Log in
Register|
pash413 ★ India, 2011-07-30 17:50 (5433 d 13:26 ago) Posting: # 7241 Views: 11,128 |
|
|
Dear All, We are conducting bioequivalence studies on patients, carried out at different clinical sites. As the biostudies are conducted on cancer patients, the dosing is planned [either single (one) or in group] at different clinical sites based on the availability of patients. Even at the same clinical site the dosing is done for different group which is separated by 1- 3 months. As per FDA guidance, if a crossover study is carried out in two or more groups of subjects (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 multi-group nature of the study. Hence we are planning to include ‘group effect' in our statistical model to evaluate whether the group effect is present or not. We have worked out the following approaches for choosing a ‘group’, which we are mentioning below along with the probable concern that may occur by doing so.
Edit: Category changed. [Helmut] |
|
ElMaestro ★★★ Denmark, 2011-07-30 19:20 (5433 d 11:56 ago) @ pash413 Posting: # 7243 Views: 8,487 |
|
|
Hi pash314, that's actually a very good post, I think. I never actually gave such issues much thought, and generally when I have been involved in group issues they have been either centers as fixed factors or timing as fixed ("early" vs "late", "week 30" vs "week 36" etc), coded by ones and zeros in the model matrix (or whatever the contrast coding is). I guess you are right, all three options will possibly apply to your situation, so you have a choice. I do not think one is prefereable to the others, but that will possibly depend. If you only dose one (or few) subject(s) per day, then I think it is overkill to use day as fixed. In that case I would bin them aiming at e.g. 6 per group. Depends on your total sample size also. I do not have any good reason to say 6 is way better than 5 or 7, to be honest. The more per bin, the fewer bins but the better than chance to detect differences between groups. Can you tell more about your sample size, recruitment timings, centers etc? Because this phenomenon is a between-subject factor and your trial is crossover, the end result (assessment of BE) should not be affected regardless of how you specify the grouping. The reason why FDA wishes applicants to extract such info because it may highlight when something is fishy. If one group's LSMean comes out a lot different from the others, a visit by the men in black can be anticipated. Save the forest, eat a beaver. — Pass or fail! ElMaestro |
|
pash413 ★ India, 2011-08-01 16:42 (5431 d 14:33 ago) @ ElMaestro Posting: # 7246 Views: 8,158 |
|
|
Dear ElMaestro The sample size for our biostudy was 48 and the recruitment of patients had been done over a period of 8 months in 5 centers. The centers were also located in different region (i.e. not in the same city/ state). Edit: Full quote removed. Please delete anything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut] |
|
ElMaestro ★★★ Denmark, 2011-08-01 19:12 (5431 d 12:04 ago) @ pash413 Posting: # 7247 Views: 8,225 |
|
|
Hello pash413, ❝ The sample size for our biostudy was 48 and the recruitment of patients had been done over a period of 8 months in 5 centers. The centers were also located in different region (i.e. not in the same city/ state). OK then I would perhaps suggest you to do a primary analysis with center as group factor (5 levels) in order to demonstrate that the T/R does not change significantly between centers. As secondary analysis one with month as fixed (8 levels) to prove that the T/R does not change through months. Never hit a man with glasses. Hit him with a baseball bat. — Pass or fail! ElMaestro |
|
pash413 ★ India, 2011-08-06 13:23 (5426 d 17:53 ago) @ ElMaestro Posting: # 7274 Views: 8,130 |
|
|
Dear ELMaestro, Thanks for your suggestion. The only concern we gave that if we are evaluating the group effect in both ways i.e.'Center' as group factor as primary analysis and 'month' as group factor as secondary analysis, if one of the evaluation is shown the existence of group effect and another doesn't, then what should be our conclusion. For example if group effect is shown for data between month and for 'center' no such effect is observed, then can we conclude that there is no group effect and combine the data of all 5 group (each for one center) for statistical analysis? If the situation is reverse i.e group effect is not shown secondary analysis(month) & shown for 'center', what would be our conclusion & action plan? Kindly advise. |
|
ElMaestro ★★★ Denmark, 2011-08-06 15:48 (5426 d 15:28 ago) @ pash413 Posting: # 7275 Views: 7,997 |
|
|
Dear pash413, in the absence of guidance covering your question, the following is my immediate reflection over your question. I am sure someone on this forum has a more qualified opinion than mine. Month significant as a factor: Many diseases vary in severity over the year. Seasonal variations in fed status and body weight/plasma volume. Drug products degrade with time. You may do an audit, but this is very difficult in practice, because multiple centers are involved. It means the auditor would need to go to each center and dig out the subjects from the outlying month (possibly compare with subjects from 'normal' months, too). Center significant as a factor: This is worse. Here I would dig into the data and compare group baselines of sex, body weight, biochemistry and what other info you have available across the centers. Depending on the how the data look, do an audit. If a center has been fiddling, taking it out is justified. To cover all bases, make sure such an unlikely event is per protocol. — Pass or fail! ElMaestro |
Ing. Helmut Schütz