Yura ★ Belarus, 2016-11-23 09:50 (2703 d 18:04 ago) Posting: # 16808 Views: 4,974 |
|
Dear All, We spend in replicate design TRR / RTR / RRT three repeated measurements. Question: how to take the level of alpha and confidence interval for the point estimate calculation between measurements (T-R1, T-R2, R1-R2)? I assumed an alpha of 0.017 and CI 96.6. Thanks for your comments in advance Above - for Cmax. AUCt - without reference drug - Alpha 0.025 and CI 95,0? Edit: Merged with a later post. You can edit your OP within 24 hours (see here). [Helmut] |
ElMaestro ★★★ Denmark, 2016-11-24 22:35 (2702 d 05:18 ago) @ Yura Posting: # 16810 Views: 4,092 |
|
Hello Yura, ❝ We spend in replicate design TRR / RTR / RRT three repeated measurements. Question: how to take the level of alpha and confidence interval for the point estimate calculation between measurements (T-R1, T-R2, R1-R2)? I assumed an alpha of 0.017 and CI 96.6. Something sounds a bit odd here. TRR/RTR/RRT with two different Ref's is not a replicate design; alpha adjustment is in that situation usually not what you want to do. But I would be entirely wrong, depending on the question you are actually asking in your trial. What is the purpose of the R1-R2 CI? Let's get some more info. — Pass or fail! ElMaestro |
Yura ★ Belarus, 2016-11-25 08:51 (2701 d 19:03 ago) @ ElMaestro Posting: # 16811 Views: 4,193 |
|
Hello, ElMaestro Design - partial replicate. Each volunteer experience once the tablet T and twice R. Receive individual BE. For the difference T - R (1) and T - R (2) is carried out and the point estimate is based CI. If extend beyond BE, using the R (1) - R (2) - to expand CI, but only for Cmax. Thanks |
Helmut ★★★ Vienna, Austria, 2016-11-25 14:00 (2701 d 13:54 ago) @ Yura Posting: # 16813 Views: 4,316 |
|
Hi Yura, since Belarus as a member of the EAEU (which in its GL closely followed the EMA’s rules for reference-scaling by Average Bioquivalence with Expanding Limits – ABEL), according “to the book” no adjustment of α is required (i.e., 90% CI of T vs. R). However, when you suspect multiplicity issues – which might lead to an inflation of the Type I Error – you are right (Labes and Schütz1, Muñoz et al.2, Wonnemann et al.3)! Adjusting α in such a way that the consumer risk is preserved at 0.05 is provided in the open-source package PowerTOST , function scABEL.ad() .No adjustment of α is required for PK metrics assessed by (conventional unscaled) ABE (like AUC). Adjustment for PK metrics intended for reference-scaling (like Cmax) depends on the CVwR and – to a minor degree – on the sample size. I would not recommend Bonferroni’s correction (i.e., α 0.025) because generally it is unnecessarily conservative and negatively impacts power. BTW, in rare cases (extremely high sample sizes) you would have to go below 0.025…
— 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 |
Yura ★ Belarus, 2016-11-25 20:28 (2701 d 07:26 ago) @ Helmut Posting: # 16814 Views: 4,045 |
|
thank you, Dear Helmut |