Bioequivalence and Bioavailability Forum • SAS Procedure on Partial Replicate (TRR) Study Data

jag009
★★★

NJ,
2013-04-02 23:11
(4424 d 23:45 ago)

Posting: # 10332
Views: 11,468

SAS Procedure on Partial Replicate (TRR) Study Data [RSABE / ABEL]

Hi all,

It's a dumb question but I might as well ask..

According to the FDA progesterone guidance, the 3-sequences for a 3-way partial replicate study (TRR) is:
TRR
RTR
RRT

I was reviewing an old study report and found out that the CRO chose the following sequences for a 3-way partial rep study; A=Test, B=Ref. I dont know why they did this :confused:

but it was before my time here.
BAB (RTR)
ABB (TRR)
BBA (RRT)

Question. Would scale average BE (SCABE) computation in SAS yield different results if the period and sequence identifier for Ref1 and Ref2 are messed up (ie: CRO inadvertently used the FDA sequences for stats but used their own sequences for dosing)? Why I ask? See below...

I set up my SAS code as per FDA's example and assigned what is Ref1 and Ref2 as per CRO's BAB, ABB, BBA randomization sequences. I compared my PROC GLM SAS output for dlat to that generated by CROs and the outcome was different. The results for SCABE (theta, sWR, Critbound, Ratio) did not match those generated by the CRO.

Now comes the interesting part. When I set up my code as per FDA, exactly AS IS, meaning even the sequence and period assignment for Ref1 and Ref2 (meaning following TRR, RTR, RRT instead of CRO's BAB, ABB, BBA), I ended up getting the same SCABE results (theta, sWR, Critbound, Ratio) as the CRO's. :angry:

Thanks
John

Edit: Category changed. [Helmut]

Helmut
★★★

Vienna, Austria,
2013-04-03 15:42
(4424 d 07:14 ago)

@ jag009
Posting: # 10333
Views: 9,754

Confusing…

Post reply

Hi John!

❝ According to the FDA progesterone guidance, the 3-sequences for a 3-way partial replicate study (TRR) is:

❝ TRR

❝ RTR

❝ RRT

❝ […] the CRO chose the following sequences for a 3-way partial rep study; A=Test, B=Ref. A=Test, B=Ref. I dont know why they did this

❝ BAB (RTR)

❝ ABB (TRR)

❝ BBA (RRT)

Are you trying to confuse me? The sequences are the same; the order is irrelevant. Are you not happy with the “A” and “B”? Such a coding would also work:
hocus pocus pocus
pocus hocus pocus
pocus pocus hocus

I guess the sequences were randomized. By chance it might be that the first three subjects had sequences TRR|TRR|TRR. As long as you had ⅓ of subjects allocated to each of the three sequences no problem.

❝ Would scale average BE (SCABE) computation in SAS yield different results if the period and sequence identifier for Ref1 and Ref2 are messed up (ie: CRO inadvertently used the FDA sequences for stats but used their own sequences for dosing)?

Duno what you mean by “FDA sequences”. Of course any administration different from the randomization used in the evaluation would screw up everything. Imagine you have a 2×2×2 cross-over in two (!) subjects. They were dosed (1) RT and (2) TR. Now you assume the sequences were TR and RT. What will you get? ;-)

Can you post the sequences of e.g., the first twelve subjects?

—
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

jag009
★★★

NJ,
2013-04-03 18:06
(4424 d 04:50 ago)

@ Helmut
Posting: # 10335
Views: 9,860

Confusing…

Post reply

Hi Helmut,

What I was trying to say is:

FDA guidance shows:
Sequence 1=TRR
Sequence 2=RTR
Sequence 3=RRT

CRO used the following:
Sequence 1=BAB (RTR)
Sequence 2=ABB (TRR)
Sequence 3=BBA (RRT)

When you look at the SAS code from the Progesterone guidance for 3-way partial replicate. The section where you generate 3 datasets, 1 for test, 1 for ref1 and 1 for ref2.

