Bioequivalence and Bioavailability Forum 07:05 CET

Main page Policy/Terms of Use Abbreviations Latest Posts

 Log in |  Register |  Search

Babe_Ruth
Junior

USA,
2019-02-08 18:32

Posting: # 19889
Views: 287
 

 ANOVA fixed and random effects [General Sta­tis­tics]

What's the appropriate ANOVA model for a two treatment, two period, fixed sequence study?

It's a food effect study where the subject takes the study drug under fasted conditions, 14-day washout, then takes study drug under fed conditions. The proposed is ANOVA model is to analyze log-transformed Cmax and AUClast with terms for treatment condition (Fasted and Fed) as a fixed effect and subject as a random effect.

Is this appropriate, and why or why not?

Compared to the standard 2x2 crossover study, what are the limitations of the above study design in terms of assessing inter- and intra-subject variability?
ElMaestro
Hero

Denmark,
2019-02-08 19:00

@ Babe_Ruth
Posting: # 19890
Views: 268
 

 ANOVA fixed and random effects

Hello to you too, Babe_Ruth :-D


» It's a food effect study where the subject takes the study drug under fasted conditions, 14-day washout, then takes study drug under fed conditions. The proposed is ANOVA model is to analyze log-transformed Cmax and AUClast with terms for treatment condition (Fasted and Fed) as a fixed effect and subject as a random effect.

» Is this appropriate, and why or why not?

That might do, but why not crossover? You'd gain information or at least have an opportunity to impress an agency with design complications :-) and you'd potentially have less deficiency letters.

» Compared to the standard 2x2 crossover study, what are the limitations of the above study design in terms of assessing inter- and intra-subject variability?

If you do it your way, then the most straightforward evaluation is a (paired) t-test approach or equivalently a linear model with two fixed effect (Condition, subject). It will give the exact same result.

There are situations when you'd be having an advantage of the true crossover. For example, imagine there were a tolerability issue, such as metformin IR in fasted naïve subjects who get gastric upsets. You'd potentially see it already in period 1 in a crossover because half the subjects will be fasted then. And so forth.

In both cases you can argue that the model residual (or pooled variance, in case of t-test) is a measure of intra-subject variability. In both cases you will be able to extract also a quantity for total variability observed between subjects. For Feinschmeckers, the latter may not always be rightfully called the between-subject variability since it has components of both within and between whichever way you measure it. But this is also a bit about semantics and personal preferences and habits.
The crossover gives you also info about period and sequence effects which for all practical purposes might (or might not) be of interest to the person assessing it. A between-period time of 14days should suffice in most cases, no doubt about it, at least in terms of he food phenomena, and hopefully also for the elimination of the drug.

if (3) 4

x=c("Foo", "Bar")
b=data.frame(x)
typeof(b[,1]) ##aha, integer?
b[,1]+1 ##then let me add 1



Best regards,
ElMaestro

"(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures." New York Times (ed.), June 9, 2018.
Babe_Ruth
Junior

USA,
2019-02-08 19:53

@ ElMaestro
Posting: # 19891
Views: 262
 

 ANOVA fixed and random effects

I appreciate the knowledge and perspectives, ElMaestro! I was thinking of writing a "Hi" in my initial post, but thought it would detract from the central questions, haha.

» That might do, but why not crossover? You'd gain information or at least have an opportunity to impress an agency with design complications :-) and you'd potentially have less deficiency letters.

The sponsor specified this fixed-sequence design in their protocol.A little more context regarding this study: Phase I, healthy subjects, with SAD (n=30), food effect (n=12), and MAD (n=24) parts. I don't know why they designed it to have all subjects take the drug under fasted and then fed conditions. I can't think of any advantage fixed sequence over 2 sequence in this study. Have you ever seen instances where one sequence is advantageous over two sequence?

» If you do it your way, then the most straightforward evaluation is a (paired) t-test approach or equivalently a linear model with two fixed effect (Condition, subject). It will give the exact same result.

Interesting, I'll look into this. Lots of terms and names I don't know in your last few paragraphs :D

Typically, I focus on specific PK sections of the study, but I'm working on learning the study design and stats aspects, so I can pull a bigger picture together.


Edit: Also, one last question. Somebody proposed the ANOVA model with terms for treatment condition (Fasted and Fed) as fixed effects, and subject nested within treatment condition (fasted and fed) as a random effect. Does this random effect make sense?
ElMaestro
Hero

Denmark,
2019-02-08 20:24

@ Babe_Ruth
Posting: # 19892
Views: 289
 

 ANOVA fixed and random effects

Hi Babe_Ruth,

» The sponsor specified this fixed-sequence design in their protocol.A little more context regarding this study: Phase I, healthy subjects, with SAD (n=30), food effect (n=12), and MAD (n=24) parts.

Yikes. Do you mean single ascending doses, mutiple ascending doses? That would likely make this a first in man, potentially?? I am not sure I would like to offer opinions on those. Is the API very well known??

