Sample distribution of etas

Dear NMusers, When running simulation, I presume NONMEM samples etas from prior distribution N(0, omega^2), if I am correct? Is there a way to ask NONMEM to sample etas from posterior distribution (conditional on observed data)? Any ideas would be appreciated. Yours sincerely, Tingjie Guo

Hi Tingjie, There is a way, but it requires some work. The posterior distribution is not necessarily a normal distribution. It can be approximated through the inverse FIM. This approximation is written by NONMEM in the .phi output file, as variance on the estimated conditional mode. You can then perform Monte Carlo sampling from the conditional modes and varcov matrix to obtain 1000 eta vectors for each subject. You can add these to your data file, and adapt your control stream to use the simulated ETA's from your data file. Monolix allows to output a simulated eta distribution using MCMC, sampled from the individual log-likelihood (which may not be normal-distributed). See http://monolix.lixoft.com/tasks/conditional-distribution/ for more information. I am not aware of similar functionality for NONMEM. Best regards, Ruben Faelens Op wo 13 mrt. 2019 13:41 schreef Tingjie Guo <[email protected]>: > Dear NMusers, > > > > When running simulation, I presume NONMEM samples etas from prior > distribution N(0, omega^2), if I am correct? Is there a way to ask NONMEM > to sample etas from posterior distribution (conditional on observed data)? > Any ideas would be appreciated. > > > > Yours sincerely, > > Tingjie Guo >

Tingjie: In NONMEM the most efficient and versatile way of obtaining samples from the posterior distribution for any parameters you desire is that shown by example 8 in the intro7.pdf manual in the section: Example 8: Sample History of Individual Values in MCMC Bayesian Analysis Robert J. Bauer, Ph.D. Senior Director Pharmacometrics R&D ICON Early Phase 820 W. Diamond Avenue Suite 100 Gaithersburg, MD 20878 Office: (215) 616-6428 Mobile: (925) 286-0769 [email protected]<mailto:[email protected]> http://www.iconplc.com/