SP ☆ India, 20190729 12:31 (380 d 08:13 ago) Posting: # 20465 Views: 1,954 

Hello Helmut/All, Can anyone help me with your experience or suggestions. Which sequence should i prefer for crossover study design to Reference Replicate and Two Different Test product (T1 and T2). Thank You!!!! Edit: Category changed; see also this post #1. [Helmut] 
Helmut ★★★ Vienna, Austria, 20190729 15:25 (380 d 05:19 ago) @ SP Posting: # 20466 Views: 1,704 

Hi SP, » Can anyone help me with your experience or suggestions. Which sequence should i prefer for crossover study design to Reference Replicate and Two Different Test product (T1 and T2). I have seen this one: T_{1}RT_{2}R  RT_{1}RT_{2} Evaluation was done after excluding the respective other Ttreatment, which gave two partial replicates with incomplete blocks (the pooled analysis will lead to limbo for unequal variances): T_{1}R•_{ }R  RT_{1}R•_{ } and•_{ }RT_{2}R  R•_{ }RT_{2} As usual with partial replicates, the FDA’s covariance specification (in SASlingo FA0(2) ) did not converge for PK metrics with s_{WR} <0.294 (no RSABE). Furthermore, one has to assume lacking period effects, since neither R nor the two test are administered in all periods. The evaluation was not easy. Since nobody knows whether or not there were true period effects, the outcome was doubtful at least. I would not go there.Maybe it is better to opt for a modified Williams’ design T_{1}T_{2}RR  T_{2}RT_{1}R  RT_{1}RT_{2}  RRT_{2}T_{1} which givesT_{1}•_{ }RR  •_{ }RT_{1}R  RT_{1}R•_{ }  RR•_{ }T_{1} and•_{ }T_{2}RR  T_{2}R•_{ }R  R•_{ }RT_{2}  RRT_{2}•_{ } At least balanced for T_{1} and T_{2} (once in every period). However, we have R twice in periods 1&2 and only once in periods 3&4. If you really want to have everything balanced, you might end up with even more sequences. The FDA argues against more than two sequences (confounded effects) anyway. Duno. Never have been there. I would perform two separate three period full replicate studies: T_{1}RT_{1}  RT_{1}R and T_{2}RT_{2}  RT_{2}R No statistical pitfalls. As a bonus you get also the intrasubject variabilities of T_{1} and T_{2}. If T_{1}/R and T_{2}/R are similar, select the one with lower variability for the pivotal study (full replicate, please – the partial replicate is crap). — Diftor heh smusma 🖖 Helmut Schütz The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes 
abhimanyu ☆ Banglore, 20190803 11:54 (375 d 08:51 ago) @ Helmut Posting: # 20469 Views: 1,561 

Hi Helmut/SP, » Maybe it is better to opt for a modified Williams’ design T_{1}T_{2}RR  T_{2}RT_{1}R  RT_{1}RT_{2}  RRT_{2}T_{1} which gives» T_{1}•_{ }RR  •_{ }RT_{1}R  RT_{1}R•_{ }  RR•_{ }T_{1} and» •_{ }T_{2}RR  T_{2}R•_{ }R  R•_{ }RT_{2}  RRT_{2}•_{ } » At least balanced for T_{1} and T_{2} (once in every period). » However, we have R twice in periods 1&2 and only once in periods 3&4. If we consider modified Williams’ design with this sequence T1T2RR  T2RT1R  RT1RT2  RRT2T1 Then, What would be change in Proc GlM / Proc Mixed for Calculate Swr, 95% Upper bound (Scaled), and 90% CI (Unscaled). Thanking you in advance!! — Abhimanyu 
d_labes ★★★ Berlin, Germany, 20190803 17:35 (375 d 03:09 ago) @ abhimanyu Posting: # 20470 Views: 1,530 

Dear Abhimanyu, » ... » What would be change in Proc GlM / Proc Mixed for Calculate Swr, 95% Upper bound (Scaled), and 90% CI (Unscaled). Nothing since the effects in GLM or mixed are not dependent from the used sequences, only the number of levels of the effect "sequence" or "period" changes. But this has not to be specified in the code (SAS or others). It will be taken automatically from the dataset you use. Thus you have: log(PK) ~ period + sequence + subject(sequence)  for calculation of sWR log(PK) ~ treatment + period + sequence + subject(sequence)  for 90% CIin R lingo. In SAS you have to modify the model statement in Proc GLM accordingly.— Regards, Detlew 
abhimanyu ☆ Banglore, 20190805 07:27 (373 d 13:18 ago) @ d_labes Posting: # 20471 Views: 1,518 

Dear Detlew, » log(PK) ~ period + sequence + subject(sequence)  for calculation of sWR» log(PK) ~ treatment + period + sequence + subject(sequence)  for 90% CIThank you for providing valuable information. Moreover, how to validate answers? Kindly suggest available method and References for the same. Thanking you Regards, Abhimanyu 