Dear Allstats,
Does anyone know of a way of assessing dose-response in data with the
following characteristics:
(1) We have 4 groups of patients- Group 1 members took drug A, group 2 took
drug B, whilst Group 3 took both and Group 4 took drug X. Group members
could continue to take the corresponding drug(s) until death/censoring.
(2) Previous analysis suggest that Patients on drugs A or B or both are less
at risk of premature death than those on X. Dosage is measured by number of
prescriptions (as a proxy).
(3) The aim is to assess whether for each of A and B, there is a dose
response, evidence of efficacy being already reasonable.
(4) Followup starts six months from start of therapy for a maximum of three
years .
THE MAIN STATISTICAL ISSUE:
As expected, cummulative dosage (like DURATION OF EXPOSURE) is highly
correlated with time. However, the efficacy of each drug is expected to
reduce the risk of death, and thus enabling such patients to have longer
time to take in more of the drug. In a conventional dose-response analysis
where exposure increases risk, it is often reasonable to assume and then
assess whether, the longer the duration of exposure or indeed, the bigger
the cumulative dosage, the higher the expected risk of failure. THE
SITUATION HERE IS CLEARLY DIFFERENT. BESIDES, THERE IS NO ALTERNATIVE WAY OF
MEASURING DOSAGE!
QUESTION:
In view of the later, is it reasonable to simply include as additional
covariates, categories of the cumulative dosage in a Proportional Hazards
model with all likely confounders (including the duration of unexposure,
i.e. gap between last therapy and end of follow-up)?
Thanks
F Kiri
_____________________________________________________________________________________
Get more from the Web. FREE MSN Explorer download : http://explorer.msn.com
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|