Log in
Register|
jag009 ★★★ NJ, 2013-12-19 18:41 (4573 d 15:37 ago) (edited on 2013-12-19 20:13) Posting: # 12081 Views: 11,619 |
|
|
Hi guys, It seems like I encountered another problem with FDA's SAS code for calculating 90% CI using the PROC Mixed procedure. The parameters did not qualify for RSABE assessment. This time I don't have any missing data. Sorry Helmut if the data is too long. dataset is space limited. SAS and WinNonlin gurus, please confirm? My SAS output showed: For AUCt: WARNING: Did not converge. For AUCi (I don't think this is a problem(?)): NOTE: Convergence criteria met but final hessian is not positive definite. NOTE: Asymptotic variance matrix of covariance parameter estimates has been found to be singular and a generalized inverse was used. Covariance parameters with zero variance do not contribute to degrees of freedom computed by DDFM=SATTERTH. The fields are: Subject, Sequence, Period, Treatment, AUCt, AUCi Sequence: 1 = AB1B2, 2=B1AB2, 3=B1B2A 1 2 2 1 1986.987 2028.539Thanks John |
|
Helmut ★★★   Vienna, Austria, 2013-12-20 05:13 (4573 d 05:05 ago) @ jag009 Posting: # 12083 Views: 9,998 |
|
|
Hi John, seems to be a pivotal study this time? PHX/WNL6.3: No problem with FA0(2), converged after seven iterations, no warnings. PE 90% CI CVWRTake the CVs with a grain of salt. Exactly the same results with FA0(1). Would you mind trying that in SAS as well?— Dif-tor heh smusma 🖖🏼 Довге життя Україна! ![[image]](https://static.bebac.at/pics/Blue_and_yellow_ribbon_UA.png) Helmut Schütz ![[image]](https://static.bebac.at/img/CC by.png) The quality of responses received is directly proportional to the quality of the question asked. 🚮 Science Quotes |
|
jag009 ★★★ NJ, 2013-12-20 18:12 (4572 d 16:06 ago) @ Helmut Posting: # 12089 Views: 9,901 |
|
|
Hi Helmut, When I switched to FA0(1) then it worked. using FA0(1): AUCt: 96.77% 91.88-101.93% AUCi: 99.71% 95.25-104.38% So same as yours  FA0(2) works for AUCi but no convergence for AUCt. For AUCi, results with FA0(2) was the same as FA0(1). Question is why FA0(2) didn't work?  I am not a saint in statistics yet so I don't know  ❝ seems to be a pivotal study this time? Yes, they didn't listen and went with 3/4 of the subjects (of this current panel) last time, guess what happened... Thanks John |
|
Helmut ★★★ Vienna, Austria, 2013-12-20 18:37 (4572 d 15:41 ago) @ jag009 Posting: # 12091 Views: 10,084 |
|
|
Hi John, ❝ When I switched to FA0(1) then it worked. […] same as yours. ❝ ❝ Question is why FA0(2) didn't work? Overspecified model – see this lengthy thread. I think you have two options:
❝ I am not a saint in statistics yet so I don't know. Me not either. But I’m pretty sure the partial replicate is crap. — 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, 2013-12-20 22:24 (4572 d 11:54 ago) @ Helmut Posting: # 12092 Views: 9,888 |
|
|
Thanks Helmut, Now I wanna see what the CRO comes back with the SAS stat output... John |
|
d_labes ★★★ Berlin, Germany, 2013-12-23 10:18 (4570 d 00:01 ago) @ Helmut Posting: # 12093 Views: 9,785 |
|
|
Dear Helmut! ❝ ... But I’m pretty sure the partial replicate is crap. That's a sort of political incorrect statement about someone who can't answer back .I'm not "pretty sure" if you are right here. In my opinion it is crap that the FDA insists on using the SAS Proc Mixed code for full replicate designs also for the partial replicate design. I always wondered why. When it comes to the scaled ABE evaluation they don't use it but revert to the evaluation of the scaled ABE criterion via intra-subject contrasts. Why not use them also for ABE if the criteria for scaling are not fulfilled? Ok, missings or imbalance may have an impact on the ABE decision if one uses this (method of moments) method. But this also applies if scaling is allowed since the ABE criterion is integral part of the scaled ABE criterion. So why do they insist on using the SAS Proc Mixed code for the ABE decision? Sort of political correctness because it has a long history being published already in 2001 for ABE? And do they really insist? Or are other models (beside FA0(1) we have additionally this one) also acceptable if described properly? As I see this: No one knows at the moment. May be it would help if not only John is asking them. IMHO the partial replicate design itself is not worse or better than others, if evaluated properly and not literally. — Regards, Detlew |
|
Helmut ★★★ Vienna, Austria, 2013-12-23 15:06 (4569 d 19:12 ago) @ d_labes Posting: # 12097 Views: 9,825 |
|
|
Dear Detlew! ❝ ❝ ... But I’m pretty sure the partial replicate is crap. ❝ ❝ That's a sort of political incorrect statement about someone who can't answer back Okay, okay. Though they could. After all guys from the FDA were able to send me private e-mails asking for R-code evaluating replicate design studies.  Agree with all you other points. — 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, 2013-12-20 15:32 (4572 d 18:46 ago) @ jag009 Posting: # 12087 Views: 9,902 |
|
|
Hi John, Generally when convergence issues pop up: Rescale, apply a different minimization Al Gore Rhythm, try to manually select appropriate starter values. But isn't all this due to SAS trying to create a within-s for A in the covariance matrix, which can't be optimized? It might very well be syntax-dependent, but I am not a SAS user so I do not know which syntax to apply to make SAS understand that only a within-s for B makes sense in V. Oh, and on a completely different note: Check AUC-extrapolations of subjs 67, 82. — Pass or fail! ElMaestro |
|
jag009 ★★★ NJ, 2013-12-20 18:16 (4572 d 16:02 ago) @ ElMaestro Posting: # 12090 Views: 9,873 |
|
|
Hi ElMaestro, ❝ But isn't all this due to SAS trying to create a within-s for A in the covariance matrix, which can't be optimized? It might very well be syntax-dependent, but I am not a SAS user so I do not know which syntax to apply to make SAS understand that only a within-s for B makes sense in V. So the SAS code is deficient? FDA should point out the need for adjustment I think. Maybe I am too much of a novice in this aspect... ❝ Oh, and on a completely different note: Check AUC-extrapolations of subjs 67, 82. Yeah I know, the PK data was pulled from the CRO so it was just a snapshot on my behalf for now. They just use WinNonlin to estimate t1/2 and roll out the results. Oh I encountered another sneaky thing with a CRO, on the bioanalytical side. That's another story, maybe after Xmas I will bring it up... John |
|
d_labes ★★★ Berlin, Germany, 2013-12-23 11:14 (4569 d 23:04 ago) @ jag009 Posting: # 12094 Views: 9,988 |
|
|
Dear John! Evaluation for both AUCt and AUCi converged without any change to the FDA code within 6-7 iterations (SAS 9.2). In both cases: NOTE: Convergence criteria met but final hessian is not positive definite.90% CIs exactly the same as Helmut's. BTW: FA0(1) gives astonishing the same 90% CI's as FA0(2), without any (small) deviation. Even the CVwR are identical to the second decimal. — Regards, Detlew |
Ing. Helmut Schütz