RE: ETABAR p-value
Dear Leonid,
In your case there is no reason to expect the mean of the etabar to be zero,
so a test of it does actually not make sense. You have pronounced shrinkage
in your posthoc etas and then you don't know what the distribution should be
(apart from expected median of zero). We have noted mean posthoc eta
significantly different from zero even when the model is correct (see
reference below for some more discussion on the "uselessness" of posthoc
etas).
http://www.aapspharmaceutica.com/search/abstract_view.asp?id=941&ct=06Abstra
cts
Best regards,
Mats
Mats Karlsson, PhD
Professor of Pharmacometrics
Div. of Pharmacokinetics and Drug Therapy
Dept. of Pharmaceutical Biosciences
Faculty of Pharmacy
Uppsala University
Box 591
SE-751 24 Uppsala
Sweden
phone +46 18 471 4105
fax +46 18 471 4003
[EMAIL PROTECTED]
Quoted reply history
-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
Behalf Of Leonid Gibiansky
Sent: Tuesday, February 20, 2007 20:49
To: Mats Karlsson
Cc: [email protected]
Subject: Re: [NMusers] ETABAR p-value
Mats,
Thanks for your reply. I think I found where the problem was: to estimate
the probability of
observed ETAbar, I used the distribution with the variance estimated by the
nonmem (0.97) while
t-test uses variance estimated from the data (0.11784 in this case). In this
particular case,
variance estimated from the data is much lower than OMEGA(3,3) because the
POSTHOC ETA3 distribution
is rather non-normal, resulting in a small p-value.
Thanks
Leonid
Mats Karlsson wrote:
> Dear Leonid,
>
> The etabar test is a t-test of the mean of the posthoc etas. I would not
> discard a model just because of this not being the case as there may be
> other reasons than misspecification for a no-zero mean of posthoc etas.
Only
> when data are very rich and there is no shrinkage (or when eta shrinkage
is
> identically large for both positive and negative etas) would we expect the
> mean of posthoc etas to be zero.
>
> Best regards,
> Mats
>
>
> Mats Karlsson, PhD
> Professor of Pharmacometrics
> Div. of Pharmacokinetics and Drug Therapy
> Dept. of Pharmaceutical Biosciences
> Faculty of Pharmacy
> Uppsala University
> Box 591
> SE-751 24 Uppsala
> Sweden
> phone +46 18 471 4105
> fax +46 18 471 4003
> [EMAIL PROTECTED]
>
>
>
>
>
>
>
>
>
> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]
On
> Behalf Of Leonid Gibiansky
> Sent: Tuesday, February 20, 2007 18:25
> To: [email protected]
> Subject: [NMusers] ETABAR p-value
>
> Dear All,
> Could anyone help me to interpret ETAbar p value? I have:
>
> TOT. NO. OF OBS RECS: 1486
> TOT. NO. OF INDIVIDUALS: 213
>
> ETABAR: -0.52E-02 -0.27E-01 -0.93E-01
> P VAL.: 0.90E+00 0.24E+00 0.81E-04
>
> ETA1 ETA2 ETA3
> ETA1 4.08E-01
> ETA2 2.26E-01 2.31E-01
> ETA3 0.00E+00 0.00E+00 9.70E-01
>
> which looks too low for me for eta3. I checked that p of abs(mean(eta)) >
> 0.093 is about 0.17 for
> normally distributed variable with SD=sqrt(0.97) and about 200 patients.
>
>> sum1 <- 0
>> for(i in 1:1000000) if(abs(mean(rnorm(213,0,sqrt(0.97))))> 0.093) sum1
<-
> sum1+1
>> print(sum1/1000000)
> [1] 0.168624
>
>
> How exactly this p-value is computed (NONMEM V) ?
>
> Thanks
> Leonid
>
>
>
>
>
>