» I don't know why they designed it to have all subjects take the drug under fasted and then fed conditions. I can't think of any advantage fixed sequence over 2 sequence in this study. Have you ever seen instances where one sequence is advantageous over two sequence?

I don't recall that. I am not so often involved in such studies. But simple is beautiful. Until a regulator thinks otherwise.

» Edit: Also, one last question. Somebody proposed the ANOVA model with terms for treatment condition (Fasted and Fed) as fixed effects, and subject nested within treatment condition (fasted and fed) as a random effect. Does this random effect make sense?

That is correct, in principle. Subjects are actually directly treated as random when you do the t-test thing, while when you model it, and you get the exact same result, then the subject is modeled as fixed effect with PROC GLM (or equivalent), even when a random statement is used.
"Subject in sequence" is a very popular term on this forum. Usually we number subjects uniquely, i.e. 1, 2, 3....N and then you don't need to worry about it when evaluating the model. A given subject is the given subject, with no opportunity for confusion (apart from "Hey subject 7 looks like my uncle" and that sorta thing).
However, you might decide that you want to violate GCP § 1.58 (for reasons entirely beyond my comprehension) and decide that you want to call subjects 1,2, 3 and so forth in both sequences. I.e. there is a "subject 1" in the sequence group where they are fasted before being fed, and also a "subject 1" in the sequence group where they are fed before being fasted. Now, to make sure the stats software can distinguish between those (after all they are not identical subjects [or at least I sure hope for everyone's sake they aren't]), you apply syntax to the effect of making sure the stats is done correctly so that there is no confusion about the uniqueness of both "subject 1" 's and so on. That syntax and model fitting is what "subject in sequence" is all about, so it only has relevance if you don't number subjects uniquely.
However, it apparently does sound rather impressive when a guy in a penguin suit in the board room raises a finger and asks if subjects are nested in sequence. It may be the kind of tricks that make you popular with the opposite gender and the CSO. I tend to bang my head into the table when it happens. Am taking anger management classes and large doses of Schützomycin because of those types (it's good for my gonorrhoea as well).

Oh, it seems I digressed a wee bit. Sorry. Let me just go and fetch my medication now.

if (3) 4

x=c("Foo", "Bar")
b=data.frame(x)
typeof(b[,1]) ##aha, integer?
b[,1]+1 ##then let me add 1



Best regards,
ElMaestro

"(...) targeted cancer therapies will benefit fewer than 2 percent of the cancer patients they’re aimed at. That reality is often lost on consumers, who are being fed a steady diet of winning anecdotes about miracle cures." New York Times (ed.), June 9, 2018.
Babe_Ruth
Junior

USA,
2019-02-08 20:46

@ ElMaestro
Posting: # 19893
Views: 243
 

 ANOVA fixed and random effects

Hi ElMaestro

» Yikes. Do you mean single ascending doses, mutiple ascending doses? That would likely make this a first in man, potentially?? I am not sure I would like to offer opinions on those. Is the API very well known??

Yes, that's what I meant with SAD and MAD. Yes, first in human. API is not known, but the its target is pretty hot these days. They call her Jaki.

» "Subject in sequence" ... only has relevance if you don't number subjects uniquely.

:surprised: whoa, mind blown. Applied to my question above, "Subject as a random effect" is the same as "Subject nested in treatment condition (Fed/Fasted)"? Given that subject numbers are all unique.

» I tend to bang my head into the table when it happens.

This explains both your intelligence and your insanity =D
Ohlbe
Hero

France,
2019-02-08 22:58

@ Babe_Ruth
Posting: # 19894
Views: 240
 

 Food effect in FIM

Dear Babe_Ruth and ElMaestro,

» » Yikes. Do you mean single ascending doses, mutiple ascending doses? That would likely make this a first in man, potentially?? I am not sure I would like to offer opinions on those. Is the API very well known??
»
» Yes, that's what I meant with SAD and MAD. Yes, first in human.

OK, this explains your study design. See EMA's FIM guideline, section 8.2.2 page 15:
Other single dose parts (e.g. food interaction) could be conducted in parallel to the SAD part provided the dose chosen and the expected exposure are equal to or lower than that which was reached in a concluded preceding SAD cohort where all relevant data has been reviewed and no dose escalation stopping criteria were met.


BTW Helmut: the guidance page still points to the old 2007 version of the guideline, not the 2018 version.


Edit: THX, added. [Helmut]

Regards
Ohlbe
Activity
 Thread view
Bioequivalence and Bioavailability Forum |  Admin contact
19,187 posts in 4,084 threads, 1,308 registered users;
online 7 (1 registered, 6 guests [including 6 identified bots]).

In these days, a man who says a thing cannot be done
is quite apt to be interrupted by some idiot doing it.    Elbert Green Hubbard

The BIOEQUIVALENCE / BIOAVAILABILITY FORUM is hosted by
BEBAC Ing. Helmut Schütz
HTML5 RSS Feed