BE-proff
●    

2015-11-05 19:31
(3065 d 21:59 ago)

Posting: # 15615
Views: 8,651
 

 Factors influence in Anova [General Sta­tis­tics]

Hi All,

Unless I am mistaken ANOVA assesses influence of different factors (subjects, treatment, period, sequence) on variation of PK-parameters.

In my practice only subjects have always had significant impact on variation.

What can be reasons for signficant influence of treatment, period and sequence?;-)

Is it bad or not?:confused:
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-11-05 20:33
(3065 d 20:57 ago)

@ BE-proff
Posting: # 15617
Views: 7,818
 

 Factors influence in Anova

Hi BE-proff,

❝ […] ANOVA assesses influence of different factors (subjects, treatment, period, sequence) on variation of PK-parameters.


OK; you are talking about a crossover.

❝ In my practice only subjects have always had significant impact on variation.


Anything else would be a big surprise. Subjects are different indeed.

❝ What can be reasons for signficant influence of treatment, period and sequence? ;-)


Treatment
Happens if the the CV is smaller + the T/R is more far away from 100% than assumed in study planning; the con­fi­dence interval does not include 100%. Reasons: The formulation and/or the study planning (over­powered?)
  • If the CI is still within the acceptance range, the difference is statistically significant, but not clinically relevant. No worries. You only burned money.
  • If the CI overlaps the AR, think twice about repeating the study in a larger sample size. IMHO, re­for­mu­lation is the way to go.
  • If the CI lies completely outside the AR, bioinequivalence is proven. End of story.
Period
Doesn’t matter at all. Since both R and T are affected to the same degree, true differences will mean out. Try it: Take one of your datasets and multiply all results of the second period by 1,000. The PE and CI of T/R will be exactly the same like in the original dataset. Of course the period effect comes out highly sig­ni­fi­cant. Reasons: Generally unknown (subjects more relaxed in the 2nd period, lunar phase, thunderstorm, broken AC, whatsoever).

Sequence
Better called unequal carryover. It essentially means that subject’s PK response might be different if they receive formulations in the order RT as compared to the order TR. Tricky. If there would be a true sequence effect we cannot get an unbiased estimate of the treatment effect (technically these effects are “con­founded”). That’s bad.
  • In the past (based on Grizzle 1965) the common approach was to discard all data of the second period and evaluate the data of the first period as a parallel design. Catastrophic loss in power.
  • Freeman showed in 1989 that this approach is statistically flawed. Stephen Senn devoted this issue a major part of one of his books. There is no statistical method which can correct for unequal carry­over. It can be only avoided by design (i.e., a sufficiently long washout).
  • In a large meta-study significant sequence effect showed up at approx. the level of the test (i.e., if you test with an α of 0.1 in 1/10 studies).
For references see this post. The EMA learned the lesson:

A test for carry-over is not considered relevant and no decisions regarding the ana­ly­sis (e.g. analysis of the first period only) should be made on the basis of such a test. The potential for carry-over can be directly addressed by examination of the pre-treatment plasma concentrations in period 2 (and beyond if applicable).


❝ Is it bad or not? :confused:


Define bad.

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
felipeberlinski
☆    

Brazil,
2015-11-05 20:56
(3065 d 20:34 ago)

@ Helmut
Posting: # 15618
Views: 7,472
 

 Factors influence in Anova

Hi Helmut and others

In case of performing a study with an endogenous substances such as hormones what would be the approach to justify an ocurrance of a sequence effect for HA? Premisse: You have an adequate washout period, and you have found this effect....:crying:

Some stats that I have contact suggested a bootstrap analysis to check if it was a random or a real effect...

What else could be done?
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-11-05 21:27
(3065 d 20:04 ago)

@ felipeberlinski
Posting: # 15619
Views: 7,538
 

 Factors influence in Anova

Hi Felipe,

❝ In case of performing a study with an endogenous substances such as hormones what would be the approach to justify an ocurrance of a sequence effect for HA? Premisse: You have an adequate washout period, and you have found this effect....:crying:


Endogenous compounds are of a nasty kind. I can only suggest to keep in mind that there might be a feedback loop. It’s not only laking concentrations in the higher periods which counts. Keep the wash­out as long a possible. If you followed all that (as you did), maybe the Type I Error hit (you got a significant effect, which is not there). You can try to claim that. Not sure whether regulators(s) will buy it.

If you want to be sure, I’m afraid there is no way around a parallel design. In all (!) crossover studies of biosimilars I’m aware of there was a significant sequence effect. Something happened to the body. Given that it is beyond me why biosimilar-guidelines recommend a crossover design as the “gold standard”.

❝ Some stats that I have contact suggested a bootstrap analysis to check if it was a random or a real effect...


Sounds like black magic to me. But I’m not a statistician. :-D

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,
2015-11-05 22:28
(3065 d 19:03 ago)

@ felipeberlinski
Posting: # 15621
Views: 7,436
 

 What else?

Hi f.b.,

❝ In case of performing a study with an endogenous substances such as hormones what would be the approach to justify an ocurrance of a sequence effect for HA? Premisse: You have an adequate washout period, and you have found this effect....:crying:

(...)

❝ What else could be done?


If P(Seq) is low then it means your metric objectively differed between the two sequences. Perhaps the randomisation "was not effective"; check your descriptive statistics. If all the fat ones ended up in TR and all the Jane Fonda types were in RT then that would be one obvious reason, but you cannot prove it as a hindsight consideration. It would however, be a likely explanation. Can happen by chance.

Pass or fail!
ElMaestro
BE-proff
●    

2015-11-05 22:27
(3065 d 19:03 ago)

@ Helmut
Posting: # 15620
Views: 7,396
 

 Factors influence in Anova

Dear Helmut,

Your explanation is really cool! :ok:

But one more question:

overpowered study...what do you mean? Power over 100%? :confused:
Never heard about it :-|
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2015-11-06 05:03
(3065 d 12:28 ago)

@ BE-proff
Posting: # 15622
Views: 7,467
 

 overpowering

Hi BE-proff

❝ overpowered study...what do you mean?


Generally we plan studies for 80–90% power π (where the producer’s risk of failing to demonstrate BE of a formulation which is BE: Type II Error, β = 1 – π). If you submit a protocol to the IEC which >90%, they might not accept it. Hence, “overpowered”. Sometimes rich sponsor know that the CV will be lower and/or the T/R is closer to 100% than what they present in the sample size estimation. Pro­tocol approved, study done, low risk of failure. They call it “to be on the safe side”. I call it playing dice with the health of subjects.

Sometimes the minimum sample size given in guidelines will lead to high power anyway. Example – T/R 0.95, 2×2 cross­over, AR 80–125%; CVs which will lead to >90% power:

 n   CV%
12  ≤13.3 
most regulations, Russian “Red Book”
18  ≤16.6  2008 Russian GL
20  ≤17.9  South Africa MR
24  ≤19.8  Brazil, Saudia Arabia (unless justified otherwise)

The chance of a significant treatment effect increases with power.
Imagine: n 134, T/R 0.97, CV 15%. The 90% CI will be 94.12–99.97%. 100% not included, significant treatment effect (p 0.0485), bingo. Is it relevant? Not at all. You can be >99.9999999999999% sure that the true T/R-ratio is not below 80%…

❝ Power over 100%? :confused:


Negative producer’s risk‽

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
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,638 registered users;
98 visitors (0 registered, 98 guests [including 7 identified bots]).
Forum time: 17:31 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