FO versus FOCE

From 73532.1742@compuserve.com Fri Aug 4 10:39:06 1995 Subject: FO versus FOCE Hi colleagues and friends, I have been contemplating the following question for some time, and would like to get your input: When searching for a basic PK model, I have encountered the situation that FOCE fits the data "better" (as a tighter more symmetrical envelope of WRES versus the predicted dependent variable and versus the independent variable) than FO. The next step is the search for relevant covariables (such as body weight, creatinine clearance, etc.). Here the question: Have I impeded or improved my chances to find relevant covariables (in the sense of fixed factors which explain interindividual variability) by using FOCE instead of FO? I have left this questions unanswered, and chose to search with FO simply because the data sets are mostly so large that a search (which may involve dozens of NONMEM runs) with FOCE is too time- consuming. Thanks for your thoughts! Joachim Grevel, PhD BAST Inc. 8609 Cross Park Drive Austin, TX 78754 Tel: 512 837-8214 Fax: 512 834-7767 e-mail: 73532.1742@compuserve.com **** From stuart Wed Aug 9 09:39:35 1995 Dr. Grevel recently posted a question on the Net, and Alison Boeckman brought it to my attention, asking if I cared to respond. My reply is marked with (>). Here's the question: Have I impeded or improved my chances to find relevant covariables (in the sense of fixed factors which explain interindividual variability) by using FOCE instead of FO? >Probably will only increase your chances of finding reasonable >covariates by using FOCE, since you also say you have noticed >some particular problem with the fit using FO. However, misfit >with FO can go away as you introduce covariates. I have left this questions unanswered, and chose to search with FO simply because the data sets are mostly so large that a search (which may involve dozens of NONMEM runs) with FOCE is too time-consuming. >As you observe here, it takes much more time to use FOCE. So it is >reasonable to use FO initially. But you should proceed carefully. >After a little model development, wherein you've introduced what seem >to be most, if not all, of the "major" covariable effects, you should again run >FOCE. If then you still see significant better fit with FOCE, you >probably should switch to FOCE, even though this will cost you much CPU >time. It doesn't make sense to try to finally get the model "right" when >your fitting method may be misleading you. A picture of the minor >covariable effects may be easily distorted by a poor fitting method. > >It would be interesting for others to learn what may be the nature of >your situation that may be favoring FOCE. Have you multiple doses? >A particularly nonlinear aspect of your model? Would you care to share this >info over the Users Net? Stu Beal