- On 28 Sep 2007 at 00:53:37, faith okalebo (okalebof.at.yahoo.com) sent the message
Dear All,
I am doing a pharmacokinetic study in mice and for one vehicle all
most concentrations are below the lower limit of quantitation except
the Cmax values. How does one use such data and what are the
guidelines regarding use of values that below LLOQ especially with
regard to determination of the terminal half life? I am using
WinNonLin for data analysis.
FAITH A OKALEBO
Ph.D STUDENT,
UNIVERSITY OF CAPE TOWN
[I recently updated the PharmPK archive. You might want to search for
LLOQ or similar to what has been said before - db]
Back to the Top
- On 1 Oct 2007 at 09:16:54, "Ed O'Connor" (eoconnor.-at-.matrixbioanalytical.com) sent the message
The following message was posted to: PharmPK
Values below LLOQ are values below LLOQ and should not be reported other
than "or CRO to "give" you a value but you are putting them and yourself at
some risk in using such data.
While there are alternative approaches which marginalize the concept of
LLOQ they have not be accepted by regulatory bodies as yet.
--
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
Back to the Top
- On 1 Oct 2007 at 17:40:46, "Rob ter Heine" (Rob.terHeine.-a-.slz.nl) sent the message
The following message was posted to: PharmPK
Ed,
It depends what the data is used for and whether the LLQ is really an
LLQ or just an arbritary chosen cut-off value. I can imagine that an
estimation of a value below the llq, can give you a lot of valuable
pharmacokinetic information. IMHO an estimation of a value for a PK
study might even be better than an omitted value.
Cheers,
Rob ter Heine
--
Rob ter Heine, MSc, PharmD
Department of Pharmacology, Slotervaart Hospital
Amsterdam, The Netherlands
Back to the Top
- On 2 Oct 2007 at 08:27:47, Nick Holford (n.holford.at.auckland.ac.nz) sent the message
The following message was posted to: PharmPK
Ed,
> Values below LLOQ are values below LLOQ and should not be reported
other
> than "folks
> or CRO to "give" you a value but you are putting them and yourself at
> some risk in using such data.
To my knowledge there are no regulatory guidelines that explicitly
forbid the use of measurements from using the truth seems to me to imaginary. If you know of a
regulatory guidance that forbids the use of measurements please give us the details.
There is a certainty (not a risk) that your analysis will be biased
if you selectively discard measurements below the LLOQ. This is a
very simple statistical fact. Imagine 3 sample measurements 0.9, 1.0,
1.1. The average is 1.0. Suppose the LLOQ is 1.0 and you discard the
value of 0.9. The average of the remaining values is 1.05. It is
biased (5% higher) compared with the best estimate of the mean
obtained from all the measurements.
Pharmacokineticists that use compartmental models can find themselves
concluding there is an extra 'terminal phase' compartment because of
the bias which forces all measurements of true values close to the
LLOQ to be >=LL0Q.
There are principle based methods (see below for references) for
minimising this bias if the chemical analyst is constrained by ill-
conceived SOPs and refuses to report the actual measured value.
Scientists who wish to take a more careful approach to data analysis
will encourage chemical analysts to tell the truth, the whole truth
and nothing but the truth.
Nick
Methodology: Beal S. Ways to fit a pharmacokinetic model with some
data below the quantification limit. Journal of Pharmacokinetics and
Pharmacodynamics. 2001;28(5):481-504.
Application: Hennig S, Waterhouse T, Wainwright C, Bell S, Miller H,
Charles B, et al. A D-optimal designed population pharmacokinetic
study of itraconazole capsules and solution in adults with cystic
fibrosis. www.page-meeting.org/?abstract=884. 2006;15(Abstr 884).
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
n.holford.at.auckland.ac.nz
www.health.auckland.ac.nz/pharmacology/staff/nholford
Back to the Top
- On 2 Oct 2007 at 09:06:58, sylvain.mandeville.-a-.labresearch.hu sent the message
Dear Nick,
There seems to be a difference of opinion on this forum whether you
are a pK scientist or an analytical chemist.
Both sides have valid explanation/conclusion however, at some point a
decision has to be made, what to do with value which are below the LOQ?
With the present guidelines, the LOQ must have a signal to noise (S/
N) of at least 5:1, and thus for validated methods in the range of pg/
mL or ng/mL range, most of the methods validated are very close to
this limit.
Presently, I witness that the bioanalytical methods are tailored.
For example of a BE study with a Cmax of 50 ng/mL, we would divide
this Cmax by 64 (6 half-lives) and this would determine a potential
LOQ (~0.750 ng/mL). Then during R&D this potential LOQ would be
attempted (always bearing in mind to a 5:1 S/N is required with the
extract samples and not in solution). The method is then validated
with determination of the accuracy and precision of the LOQ
throughout the validation (i.e. QC LOQ are extracted with every run
as well as a intra run determination). Once the validation is
completed, the accuracy and precision of the QC LOQ is most often
between 10-20% since your are very at the limit of the instrument
sensitivity.
Thus, I feel it is very difficult to give clear cut recommendation in
forum like this one until you actual see the validated method and
verify the S/N at the LOQ.
In addition, the FDA guideline on Bioanalytical method on page 14,
first paragraph "Estimation of concentration in unknown samples by
extrapolation of standard curves below LLOQ or above the highest
standard is not recommended. Instead, the standard curve should be
redefined or samples with higher concentration should be diluted and
reassayed." This is assuming the method is not already at the limit
of the instrument capability.