Dataset containing TEST observations:
data test;
set pk;
if trt='T';
latt=lauct;
run;
Dataset containing REFERENCE 1 observations:
data ref1;
set ref;
* if (seq=1 and per=2) or (seq=2 and per=1) or (seq=3 and per=1); *
lat1r=lauct;
run;
Dataset containing REFERENCE 2 observations:
data ref2;
set ref;
* if (seq=1 and per=3) or (seq=2 and per=3) or (seq=3 and per=2); *
lat2r=lauct;
run;

See the if statements in astericks? Those statements should be written according to the way the sequences were set up (i.e., the aboves were written based on how FDA set up the sequences meaning Sequence 1 = TRR, Sequence 2 = RTR, Sequence 3 = RRT). My suspicion is that the CRO failed to update those statements according to the way they set up their sequences but instead they used the FDA codes (if statements) above. How did I find out? I used the above codes AS IS with the CRO data and I ended up with the same SCABE results. However, when I set up the if statements the way they are supposed to be as per CRO's sequences (Sequence 1 = BAB, Sequence 2 = ABB, Sequence 3 = BBA) then the results were different.

As I go further down the SAS code, the PROC GLM analysis for dlat (below). Would incorrect sequence assignment leads to different results (Perhaps d_labes or ElMaestro can help since they use SAS?)

Intermediate analysis - dlat
proc glm data=scavbe;
class seq;
model dlat=seq;
ods output overallanova=dglm1;
ods output NObs=dglm3;
title1 'scaled average BE';
run;

Can I post the data up (of course no labels)? 72 subjects though. Just Cmax.

I hope I am wrong but I can't confirm so need a 2nd pair of hands. :-)

Thanks
John

Helmut
★★★

Vienna, Austria,
2013-04-03 23:49
(4423 d 23:07 ago)

@ jag009
Posting: # 10336
Views: 9,730

Confusing…

Post reply

Hi John!

❝ My suspicion is that the CRO failed to update those statements according to the way they set up their sequences but instead they used the FDA codes (if statements) above.

Wonderful.
Personally I never code sequences with numbers but characters instead. The code is much easier to check. Pseudocode:

R₁: (seq='TRR' ∧ per=2) ∨ (seq='RTR' ∧ per=1) ∨ (seq='RRT' ∧ per=1)

R₂: (seq='TRR' ∧ per=3) ∨ (seq='RTR' ∧ per=3) ∨ (seq='RRT' ∧ per=2)

❝ Can I post the data up (of course no labels)? 72 subjects though. Just Cmax.

Sure. Separate columns by single blanks. Columns (any order you like): subject, sequence, period, treatment, Cmax. If you want to do me a favor code sequences with three character strings (not 1, 2, 3).

—
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

jag009
★★★

NJ,
2013-04-04 01:01
(4423 d 21:55 ago)

@ Helmut
Posting: # 10337
Views: 9,971

Confusing…

Post reply

Thanks Helmut.

Hope this is the way you wanted (space delimited). I swapped the Sequence# 1,2,3 as A,B,C. Therefore sequence A=BAB (or RTR), B=ABB (or TRR), C=BBA (or RRT). A=Test, B=Reference. The columns are Subject, Period, Drug, Sequence, LnCmax. There are two missing values "."


