Back to the Top
Dear all
I would like to know that are we allowed to weight some (or all) points
of calibration curve when we are conducting a bioequivalence study. What
is the FDA guidance in this regard ? --
Dr.Abolfazl Mostafavi
Faculty of Pharmacy and Pharmaceutical Sciences Isfahan University of
Medical Sciences
Isfahan, I.R. Iran
Back to the Top
[Two replies - I hope this isn't a repeat - the network is a little
strange right now - db (strange == I don't know what is happening ;-) ]
From: "Edward O'Connor"Date: Tue, 29 Jan
2002 15:59:44 -0600
To: david.-a-.boomer.org
Subject: Re: PharmPK Statistical weight for calibration data
The following message was posted to: PharmPK
weighting should be used to get the best possble fit. Weighting will
impact the accuracy but will not alter %CV. On a negative side,
weighting will decrease the r. You have to "weigh" this impact against
the increase in accuracy. Weighting may also be done within the PK
program.
---
From: jose-antonio.allue.-a-.beaufour-ipsen.com Date: Wed, 30 Jan 2002
07:30:54 +0100
To: david.aaa.boomer.org
Subject: Re: PharmPK Statistical weight for calibration data
The following message was posted to: PharmPK
Dear Abolfazl:
I don«t know why BE studies should different from the rest. If you
demonstrate that response«s variance (response=Peak area, ratio,
cps....) is not constant along the concentration range of your
calibration curve, then you are not allowed, but oblied, to use the
correct weighting factor. Otherwise you won«t be able to estimate
correctly some of the points.
JosŽ Antonio AlluŽ
Mass Spectrometry Laboratory
Metabolism & Pharmacokinetics Service
Research & Development Department
IPSEN-PHARMA S.A. Laboratories
Ctra. Laureˆ Mir— 395
Sant Feliu de Llobregat, Barcelona, Spain
Back to the Top
The following message was posted to: PharmPK
The answer to the original question can be found in the FDA guidance:
Bioanalytical Method Validation Issued 5/2001, Posted 5/22/2001)
at http://www.fda.gov/cder/guidance/index.htm
The following is verbatim from the subsection on Calibration
Curve/Standard Curve/Concentration-Response from the guidance:
"Selection of weighting and use of a complex regression equation should
be justified."
Two additional comments to Ed O'Connor's message:
r is no longer requested as it can be misleading.
The best-fit model is inappropriate. Instead we specifically used the
following phrase:
"The simplest model that adequately describes the concentration-response
relationship should be used."
To clarify this I'll use an example.
Take a 3 point calibration curve where the true underlying relationship
between concentration and response is linear. If there is variability in
measurement, then when making a calibration curve the 3 points won't
fall on a straight line. However a second order polynomial will fit the
3 points without any error. Consequently, the second order polynomial is
the best fit to the data, because there's no error. However the 1st
order polynomial (linear model) is simpler, is a more appropriate model,
and should be used.
This becomes even more obvious with more standard concentrations e.g. 6
standards may have the best fit with a 5th order polynomial, but a
linear model may be adequate and appropriate. Consequently, the best fit
to the standard curve won't give a good measurement for an unknown
sample.
Ron Kavanagh
Back to the Top
The following message was posted to: PharmPK
I agree with Ron's points. One addition: One can use Akaike's or
Schwartz' Information Criteria for the goodness of fits. These criteria
use sum squares plus the number of parameters of the fit. So, one might
have a good fit with a complex expression, but will have too many
parameters in the fit. A simpler model will have less no. of parameters
and hance give beter fits with Akaike or Schwartz criteria.
Prof.Dr. lbeyi AÝabeyoÝlu
ilbeyi.-at-.tr.net
Dept.Pharmaceutical Technology
Faculty of Pharmacy
Gazi University
Ankara,Turkey.
Back to the Top
Dear Dr. O'Connor:
It is not clear to me what you mean when you say that "weighting should
be used to get the best possible fit". Is this an art? How does one go
about it?
On the other hand, careful weighting of data by reasonable measures of
its credibility appears to be most useful. Data should be weighted, we
would suggest, according to the precision with which it is measured.
Many would therefore suggest that it is useful to first start with the
assay one is using to get the measurements, and then to determine the
precision of the assay over its entire working range. Data measured
precisely are more credible than other data. The Fisher information of a
data point (the reciprocal of the variance with which each point is
measured) is used by many as a good criterion for initial weighting of
data. This means that we ideally need to know the standard deviation
(SD, and variance) of each data point we measure. Since the SD now
becomes a function of the concentration, one cannot say any more that
the data are independent of the state of the system.
In addition to the assay error, there are all the other errors that
surround each subject or patient, that reflect the precision, or lack of
it, in the therapeutic environment that surrounds each patient and
his/her care. These are the errors in preparation and administration of
each dose, the errors in recording when each dose was given. similar
errors in recording the times at which serum (or other) samples were
drawn, structural model misspecification, and any possible changes in
the patient's parameter values during the period of data analysis.
