Back to the Top
The following message was posted to: PharmPK
Dear Pharmacokinetist,
I was analyzing a PK data using Non Compartmental analysis, I compared
the oral with IV bolus. The values of oral and IV bolus were the same. I
have about 6 patients data unable to extrapolate terminal lambda_z to
infinity. I felt the sample was drawn 8 hours instead of 24 hours. The
drugs are camptothecin derivatives. The drugs were given orally. In
previous publications, samples were drawn for 24 hours. Another concern
was the processing of the sample due to stability of the lactone ring
and
the carboxylate?
However, I had problem with Compartmental modeling for comparison of the
output values. I had models #3, #4, #5 that fit the data with the same 6
patients unable to get their output values estimate. I used model #6,
and I had different 4 patients data unable to get their output values
estimate. The patients data could not fit any other Compartmental models
except as mentioned above.
Is it appropriate to compare the output values of Non Compartment with
Compartmental? My curiosity is that the output values of Non Compartment
and Compartment will not be the same due to their linearity/non
linearity
with estimate, differential equation used and setting of the each
module?
Can I used oral and IV bolus for comparison output values estimate of
Compartmental models and, if not why? What can I used for comparison of
a
Compartmental model output values estimate? Please, I will appreciate
your time and effort/references to answer my question. Thanks alot!
Ollie Anum, Pharm.D.
Back to the Top
The following message was posted to: PharmPK
ck34.at.juno.com wrote:
> I was analyzing a PK data using Non Compartmental analysis, I compared
> the oral with IV bolus. The values of oral and IV bolus were the same.
> I
> have about 6 patients data unable to extrapolate terminal lambda_z to
> infinity. I felt the sample was drawn 8 hours instead of 24 hours. The
> drugs are camptothecin derivatives. The drugs were given orally. In
> previous publications, samples were drawn for 24 hours. Another concern
> was the processing of the sample due to stability of the lactone ring
> and
> the carboxylate?
I do not know what you mean by "the values of the oral and IV bolus
were the same". Do you mean the dose was the same? Or the sampling
times were the same? Or the estimates of parameters (e.g. clearance)
were the same (unlikely)? Or what?
> However, I had problem with Compartmental modeling for comparison of
> the
> output values. I had models #3, #4, #5 that fit the data with the same
> 6
> patients unable to get their output values estimate. I used model #6,
> and I had different 4 patients data unable to get their output values
> estimate. The patients data could not fit any other Compartmental
> models
> except as mentioned above.
I think the models you refer to in WinNonLin are one compartment,
first-order elimination with first-order absorption with/without
lagtime. There are many reasons why WNL may fail to produce parameter
estimates. Some due to intrinsic problems with the data and some due to
user choices. One of the key advantages of compartmental modelling over
NCA/SHAM Walt Disney style PK approaches is that it encourages the
investigator to look at the data and to try to understand how the data
could have arisen and why different people may need different models.
The blind NCA approaches can easily lead to the complacent conclusion
that the data is clean and all subjects only differ randomly in their
estimates.
> Is it appropriate to compare the output values of Non Compartment with
> Compartmental? My curiosity is that the output values of Non
> Compartment
> and Compartment will not be the same due to their linearity/non
> linearity
> with estimate, differential equation used and setting of the each
> module?
> Can I used oral and IV bolus for comparison output values estimate of
> Compartmental models and, if not why? What can I used for comparison of
> a
> Compartmental model output values estimate?
You are not clear about what you mean by output values. WNL produces
tables and tables of output from a compartmental model analysis.
However, let's assume you are mainly interested in the essential PK
parameters of clearance (CL) and steady state volume of distribution
(Vss). To start with I would assume the PK system is linear and
therefore would expect these parameters to be independent of route of
administration or whether NCA or compartmental methods were used. Note
that this assumes you account for oral bioavailability when computing
CL and Vss after oral doses.
The models you mention are linear and do not use differential equations
so I don't know why you mention this. However, the fun part of PK is
the challenge of detecting deviations from the simple assumptions of
linearity that are implicit in NCA methods for estimating CL and Vss.
Inspection of the data can suggest there may be non-linearities which
you might be able to model e.g. capacity limited elimination, saturable
plasma protein binding, aborption rate dependent first pass extraction,
dose dependent rate of absorption, etc.
