Bioequivalence and Bioavailability Forum

preyes323
☆

2010-07-12 19:12
(5422 d 03:08 ago)

Posting: # 5608
Views: 15,473

SABE Reference Variability and CI Computation [General Statistics]

Hi,

I am currently studying on Pharmacokinetic Analysis. My teacher gave me a sample Cmax Data for use in studying. I was tasked to compute for the bioequivalence using the protocol in this link:

FDA Draft Guideline

I got the following values for the reference variability and CI.

reference = 0.378573
CI = 34.55646

According to him these are the wrong values. It should have been 0.452 and -0.16794 respectively.

I have tried recomputing it a number of times already. I also used the SAS code as indicated in the guide. Am I missing something? Is there supposed to be some other implied computation that I am not doing? Below is a link to some of the files I have used for computation as well as the sample data he gave.

Thank you very much in advance to anyone who can help me.

Thanks and Regards,
Paolo

Cmax Reference Variability.zip

Helmut ★★★ Vienna, Austria, 2010-07-12 19:26 (5422 d 02:54 ago) @ preyes323 Posting: # 5609 Views: 14,305	Dataset Post reply
	Dear Paolo! I'm not gifted with but I'm missing the test formulation in your dataset. Can you post the entire stuff? — 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

preyes323 ☆ 2010-07-12 20:20 (5422 d 02:00 ago) @ Helmut Posting: # 5610 Views: 14,341	Dataset Post reply
	Hi, Thanks for the quick reply. Here is the link to the datasets and codes I used. Datasets.zip Thanks, Paolo Edit: Full quote removed. Please delete anything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut]

Helmut
★★★

Vienna, Austria,
2010-07-12 20:38
(5422 d 01:43 ago)

@ preyes323
Posting: # 5611
Views: 14,499

Data format

Post reply

Dear Paolo!

❝ Here is the link to the datasets and codes I used.

OK, with SAS' native binary data format you are leaving everybody without SaS out in the rain. ;-)

Can you please upload data as SAS Transport Files (*.xpt) - which can be imported by (some) other statistical software, or - even better - in Character Separated Format (*.csv)?

—
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

preyes323 ☆ 2010-07-13 03:16 (5421 d 19:05 ago) @ Helmut Posting: # 5612 Views: 14,333	Data format Post reply
	Hi, I am very sorry for that. Attached are the datasets again, this time with the *.xpt file type. Thanks, Paolo (2) Datasets.zip Edit: Full quote removed. Please delete anything from the text of the original poster which is not necessary in understanding your answer; see also this post! [Helmut]

Helmut
★★★

Vienna, Austria,
2010-07-13 04:29
(5421 d 17:51 ago)

@ preyes323
Posting: # 5613
Views: 14,511

SAS and/or Phoenix/WinNonlin-experts around?

Post reply

Dear Paolo!

Had a quick look in Phoenix6.1 (average BE, too late to get FDA's model running).
Fixed: SEQUENCE+TREATMEN+PERIOD
Repeated Specification:
PERIOD
Variance Blocking Variables (Subject): SUBJECT
Group: TREATMEN
Type: Variance Components
Random Effects: TREATMEN
Variance Blocking Variables (Subject): SUBJECT
Type: Banded No-Diagonal Factor Analytic (f)
Number of factors (f): 2

I got:

Warning 11091: Newton's algorithm converged with modified Hessian. Output is suspect.

Model may be over-specified. A simpler model could be tried.

and sigma²_WR 0.1861995 (CV_WR 0.452397)

If I try IBE/PBE I got:

*** WARNING 11121: Subject X had incomplete design and was discarded.

*** WARNING 11122: Sequence RTR had less than two complete subjects and was discarded.

*** WARNING 11122: Sequence TRR had less than two complete subjects and was discarded.

*** ERROR 11107: Fewer than two sequences had complete design. Program termininating.

X in the warning above is the ID of all subjects except ones in sequence RRT. That was it. :confused:

P.S.: Please delete anything from the text of the original poster which is not necessary in understanding your answer; see also this post!

—
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

d_labes
★★★

Berlin, Germany,
2010-07-13 14:15
(5421 d 08:06 ago)

@ Helmut
Posting: # 5615
Views: 14,304

Phoenix/WinNonlin-experts around?

Post reply

Dear Helmut!

❝ I got:

❝ Warning 11091: Newton's algorithm converged with modified Hessian.