Maybe there is no guidelines for pK analysis and the use of values
below LOQ, however the bioanalytical laboratory which are following
the FDA guidelines have certain limit the validation must adhere to.
Finally, I don't know if you have been ever subjected to an FDA
audit? from my personal experience during bioanalytical audit, don't
report values which are below LOQ, simply flag them as BLQ (below the
limit of quantitation) if you don't want to received a citation (i.e.
a 483)
I hope this explanation will help you understand the analytical point
of view on this debate.
Best regards,
Sylvain
Sylvain Mandeville, Ph.D.
Back to the Top
- On 2 Oct 2007 at 09:11:18, Sven.Poetzsch.-a-.merck.de sent the message
The following message was posted to: PharmPK
Dear Nick,
Even as an analytical scientist I understand your concerns ;-).
Nevertheless, to cite the FDA Guidance for Industry on Bioanlytical
Method
Validation:
"The calibration (standard) curve should cover the expected unknown
sample
concentration range in addition to a calibrator sample at LLOQ.
Estimation
of concentration in unknown samples by extrapolation of standard curves
below LLOQ or above the highest standard is not recommended. Instead,
the
standard curve should be redefined..."
So, what can we do for you. Push the Analyst to set up an analytical
method
that can resolve a LLOQ that is 1/500 or 1/1000 of Cmax.
Then the LLOQ discussion will end up with the result that the expected
analytical error will be greater than the inaccuracy due to not
explicitly
defined concentrations below the LLOQ and both "parties" will hopefully
feel OK.
Best wishes
Sven
Dr. Sven Poetzsch
Head of Laboratory
Central Analytical R & D . GLP - quantitative Bioanalytics
Email: sven.poetzsch.at.merck.de
Back to the Top
- On 2 Oct 2007 at 13:18:11, "Satya Jadhav" (satya.jadhav.at.gmail.com) sent the message
The following message was posted to: PharmPK
Can't we use the extrapolated values (below LOQ) for PK analysis
while reporting them as BLQ to regulatory authorities???
Back to the Top
- On 3 Oct 2007 at 09:19:41, "Stephen Duffull" (stephen.duffull.-a-.stonebow.otago.ac.nz) sent the message
The following message was posted to: PharmPK
Sven, Sylvain and others...
We have this discussion about every 6-12 months. It never resolves -
and
presumably won't resolve in the foreseeable future.
If a sample is below LLOQ why not do both things?
i.e. report the actual value and report that it is below LLOQ at the
same
time. Then the person who gets the sample can determine what to do
with it.
It should be noted, as well, that a sample that is claimed to be
below LLOQ
isn't necessarily actually below LLOQ. So you have the additional
choice of
repeatedly analysing the sample until you find a value that is above the
LLOQ and then reporting it (ignoring all the lower values of course)
- but I
think we all agree that this seems inappropriate and biased.
Interestingly,
this is exactly what happens in the real life, i.e. where many many
samples
(different ones instead of repeated measures) are analysed and only
those
that are above LLOQ (by chance or otherwise) are reported.
Going back to my thought experiment above about repeatedly measuring
an LLOQ
sample, say 100 or more times, and in this new thought experiment we
actually report either the value or "that this would then give a very good description of the missing
sample and
not contravene the *recommendation* of the FDA. Alternatively, just
report
both items (the value and "
Regards
Steve
--
Professor Stephen Duffull
Chair of Clinical Pharmacy
School of Pharmacy
University of Otago
PO Box 913 Dunedin
New Zealand
E: stephen.duffull.at.otago.ac.nz
Back to the Top
- On 3 Oct 2007 at 09:23:01, Nick Holford (n.holford.-a-.auckland.ac.nz) sent the message
The following message was posted to: PharmPK
Sylvain and Sven,
Thanks for responding to my challenge for someone to point out where
a regulatory authority explicitly forbids the use of measurements
below the LLOQ.
You both cite the same FDA Bioanalytical Method Validation guidance
(2001) "Estimation of concentration in unknown samples by
extrapolation of standard curves below LLOQ ... is not recommended.
Instead, the standard curve should be redefined..."
Please note that FDA wisely does not forbid the use of measurements
below the LLOQ but simply recommends using a standard curve with
lower concs to avoid extrapolation.
Sylvain discusses the signal to noise ratio of 5:1 to define the LLOQ
(or a coefficient of variation of 20% as it is defined by FDA). This
criterion for defining the LLOQ is not related to the use of the
measurements. In the FDA guidance it is there simply so that chemical
analysts can tick the box to show they know the properties of their
assay and can claim their assay is validated. Note that validation
says nothing about the purpose to which the measurements are put.
The naive interpretation that samples less than LLOQ should not be
reported but just flagged as BLQ reflects ignorance of how
measurements are used. As I pointed out in this thread previously
this interpretation necessarily leads to bias in the use of
concentrations close to LLOQ and can easily lead to selection of the
wrong model structure to predict the pharmacokinetics.
> There is a certainty (not a risk) that your analysis will be
biased if you selectively
> discard measurements below the LLOQ. This is a very simple
statistical fact. Imagine 3
> sample measurements 0.9, 1.0, 1.1. The average is 1.0. Suppose the
LLOQ is 1.0 and you
> discard the value of 0.9. The average of the remaining values is
1.05. It is biased (5%
> higher) compared with the best estimate of the mean obtained from
all the measurements.
