Log in
Register|
zan ☆ US, 2014-01-30 01:08 (4179 d 00:09 ago) Posting: # 12290 Views: 6,528 |
|
|
All, My knowledge to obtain an accurate intrasubject CV% is to conduct a replicated crossover design dosing same formulation twice. From a 2x2 crossover study where signiciant treatment effects exist, eg. a significant food-effect study, we are still able to obtain the intrasub CV% from the output. As I was going to use these intrasub CV% to help biostats to est sample size for a BE study, someone challenged that these number are not real and are likely inflated due to the signficant treatment effect. Is it true? I am wondering does it invalidate the use of intrasub CV% in studies with significant tx effect or these estimates are still close to those from a replicated study. Many thanks and best regards zan Edit: Category changed. [Helmut] |
|
Helmut ★★★   Vienna, Austria, 2014-01-30 03:12 (4178 d 22:05 ago) @ zan Posting: # 12292 Views: 5,508 |
|
|
Hi Zan, ❝ My knowledge to obtain an accurate intrasubject CV% is to conduct a replicated crossover design dosing same formulation twice. If you talk about the intra-subject CV of treatments, yes. ❝ From a 2x2 crossover study where signiciant treatment effects exist, eg. a significant food-effect study, we are still able to obtain the intrasub CV% from the output. Correct. In the 2×2 the ISCV is actually derived from a common variance of the treatment ones. If treatment variances are substantially different (e.g., high for a ‘bad’ reference and low for a ‘good’ test), the residual error – and therefore the CI – will be inflated. That’s why we need high sample sizes for 2×2 Xovers of HVDPs – the test will be punished for the reference’s CV. In (fully) replicated design together with RSABE in this case you will get a reward in terms of sample size. But that’s somehow OT. ❝ As I was going to use these intrasub CV% to help biostats to est sample size for a BE study, someone challenged that these number are not real and are likely inflated due to the signficant treatment effect. Is it true? I don’t think so. Imagine the distribution (normal or lognormal, doesn’t matter). It is defined by two parameters, the mean and the variance. These two are independent. ![[image]](img/uploaded/image249.png) ![[image]](img/uploaded/image250.png) — Dif-tor heh smusma 🖖🏼 Довге життя Україна! ![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png) Helmut Schütz ![[image]](https://static.bebac.at/img/CC by.png) The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
zan ☆ US, 2014-01-30 19:26 (4178 d 05:51 ago) @ Helmut Posting: # 12297 Views: 5,461 |
|
|
❝ […] In the 2×2 the ISCV is actually derived from a common variance of the treatment ones. If treatment variances are substantially different (e.g., high for a ‘bad’ reference and low for a ‘good’ test), the residual error – and therefore the CI – will be inflated. That’s why we need high sample sizes for 2×2 Xovers of HVDPs – the test will be punished for the reference’s CV. In (fully) replicated design together with RSABE in this case you will get a reward in terms of sample size. But that’s somehow OT. Very interesting. I am wondering if an estimated CVintra from a 2×2 study (test two different formulation) is still a good estimate when the Cmax or AUC difference between the two formulations are <30%, assumming this 30% is not substantially different? Thanks zan Edit: Full quote removed. Please delete everything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut] |
|
ElMaestro ★★★ Denmark, 2014-01-30 09:38 (4178 d 15:39 ago) @ zan Posting: # 12293 Views: 5,490 |
|
|
Hi Zan, ❝ As I was going to use these intrasub CV% to help biostats to est sample size for a BE study, someone challenged that these number are not real and are likely inflated due to the signficant treatment effect. Is it true? I am wondering does it invalidate the use of intrasub CV% in studies with significant tx effect or these estimates are still close to those from a replicated study. Treatment is a fixed effect. When we maximise the likelihood of the model given the observed data, we figure out two constants for the treatments (and for the other fixed effects) which help towards minimisation of the sums of squares. A significant treatment effect is thus not causing a statistical inflation of the residual sums of squares per se but will increase the difference for the treatment LS Means or effect vector values. Or to say it differently: A treatment effect just increases the total/null/model-free sums of squares but not the residual. — Pass or fail! ElMaestro |
|
Helmut ★★★ Vienna, Austria, 2014-01-30 17:30 (4178 d 07:47 ago) @ ElMaestro Posting: # 12294 Views: 5,523 |
|
|
Hi ElMaestro, mindblowing wording.  ❝ When we maximise the likelihood of the model given the observed data, we figure out two constants for the treatments […] which help towards minimisation of the sums of squares. <nitpicking> Maximizing the likelihood ≠ minimizing the sum of squares. </nitpicking>These are two different approaches. The fact that we get similar – if not identical – estimates in most cases is an amazing technical coincidence. — Dif-tor heh smusma 🖖🏼 Довге життя Україна! Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
ElMaestro ★★★ Denmark, 2014-01-30 18:08 (4178 d 07:09 ago) @ Helmut Posting: # 12295 Views: 5,426 |
|
|
Hi Hötzi, ❝ <nitpicking> Maximizing the likelihood ≠ minimizing the sum of squares. </nitpicking>❝ ❝ These are two different approaches. The fact that we get similar – if not identical – estimates in most cases is an amazing technical coincidence. That's a fair comment to raise. I see it exactly oppositely: With a single variance component minimisation of SS is the technical shortcut to finding the actual max likelihood solution as long as a period value is not missing. — Pass or fail! ElMaestro |
Ing. Helmut Schütz