i**f
发帖数: 1195 | 1 the noncompartmental method using trapezoidal rule can give the estimates of
CL,V,AUC etc.. and the standard deviation of the estimates. However I am
confused whether it can estimate the IIV or IOV?
Any input is greatly appreciated! |
a*******o
发帖数: 280 | 2 noncompartmental PK approach is still a semi-model based approach with
parameterizations such as
CL=F*Dose/AUC
of course you can estimate the IIV or IOV for parameters such as CL.
Unbelievable, I still know this stuff!!!
of
【在 i**f 的大作中提到】
: the noncompartmental method using trapezoidal rule can give the estimates of
: CL,V,AUC etc.. and the standard deviation of the estimates. However I am
: confused whether it can estimate the IIV or IOV?
: Any input is greatly appreciated!
|
i**f
发帖数: 1195 | 3
【在 a*******o 的大作中提到】
: noncompartmental PK approach is still a semi-model based approach with
: parameterizations such as
: CL=F*Dose/AUC
: of course you can estimate the IIV or IOV for parameters such as CL.
: Unbelievable, I still know this stuff!!!
:
: of
|
i**f
发帖数: 1195 | 4 Thank you for the input. could you enlighten more?
calculate CL=F*Dose/AUC for each subject
average=》mean of CL (CL_bar), CL_bar is the estimate of the population CL
calculate: sum(CL_bar-CL)^2/n-1=>standard deviation of CL,did you mean this
is the IIV for CL?
Thanks!
【在 a*******o 的大作中提到】
: noncompartmental PK approach is still a semi-model based approach with
: parameterizations such as
: CL=F*Dose/AUC
: of course you can estimate the IIV or IOV for parameters such as CL.
: Unbelievable, I still know this stuff!!!
:
: of
|
c*******g
发帖数: 695 | 5 I think probably you need the geological mean of CL
the Population PK ppl usually use geological mean, because
the CL distribution may be screwed.
Try to look at the distribution of your CLs then make the decision
this
【在 i**f 的大作中提到】
: Thank you for the input. could you enlighten more?
: calculate CL=F*Dose/AUC for each subject
: average=》mean of CL (CL_bar), CL_bar is the estimate of the population CL
: calculate: sum(CL_bar-CL)^2/n-1=>standard deviation of CL,did you mean this
: is the IIV for CL?
: Thanks!
|
y**g
发帖数: 197 | 6 also can try bootstrapping to find variance.
【在 c*******g 的大作中提到】
: I think probably you need the geological mean of CL
: the Population PK ppl usually use geological mean, because
: the CL distribution may be screwed.
: Try to look at the distribution of your CLs then make the decision
:
: this
|
a*******o
发帖数: 280 | 7 that approach is called naive average
before nonlinear mixed effects modeling was popularized by Sheiner
that was how people characterize the interindividual variability IIV.
however, right now, people more use IIV refering to some ETA terms in NONMEM
.
If you have suffient data, run a NONMEM, good for your resume and job
hunting.
LOL
what a break from the life of two by two matrix, porter's 5, and NPV.
this
【在 i**f 的大作中提到】
: Thank you for the input. could you enlighten more?
: calculate CL=F*Dose/AUC for each subject
: average=》mean of CL (CL_bar), CL_bar is the estimate of the population CL
: calculate: sum(CL_bar-CL)^2/n-1=>standard deviation of CL,did you mean this
: is the IIV for CL?
: Thanks!
|
c*******g
发帖数: 695 | 8
~~~~~~~~~~~~~Still don't understand this term
any good reference for interpretation?
Thanks
【在 y**g 的大作中提到】
: also can try bootstrapping to find variance.
|
a*******o
发帖数: 280 | 9 check out Nick Holford's work. I beleive he loved it.
As matter of fact, the orginal developer of NONMEM, Sheiner and Beal, did
not like the idea of applying Boostrapping with nonlinear mixed effects data
at all.
In the case of pharmacokinetis, most of times, boostrapping was used because
insufficient sample size, and you can not get a good statistical inference
from it.
【在 c*******g 的大作中提到】
:
: ~~~~~~~~~~~~~Still don't understand this term
: any good reference for interpretation?
: Thanks
|
i**f
发帖数: 1195 | 10 thank you yoda! you made it clearer.
from statistical point of view, IIV is the variance of a random variable(
random effect) in the nonlinear mixed effects model frame.The ETA you
mentioned is an estimate of the variance. There is noway to estimate this
variance using the naive average.