Currently, all these are lumped together under the heading of
intra-individual variability.
All these things are factors in the real environment that surrounds each
subject or patient. They are weighting factors to be carefully
determined by thoughtful analysis of the assay and the data. They are
not something to be simply adjusted in order to get the best fit. Our
aim is to understand Mother Nature, not to mess with her. More info can
be found on our web site, www.lapk.org, under teaching topics, and assay
errors, and population modeling, especially sections 12 and 14. The
general idea is to get the most informed, and correctly skeptical, model
of the process, for intelligent action to be taken based on it.
Very best regards,
Roger Jelliffe
Roger W. Jelliffe, M.D. Professor of Medicine, USC USC Laboratory of
Applied Pharmacokinetics 2250 Alcazar St, Los Angeles CA 90033, USA
email= jelliffe.-at-.hsc.usc.edu Our
web site= http://www.usc.edu/hsc/lab_apk
Back to the Top
The following message was posted to: PharmPK
Roger Jelliffewrote:
>>"...Data measured precisely are more credible than other data...">>
Well , Roger, with all the respect to you and to the general wisdom of
this statement, I have seen in my life so many precisely inaccurate data
(especially when I wear a regulator's hat) that I do not believe in
credibility without a validation any more.
I found that even imprecise results, but obtained with the calibrated
method and validated with external standards, are much more credible
than very precise data (sometimes already certified by QA/QC!) but
obtained by a flawed method (or rigged by a cheat).
Best regards.
Janusz Z. Byczkowski, Ph.D.,D.Sc.,D.A.B.T.
Consultant
212 N. Central Ave.
Fairborn, OH 45324
e-mail januszb.aaa.AOL.com
homepage: http://members.aol.com/JanuszB/index.html JZB Consulting web
site: http://members.aol.com/JanuszB/consult.htm --
Back to the Top
The following message was posted to: PharmPK
Roger,
I couldn't agree more.
What also has to be brought into the equation is how experience and
judgement can be considered weighed variables. Data that looks
reasonable on the surface, can often lead to erronious analysis
resulfts. We cannot get away from looking at modeled results and using
our experience to decide if they make sense or not.
Our students at times seem to have problems with this and it is one of
the prime things we try to impart. Does the answer make sense? Would you
expect a 95 y/o F with a creatinine of 1.5mg% to have a gentamicin t1/2
of 3 hours? Even if the analysis program says so?
This is where the weighing of experience and judgement comes in. After
all, the model is only as good as the data.
Bob
Back to the Top
As a relative newcomer (from Physiology) to this field of PK analysis,
this whole debate about weighting is a fascinating exercise.
Further to the remarks by Brennan & Co.- I am curious - what do you do
after suspecting an unlikely/improbable/unreasonable estimate? Do you
adjust the estimates? If so, how? What experience is required?
>From my background I would suspect the model not the data. This
requires a different type of experience, one that is documented.
Back to the Top
Dear Dr. O'Connor:
It is not clear to me what you mean when you say that "weighting
should be used to get the best possible fit". Is this an art? How does
one go about it?
On the other hand, careful weighting of data by reasonable
measures of its credibility appears to be most useful. Data should be
weighted, we would suggest, according to the precision with which it is
measured. Many would therefore suggest that it is useful to first start
with the assay one is using to get the measurements, and then to
determine the precision of the assay over its entire working range. Data
measured precisely are more credible than other data. The Fisher
information of a data point (the reciprocal of the variance with which
each point is measured) is used by many as a good criterion for initial
weighting of data. This means that we ideally need to know the standard
deviation (SD, and variance) of each data point we measure. Since the SD
now becomes a function of the concentration, one cannot say any more
that the data are independent of the state of the system.
In addition to the assay error, there are all the other errors
that surround each subject or patient, that reflect the precision, or
lack of it, in the therapeutic environment that surrounds each patient
and his/her care. These are the errors in preparation and administration
of each dose, the errors in recording when each dose was given. similar
errors in recording the times at which serum (or other) samples were
drawn, structural model misspecification, and any possible changes in
the patient's parameter values during the period of data analysis.
Currently, all these are lumped together under the heading of
intra-individual variability.
All these things are factors in the real environment that
surrounds each subject or patient. They are weighting factors to be
carefully determined by thoughtful analysis of the assay and the data.
They are not something to be simply adjusted in order to get the best
fit. Our aim is to understand Mother Nature, not to mess with her. More
info can be found on our web site, www.lapk.org, under teaching topics,
and assay errors, and population modeling, especially sections 12 and
14. The general idea is to get the most informed, and correctly
skeptical, model of the process, for intelligent action to be taken
based on it.
Very best regards,
Roger Jelliffe
Roger W. Jelliffe, M.D. Professor of Medicine, USC
USC Laboratory of Applied Pharmacokinetics
2250 Alcazar St, Los Angeles CA 90033, USA
email= jelliffe.-a-.hsc.usc.edu
Our web site= http://www.usc.edu/hsc/lab_apk
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)