Oiinkie
☆    

The Netherlands,
2013-07-22 18:03
(3902 d 04:35 ago)

Posting: # 11032
Views: 10,173
 

 Two-stage three-treatment BE study? [Two-Stage / GS Designs]

Dear members,

After having performed numerous dissolution tests, a pilot PK study (failed, with different test formulations than currently developed) and a study in a gastrointestinal model, we have arrived at two candidate test formulations for a certain MR matrix tablet (generic of a reference tablet, not a one-to-one copy). As we are still rather uncertain which test formulation to take to a pivotal BE study (but wanting to avoid a new pilot PK study as we are ready for process validation and a pivotal study, and do not want to "waste" time/money), we would like to perform a two-stage pivotal BE study with three treatments (T1 vs T2 vs R; Williams' design).

The idea is to use the first stage as an internal pilot in order to select the best matching test formulation and estimate a total sample size (partly) based on the results (CV) of the first stage. The "bad" test formulation would be dropped from the study after the interim thereby making the second stage a two treatment study. This approach would not only save us time/money, but also allows us to make an educated decision on the test formulation. Moreover, dropping a "bad" test formulation and estimating a sample size based on results of the "best" test formulation, thereby exposing a limited amount of healthy volunteers to an inferior product and including only the amount of subjects required to demonstrate BE, may be regarded as ethical...

However, in my opinion, such a design sounds pretty dodgy, more from a statistical than from an ethical perspective... Is such a design acceptable from a regulatory (EU) point of view? What are the statistical implications (e.g. papers of Potvin et al.: spending of alpha, sample size estimation)? Is a two-stage study with three treatments allowed (not to mention dropping one treatment after the interim)?

Or are we back at square one having to choose between:
  • performing a pilot study with current test formulations to select the best test formulation (and subsequently performing a standard 2x2x2 single-stage pivotal study on the best test with a sample size based on the results of the pilot [conservative planning, CV not carved in stone ;-)]), and
  • performing a two-stage two-treatment pivotal study at risk with the test product we "think" performs best in comparison to the reference product?
Many thanks in advance for your ideas, brainwaves, comments...

Regards,

Oiinkie
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2013-07-22 18:33
(3902 d 04:05 ago)

@ Oiinkie
Posting: # 11033
Views: 8,971
 

 Bingo!

Hi Oiinkie,

welcome to the party! This is one of the questions I hear most often after one of my presentations on two-stage designs. First: Nothing is published so far. Stop searching. The GL tells us:

[…] appropriate steps must be taken to preserve the overall type I error of the experiment and the stopping criteria should be clearly defined prior to the study.
[…] the choice of how much alpha to spend at the interim analysis is at the company’s discretion. The plan to use a two-stage approach must be pre-specified in the protocol along with the adjusted significance levels to be used for each of the analyses.


Essentially you would have to set up a framework similar to the published papers (Potvin et al. 2008, Montague et al. 2011, Fuglsang 2013; forget Karalis & Macheras 2013) and demonstrate by simulations (sorry!) that the overall α-level is maintained. Points to consider:
  • In the 3-treatment part with the usual model you have a common variance – which might be larger/smaller then the one in the second part.
  • An option would be to use EMA’s approach of “leaving one out”; you would get only the pooled variance of the remaining pair.
  • You should introduce at least a statistical futility rule based on the ratio. If T1/R and T2/R are similar (whatever that means), consider dropping the one with higher variability.
Just some desultory thoughts.

P.S.: Expect a couple of weeks running the sim’s 24/7 on a single machine. :-(

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
Oiinkie
☆    

The Netherlands,
2013-07-29 12:41
(3895 d 09:57 ago)

@ Helmut
Posting: # 11097
Views: 8,638
 

 Bingo!

Dear Helmut,

Many thanks for sharing your thoughts.

❝ P.S.: Expect a couple of weeks running the sim’s 24/7 on a single machine. :-(


Unfortunately, I do not have the resources or knowledge to do this at the moment. Would be a nice project in the future though...

