Bioequivalence and Bioavailability Forum

Hi KG,

❝ Dear Sir,

      ↑↑↑↑ Not interested in opinions of female members of the Forum?

❝ Why Geometric mean is taken in to consideration in BE studies. Why not arithmetic mean??

There is a consensus – for decades – that AUC and C_max (like many biological variables) follow a log­nor­mal (rather than a normal) distribution. Whereas the arithmetic mean is the best unbiased estimator of location for the normal distribution, the best estimator for the lognormal distribution is the geometric mean.
Justification for the lognormal distribution in BE:

• Negative concentrations are not possible.
  
• Empiric distributions of AUC and C_max are skewed to the right.
  
• Serial dilutions in bioanalytics lead to multiplicative errors.

See this example for the distribution of AUC-values (437 subjects from pooled studies).

<nitpicking>

The distribution of data is not relevant, only the intra-subject residuals from the model. How­ever, using the geometric means – or adjusted means in case of imbalanced sequences (SAS-lingo: least squares means) – is in line with the log-transformation / multiplicative model.

</nitpicking>