Re: Data Construct / ADVAN selection

From: Matthew Riggs <riggsmm@yahoo.com> Subject: Re: Data Construct / ADVAN selection Date: Wed, 11 Apr 2001 05:58:38 -0700 (PDT) Amir, I would start by referring you to recent answer provided by Lewis Sheiner to a similar question (copied below)... " 3. The additive etas may give you trouble in the posthoc step; it's probably safer to use $PK V=THETA(1)*EXP(ETA(1)) CL=THETA(2)*EXP(ETA(2)) 4. You should probably start with a full OMEGA, not a diagonal one; you can then see if a diagonal one suffices. Thus you might start with $OMEGA BLOCK(2) 0.25 .1 0.25 LBS." Additionally, with this relatively rich data set you may want to try the first-order conditional estimation method (Method=1) with interaction versus the default first-order method. In your case, you may change the default using: $EST SIGDIGITS=4 MAXEVAL=999 METHOD=1 INTERACTION PRINT=5 My final suggestion would be to define your typical value (poulation mean) VSS as TVSS, rather than TVVSS. NONMEM will only read 4 letters of a user defined variable. This may help your Bayesian estimated (POSTHOC) individual VSS estimates. Another diagnostic: Was the gradient for ETA4 zero? Or was the final estimate for ETA4 zero or the same as the initial estimate? Hope this helps. Matthew Riggs