Log in
Register|
Ken Peh ★ Malaysia, 2013-04-05 16:00 (4822 d 09:41 ago) Posting: # 10350 Views: 10,121 |
|
|
Dear All, We can run a 3-way crossover study comparing a test product in different dosage forms (capsule and tablet) and a reference product. For highly variable drug, we need to carry out replicate design, either partial or full replicate. How to fit one reference product and one test product in different dosage forms into replicate design ? Any idea ?  Thank you. Regards, Ken |
|
jag009 ★★★ NJ, 2013-04-05 17:46 (4822 d 07:55 ago) @ Ken Peh Posting: # 10355 Views: 7,844 |
|
|
Hi ken. ❝ For highly variable drug, we need to carry out replicate design, either partial or full replicate. How to fit one reference product and one test product in different dosage forms into replicate design ? Any idea ? You mean setting up the study treatment sequences for T1=capsules, T2=Tablet, Reference (replicate)? John |
|
Ken Peh ★ Malaysia, 2013-04-05 22:26 (4822 d 03:15 ago) @ jag009 Posting: # 10362 Views: 7,756 |
|
|
Dear John, Yes, treatment sequences for T1=capsules, T2=Tablet and Reference. Regards, Ken |
|
ElMaestro ★★★ Denmark, 2013-04-05 19:37 (4822 d 06:03 ago) @ Ken Peh Posting: # 10358 Views: 7,841 |
|
|
Hello Ken, ❝ We can run a 3-way crossover study comparing a test product in different dosage forms (capsule and tablet) and a reference product. ❝ ❝ For highly variable drug, we need to carry out replicate design, either partial or full replicate. How to fit one reference product and one test product in different dosage forms into replicate design ? Any idea ? If I get you right, then Test(Caps)-Test(Tabs)-Ref-Ref and (a subset of) permutations thereof will work. Provided that I haven't misunderstood the latest trends in sequence coding defined by possible descendants of Ötzi this is equivalent (no pun intended) to Hocus(Caps)-Hocus(Tabs)-Pocus-Pocus or something like that plus permutations. — Pass or fail! ElMaestro |
|
Ken Peh ★ Malaysia, 2013-04-05 22:23 (4822 d 03:18 ago) @ ElMaestro Posting: # 10361 Views: 7,801 |
|
|
Dear ElMaestro, ❝ If I get you right, then Test(Caps)-Test(Tabs)-Ref-Ref and (a subset of) permutations thereof will work. Yes. Do you mean similar to fully replicate design (2x2x4)? So, the design is as follows :- Sequence 1 R - T(cap) - T(tab) - R Sequence 2 T(cap)- R - R - T(tab) Are the statistics and computations same as that of fully replicate design? Thank you. Regards, Ken |
|
ElMaestro ★★★ Denmark, 2013-04-06 02:28 (4821 d 23:13 ago) (edited on 2013-04-06 10:53) @ Ken Peh Posting: # 10364 Views: 7,717 |
|
|
Hi Ken, ❝ Yes. Do you mean similar to fully replicate design (2x2x4)? So, the design is as follows :- ❝ ❝ Sequence 1 R - T(cap) - T(tab) - R ❝ Sequence 2 T(cap)- R - R - T(tab) That is one way of doing it. I imagine Detlew will recoil in disgust and mention something about n'th order effects so let's not talk too loud for the sake of his blood pressure. Ideally, in the best of all worlds to make everybody happy you would need all 12 sequences and also balance between them but of course that's absurdly difficult to achieve with certainty. ❝ Are the statistics and computations same as that of fully replicate design? In principle yes. I believe -could be extremely wrong of course- that this is a normal linear mixed effects model with stotal,tabs2 on the diagonal of the covariance matrix where observations are with the test tabs. stotal,caps2 on the diagonal where observations are with the test capsules. sintra,Ref2+sinter,Ref2 on the diagonal where observations are with Ref. sinter,Ref2 off-diagonal where Ref's coincide for a given subject. 0 elsewhere. All in all four variance components. Of course EMA's principle of mandatory fixed factors screws it up, but I am sure it can all be worked out with a linear model too. No idea how to request that in SAS or WNL or R, or how to implement a linear model for the same. I can only do it manually in C. — Pass or fail! ElMaestro |
|
d_labes ★★★ Berlin, Germany, 2013-04-08 13:57 (4819 d 11:43 ago) @ ElMaestro Posting: # 10368 Views: 7,624 |
|
|
