jag009
★★★

NJ,
2014-12-05 21:57
(3400 d 13:10 ago)

Posting: # 13991
Views: 13,851
 

 Concerta Guidance: Sub x Form inter­action vari­ance test [Regulatives / Guidelines]

Thread locked
Hi everyone,

Question regarding the sub*Form interaction variance test required for Concerta draft guidance...

"Subject-by-formulation interaction variance, σ2D2I-0.5*(σ2wt + σ2wr)

In Step 4..."As per APPENDIX A in the FDA 2001 bioequivalence guidance, the recommended allowance for σ2D is 0.03."


They are referring to σ2D value being no greater than 0.03 right? Can the number be negative, i.e, -0.06?

Thanks
John
AngusMcLean
★★  

USA,
2014-12-06 02:28
(3400 d 08:39 ago)

@ jag009
Posting: # 13993
Views: 12,233
 

 Concerta Guidance: Sub x Form interaction variance test

 
I am thinking "no". It is from a (sum of squares) so by definition it cannot be negative.

If you are getting a negative value then I would check the formula you are using to do the calculation and the calculations.

Also I would look at your input to make sure your PK parameters are correctly entered. If you are using hypothetical parameters then I would examine them with a view to altering them to make them more realistic.

Are you using hypothetical parameters?

Angus
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-12-06 02:53
(3400 d 08:13 ago)

@ AngusMcLean
Posting: # 13994
Views: 12,518
 

 Concerta Guidance: Sub x Form inter­action vari­ance test

 
Hi Angus & John,

❝ I am thinking "no". It is from a (sum of squares) so by definition it cannot be negative.


I’m thinking “yes”. IMHO it’s the difference of the between- and pooled within-subject variances. Even in 2×2 cross­overs sometimes CVintra > CVinter. Then (and here as well) you would get a negative variance component ([image] search the forum). I guess SAS forces it to zero and Phoenix screams for help. If you have a “real world” dataset you could try the explicit formulas from the guidance to check the values of σ²I, σ²WT, and σ²WR.

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
jag009
★★★

NJ,
2014-12-07 21:58
(3398 d 13:09 ago)

(edited by jag009 on 2014-12-07 22:10)
@ Helmut
Posting: # 14005
Views: 12,300
 

 Concerta Guidance: Sub x Form inter­action variance test

 
Hi folks,

❝ I’m thinking “yes”. IMHO it’s the difference of the between- and pooled within-subject variances. Even in 2×2 cross­overs sometimes CVintra > CVinter...


I used the ema 4 way fully replicate dataset (Cmax) (I don't have the link now but I am sure you know what I am referring to) and tried it with my revised FDA RSABE SAS code. Briefly I obtained withsubject variances for T & R from dlats and the withinsubject variance for I from ilat (this might be my problem?). As per concerta guidance the sub*form interaction variance is,

Mi - 0.5* (Mt + Mr).

Using this formula my value was -0.074978. Can someone try this with SAS and let me know? Oh my withinsubject variances for T and R were 0.11654 and 0.19931. The withsubject variance for I was 0.08295 from ilat.

Update: I think I know what my potential problem is... I think I misinterpreted the withinsubject variance for I. For withinsubject variance T & R, is derived from estimate/2 in SAS, but for I its just the estimate?

Sorry I am typing from a tablet... ;-)

John
AngusMcLean
★★  

USA,
2014-12-08 02:35
(3398 d 08:32 ago)

@ jag009
Posting: # 14007
Views: 12,028
 

 Concerta Guidance: S×F vari­­ance test

 
if you are using PHOENIX then you can calculate MI FROM:

MI=(σ 2 WT + σ 2 WR )/2 + σ2D

So you check your calculation for MI that way since the other variables are provided by PHOENIX output,


I am doing the much the same thing you are doing, but using Phoenix with the 4 period Full ABAB data set (progesterone) with the FDA template. I understand from Helmut that the progesterone data set is a simulated one.