❝ Output is suspect.

❝ Model may be over-specified. A simpler model could be tried.

❝ and sigma²_WR 0.1861995 (CV_WR 0.452397)

This is the continuation of our earlier discussion: The FDA model (and your specification is some sort of :yes:

) is over-specified for the extra-reference design.

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2010-07-13 15:42
(5421 d 06:39 ago)

@ d_labes
Posting: # 5616
Views: 14,257

Phoenix/WinNonlin-experts around?

Post reply

Dear D. Labes!

❝ This is the continuation of our earlier discussion: The FDA model

❝ (and your specification is some sort of :yes: ) is over-specified for the

❝ extra-reference design.

Yes, I know. ;-)

I've read exactly this discussion before - but it was too late in the night to start a fight with Phoenix' LinMix Engine, a beast at least as biting as your pet...

I found it interesting that ABE's CV_WR 0.452397 matched Paolo's teacher's 0.452 (though not sure what he meant by 'reference variability').

I'm hoping Simon will pay us a visit.

—
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

d_labes
★★★

Berlin, Germany,
2010-07-13 17:11
(5421 d 05:10 ago)

@ Helmut
Posting: # 5618
Views: 14,600

Mixed interest

Post reply

Dear Helmut!

❝ I found it interesting that ABE's CV_WR 0.452397 matched Paolo's

❝ teacher's 0.452 (though I'm not sure what he means by 'reference

❝ variability').

Then you may find this also interesting :cool:

.

Results of Proc MIXED FDA code:

...

The Mixed Procedure



                     Iteration History



Iteration    Evaluations    -2 Res Log Like       Criterion

        0              1       306.93584974

        1              3       274.26626278      8.89107758

...

        4              1       272.91032505      0.00000213

        5              1       272.91032497      0.00000001



   Convergence criteria met but final hessian is not positive

                           definite.



                        Estimated G Matrix



 Row    Effect       Treatment    Subject        Col1        Col2



   1    Treatment    R             1           0.3313      0.2581

   2    Treatment    T             1           0.2581      0.3596



        Covariance Parameter Estimates



Cov Parm     Subject    Group          Estimate



FA(1,1)      Subject                     0.5756

FA(2,1)      Subject                     0.4485

FA(2,2)      Subject                     0.3981

Residual     Subject    Treatment R      0.1862  (CV=0.4524)

Residual     Subject    Treatment T      0.3269

...

And just more interesting: Analysis according to D. Brown (EMA) of data under treatment with R only, employing an ANOVA with period, sequence and subject within sequence effects (all fixed):

...

The GLM Procedure



Dependent Variable: ln_Cmax   ln_Cmax



                                Sum of

Source              DF         Squares     Mean Square    F Value    Pr > F



Model               45     38.86413212      0.86364738       4.73    <.0001

Error               42      7.66318204      0.18245672

Corrected Total     87     46.52731416

...

(All colors by me).
Very near to the MIXED results. But maybe this coincidence is by chance.
Note also that this is not the same as the analysis via intra-subject contrasts R1-R2 with
s²_WR= 0.143318, s_WR= 0.378573 (Paolos result), CV=0.392551.

But never trust estimates with questionable convergence in REML (WINNONLIN definitely warns at least "Output is suspect", R's lme() will throw an error if not converged and will not give you any parameter estimate or will not give any CI for the covariance parameters if VarCov is not positive definite, but the incredible tells all estimates as if nothing happens :angry:

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2010-07-13 21:19
(5421 d 01:01 ago)

@ d_labes
Posting: # 5620
Views: 14,328

Mixed interest

Post reply

Dear D. Labes!

❝ ...

❝ The Mixed Procedure

❝

❝ Iteration History

❝

❝ Iteration Evaluations -2 Res Log Like Criterion

❝ 5 1 272.91032497 0.00000001

OK, I end up after 7 iterations at:
-2* REML log(likelihood) 251.724
Same result if use the default convergence criterion (10^-10), yours ((10^-8), or approach numeric resolution.

❝ Covariance Parameter Estimates

❝

❝ Cov Parm Subject Group Estimate

❝

❝ FA(1,1) Subject 0.5756

❝ FA(2,1) Subject 0.4485

❝ FA(2,2) Subject 0.3981

❝ Residual Subject Treatment R 0.1862 (CV=0.4524)

❝ Residual Subject Treatment T 0.3269

Closer match than we had in our previous example...

