Bioequivalence and Bioavailability Forum • Help with finding a reference for...

Wisboy
☆

2013-10-23 22:23
(4205 d 01:34 ago)

Posting: # 11743
Views: 7,228

Help with finding a reference for... [Power / Sample Size]

Hello,

I need help finding a reference that gives methodology and assumptions for estimating between and within subject variabilities from total variability for a BE study.

Here are my specific conditions:

I have a variance estimate from a prior study. This study was a parallel design.

I need to use this variability to generate a power calc (samp size) for a x-over study. So I need to estimate the within subject variability using the variability estimate I have from the prior study.

For the purposes of this discussion, I am going to assume the variability estimate I have is the total variability we would expect to see when testing this drug.

I will also assume: total variability = between subject variability + within subject variability

Based on my experience I have found the split to be, in general, something like this:

Between subject variability = 50% to 70% of total variability
Within subject variability = 30% to 50% of total variability

I need help finding a reference that supports these assumptions. I recall having a reference many years ago but it has disappeared over the years.

Over the past 15+ years, these assumptions have served me well and to my recollection, have not been significantly off when powering a definitive study. I do understand that within subject variability can be larger than the between subject variability but these are not typical results and are rare in my experience. And if this is observed and the numbers are correct, there are big problems with the compound being tested!

Ahy help with finding a reference is much appreciated!

Thanks in advance!
Jack

Helmut
★★★

Vienna, Austria,
2013-10-24 02:06
(4204 d 21:51 ago)

@ Wisboy
Posting: # 11745
Views: 6,597

Variances from “lower level” designs

Post reply

Hi Jack,

❝ I need help finding a reference that gives methodology and assumptions for estimating between and within subject variabilities from total variability for a BE study.

Stop searching. IMHO there is none. See this presentation (slides 18–20).

❝ For the purposes of this discussion, I am going to assume the variability estimate I have is the total variability we would expect to see when testing this drug.

So far so good.

❝ I will also assume: total variability = between subject variability + within subject variability

If you mean by variability the variances (and not the CVs), you are almost there.

❝ Between subject variability = 50% to 70% of total variability

❝ Within subject variability = 30% to 50% of total variability

OK, this is one of the often quoted ‘rules of thumb’ which have no statistical background. If it worked in your experience, you have been lucky.
See slide 20 for other examples and this thread as well.

A justification of any ‘rule’ would a require a sufficiently high correlation between intra- and inter-subject variability. What I have seen in my studies (checked ~150 until I got bored) is R² of 0.5…

❝ […] I recall having a reference many years ago but it has disappeared over the years.

Too bad. Would have been nice to dissect it.

❝ Over the past 15+ years, these assumptions have served me well and to my recollection, have not been significantly off when powering a definitive study.

Again: Lucky you!

❝ I do understand that within subject variability can be larger than the between subject variability but these are not typical results and are rare in my experience. And if this is observed and the numbers are correct, there are big problems with the compound being tested!

Not necessarily the drug, but it might be the formulation as well. No very uncommon for PPIs.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

Wisboy
☆

2013-10-24 20:06
(4204 d 03:50 ago)

@ Helmut
Posting: # 11755
Views: 6,469

Variances from “lower level” designs

Post reply

Thanks for your responses Helmut.

❝ If you mean by variability the variances (and not the CVs), you are almost there.

Yes - variances - I only use CVs to help explain the magnitude of the variability.

❝ OK, this is one of the often quoted ‘rules of thumb’ which have no statistical background. If it worked in your experience, you have been lucky.

Please explain what you mean by "lucky"? In your experience do you find this not to be true more times than not?

❝ A justification of any ‘rule’ would a require a sufficiently high correlation between intra- and inter-subject variability. What I have seen in my studies (checked ~150 until I got bored) is R² of 0.5…

I will have to take a look at this. R² = 0.5 is not very good. I have worked on well over 500 BE type studies so I have a lot of data.

❝ Too bad. Would have been nice to dissect it.

Still looking – if I find it, I will share it with you.

❝ Again: Lucky you!

Here again, please explain what you mean by "lucky"?

