From: "Batul Parta"
Subject: [NMusers] SD values 
Date: Wed, May 4, 2005 5:04 pm 


I apologize if my question is very simple. Is there a
simple way to estimate the SD and coefficient of
variation of a model parameter estimate in NONMEM? As
an example, if I estimate CL and VC, how do I
calculate the SD of these values?

Thank you all,


From:  "Nick Holford"
Subject: Re: [NMusers] SD values 
Date: Thu, May 5, 2005 3:19 pm


There is some potential for confusion when trying to define the SD and CV of a
parameter like CL when using NONMEM.

It is common to describe the between subject variability of CL in terms of a
variance (estimated as one of the OMEGA parameters).  If the model used for CL is
the usual exponential one e.g.


then the apparent coefficient of variation of between subject variability in CL is
the square root of OMEGA describing the variance of ETA(CL). This would be the same
as the standard deviation of ETA(CL).

If you are lucky then NONMEM may complete the $COVARIANCE step. The NONMEM output
listing contains asymptotic standard error estimates for each of the parameters. You
can compute the coefficient of variation for the uncertainty in the parameter
estimate by dividing the standard error by the parameter value. This CV describes
uncertainty in the parameter which is very different from the CV derived from OMEGA
which describes the variability in the parameter.