To avoid the "too progressive" design of a two-stage three-treatment study, we will go for a three-treatment pilot and base our decision for the test formulation on this study.

Regards,

Oiinkie
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2013-07-29 13:32
(3895 d 09:05 ago)

@ Oiinkie
Posting: # 11099
Views: 8,699
 

 Bingo!

Hi Oiinkie,

❝ To avoid the "too progressive" design of a two-stage three-treatment study, we will go for a three-treatment pilot and base our decision for the test formulation on this study.


Good idea. Given the strange responses I have seen in the recent past from European regulators even on straightforward two-stage designs I would not recommend such an approach – unless a framework is published. Nevertheless, a tempting idea which deserves some attention, IMHO.
(:waving: ElMaestro?)

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes
ElMaestro
★★★

Denmark,
2013-07-29 14:06
(3895 d 08:32 ago)

@ Helmut
Posting: # 11100
Views: 8,761
 

 Bingo!

Hi Hötzi,

❝ Given the strange responses I have seen in the recent past from European regulators even on straightforward two-stage designs I would not recommend such an approach – unless a framework is published. Nevertheless, a tempting idea which deserves some attention, IMHO.

❝ (:waving: ElMaestro?)


Si Señor... I'll be happy to write some software which does the EMA thing. But I am short of time at the moment. I am more of a sailor than a simulant these days. Longish WHO trip to the far east coming up as well as a few conferences and blah blah blah.

Simplest case would be to take Potvin's meffuds B and C and to just implement a kinda "pick one" decision at stage 1. The criteria for picking the right one after stage 1 might be based on the PE. But of course it would be subject to debate if apparent variability should also be somehow considered. We could -and this is just a stray thought- do a power calc after stage 1 using the observed PE and then define the winner as the one that is associated with the highest power at this stage. We could of course also just apply the expected PE and it would boil down to lowest Variance wins. Not sure, haven't thought it through at all. And we will need to consider two type I errors, of course.
Oh wait, we also need to consider "chance of one product passing BE criteria", so I guess we'll have three T1Es. Darn... this quickly gets complicated.


Lemme hear some creative thoughts or alternatives. I don't think this is very difficult but it will take a bloody lot of time.

Pass or fail!
ElMaestro
ElMaestro
★★★

Denmark,
2013-07-29 18:50
(3895 d 03:48 ago)

@ Helmut
Posting: # 11115
Views: 8,690
 

 Terms in Anova

Hi all,

couldn't help but had to start thinking about the implementation of this.

Now directly for the tricky stuff:

First:
We start out with 3 treatments at stage 1. This means that at stage 1 we have 6 different sequences.
After stage 1 we take a formulation and put it into the trash can. Stage 2 therefore has 2 treatments and thus also 2 sequences. This implies that we have to speak of sequence in stage rather than just sequence.
Does anyone agree?

Second:
The following terms should be evaluated:
  • Subject (no I am not nesting them in something because I code them uniquely)
  • Sequence in Stage (with nesting; we could also just give them factor levels like 1,2,3,4,5,6,7,8 and observe that 1..6 apply to stage 1 and 7..8 apply in stage 2.
  • Period in stage (with nesting; 1..3 at stage 1 and 1..2 in stage 2. We could also nominate them 1..5)
  • Treatment
  • Stage
And Sequence in its own right would be a completely irrelevant term - EMAs Q&A is not applicable here.

Does anyone agree?

Third:
Here's a dataset
- the owner of Fuglsang Pharma (a miserable person of generally dubious character) kindly allowed me to upload it there :yes:
You can save it somewhere and open it in R like this:

A=read.table("testout.txt", header=T) # reads the data

B=subset(A, Trt!=2) ## we now exclude treatment 2; we only wish to look at treatment 1 vs 3

M1=lm(B$lnPK~0+
    as.factor(B$Trt)
   +as.factor(B$Seq):as.factor(B$Stg)
   +as.factor(B$Subj)
   +as.factor(B$Per):as.factor(B$Stg)
   +as.factor(B$Stg))

anova(M1)
# Quite nice; we just need to deal with three terms,
# one of them nested before
# everything else is uniquely determined. 