>
> Pharmacokineticists that use compartmental models can find
themselves concluding there
> is an extra 'terminal phase' compartment because of the bias which
forces all
> measurements of true values close to the LLOQ to be >=LL0Q.
Because pharmacokineticists have available to them the statistical
tools to deal with measurements with any signal to noise ratio they
do not need to use the LLOQ value used for assay validation. These
are two different worlds and the criterion for assay validation is of
little direct importance to the pharmacokinetic analyst.
A somewhat different (but connected) issue is the limit of detection
(LOD). This is really where the problem lies for the chemical
analyst. The LLOQ and the LOD are defined by FDA (2001) as follows
Lower limit of quantification (LLOQ): The lowest amount of an analyte
in a sample that can be
quantitatively determined with suitable precision and accuracy.
Limit of detection (LOD): The lowest concentration of an analyte that
the bioanalytical procedure
can reliably differentiate from background noise.
The FDA (2001) Guidance gives quantitative guidance for LLOQ but I
have not found this for LOD (I could have missed it so let me know if
you find it).
What the pharmacokineticist needs in order to perform an unbiased PK
analysis is unbiased measurements. As I have pointed out above the
practice of selective reporting of concentrations near the LLOQ
provides biased information. I appreciate the chemical analyst
concern that at some point they are less than the LOD and do not feel
this measurement can have useful information. But surely the
interpretation of the information should be in the hands (and brain)
of the persons responsible for the PK analysis?
This is why I ask for the chemical analyst to report the truth the
whole truth and nothing but the truth. Midguided interpretation of
regulatory statements is a cause of real problems for PK analysts and
eventually for those who rely upon PK predictions for the safe and
effective use of medicines.
Sylvain asks:
> Finally, I don't know if you have been ever subjected to an FDA
> audit? from my personal experience during bioanalytical audit, don't
> report values which are below LOQ, simply flag them as BLQ (below the
> limit of quantitation) if you don't want to received a citation (i.e.
> a 483)
Fortunately I have not been invaded by the FDA police. As in all
dealings with police it is best not to argue when they have a gun
with them ('483'). But you should bring the arbitrary and capricious
behaviour of the police officer to the attention of the police
administration and to your lawmaker. Most regulatory authorities have
reasonable administrators if you know where to look (e.g. OCPB at
FDA). We need to make the case for clearer guidance for the way the
FDA police operate.
Reference
FDA. Bioanalytical Method Validation. http://wwwfdagov/cder/guidance/
4252fnl.htm 2001.
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
n.holford.at.auckland.ac.nz
www.health.auckland.ac.nz/pharmacology/staff/nholford
Back to the Top
- On 2 Oct 2007 at 16:28:34, "Ed O'Connor" (eoconnor.aaa.matrixbioanalytical.com) sent the message
The following message was posted to: PharmPK
The danger is that you might forget you did that; that the PK parameter
estimates you report rely on data you really did not have.
You will explain that you obtained the data from a very cooperative
analyst and both you and the bioanalyst will sport 483s at the very
least.
PK scientists can and should drive the bioanalytical process and push
the LLOQ to the lowest possible limit, incorporating larger initial
sample volume, sample enrichment procedures, etc.
---
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
Back to the Top
- On 2 Oct 2007 at 17:02:44, Roger Jelliffe (jelliffe.-at-.usc.edu) sent the message
The following message was posted to: PharmPK
Dear Sylvain:
Nick is right. He has the science. The correct measure of credibiity
is the Fisher information of a data point, NOT the CV%. In addition,
as a the measurement gets smaller, the assay SD and the variance
usually get smaller, and the Fisher info, which is 1/the variance
with which the measurement was made, is the correct and much more
useful measure of the error. Further, while the CV% gets larger and
does to infinity as the measurement goes to zero, the SD, and the
Fisher info, are always finite. There is always a valid measure of
credibility all the way down to ZERO. There are NO measurements to
censor! At least for PK work, there is NO LLOQ! This has been an
illusion perpetrated by those who look ONLY at the CV%.
Think about an assay with a CV of 10%. Think about a measurement of
10 units. The SD is 1, the variance is 1, and the Fisher info is 1.
Now double the measurement to 20. The SD is 2, the variance is 4, and
the Fisher info is 1/4. This is why the Fisher info is quite
different from the CV%. The measurement is really less precise,
although the CV% is the same.
Now take an assay with a constant SD of 1 unit - equally precise at
all measurements. Then the CV is 10% at a measurement of 10 units. At
20. it is 5%. At 5 it is 20%, at 2 it is 50%, at 1 it is 100%, and at
0.1 it is 1000%. What is the meaning of a signal to noise ratio here?
The actual precision of the assay, at all the concentrations, is
exactly the same!
That is for an ideal assay. Most assays have a positive SD intercept
at zero, reflecting machine noise. The SD may sometimes get smaller
as the concentrations rise for a bit, as with EMIT assays, for
example, but it gets larger sooner or later, usually with an upward
concavity. This can easily be fitted with a polynomial of 2nd order
in most cases, and the polynomial can be stored in software to fit
the data correctly.
It is not practical to do this every day, but every so often, with
new reagents, for example, whenever you want to check the assay
error, you can easily do it with such a polynomial. If there is no
drift, add the new points to the polynomial. If there is drift, use
the new polynomial.
