Log in
Register|
arl_stat ★ India, 2013-03-25 13:32 (4829 d 11:53 ago) Posting: # 10272 Views: 8,356 |
|
|
Dear All, We are in the process of conducting Bioequivalence Study in two batches. The Anova model is as follows. Batch, Sequence, treatment, batch*treatment and Period(batch), and Subject(sequence*Batch) as factors. Plz let me know if this model can be considered for conducting Bioequivalence Study in two batches? Also if this model accepted by the Regulatory authorities? Thanks in advance. |
|
ElMaestro ★★★ Denmark, 2013-03-25 16:23 (4829 d 09:02 ago) @ arl_stat Posting: # 10275 Views: 7,348 |
|
|
Hello Arl_stat, ❝ We are in the process of conducting Bioequivalence Study in two batches. The Anova model is as follows. ❝ Batch, Sequence, treatment, batch*treatment and Period(batch), and Subject(sequence*Batch) as factors. You forgot to give some very important info there. Is this a crossover? Are there two ref. batches and one test batch? Only two ref. batches and zero test batch (if the latter then batch*treatment might not make sense)? One ref. batch per subject only? What do you wish to accomplish with batch*treatment? — Pass or fail! ElMaestro |
|
Ohlbe ★★★ France, 2013-03-25 17:50 (4829 d 07:34 ago) @ arl_stat Posting: # 10276 Views: 7,290 |
|
|
Dear arl_stat, ❝ We are in the process of conducting Bioequivalence Study in two batches. What do you mean with "batches" exactly: 2 different batches of the same drug product ? Or two different groups of subjects, pooled in the end ? Regards Ohlbe — Regards Ohlbe |
|
arl_stat ★ India, 2013-03-25 19:27 (4829 d 05:58 ago) @ Ohlbe Posting: # 10277 Views: 7,253 |
|
|
Hello All, It is 2 treatment, 2 period, 2 sequence crossover study (N=50). And batches considered are as follows: Subject no. 1 to 25 as Batch-I AND Subject no. 26 to 50 as Batch-II. Thanks. |
|
ElMaestro ★★★ Denmark, 2013-03-25 20:34 (4829 d 04:51 ago) @ arl_stat Posting: # 10278 Views: 7,223 |
|
|
Thank you, arl_stat, ❝ It is 2 treatment, 2 period, 2 sequence crossover study (N=50). And batches considered are as follows: Subject no. 1 to 25 as Batch-I AND Subject no. 26 to 50 as Batch-II. In that case you have batch as a fixed between-factor where the levels correspond to groups separated in time, and I would avoid Treatment*Batch (~"treament in batch") unless you have some sneaky plan with that interaction. — Pass or fail! ElMaestro |
|
arl_stat ★ India, 2013-03-26 06:35 (4828 d 18:49 ago) @ ElMaestro Posting: # 10280 Views: 7,187 |
|
|
Hi ElMaestro, I am planning to perform Anova including Treatment*Batch first. Further if there is no significant Treatment*Batch effect then to again perform Anova excluding Treatment*Batch effect. Correct me if I am wrong. Also it would be appreciable if can someone explain me these Anova Factors(i.e. Batch, Sequence, treatment, batch*treatment and Period(batch), and Subject(sequence*Batch) ) Thanks. |
|
ElMaestro ★★★ Denmark, 2013-03-26 10:04 (4828 d 15:21 ago) @ arl_stat Posting: # 10283 Views: 7,172 |
|
|
Hello arl_stat, ❝ I am planning to perform Anova including Treatment*Batch first. Further if there is no significant Treatment*Batch effect then to again perform Anova excluding Treatment*Batch effect. Correct me if I am wrong. Also it would be appreciable if can someone explain me these Anova Factors(i.e. Batch, Sequence, treatment, batch*treatment and Period(batch), and Subject(sequence*Batch) ) Your proposal is perhaps not completely in line with model reduction habits in the pharmaceutical industry. Some companies might use Akaike or Schwarz info criteria to select factors that go into the trash can. Anyways, no personal objection from me. Explanation of Anova factors...? I assume you mean the interactions or nestings? Let's take Subject(Sequence*batch); in this case imagine you have -subject 1,2,3,4 in batch 1 of sequence 1. -subject 1,2,3,4 in batch 2 of sequence 2. -subject 1,2,3,4 in batch 1 of sequence 2. -subject 1,2,3,4 in batch 2 of sequence 1. All these subjects are different subjects, so they are just coded in a dumb way. It des not make sense to ask what the fixed effect of subject 3 is if you don't specify which of the four subject 3's you think of. In such a situation it makes sense to apply Subject(batch * sequence). The unreduced model matrix will have 16 columns for the fixed treatment levels for that (4+4+4+4 as indicated above) - you will then loose a couple of them due to the df redundancy. How many depends on whether you have requested an intercept and which factors are preceding Subject (batch * sequence) in your specification. — Pass or fail! ElMaestro |
