Re: OMEGA HAS A NONZERO BLOCK

From:Nick Holford Subject: Re: [NMusers] OMEGA HAS A NONZERO BLOCK Date:Sat, 05 Oct 2002 09:47:46 +1200 Lewis, Thanks for sermonizing on the Bayesian perspective. I had been trying to advocate this all along in my contributions to this thread (based largely on my conclusion from a similar thread last year). However, you do not seem to recognize this in your comments below and I wonder why you seem to cast me in the role of being "hyper-subjective"? Perhaps it is because there is another element of uncertainty here -- how to solve this problem in practice. NONMEM has some limited ability to do a sort of Bayesian estimation (Stuart asserts the use of PRIOR is not really Bayesian). To my knowledge nobody has ever applied NONMEM with priors on the covariance although it is clear how this might be done. Given that the NONMEM prior method is unsupported, undocumented and has almost no literature support it is not surprising that many people would be reluctant to use it for real problems. Fixing a parameter to a reasonable "hyper-subjective" prior is a well known pragmatic method for data description (and recommended by FDA guidance e.g. fixing KA). As a practical solution I think that this is within the comfort zone of most who have to face this problem and want to use NONMEM to solve it. I quite agree that one should be more uncomfortable when relying on this fixed value for simulation. I am still hoping that Steve Duffull, who started this thread rolling, will come back on line sometime and tell us his experiences of looking at this same problem using WinBugs. Nick Nick Holford, Divn Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand email:n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.health.auckland.ac.nz/pharmacology/staff/nholford/