Anyone agrees?
(I guess I will receive a beating for this)

Pass or fail!
ElMaestro
ElMaestro
★★★

Denmark,
2013-07-30 00:37
(3894 d 22:01 ago)

@ ElMaestro
Posting: # 11120
Views: 8,627
 

 annoying imbalance

Hi all,

❝ – Sequence in Stage (with nesting; we could also just give them factor levels like 1,2,3,4,5,6,7,8 and observe that 1..6 apply to stage 1 and 7..8 apply in stage 2.


Dammit.:crying: I just tried to work out the equations for the various sums of squares. Imbalance on the sequence part (6 in stage 1, 2 in stage 2) implies that I can probably not derive a residual ss by means of a simple formula. Then the only option (at least for me) will be the matrix way forward and that will slow the whole thing considerably down. Might as well just set this up in R then?!?

Does anyone feel like wasting some days and nights on deriving a formulae for residual SS now that I admit algebraic defeat?

Pass or fail!
ElMaestro
jatkins_5
☆    

Philippines,
2014-10-30 04:53
(3437 d 16:44 ago)

@ Helmut
Posting: # 13814
Views: 7,690
 

 Bingo!

An option would be to use EMA’s approach of “leaving one out”; you would get only the pooled variance of the remaining pair.


Good day sirs!

In the above quote, is it statistically valid to establish ANOVA using drugs A and B only (as if done like a 2-way 2-seq crossover) in stage 1 (Potvin Method C) even though the study was done on a 3-way 6-seq crossover (Williams design) with drugs A, B and C? (asssuming the protocol specifically states that sequential BE shall be done on drugs A and B only; drug C shall be done for descriptive statistics determination only). Also, may I ask what specific EMA guideline mentioned the above approach so I could explore the topic further. I know a lot of discussion has already been poured on sequential BE, but I couldn't find a specific answer to my query. Many thanks in advance!

jerry
ElMaestro
★★★

Denmark,
2014-10-30 09:59
(3437 d 11:39 ago)

@ jatkins_5
Posting: # 13815
Views: 7,739
 

 Yes, however...

Hi Jerry,

❝ In the above quote, is it statistically valid to establish ANOVA using drugs A and B only (as if done like a 2-way 2-seq crossover) in stage 1 (Potvin Method C) even though the study was done on a 3-way 6-seq crossover (Williams design) with drugs A, B and C? (asssuming the protocol specifically states that sequential BE shall be done on drugs A and B only; drug C shall be done for descriptive statistics determination only). Also, may I ask what specific EMA guideline mentioned the above approach so I could explore the topic further. I know a lot of discussion has already been poured on sequential BE, but I couldn't find a specific answer to my query. Many thanks in advance!


Yes, I would assume you can actually do that across some of the EU agencies.
But how could you ever be in a situation where you'd wish to study a third product descriptively in a form of BE trial? Not sure I get everything here, and if you are asked how this approach influences the type I error do you have a good answer prepared?

Pass or fail!
ElMaestro
jatkins_5
☆    

Philippines,
2014-11-01 00:20
(3435 d 21:18 ago)

@ ElMaestro
Posting: # 13816
Views: 7,436
 

 Yes, however...

Hi ElMaestro,

❝ Yes, I would assume you can actually do that across some of the EU agencies.


Thanks, thats good to know..

❝ But how could you ever be in a situation where you'd wish to study a third product descriptively in a form of BE trial?


We'd actually want to compare C's pk data with A, in a non statistical way, for future marketing purposes.


Not sure I get everything here, and if you are asked how this approach influences the type I error do you have a good answer prepared?

I think it's time we review our stats for this. I haven't got a prepared answer, really. Perhaps a simulation might work?

Thank you very much for your valuable reply!

Jerry
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
76 visitors (0 registered, 76 guests [including 7 identified bots]).
Forum time: 21:38 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5