If your T/R ratio is around 80% unless you're planning on running a
replicate study in order to use the EMEA's widened confidence limits,
this looks like a poor drug. If the 1-arm pilot you ran was 20
subjects, I would wonder about whether or not the true Test/Reference
ratio was captured. You will get a better idea from a crossover. All
you can do at this point is hope the your pilot doesn't represent the
general population. One of these days I'm going to find the power
equation for the Test/Reference ratio alone passing BE limits
(appropriate for Candian studies on Cmax for Part A drugs) and then I
can actually put some numbers behind calculating power based on sample
size. If you already have pilot data, you can try bootstrapping the
data to see what happens.
> I was reading some discussions
> regarding approximation of INTRA-subject when INTER-subject
> variability is known but I didn't find anything useful. As a see it
> this INTER-subject is in most cases higher than INTRA-subject. Do you
> have any other thoughts?
Here are my thoughts: you don't know intraCV if you've run a 1-arm
pilot, period. However, as a starting point, it is pretty safe to say
that intraCV won't be larger than the interCV. I've only seen that
happen VERY infrequently, for example when there is an outlier in a
crossover study who throws off the whole analysis. There are a few
rules of thumb out there that people cringe about using, like
IntraCV=0.5*IntraCV, Intracv=0.6*intraCV, etc. These estimates don't
hold any water statistically and are better than a shot in the dark,
in the absence of any other data. For instance, if my GPS isn't
working I might roll down the window at the next stoplight and ask the
person next to me where a street is. Odds are pretty good that they
will point me in the right direction. It's better than nothing.
Notwithstanding, a small 1-arm pilot data (e.g. 6-12 subjects) would
be a poor representation of the variability, and taking a cut of this
to represent intraCV
would be reckless. This is why a crossover/replicate pilot would
help you.
I don't envy your position. You have to make a decision about a high
variability drug, you don't have enough time to run a proper study,
and your pilot data doesn't look so great. Here is an idea: consider
running a 2-period replicate pilot with around 24 subjects (4 groups:
AA, AB, BA, BB). This would be 6 subjects per group, which is lean in
terms of power, but if the intraCV is really over 70% then it will be
great justification for using those 75-133% goalposts. I haven't seen
this design for pivotals. The Question & Answers EMEA document
specifies in order to use widened confidence intervals, you need to
demonstrate high variability (intraCV > 30%) by running a replicate
pilot study. Normally this would be a 3- or 4-period replicate study,
but you have no time. A 2-period replicate will provide you with an
estimate of intraCV for both Test and Reference products. You've
already said there is no time to wait for a second period, but you've
already
run a parallel pilot, what are you hoping to gain from another one?
It sounds like a waste of time and money to me. This design would
never fly for a pivotal (you actually lose power compared to a 2-way
crossover) but on the bright side, in addition to giving you an
estimate of intraCV. The drawback is that I've just told you above
that an intraCV calculated from 12 subjects isn't reliable, and in the
same breath I recommended a design with 6 subjects to estimate intraCV
for treatment A, and 6 for treatment B. It will be better than 0.6*CV
though, and from a regulatory perspective you've satified running a
replicate study in only 2 periods.
Good luck with whatever you decide,
-Dave
--
David Dubins, Ph.D., B.Eng.
Global Bioequivalence Consulting
Assistant Professor, Leslie Dan Faculty of Pharmacy
University of Toronto
Back to the Top
The following message was posted to: PharmPK
Dear Dave,
It seems that my explanations are pretty poor, cause I didn't mention
that example used for T/R ratio with high variability was in fact
totally different drug and I used it just to see if in this kind of
situation T/R ratio is reliable. This example was a cross-over study.
Example had nothing to do with possible predose concentrations in pilot
study that we were discussing all the time.
Sorry for this draw back as I didn't realize that this can be a bit
confusing.
For the drug that I have to run a crossover pilot study and I don't have
enough time for the wash out period I don't have intra-subject
variability nor T/R ratio and these are two main numbers that I am
seeking.
I only have inter-subject variability from one arm pilot study (with
reference only) that was intended to see how long t1/2 really is
(literature data were confusing on that as well), so that adequate
washout can be employed in pivotal study if we decide to go with
cross-over design.
Hope this is more clear.
Regards,
Lidija
Want to post a follow-up message on this topic?
If this link does not work with your browser send a follow-up message to PharmPK@boomer.org with "Predose concentrations" as the subject | Support PharmPK by using the |
Copyright 1995-2011 David W. A. Bourne (david@boomer.org)