Final variance parameter estimates:

                 lambda(1,1)_11     0.575597

                 lambda(1,2)_11     0.448489

                 lambda(2,2)_11     0.355511

Var(PERIOD*TREATMEN*SUBJECT)_21     0.186160

Var(PERIOD*TREATMEN*SUBJECT)_22     0.359059

❝ And just more interesting: Analysis according to D. Brown (EMA)...

❝ ...

❝ The GLM Procedure

❝

❝ Dependent Variable: ln_Cmax ln_Cmax

❝

❝ Sum of

❝ Source DF Squares Mean Square F Value Pr > F

❝

❝ Model 45 38.86413212 0.86364738 4.73 <.0001

❝ Error 42 7.66318204 0.18245672

❝ Corrected Total 87 46.52731416

Confirmed.


       Total Observations :    88

        Observations Used :    88

              Residual SS :     7.66318204

              Residual df :    42

Final variance parameter estimates:

            Var(Residual)       0.18245672

❝ Very near to the MIXED results. But maybe this coincidence is by chance.

Who knows?

❝ But never trust estimates with questionable convergence in REML (WINNONLIN

❝ definitely warns at least "Output is suspect", R's lme() will throw an error if

❝ not converged and will not give you any parameter estimate or will not give

❝ any CI for the covariance parameters if VarCov is not positive definite, but

❝ the incredible tells all estimates

❝ as if nothing happens :angry: ).

Well, if I ask PHX/WNL for intermediate results (besides many pages of Matrazen aka matrices) I got an additional
Warning 11090: Asymptotic covariance matrix not computed. Information matrix is deemed singular.

—
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

d_labes
★★★

Berlin, Germany,
2010-07-14 12:41
(5420 d 09:39 ago)

@ Helmut
Posting: # 5621
Views: 14,374

lme() answer and beyond ...

Post reply

Dear Helmut!

❝ OK, I end up after 7 iterations at:

❝ -2* REML log(likelihood) 251.724

❝ Same result if use the default convergence criterion (10^-10),

❝ yours ((10^-8), or approach numeric resolution.

Here the results using R's lme().
Code used:

model2 <- lme(log(Cmax) ~ tmt + period + sequence,

              random=list(subject=pdSymm(form= ~tmt-1)),

              weights=varIdent(form = ~ 1 | tmt), 

              data=PKparms, method="REML", na.action=na.omit) 

summary(model2)

Answer:

Linear mixed-effects model fit by REML

 Data: PKparms 

      AIC     BIC   logLik

  291.654 320.096 -135.827  # -2*LL = 271.654! Seems WNL computes different. 



Random effects:

 Formula: ~tmt - 1 | subject

 Structure: General positive-definite

         StdDev   Corr 

tmtR     0.574772 tmtR 

tmtT     0.705001 0.642

Residual 0.431758           # s2=0.186415



Variance function:

 Structure: Different standard deviations per stratum

 Formula: ~1 | tmt 

 Parameter estimates:

       R        T 

1.000000 0.993312           # very interesting: variability of T lower then R.



Fixed effects: log(Cmax) ~ tmt + period + sequence 

               Value Std.Error DF  t-value p-value

(Intercept)  5.60475 0.1867666 86 30.00937  0.0000

tmtT        -0.14753 0.1153259 86 -1.27927  0.2042

period       0.08578 0.0534769 86  1.60402  0.1124

sequenceRTR  0.17906 0.2209845 41  0.81027  0.4225

sequenceTRR -0.28646 0.2297086 41 -1.24704  0.2195

Ok the random effects parameters can't directly compared, only via th G matrix.


getVarCov(model2)

Random effects variance covariance matrix

        tmtR    tmtT

tmtR 0.33036 0.26032          # SAS: 0.3313  0.2581

tmtT 0.26032 0.49703          #      0.2581  0.3596



intervals(model2,which="var-cov",level=0.95)

Error in intervals.lme(model2, which = "var-cov", level = 0.95) : 

  Cannot get confidence intervals on var-cov components: Non-positive

  definite approximate variance-covariance

Seems that the difference in intra-subject variability is absorbed by the G matrix, with no loss in the fit (nearly identical -2*LL).