Here is my result from Phoenix for AUC

σ2D =-0.80391.

So I have a negative value and I wonder if it is due to imperfections associated with simulating or creating your own data set.

Why not try that set in SAS and let me know what you see.

Angus


Edit: Merged from a later post. Note that you can edit your original post within 24 hours. [Helmut]
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-12-08 15:12
(3397 d 19:54 ago)

@ AngusMcLean
Posting: # 14012
Views: 11,994
 

 Concerta Guidance: S×F vari­­ance test

 
Hi Angus,

❝ I am doing the much the same thing you are doing, but using Phoenix with the 4 period Full ABAB data set (progesterone) with the FDA template. I understand from Helmut that the progesterone data set is a simulated one.


Correct. There is nothing like a “progesterone dataset”. The one you and John are referring to is from EMA’s Q&A-document.

❝ Why not try that set in SAS and let me know what you see.


John’s results were obtained in SAS.

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
jag009
★★★

NJ,
2014-12-08 18:40
(3397 d 16:27 ago)

@ Helmut
Posting: # 14015
Views: 12,198
 

 Update: S×F vari­­ance test

 
Hi Angus, Helmut and others,

Either I am not thinking straight (Monday...) or something else but my outcome is not the same as Angus'. Working off the EMA 4-way Fully replicate dataset (yes same Cmax Dataset I) in Winnonlin to get the sub*form interaction variance as per Helmut's previous post in which the final variance parameter lambda(2,2)_11 is the sub*form interaction, I got the following output:

Dependent     Parameter                            Estimate
Ln(Cmax)      lambda(1,1)_11                       0.82840679
Ln(Cmax)      lambda(1,2)_11                       0.85299475
Ln(Cmax)      lambda(2,2)_11                       5.63E-07
Ln(Cmax)      Var(Period*Formulation*Subject)_21   0.11739425
Ln(Cmax)      Var(Period*Formulation*Subject)_22   0.20211805


So the sub*form interaction variance σD2 is 5.63E-07? My σWT2 and σWR2 matched (see my previous post above, close enough). Working backward with the winnonlin σD2=5.63E-07, σI2=0.159756713?

Angus, how did you arrive at σD2 =-0.80391.

My covariance residual from ilat (SAS) was 0.165897781

John
AngusMcLean
★★  

USA,
2014-12-08 19:09
(3397 d 15:58 ago)

@ jag009
Posting: # 14017
Views: 11,849
 

 Update: S×F vari­­ance test

 
if you look at my post I think I said it was the AUC paramater I was working with from the 4 period FDA ABAB template using the simulated data set.

I have not looked at the Cmax, but I will.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-12-08 19:18
(3397 d 15:49 ago)

@ AngusMcLean
Posting: # 14019
Views: 11,939
 

 EMA’s dataset I

 
Hi Angus,

are you trying to confuse us? EMA’s dataset I gives only one variable, called “DATA” – or if you want to work directly with the transformed ones “logDATA” (Q&A, p 24–29).

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
AngusMcLean
★★  

USA,
2014-12-08 20:55
(3397 d 14:12 ago)

@ Helmut
Posting: # 14024
Views: 11,979
 

 EMA’s dataset I

 
HELMUT: No not at all! I looked at the magnitude of the number and assumed that it was an AUC. If a mass unit is in picograms and it runs to thousands then usually, but not always, it is expressed in nano-gram/mL.

Angus
jag009
★★★

NJ,
2014-12-08 21:41
(3397 d 13:26 ago)

@ AngusMcLean
Posting: # 14025
Views: 11,886
 

 Update: S×F vari­­ance test

 
My bad, the data column is not parameter labeled, but I arbitrary labeled it as Cmax:-D. Anyhow, there is only one data column (non-transformed) and one ln-tranformed data column...

John
AngusMcLean
★★  

USA,
2014-12-08 19:18
(3397 d 15:49 ago)

@ jag009
Posting: # 14020
Views: 12,171
 

 Update: S×F vari­­ance test

 
I have a recent email from Ana over at Pharsight: Please see below and let me know your thoughts. Evidently there has been confusion in the past and I am wondering whether it has been resolve (below).

