Re: [Fwd: CLIN PHAR STAT: Mixed Vs Fixed]

Date: Wed, 09 Aug 2000 08:14:25 +1200 From: Nick Holford <n.holford@auckland.ac.nz> Subject: Re: [Fwd: CLIN PHAR STAT: Mixed Vs Fixed] Stephen, Stephen Senn wrote: > > I am very ignorant on PK/PD but the analogy here would seem to > be not in the number of patients but in the number of > measurements per patient. In this context, there may be a bias in > variance estimation and associated inferential statistics (CI, P- > values (ugh!) etc) for sparse sampling. However, it depends on the > way you set the model up. If you impose a common residual > variance for each patient then the problem largely disappears. It seemed to me that Mats had suggested a solution to this problem which was to explicitly use a DIFFERENT residual variance for each patient. The usual NONMEM model assumes a common residual error for each patient. What do you think the consequences of using different residual error for each patient might be? > Of course, fully Bayesian methods put a prior on everything and so > deal with this problem. (Or at least appear to deal with it.) I had suggested using the NONMEM Bayesian prior approach earlier in this thread. NONMEM allows one to put a normal (or lognormal) prior on structural model parameters (THETA), an inverse Wishart prior on random effects parameters (OMEGA) and normal/lognormal prior on residual error parameters (SIGMA) (by indirection using THETA). How well does this approach your definition of "fully" Bayesian? [Please reply to nmusers <nmusers@c255.ucsf.edu>] -- Nick Holford, Division of Pharmacology & Clinical Pharmacology University of Auckland, Private Bag 92019, 85 Park Road, Auckland, NZ email: n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.htm