Question regarding Parameter Freezing for Simulation: Should I fix both Fixed and Random Effects?

Phoenix Certara Forums Modelling and Simulation
,
1

Hi everyone,

I am currently working on a simulation step using my final PK/PD model in Phoenix NLME. I have a question regarding the best practice for “freezing” parameters before running the simulation.

When transitioning from the final estimation to the simulation stage:

  1. Fixed Effects (THETAs): I understand these should be fixed to the final estimates.

  2. Random Effects (OMEGAs and SIGMAs): Should I also freeze the Variance-Covariance matrix (Omega) and the Residual Error (Sigma) values?

My goal is to perform a population simulation to predict concentration and suggest dose.

In my current setup, I am considering fixing all parameters to ensure the simulation strictly follows the final model’s estimates. Is it standard practice to freeze both, or are there specific scenarios where the random effects should remain “unfrozen”?

I would appreciate any guidance or links to best practice resources on this.

Thank you in advance!

Virunya

2

Typically, the workflow for a sequential PK/PD model is to try to fix “individual parameters” to the previously estimated values. The goal is to fix parameters based on the model purpose and data support. So it seems here if the goal is to suggest dose and perhaps covariate testing or parameter identifiability, then you may want to consider fixing only the poorly informed fixed effects parameters. When you fix both fixed and random effects, you’ll ignore BSV, which may be not be appropriate for population predictions or trial simulations. TLDR: no, but if you have to constrain, I’d prefer fixed effects with a priori information and justify via diagnostics.

3

True, Adevanathan12. Dear Varunya, if you freeze Random Effects (OMEGAs and SIGMAs) then you will not get IIV and residual error loosing uncertainty and main purpose of Population PK.

Thank you.

Bhim