"Some additional discussion about the questions that you asked about BE has taken place within Pharsight. I just wanted to give you an update.

1. We entered an enhancement on our tracking system “Provide additional output in Pop/Indiv Bioequivalence - sigmaD, sigmaWT (for subject by formulation interaction variance)”. WinNonlin is calculating this but not providing it in the output.
2. For certain designs the documentation has some information that might be useful to you:

For an RTRT/TRTR study, if the default model for replicated crossovers is used, which has a random model using type FA0(2), look at the Parameter and the Group_Level columns on the Parameter Key output workbook to determine which variance parameter is for the reference formulation and which is for the test formulation. In the following equations, it is assumed that the first output parameter is for the test formulation.

The within-subject variance of the test, sig_WT^2, and the within-subject variance of the reference, sig_WR^2, are in the output as:
sig_WT^2 = Var(Period*Formulation*Subject)_21
sig_WR^2 = Var(Period*Formulation*Subject)_22

The intra-subject CV's can be computed as:
IntraCV_T = 100%*sqrt(exp(sig_WT^2)-1)
IntraCV_R = 100%*sqrt(exp(sig_WR^2)-1)

The between-subject variance of test, sig_BT^2, the between-subject variance of reference, sig_BR^2, and the G matrix for the random model, can be computed as follows. This still assumes that the test formulation precedes the reference. The '_11' notation on the lambda parameters (which indicates that they are for the first random model and first group) has been dropped to improve readability:

G(1,1) = sig_BT^2 = lambda(1,1)^2
G(2,2) = sig_BR^2 = lambda(2,1)^2+lambda(2,2)^2
G(1,2) = rho*sig_BT*sig_BR = lambda(1,1)*lambda(2,1) (= between subject covariance for test and reference)

The subject-by-formulation interaction variance in general is sig_D^2 = sig_BT^2 + sig_BR^2 - rho*sig_BT*sig_BR, and can be computed from the above equations as:
sig_D^2=G(1,1)+G(2,2)-2*G(1,2)

The inter-subject CV's can be computed as:
InterCV_T = 100%*sqrt(exp(sig_BT^2)-1)
InterCV_R = 100%*sqrt(exp(sig_BR^2)-1)

If the CSH variance structure is used instead of FA0(2), the following equations change. This still assumes the first output parameter is for test and the second for reference.

G(1,1) = sig_BT^2 = cshSD(1)^2
G(2,2) = sig_BR^2 = cshSD(2)^2
G(1,2) = cshCorr * cshSD(1) * cshSD(2)"


The parameter name or units is not in the file I have: it is from the FDA template, it is the same file for EMEA.

Here is the output from Phoenix. Now one has to look at Ana's email

Dependent Units Parameter Estimate
Data lambda(1,1)_11 0.85299471
Data lambda(1,2)_11 0.82840678
Data lambda(2,2)_11 8.34E-07
Data Var(Period*Formulation*Subject)_21 0.20211811
Data Var(Period*Formulation*Subject)_22 0.11739421


Herr is my paramater key and the REFERENCE IS first. My file is RTRT

Dependent Units Parameter Source Type Bands Subject_Term Group_Term Group_Level
Data lambda(1,1)_11 Random No-Diag Factor Analytic 2 Subject
Data lambda(1,2)_11 Random No-Diag Factor Analytic 2 Subject
Data lambda(2,2)_11 Random No-Diag Factor Analytic 2 Subject
Data Var(Period*Formulation*Subject)_21 Repeated Variance Components Subject Formulation Formulation_R
Data Var(Period*Formulation*Subject)_22 Repeated Variance Components Subject Formulation Formulation_T


That is why the parameters are switched from yours.


