Bioequivalence and Bioavailability Forum

Hi Raja,

❝ Define and differentiate Compartmental and Noncompartmental.

You have some data of plasma conc. versus time. These can be fit to a model (e.g. by least squares) that make some assumptions in terms of compartments and absorption lag-time etc. That's compartmental modeling. In that case you can calculate an area under the curve or max conc. on basis of the resulting constants (and the model they belong to).
On the other hand, you can also look at it without making any assumptions about the number of compartments or anything physiological. That's non-compartmental. The AUC can be estimated by means of summed trapez-areas and Cmax is the maximum observed value.

❝ Define and differentiate Parametric and Non-parametric.

Parametric statistics is what you use if you have a reason to assume you know with a good approximation the distribution of your data (or their residuals if we talk models). Non-parametric is if you do not know the distribution. Example of parametric statistics: Often in BE with Cmax and AUC it appears reasonable (after a transformation) to think that the data/residuals are normally distributed, and a confidence interval is constructed on basis of the observations and that same assumption.
Example of non-parametric: Tmax follows a generally unknown distribution. It cannot be smaller than zero, but it can be zero. Thus log transform leads nowhere, and a normal distribution cannot be assumed. People interested in testing it then resort to techniques that assume nothing about the distribution.

❝ Define and differentiate Linear and Non-linear.

Non-linear:
Expressions that are to be solved for "X" are nonlinear if the expressions involve anything other than constants and coefficients in front of X, such as X² or sin(X).
Linear:
Expressions that are to be solved for "X" are linear if the expressions involve just constants and coefficients in front of X.
Example: In a 2,2,2-BE trial with 20 subjects you have 20 subject constants, 2 treatment constants, 2 period constants, and 2 sequence constants. All in all, that's ...hang on, just need to fire up my spreadsheet... oh yeah, 26 constants¹. Think of each of these constants as X. We make 40 equations that each combine (sets of²) these 26 constants, then apply some algorithmic fitting to find the most likely values for the 26 constants.
Note that although log-transformation is used for the observations (Cmax, AUC) the system is linear for the constants we wish to estimate.

Best regards
EM.

1: Technicality: some of them can be derived from others.
2: Technicality: depending on contrast coding.