(183b) Predicting Roller Compaction Process Capability

Roller compaction is a common processing option in the tableting process of small molecule pharmaceutical compounds.  Because roller compaction can be described using a physical model of the process, it is possible to predict manufacturing behavior from laboratory experiments.  However, it is desired to estimate the process robustness incorporating uncertainty at the defined manufacturing conditions.  This presentation will describe an application of probabilistic modeling to leverage process models, lab data, manufacturing data, and uncertainty to best incorporate potential sources of process variability.  

Using Markov Chain Monte Carlo, the process model can be regressed to laboratory data or pilot manufacturing data, resulting in an estimate of the possible outcomes of the measured process quality attributes, i.e. ribbon density.  These outcomes are generated from a Bayesian network that integrates all the sources of variability (measurement error, model error, and input variability) associated with the process, predicting process capability (Cpk) and identifying any hidden deficiencies in the input data.

The framework quantifies the expected Cpk at the time of process qualification, increases the fundamental scientific knowledge of the process and through a sensitivity analysis, provides a rational mechanism to prioritize resources for future measurements.