Bioequivalence and Bioavailability Forum

jag009
★★★

NJ,
2024-10-14 03:44
(58 d 11:32 ago)

Posting: # 24225
Views: 1,394

Question about computing 90% CI w data from 2 separate studies [General Statistics]

Hi all,
I would like to do a meta BE analysis with treatment datasets (concentration vs time) from 2 separate BE studies that were conducted at different locations and at different time. The sample sizes are not the same (one is 36 and the other is 48). Both datasets (A from study 1, and B from study 2) have the same sampling time scheme and the same number of timepoints. For this analysis I would like to calculate the ratio of geometric means and 90% Geo CIs for Cmax and AUCs in SAS using a parallel study BE design setup. Can someone let me know if the following is correct? Do I need to check for Center and Center by treatment interaction as well?

  proc mixed data=PKEST;

  by paramn param;  /* Param is AUCs, or Cmax */

  class subject seq period trt center;

  model logval= seq period trt center center*trt/ddfm=satterth;

  random subject(seq); /* should this collapse to "random subject" since technically it would be single sequence for parallel design? */

  estimate  " A - C " TRT 1 -1 /cl alpha=0.1;

  lsmeans TRT /cl alpha=0.1; /* alpha is 0.1 since this is a parallel design comparison */

  ods output lsmeans=lsmeans;

  ods output covparms=covparms;

  ods output estimates=ests;

  run;

Thx
J

dshah
★★

India,
2024-10-14 10:38
(58 d 04:37 ago)

@ jag009
Posting: # 24226
Views: 1,238

Question about computing 90% CI w data from 2 separate studies

Post reply

Hi J!

Apologies but a further question on your question-

❝ Both datasets (A from study 1, and B from study 2) have the same sampling time scheme and the same number of timepoints. For this analysis I would like to calculate the ratio of geometric means and 90% Geo CIs for Cmax and AUCs in SAS using a parallel study BE design setup.

Are the reference and test lot same in both the studies? If no- then what is the purpose of further statistics? If yes- why was there a difference in Ratio if any?

Divyen

jag009 ★★★ NJ, 2024-10-14 17:40 (57 d 21:36 ago) @ dshah Posting: # 24229 Views: 1,186	Question about computing 90% CI w data from 2 separate studies Post reply
	Hi Divyen, ❝ Are the reference and test lot same in both the studies? If no- then what is the purpose of further statistics? If yes- why was there a difference in Ratio if any? Both ref and test lots are different between the two studies. Thx J

BEQool
★

2024-10-14 13:37
(58 d 01:39 ago)

@ jag009
Posting: # 24228
Views: 1,211

Question about computing 90% CI w data from 2 separate studies

Post reply

Hello J,

I would just use Treatment (or Center) as a factor in the model (as normally for parallel designs). I think that it doesnt matter what you use (Treatment or Center) because they are correlated - so all subject that got Treatment A are in one Center and all subjects that got Treatment C are in the other Center. So Treatment effect is the same as Center effect and you cannot distinguish them. They are confounded.
Similarly, you cannot get Center*Tretament interaction.
I doubt that your model even gives you anything? Do you get any results?

Hello Divyen,

❝ Are the reference and test lot same in both the studies? If no- then what is the purpose of further statistics? If yes- why was there a difference in Ratio if any?

Just courious, do you most of the time get the same Ratio if you repeat the study with the same lot of Test and Reference? Additionally, what is for you a "different" or the "same" ratio"? How small/big does the difference has to be that you say there is a difference in ratio? :-)

Regards
BEQool

jag009
★★★

NJ,
2024-10-14 17:45
(57 d 21:31 ago)

@ BEQool
Posting: # 24230
Views: 1,173

Question about computing 90% CI w data from 2 separate studies

Post reply

Hi BEQool,

>❝ I doubt that your model even gives you anything? Do you get any results?

Oh, I got a bunch of non-estimables because there were not enough degrees of freedom left. Obviously, I had more terms in the model than degrees of freedom. I think I got only the LSMeans of the 2 treatments, but nothing else could be estimated.

Thanks again for your feedback. I am going to try the following codes later today. What do you think?

PROC MIXED data=BEdata;

CLASSES SUBJECT TRT;

MODEL logval = TRT/ DDFM=SATTERTH;

RANDOM TRT/TYPE=FA0(2) SUB=SUBJECT G;

REPEATED/GRP=TRT SUB=SUBJECT;

ESTIMATE 'T vs. R' TRT 1 -1/CL ALPHA=0.1;

lsmeans TRT/ cl alpha=0.1; * <-- for parallel design studies, alpha=0.1;

ods output lsmeans=lsmeans;         * Least Squares means Output;

ods output covparms=covparms;       * Covariance output;

ods output estimates=ests;          * Estimate of differences between treatments;

Thx
J

BEQool
★

2024-10-18 14:03
(54 d 01:13 ago)

@ jag009
Posting: # 24240
Views: 938

Question about computing 90% CI w data from 2 separate studies

Post reply

Hello J,

❝ I am going to try the following codes later today.

Did you try it? How did it go?

❝ What do you think?

❝ PROC MIXED data=BEdata;

❝ CLASSES SUBJECT TRT;

❝ MODEL logval = TRT/ DDFM=SATTERTH;

❝ RANDOM TRT/TYPE=FA0(2) SUB=SUBJECT G;

❝ REPEATED/GRP=TRT SUB=SUBJECT;

❝ ESTIMATE 'T vs. R' TRT 1 -1/CL ALPHA=0.1;

❝ lsmeans TRT/ cl alpha=0.1; * <-- for parallel design studies, alpha=0.1;

❝ ods output lsmeans=lsmeans; * Least Squares means Output;

❝ ods output covparms=covparms; * Covariance output;

❝ ods output estimates=ests; * Estimate of differences between treatments;

I dont use SAS a lot and am therefore not an experienced user but I think you dont need statements in blue because you have a parallel design with just one factor Treatment.

And if you want to estimate T vs. R and have it coded like that, you should switch "1 -1" to "-1 1", otherwise with code "1 -1" you will get R vs. T and not T vs. R.
If you have it coded A vs. C (like in your first post) then "1 -1" is correct (it depends on the alphabetical order).

Regards
BEQool

dshah
★★

India,
2024-10-15 08:11
(57 d 07:05 ago)

@ BEQool
Posting: # 24232
Views: 1,103

Question about computing 90% CI w data from 2 separate studies

Post reply

Hi BEQool!

❝ Just courious, do you most of the time get the same Ratio if you repeat the study with the same lot of Test and Reference? Additionally, what is for you a "different" or the "same" ratio"? How small/big does the difference has to be that you say there is a difference in ratio? :-)

The reason for asking the ratio was- I am not getting purpose of such exercise. If power is insufficient, then easiest is bootstrapping or modelling. Generally, the ratio should not change much if the power of study was good. So asked for difference in ratio of two studies.

Divyen

Question about computing 90% CI w data from 2 separate studies [General Sta­tis­tics]

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies

Question about computing 90% CI w data from 2 separate studies [General Statistics]