The problem is that up to now, people never DID anythng with the data
except to LOOK at it and get an IMPRRESSION about the meaning ot the
result - low, OK, or high, for example. The visual impression of the
constant angle given by a graph of assay result versus constant CV%,
is powerfully misleading. Again, the SD is the thing, and the
variance, and the Fisher info. They are always finite, always a
reliable measure of credibility, and there is nothing to censor or
withold. As Steve Duffull says, give them both. It is SO easy, and is
good SCIENCE besides. It is probably the best thing to do for now.
People now are really ACTING QUANTITATIVELY on the data, and they
need, and can easily get, thte SD and the Fisher info, which they
need, and the lab needs to give them. Who are the lab people to
withold ANY data, anyway? If I were a Medicare administrator or an
accounts manager in an HMO, I would not pay for a TDM result that was
reported as ways. It is embarrassingly easy to do, and everyone is happy, as no
data are censored.
All the best,
Roger Jelliffe
Back to the Top
- On 3 Oct 2007 at 13:55:56, Nick Holford (n.holford.at.auckland.ac.nz) sent the message
The following message was posted to: PharmPK
Ed,
> The danger is that you might forget you did that; that the PK
parameter
> estimates you report rely on data you really did not have.
> You will explain that you obtained the data from a very cooperative
> analyst and both you and the bioanalyst will sport 483s at the very
> least.
Alzheimer's is indeed a worry for us all. Writing stuff down in
reports can help.
> PK scientists can and should drive the bioanalytical process and push
> the LLOQ to the lowest possible limit, incorporating larger initial
> sample volume, sample enrichment procedures, etc.
I think you mean 'CA scientists' ie. chemical analysts. The PK
scientists are the ones who want the truth the whole truth and
nothing but the truth from the CA scientists. The PK scientists don't
ever need the LLOQ.
Nick
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
n.holford.-a-.auckland.ac.nz
www.health.auckland.ac.nz/pharmacology/staff/nholford
Back to the Top
- On 2 Oct 2007 at 20:23:36, "Tata, Prasad N" (Prasad.Tata.-a-.Covidien.com) sent the message
The following message was posted to: PharmPK
Why not we use values below LLOQ? The answer is sure, except where you
are using them. If it is for your own academic/scientific mulch the
answer is YES, if you want to get some regulatory mileage the answer is
NO you play by their rules... NONMEM population may disagree with my
opinion.
Prasad Tata
Saint Louis, MO
Back to the Top
- On 3 Oct 2007 at 10:01:24, "Peris Walter" (Walter.Peris.aaa.rottapharm.com) sent the message
The following message was posted to: PharmPK
Dear all.
I'm new in the list and I beg your pardon for my terrible English.
I want to give my two cents about LLOQ question.
LOQ, IMHO, has different meanings, depending the different objectives
you have in your test.
You can give it a statistical or a graphical value or a value
depending from the test you are running (i.e. if in your test it's
not important to search values of drug below 5 ng/mL, it is without
any sense to run a method having a LLOQ at 0.1 ng/mL. In this case,
you are running a LC method having a used LLOQ above a physical LLOQ.).
Notwithstanding, from a chemical point of view, the "analytical"
meaning of LOQ is the LOWER value graphically quantifiable
(difference from LC noise, in a range of 10 minutes, around your
important chromatographic signal).
So, for a PK, you can obtain any value below LLOQ only, and only if,
the signal of the drug is REALLY different from noise (as above
stated, if you are using a LLOQ larger than a real LLOQ).
That means that if you have a LLOQ statistically calculated, i.e. on
CV%, and your peak is below this LLOQ, but significantly different
from noise (i.e. S/N > 10), you can use it remembering, in any case,
that you are using a LLOQ value (with any possible error coming from
its statistical analysis and any limit defined from guidelines).
But if your LC signal is not different from noise (graphically, LOQ
must has a LC signal different from noise between 5 and 10 fold) to
use a LLOQ value is without any physical and chemical sense.
In this case, you are not able to differentiate the peak from
chromatographic noise; this kind of data must be rejected.
Regards
Walter Peris
Back to the Top
- On 3 Oct 2007 at 09:53:14, "Ed O'Connor" (eoconnor.at.matrixbioanalytical.com) sent the message
The following message was posted to: PharmPK
That is why the PK scientists need to drive that aspect (LLOQ). The
analyst may be driven to find the simplest and least expensive LLOQ, one
which involves a minimum of sample processing (minimum cost in time and
materials).
If a given LLOQ is deemed not acceptable by the PK, the PK should drive
the process to include sample enrichment steps or perhaps entirely
different analytical approaches.
The approach advocated by Roger Jelliffe has a sound basis, but it needs
to be put in place and accepted by the analytical community and the
regulatory agencies.
--
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
Back to the Top
- On 3 Oct 2007 at 16:45:13, Roger Jelliffe (jelliffe.-a-.usc.edu) sent the message
The following message was posted to: PharmPK
Dear all:
I think the thing to do is to get rid of the idea of LLOQ entirely,
at least for PK work. Up to now, people have thought of the specimen
itself as the single source of information. Because of this, they
want to know if there is a drug (or some other substance PRESENT OR
NOT. Accordingly, people have liked to see a reasonable signal to
noise ratio, such as 2, 3, or more SD above the blank. That then
becomes "the LLOQ". Unfortunately, there is only a statistical
probability of something being there based on how many SD above the
blank the result is. All that is negotiation, not science. I like
some wine, you lilke another. Same thing. Not science.