 1 1 B B 7.423568444

 1 2 A B 7.70210434

 1 3 B B 6.964135612

 2 1 A A 8.009363077

 2 2 B A 7.408530567

 2 3 B A 6.631474629

 3 1 B B 7.930206207

 3 2 A B 8.327726166

 3 3 B B 7.503840747

 4 1 A A 7.668093709

 4 2 B A 8.233237501

 4 3 B A 6.741110202

 5 1 B C 7.409741954

 5 2 B C 8.814033202

 5 3 A C 8.019612794

 6 1 B C 7.605392365

 6 2 B C 7.402451521

 6 3 A C 7.733245647

 7 1 B B 6.800504088

 7 2 A B 7.893945138

 7 3 B B 7.724004657

 8 1 B B 6.909753282

 8 2 A B 8.226305988

 8 3 B B 7.34536484

 9 1 A A 8.00836557

 9 2 B A 7.954021087

 9 3 B A 7.832014181

10 1 B C 6.833462674

10 2 B C 6.521651524

10 3 A C 7.372746366

11 1 B C 7.817222786

11 2 B C 7.221105098

11 3 A C 8.280204233

12 1 A A 7.319202459

12 2 B A 6.36233821

12 3 B A 7.377133713

13 1 B C 8.002359546

13 2 B C 8.058960018

13 3 A C 8.404024493

14 1 A A 7.806696373

14 2 B A 6.813884063

14 3 B A 6.898613621

15 1 B C 8.218517577

15 2 B C 8.068089626

15 3 A C 8.405591015

16 1 B B 7.705262475

16 2 A B 8.090708716

16 3 B B 7.864804003

17 1 A A 7.369600721

17 2 B A 7.126890809

17 3 B A 7.305860033

18 1 B B 7.00033446

18 2 A B 7.805474625

18 3 B B 7.320526962

19 1 B B 7.037027615

19 2 A B 7.393263095

19 3 B B 7.389563954

20 1 B B 6.326864781

20 2 A B 8.18729927

20 3 B B 6.245331271

21 1 B C 8.106212903

21 2 B C 7.37337431

21 3 A C 7.910590612

22 1 B C 6.941190055

22 2 B C 7.062191632

22 3 A C 8.666130304

23 1 A A 7.898782357

23 2 B A 8.057377489

23 3 B A 7.982416347

24 1 A A 6.797270546

24 2 B A 6.626850191

24 3 B A 6.817721156

25 1 B C 8.13973228

25 2 B C 7.522400231

25 3 A C 8.25997566

26 1 B B 6.91274282

26 2 A B 7.453561872

26 3 B B 7.062191632

27 1 A A 5.835979644

27 2 B A 5.691709893

27 3 B A 6.21080086

28 1 A A 7.286191715

28 2 B A 7.444248649

28 3 B A 8.540128163

29 1 B C 7.884952946

29 2 B C 7.753623547

29 3 A C 8.418918622

30 1 B B 7.774015077

30 2 A B 7.513709248

30 3 B B 7.582229194

32 1 B B 6.659678461

32 2 A B 7.832807517

32 3 B B 7.512617545

33 1 B C 8.567315801

33 2 B C 7.808729307

33 3 A C 8.350902452

34 1 A A 7.54009032

34 2 B A 6.455041318

34 3 B A 6.576051654

35 1 A A 7.796880343

35 2 B A 6.757048662

35 3 B A 7.252762418

36 1 B B 7.18690102

36 2 A B 7.440733707

36 3 B B 6.91671502

37 1 A A 7.586296307

37 2 B A 7.557994959

37 3 B A .

38 1 B C 6.317706056

38 2 B C 7.266827348

38 3 A C 7.754052639

39 1 B C 6.161840033

39 2 B C 6.464432528

39 3 A C 7.03438793

40 1 B B 7.134093721

40 2 A B 7.966239777

40 3 B B 7.983440063

41 1 A A 7.01301579

41 2 B A 6.858144778

41 3 B A 6.629759476

42 1 B B 7.381501895

42 2 A B 7.994632311

42 3 B B .

