Purpose: Development of reliable mathematical models to predict skin permeability remains a challenging objective. This article examines some of the existing algorithms and critically evaluates their statistical relevance.
Methods: Complete statistics were recalculated for a number of published models using a stepwise multiple regression procedure. The predictivity of the models was obtained by cross-validation using a "leave-one-out" deletion pattern. The relative contribution of each independent variable to the models was calculated by a standardization procedure.
Results: The heterogeneity of the data in terms of skin origin and experimental conditions has been shown to contribute to the residual variance in existing models. Furthermore, rigorous statistics demonstrate that some published models are based on nonsignificant parameters. As such, they afford misleading mechanistic insight and will lead to over-interpretation of the data.
Conclusions: The large number of published models reflects the need for predictive tools in cutaneous drug delivery and toxicology. However, such models are more reliable when confined within well-defined chemical classes, and their applicability is often limited by the narrow property space of the set of permeants under study.