Bayesian predictions of probabilities of out-of-specification drug products during a stability analysis

Out-of-specification products (OOS) are critical issues for drug manufacturers. They may cause serious financial and legal consequences for companies. Manufacturers are increasingly undergoing regulatory pressures to develop and use modern quality management techniques to evaluate and control the risks of such products in their productions, both during stability studies and after the drug release. One way to address the situation is to assess the level of risks of OOS associated with a chosen shelf-life of a product. To do so, drug manufacturers need reliable degradation models from which the probabilities of occurrence of OOS can be computed. This work proposes and evaluates two Bayesian models to predict the probabilities of OOS in pharmaceutical productions based on stability data. The first method computes the expectation of the probability of OOS based on the posterior predictive distribution of a quality attribute. The second method computes the posterior predictive distribution of the probability of OOS as function of the parameters of the drug degradation model, and summarizes this distribution with its mode, median, mean and credibility intervals. The performances of the two methods are evaluated and compared. Integrating such models in existing shelf-life estimation methods would considerably enhance the control of OOS risks.