43 1 A A 7.183111702

43 2 B A 7.733245647

43 3 B A 7.897668151

44 1 B B 7.018401799

44 2 A B 8.058960018

44 3 B B 7.17930797

45 1 A A 9.119868469

45 2 B A 7.198183577

45 3 B A 8.538367427

46 1 B C 6.677209398

46 2 B C 6.493148513

46 3 A C 7.244941546

47 1 B B 8.872627366

47 2 A B 8.068402959

47 3 B B 8.886685639

48 1 B C 6.97354302

48 2 B C 7.290974778

48 3 A C 8.349247801

49 1 B C 7.65349491

49 2 B C 7.507690078

49 3 A C 7.670428522

50 1 A A 7.463363046

50 2 B A 6.730062991

50 3 B A 7.189922171

51 1 B B 6.254789888

51 2 A B 6.856882949

51 3 B B 6.899118086

52 1 A A 7.357556201

52 2 B A 7.107425474

52 3 B A 7.737180078

53 1 B C 7.99530662

53 2 B C 8.697178688

53 3 A C 8.286017468

54 1 B B 8.086102536

54 2 A B 7.163172391

54 3 B B 8.051659557

55 1 B B 7.687080156

55 2 A B 8.158516245

55 3 B B 6.856567242

56 1 B B 6.683986532

56 2 A B 8.06022424

56 3 B B 6.187647917

57 1 B C 7.03438793

57 2 B C 7.823645931

57 3 A C 7.23489842

58 1 A A 7.437206367

58 2 B A 7.002155954

58 3 B A 7.237059026

59 1 A A 9.036344064

59 2 B A 7.326465614

59 3 B A 8.185350223

60 1 B C 5.927725706

60 2 B C 5.966659428

60 3 A C 6.293049525

61 1 B B 7.848933726

61 2 A B 8.313361951

61 3 B B 7.973844376

62 1 B C 8.921857979

62 2 B C 8.868694777

62 3 A C 8.620291149

63 1 A A 7.528331767

63 2 B A 7.434847875

63 3 B A 7.440733707

64 1 B C 7.643482907

64 2 B C 7.878534196

64 3 A C 7.242797923

65 1 A A 8.339261983

65 2 B A 7.467371067

65 3 B A 7.740229525

66 1 B B 6.801616683

66 2 A B 7.126890809

66 3 B B 7.836369761

67 1 B B 6.373149127

67 2 A B 7.858254182

67 3 B B 6.900529234

68 1 B B 6.593455409

68 2 A B 7.203405521

68 3 B B 6.814652661

69 1 A A 6.849914279

69 2 B A 6.321127942

69 3 B A 6.502190166

70 1 B C 6.58271742

70 2 B C 8.02289687

70 3 A C 7.681099002

71 1 A A 8.004699511

71 2 B A 6.869637957

71 3 B A 7.244941546

72 1 B C 8.698514248

72 2 B C 8.883917471

72 3 A C 7.960323629

Thanks
John

Edit: Table sorted by subject and period. [Helmut]

Helmut
★★★

Vienna, Austria,
2013-04-04 01:22
(4423 d 21:34 ago)

@ jag009
Posting: # 10338
Views: 9,692

Confusing…

Post reply

Hi John!

❝ Hope this is the way you wanted (space delimited).

Not quite but I think I can handle that.

❝ I swapped the Sequence# 1,2,3 as A,B,C. Therefore sequence A=BAB (or RTR), B=ABB (or TRR), C=BBA (or RRT). A=Test, B=Reference. The columns are Subject, Period, Drug, Sequence, LnCmax. There are two missing values "."

Your coding somehow still matches the post’s subject line.*

1 1 B B 7.423568444

1 2 A B 7.70210434

1 3 B B 6.964135612

B B? Is that the reference twice? Oh no, it codes for treatment R in the 3^rd column and sequence TRR in the 4^th …

For clarity – and substituting treatments A→T, B→R and sequences A→RTR, B→TRR, C→RRT – I would have coded somefink like:

1 1 R TRR 7.423568444

1 2 T TRR 7.70210434

1 3 R TRR 6.964135612

With such a code you do not need to look at the head of the table in order to understand what is going on in each row. Much easier for error snooping …

… but hey,
1 1 R TRR 7.423568444 → R is not in period 1
1 2 T TRR 7.70210434 → T is not in period 2
1 3 R TRR 6.964135612 → OK, R is in period 3

Can you check please?

If I trust the sequences more and recode treatment according to period, e.g.,

1 1 T TRR 7.423568444

1 2 R TRR 7.70210434

1 3 R TRR 6.964135612

I got in Phoenix 6.3:
pointest 0.86899272 (passes; within [0.8, 1.25])
sWR 0.45854843 (≥0.294; perform RSABE. CV_WR 48.37%)
critbound -0.092689149 (≤0: passes RSABE)

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

—
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

jag009
★★★

NJ,
2013-04-04 17:59
(4423 d 04:57 ago)

@ Helmut
Posting: # 10345
Views: 9,686

Confusing…

Post reply

Hi Helmut,

Sorry for the typos. Here is the dataset again. I coded the sequence and treatments as they are (A,B). BAB is sequence 1, ABB is sequence 2, BBA is sequence 3. Subject, period, treatment, sequence, lnCmax. (see my next message). Thanks John

