Statistical approaches to stability protocol design.

In stability protocols, data are usually visualized as being generated, and stability evaluation is accomplished at a point in time when sufficient data have been accumulated. Often, data are simply treated by the "statistically best fit" and, as a consequence, statements describing some batches as being first order and some being zero order are frequently used. From a scientific point of view, it is more advantageous at the preformulation stage to ascertain what the stability profile should be (i.e., what the mechanism is) and then apply the statistics to this format. Examples are given of pH profiles, Arrhenius plotting, and dissolution data. In the first case, the use of fractional factorials (a matrix approach) is suggested.