For toxicology, and for many other applications, where the sample is
the ONLY surce of information, this may be the best one can do.
However, most physicians also know this, and if you give them that
information, they are probably just as smart al the lab guys at
putting things together. Is it the lab guys that don't wish to have
to report errors?????
However, for PK work, we know when the drug was given (pretty well)
and when the sample was drawn (pretty much). Because of this EXTRA
INFORMATION, usually requested on the lab slip (and for just this
reason!) and the fact that most drugs go away with half times, one
never gets rid of the last molecule, and so we know that the drug IS
PRESENT. We are not asking the same question as for toxicology - is
it there or not? Instead, since we know it is there, we are now
asking HOW MUCH IS PRESENT? When you cannot tell the peak from the
noise, then you know the result is close to zero. DO NOT REJECT IT!
WHO ARE YOU TO WITHOLD SUCH A RESULT? Physicians can also think. That
data should NEVER be rejected, as Walter Peris suggests. Walter, I
think you misunderstand the situation.
Get rid of thte LLOQ, at least for PK work. No one needs it! Instead,
give us what you already have obtained before you corrupted it - the
assay SD, not the CV%. Of course, push the assay to be as sensitive
as possible. But the IDEA of an LLOQ is an illusion that comes from
not understanding the new scientific uses to which the data is now
being put. Get with it, you guys! PLEASE!
Best regards to all,
Roger Jelliffe
As Nick says, we all want, and deserve, the truth, the entire truth,
and nothing but the truth.
Back to the Top
- On 4 Oct 2007 at 09:28:58, Azucena Aldaz (aaldaz.-a-.unav.es) sent the message
O.K. Dr.Jelliffe. You're right. The people who quantify drugs must be
trained at least in some basic pharmacokinetics concepts to
distinguish between toxicology and pharmacokinetics needs. The tools
are to be used to obtain the best results. Azucena Aldaz.
Back to the Top
- On 4 Oct 2007 at 08:26:29, "Michael Neely" (mneely.-at-.usc.edu) sent the message
The following message was posted to: PharmPK
"...the PK parameter estimates you report rely on data you really did
not have..."
If the data are reported with the corresponding (un)certainty, i.e. SD
at each point, such as Roger or Nick suggest, then you really do have
the data. It is when pharmacokineticists are forced to impute or ignore
points reported as problems.
Michael Neely, MD
Laboratory of Applied Pharmacokinetics
University of Southern California
Los Angeles, CA
Back to the Top
- On 5 Oct 2007 at 09:35:02, David Bourne (david.-a-.boomer.org) sent the message
[I think it is time to give this topic a rest. I have a few more
responses that I'll include in this message. If you want more
information about this topic you might consult the archives where
this topic has been discussed before. If you wish to follow-up with
an individual you might do that (of course) off-line (i.e. directly)
- db]
From markmiltonpharmaceutical.at.gmail.com
Having viewed this discussion with interest, I would like to make the
following observations and comments.
Azucena - If you are working in an academic insititute then I agree
with your statement "The poeple who quantify drugs must be trained at
least in some basic pharmacokinetic concepts to distinguish between
toxicology and pharmacokinetics needs." If you are working in a
pharmaceutical company, I would have to disagree with your comment
since most of the data generated will be generated by CROs who will
do what is asked of them by the sponsor. In fact, they can get in
big trouble if they do anything other than treat the data in a
blinded manner (see the FDA warning letters to MDS).
Roger - I agree with your point that TK needs and different from PK
needs. However, I think that your comments implying that the
bioanalytical data have been corrupted by the bioanalysts is a bit
strong.
Ed - I agree with your comment that if you need the data that are <
LLOQ, then it may be time to look for a new assay.
Prasad - I totally agree with your comment that if the data will be
used in a Regulatory submission you need to play by the Regulatory
Agency's rules (or their desires).
Nick - I agree that you need to clearly state in the report what data
you used and how you used it. I also agree that you can always
appeal to an FDA reviewer's supervisor and hope that he will support
your agrument. However, reality is that such an approach very rarely
works and if anything puts you in a worse situation than you were in
before (again, see the MDS issue).
Nick - How often is it that you need to use concentration data that
were deemed to be < LLOQ by the bioanalyst? Is it frequent or rare?
When you used such data, how often did you arrive at meaningfully
different (not mathematically different) answers? Just curious.
All - beware that you will need to reap what you sow. If you want
data that is < LLOQ from an assay, you must also accept such data
from control (placebo) dosed subjects and animals. The Regulatory
Agencies will not accept the concept that "those subjects/animals
were in the control group and therefore there can't be drug in the
samples." The EMEA has issued a guideline on this topic (
http://www.emea.europa.eu/pdfs/human/swp/109404en.pdf
). The worst outcome of finding drug in your control samples is that
you may need to repeat the Tox studies. The guideline does mention
LLOQ in one place (Section 4) where it states that "trace levels of
contamination that are below the lower limit of quantification may in
principle be considered as nonrelevant."
All - when you are dealing with very low concentrations (i.e. below
LLOQ) how do you distinguish between sample carryover or
contamination and the presence of drug in the sample that was collected?
