Differences in Covariance analysis between FOCE vs. IMPMAP

4 messages 4 people Latest: Sep 13, 2010

Differences in Covariance analysis between FOCE vs. IMPMAP

From: Andreas Steingötter Date: September 07, 2010 technical
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]

RE: Differences in Covariance analysis between FOCE vs. IMPMAP

From: Robert Bauer Date: September 07, 2010 technical
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]

RE: Differences in Covariance analysis between FOCE vs. IMPMAP

From: Dieter Menne Date: September 13, 2010 technical
Robert, when trying your suggestion of using full omega instead of a band, I got the message below with the standard compiled nm7.1.2. Is there anything I can do? 0OMEGA HAS BLOCK FORM: 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0DEFAULT OMEGA BOUNDARY TEST OMITTED: NO 0SIGMA HAS SIMPLE DIAGONAL FORM WITH DIMENSION: 2 0DEFAULT SIGMA BOUNDARY TEST OMITTED: NO 0INITIAL ESTIMATE OF THETA: 0.1000E-01 0.6000E+01 0.8500E+02 0.1399E+01 0.5500E+00 0.2299E+01 0.1300E+00 0.4200E+02 0.9600E-01 0.3600E+02 0.4000E+01 0.6799E+01 0.6099E+01 0.1600E+02 0.1500E+01 0.4600E+05 0.6899E+00 0.6100E+02 0.1000E+02 -0.2000E-01 0INITIAL ESTIMATE OF OMEGA: BLOCK SET NO. BLOCK FIXED 1 NO DIMENSION OF BLOCK EXCEEDS 8
Quoted reply history
From: [email protected] [mailto:[email protected]] On Behalf Of Bauer, Robert Sent: Tuesday, September 07, 2010 10:13 PM To: Steingötter Andreas; nmusers Subject: RE: [NMusers] 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

RE: Differences in Covariance analysis between FOCE vs. IMPMAP

From: Thomas Ludden Date: September 13, 2010 technical
Bob: I have not found a test problem with an OMEGA BLOCK greater than 8. The largest I found was for 8 but I am still looking. However, this control stream will definitely not run but I do not think the problem is with the size of the block. The zero off-diagonal elements in the last row should terminate the run. Also, I believe the correlations for the rest of the matrix are all 1 and this could also be problematic. The user may not have looked through the rest of the file. I think the entire output (.res) file as well as the cs and data file if possible should be requested. Tom
Quoted reply history
________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Dieter Menne Sent: Monday, September 13, 2010 4:41 AM To: [email protected] Cc: Bauer, Robert Subject: RE: [NMusers] Differences in Covariance analysis between FOCE vs. IMPMAP Robert, when trying your suggestion of using full omega instead of a band, I got the message below with the standard compiled nm7.1.2. Is there anything I can do? 0OMEGA HAS BLOCK FORM: 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0DEFAULT OMEGA BOUNDARY TEST OMITTED: NO 0SIGMA HAS SIMPLE DIAGONAL FORM WITH DIMENSION: 2 0DEFAULT SIGMA BOUNDARY TEST OMITTED: NO 0INITIAL ESTIMATE OF THETA: 0.1000E-01 0.6000E+01 0.8500E+02 0.1399E+01 0.5500E+00 0.2299E+01 0.1300E+00 0.4200E+02 0.9600E-01 0.3600E+02 0.4000E+01 0.6799E+01 0.6099E+01 0.1600E+02 0.1500E+01 0.4600E+05 0.6899E+00 0.6100E+02 0.1000E+02 -0.2000E-01 0INITIAL ESTIMATE OF OMEGA: BLOCK SET NO. BLOCK FIXED 1 NO DIMENSION OF BLOCK EXCEEDS 8 From: [email protected] [mailto:[email protected]] On Behalf Of Bauer, Robert Sent: Tuesday, September 07, 2010 10:13 PM To: Steingötter Andreas; nmusers Subject: RE: [NMusers] 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