❝ Not necessarily the drug, but it might be the formulation as well. No very uncommon for PPIs.

OK - I guess I should have been more specific and said treatment regimen - to me this means the drug, formulation, whatever else goes along with it. IMO - if the within > between, you have no control over exposure in any given patient and this is bad. Thus, big problems and more than likely, a dead "treatment regimen".

thanks again!

Helmut
★★★

Vienna, Austria,
2013-10-25 17:35
(4203 d 06:22 ago)

@ Wisboy
Posting: # 11769
Views: 6,681

Prediction of with-subject variability

Post reply

Hi Wisboy,

❝ ❝ OK, this is one of the often quoted ‘rules of thumb’ which have no statistical background. If it worked in your experience, you have been lucky.

❝ Please explain what you mean by "lucky"? In your experience do you find this not to be true more times than not?

OK, it is quite natural that between-subject variability is larger than with-subject variability (how much larger?). That’s the main reason we opt for cross-over designs. By “lucky” I meant that the “rule of thumb” might be misleading in some cases. If you base your sample size on a fixed ratio between/within you might overpower your Xover study (you pass BE – but waste money) or fail (if the true ratio is lower than assumed).

❝ I will have to take a look at this. R² = 0.5 is not very good. I have worked on well over 500 BE type studies so I have a lot of data.

Hey, that would be interesting. Go for it!

Amazingly last year I reviewed a manuscript submitted to the International Journal of Clinical Pharmacology and Therapeutics. Based on the (univocal) reviewers’ recommendations the editor did not accept the paper for publication. The authors had the balls to submit it to another journal (bad practice) and succeeded…

Ramírez E, Abraira V, Guerra P, Borobia AM, Duque B, López JL, Mosquera B, Lubomirov R, Carcas AJ, and J Frías. A Preliminary Model to Avoid the Overestimation of Sample Size in Bioequivalence Studies. Drug Res. 2013;63(2):98-103. doi:10.1055/s-0032-1333296.

I leave it to you to find out why I consider it to be crap.

Actually I fail to see any underlying physiological processes which would support a dependency of within- and between-subject variability. Imagine the extremes:

Complete absorption, no metabolization
Concentration primarily is linked to the volume(s) of distribution, which itself might be linked to anthropometric variables like body weight, surface area, or sex. Clearance is not an issue in BE (might be linked to age and sex).
Incomplete absorption, first-pass and/or presystemic metabolization, genetic polymorphism
All of the above + both variable absorption/elimination.

Even if you succeed in deriving a model describing intra/inter across different drugs* I expect the confidence intervals of its parameters to be so wide that I seriously doubt its applicability in practice.

❝ ❝ Not necessarily the drug, but it might be the formulation as well. No very uncommon for PPIs.

❝

❝ OK - I guess I should have been more specific and said treatment regimen - to me this means the drug, formulation, whatever else goes along with it. IMO - if the within > between, you have no control over exposure in any given patient and this is bad. Thus, big problems and more than likely, a dead "treatment regimen".

Maybe. Or simply a lousy formulation.

Or even worse: Across different formulations within the same drug. For example diclofenac has a very low CV_intra if administered as a solution (<10%), and increases in the order effervescent tablet → oral dispersion → suppositories → IR → enteric coated → topical. I haven’t checked my data, but assume that CV_inter does not increase to the same extent.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

ElMaestro
★★★

Denmark,
2013-10-25 22:08
(4203 d 01:49 ago)

@ Helmut
Posting: # 11774
Views: 6,499

slightly off-topic

Post reply

Hi Hötzi,

❝ Amazingly last year I reviewed a manuscript submitted to the International Journal of Clinical Pharmacology and Therapeutics. Based on the reviewers’ recommendations the editor did not accept the paper for publication.

Funny. I think I know another person who was a reviewer for that ms. Sounds like there was agreement between the reviewer opinions. :-D

❝ The authors had the balls to submit it to another journal (bad practice) and succeeded…

Even funnier, I think that's a very common practice for people working in academia. Submit first to a journal with a high impact factor (yes I know, sigh & yawn!), then titrate your way down to the appropriate level/journal by trial and error until acceptance. I think my personal record is 6 journals :-D

- it was a paper dealing with codon usage in bacteria. Not indicating in any way that I like impact factors by mentioning this.