Finally, I would like to put the following scenario to you all. You
are undergoing a random drug test at work. The LLOQ for the
quantitation of cocaine in plasma was 5 ng/mL. Your sample showed a
concentration that was < LLOQ. Your boss asked what the "real"
number was and was told that it was 2.5 ng/mL. You are then
summarily fired. Was this an appropriate use of the plasma
concentration data or was it a misuse or the data? No need to
answer, it is a rhetorical question.
Hopefully we can lay this topic to rest (i.e. agree to disagree) and
when it arises again, simply point the person that asked the question
to the previous discussions.
Regards
Mark Milton
--
From stefan.soback.-at-.gmail.com
Dear all,
The previous round of LLOQ discussion was quite interesting. I think
some realities should be acknowledged. There are drug concentrations
present in the organism above the LLOQ, between LLOQ and LLOD and
below LLOD. The concentrations below the LLOD are not incorporated in
the PK analysis, because the instrument did not give any signal (but
they are present nevertheless).
The bias Nick described at the LLOQ (upward trend) will repeat itself
at the LLOD if LLOQ is disregarded. Only this time it is not chemist
induced but instrument dependent. One also has to keep in mind that
the difference between LLOQ and LLOD is not big. Usually LLOD is
around 1/3-1/2 of the LLOQ. Before we discard the LLOQ and/or the
LLOD as unscientific and harmful, some things should be considered.
For an analytical chemist to provide concentrations below LLOQ is
about the same as for a physician to use homeopathic drugs instead of
licenced ones. Why should we throw out the basic concept of LLOQ only
because this discussion? The issue here appears to be basically about
data rescue after things went bad. Anyone in his right mind should
plan a study using an analytical method that can easily deal with the
anticipated and relevant concentrations. Clearly the method
performance requirements must be determined before the study. That in
my understanding would be science. All the rest would then be fixing
or something like that.
A determined concentration is a unit of a distribution resulting from
an often lengthy process of sample preparation and instrument
analysis. The distribution is poorly known below LLOQ because it was
not interesting in the first place. When data rescue becomes a
necessity I would suggest reanalysis of the sample in sufficient
number of times and to select the median value (preferably) or to
revalidate the method in that specific range.
Best regards,
Stefan
--
From Walter.Peris.-at-.rottapharm.com
Dear Roger,
thanks a lot for your answer.
> I think the thing to do is to get rid of the idea of LLOQ entirely,
> at least for PK work. Up to now, people have thought of the specimen
> itself as the single source of information. Because of this, they
want
> to know if there is a drug (or some other substance PRESENT OR NOT.
> Accordingly, people have liked to see a reasonable signal to noise
> ratio, such as 2, 3, or more SD above the blank. That then becomes
> "the LLOQ". Unfortunately, there is only a statistical probability of
> something being there based on how many SD above
> the blank the result is. All that is negotiation, not science. I like
> some wine, you lilke another. Same thing. Not science.
Please, Roger, I want to explain my point of view.
Mainly, my observation was only an "analytical observation"; on the
other hand, it's important to avoid to confuse two different situation:
analytical data and PK (o tox) data.
>From a PK point of view, you can use any approximation or decision you
can find useful. From an analytical point of view, you have to decide
what is "real" and what is "fiction". On a "real" data, you can drive
any possible hypotheses or conclusion. On a "fantastic" data, you don't.
Following this line, you have two different analytical situations: lower
limit of quantisation and limit of detection.
All analytical requirements define a peak as "present" only if it is
different from noise for at least 3 fold.
In particular, the ICH Guidelines Q2A (Definitions and Terminology),
states that:
Detection limit: the DL of an individual analytical procedure is the
lowest amount of analyte in a sample which can be detected but not
necessarily quantitated as an exact value.
Quantitation limit: the QL of an individual analytical procedure is the
lowest amount of analyte in a sample which can be quantitatively
determined with suitable precision and accuracy.
[cut]
The ICH Guideline Q2B defines for DL five different ways to obtain it:
based on visual evaluation, based on S/N (in particular, for this method
the guideline defines as acceptable a S/N ratio of 3 or 2/1), based on
the SD of response and slope, based on SD from blank and based on the
calibration curve.
For QL, the guideline defines analogue method, with a difference in S/N,
i.e., of 10/1 as typical value.
Under these values, isn't physically possible to say: "This is a peak,
having this area, and not noise".
This values depend from instrument, software, chromatographic parameters
setting, operator, lab SOP, test requirements and so on.
When you have fixed these hits, you can affirm, with no any possible
doubt, that a particular peak is present and quantifiable.
This is the theory.
In the Guidance for industry, Bioanalytical Method Validation (May 2001)
is present only S/N (5 times vs blank) definition for LQ (LD isn't
present) and this can produce a misunderstanding. To quantify a peak its
necessary that the peak exists!
On a set of data, you can perform any kind of statistical work, but you
MUST have the data.
Using data coming from a BLOQ response, you risk to obtain data without
any sense because below 5 times the blank its possible to have a not
true response.
It is FUNDAMENTAL to assure that the signal coming from the LC apparatus
was different from noise.
This is the point that I want to fix, if we are speaking of a LLOQ close
the limit of sensitivity of the method and/or the instrument.
> For toxicology, and for many other applications, where the sample is
> the ONLY surce of information, this may be the best one can do.
> However, most physicians also know this, and if you give them that
> information, they are probably just as smart al the lab guys at
putting
> things together. Is it the lab guys that don't wish to have to report
> errors?????