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.-a-.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Dear Ollie,
concerning your question about how to compare output estimates from
compartmental models, you usually rely on AIC and SC ,WSSR and residual
plots,
least values of SC, AIC, WSSR show better model(more fit) , look for
the plots
in which the difference between the observed and predicted is minimal,
also
check the residual plots for abscence of patterns like u or V shapes ,
Taken
all together, you would be able to decide on the best model, provided
you use
the same weighting.
hope this helps
Noha
Noha Nabil Salama
Ph.D. candidate
Pharmacokinetics & Biopharmaceutics Lab,
School of Pharmacy, university of Maryland.
100 Penn St., Baltimore, MD 21201
Tel:410-706-7388
Back to the Top
The following message was posted to: PharmPK
Dear Pharmacokinetists,
The output values estimates using the Compartmental modeling, I meant
are
CL, VD/VDss, T1/2, and rate constant (k). I was doing the comparison of
CL, VD, T1/2, K with Non Compartmental analysis versus Compartmental
models (1, 3, 4, 5, 6). The output values of Non Compartmental analysis
and any of the Compartmental models were not the same except the AUC. I
used the same data with same doses & concentrations using Non
Compartmental analysis comparison of oral and IV models; CL, VD/VDss,
T1/2, K were the same. For publication purpose and FDA recommendations,
what should I use to double check for accuracy of any Compartmental
models? Your time and effort/references will be highly appreciated.
Thanks a lot!
Dr. Holford and Dr. Salama, thanks for your feedback.
Ollie
Back to the Top
The following message was posted to: PharmPK
Ollie,
ck34.at.juno.com wrote:
> The output values estimates using the Compartmental modeling, I meant
> are
> CL, VD/VDss, T1/2, and rate constant (k). I was doing the comparison of
> CL, VD, T1/2, K with Non Compartmental analysis versus Compartmental
> models (1, 3, 4, 5, 6).
What you call output values are known more specifically as the
pharmacokinetic parameter estimates. For the one compartment
disposition model there are only two parameters e.g. CL, Vss. The other
parameters, e.g. T1/2, k, are secondary (WNL terminology) and can be
ignored when comparing estimates from different analysis methods.
> The output values of Non Compartmental analysis
> and any of the Compartmental models were not the same except the AUC.
You seem to be comparing estimates from NCA and compartmental models.
What does "not the same" mean? You cannot mean they are identical. Do
you mean that the means are not the same when you test the null
hypothesis e.g. with a t-test?
Note: Whatever you mean by "same" if the AUC is the same then the CL
must be the same!
> I
> used the same data with same doses & concentrations using Non
> Compartmental analysis comparison of oral and IV models; CL, VD/VDss,
> T1/2, K were the same.
Here you are comparing oral and IV models using NCA. I would only
expect the raw (as output by WNL) estimates of CL and Vss to be the
"same" if bioavailability was 1.
> For publication purpose and FDA recommendations,
> what should I use to double check for accuracy of any Compartmental
> models?
Accuracy of parameter estimates cannot be checked (or double checked).
You don't know the true parameter values so you cannot determine
accuracy. You can only estimate the imprecision of the estimates.
Imprecision is reflected in the standard error of the estimate but this
is only a rough guide to adequate precision. Other aspects of model
performance are best evaluated by making sure that the predicted time
course of concs adequately follows the observed concentrations. In this
context "adequately" must be defined by you with the intended purpose
of the model in mind. The purpose should preferably be something
scientific not simply to satisfy FDA or journal editors.
Nick
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.at.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Hi Ollie,
> However, I had problem with Compartmental modeling for comparison of
> the
> output values. I had models #3, #4, #5 that fit the data with the same
> 6
> patients unable to get their output values estimate. I used model #6,
> and I had different 4 patients data unable to get their output values
> estimate. The patients data could not fit any other Compartmental
> models
> except as mentioned above.
>
Nick wrote:
There are many reasons why WNL may fail to produce parameter
estimates. Some due to intrinsic problems with the data and some due to
user choices.
