I would concentrate on improving the model: FOCEI should be fine (unless
you would like to use random effect on the lag time: then may not work
well). Some options to improve absorption part are the time-variable KA
(implemented using MTIME option) or sequential zero-order then first
order absorption.
Thank you
Leonid
On 8/16/2024 9:53 AM, Qiuyue Li wrote:
Dear nmusers,
I have a question regarding the selection of a model using different
estimation methods. My dataset consists of 20% of the data from a
relatively extensive sampling scheme (about 6 data points per
subject, with 1-2 points during the absorption phase) and the rest from
sparse samples (about 3 data points per subject, with one point near the
Tmax). I initially employed the FOCEI estimation method, but the GOF
plots exhibited underestimations in the PRED values. The pcVPC plots
indicated an underestimation of the absorption phase, where the median
observed concentrations were higher than the 95% CI of the simulated
median concentrations.
When I switched the SAEM estimation method for the same structure model,
there was minimal improvement in the GOF plots. However, the pcVPC
demonstrated a better fit, with the median observed concentrations
falling within the 95% CI of the simulated median concentrations.
For the absorption model, I tried a first-order absorption model with
lag time and a transit compartment absorption model. When switching from
the FOCEI algorithm to the SAEM algorithm, both models showed similar
trends in the VPC plots.
Given these observations, would the SAEM estimation method be more
appropriate? What might explain the differences between these two
approaches? Additionally, when using the FOCEI algorithm, MAXEVAL= 0 is
set to obtain the empirical Bayes estimates (EBEs) of individual PK
parameters and derive exposure. How should this be handled when using
the SAEM algorithm?
Thanks very much for your explanation.
Best Regards
Qiuyue