Edit: Merged with two later post. Note that you can edit your original post within 24 hours. [Helmut]
jag009
★★★

NJ,
2014-12-10 18:01
(3395 d 17:06 ago)

@ AngusMcLean
Posting: # 14047
Views: 11,712
 

 Question

 
Hi Angus and other gurus,

I have a question...

❝ G(1,1) = sig_BT^2 = lambda(1,1)^2

❝ G(2,2) = sig_BR^2 = lambda(2,1)^2+lambda(2,2)^2

❝ G(1,2) = rho*sig_BT*sig_BR = lambda(1,1)*lambda(2,1) (= between subject covariance for test and reference)


Why Sig_BR^2 is the sum of 2 terms? What does Lamba(2,2) represent?

Thanks
John
AngusMcLean
★★  

USA,
2014-12-10 19:34
(3395 d 15:33 ago)

@ jag009
Posting: # 14050
Views: 11,688
 

 Question

 
ANA is the GURU:

SEE HER EMAIL

G(2,2) = sig_BR^2 = lambda(2,1)^2+lambda(2,2)^2

G(1,2) = rho*sig_BT*sig_BR = lambda(1,1)*lambda(2,1) (= between subject covariance for test and reference)

THE SIGMA SQUARED D calculation is shown in one of my posts from yesterday. The co variate term is subtracted from the sum of the other two parameters.

From Helmut's post above to the European document the following narrative is obtained.

This model allows a different subject effect for each formulation (i.e. a subject by formulation interaction), and therefore has 5 variance terms:


(1) Within subject for reference
(2) Within Subject for Test
(3) Between Subject for Test
(4) Between Subject for Reference
(5) Covariance for between subject test and reference

It also states the last three are "combined" to give the subject ×formulation interaction variance component.

From Ana’s letter the "combination" involves the addition of (3+4)-5.
kumarnaidu
★    

Mumbai, India,
2014-12-09 07:48
(3397 d 03:19 ago)

@ jag009
Posting: # 14031
Views: 11,952
 

 Update: S×F vari­­ance test

 
Hello friends, I am getting below output using this SAS code. I dont know why I am getting value 0 (I think it is a rounded value) for FA(2,2).
Please guide me.

proc mixed data=sample method=reml ITDETAILS CL=WALD ALPHA=0.1;
class Subject sequence period formulation;
model logDATA=sequence period formulation/ddfm=satterth;
random formulation/type=FA0(2) subject=subject G;
repeated/group=formulation subject=subject;
lsmeans formulation/pdiff cl alpha=0.1;
estimate 'T-R' formulation 1 -1 /CL ALPHA=0.1;
ods output Estimates=BLUEUN CovParms=COVUN AsyCov=ascovun;
run;


Dependent  Parameter                   Estimate
Ln(data)   FA(1,1)                     0.8530
Ln(data)   FA(1,2)                     0.8284
Ln(data)   FA(2,2)                     0
Ln(data)   Res(Formulation R*Subject)  0.2021
Ln(data)   Res(Formulation T*Subject)  0.1174

Kumar Naidu
jag009
★★★

NJ,
2014-12-09 16:16
(3396 d 18:51 ago)

(edited by jag009 on 2014-12-09 16:33)
@ kumarnaidu
Posting: # 14036
Views: 11,914
 

 Update: S×F vari­­ance test

 
Hi,

Why don't you just use the 4-way Fully Replicate SAS code from the FDA progesterone draft guidance and revise it according the instructions given in the new Concerta draft guidance?

John
AngusMcLean
★★  

USA,
2014-12-09 18:47
(3396 d 16:20 ago)

@ jag009
Posting: # 14038
Views: 11,804
 

 Update: S×F vari­­ance test

 
From above message the output 4 period replicate FDA template is repeated as follows:

Dependent Parameter Estimate
ln Cmax lambda(1,1)_11 is 0.82840679
ln Cmax lambda(1,2)_11 is 0.85299475
ln Cmax lambda(2,2)_11 is 5.63E-07
ln Cmax Var(Period*Formulation*Subject)_21 is 0.11739425
ln Cmax Var(Period*Formulation*Subject)_22 is 0.20211805