Quintessence for scABE using the EMA approach:


                                              -- Widened (EMA)

             ----- 90% CIs -----              acceptance range --

           point est.  lower  upper     s²_WR    lower   upper

FDA (MoM)   0.8746    0.7216  1.0605  0.14332   0.7500  1.3334

SAS (FDA)   0.8632    0.7037  1.0498  0.1862    0.7204  1.3881

WNL             ?                     0.18616   0.7204  1.3881

lme()       0.8628    0.7123  1.0452  0.18642   0.7203  1.3884

Confusion

: scABE not proven.

Astonishing enough the 95% upper confidence interval of the linearized scABE criterion
(µ_T-µ_R)²-theta²*sigma²_WR = -0.03751
is negative if theta=0.76 (EMA and recommended by Tothfalusi et.al.) and the MOM terms are used, if I calculated right. And this would indicate scABE proven.

Tothfalusi et.al.
Evaluation of Bioequivalence for Highly Variable Drugs with Scaled Average Bioequivalence
Clin. Pharmacokinet. 2009; 48 (11): 725-743

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2010-07-14 15:47
(5420 d 06:33 ago)

@ d_labes
Posting: # 5624
Views: 14,183

lme() answer and beyond ...

Post reply

Dear D. Labes,

let's continue!

           -2 REML log(LikH)   AIC       BIC

SAS (FDA)      272.910          ?         ?

PHX/WNL        251.724       273.724   304.923

lme()          271.654       291.654   320.096

❝ Seems WNL computes different.

Hhm; according to Phoenix 1.1 users' Guide:

Akaike's Information Criterion (AIC)
The Linear Mixed Effects object uses the smaller-is-better form of Akaike's Information Criterion:

AIC = –2L_R + 2s
where L_R is the restricted log-likelihood function evaluated at final fixed parameter estimates beta-hat and the final variance parameter estimates theta-hat, and s is the rank of the fixed effects design matrix X plus the number of parameters in theta (i.e., s = rank(X) + dim(theta)).
Schwarz's Bayesian Criterion (SBC)
The Linear Mixed Effects object uses the smaller-is-better form of Schwarz's Bayesian Criterion:

SBC = –2L_R + slog(N-r)
where L_R is the restricted log-likelihood function evaluated at the final estimates beta-hat and theta-hat,
N is the number of observations used,
r is the rank of the fixed effects design matrix X, and
s is the rank of the fixed effects design matrix X plus the number of parameters in theta
(i.e., s = rank(X) + dim(theta)).

                                                  Widened (EMA)

                      --- 90% CIs ---           acceptance range

            point est.  lower  upper    s²_WR      lower   upper

FDA (MoM)   0.8746     0.7216  1.0605  0.14332   0.7500  1.3334

SAS (FDA)   0.8632     0.7037  1.0498  0.1862    0.7204  1.3881

PHX/WNL     0.8632     0.7096  1.0500  0.18616   0.7204  1.3881

lme()       0.8628     0.7123  1.0452  0.18642   0.7203  1.3884

—
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

d_labes
★★★

Berlin, Germany,
2010-07-15 14:01
(5419 d 08:19 ago)

@ Helmut
Posting: # 5635
Views: 14,420

AIC, BIC and that all ...

Post reply

Dear Helmut,

❝ let's continue!

           -2 REML log(LikH)   AIC       BIC

SAS (FDA)      272.910       282.9     291.8

PHX/WNL        251.724       273.724   304.923

lme()          271.654       291.654   320.096

❝ ❝ Seems WNL computes different.

Seems SAS computes different.
According to the SAS help on Proc MIXED:

AIC = -2*LL + 2*d
BIC = -2*LL + d*log(n)

Here LL denotes the maximum value of the (possibly restricted) log likelihood, d the dimension of the model, and n the number of observations. In SAS 6 of SAS/STAT software, n equals the number of valid observations for maximum likelihood estimation and n-p for restricted maximum likelihood estimation, where p equals the rank of X. In later versions, n equals the number of effective subjects as displayed in the "Dimensions" table, unless this value equals 1, in which case n equals the number of levels of the first random effect you specify in a RANDOM statement. If the number of effective subjects equals 1 and you have no RANDOM statements, n then reverts to the SAS 6 values. For restricted likelihood estimation, d equals q, the effective number of estimated covariance parameters. In SAS 6, when a parameter estimate lies on a boundary constraint, then it is still included in the calculation of d, but in later versions it is not. The most common example of this behavior is when a variance component is estimated to equal zero. For maximum likelihood estimation, d equals q+p.