1 1 B BAB 7.42357

1 2 A BAB 7.7021

1 3 B BAB 6.96414

2 1 A ABB 8.00936

2 2 B ABB 7.40853

2 3 B ABB 6.63147

3 1 B BAB 7.93021

3 2 A BAB 8.32773

3 3 B BAB 7.50384

4 1 A ABB 7.66809

4 2 B ABB 8.23324

4 3 B ABB 6.74111

5 1 B BBA 7.40974

5 2 B BBA 8.81403

5 3 A BBA 8.01961

6 1 B BBA 7.60539

6 2 B BBA 7.40245

6 3 A BBA 7.73325

7 1 B BAB 6.8005

7 2 A BAB 7.89395

7 3 B BAB 7.724

8 1 B BAB 6.90975

8 2 A BAB 8.22631

8 3 B BAB 7.34536

9 1 A ABB 8.00837

9 2 B ABB 7.95402

9 3 B ABB 7.83201

10 1 B BBA 6.83346

10 2 B BBA 6.52165

10 3 A BBA 7.37275

11 1 B BBA 7.81722

11 2 B BBA 7.22111

11 3 A BBA 8.2802

12 1 A ABB 7.3192

12 2 B ABB 6.36234

12 3 B ABB 7.37713

13 1 B BBA 8.00236

13 2 B BBA 8.05896

13 3 A BBA 8.40402

14 1 A ABB 7.8067

14 2 B ABB 6.81388

14 3 B ABB 6.89861

15 1 B BBA 8.21852

15 2 B BBA 8.06809

15 3 A BBA 8.40559

16 1 B BAB 7.70526

16 2 A BAB 8.09071

16 3 B BAB 7.8648

17 1 A ABB 7.3696

17 2 B ABB 7.12689

17 3 B ABB 7.30586

18 1 B BAB 7.00033

18 2 A BAB 7.80547

18 3 B BAB 7.32053

19 1 B BAB 7.03703

19 2 A BAB 7.39326

19 3 B BAB 7.38956

20 1 B BAB 6.32686

20 2 A BAB 8.1873

20 3 B BAB 6.24533

21 1 B BBA 8.10621

21 2 B BBA 7.37337

21 3 A BBA 7.91059

22 1 B BBA 6.94119

22 2 B BBA 7.06219

22 3 A BBA 8.66613

23 1 A ABB 7.89878

23 2 B ABB 8.05738

23 3 B ABB 7.98242

24 1 A ABB 6.79727

24 2 B ABB 6.62685

24 3 B ABB 6.81772

25 1 B BBA 8.13973

25 2 B BBA 7.5224

25 3 A BBA 8.25998

26 1 B BAB 6.91274

26 2 A BAB 7.45356

26 3 B BAB 7.06219

27 1 A ABB 5.83598

27 2 B ABB 5.69171

27 3 B ABB 6.2108

28 1 A ABB 7.28619

28 2 B ABB 7.44425

28 3 B ABB 8.54013

29 1 B BBA 7.88495

29 2 B BBA 7.75362

29 3 A BBA 8.41892

30 1 B BAB 7.77402

30 2 A BAB 7.51371

30 3 B BAB 7.58223

32 1 B BAB 6.65968

32 2 A BAB 7.83281

32 3 B BAB 7.51262

33 1 B BBA 8.56732

33 2 B BBA 7.80873

33 3 A BBA 8.3509

34 1 A ABB 7.54009

34 2 B ABB 6.45504

34 3 B ABB 6.57605

35 1 A ABB 7.79688

35 2 B ABB 6.75705

35 3 B ABB 7.25276

36 1 B BAB 7.1869

36 2 A BAB 7.44073

36 3 B BAB 6.91672

37 1 A ABB 7.5863

37 2 B ABB 7.55799

37 3 B ABB .

38 1 B BBA 6.31771

38 2 B BBA 7.26683

38 3 A BBA 7.75405

39 1 B BBA 6.16184

39 2 B BBA 6.46443

39 3 A BBA 7.03439

40 1 B BAB 7.13409

40 2 A BAB 7.96624

40 3 B BAB 7.98344

41 1 A ABB 7.01302

41 2 B ABB 6.85814

41 3 B ABB 6.62976

42 1 B BAB 7.3815

42 2 A BAB 7.99463

42 3 B BAB .