I can explain the problems due to user chices:
Compartment models in WNL are VERY sensitive for the initial guess for
the
curve fitting procedure, your initial values for the parameters must be
close enough
to the real ones in order for Gauss-Newtons method (the default
curve-fitting
procedure) to converge (ie give you a result). WNL has two more
options for the
curve-fitting: 2) Gauss-Newton with Levenberg-Marquarts improvement
(this is not
working so much better than Gauss-Newton in WNL, I wonder why) and 3)
Nelder-Mead will
work no matter what, but it will take an eternity to complete the curve
fitting.
Otilia Lillin
M.Sc. Scientific Computing
Södersjukhuset Stockholm, Anestesikliniken
email otilia.aaa.nada.kth.se
Back to the Top
The following message was posted to: PharmPK
Dear Pharmacokineticists,
Thanks for your feedback regarding the comparison of WNL NCA and
Compartmental Models.
Dr. Holford, when I compared V/F, Cl/F, T1/2, K (PK parameters) of NCA
with any of the compartmental models, the PK parameters output values
were not identical. One compartmental model I used were model 3,4,5 & 6.
Model 5 had the best fitting curve with less standard error. Could you
please give the reasons why one, two, three compartments of the PK
models
should or should not be used to compare NCA. Also, I have used WNL to
analyze other PK data including etoposide IV bolus which fit two
compartments of the PK models. I compared the PK parameters output
values
of NCA with the two compartments PK model that fit the data, the PK
parameters were not identical. Regarding your last e-mail question,
please I was not talking about "statistic mean" comparison of NCA and
compartmental models and sorry for the misunderstanding. I meant PK
parameters output.
Dr. Suresh, I have tried to send you an e-mail on several occasions but
the e-mail was forwarded back to me. I am unable to send you the excel
raw data. Thanks for your request.
Thanks to everyone for your contributions to answer my questions. I am
looking forward to get the reasons why we should or should not compare
NCA with any of the compartmental models.
Ollie
Back to the Top
The following message was posted to: PharmPK
Ollie,
ck34.at.juno.com wrote:
>
> Dr. Holford, when I compared V/F, Cl/F, T1/2, K (PK parameters) of NCA
> with any of the compartmental models, the PK parameters output values
> were not identical. One compartmental model I used were model 3,4,5 &
> 6.
> Model 5 had the best fitting curve with less standard error. Could you
> please give the reasons why one, two, three compartments of the PK
> models
> should or should not be used to compare NCA.
There is little point to comparing compartmental *models* with NCA
(this is comparing apples with oranges except for the modelling
assumptions made in predicting CL and Vss from the AUC and AUMC
statistics). But you can compare some of the *parameter estimates*
obtained from the 2 methods e.g. CL and Vss.
> Also, I have used WNL to
> analyze other PK data including etoposide IV bolus which fit two
> compartments of the PK models. I compared the PK parameters output
> values
> of NCA with the two compartments PK model that fit the data, the PK
> parameters were not identical. Regarding your last e-mail question,
> please I was not talking about "statistic mean" comparison of NCA and
> compartmental models and sorry for the misunderstanding. I meant PK
> parameters output.
It would be much clearer to me if you gave specific numerical examples
of the parameter estimates obtained by the two methods. I could then
maybe understand what you mean by "not identical". Nobody should expect
the estimates from real data to be identical!
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.-at-.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Dear Pharmacokineticists,
Dr. Holford, thank again for your reply. Examples of 3 subjects PK
parameters Cl/F (15.05, 10.98, 29.99 ), V/F (44.67, 37.9, 75.53), K
(0.33, 0.28, 0.39), T1/2 (2.06, 2.39, 1.75) respectively for the
Compartmental model. The same 3 subjects PK parameter CL/F (7.51, 3.99,
18.86), V/F (117.13, 105.69, 117.87), K (0.06, 0.04, 0.11), T1/2 (10.82,
18.35, 6.54,) respectively for NCA. This is for Drug #1
For drug #2, 3 subjects PK parameters CL/F (3.86,15.76, 5.23, V/F
(10.8,
59.6, 19.03), K ( 0.35, 0.26, 0.27), T1/2 (1.94, 2.6, 2.52) respectively
for Compartmental model. The same subjects PK parameters Cl/F (2.38,
15.67, 3.8) V/F (26.8, 109.34, 41.47), K (0.0887, 0.1433, 0.0917), T1/2
(7.82, 4.84, 7.56) respectively for NCA. I hope you will not have
problems reading the PK parameters as listed above.