A very concise and clear description of the calculations in SAS 9.2 :confused:

. That recognize who will or can.

Of course the R's lme() values are different from SAS's. The R folks undertake each effort to do things not the <$ineffable$> way :-D

—
Regards,

Detlew

d_labes
★★★

Berlin, Germany,
2010-07-13 13:51
(5421 d 08:29 ago)

@ preyes323
Posting: # 5614
Views: 14,606

To Err is Human

Post reply

"Good-Nature and Good-Sense must ever join;
To err is human, to forgive divine."
Alexander Pope (1688-1744)

Hi Paolo,
after looking at your data and your [image]

code some observations:

Your implementation of the intra-subject contrast T-R (ilat in the guidance) is faulty. You must calculate T-0.5(R1+R2).
The SAS code given in the guidance is half-cooked. See here and here.
First: To get a 95% upper CI of the scaled ABE criterion IMHO you must use alpha=0.05 in the intermediate analysis of ilat.
Second: The ominous x in the IGLM2 part of the code is IMHO wrong. What wee need here is (Y_T-Y_R)² and this would translate to me to x=estimate**2 in the SAS code.
Third: The output in dglm1 from the intermediate analysis of (R1-R2) contains 3 rows of the ANOVA. We need only the row with the error term. Add a where source="Error"; in the corresponding datastep to calculate s_WR.
Regarding the value for that I don't find an error. So dispute with your teacher .
Try to check your SAS results (validate ) by implementing the necessary formulas from scratch f.i. in R. See here for some useful code snippets.

Hope this helps.

BTW: Can you tell us where the data came from?
And from where your teacher knows the solution?

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2010-07-13 15:45
(5421 d 06:35 ago)

@ d_labes
Posting: # 5617
Views: 14,257

To Err is Human, but...

Post reply

Dear D. Labes!

❝ "To err is human, to forgive divine."

❝ Alexander Pope (1688-1744)

"To err is human, but to really foul things up
requires a computer."
(Anonymous in Farmer's Almanac, 1978)

—
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

d_labes ★★★ Berlin, Germany, 2010-07-13 17:47 (5421 d 04:33 ago) @ Helmut Posting: # 5619 Views: 14,168	... to Arr is Pirate Post reply
	Dear Helmut! This is one for our captain: "To Err is Human, to Arr is Pirate " 19. September each year — Regards, Detlew

ElMaestro
★★★

Denmark,
2010-07-14 23:03
(5419 d 23:17 ago)

@ d_labes
Posting: # 5629
Views: 14,068

Brilliant page!!!

Post reply

Dear d_labes,

❝ 19. September each year

This is an awesome page. Must immediately update my bookmarks. Who on earth knew that there was a pirate version of Google?

And the loveable "Damn ye, yellow-bellied sapsuckers, I'm a better man than all of ye milksops put together" gives me the impression that Blackbeard was aspiring hard to a position in an agency somewhere.

Many thanks and best regards,
EM.

Edit: Ahoy, pirrrate version of Google linked, matey ! [Ohlbe]

—
Pass or fail!
ElMaestro

preyes323
☆

2010-07-14 16:45
(5420 d 05:36 ago)

@ d_labes
Posting: # 5626
Views: 14,052

To Err is Human

Post reply

Hi All,

Thanks for all the replies. I have lots to catch up on. :-D

I will be going through all your posts first. :-)

From the comments I am reading, I guess I shouldn't have taken the guidelines at face value then...

With regards to the answer my teacher gave, he said that the answer he gave me was already FDA approved and would say nothing more. I tried asking for the solution but he said he was not provided with it. He did not bother to discuss further with me since he said this problem was only for our practice purposes.

Again, thanks for all the replies. I'll be posting my findings again after going through all the information you guys provided.

Thanks,
Paolo

Helmut
★★★

Vienna, Austria,
2010-07-14 17:17
(5420 d 05:03 ago)

@ preyes323
Posting: # 5627
Views: 14,075

Heads up!

Post reply

Dear Paolo!

❝ […] my teacher […] did not bother to discuss further with me since he said this problem was only for our practice purposes.

(Some) teachers are funny persons, aren't they?
If the problem is 2×2 and you get 5 and he claims 3 and there are rumors that there's another result like 4 - don't worry: It's for practice purposes only!