One can still characterize the IIV by calculating the standard deviation of
CL but it is not the same thing as the variance talked above.
I am pretty good at NONMEM：）and I am writing my own algorithm to implemen
【在 a*******o 的大作中提到】
: that approach is called naive average
: before nonlinear mixed effects modeling was popularized by Sheiner
: that was how people characterize the interindividual variability IIV.
: however, right now, people more use IIV refering to some ETA terms in NONMEM
: .
: If you have suffient data, run a NONMEM, good for your resume and job
: hunting.
: LOL
: what a break from the life of two by two matrix, porter's 5, and NPV.
:
|
|
|
i**f
发帖数: 1195 | 11 I met this guy once..very interesting....and very insulting to people with
math/stat background...
data
because
inference
【在 a*******o 的大作中提到】
: check out Nick Holford's work. I beleive he loved it.
: As matter of fact, the orginal developer of NONMEM, Sheiner and Beal, did
: not like the idea of applying Boostrapping with nonlinear mixed effects data
: at all.
: In the case of pharmacokinetis, most of times, boostrapping was used because
: insufficient sample size, and you can not get a good statistical inference
: from it.
|
i**f
发帖数: 1195 | 12 i think you meant geometric mean. you are right! Cls are always positive and
the distribution is often skewed (like a lognormal). geomean is a better
choice than mean.
【在 c*******g 的大作中提到】
: I think probably you need the geological mean of CL
: the Population PK ppl usually use geological mean, because
: the CL distribution may be screwed.
: Try to look at the distribution of your CLs then make the decision
:
: this
|
c*******g
发帖数: 695 | 13
and
~~~~~~~~~~~Yes it is what I mean
【在 i**f 的大作中提到】
: i think you meant geometric mean. you are right! Cls are always positive and
: the distribution is often skewed (like a lognormal). geomean is a better
: choice than mean.
|
a*******o
发帖数: 280 | 14 interesting, theoretically, non-liner mixed effects modeling is kind of in
the middle of frequentist and bayesian frames. Based on my experiences from
my previous career, FO in NONMEM is reasonable robust, and empirical bayes
can be easily obtained through POSTHOC option.
Anyway,I am sure you will do a great job. :-)
of
implement
【在 i**f 的大作中提到】
: thank you yoda! you made it clearer.
: from statistical point of view, IIV is the variance of a random variable(
: random effect) in the nonlinear mixed effects model frame.The ETA you
: mentioned is an estimate of the variance. There is noway to estimate this
: variance using the naive average.
: One can still characterize the IIV by calculating the standard deviation of
: CL but it is not the same thing as the variance talked above.
: I am pretty good at NONMEM：）and I am writing my own algorithm to implemen
|
y**g
发帖数: 197 | 15
of
==============================
what kind of your algorithm for Bayesian
analysis? is it different with the one used in WinBug(MCMC+GIBBS SAMPLING)?
implement
【在 i**f 的大作中提到】
: thank you yoda! you made it clearer.
: from statistical point of view, IIV is the variance of a random variable(
: random effect) in the nonlinear mixed effects model frame.The ETA you
: mentioned is an estimate of the variance. There is noway to estimate this
: variance using the naive average.
: One can still characterize the IIV by calculating the standard deviation of
: CL but it is not the same thing as the variance talked above.
: I am pretty good at NONMEM：）and I am writing my own algorithm to implemen
|
i**f
发帖数: 1195 | 16 hehe, the empirical bayesian individual estimates are not real bayesian
estimates...
from
【在 a*******o 的大作中提到】
: interesting, theoretically, non-liner mixed effects modeling is kind of in
: the middle of frequentist and bayesian frames. Based on my experiences from
: my previous career, FO in NONMEM is reasonable robust, and empirical bayes
: can be easily obtained through POSTHOC option.
: Anyway,I am sure you will do a great job. :-)
:
: of
: implement
|
i**f
发帖数: 1195 | 17 The theoretical background is the same, combined standard Gibbs and a more
general MH algorithm. However
by writing you own algorithm, it is much more flexible than using the build-
in packages from those
statistical software, ie winbug pkbug,etc. :)
Bayesian
)?
【在 y**g 的大作中提到】
:
: of
: ==============================
: what kind of your algorithm for Bayesian
: analysis? is it different with the one used in WinBug(MCMC+GIBBS SAMPLING)?
:
: implement
|