> However, for PK work, we know when the drug was given (pretty well)
> and when the sample was drawn (pretty much). Because of this EXTRA
> INFORMATION, usually requested on the lab slip (and for just this
> reason!) and the fact that most drugs go away with half times, one
never
> gets rid of the last molecule, and so we know that the drug IS
PRESENT. We > are not asking the same question as for toxicology - is it
there or not?
> Instead, since we know it is there, we are now asking HOW MUCH IS
PRESENT? > When you cannot tell the peak from the noise, then you know
the result
> is close to zero. DO NOT REJECT IT!
I think that here we have a little conceptual difference.
You think that these data are "close" to zero.
IMHO, these data are, probably, NOT DIFFERENT from zero.
> WHO ARE YOU TO WITHOLD SUCH A RESULT? Physicians can also think. That
data
> should NEVER be rejected, as Walter Peris suggests. Walter, I think
you
> misunderstand the situation.
Maybe yes.
I'm not a PK specialist; I'm just a poor chemist.
:-)
> Get rid of thte LLOQ, at least for PK work. No one needs it! Instead,
give > us what you already have obtained before you corrupted it - the
assay SD, > not the CV%.
> The SD assay, in LLOQ zone, cab be not very useful.
> Of course, push the assay to be as sensitive as possible. But the
IDEA
> of an LLOQ is an illusion that comes from not understanding the new
> scientific uses to which the data is now being put. Get with it, you
guys! > PLEASE!
Roger, I'm sorry but I'm not in accordance with this point of view.
LLOQ isn't an idea, it's a fact.
And the fact is that a peak BELOW the DL (remember that a S/N for QL of
5 is close to DL) is a ghost peak; using it, you can obtain ghost
results.
It's a your choice to use, or not, this data, and how, but it's
important to understand what kind of number you have in your hands.
Best regards to all,
Walter Peris
-
The following message was posted to: PharmPK
Hello Roger.
> There are NO measurements to censor!
OK, can you define what do you mean with "measurements"?
> At least for PK work, there is NO LLOQ! This has been an
> illusion perpetrated by those who look ONLY at the CV%.
Not at all.
> Think about an assay with a CV of 10%. Think about a measurement of
> 10 units. The SD is 1, the variance is 1, and the Fisher info is 1.
> Now double the measurement to 20. The SD is 2, the variance is 4, and
> the Fisher info is 1/4. This is why the Fisher info is quite
> different from the CV%. The measurement is really less precise,
> although the CV% is the same.
OK.
And what about ACCURACY?
> Now take an assay with a constant SD of 1 unit - equally precise at
> all measurements.
Any method can be precise but NOT accurate.
> Then the CV is 10% at a measurement of 10 units. At
> 20. it is 5%. At 5 it is 20%, at 2 it is 50%, at 1 it is 100%, and at
> 0.1 it is 1000%. What is the meaning of a signal to noise ratio here?
> The actual precision of the assay, at all the concentrations, is
> exactly the same!
The most important question is: what about accuracy?
> That is for an ideal assay. Most assays have a positive SD intercept
> at zero, reflecting machine noise.
An intercept in the curve is due, mainly, at an interference in the
matrix.
> The SD may sometimes get smaller
> as the concentrations rise for a bit, as with EMIT assays, for
> example, but it gets larger sooner or later, usually with an upward
> concavity. This can easily be fitted with a polynomial of 2nd order
> in most cases, and the polynomial can be stored in software to fit
> the data correctly.
> It is not practical to do this every day, but every so often, with
> new reagents, for example, whenever you want to check the assay
> error, you can easily do it with such a polynomial. If there is no
> drift, add the new points to the polynomial. If there is drift, use
> the new polynomial.
> The problem is that up to now, people never DID anythng with the data
> except to LOOK at it and get an IMPRRESSION about the meaning ot the
> result - low, OK, or high, for example. The visual impression of the
> constant angle given by a graph of assay result versus constant CV%,
> is powerfully misleading. Again, the SD is the thing, and the
> variance, and the Fisher info. They are always finite, always a
> reliable measure of credibility, and there is nothing to censor or
> withold.
>From a theoretic point of view, this is correct.
But what kind of measure you have in your hands?
Have you a number or have you a zero?
> As Steve Duffull says, give them both. It is SO easy, and is
> good SCIENCE besides. It is probably the best thing to do for now.
> People now are really ACTING QUANTITATIVELY on the data, and they
> need, and can easily get, thte SD and the Fisher info, which they
> need, and the lab needs to give them. Who are the lab people to
> withold ANY data, anyway?
An analyst has the responsibility to give out a DATA, not a zero.
> If I were a Medicare administrator or an
> accounts manager in an HMO, I would not pay for a TDM result that was
> reported as
In ANY analytical method there is a sensitivity limit; you don't go
below this limit only for decree.
Any technique has a limit and you MUST take it in account.
You have any reason to ask for a better limit and for a better
sensitivity, but you cannot force the physic or the actual technology.
> Steve Duffull is also right. Give it to them both
> ways. It is embarrassingly easy to do, and everyone is happy, as no
> data are censored.
Nobody wants to censure data.
Simply, no analyst likes to give as a value any measurement that he is
not able to differentiate from zero.
The LLOQ symbolism, really, means:
"I'm sorry, but my actual analytical conditions are not able to
distinguish this value from zero; if you need a lower value, PROBABLY
it's necessary to work to perform better conditions/to spent money to
buy a more sensible instrument/to wait that the technology will produce
a new instrument more sensible than that".
