Back to the Top
Hi All,
When analyzing data from Phase 1 studies, I occasionally see an
unexpected high value for one subject and an unexpected low value for
another subject, at the same time and day of blood draw. These I
consider to be labeling errors, probably occurring in the lab after
the samples are centrifuged.
My approach is to plot the data as received for the two subjects in
one plot, and again after having interchanged the values. I also make
a plot of all subjects in the group who were dosed that day to make
sure that the switch could have been made between these particular two
subjects only and that there is no other switch thinkable. Once this
is established I go ahead using the interchanged values for AUC
calculations etc. And all of this is of course accurately and openly
documented in the study report; the values from the Analytical
Laboratory are listed as received and we flag the (odd) ones that were
interchanged. I see no harm in this approach, as long as it is
documented. I see much more harm in making incorrect calculations,
when my common sense tells me that there is an error in the data set.
This of particular importance when evaluating the kinetic profile of a
drug in Phase 1 studies.
This approach may not acceptable to the FDA (I hear) when analyzing
BE or food effect studies in healthy volunteers, but am not aware of
any guidance on this. Does anyone have any experience with this issue
e.g. an analysis / study result of a BA/BE/ Food effect study was
rejected because of this, or did you actually get written or verbal
information from the FDA saying that data have to be used as received,
no matter what?
Thanks a lot in advance for any help,
Frieda
Back to the Top
Dear Frieda,
We experienced swapping issue with subjects on placebo during phase 1
multiple ascending dose study. This was occurred mainly at Cmax point
(2 out of 6 subjects on active in middle dose group). We decided to
calculate only Thalf and Kel from those subjects.
Regards,
Jignesh
Back to the Top
Dear Frieda,
I am not aware of any written regulatory guidance that allows changing
data just because it doesn't look the way we would like it to look.
However, without revealing proprietary information, I can say I have
seen successful applications of your proposed approach during nearly
20 years of work at multiple pharma companies and CROs while I worked
at Syntex, Roche, and Elan. Many of the elements you mention are key
to success:
1. Clearly and transparently show calculations and results both
with and without the switch.
2. Studies that produce the data are well powered, such that
calculations on data as-is do not alter the general conclusions.
3. Having additional circumstantial evidence that supports the
suspicion (contemporaneous blood draws, subjects, etc).
4. Values that differ so widely from expectations that explaining
them becomes even more difficult than justifying a change (e.g., high
drug in placebo and zero drug at Cmax in profile of a subject dosed at
same time).
5. Other measurable markers or properties of the sample that
conclusively identify the donor of the sample.
6. A slightly modified approach is simply to "censor" (exclude)
the suspicious values in a calculation, rather than switching them.
The argument for switching still suffers from lack of causal evidence,
and leads to suspicion of wider but less obvious sample mishandling
(Which other samples may have been switched, but are not as obvious
from the data? How pervasive was the erroneous procedure?). In some
cases, pharmacokineticists will simply present the data as labeled for
the primary calculation, if it still is consistent with other study
conclusions, and flag the values as suspicious to let a reviewer know
that they recognize the values as suspicious.
I would note also that although such events are generally rare, I have
generally seen the evidence for the mislabeling more often associated
with the sample collection process than the bioanalytical lab, though
certainly not 100%.
I'd be interested to hear others' perspectives.
Tom
Thomas L. Tarnowski, Ph.D.
Bioanalytical Development
Elan Pharmaceuticals, Inc.
800 Gateway Boulevard
South San Francisco, CA 94080
thomas.tarnowski.aaa.elan.com
Back to the Top
The following message was posted to: PharmPK
Dear Frieda!
You wrote:
> When analyzing data from Phase 1 studies, I occasionally see an
> unexpected high value for one subject and an unexpected low value for
> another subject, at the same time and day of blood draw. These I
> consider to be labeling errors, probably occurring in the lab after
> the samples are centrifuged.
>
> My approach is to plot the data as received for the two subjects in
> one plot, and again after having interchanged the values. I also make
> a plot of all subjects in the group who were dosed that day to make
> sure that the switch could have been made between these particular
two
> subjects only and that there is no other switch thinkable. Once this
> is established I go ahead using the interchanged values for AUC
> calculations etc. And all of this is of course accurately and openly
> documented in the study report; the values from the Analytical
> Laboratory are listed as received and we flag the (odd) ones that
were
> interchanged. I see no harm in this approach, as long as it is
> documented. I see much more harm in making incorrect calculations,
> when my common sense tells me that there is an error in the data set.
> This of particular importance when evaluating the kinetic profile
of a
> drug in Phase 1 studies.
I agree with you. This also demonstrates the importance of a (blinded)
plausibility review of analytical data.
