Statistical Experimental Design

Statistical Experimental Design

INTRODUCTION

Statistical experimental design has been employed for optimization of complex multivariate processes central to pharmaceutical product development for several decades (Box et al., 1978; Cochran and Cox, 1957). The intent of experimental design is to rapidly and efficiently study many parameters to identify the combination of conditions that most efficiently and reproducibly generates a product with desired properties. These properties are most frequently uniform and reproducible drug delivery from a stable dosage form, but can extend to more subtle phenomena such as control of particular physicochemical proper-ties or even cost and time efficiency of the process.

The general evolution of experimental design techniques begins with conventional randomized or Latin square approaches, factorial design and fractional factorial design, which yields significant and useful information with regard to the limits of input variables with respect to particular output prop-erties. The results of these studies can be employed to identify regions of combinations of input parameters, so-called design space, that give rise to desirable output properties. Assessment by central composite experimental design yields more information regarding the curvature of design space, ulti-mately leading to complete response surface maps, which allow interpolation of changes in output as a continuous dependent function of the independent variable inputs. The following sections will describe each of these approaches in more detail. All lead to the concept of process design space, which leads to normal operating range from which product specifications can be developed.

Once input variables have been identified, according to the approach described in chapter 17, experiments can be designed that evaluate their con-tribution to the critical quality attributes of the product under development. Therefore, it is important to identify the output parameters and the techniques (see chap. 19) that will be employed to measure these properties to allow their response to the input variables to be characterized. Having considered these practical elements of the experimental design, the statistical approach has then to be selected for its relevance to the process under consideration.