(238c) A Systematic Model Development Framework for Batch Crystallization Using Iterative Model-Based Experimental Design Approach

Batch crystallization remains to be a predominant separation and purification technique employed in the production of synthetic drug substances and its intermediates. Increasing expectations on the drug quality in the market place due to stringent regulations introduced new challenges for crystallization processes related to improving the physical attributes of product crystals (e.g. size, shape, purity, size distribution etc.) and associated performance characteristics (e.g. dissolution, flowability, downstream processing). Modeling the crystallization processes provides better insights into process dynamics through design space exploration and insilico studies. However, a reliable and robust predictive model is necessary for such simulation studies and for optimization and control applications. Quality-by-design (QbD) approach is the usual industrial practice for crystallization modeling where in a set of crystallization experiments are designed either by prior-knowledge/experience or traditional statistical design of experiments (DoE) based on the effect of critical process parameters (CPP) (e.g. temperature, seed, supersaturation etc.) on the critical quality attributes (CQA). However, the model calibration with larger set of CPPs could result in designing larger set of experiments that is time, material and labor intensive. A quality-by-control¹ framework was recently proposed that works on a feedback control strategy for rapid process development and is more suitable for processes involving complex procedures such as temperature cycling. Inspite of some previous reported literature on application of optimal experimental design^2,3,4 to crystallization processes, there is no generic framework proposed for modeling batch cooling crystallization processes that guides to identify best model and precise parameters with minimum effort and resources.

The focus of this work is based on a two-tier strategy with optimal experimental design for sequential model discrimination and parameter precision. The experimental design strategy is performed iteratively based on a previously available experimental data set(s) to design a new experiment based on a performance criterion. The new experiment is executed to collect experimental observations that are appended to the experimental data set and evaluated for the targeted objectives. Thus, the iterative model development approach (IMED) iterates design, execution and evaluation steps automatically to determine a best mechanistic model and corresponding parameters. The proposed workflow is demonstrated for modeling an industrial pharmaceutical compound from Takeda pharmaceutical International Co.

The modeling of batch cooling crystallization of the compound is carried out with a carefully designed DoE for initial parameter estimation considering seeded experiments with varying seed loads, initial supersaturation and linear cooling ramps. Model database for initial model identification and discrimination steps consist of various nucleation, growth, agglomeration mechanisms and corresponding mathematical expressions. D-optimality criterion applied for model refinement to find optimally informative experiments is successful in minimizing uncertainties of the estimated crystallization model parameters. Based on the obtained results, it can be inferred that IMED proves to be a promising tool for discrimination of many rival models and further refinement of model parameters. This two-tier model development approach presented can serve as a generic framework for batch crystallization processes.

Acknowledgement

Financial Support from Takeda Pharmaceuticals International Co. is gratefully acknowledged.

References

1. Szilagyi, A. Eren, J.L. Quon, C.D. Papageorgiou, Z.K. Nagy, 2020, Application of model-free and model-based Quality-by-Control (QbC) for the efficient design of pharmaceutical crystallization processes, Cryst. Growth & Des., 20, 6, 3979–3996.
2. H. Chung, D. L., Ma., R. D., Braatz, 2000, Optimal model-based experimental design in batch crystallization. Chemometrics and Intelligent Laboratory Systems, 50(1), 83-90.
3. H. Chen, S., Bermingham, A.H., Neumann, H.J. Kramer, S.P., Asprey, 2004, On the design of optimally informative experiments for dynamic crystallization process modeling, Industrial & engineering chemistry research, 43(16), 4889-4902.
4. Pal, B. Szilagyi, Burcham, C.L., D.J. Jarmer, Z.K. Nagy, 2021, Iterative model‐based experimental design for spherical agglomeration processes. AIChE Journal, 67(5), e17178.