From Ana’s message the value for subj. formulation interaction(sig_D^2) is as follows:

(0.82840679)+(5.63E-07)-2*(0.85299475)= -0.87758

Here are my results from the same data set in Phoenix: my input structure is different
Data lambda(1,1)_11 0.85299471
Data lambda(1,2)_11 0.82840678
Data lambda(2,2)_11 8.34E-07
Data Var(Period*Formulation*Subject)_21 0.20211811
Data Var(Period*Formulation*Subject)_22 0.11739421

I used BABA for the input for the first subject. If you look at the Parameter Key and Group Levels on the Parameter Key Output Workbook in Phoenix I have first output parameter as reference not test. That is why there is transposition in the parameter values.
jag009
★★★

NJ,
2014-12-09 19:01
(3396 d 16:06 ago)

@ AngusMcLean
Posting: # 14039
Views: 11,955
 

 Update: S×F vari­­ance test

 
Hi Angus,

❝ From above message the output 4 period replicate FDA template is repeated as follows:


Confirmed. I followed your email (Ana's) in Winnonlin and arrived with the same conclusion. Now I just need to figure it out in SAS.

John
AngusMcLean
★★  

USA,
2014-12-09 19:30
(3396 d 15:37 ago)

@ jag009
Posting: # 14040
Views: 11,740
 

 Update: S×F vari­­ance test

 
Thanks for the update: SAS will give a very similar result. According to Helmut SAS assigns 0 to negative variance values and we see this in the post from India. We have to keep in mind that this data set is simulated and this is surely the source of the problem?

Once you complete SAS work then we can compare results of the upper confi­dence interval calculation for sigmaD2 comparison given in the Guidance. I have already an Excel spreadsheet plus chi squared from another program to provide the upper CI. I need to verify my results by comparison with someone else.

So far what I see is very high values for upper CI, but this could be due to lack of an actual data set
jag009
★★★

NJ,
2014-12-18 16:31
(3387 d 18:36 ago)

@ AngusMcLean
Posting: # 14111
Views: 11,112
 

 Update: S×F vari­­ance test

 
Hi Angus,

❝ Thanks for the update: SAS will give a very similar result. According to Helmut SAS assigns 0 to negative variance values and we see this in the post from India. We have to keep in mind that this data set is simulated and this is surely the source of the problem?


What value did you arrive at for the degree of freedom (n-s) used to calculate the 95 CI? I am getting something like 67. Yes, with the same EMA dataset.

Thanks
John
AngusMcLean
★★  

USA,
2014-12-18 17:12
(3387 d 17:55 ago)

@ jag009
Posting: # 14112
Views: 11,055
 

 Update: S×F vari­­ance test

 
John: please outline the steps you made to arrive at the number of degrees of freedom for confidence intervals in Phoenix. I have not as yet calculated them.

Angus
jag009
★★★

NJ,
2014-12-22 18:28
(3383 d 16:39 ago)

@ AngusMcLean
Posting: # 14136
Views: 10,984
 

 Update: S×F vari­­ance test

 
Hi Angus,

Before I go further, just want to clarify, did you go through the exercise 1) as per FDA's concerta guidance, calculating Iij, Tij, Rij, YijR, YijT etc etc and then obtain the within-subject variances for T and R in Phoenix?
2) Or you just threw the EMA dataset (ln-transformed) into Phoenix ABE routine and obtain #s from The G matrix output? I imagine you would say #2?

I don't think approach #2 will give the right answer and the d.f. which is needed to compute the 95% CB as per FDA instruction.

Helmut and I goofed around with both scenarios (in Phoenix, and I did in SAS) and the results were a bit amusing (I will let Helmut to summarize if he wants).

John
AngusMcLean
★★  

USA,
2014-12-22 20:32
(3383 d 14:35 ago)

@ jag009
Posting: # 14137
Views: 11,015
 

 Update: S×F vari­­ance test

 
John: what I did was ran the RSABE template with the EMEA data: first ABE is evaluated prior to RSABE in Phoenix.

