I am fitting a parent-metabolite model simultaneously. I got a return code of 1 for my basemodel (first order absorption+2 cmt for parent+2 cmt for the metabolite with the fraction metabolized (Fm) fix to a literature value). As eta correlation was observed between Vparent_central and CLparent, omega block between them was added and the OFV dropped significantly (dOFV=-20). However, my return code turns to 3.
a return code of 1 is obviously more desirable than a return code of 3, but even with a return code of 3 it already has a better OFV and should be selected as the better model than the return code=1 model. There might still be room for a bit of improvement and that could further improve OFV. For example, you could take QRPEM first to get a better starting point for a second run with FOCE ELS.
As the model parameters become less identifiable for various reasons you will start getting return code 3 with less likelihood to be able to get standard error of estimates.
In addition to Bernd suggestions I would add:
try to make the model simpler by fixing a parameter or removing a parameter that is no longer needed.
Also, when you have a current best model try to run a bootstrap on it so you can assess the confidence interval on parameters to see if any has big uncertainty.
First I tried QRPEM and got the error message below:
FATAL ERROR IN QRPEM/IMPEM!
Model not suitable for QRPEM analysis
Possibly non linear covariate model or
some other unimplemented feature
Is that because my model is too complicated? My structure model or random effect model?
I agreed with Samer, after adding omega block, in addition to return code from 1 to 3, the condition number from 800+ to 10^8. What should I consider first, go back to the model without omega block or remove other unnecessary ETAs, like nVparent_preph and nCLparent_preph?