Re: SAEM and IMP
Hi Pavel
I have experienced a similar problem. In my case, the
following code for IMP after SAEM (using NM7.3) greatly reduced the
Monte Carlo OFV noise from variations of about +/- 60 points to variations of
+/-
6 points (though still not good enough for covariate testing):
$EST METHOD=IMP LAPLACE INTER NITER=15 ISAMPLE=3000 EONLY=1 DF=7
IACCEPT=0.3
ISAMPEND=10000 STDOBJ=2 MAPITER=0 PRINT=1 SEED=123456 RANMETHOD=3S2
The settings are explained in the NM7.3 guide. If you are using
NM7.3, you can also try IACCEPT=0.0 whereupon "NONMEM will determine the
most appropriate IACCEPT level for each subject". Of course the settings
for DF and IACCEPT in the above code will depend on the type of data you have.
Which brings me to my own question. If I have both continous and categorical
DVs in the dataset (which would mean different optimal settings) and I
am using F_FLAG accordingly, what would the 'right' values of DF and
IACCEPT be? I have noticed that the DF automatically chosen by NONMEM for
individuals in the dataset can vary from 0-8 and this appears to be random.
Emmanuel
Quoted reply history
From: Pavel Belo <[email protected]>
>To: [email protected]
>Sent: Thursday, May 15, 2014 11:01 AM
>Subject: [NMusers] SAEM and IMP
>
>
>Hello NONMEM Users,
>
>As SAEM does not provide a useful objective function, the manuals
>recommend using IMP after SAEM. It works well in many cases when IMP
>works well. When IMP works well, SAEM is not always needed. SAEM is
>really needed when the other methods do not work well.
>
>The issue is that there are hard cases when SAEM works very well and IMP
>does not work at all. SAEM provides meaningful and consistent PK/PD
>parameters across very different runs, while IMP provides objective
>function, which varies so greatly that it looks meaningless. Another
>potential issue with IMP is that even when it works well with a problem,
>it occasionally provides low values of objective functions after SAEM or
>as the first estimate (less frequently in the middle of IMP run) and
>then becomes unstable or jumps to much higher objective function and
>then converges to something between the low and the high values for a
>long time. It almost looks like IMP does not show some kind of
>integration/computation errors and keeps running providing a funny
>objective function.
>
>It seems like we cannot estimate objective function when SAEM runs well
>and IMP does not. It reduces the value of SAEM. Is there a way around
>it?
>
>Thanks,
>Pavel
>
>
>