Date: Thu, 15 May 2008 14:36:45 -0400
1) I would not call a OF drop of 11 points a large change; in fact, I'm
betting that you can not even see a difference in your goodness-of-fit
plots. What effect did the inclusion of this covariate have on your eta -
estimate for CL? If it decreased (e.g., before you added the covariate you
CV% was 27%, and after you added the covariate it was 18%) that's what you
want - your covariate is explaining some part of the residual error for
this parameter. If not, I'm not sure I'd keep the covariate in the model
even with an 11 point drop.
2) R2 is a terrible goodness-of-fit metric and should not be used. See
Sheiner LB and Beal SL. Journal of Pharmacokinetics and Biopharmaceutics
(1981) 9:503-512. for the reasons why.
mdkzone_at_aol.com
Sent by: owner-nmusers_at_globomaxnm.com
15-May-2008 10:23
To
nmusers_at_globomaxnm.com
cc
Subject
[NMusers] r Squared from NONMEM
Can anyone tell me how to compute an r-squared (or equivalent) from NONMEM
output.
I have a covariate that markedly diminishes the objective function value
(decrease by 11.5
with an associated chi-square p-value of 0.001); however, I don't know how
to determine
how much of the overall variability inclusion of this covariate accounts
for. In a
simple linear regression we get r-squared. What parameter should I look
at in a
NONMEM output?
Any assistency you could provide would be very much appreciated.
Plan your next roadtrip with MapQuest.com: America's #1 Mapping Site.