(T1130-08-51) Multivariate Analysis of Granule Size Data from Continuous Granulation Screening and Response Surface Studies: A Tool for Process Understanding

Purpose: Multivariate data analysis is pivotal for providing maximum process understanding which is essential in developing effective continuous manufacturing equipment (CM). Twin-screw granulators are commonly used for continuous wet granulation and are typically evaluated by their ability to increase and maintain granule size. Granule size data represented as a granule size distribution (GSD) captures critical information which can help assess granulator performance. However, use of a single type of metric such as d-values, which define percentile values for granule size is insufficient to capture most of the information embedded within the granule size data. D-values were originally intended to describe a normal distribution, but GSDs are infrequently normal or log-normal. The necessity of multivariate particle size analysis has been demonstrated with API particle size data.¹ The current work seeks to contextualize the application of multivariate analysis to granule size data for CM process development. Screening and response surface studies were performed with a novel twin-screw granulation system. Results from each study were analyzed via d-values and principal component analysis (PCA). Comparing the PCA and d-value results demonstrates the additional process understanding provided by multivariate analysis. Capturing GSD variance with PCA is advantageous since d-values are not always representative of GSD shape. Principal component (PC) loadings plots are used to demonstrate the region of the distribution captured by each PC, while PC scores plots are used to show the effect of process parameters. The relationship between parameter significance and PC score plot trends highlight the value of PCA as a viable data analysis approach for granule size data in the context of process development.

Methods: The Dome-Ex Granulator and DG Dryer system (Dalton, Japan) granulated a model theophylline formulation. A definitive screening design (DSD) and a response surface methodology (RSM) were calculated in JMP (15.2.0, SAS, USA). Four process parameters were screened with the DSD, facilitating an RSM study with two parameters. The Canty SolidSizer (J.M. Canty, USA) measured the granule size for each sample using dynamic image analysis. A volume-based GSD was plotted and the dv10, dv50, and dv90 (10^th, 50^th, and 90^th, percentile, respectively) were calculated using MATLAB (R2022b, MathWorks, USA) for each sample. Additionally, data preprocessing (mean centering) and application of PCA on the granule size data was performed using PLS_Toolbox (R8.8.1, Eigenvector Research, USA). Size fractions [fines (0-300um), small granule (300-600um), oversize ( >1500um)] were incorporated as DSD outputs to facilitate the results comparison.

Results: The DSD yielded parameter significance for each output (three d-values, three PC scores, three size fractions). Table 1 identifies the significant parameters for each DSD output. The notable similarity is highlighted in green. The differences in parameter significance for the remaining d-values and PC scores suggest that the information captured by the d-values may not be consistently representing specific regions of the distribution. The limitations of d-value analysis manifest in the plot in figure 1A. Limited information regarding the effect of L/S on each d-value can be derived from the trends. The loadings plot in figure 1B shows the shape of variance captured by each PC. The regions of the distribution correlating to the size fractions are highlighted to show the ability of PCA to capture the relevant variance in the GSDs. The score plots in figure 1C and 1D are labeled according to L/S and screw speed, respectively. The RSM results are shown in figure 2. The surface plots in figure 2A and 2B show the effects of screw speed and L/S, which leads to the importance of PCA. D-value analysis suggests an effect of both screw speed and L/S on dv10. The surface plot for PC1 is similar, however, the scores plot in figure 2C and 2D reveals the true effects of each parameter. L/S is shown to affect the GSD by the trend in the PC1 vs PC2 plot, while the lack of trend with screw speed likely indicates a change in process dynamics.

Conclusion: Throughout this work, PCA is shown as a viable data analysis method for analyzing granule size data from process development studies. The DSD study showed the ability of PCA to capture the relevant variance in the GSDs. The scores plot in both the DSD and RSM studies demonstrated the increase in process understanding afforded by PCA. The PC scores trend with L/S suggests a true effect on GSD, while the lack of trend with screw speed suggests a potential process dynamic change. This information increases the process understanding from what is provided by d-values. This highlights the potential for accelerating the development and implementation of CM granulation processes with multivariate data analysis of GSDs.

References: 1. Gamble, J. F., Akseli, I., Ferreira, A. P., Leane, M., Thomas, S., Tobyn, M., & Wadams, R. C. (2023). Morphological distribution mapping: Utilisation of modelling to integrate particle size and shape distributions. International Journal of Pharmaceutics, 635, 122743.

Acknowledgements: The authors wish to gratefully acknowledge the support from LCI for this work.

Table 1: Parameter significance at α=0.05 calculated according to the DSD. Important comparison highlighted in green.

Figure 1: 1A-Trends in d-values according to L/S change. 1B-PC loadings with size fraction regions highlighted (fines=green, small granule=purple, oversize=light blue). 1C-PC1 vs PC2 scores plot labeled by L/S. 1D-PC1 vs PC2 scores plot labeled by screw speed.

Figure 2: 2A and 2B-RSM surface plots for dv10 and PC1, respectively. 2C-PC1 vs PC2 scores plot labeled by L/S. 2D-PC1 vs PC2 scores plot labeled by screw speed.