Re: [NMusers] Additional predicted levels (EVID=2) changed objective function values

From: Denney, William S. <William.S.Denney_at_pfizer.com>
Date: Mon, 15 Feb 2016 14:05:21 +0000

Hi Zheng,

My first guess is that you have a time-varying covariate in your data set, and the way that NONMEM handles time varying covariates in the data set is that they are kept fixed until an instantaneous change when the new record appears. If your EVID=2 records change the interpretation of the covariate (either carrying an observation backward or interpolating it), then an objective function change is expected.

If you can post your control stream and a part of your data set, it would be easier to help.

Thanks,

Bill

On Feb 15, 2016, at 5:46, "Zheng Liu" <Zheng.Liu_at_rch.org.au<mailto:Zheng.Liu_at_rch.org.au>> wrote:


Dear NMusers,


I intended to print out more predicted levels at the time points without observations. Therefore I inserted those lines with EVID=2 in the data set. Surprisingly after doing so, the objective function value and the fitted parameter values changed more or less. This should not be rational. Could anyone explain the reason and how to avoid it? Thanks a lot in advance!


Best regards​


​ Zheng Liu, Ph.D.
Pharmacometrician (postdoc), Melbourne Royal Children's Hospital
email: Zheng.Liu_at_rch.org.au<mailto:Zheng.Liu_at_rch.org.au>
Received on Mon Feb 15 2016 - 09:05:21 EST

This archive was generated by hypermail 2.3.0 : Fri Sep 27 2019 - 16:48:56 EDT