Bioequivalence and Bioavailability Forum

vijay
☆

2009-04-21 04:03
(5847 d 04:37 ago)

Posting: # 3571
Views: 8,792

SAS: LSMEANS and ESTIMATE options [Software]

Post reply

Hello All,
I have a few questions regarding the implementation and interpretation of PROC MIXED.

What is the default multiple pairwise comparison adjustment used in PROC MIXED when we specify "LSMEANS TRT/pdiff cl" where we have more than 2 treatments? The SAS manual says that there is a default adjustment of all pairwise differences, but does not state what it is. So, I am assuming that confidence intervals produced in output "Difference of least square means" is based on some pairwise adjustment (This is not labeled ). But let's say my LSMEANS statement is “LSMEANS TRT/pdiff cl ADJUST=BON", which is bonferroni adjustment, the output states that adjustment has been made and gives both the adjusted and unadjusted values for the confidence intervals. As such is there a recommended procedure when more than 2 treatments are used.
Is it redundant to use the ESTIMATE statement with contrasts, "ESTIMATE 't-r' -1 1/cl alpha=0.1' and 'LSMEANS TRT/pdiff cl alpha=0.1' in the same model. I notice that the point estimates and it's corresponding confidence intervals at the same alpha level given in the ESTIMATE part of the output for the contrast is similar to "Difference of LSMEANS" output unless a specific ADJUST= option is provided in the LSMEANS statement.
I am frequently getting almost similar estimates for the standard errors of the point estimates in both the ESTIMATE and LSMEANS output. The SAS manual states, "The approximate standard errors for the LS-mean is computed as the square root of L(X'V^-1X)^-L'." I am assuming that the similar standard errors for the point estimates has something to do with the specification of the L matrix given that the X comes from the data. Please correct me if I am wrong. (One can find such similar estimates of SE in all outputs in chapter 4 of "Bioequivalence and Statistics in Clinical Pharmacology by Patterson and Jones").
Finally, I may be totally naïve in asking this question but I still need to clarify . Does one use the actual AUC and Cmax parameters or the dose normalized AUC and Cmax parameters while assessing bioequivalence. Is there any difference? (I should have checked myself, just realized while writing this ). If we consider a scenario where a lesser dose of a new formulation (say 7.5 mg) is being tested against an old formulation (say 10 mg), then we do not need to dose normalize on the presumption that the new formulation is designed to perform as good as the old formulation but with a lesser dose due to better bioavailability.

Thanks in advance.
Vijay

d_labes
★★★

Berlin, Germany,
2009-04-21 10:58
(5846 d 21:43 ago)

@ vijay
Posting: # 3572
Views: 7,543

ADJUST does not adjust

Post reply

Dear Vijay,

❝ 1. What is the default multiple pairwise comparison adjustment used in PROC MIXED when we specify “LSMEANS TRT/pdiff cl” where we have more than 2 treatments? The SAS manual says that there is a default adjustment of all pairwise differences, but does not state what it is. [...]

The entry in the SAS manual is totally misleading here :-(

. Proc MIXED uses the same default as Proc GLM namely ADJUST=T, which is an oxymoron because this option does not apply any multiplicity adjustment but merely calculates each pairwise confidence intervals using the appropriate quantiles of the student's t-distribution with unadjusted alpha.

Because the ESTIMATE ... statement handles only one contrast at a time, the confidence intervals are also based on t-distribution without any multiplicity adjustment.

❝ 2. Is it redundant to use the ESTIMATE statement with contrasts, “ESTIMATE ‘t-r’ -1 1/cl alpha=0.1’ and ‘LSMEANS TRT/pdiff cl alpha=0.1’ in the same model. [...]

From the above said it is redundant (or better one of both is enough) unless you specify any adjustment for multiplicity. In that case it is also superfluous to use ESTIMATE ... and LSMEANS ... together because then you are interested in the multiplicity adjusted results and not in the pairwise unadjusted.

For your example statements it is always redundant, because for studies with 2 treatments any adjustment method degenerates to the case without adjustment.

❝ 3. I am frequently getting almost similar estimates for the standard errors of the point estimates in both the ESTIMATE and LSMEANS output.

If not, your SAS installation is in error :-P

.
Provided your ESTIMATE ... statement defines the same contrast as the LSMEANS difference.

Point 4. I will leave over for others. It is another pair of shoes.
But SEARCH! There are some threads in this Forum dealing with dose / potency correction.

—
Regards,

Detlew

vijay
☆

2009-04-21 18:10
(5846 d 14:30 ago)

@ d_labes
Posting: # 3577
Views: 7,332

ADJUST does not adjust

Post reply

Dear d_labes,
Thanks for the clarification. I would like to rephrase question 3 as it seems I did not get the correct message out.
Considering I am using only the LSMEANS option for differences between more than 2 treatments (lets say: A-B, B-C & A-C) then, "I often see similar standard errors for each of those differences even though their point estimates may be very different" I am citing an example output from chapter 4 of “Bioequivalence and Statistics in Clinical Pharmacology by Patterson and Jones”); Example 4.5 for 3 treatments


         Differences of Least Squares Means



                                              Standard

Effect     formula    _formula    Estimate       Error      

formula    R          S            -0.3410     0.03851     

formula    R          T            -0.1497     0.03849     

formula    S          T             0.1912     0.03871

I removed the last few columns of the code, but the output comes from this statement;
lsmeans formula/pdiff cl alpha=0.1;.

This is at least one case where the standard errors differ in the 3rd decimal place, but more than often I see the same standard errors. I am just curious as to why this happens?

Regarding the last point, Thanks HS. I will search more before I get back.

Thanks,
Vijay

d_labes ★★★ Berlin, Germany, 2009-04-21 18:18 (5846 d 14:22 ago) @ vijay Posting: # 3579 Views: 7,490	Missingness missed Post reply
	Dear Vijay, example 4.5 in the Patterson and Jones book has a lot of missing values in it. These are the source for your observation. — Regards, Detlew

Helmut
★★★

Vienna, Austria,
2009-04-21 16:06
(5846 d 16:35 ago)

@ vijay
Posting: # 3576
Views: 8,483

Suprabioavailability

Post reply

Dear Vijay!

❝ 4. Finally, I may be totally naïve in asking this question but I still need to clarify :-) . Does one use the actual AUC and Cmax parameters or the dose normalized AUC and Cmax parameters while assessing bioequivalence.

The former.

❝ Is there any difference?

The definition of bioequivalence (for different wordings see the Guidelines) includes the same molar dose giving the same rate and extent of bioavailability.

❝ If we consider a scenario where a lesser dose of a new formulation (say 7.5 mg) is being tested against an old formulation (say 10mg), then we do not need to dose normalize on the presumption that the new formulation is designed to perform as good as theold formulation but with a lesser dose due to better bioavailability.

Wonderful, that’s an improvement! The technical term is “Suprabioavailability” (see the EMEA’s NfG Section 5.5). If you succeed, most likely you have to run additionally a (small) clinical study to show safety/efficacy as well. IIRC, an example was glibenclamide, where a formulation of 3.5 mg with micronized particles gave a similar BA as the conventional formulation with 5 mg.

—
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

SAS: LSMEANS and ESTIMATE options [Software]

ADJUST does not adjust

ADJUST does not adjust

Missingness missed

Suprabioavailability