(343c) Model Based Design of Experiments for Pharmaceutical Process Development

Design of experiments (DoE’s) are widely used experimental methods to build process understanding.  However, the classical design of experiments methodology is optimal only for systems with linear responses and where no prior knowledge exists for the system.  Classical factorial designs are common within pharmaceutical drug substance process development despite the fact that these processes are non-linear in nature (i.e. temperature, chemical kinetics).  Model based design of experiments allows the researcher to maximize, based on any prior knowledge, the information content of a future series of experiments.  Before drug substance DoE’s are planned, knowledge of the process is available from, at a minimum, route scouting experiments.  Even for reactions where the mechanism is not known, credible options can often be proposed from first principles.  From the proposed mechanism(s), a family of models can be drafted.  Thus, the combination of any amount of data with the experience and judgment of the researcher can be codified into a model more appropriate than the “no prior knowledge” DoE assumption. 

A number of model-based experimental design methodologies have been developed that address the needs of experimental design within the pharmaceutical industry.  This paper will report model based design of experiments applied to drug substance synthesis and include examples of optimal experimental design ranging from identifying under-explored experimental conditions to Bayesian updating to refine reliable process operating regions.