Therefore after completing the example in both ABE and RSABE I went to the Average BE results tab to get the G matrix values.


Data lambda(1,1)_11 0.85299471
Data lambda(1,2)_11 0.82840678
Data lambda(2,2)_11 8.34E-07
Data Var(Period*Formulation*Subject)_21 0.20211811
Data Var(Period*Formulation*Subject)_22 0.11739421

The structure of the EMNEA input data is RTRT not TRTR so paramater values are transposed. {It should be TRTR for input}

Do you follow? Please comment on above.

Subsequently in Excel using the values above I can create MI from Wt and WR AND SIGMA.

MI = (σ2WT + σ2WR )/2 + σ2D

I have a spreadsheet that can perform subsequent calculations except that I have to calculate chi squared in another program (NCSS)
jag009
★★★

NJ,
2014-12-22 21:19
(3383 d 13:48 ago)

(edited by jag009 on 2014-12-23 04:44)
@ AngusMcLean
Posting: # 14138
Views: 11,366
 

 Update: S×F vari­­ance test

 
Hi Angus,

❝ John: what I did was ran the RSABE template with the EMEA data: first ABE is evaluated prior to RSABE in Phoenix.


So you followed the template which Helmut generated for Phoenix (or his instruction in calculating RSABE)?

I don't think you should look at the ABE analysis but instead look through the RSABE part of the worksheet and results. The RSABE part of the worksheet will give you the within-subject variances for T and R and I still believe the within-subject variance for I (or so called Ilat in the SAS code) is obtained from there also (and you will also get the df there). I think that path is how FDA wants our conclusion to arrive at.

Email me (you know how right) and I will give you more info. Maybe then we can see if we can come to an agreement before going further.

John
AngusMcLean
★★  

USA,
2014-12-22 21:45
(3383 d 13:22 ago)

@ jag009
Posting: # 14141
Views: 11,183
 

 Update: S×F vari­­ance

 
John: I got the updated RSABE template from Ana Henry of Phoenix: I do not know the extent of Helmut's involvement. I looked at the RSABE results part of the template aa you suggest. I see below. I do not know what Cinv is?


dlat 69 (df) 0.11653964 (variance WR) 89.391208 (Cinv)

Any comment?

I know it is also possible to run individual BE in this template. Is this what you have in mind.

Angus
jag009
★★★

NJ,
2014-12-22 21:51
(3383 d 13:15 ago)

@ AngusMcLean
Posting: # 14142
Views: 10,976
 

 Update: S×F vari­­ance

 
Hi Angus,

Helmut generated a worksheet (or instruction to generate one) a while while back which enables one to analyze RSABE (Search the forum, you will see posts about it). I don't have Ana's template. Is it freely distributed? If so, I can get it, but where?

Cinv is to get the values from the chi-square distribution.


John
AngusMcLean
★★  

USA,
2014-12-22 22:10
(3383 d 12:57 ago)

@ jag009
Posting: # 14143
Views: 10,996
 

 Update: S×F vari­­ance

 
No: you must use Anas updated worksheet, since it is updated and works better. I do not follow how you get MI or what you do with Cinv. I know you need the df for chi squared but I do not follow how to use Cinv to get the value from Chi squared. I have a program that used the mean values plus the probability and the Mean values of Mi, Mt and Mr.

http://tinyurl.com/p6xmzwv


Edit: I shortend the 199 character original URL. [Helmut]
AngusMcLean
★★  

USA,
2014-12-22 22:36
(3383 d 12:31 ago)

@ jag009
Posting: # 14144
Views: 10,914
 

 Update: S×F vari­­ance

 
John: from the RSABE worksheet. Is this MI . it is called Ilat.