Best regards
Walter Peris
-
The following message was posted to: PharmPK
Hi Ed.
> The danger is that you might forget you did that; that the PK
parameter
> estimates you report rely on data you really did not have.
> You will explain that you obtained the data from a very cooperative
> analyst and both you and the bioanalyst will sport 483s at the very
> least.
> PK scientists can and should drive the bioanalytical process and push
> the LLOQ to the lowest possible limit, incorporating larger initial
> sample volume, sample enrichment procedures, etc.
This it's true.
However, we all have a limit: the technology.
30 years ago, you all were happy to have a drug concentration in plasma
measured at micrograms magnitude.
Today, we are happy to have concentration ranging around pg or fg.
But if you want see zepto or yoctograms, probably you have to wait few
years again.
And this, independently from what you needs, you wishes and/or my
ability.
Best regards
Walter Peris
-
The following message was posted to: PharmPK
Hi Ed.
> The danger is that you might forget you did that; that the PK
parameter
> estimates you report rely on data you really did not have.
> You will explain that you obtained the data from a very cooperative
> analyst and both you and the bioanalyst will sport 483s at the very
> least.
> PK scientists can and should drive the bioanalytical process and push
> the LLOQ to the lowest possible limit, incorporating larger initial
> sample volume, sample enrichment procedures, etc.
This it's true.
However, we all have a limit: the technology.
30 years ago, you all were happy to have a drug concentration in plasma
measured at micrograms magnitude.
Today, we are happy to have concentration ranging around pg or fg.
But if you want see zepto or yoctograms, probably you have to wait few
years again.
And this, independently from what you needs, you wishes and/or my
ability.
Best regards
Walter Peris
The following message was posted to: PharmPK
Hi Nick.
> The naive interpretation that samples less than LLOQ should not be
> reported but just flagged as BLQ reflects ignorance of how
> measurements are used.
Not at all.
BLQ simply means that, in these specific analytic conditions, was not
possible to obtain a numerical value.
> As I pointed out in this thread previously
> this interpretation necessarily leads to bias in the use of
> concentrations close to LLOQ and can easily lead to selection of the
> wrong model structure to predict the pharmacokinetics.
My question is:
Is more important to perform a PK analysis using a larger set of data,
at least using not real values, or perform a PK analysis with a smaller
set of data, but more accurate?
In general, the BLQ values are related to high times, influencing
heavily the AUC, and I understand that this is very important, but if we
have a objective limit in the technique, what do you prefer?
> A somewhat different (but connected) issue is the limit of detection
>(LOD). This is really where the problem lies for the chemical
analyst.
Precisely.
This is not only a key point.
This IS THE POINT.
> The LLOQ and the LOD are defined by FDA (2001) as follows
> Lower limit of quantification (LLOQ): The lowest amount of an analyte
> in a sample that can be
> quantitatively determined with suitable precision and accuracy.
> Limit of detection (LOD): The lowest concentration of an analyte that
> the bioanalytical procedure
> can reliably differentiate from background noise.
OK.
> The FDA (2001) Guidance gives quantitative guidance for LLOQ but I
> have not found this for LOD (I could have missed it so let me know if
> you find it).
In that guide, a LOD acceptable value is missed.
In another message, I give the values for the analytical guidance (Q2B),
non bioanalytical.
In general, for a S/N method, a peak is detectable if the difference vs
blank noise is 2/3:1 and the peak is quantifiable if the S/N is not
lower than 10:1. In this case, the accuracy is acceptable for chemical
analytical requirements.
In a bioanalytical method, the analytical requirements are less
restrictive and the LOQ values is reduced to 5:1; nothing is stated for
LOD.
In this prospective, it's objectively very difficult differentiate a
peak from noise if you want to go below LOQ.
> What the pharmacokineticist needs in order to perform an unbiased PK
> analysis is unbiased measurements.
That it's correct.
But for definition of LLOQ, any value below LLOQ has, at least, a big
bias.
Unfortunately, any value below LLOQ has a second negative aspect: it
cannot be different from zero.
Are you sure to want this kind of data?
> As I have pointed out above the
> practice of selective reporting of concentrations near the LLOQ
> provides biased information.
Not only. Probably, this value is zero.
> I appreciate the chemical analyst
> concern that at some point they are less than the LOD and do not feel
> this measurement can have useful information. But surely the
> interpretation of the information should be in the hands (and brain)
> of the persons responsible for the PK analysis?
This is the reason because a report shows LLOQ symbol.
In another message I wrote:
*******
The LLOQ symbolism, really, means:
"I'm sorry, but my actual analytical conditions are not able to
distinguish this value from zero; if you need a lower value, PROBABLY
it's necessary to work to perform better conditions/to spent money to
buy a more sensible instrument/to wait that the technology will produce
a new instrument more sensible than that".
*******
> This is why I ask for the chemical analyst to report the truth the
> whole truth and nothing but the truth.
The truth is:
This value is over my actual possibilities to obtain a real and accurate
value.
If I'm not in conditions to give you a value, I'll never give you a
value.
It's not my job make you happy; my job is give you real data.
Walter Peris
[An interesting challenge might be to develop a consensus report. But
I wonder if the analytical chemist and the PK analyst can agree. As
one who does both from time to time I don't see a real conflict but
I'm not reporting to a government agency for drug approval and my
experiments can be repeated, redesigned if necessary - db]
Back to the Top