(417f) Design Space, Models and Model Uncertainty

The concept of Design Space plays a central role in the current thinking on the regulation of pharmaceutical production processes. Most such processes involve large numbers of external disturbances and/or potential control manipulations. This results in multidimensional Design Spaces, whose effective determination and exploration require the use of process models. A range of different categories of models is currently being used, including empirical correlations, statistical non-parametric models, data-based models and first-principles models.

This paper focuses on first-principles models, of the kind that are finding increasing use in pharmaceutical applications. We consider the relation between the Design Space and the concept of ?process flexibility? which has been studied extensively in the process systems engineering literature since the early 1980s. We are particularly interested in the extent to which existing flexibility analysis techniques can be applied to the determination of the Design Space, both in principle and in practice, given the transient and multistage nature of most pharmaceutical processes.

Design Spaces are supposed to provide assurances that the process will deliver products of the right quality provided the operation is maintained remain within a given envelope. However, like the result of any other model-based computation, the reliability of any Design Space determined on the basis of a model depends on the accuracy of the model itself. In other words, the Design Space is not a well delineated region, but a probabilistic one, with each point in the multivariable space of operation of the process being characterized only by a probability of belonging to the Design Space. We consider techniques for computing probabilistic Design Spaces based on quantitative information on model uncertainty derived during the formal validation of such models against experimental data sets.