|
arl_stat ★ India, 2013-04-01 15:09 (4822 d 11:16 ago) @ ElMaestro Posting: # 10320 Views: 7,047 |
|
|
Hello ElMaestro, Thanks ElMaestro for your quick reply. ❝ Your proposal is perhaps not completely in line with model reduction habits in the pharmaceutical industry. I am also confused about Anova model selection. This is a crossover study carried out in two groups of subjects(25 subjects in each group) within a month (e.g., if for logistical reasons only a limited number of subjects can be studied at one time), the statistical model should be modified to reflect the multigroup nature of the study. In particular, the model should reflect the fact that the periods for the first group are different from the periods for the second group. I assume you mean the interactions or nestings? - batch *treatment : It is the interaction effect. - Period (batch): It is Period nested within Batch effect. - Subject (sequence*Batch): it is Subject nested Within (sequence*Batch)effect. - batch : It is Simply Batch effect. ❝ Let's take Subject(Sequence*batch); in this case imagine you have ❝ ❝ -subject 1,2,3,4 in batch 1 of sequence 1. ❝ -subject 1,2,3,4 in batch 2 of sequence 2. ❝ -subject 1,2,3,4 in batch 1 of sequence 2. ❝ -subject 1,2,3,4 in batch 2 of sequence 1. ❝ All these subjects are different subjects, so they are just coded in a dumb way. It des not make sense to ask what the fixed effect of subject 3 is if you don't specify which of the four subject 3's you think of. In such a situation it makes sense to apply Subject(batch * sequence). The unreduced model matrix will have 16 columns for the fixed treatment levels for that (4+4+4+4 as indicated above) - you will then loose a couple of them due to the df redundancy. How many depends on whether you have requested an intercept and which factors are preceding Subject (batch * sequence) in your specification. Sorry but, I did not get your explanation. |
|
ElMaestro ★★★ Denmark, 2013-04-01 15:27 (4822 d 10:57 ago) @ arl_stat Posting: # 10321 Views: 7,072 |
|
|
Hello arl stat, ❝ In particular, the model should reflect the fact that the periods for the first group are different from the periods for the second group. Yes, that is no issue as long as subjects are numbered uniquely. ❝ Sorry but, I did not get your explanation. OK, Subject(Sequence*batch) again... You have two sequences. You have two batches. You have 25 subjects in each batch and these are uniquely coded. 1 to 25 are the subjects in batch 1 and 26 to 50 or something like that are the subjects in batch 2. When you ask for a model with Subject(Sequence*batch) then you actually ask the engine to derive effect estimates for Subject 1 in sequence 1 in batch 1, Subject 1 in sequence 1 in batch 2, subject 2 in sequence 1 in batch 1 ... subject 50 in sequence 2 in batch 2. That is all in all 100 effects. Add to that the verious other effects of terms like period and treatment. You realise you now are asking for a number of effect estimates that exceed the number of observations, so is a silly mess. For example, subject 50 is in batch 2 so it doesnt make sense to ask for an effect of subject 50 in sequence 1 in batch 1 and so forth. The whole Subject(Sequence*batch) only makes sense if you code subjects nonuniquely. Then you may have subjects 1..12 in sequence 1 of batch 1, subject 1..13 in sequence 2 of batch 1, subjects 1..13 in sequence 1 of batch 2, and subject 1..12 in sequence 2 of batch 2. Suddenly it now makes sense to speak of Subject(Sequence*batch). But of course in real life noone will number subjects nonuniquely for at least two reasons.
Let me add, it is not leading to erronoeus results per se to use Subject(Sequence * batch) when uniquely coded subjects are used because the model matrices will be reduced to full rank internally by the lexer or engine before the fit is done. — Pass or fail! ElMaestro |
|
ElMaestro ★★★ Denmark, 2013-04-02 21:28 (4821 d 04:56 ago) @ ElMaestro Posting: # 10331 Views: 6,946 |
|
|
Apologies, I just realised that 50 x 2 x 2 happens to be slightly more than 100. I hope the rest of my post makes a wee bit more sense.  — Pass or fail! ElMaestro |
Ing. Helmut Schütz