Bioequivalence and Bioavailability Forum • Replicate designs

Helmut
★★★

Vienna, Austria,
2013-11-22 17:11
(4173 d 23:14 ago)

Posting: # 11945
Views: 7,100

Replicate designs [🇷 for BE/BA]

Dear userRs!

Yesterday I received an e-mail from the FDA’s CDER/OGD:
<snip>

█ ████ ███████ ████ ██ █████████████ ██ ████████████████████████████ ███ █████ ████ █████████ ███████████. █ ██████ ██████████ ████ ███ ████ ████ █████ █████████████ █████████ ██ ███ ██████. ████ ████████████ ████████ ██ ███ ███ █ █████ ███ ████ ████████ ██ ████ ██ ████ █████████████ ███ ██ ████████████ █ ██████ ██ ███ ███ █████████. █ ████ ███ █████████ ██ ███████ ██ ████ ███ ██ ███ ███ █████ ████ ██ ██ ███ ████ ███████ █████████ ███ █████.

</snip>

Well, I don’t. Given the numerous discussions we had about lme(), nlme:lme(), and lme4:lmer() it seems that it not possible to reproduce FDA’s SAS-code in R (df limbo). Something new from any of you?

PS: Since Donald Schuirmann performed his simulations for NTID-scaling (RTRT|TRTR) in R, FDA should have some code, right?
PPS: Maybe the e-mail was triggered by this or that?

Edit 2013-12-04: Quoted e-mail blacked out as requested by the FDA.

—
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,
2013-11-23 16:10
(4173 d 00:15 ago)

@ Helmut
Posting: # 11947
Views: 5,950

Replicate designs

Post reply

Hi Hötzi,

I have very little to add here. There is still no option to do the equivalent of DDFM=Satt in R for any of the available packages, as far as I know.
The very few discussions I have seen about this issues tend not to discuss what the Satt denominator actually does or what the alternatives are and how they perform. I'd love to understand it. The SAS documentation is not so helpful.

❝ PS: Since Donald Schuirmann performed his simulations for NTID-scaling (RTRT|TRTR) in R, FDA should have some code, right?

Yes, but it might be much, much easier to email you and ask than going through the internal bureaucratic processes at FDA. :-D

—
Pass or fail!
ElMaestro

Helmut
★★★

Vienna, Austria,
2013-11-23 16:32
(4172 d 23:54 ago)

@ ElMaestro
Posting: # 11948
Views: 5,989

df limbo

Post reply

Hi ElMaestro,

❝ There is still no option to do the equivalent of DDFM=Satt in R for any of the available packages, as far as I know.

Yep. Some background here, especially:

Because the primary authors of lme4 are not convinced of the utility of the general approach of testing with reference to an approximate null distribution, and because of the overhead of anyone else digging into the code to enable the relevant functionality (as a patch or an add-on), this situation is unlikely to change in the future.

No Satterthwaite approximation; at least the Kenward & Roger approximation^{1 (why not 2?)} is available in the package pbkrtest (reference). But this is not what FDA wants…

Kenward MG, Roger JH. Small Sample Inference for Fixed Effects from Restricted Maximum Likelihood. Biometrics. 1997;53(3):983–97.
Kenward MG, Roger JH. An Improved Approximation to the Precision of Fixed Effects from Restricted Maximum Likelihood. Comp Stat Data Anal. 2009;53(7):2583–95.

—
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

yjlee168
★★★

Kaohsiung, Taiwan,
2013-11-23 20:58
(4172 d 19:27 ago)

@ Helmut
Posting: # 11949
Views: 5,911

df limbo

Post reply

Dear Elmaestro and Helmut,

❝ ❝ There is still no option to do the equivalent of DDFM=Satt in R for any of the available packages, as far as I know.

How about the package 'lmerTest'? I just ran the example, and got

...

anova(m)

Analysis of Variance Table of type 3 with  Satterthwaite  approximation for degrees of freedom

            Df Sum Sq Mean Sq F value  Denom    Pr(>F)    

Gender       1  3.844  3.8437  0.8790  78.98 0.3513471    

Information  1  6.520  6.5201  1.4909 562.97 0.2225805    

Product      3 91.807 30.6024  6.9979 562.97 0.0001257 ***

---

Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Don't know if it is what FDA wants.

—
All the best,
-- Yung-jin Lee
bear v2.9.2:- created by Hsin-ya Lee & Yung-jin Lee
Kaohsiung, Taiwan https://www.pkpd168.com/bear
Download link (updated) -> here

ElMaestro ★★★ Denmark, 2013-11-25 11:06 (4171 d 05:19 ago) @ yjlee168 Posting: # 11960 Views: 5,796	df limbo Post reply
	Hi yjlee168, ❝ How about the package 'lmerTest'? It certainly looks like it produces denominator DFs of the Satterthwaite variety. Very interesting, thanks for posting this. Do you by any chance know what the interpretation is and/or if the Satterthwaite df-type is particularly suited to a BE-scenario? — Pass or fail! ElMaestro

yjlee168
★★★

Kaohsiung, Taiwan,
2013-11-25 11:22
(4171 d 05:03 ago)

(edited on 2013-11-25 17:47)
@ ElMaestro
Posting: # 11961
Views: 5,709

df limbo

Post reply

Dear Elmaestro,

Thanks. It said that "...anova table the same as of class merMod but with additional F statistics and p-values calculated based on Satterthwaite’s approximations..." in its pdf file. and so

library(lmerTest)

...

m <- lmer(y ~ formula, data=xxx)

anova(m)  ### which means anova(m, ddf="Satterthwaite",type=3 or 1, ...) by default

...

I don't have SAS and thus cannot validate if it will reproduce the same results as SAS. However, it should be an interesting package.

❝ ❝ How about the package 'lmerTest'?

❝

❝ It certainly looks like it produces denominator DFs of the Satterthwaite variety. Very interesting, thanks for posting this. Do you by any chance know what the interpretation is and/or if the Satterthwaite df-type is particularly suited to a BE-scenario?

Type 3 or 1: Type 3 or 1 anova table with denominator degrees of freedom calculated based on Satterthwaite's approximation.

I would like to make sure if it is what we are looking for first before going further. After all, I am not a statistician at all. More comments or opinions should be helpful.

—
All the best,
-- Yung-jin Lee
bear v2.9.2:- created by Hsin-ya Lee & Yung-jin Lee
Kaohsiung, Taiwan https://www.pkpd168.com/bear
Download link (updated) -> here