Non scholae, sed vitae discimus
We do not learn for school, but for life
:blahblah:

❝ Again, thanks for all the replies. I'll be posting my findings again after going through all the information you guys provided.

Good luck! On the contrary to your teacher we don't have the right (?!) answer yet. :-D

—
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

d_labes ★★★ Berlin, Germany, 2010-07-15 13:05 (5419 d 09:15 ago) @ preyes323 Posting: # 5633 Views: 14,073	To Err is Teacher Post reply
	Dear paolo, if your teachers result was FDA approved it is sainted. No need to dare to ask where it comes from. Follow Helmut's piece of advice and learn this lesson for your further live: If you are asked, whatever question, always answer "It is FDA approved!" . Holy mackerel! (in German, not literally: Heiliger Bimbam!). How ignorant can it yet be? — Regards, Detlew

preyes323
☆

2010-07-16 04:28
(5418 d 17:52 ago)

@ d_labes
Posting: # 5643
Views: 14,260

To Err is Human

Post reply

Hi All,

❝ Your implementation of the intra-subject contrast T-R (ilat in the guidance) is faulty. You must calculate T-0.5(R1+R2).

My bad on this. I went through my code a number of times but didn't even notice this!!! :-(

❝ • The SAS code given in the guidance is half-cooked. See here and here.

❝ First: To get a 95% upper CI of the scaled ABE criterion IMHO you must use

❝ alpha=0.05 in the intermediate analysis of ilat.

❝ • Second: The ominous x in the IGLM2 part of the code is IMHO wrong. What

❝ wee need here is (Y_T-Y_R)² and this would translate to me to x=estimate**2

❝ in the SAS code.

I revised this two as well (see way below for my codes :-D

). After revising, I ran my models however I got a different result than the one gotten by d_labes (see below). Mine is 0.023485. Is there something wrong with my implementation of the FDA guideline for the crit bound?

❝ (µ_T-µ_R)²-theta²*sigma²_WR = -0.03751

I am also confused with the difference between the S2wr results by the ABE SAS code by FDA (page 9 of progesterone guideline) using PROC Mixed and the S2wr formula for intra-subject contrasts. I tried using both as well. I produced the results that was earlier posted. Why is it that Sigma2WR values are different? Is this ok? I read in your other posts the Progesterone guidance makes use of MoM. Is MoM the approach really used for SABE then?

Thanks again.

Thanks and Regards,
Paolo

Code for IGLM


proc glm data=SABE.scavbe_Cmax;

   class sequence;

   model ilat = sequence/clparm alpha =0.05;

   estimate 'average' intercept 1 sequence 0.3333333333 0.3333333333 0.3333333333;

   ods output overallanova=SABE.iglm1_Cmax;

   ods output Estimates=SABE.iglm2_Cmax;

   ods output NObs=SABE.iglm3_Cmax;

   title1 'scaled average BE';

run;



Data SABE.iglm2_Cmax;

   set SABE.iglm2_Cmax;

   pointest=exp(estimate);

   x=estimate**2;

   boundx=(max((abs(LowerCL)),(abs(UpperCl))))**2;

run;

Code for Crit Bound

data SABE.upper_cmax;

   merge SABE.dglm1_cmax SABE.iglm2_cmax;

   theta=((log(1.25))/0.25)**2;

   y=-theta*s2wr;

   boundy=y*dfd/cinv(0.95,dfd);

   sWR=sqrt(s2wr);

   critbound=(x+y)+sqrt(((boundx-x)**2)+((boundy-y)**2));

run;

d_labes
★★★

Berlin, Germany,
2010-07-16 12:42
(5418 d 09:39 ago)

@ preyes323
Posting: # 5644
Views: 13,950

To Err is Human

Post reply

"If debugging is the process of removing bugs,
then programming must be the process of putting them in."
(Edsger W. Dijkstra)

Dear Paolo,

❝ »... You must calculate T-0.5(R1+R2).

❝ My bad on this. I went through my code a number of times but didn't even notice this!!! :-(

See the subject of these posts and the slogan of he day :-D

.
Or another one in German "Man sieht den Wald vor lauter Bäumen nicht!" (Rough translation: Don't see the wood for the trees.)

❝ I revised this two as well (see way below for my codes :-D ). After revising, I ran my models however I got a different result than the one gotten by d_labes (see below). Mine is 0.023485.

