Bioequivalence and Bioavailability Forum

Hi again,

❝ Adjusted means are used if the design is imbalanced (in the balanced case, geometric means = adjusted means).

Agreed, adjusted means (LSMeans) introduce another source of confusion. Let's say we have a BE-study unbalanced so as to have different numbers of subjects in the sequences (and/or the groups cf recent discussion). LSMeans are averages of averages in the sequences (groups).

Example:
If the mean of log AUC is 10.0 in sequence 1 (n=10) and the mean of log AUC is 2.0 in sequence 2 (n=2) then the LSMean is 0.5*(10.0+2.0)=7.0.

Whether LSMeans are more meaningful than Means is debated widely, but since LSMeans were invented by SAS those statisticians who leaned on SAS during their education tend to favour LSMeans over Means. Fortunately we don't all have to be sheep - unless we want to submit our dossiers to EU regulators of course . Type III SS is a good further example of this phenomenon.
In the example above one would argue that the value in sequence 1 is better estimated (n=10) than the value in sequence 2 (n=2) so it would seem unfair to weight them equally when we work out an overall meaningful mean. Terrible wording, sorry.

I do not have the answer as to why LSMeans would be more relevant than Means, or the other way around for that matter.

Best regards,
EM.