RE: Differences in Covariance analysis between FOCE vs. IMPMAP
Andreas:
The IMP/IMPMAP method of standard error assessment for OMEGA band matrices may
be less reliable. I am presently investigating if there is a bug pertaining to
it, and if so, I shall have it fixed for the next version. Meanwhile, Monte
Carlo EM methods do quite well with full OMEGA matrices, so you may wish to
repeat the SAEM/IMP method with full OMEGA.
Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics
ICON Development Solutions
Tel: (215) 616-6428
Mob: (925) 286-0769
Email: [email protected]
Web: www.icondevsolutions.com
Quoted reply history
________________________________
From: [email protected] [mailto:[email protected]] On
Behalf Of Steingötter Andreas
Sent: Tuesday, September 07, 2010 9:56 AM
To: nmusers
Subject: [NMusers] Differences in Covariance analysis between FOCE vs. IMPMAP
Dear NONMEM user
The discussion about negative Eigenvalues during IMPMAP covariance analysis
made me add this method after my standard FOCE run:
$OMEGA BLOCK(2)
0.3 ;LOGETA_K40
0.1 0.3 ;LOGETA_K43
$OMEGA BLOCK(2)
0.3 ;LOGETA_K30
0.1 0.3 ;LOGETA_K34
$EST METHOD=0 PRINT=10 INTERACTION NSIG=3 SIGL=12
$EST METHOD=COND PRINT=10 INTERACTION NSIG=3 SIGL=12 MAXEVAL=3000
FOCE minimization and covariance analysis was SUCCESSFULL resulting in an OFV
of 6239. Standard errors, theta and eta correlations looked reasonable:
THETA:se% = 4.8 8.5 22.7 8.6 19.9
OMEGA:se% = 37.3 45.5 58.9 32.2 49.1
SIGMA:se% = 30.8 17.5
After adding the additional IMPMAP method my OFV becomes 6252 (not critical I
think), standard errors increase (for one even dramatically), all thetas are
correlated and eta correlations are mirrored in the band matrix:
$EST METHOD=0 PRINT=10 INTERACTION NSIG=3 SIGL=12
$EST METHOD=COND PRINT=10 INTERACTION NSIG=3 SIGL=12 MAXEVAL=3000
FILE=FOCE_2WAY.TXT
$EST METHOD=IMPMAP INTERACTION EONLY=1 ISAMPLE=1000 NITER=5 FILE=IMPMAP_2WAY.TXT
THETA:se% = 13.1 17.7 28.8 34.0 20.6
OMEGA:se% = 59.9 43.9 26.0 248.4 84.3
SIGMA:se% = 40.3 18.3
What is happening here? What is my mistake here? Was I just lucky with the
FOCE?
Thanks for your help and comments
Andreas
----------------------------------------------------------------------------------------------------------------------------------------
Andreas Steingötter, PhD
Division for Gastroenterology and Hepatology
Institut for Biomedical Engineering
Department of Internal Medicine
Divisions of Bioimaging and MRI Technology
University Hospital Zurich
University and ETH Zurich
Rämistrasse 100
Gloriastrasse 35
CH - 8091 Zurich
CH - 8092 Zurich
Tel. +41 44 255 5684
Tel. +41 44 255 5684
Fax +41 44 255 4591
Fax +41 44 632 1193
Email [email protected]
Email [email protected]