Both drugs #1 and #2 are in the same class of drugs. I just would like
to
have every one's opinion. Thanks again!
Ollie.
Back to the Top
The following message was posted to: PharmPK
Ollie,
> Dr. Holford, thank again for your reply. Examples of 3 subjects PK
> parameters Cl/F (15.05, 10.98, 29.99 ), V/F (44.67, 37.9, 75.53), K
> (0.33, 0.28, 0.39), T1/2 (2.06, 2.39, 1.75) respectively for the
> Compartmental model. The same 3 subjects PK parameter CL/F (7.51, 3.99,
> 18.86), V/F (117.13, 105.69, 117.87), K (0.06, 0.04, 0.11), T1/2
> (10.82,
> 18.35, 6.54,) respectively for NCA. This is for Drug #1
>
> For drug #2, 3 subjects PK parameters CL/F (3.86,15.76, 5.23, V/F
> (10.8,
> 59.6, 19.03), K ( 0.35, 0.26, 0.27), T1/2 (1.94, 2.6, 2.52)
> respectively
> for Compartmental model. The same subjects PK parameters Cl/F (2.38,
> 15.67, 3.8) V/F (26.8, 109.34, 41.47), K (0.0887, 0.1433, 0.0917), T1/2
> (7.82, 4.84, 7.56) respectively for NCA. I hope you will not have
> problems reading the PK parameters as listed above.
>
I have reformatted some of your numbers so they are easier for me to
read (tables are better than lists).
CL/F T1/2
Drug ID CMT NCA CMT NCA
1 1 15.05 7.51 2.06 10.82
1 2 10.98 3.99 2.39 18.35
1 3 29.99 18.86 1.75 6.54
2 1 3.86 2.38 1.94 7.82
2 2 15.76 15.67 2.6 4.84
2 3 5.23 3.8 2.52 7.56
There is systematic major underestimation of the NCA half-lives when
using CMT (compartment modelling).
Looking at your CL/F values which should be readily estimated by both
methods it seems that there is reasonable agreement for drug 2 but
rather poor for drug 1 (NCA estimates are about 50% of CMT). If you
look at the predicted time course of concs using CMT I expect you will
see that the predictions are seriously underestimating the actual
observations (because the AUC for the CMT predictions must be about
half the AUC obtained using NCA). If this is not the case then your NCA
AUC estimates are wrong -- perhaps due to over-estimation of the
terminal half-life.
I would not trust any of your estimates until you can get the
predictions to match the observations and remove the systematic large
differences between CMT and NCA half-lives and CL/F. IMHO you are
wasting your time looking at the CMT parameter estimates until you get
a reasonable graphical fit to your data.
Nick
--
Nick Holford, Dept Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New
Zealand
email:n.holford.-a-.auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556
http://www.health.auckland.ac.nz/pharmacology/staff/nholford/
Back to the Top
The following message was posted to: PharmPK
Dr. Holford,
Thanks for your reply. I agreed with you that you can not compare apples
and oranges.
Regarding your last e-mail, you stated " systematic major underestimate
of the NCA half-life when using CMT". The observed AUC for both NCA and
CMT were identical (e.g 50mg/L/hr for both NCA and CMT). We should
realized that WNL NCA method uses AUCINF to calculate PK parameters
while
WNL CMT methods does not. Also, when I reviewed the NCA data some
subjects have high AUCINF with high half-life, less CL. All subjects
with
out high AUCINF, their PK parameters were reasonable. Clinically, I
assumed that the reasons for high AUCINF on some subjects were due to
comorbid factors and the drug was unable to eliminate rapidly compared
to
other subjects. When I used AUC to calculate the PK parameters for both
NCA and CMT, the PK parameters were identical. For drug #1, AUCINF were
over two times, 5 times, 10 time etc., higher than AUC.
Some published articles calculated CL, half-life of drug #1 with out
using computer. I matched the predictions and observations concs with
the
published articles, there were similar graphs. The predictions time
course of concs using CMT were similar to actual observations. It is the
reason that the observed AUC for both NCA and CMT were identical. Could
you please explain what you meant by "IMHO"? Thanks!
Ollie
PharmPK Discussion List Archive Index page
Copyright 1995-2010 David W. A. Bourne (david@boomer.org)