Dear Mohamed,
A logit transformation of data will transform your VAS scores from - infinity
to + infinity.
Varun Goel
--- On Mon, 6/16/08, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
From: [EMAIL PROTECTED] <[EMAIL PROTECTED]>
Subject: [NMusers] Log Transformation in NONMEM
To: nmusers@globomaxnm.com
Date: Monday, June 16, 2008, 5:17 PM
Dear all,
I developing a mixed effects PK/PD model for VAS sleepiness reported
by 20 healthy volunteers after 2 mg oral lorazpam administration.
Since the VAS scale is bound, the distribution of VAS scores at
various time points is right skewed. However, when I look at the
distribution of my WRES in my model it is only very slighlty right
skewed.
My questions are:
1) Do we base the need to do a Log transformation on diagnostics from
the data (e.g. score distributions) or diagnostics of the model (e.g.
WRES).
2) Also I have zero baseline data. I realize there is a need to bias
the data with a constant if I am to Log transform. I added a small
constant (0.01), but in my concordance plots I get all my baseline
points below the line of concordance. Is this because the Log
transformation treats data between 0 and 1 differently from numbers
higher then 1, i.e. (taking the log 0f 0.01 results in a negative
number, while doing so for numgers greater then 1 results in positive
numbers.
Would following the log-transformed model that introduces an
additional theta to account for systematic bias be applicable in this
scenario
(reported by by Beal, JPP 2001;28:481-504)? Has anybody tried this?
M = THETA(n)
Y = LOG(F+M) + (F/(F+M))*EPS(1) + (M/(F+M))*EPS(2)
3) My maximum VAS is 100 mm (or a log of 2). Yet when I log
transform, I get predictions with Log values higher then 2. Is there
any way to place a constraint so as not to get predictions higher than
log 2.
Thanks in advnace, Mohamed
Mohamed A. Kamal, Pharm.D.
Ph.D. Candidate
Department of Pharmaceutical Sciences
University of Michigan
