Back to the Top
A data analysis procedure is always limited by the assay. Any value below
the LOQ should be assigned the value "0." Using the value of 1/2 of the
concentration at the current LOQ requires a validation at the new
concentration. Perhaps re-validating a new standard curve over a limited
range at the lower portion of the total curve with twice the volume would
allow obtaining measurement of lower values. Thus, one way of proceeding is
to have 2 standard curves: 1 regular, and the other prepared with twice the
volume. Select the samples with lower conc to be run with twice the volume.
Armen P. Melikian
armen_melikian.-a-.berlex.com
Back to the Top
Ray,
One approach is to try setting the 'detected' values to the LOQ, zero,
something inbetween and missing. If the changes have a big effect on the
model,
you need better data.
One way of getting this is to pester the bioanalyst. Bioanalysts tend
to follow guidelines on setting LOQs which run something like this:
"Compute a confidence interval (really a prediction interval -
important
difference, but not relevant in this posting), this will be a pair of
hyperbolic
curves above and below the calibration line. Where the upper one
intercepts the
y-axis draw a horizontal line to the lower curve, then a vertical line down to
the x-axis (concentration) place the LOQ there, since lower y-values
(instrument
readings) imply a possibly negative concentration."
My view of this is as follows:-
The limits arise from a mathematical model, and any scientist knows
that
if you push a model to extremes it will produce physically unrealistic
predictions. To keep life simple, we tend to put up with this if it does not
happen too often and just say "the model is no good in this region".
If we actually need the model to produce results in this difficult
region, we should beat on the doors (or the heads) of the folk who provided the
model, and demand a better one. This may imply more sophistication, and
greater
effort, but can usually be done.
In this case, the model comes from statisticians (I am one), and we can
indeed do better. The usual guidelines are based on a calculation which
assumes constant and symmetrical measurement variation. Both assumptions are
unrealistic at low concentrations. The process of assigning an LOQ therefore
rests on an abuse of statistics.
For many assays, if correctly evaluated, the LOQ and limit of detection
should be very nearly the same. Usually the estimated precision of the assay
cannot be fully described by either standard deviation or coefficient of
variation, and if you are given symmetrical confidence limits for concentration
estimates near zero, they are probably wrong.
My own opinions, not the Company's.
Frank House.
Back to the Top
>
> In similar situations with clinical PK studies, I have ascribed zero to those
> values.
I am afraid there is no simple answer to the problem of how to deal with
quantities that are below the limit of quantitation. But I am pretty
sure that the assuming they are zero is likely to be your worst strategy
because you can be sure assuming any kind of reasonable model for
elimination that the conc will not be zero.
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, Private Bag 92019, Auckland, New Zealand
email:n.holford.-a-.auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556
http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.html
Back to the Top
I think it is impossible to comment without first knowing what you hope
to gain from your TK dataset.
Typically, TK studies are performed to estimate drug exposure in a
toxicology study. If that is your primary goal, look at your
non-compartmental results when you have used a value of zero and when you
have used the LOQ (or LOQ/2). If the different methods provide different
conclusions about the toxicology exposure relationship, your situation
needs more evaluation. It also may mean that you have used too high a
dose in your toxicology study, and it definitely means you need a better
assay.
If you are trying to obtain primary PK information from your dataset,
remember that this type of study is typically capable of reliably
providing only steady-state clearance (or CL/F). If all your data wasactual measurements you can refine your estimate a bit more. This is
ample information to design future PK studies, to gain a first guess to
the disposition of your drug, and to refine your assay specifications
(aka Wish List). Good-Luck
Regards,
Jeffrey Wald, Ph.D.
_____________________________________________________________
Quintiles, Incorporated
Post Office Box 13979
Research Triangle Park, North Carolina
27709-3979
Phone 919-941-7245
Fax 919-941-0493
Back to the Top
Please consider what the TK data is intending to do, I.e. to provide drug
exposure information in relevant animal species and to relate the
exposure to drug safety, ultimately in man. These studies are not
necessarily designed to accurately characterize the PK profile in animals.
Assignment of the LOQ concentration to samples with concentrations
between LOQ and Limit of Detection may actually overpredict the AUC
estimate. This is Ok, though, because a more conservative estimate is
obtained from the standpoint of how the animal exposures relate to
exposure in man.
Since values below LOQ must be handled in some way, and since limited
analytical resources and time may not justify revalidation at lower
concentrations, this approach allows all available data to be used to
make the conservative decision.
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)