Bioequivalence and Bioavailability Forum • Dose Proportionality

Dose Proportionality [Regulatives / Guidelines]

posted by Ben – Germany, 2012-03-25 22:18 (5208 d 09:20 ago) – Posting: # 8333
Views: 1,538
(edited on 2012-03-26 17:06)

Dear all,

I am wondering about some details regarding the investigation of dose proportionality and am hoping for some thoughts on that (although this might be a bit off-topic).

Gough et al (1995) wrote that one should (first) investigate a Power model where beta varies between subjects. They suggested two ways for implementing that: (i) two-stage using ANOVA including (fixed effects) dose and subject-by-dose interaction and (ii) a random regression model (beta_i and s_i both random terms). Let's say we only want to use the second because in case of missing data this one allows for a more precise estimate. Now, the question is how can we assess a departure from a common slope for each subject by using this model (ii)? In their paper the refer to the test of subject-by-dose interaction but this one uses fixed effects, so option (i). (Should we apply model (i) to the data just to check for this departure?). They also refer to the estimated variance component in the mixed model, which is said to be "small (ratio to residual variance 0.23)". I guess they mean σ²_beta / σ²_eps. What kind of threshold is sufficiently small?
When should we use the usual ANOVA model? Only in the two-dose case? To identify a dose range for further investigation using the Power model (but the latter can be done by playing around with the data set restricting for some doses and just using the Power model...)? Maybe only in case we get the impression (graphical investigation) that the linear relationship between log(dose) and log(response) is not given...

Thanks and best regards,
Ben

Complete thread:

Dose ProportionalityBen 2012-03-25 20:18