Interpretation and Optimization of the Dissolution Specifications for a Modified Release Product with an In Vivo–In Vitro Correlation (IVIVC)

Abstract

This article considers the in vivo significance attached to in vitro dissolution testing. Almost invariably, the in vitro dissolution test is interpreted in terms of bioequivalence. The literature that describes methods for setting in vitro dissolution specifications is reviewed. The most common interpretation of these specifications is a deterministic one, that is, those batches passing the dissolution specifications would be bioequivalent with the reference if tested in vivo and those failing the dissolution specifications would not be bioequivalent if tested in vivo. Due to random variation, the deterministic interpretation is not appropriate. Instead, we need to consider the conditional probability that a batch that has passed the in vitro dissolution test would demonstrate bioequivalence if tested in vivo, and that a batch known to have failed the in vitro dissolution test would demonstrate bioinequivalence if tested in vivo. One way to estimate these probabilities is by means of a simulated experiment in which the production and testing (in vivo and in vitro) of a large number of batches is computer simulated. Such a simulation can only be performed if the relationship between the in vitro dissolution characteristics and the in vivo performance of the product has been modeled. These models are generally referred to as in vivo–in vitro correlations (IVIVC). The results of one such experiment are described. The above-mentioned conditional probabilities are shown to depend on the choice of dissolution specifications. This result leads to the notion of optimal dissolution specifications. However, both of the conditional probabilities cannot be maximized simultaneously. The probability of making a correct decision on the basis of the in vitro dissolution test is introduced as a possible optimality criterion. This probability is a linear combination of the two conditional probabilities of interest. Using this criterion, the optimal dissolution specifications can be found by searching over the multidimensional space defined by the half width of each interval used in the specifications to find the combination that maximizes this probability. This process is demonstrated using the Nelder-Mead search routine. The choice of dissolution specifications has profound implications for the routine production of the product because if the specifications were very narrow the probability of a batch passing would be low, resulting in a low hit rate. The same computer program used to perform the simulation experiment can be used to estimate the hit rate. Furthermore, it can be used to explore the magnitude of changes required in the parameters describing the test product (particularly variability) to increase a low hit rate to an acceptable level. © 2004 Wiley-Liss, Inc. and the American Pharmacists Association