—
Pass or fail!
ElMaestro

Helmut
★★★

Vienna, Austria,
2013-10-26 16:51
(4202 d 07:05 ago)

@ ElMaestro
Posting: # 11776
Views: 6,473

completely off-topic

Post reply

Hi ElMaestro,

❝ Funny. I think I know another person who was a reviewer for that ms. Sounds like there was agreement between the reviewer opinions. :-D

I have heard that the editor faced serious problems finding victims willing to review this manuscript. Might well be that reviewer #1 suggested #2. :pirate:

❝ […] I think that's a very common practice for people working in academia. Submit first to a journal with a high impact factor […], then titrate your way down to the appropriate level/journal by trial and error until acceptance.

Publish or parish. Confirmed:
Int J Clin Pharmacol Ther (1.18) → Drug Res (0.599)…

❝ Not indicating in any way that I like impact factors […]

True. BTW, in our field the champion is Les Benet with a citation index of 16,000+. His paper¹ about volumes of distribution was cited 900+ times and is still the top cited one out of the more than 18,000 articles published in J Pharm Sci since 1965. Another one² was cited 600+ times.

Benet LZ, Galeazzi RL. Noncompartmental Determination of the Steady-State Volume of Distribution. J Pharm Sci. 1979;68(8):1071–4. doi:10.1002/jps.2600680845.
Rowland M, Benet LZ, Graham GG. Clearance concepts in pharmacokinetics. J Pharmacokinet Biopharm. 1973;1(2):123–36. doi:10.1007/BF01059626.

If you are interested in history:

Benet L. Benet LZ and Galeazzi RL: Noncompartmental Determination of the Steady-State Volume of Distribution, J Pharm Sci 68, 1071–1074, 1979—the Backstory. AAPS J. 2012;14(2):164–67. doi:10.1208/s12248-012-9326-9, [image] free resource.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

ElMaestro
★★★

Denmark,
2013-10-26 20:22
(4202 d 03:35 ago)

@ Helmut
Posting: # 11778
Views: 6,383

completely off-topic

Post reply

Hi Hötzi,

❝ Publish or parish. Confirmed: Int J Clin Pharmacol Ther (1.18) → Drug Res (0.599)…

I recently made it in reverse. Rejected in a journal with IF 2.2 and accepted with minor modifs in a journal with IF 4.3. All in crackpot-style. You may soon see it published in the Scandinavian Journal of Obfuscated, Obsolete and Manipulated Science.

❝ If you are interested in history:

❝ Benet L. Benet LZ and Galeazzi RL: Noncompartmental Determination of the Steady-State Volume of Distribution, J Pharm Sci 68, 1071–1074, 1979—the Backstory. AAPS J. 2012;14(2):164–67. doi:10.1208/s12248-012-9326-9, [image] free resource.

Hey, thanks for that. It's great reading. :cool:

—
Pass or fail!
ElMaestro

Helmut
★★★

Vienna, Austria,
2013-10-27 15:58
(4201 d 06:59 ago)

@ ElMaestro
Posting: # 11781
Views: 6,350

completely off-topic

Post reply

Hi ElMaestro,

❝ […] You may soon see it published in the Scandinavian Journal of Obfuscated, Obsolete and Manipulated Science.

Sounds interesting! Is this journal a spin-off from the “Annals of Improbable Research”?

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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

Helmut
★★★

Vienna, Austria,
2013-10-31 02:12
(4197 d 20:45 ago)

@ Wisboy
Posting: # 11828
Views: 6,311

Variances from “lower level” designs

Post reply

Hi Jack,

found an old one (I have just the abstract of a presentation held at the AAPS Anual Meeting 2000):

Masson É, Roy G, Lapointe C, Abolfathi Z. A Regression Model to Predict Intrasubject Variability for Bioequivalence Studies.

The correlation (even for the covariate model) looks familiar. ;-)

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

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