Dependent Units Title Estimate StdError Denom_DF T_stat P_value Conf_Level T_critical Lower_CI Upper_CI
ilat Average Sequence Effects* -0.14376536 0.049080231 67 -2.9291907 0.004641283 90 1.6679161 -0.22562707 -0.061903649
jag009
★★★

NJ,
2014-12-22 23:18
(3383 d 11:49 ago)

@ AngusMcLean
Posting: # 14145
Views: 10,824
 

 Update: S×F vari­­ance

 
Hi Angus,

I have to look at Ana's template and see....

Here are my results from SAS. I need to double check too.

s2wr      s2wt      s2wi      s2D         95% UCB for s2D
0.19931   0.11654   0.16590   .007971169   0.054627


John
AngusMcLean
★★  

USA,
2014-12-23 00:26
(3383 d 10:41 ago)

@ jag009
Posting: # 14146
Views: 11,192
 

 Update: S×F vari­­ance

 
John: I can confirm two of them from the RSABE worksheet. Remember WT and WR are transposed in the output.

ILAT = 0.165897= S2WI

dlat Var WR = 0.1165394

I cannot find WT in RSABE WORKSHEET. Also sigma2D and the upper CB are not clear how to obtain.
jag009
★★★

NJ,
2014-12-23 05:43
(3383 d 05:24 ago)

@ AngusMcLean
Posting: # 14147
Views: 11,024
 

 Update: S×F vari­­ance

 
Hi Angus,

❝ John: I can confirm two of them from the RSABE worksheet. Remember WT and WR are transposed in the output.


❝ ILAT = 0.165897= S2WI


❝ dlat Var WR = 0.1165394



❝ I cannot find WT in RSABE WORKSHEET. Also sigma2D and the upper CB are not clear how to obtain.


If I remember correctly (I loaded the template in my office laptop for a min before going home), you can't because Ana's template is designed to analyze RSABE for fully replicate (and partial replicate too?) so it only has workaround for extracting the within-subject variance & SD for reference. You have to generate a workaround to get the within-subject var/SD for test. Yes it can be done :)

John
AngusMcLean
★★  

USA,
2014-12-23 16:24
(3382 d 18:43 ago)

@ jag009
Posting: # 14149
Views: 11,124
 

 Update: S×F vari­­ance

 
John: you can change the code in the AVERAGE BE part of the worksheet and get it to display WT= 0.11739421.

But I think you mean to perform a workaround in the RSABE part of the worksheet.

Right now I do, not see how to do it?

Angus


Edit: Thread closed [image]. Please continue over there. [Helmut]
AngusMcLean
★★  

USA,
2014-12-09 22:03
(3396 d 13:04 ago)

@ jag009
Posting: # 14042
Views: 11,813
 

 Update: S×F vari­­ance test

 
From Helmut's post above to the European document the following narrative is obtained.

This model allows a different subject effect for each formulation (i.e. a subject by formulation interaction), and therefore has 5 variance terms:

(1) Within subject for reference
(2) Within Subject for Test
(3) Between Subject for Test
(4) Between Subject for Reference
(5) Covariance for between subject test and reference

It also states the last three are "combined" to give the subject × formulation interaction variance component.

From Ana’s letter the "combination" involves the addition of (3+4)-5.
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2014-12-09 19:49
(3396 d 15:18 ago)

@ AngusMcLean
Posting: # 14041
Views: 11,762
 

 Coding

 
Hi Angus,

❝ That is why there is transposition in the parameter values.


Your coding is correct.

[image]

It is a mystery why EMA coded sequences ABAB|BABA and formulations T and R.

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
jag009
★★★

NJ,
2014-12-09 23:24
(3396 d 11:43 ago)

@ Helmut
Posting: # 14043
Views: 11,800
 

 working....

 
This is becoming fun...

John
ElMaestro
★★★

Denmark,
2014-12-06 12:12
(3399 d 22:55 ago)

@ jag009
Posting: # 13995
Views: 12,219
 

 Concerta Guidance: S×F vari­­ance test

 
Hi John,

❝ They are referring to σ2D value being no greater than 0.03 right? Can the number be negative, i.e, -0.06?