43 1 A ABB 7.18311

43 2 B ABB 7.73325

43 3 B ABB 7.89767

44 1 B BAB 7.0184

44 2 A BAB 8.05896

44 3 B BAB 7.17931

45 1 A ABB 9.11987

45 2 B ABB 7.19818

45 3 B ABB 8.53837

46 1 B BBA 6.67721

46 2 B BBA 6.49315

46 3 A BBA 7.24494

47 1 B BAB 8.87263

47 2 A BAB 8.0684

47 3 B BAB 8.88669

48 1 B BBA 6.97354

48 2 B BBA 7.29097

48 3 A BBA 8.34925

49 1 B BBA 7.65349

49 2 B BBA 7.50769

49 3 A BBA 7.67043

50 1 A ABB 7.46336

50 2 B ABB 6.73006

50 3 B ABB 7.18992

51 1 B BAB 6.25479

51 2 A BAB 6.85688

51 3 B BAB 6.89912

52 1 A ABB 7.35756

52 2 B ABB 7.10743

52 3 B ABB 7.73718

53 1 B BBA 7.99531

53 2 B BBA 8.69718

53 3 A BBA 8.28602

54 1 B BAB 8.0861

54 2 A BAB 7.16317

54 3 B BAB 8.05166

55 1 B BAB 7.68708

55 2 A BAB 8.15852

55 3 B BAB 6.85657

56 1 B BAB 6.68399

56 2 A BAB 8.06022

56 3 B BAB 6.18765

57 1 B BBA 7.03439

57 2 B BBA 7.82365

57 3 A BBA 7.2349

58 1 A ABB 7.43721

58 2 B ABB 7.00216

58 3 B ABB 7.23706

59 1 A ABB 9.03634

59 2 B ABB 7.32647

59 3 B ABB 8.18535

60 1 B BBA 5.92773

60 2 B BBA 5.96666

60 3 A BBA 6.29305

61 1 B BAB 7.84893

61 2 A BAB 8.31336

61 3 B BAB 7.97384

62 1 B BBA 8.92186

62 2 B BBA 8.86869

62 3 A BBA 8.62029

63 1 A ABB 7.52833

63 2 B ABB 7.43485

63 3 B ABB 7.44073

64 1 B BBA 7.64348

64 2 B BBA 7.87853

64 3 A BBA 7.2428

65 1 A ABB 8.33926

65 2 B ABB 7.46737

65 3 B ABB 7.74023

66 1 B BAB 6.80162

66 2 A BAB 7.12689

66 3 B BAB 7.83637

67 1 B BAB 6.37315

67 2 A BAB 7.85825

67 3 B BAB 6.90053

68 1 B BAB 6.59346

68 2 A BAB 7.20341

68 3 B BAB 6.81465

69 1 A ABB 6.84991

69 2 B ABB 6.32113

69 3 B ABB 6.50219

70 1 B BBA 6.58272

70 2 B BBA 8.0229

70 3 A BBA 7.6811

71 1 A ABB 8.0047

71 2 B ABB 6.86964

71 3 B ABB 7.24494

72 1 B BBA 8.69851

72 2 B BBA 8.88392

72 3 A BBA 7.96032

jag009
★★★

NJ,
2013-04-04 19:21
(4423 d 03:34 ago)

(edited on 2013-04-04 22:15)
@ Helmut
Posting: # 10346
Views: 9,656

Confusing…

Post reply

Helmut, here is the issue with Dlat analysis between mine and CRO's. The 2 GLM procedure output are different but the Ratio, 95% Upper Bound, SwRs are ok. :confused:

However, my GLM procedure output for ilat analysis shows same results as those from the CRO.

My output.

scaled average BE - analysis of dlat for ln(Cmax)

The GLM Procedure

Dependent Variable: dlat_lcmax

Source          DF  Sum of Squares  Mean Square   F Value  Pr > F

Model            2    0.17972168     0.08986084    0.28    0.7588

Error           66   21.40250828     0.32428043

Corrected Total 68 21.58222996



R-Square Coeff Var  Root MSE  dlat_lcmax Mean

0.008327 -322.9433  0.569456    -0.176333



Source     DF   Type I SS     Mean Square  F Value  Pr > F

sequence   2    0.17972168    0.08986084    0.28    0.7588

Source     DF   Type III SS   Mean Square  F Value  Pr > F

sequence   2    0.17972168    0.08986084    0.28    0.7588



Theta   Within Subject   Within Subject   Pt Est.   95% Uppder 

          Var of Ref      SD of Ref        (T/R)    Conf. Bnd

0.79669    0.16214         0.40267        150.848     0.15460

CRO output

scaled average BE - analysis of dlat for ln(Cmax)

