Re: OMEGA HAS A NONZERO BLOCK

From:Nick Holford Subject:Re: [NMusers] OMEGA HAS A NONZERO BLOCK Date:Tue, 08 Oct 2002 08:28:50 +1300 Ken, I have several times asserted in this thread that it is never reasonable, as a domain expert in PK, to asssme that CL and V are perfectly correlated. To paraphrase your own words: > I just want to see the theory or prior data that supports this > information, otherwise, it just sounds like hyper-subjectivity to me. > Is there some scientific theory that says the correlation is > 1.0?...I doubt it. Put up or shut up :-) {That include you too, Serge) To support my own hyper-subjective viewpoint that CL and V are more likely to be positively correlated (e.g. 0.5 ) I would point out that CL reflects function which increases with size while V reflects structure which also increases with size. Given the common covariate of size then I would expect CL and V to be positively correlated and if that was all that was involved then indeed the correlation would be exactly 1. However, biology is not that simple. For example, in the neonate CL typically increases rapidly (as liver and kidney function mature) while V often decreases (as water is lost). In the elderly clearance may decrease somewhat, due to renal and hepatic function decline, while V may increase (diazepam) or decrease (digoxin) as body composition changes to more fat and less muscle. Add to these examples, the very obvious random between subject variability in CL and V and it is necessarily the case that the correlation between CL and V CANNOT be exactly 1. Size, age! , organ function (and other factors) will determine the magnitude and sign of the correlation. I do not assert that 0.5 is THE ANSWER. It is a suggestion that is the least of 3 evils if we choose not to take a fully Bayesian approach (as I have advocated and LBS has supported). The 3 evils are to fix the correlation to 1, 0 or some other number. I contend that a priori the choice of 1 cannot reflect the real world. The choice of 0 is possible but unlikely given the biology. For a typical case (not at the extremes of life) then I expect the correlation to be positive. Based on simulation of a drug resembling an aminoglycoside (in collaboration with Joga Gobburu and Diane Mould) and including the fixed and random effects mentioned above the correlation between CL and V happened to be 0.3. Analysis of a an actual adult aminoglycoside data set (with Ivan Mathews and Carl Kirkpatrick) estimated the correlation of CL and V1 to be 0.56. As a group (nmusers) I think we have frequently failed to document good estimates of the correlation of PK parameters so I have to say it is hard to point to a review from the literature that would provide data to test the theory. If someone can enlighten me with such a review I would be delighted to hear from you. If you have estimates of the correlation between CL and V and would like to send them to me then I will compile them and report back. Nick