Dear Allstat
I have a dose-response problem which I would appreciate advice on.
Many anaesthetic studies have looked at dose-response of a
drug, with respect to depth of sedation (a continuous
measure from 0-100) at a particular dose level. Repeated
measurements over time are made at several different dose
levels on the same patients, and then a sigmoid Emax model
is fitted using specialist software to estimate the EC50
level at half maximal effect (when BIS=50). The resulting
EC50 is expressed as a concentration with a 95% confidence
interval.
In the papers I have been shown as examples the data seem
to be analysed as if they are independent observations,
taking no account of the correlation between measures made
on the same patient. I have advised a clinician who wants
to repeat this methodology that it is flawed and that he
should either (i) randomly allocate a group of patients to
four separate dose levels (ie. a different group of
patients for each dose level) and analyse using standard
dose-response techniques (eg. linear regression), or (ii)
take account of the correlation using more complex
longitudinal data analysis for which he would need
statistical support.
I am not familiar with the pharmacodynamic sigmoid Emax
model and wonder if I am missing something in terms of the
design, since the above seems to be standard practice in
quite a large number of anaesthetic studies. I think what
I have suggested seems reasonable and the clinician will
probably go for option (i), but would appreciate some
reassurance from someone who has more experience with these
types of studies!
----------------------
Dr. Gill Lancaster
Division of Statistics and OR
Department of Mathematical Sciences
M & O Building
University of Liverpool
L69 3BX
Tel: 0151-794-4733
Fax: 0151-794-4754
Email: [log in to unmask]
|