The GLM Procedure

Dependent Variable: dlat_lcmax

Source            DF    Sum of Squares   Mean Square  F Value  Pr > F

Model              2      2.06868456      1.03434228    3.19   0.0476

Error             66     21.40251551      0.32428054

Corrected Total   68     23.47120008



R-Square  Coeff Var   Root MSE   dlat_lcmax Mean

0.088137  -933.9983   0.569456   -0.060970



Source   DF   Type I SS     Mean Square  F Value   Pr > F

SEQ      2    2.06868456    1.03434228     3.19    0.0476

Source   DF   Type III SS   Mean Square  F Value   Pr > F

SEQ      2    2.06868456    1.03434228     3.19    0.0476



Obs   RATIO   theta        y       boundy    sWR      critbound

1    150.848  0.79669  -0.12918 -0.099175   0.40267    0.15460

Helmut
★★★

Vienna, Austria,
2013-04-05 03:01
(4422 d 19:55 ago)

@ jag009
Posting: # 10348
Views: 9,486

2nd opinion

Post reply

Hi John,

❝ The 2 GLM procedure output are different but the Ratio, 95% Upper Bound, SwRs are ok. :confused: However, my GLM procedure output for ilat analysis shows same results as those from the CRO.

OK, lacking [image]

that’s not my cup of tea. ;-)

I updated treatments and sequences in my Phoenix project (i.e., kept the full precision of your first dataset) and got:
pointest 1.5084845
sWR 0.40266645 (CV_WR 41.96%)
critbound 0.15460335
Bad luck. Reformulate since in the unscaled analysis the lower 90% CL of 134.02% > 125.00% – demonstrating bioinequivalence.

I would opt for your analysis. Like you I got an estimate for dlat of -0.17633358 and a residual SS of 21.402516. However, these values are not further used (only the residual variance 0.32428054).

—
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

jag009
★★★

NJ,
2013-04-05 17:24
(4422 d 05:32 ago)

@ Helmut
Posting: # 10353
Views: 9,466

2nd opinion

Post reply

Hi Helmut!

❝ Bad luck. Reformulate since in the unscaled analysis the lower 90% CL of 134.02% > 125.00% – demonstrating bioinequivalence.

Yes the study was a dud :-P

I was trying to use the dataset and results to validate my SAS code. I already validated my code with another study but was trying to validate with this one again just to kill time. I didn't know I would end up with this turd from the CRO.

So my computation is correct then ;-)

I still don't understand why they didn't use the FDA sequence as is (1=TRR, 2=RTR, 3=RRT) instead of using their own (1=BAB, 2=ABB, 3=BBA). Oh well.

Thank v much for your help.

John

P.S. Did you have a good time with my professor laszlo Endrenyi during your last seminar/conference? Great guy.

d_labes
★★★

Berlin, Germany,
2013-04-05 13:04
(4422 d 09:52 ago)

@ jag009
Posting: # 10349
Views: 9,556

3rd, 4th opinion

Post reply

Dear John!

I have checked your results with a different coding approach avoiding all the bells and whistles with separating the values for Test and Ref., coding sequences numeric and getting the replicates by such monsters like

 ...

 if (seq=1 and per=3) or (seq=2 and per=3) or (seq=3 and per=2);

...

which depend naturally on the sequence numbering and is error prone.

Here my code, taking your first posted data:

Data John;

input subject period tmtAB $ seqABC $ logval;

datalines;

 1 1 B B 7.423568444

 1 2 A B 7.70210434

 1 3 B B 6.964135612

...

