Dear all,
I am fitting a ADVAN3 TRANS4 model with M3 method. The base model (MODEL_001) consists of single parameter estimates for each PK parameter and its distribution, proportional+additive residual error model and effect of BSA on V1. The only difference of MODEL_002 is that the additive component of the RUV is removed, by fixed SGM_PROP and EPS_PROP to zero.
But when fitting MODEL_002, the following error occurs at the first observation of the first subject (i.e. ID=1, rec 2), which does not happen for MODEL_001
OCCURS DURING SEARCH FOR ETA AT INITIAL VALUE, ETA=0
F OR DERIVATIVE RETURNED BY PRED IS INFINITE (INF) OR NOT A NUMBER (NAN).
So far I cannot identify the cause of the issue. ID=1, rec 2 has a DV above LLOQ so the likelihood should be computed based on the DV and the error size (i.e. F_FLAG = 0
). I have checked the syntax a few times and it seems to be irrelevant.
Please find a reproducible example in the link below. Any help is appreciated!
====================================================
SOLVED:
So, the problem is due to the dosing record, which has IPRED=0, being put at the denominator during evaluation. I tried to bypass this problem this way, which seems to be helpful at this point:
IF (F.EQ.0) CUMD=1
IF (F.GT.0) CUMD=PHI((LOQ-F)/SD)
IF (DV.GE.LOQ) THEN
F_FLAG=0
Y=F+ERR_PROP+ERR_ADD
ELSE
F_FLAG=1
Y=CUMD
ENDIF