Dear ElMaestro, ❝ ... I imagine Detlew will recoil in disgust and mention something about n'th order effects so let's not talk too loud for the sake of his blood pressure. Although a little bit hard of hearing already I have heard your whispering very well! Thanks for considering my health. But don't care. I regularly take my pills in contrast to a long-haired guy  .❝ ... I believe -could be extremely wrong of course- that this is a normal linear mixed effects model with ❝ stotal,tabs2 on the diagonal of the covariance matrix where observations are with the test tabs. ❝ stotal,caps2 on the diagonal where observations are with the test capsules. ❝ sintra,Ref2+sinter,Ref2 on the diagonal where observations are with Ref. ❝ sinter,Ref2 off-diagonal where Ref's coincide for a given subject. ❝ 0 elsewhere. Quoting Helmut's late father: “If you don’t want to know, but to believe – go to church.” IMHO that variance-covariance matrix, although you can do it in C (wow!), is not what such a design would have. What puzzles me is "0 elsewhere" . I would expect some inter-subject (co)variance terms on each off-diagonal element like that for 'normal' replicate designs, f.i. partial replicate (written for the sequence TRR, heterogeneous compound symmetry parametrization): T R Rs2bT=between variance of T, s2wT=within variance of T, sbT=sqrt(s2bT) s2bR=between variance of R, s2wR=within variance of R, sbR=sqrt(s2bR) Not in the mood to work out Kens aimed design T1/T2/R/R the same way. See below. Scott D Patterson and Byron Jones "REPLICATE DESIGNS AND AVERAGE, INDIVIDUAL, AND POPULATION BIOEQUIVALENCE" I. Estimation, Inference, and Retrospective Assessment of Performance of Novel Procedures and the Proposed FDA Methods for Bioequivalence Assessment GSK BDS Technical Report 2002 – 01 (part I) Online resource posted here no longer found, unfortunately. — Regards, Detlew |
|
ElMaestro ★★★ Denmark, 2013-04-08 14:23 (4819 d 11:18 ago) @ d_labes Posting: # 10369 Views: 7,525 |
|
|
Dear d_labes, ❝ I would expect some inter-subject (co)variance terms on each off-diagonal element like that for 'normal' replicate designs, f.i. partial replicate (written for the sequence TRR, heterogeneous compound symmetry parametrization): ❝ ❝ ❝ ❝ ❝ s2bT=between variance of T, s2wT=within variance of T, sbT=sqrt(s2bT) ❝ s2bR=between variance of R, s2wR=within variance of R, sbR=sqrt(s2bR) This, I think, does not look right; e.g. lower left and upper right should be identical; I would possibly put rho*sbT*sbR top right if the solution with rho is desired. I never understand why we want a rho but what do I really know?!? The zeros elsewhere applies whenever two observations are from different subjects. Note that I speak of V (=ZGZt+R), rather than G or R alone. Anyways, for the case of three treatments (2 R's and one T) V will, assuming data are ordered by subject, have such blocks as you suggest diagonally and zeros elsewhere. — Pass or fail! ElMaestro |
|
d_labes ★★★ Berlin, Germany, 2013-04-08 18:50 (4819 d 06:51 ago) @ ElMaestro Posting: # 10372 Views: 7,581 |
|
|
Dear Old pirate! ❝ This, I think, does not look right; e.g. lower left and upper right should be identical; Got me! I correct as follows: T R R❝ The zeros elsewhere applies whenever two observations are from different subjects. Agree. This is the assumption of independence of data from different subjects. — Regards, Detlew |
|
d_labes ★★★ Berlin, Germany, 2013-04-08 11:10 (4819 d 14:31 ago) @ Ken Peh Posting: # 10367 Views: 7,812 |
|
|
Dear Ken, ❝ Sequence 1 R - T(cap) - T(tab) - R ❝ Sequence 2 T(cap)- R - R - T(tab) To be period balanced (i.e. to be able to estimate period effects and get rid of them) each of your formulations have to occur at least once in each period. IMHO your aimed design is a mixture of two partial replicate designs (T1RR and T2RR). According to ElMeastroso: Hokus-Pokus-Fidibus-Fidibus (combined). So if you really have to use such a design derive the sequences starting from the sequences T1RR/RT1R/RRT1 and try to add T2 in a period balanced manner. F.i. T1 R R T2red: modifications to the TRR/RTR/RRT ❝ Are the statistics and computations same as that of fully replicate design? I think: no  . If you think in the direction of the FDA scaled ABE you have to evaluate two linearized ABE criterions (or a mixture?). That immediately rises the question of alpha adjustment. If you think in EMA widened ABEL direction: May be you can calculate T1-R and T2-R from ANOVA of all data and s2wR from data on R only. But again the question of alpha adjustment arises. Next question would be the regulatory acceptance. I would be very unsure about that. What do you think? I myself would not go with such a design due to the uncertainties involved. — Regards, Detlew |
Ing. Helmut Schütz