I have calculated using (is theta in your code)
theta² = ((log(1.25)/0.2936)²=(0.76)²
(BTW: Yes nitpickers, of course I know it's really (log(1.25)/sqrt(log(0.3*0.3+1)))^2=(0.7601283...)^2 :cool:

.)

This is recommended by Tothfalusi et. al. (see here for literature) to avoid a discontinuity at the switching variance of CV=30% (switch from usual 80 ... 125% acceptance range to the widened ranges) and is the regulatory constant in the new EMA bioequivalence guidance.
This discontinuity leads to a high alpha-inflation around CV=30%.

❝ I am also confused with the difference between the S2wr results by the ABE SAS code by FDA (page 9 of progesterone guideline) using PROC Mixed and the S2wr formula for intra-subject contrasts. I tried using both as well. I produced the results that was earlier posted. Why is it that Sigma2WR values are different? Is this ok?

(color by me)

Dear Paolo: Sorry but this question is plaguing us as well and is source of our continuing discussion here.
Especially we don't know if one can trust in the results of the FDA Proc Mixed code in case of the extra-reference design your data came from. :ponder:

—
Regards,

Detlew

preyes323
☆

2010-07-17 13:14
(5417 d 09:06 ago)

@ d_labes
Posting: # 5648
Views: 13,851

To Err is Human

Post reply

Hi d_labes,

❝ I have calculated using (is theta in your code)

❝ theta² = ((log(1.25)/0.2936)²=(0.76)²

❝ (BTW: Yes nitpickers, of course I know it's really

❝ (log(1.25)/sqrt(log(0.3*0.3+1)))^2=(0.7601283...)^2 :cool: .)

Yes, I included theta in my code. From what I see our main difference might be in the formula for the crit bound. The formula I used is the one from the FDA progesterone guideline. Did you use the same?

data SABE.upper_cmax;

   merge SABE.dglm1_cmax SABE.iglm2_cmax;

   theta=((log(1.25))/0.25)**2;

   y=-theta*s2wr;

   boundy=y*dfd/cinv(0.95,dfd);

   sWR=sqrt(s2wr);

   critbound=(x+y)+sqrt(((boundx-x)**2)+((boundy-y)**2));

run;

❝ Dear Paolo: Sorry but this question is plaguing us as well and is source of our continuing discussion here.

❝ Especially we don't know if one can trust in the results of the FDA Proc Mixed code in case of the extra-reference design your data came from. :ponder:

I see.

Thanks and Regards,
Paolo

d_labes ★★★ Berlin, Germany, 2010-07-19 12:00 (5415 d 10:20 ago) @ preyes323 Posting: # 5653 Views: 14,169	Regulatory constants Post reply
	Hi Paolo, mine (your theta is mine theta²): ❝ ❝ `theta² = ((log(1.25)/0.2936)²=(0.76)²` yours: ❝ `theta=((log(1.25))/0.25)2;= 0.89257422` — Regards, Detlew

ElMaestro ★★★ Denmark, 2011-02-06 12:54 (5213 d 08:26 ago) @ preyes323 Posting: # 6564 Views: 13,316	Data format Post reply
	Hi all, ❝ (2)Datasets.zip sorry to ask about this: Do any of you have the dataset used in this post in a format that is more readable (ideally, saved from Excel in a txt-file) ?? Thanks a lot for any help. — Pass or fail! ElMaestro

Helmut ★★★ Vienna, Austria, 2011-02-06 14:02 (5213 d 07:18 ago) @ ElMaestro Posting: # 6565 Views: 13,366	SAS System Viewer Post reply
	Dear ElMaestro, believe it or not - after registration you might get free SAS System Viewer — 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

SABE Reference Variability and CI Computation [General Sta­tis­tics]

Dataset

Dataset

Data format

Data format

SAS and/or Phoenix/WinNonlin-experts around?

Phoenix/WinNonlin-experts around?

Phoenix/WinNonlin-experts around?

Mixed interest

Mixed interest

lme() answer and beyond ...

lme() answer and beyond ...

AIC, BIC and that all ...

To Err is Human

To Err is Human, but...

... to Arr is Pirate

Brilliant page!!!

To Err is Human

Heads up!

To Err is Teacher

To Err is Human

To Err is Human

To Err is Human

Regulatory constants

Data format

SAS System Viewer

SABE Reference Variability and CI Computation [General Statistics]