;

run;

* recode. not really necessary but common notation;

data john;

  set john;

  if tmtAB='A' then tmt='T';

  if tmtAB='B' then tmt='R';

  if seqABC='A' then sequence='TRR';

  if seqABC='B' then sequence='RTR';

  if seqABC='C' then sequence='RRT';

  if seqABC='A' then seq=1;

  if seqABC='B' then seq=2;

  if seqABC='C' then seq=3;

run;

* replicate no. this is the KEY step;

Proc sort data=john; by subject tmtAB period; run;

data john;

  set john;

  by subject tmtAB;

  if first.tmtAB then repl=1;

  else repl+1;

run;

data john;

  set john;

  tmtrepl=tmt||put(repl,1.);

run;

* make the wide form with columns T1 R1 R2;

Proc transpose data=john out=john_t;

  by subject sequence seq;

  var logval;

  id tmtrepl;

run;

* intra-subject contrasts;

Data john_t;

  set john_t;

* the only place to change in case of full replicate;

* use then TR=0.5*(T1+T2-R1-R2);

  TR=T1-0.5*(R1+R2);

  RR=R1-R2;

  * uncomment next line to get the CRO results;

  * if seq=2 then RR=-RR;

run;

* and now: trara-trara-tra-ra the ANOVA for R-R;

Title GLM analysis of R-R;

Proc GLM data=john_T;

  class sequence;

  model RR = sequence;

run; quit;

Output omitted. It is totally the same as your results :cool:

.

Fortunately the mse's (=2*s2wR) are the same in both evaluations.
You can obtain the results of the CRO if you uncomment the statement in red above. It seems that the use of the progesterone code without considering the different numbering of the sequences has lead to such a calculation.
By luck this contrast doesn't change the mse because nobody will challenge you defining R-R as R2-R1 instead of the usual R1-R2.

The ANOVA for T-R (ilat in the progesterone guidance speak) is left as your homework :-D

. Also the various 'ODS output' statements to save the necessary terms for calculating the linearized scaled ABE criterion and its 95% CI.

BTW: 4th opinion: An R implementation using the same spirit as the SAS code above gives

Analysis of Variance Table



Response: RR

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

sequence   2  0.1797 0.08986  0.2771 0.7588

Residuals 66 21.4025 0.32428

—
Regards,

Detlew

jag009
★★★

NJ,
2013-04-05 17:41
(4422 d 05:15 ago)

@ d_labes
Posting: # 10354
Views: 9,438

3rd, 4th opinion

Post reply

Thank you SAS Guru :-)

Glad to know that I am not the insane one. I can do the Ilat part of the code :-D

Question...

❝ RR=R1-R2;

❝ * uncomment next line to get the CRO results;

❝ * if seq=2 then RR=-RR;

»...

❝ You can obtain the results of the CRO if you uncomment the statement in red above. It seems that the use of the progesterone code without considering the different numbering of the sequences has lead to such a calculation.

❝ By luck this contrast doesn't change the mse because nobody will challenge you defining R-R as R2-R1 instead of the usual R1-R2.

The CRO's sequence assignment(I added the subscripts 1&2 here just for clarity):
#1=B₁AB₂
#2=AB₁B₂
#3=B₁B₂A

Why swap RR as -RR for sequence 2 if the logistic seems to indicate B₁ is the 1st value and B₂ is the 2nd value? It's 9AM here in US so I am not zipping EtOH.

Granted that the study was a dud and the outcome was not affected by that additional line of code in the SAS code...

Thanks
John

d_labes
★★★

Berlin, Germany,
2013-04-05 18:09
(4422 d 04:47 ago)

@ jag009
Posting: # 10356
Views: 9,396

Beer swaped

Post reply

Dear John!

❝ Why swap RR as -RR for sequence 2 if the logistic seems to indicate B₁ is the 1st value and B₂ is the 2nd value? It's 9AM here in US so I am not zipping EtOH.

No no. I'm nor suggesting any swap. The correct way is using R1-R2. I only had in mind to give you a hint where the differences came from.

Just leaving for an evening beer. Thank god it's friday.

Have a nice day.

—
Regards,

Detlew

jag009 ★★★ NJ, 2013-04-05 19:18 (4422 d 03:38 ago) @ d_labes Posting: # 10357 Views: 9,425	Beer swaped Post reply
	Hi Detlew, ❝ No no. I'm nor suggesting any swap. The correct way is using R1-R2. I only had in mind to give you a hint where the differences came from. Thanks for the detective work. Too bad I can't talk to the CRO to find out the real story... Yeah it's about 5 hr from now before our Absinthe gathering here, just kidding John