A covariance cannot be negative. That should be the end of the story, but isn't: it may appear negative due to numerical optimiser phenomena even when REML is used for optimisation. The optimiser should of course warn you about it if you use decent software (given the recent events on the software scene I am not entirely sure how to define decent. Price and quality is seemingly not so well correlated :-D).

Pass or fail!
ElMaestro
AngusMcLean
★★  

USA,
2014-12-07 15:32
(3398 d 19:35 ago)

@ ElMaestro
Posting: # 14000
Views: 12,023
 

 Concerta Guidance: S×F vari­­ance test

 
IF the variance appears to be negative then that that would mean that the corresponding standard deviation is an imaginary number?
ElMaestro
★★★

Denmark,
2014-12-07 19:08
(3398 d 15:59 ago)

@ AngusMcLean
Posting: # 14003
Views: 12,166
 

 Concerta Guidance: S×F vari­­ance test

 
Hi Angus,

❝ IF the variance appears to be negative then that that would mean that the corresponding standard deviation is an imaginary number?


It's an anomaly. A little bit like when type III Sums of Squares for an anova add to a number larger than the total variation. You know that theory says it really can't be like that but it just appears so for purely technical reasons. But at the end of the day you still know the total variation is not the sum of the factors you just computed. You still report them, though. We live with the fact that we cannot account for everything so it isn't imaginary when we get such results but in some sense they just don't make much sense. I have not ever come across a negative SxF term, I think. Others?

Pass or fail!
ElMaestro
AngusMcLean
★★  

USA,
2014-12-08 00:13
(3398 d 10:54 ago)

@ ElMaestro
Posting: # 14006
Views: 12,101
 

 Concerta Guidance: S×F vari­­ance test

 
Perhaps we expect perfection from ANOVA and our expectations need to be modulated for imperfections in the applications of the model. Please allow me to give an example of a more practical situation. When I worked in the U.K I was involved in drug metabolism i.e. administration of carbon-labeled drugs to patients. One of the goals, as you may well imagine, was to account for all of the radio-labeled drug administered. Now if you accounted for 90% excreted (and or exhaled) the regulatory authority would say well where is the rest of the drug? In cases where we know the drug is not bonded to tissue then we not be perfect recovering 90%.

Another way of looking at such situations is to evaluate the the weight of the captain of a ship.
  1. Fist weigh the captain plus the ship.
  2. Then ask the captain to step off the ship.
  3. Reweigh the ship.
  4. The difference in weight provides you with the weight of the captain.
ElMaestro
★★★

Denmark,
2014-12-08 13:25
(3397 d 21:42 ago)

@ AngusMcLean
Posting: # 14008
Views: 11,992
 

 Concerta Guidance: S×F vari­­ance test

 
Hi Angus,

what does it mean when a CoA specifies that the white powder consists of 101.54% Schützomycin?

Pass or fail!
ElMaestro
nobody
nothing

2014-12-08 14:44
(3397 d 20:23 ago)

@ ElMaestro
Posting: # 14009
Views: 12,031
 

 Concerta Guidance: S×F vari­­ance test

 
Edith:
(...forget it, I didn't get the joke... :-D)

Kindest regards, nobody
AngusMcLean
★★  

USA,
2014-12-08 15:03
(3397 d 20:04 ago)

@ ElMaestro
Posting: # 14011
Views: 11,944
 

 Concerta Guidance: S×F vari­­ance test

 
it means that the precision of the assay need to be checked against the reference standard.
nobody
nothing

2014-12-08 15:32
(3397 d 19:35 ago)

@ AngusMcLean
Posting: # 14013
Views: 11,934
 

 Concerta Guidance:S×F vari­­ance test

 
Not really, it's 100% normal result for pharmaceutical quality. Pharmacop. normally requires 95% to 105% content of the respective compound...

Kindest regards, nobody
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,639 registered users;
69 visitors (0 registered, 69 guests [including 7 identified bots]).
Forum time: 11:07 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