For a nasty example see here:
http://bebac.at/lectures/Ensuring%20bioanalytical%20compliance%20of%20your%20BA-BE%20study.pdf#page=49
(Slide 49 and the next slide)
The example derives from a BE study (don't worry about the 'strange'
profiles: two-phase release formulations) where a barcode-tracking
system was in place, which was malfunctioning on this particular day and
the SOP in such a case calls for a 'four-eye-method' (ie, a second
person watches the plasma transfer after centrifugation). Unfortunatelly
no peculiarities were documented. We re-analysed the suspected samples
and two adjacent values in the profiles of both subjects - original
values were confirmed. Since citrate plasma was used most safety-lab
values could not be estimated with the exception of GGT and albumine.
Luckily the two subjects had quite different values in their pre-study
screenings, so the mix-up could be justified based on their lab-values.
> This approach may not acceptable to the FDA (I hear) when analyzing
> BE or food effect studies in healthy volunteers, but am not aware of
> any guidance on this.
Me not either - but I would expect that FDA would not accept this
approach... :-(
> Does anyone have any experience with this issue
> e.g. an analysis / study result of a BA/BE/ Food effect study was
> rejected because of this,
No (concerning European countries only).
Regards,
Helmut
--
Ing. Helmut SchA\0x00tz
BEBAC - Consultancy Services for
Bioequivalence and Bioavailability Studies
Neubaugasse 36/11
1070 Vienna, Austria
e-mail helmut.schuetz.-a-.bebac.at
web http://bebac.at/
contact http://bebac.at/Contact.htm
forum http://forum.bebac.at
Back to the Top
Why would you even interchanging the data?? The first thing your
should do is confirm with the clinic that 1) the patient did not
recive his dose or sampling early or late (not that that ever happens)
2) there was nothing mislabeled (time etc). After that, if they are
outliers note this and process the data both inclusive and exclusive
of the outlier points. Don't guess! Don't assume.
--
Ed O'Connor, Ph.D.
Laboratory Director
Matrix BioAnalytical Laboratories
25 Science Park at Yale
New Haven, CT 06511
Web: www.matrixbioanalytical.com
Email: eoconnor.aaa.matrixbioanalytical.com
Back to the Top
Ed,
The situation I am describing occurs in Phase I studies where lots of
samples are processed at the same time and place, and mistakes will be
made, you can count on it. Man is fallible and will make mistakes, on
average about 0.5 %. This cannot be prevented, however much you train
people. Even with a 100% QC someone will make a mistake somewhere.
So when in our lab the blood samples are centrifuged and the plasma is
pipetted out in pre-labeled duplicate tubes that will go in the
freezer till shipment to the Analytical Lab, very occasionally,
infrequently, an error will be made. The plasma may be pipetted into
tubes with the label for another subject who was dosed at the same
time. This can happen if the centrifuged tubes are not placed in the
correct sequential order in the rack before pipetting out, or maybe
the Eppendorf tubes were not placed in the right order. So there is no
point in checking labeling, because you cannot track where it went
wrong.
Then I get the concentration data, and a placebo subject has valueshere the concentration is 85.6 ng/mL. And one of the subjects
receiving active treatment has a nice concentration curve with a Cmax
of 197.3 ng/mL at 1 h post-dose, with a steady decline thereafter
except for the value at 3.5 h post-dose which istells me that I need to interchange these values. The probability that
the placebo subject has a true high value at one single time point is
nil. And the probability that this < LOQ value belongs to the subject
on active is also about nil, when you look at the entire data set. And
I don't believe this is assuming, or guessing, or brushing up my data
set to make it look better. The proviso is of course that it is only
done when there is no doubt whatsoever which values need to be
interchanged. The alternative is to set both values to missing, which
is more or less the same, not using a value that you sincerely believe
cannot be right, based on the set of values you have.
Frieda Ebes
Senior Pharmacokineticist
Kendle International B.V.
Bolognalaan 40
3584 CJ Utrecht
the Netherlands
Back to the Top
The following message was posted to: PharmPK
One point is that the error may have been pre-analytical, the clinic
may have annotations such that the patient was actually dosed early or
late or that the collection was early or late (day 5 rather than day
10) because the patient would not be available. I agree that if the
label was not verified before tansfer there is no use in tracking it
back. If there is the possbility of a parallel sample-one taken for
safety assessment not PK, you could use the remainder to verify the PK
results, given that that the error was not pre-analytical. Best as
Tom suggested earlier- present data with and without points excluded
as outliers.
--
Ed O'Connor, Ph.D.
Laboratory Director
Matrix BioAnalytical Laboratories
25 Science Park at Yale
New Haven, CT 06511
Web: www.matrixbioanalytical.com
Email: eoconnor.-at-.matrixbioanalytical.com
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 "Handling of suspected sample labeling errors" as the subject | Support PharmPK by using the |
Copyright 1995-2011 David W. A. Bourne (david@boomer.org)