(759d) Efficient Experiment Design and Model Generation to Assist Process Decision Making

Statistically designed experiments and empirical modeling techniques are commonly used in the pharmaceutical industry to define process ranges and enhance process knowledge. Parallel and automated experimental techniques are frequently used to execute the designs, as the experimental plan can be large. In some cases, parallel experimental techniques cannot be used due to mixing considerations or other setup specific issues. 

This work presents a model-driven experimental strategy that was used to characterize a two reaction telescoped synthesis to product an active pharmaceutical intermediate.  Additional complexity in this synthetic step includes the heterogeneity in the reactions and analytical stability issues.  This process step required labor intensive experiments which had to be executed one at a time This work describes a process characterization strategy that started with an efficient D-optimal experimental design that was subsequently augmented based on findings from exploratory data analysis. The team used a model-driven approach, where analysis of the data was partially accomplished by the studying of the resulting statistical models and fitting characteristics. The resulting models were used to guide pilot plant parameter variation and were verified by the resulting on-scale data collection. The models were also used to perform robustness analysis of the system, where the expected process outcome variability was estimated by simulating expected process parameter variability. This experimental and modeling strategy provided actionable information with a modest number of experiments that allowed the project team to make decisions in a timely manner.