[NMusers] RE: IMP method not stationary

From: Panetta, Carl <Carl.Panetta_at_stjude.org>
Date: Tue, 24 Nov 2015 21:22:17 +0000

Thanks for your reply.

I am using version 7.3.

The run described in the original e-mail used the MAPITER=0 option.

When I changed the option to MAPITER=20 (and did a total of 20 iterations) the OBJ function was very stable. So this worked well.

What is the reason why doing the MAP estimation on every step stabilizes the problem?

J. Carl Panetta, Ph.D.
Department of Pharmaceutical Sciences
St. Jude Children's Research Hospital
262 Danny Thomas Place
Memphis, TN 38105
Office: (901) 595-3172
Mobile: (901) 921-3740
Fax: (901) 595-3125

From: Bauer, Robert [mailto:Robert.Bauer_at_iconplc.com]
Sent: Tuesday, November 24, 2015 2:25 PM
To: Panetta, Carl; nmusers_at_globomaxnm.com
Subject: RE: IMP method not stationary

You may want to try a couple of things.

First, NONMEM 7.3 should be used, as improvements on IMP are always made.

Next, after the SAEM step, try


Where MAPITER=0 has NONMEM not perform a MAP step even on the first iteration, but borrows conditional means and variances as proposal density information from the previous SAEM step.

If this still does not work, then go in the opposite direction and have MAP done on every iteration:

Robert J. Bauer, Ph.D.
Vice President, Pharmacometrics R&D
ICON Early Phase
Office: (215) 616-6428
Mobile: (925) 286-0769

From: owner-nmusers_at_globomaxnm.com<mailto:owner-nmusers_at_globomaxnm.com> [mailto:owner-nmusers_at_globomaxnm.com] On Behalf Of Panetta, Carl
Sent: Tuesday, November 24, 2015 9:35 AM
To: nmusers_at_globomaxnm.com<mailto:nmusers_at_globomaxnm.com>
Subject: [NMusers] IMP method not stationary

I am having some issues with the IMP method using EONLY=1 option to generate an estimate of the objective function.

Specifically I am first running the SAEM method and the model appears to be converging well (i.e. SAEMOBJ is stable). But when I then try to get the estimate of the objective function using the IMP method the OBJ value is not stable and actually is steadily increasing. (See the NONMEM output below.) The data and model for this problem are a 3 compartment linear PK model with zero-order absorption (using ADVAN7) where each compartment represents either the parent drug or one of the two measured metabolites. Overall the fits look good and the parameter estimates are reasonable.

Note that if I only run the model with the parent drug and a one compartment model with zero-order absorption everything works as expected, i.e. the OBJ for the IMP step is very stable.

Are there setting for the IMP step that I should be modifying to help stabilize the objective function with the more complex model (i.e. drug + metabolite) vs the more basic model (parent drug alone)?

I appreciate any suggestions.


J. Carl Panetta, Ph.D.
Department of Pharmaceutical Sciences
St. Jude Children's Research Hospital
262 Danny Thomas Place
Memphis, TN 38105
Office: (901) 595-3172
Mobile: (901) 921-3740
Fax: (901) 595-3125

NONMEM Command:

Stochastic/Burn-in Mode
iteration -600 SAEMOBJ= 9495.99674546385
iteration -400 SAEMOBJ= 8603.58235406272
iteration -200 SAEMOBJ= 8723.87279412471
Reduced Stochastic/Accumulation Mode
iteration 0 SAEMOBJ= 8736.22287161761
iteration 200 SAEMOBJ= 8483.86390119489
iteration 400 SAEMOBJ= 8461.33184842421
iteration 600 SAEMOBJ= 8458.72312502707
iteration 800 SAEMOBJ= 8457.40631588857

NONMEM Command:

iteration 0 OBJ= 11357.6200237143 eff.= 825098. Smpl.= 3000. Fit.= 0.99912
iteration 1 OBJ= 11480.4689169933 eff.= ******** Smpl.= 3000. Fit.= 0.99692
iteration 2 OBJ= 11794.3704982717 eff.= ******** Smpl.= 3000. Fit.= 0.99469
iteration 3 OBJ= 11877.2905908383 eff.= ******** Smpl.= 3000. Fit.= 0.99666
iteration 4 OBJ= 12350.7139467883 eff.= ******** Smpl.= 3000. Fit.= 0.99319
iteration 5 OBJ= 12822.3794461581 eff.= ******** Smpl.= 3000. Fit.= 0.98859
iteration 6 OBJ= 13338.5294969671 eff.= ******** Smpl.= 3000. Fit.= 0.98440
iteration 7 OBJ= 13800.6919759203 eff.= ******** Smpl.= 3000. Fit.= 0.98330
iteration 8 OBJ= 14072.5744654244 eff.= ******** Smpl.= 3000. Fit.= 0.98545
iteration 9 OBJ= 14703.6086363413 eff.= ******** Smpl.= 3000. Fit.= 0.98586
iteration 10 OBJ= 15703.8492953385 eff.= ******** Smpl.= 3000. Fit.= 0.97584
iteration 11 OBJ= 16114.3684511057 eff.= ******** Smpl.= 3000. Fit.= 0.96980
iteration 12 OBJ= 15968.9435824919 eff.= ******** Smpl.= 3000. Fit.= 0.98161
iteration 13 OBJ= 16309.4158350109 eff.= ******** Smpl.= 3000. Fit.= 0.96581
iteration 14 OBJ= 16001.2089520460 eff.= ******** Smpl.= 3000. Fit.= 0.97431
iteration 15 OBJ= 16920.1964345088 eff.= ******** Smpl.= 3000. Fit.= 0.95322

ICON plc made the following annotations.
This e-mail transmission may contain confidential or legally privileged information that is intended only for the individual or entity named in the e-mail address. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution, or reliance upon the contents of this e-mail is strictly prohibited. If you have received this e-mail transmission in error, please reply to the sender, so that ICON plc can arrange for proper delivery, and then please delete the message.

Thank You,

ICON plc
South County Business Park
Dublin 18
Registered number: 145835
Received on Tue Nov 24 2015 - 16:22:17 EST

This archive was generated by hypermail 2.3.0 : Fri Sep 27 2019 - 16:47:27 EDT