Auswertung und Optimierungvon Haltbarkeitsuntersuchungen
bei pharmazeutischen Produkten
1. Mitteilung: Grundlagen Martin Holz a, Donald Dill b, Rudolph Frank c und Theo Wember d Datenanalyse Dr. Holza, Neuenburg, A & M Stabtestb, Mainz, Hoffmann-La Roche AGc, Basel (Schweiz),
und Statistikberatungd, Waltrop Evaluation and Optimization of Shelflife Estimation in Pharmaceutical Products / 1st Communication: Basics The regulatory demand to specify the shelflife of a pharmaceutical product and to investigate the factors affecting the shelflife with sufficient statistical confidence represents a challenge. To achieve this goal, the ICH guideline for drug stability testing offers wide choice for the experimental design and subsequent data analysis including matrixing and bracketing approaches. This current work illustrates how the precision of the shelflife estimation can be increased and the experimental effort decreased by individually adapted D-optimal designs together with general linear model multiple regression. Even for the non-mathematician, the generation of individual optimal designs is no longer problematic using available modern statistical software. The main advantage of a common model simultaneously evaluating all possible influential factors is discussed here in comparison with the separate linear regression for each combination of these factors.
Based on statistical power estimation, alternatives are p resented for the current FDA recommendation to generally test poolability against a significance level of
a = 0.25. Our approach assures with a defined power that relevant factor effects will be detected as significant if they truly exist. The implications of the small number of batches generally involved in development studies and the special error structure of most pharmaceutical assays are discussed.
A practical example for our approach will be presented later in part t wo of this article series in “die pharmazeutische industrie”. | 
