(W0930-01-04) Optimizing Drum Selection for Volumetric-Based Capsule Filling: A Mechanistic Model for Lactose-Based Formulations

Purpose: This study aimed to develop a mechanistic model for the selection of suitable drums for capsule filling utilizing drum filling technology across different lactose-based formulations based on material flow properties.

Methods: For this study, nine different lactose samples with varying flowability (Carr’s index ranging from 12 to 50) were thoughtfully selected, each exhibiting distinct particle size distributions and flow properties. These samples were utilized to construct a PLS model, leveraging lactose properties to predict fill weight and validate the model. Capsules were filled using the ModuC LS trolley, which was equipped with drums of three different capacities (7.5mm³, 18.8mm³, and 31.5mm³). Previous research has underscored the importance of vacuum adjustment in achieving consistent capsule fill weights and minimizing the relative standard deviation (%RSD) in fill weights. To delve deeper into this phenomenon, three vacuum levels (200hPa, 400hPa, and 600hPa) were applied during the filling process. Furthermore, the lactose formulations underwent comprehensive analysis for parameters such as D50, D90, bulk and tapped density, and flow function, utilizing the Malvern Mastersizer 3000E and Micromeritics FT4 rheometer, respectively. The primary objective of this study was to elucidate the intricate relationship between material properties and process parameters, with the ultimate goal of developing a machine-learning model capable of predicting the optimal drum size based on material properties and the desired to fill weight.

Results: Figure 1 suggests a positive correlation between vacuum pressure and puck density, while an inverse correlation is observed between vacuum pressure and the percentage relative standard deviation (% RSD). Also, Figure 1 visually explores the connection between a powder's Carr's index, its percent relative standard deviation (% RSD), and the resulting puck density during encapsulation. Across all the materials tested, there appears to be an inverse correlation between puck density and % RSD, wherein materials with higher puck density tend to have lower % RSD. This can be explained by improved powder packing during the encapsulation process. Denser packing leads to less variation in powder filling each capsule, resulting in a more consistent product. The data also suggests that Lactohale 300 and Inhalac 500 (Carr’s index: 35-37) have a relatively higher % RSD compared to the other materials. This might be attributed to their fine particle size (D50 around 5 microns). Finer particles are generally more challenging to handle during encapsulation due to factors like increased inter-particle cohesion and potential for electrostatic interactions. These factors can lead to greater variability in powder flow and packing during capsule filling, reflected in the higher % RSD observed for these materials. The PCA score plot provides a visual representation of data distribution within a reduced-dimensional space derived from principal component analysis (PCA). Each data point corresponds to a sample, positioned based on its scores on the two principal components, capturing 71% of the dataset's variance, as demonstrated in Figure 2. Clusters of closely grouped samples indicate similarities in characteristics or trends, while greater distances between points signify dissimilarities. Notably, in the depicted plot, the positive correlation between puck density and vacuum is evident, with these parameters forming a closely clustered group (Figure 2). Conversely, %RSD exhibits a negative correlation with vacuum, occupying an opposing cluster, indicative of their inverse relationship with the principal components. The positive correlation between puck density and vacuum can be attributed to densification of powder at higher vacuums. At higher vacuum settings, apparently intact slugs are prepared which improve the filling efficiency leading to less fill weightvariance. This observation underscores the contrasting effects of these parameters on the principal components, highlighting their distinct contributions to the dataset's variability. Such insights gleaned from the PCA score plot facilitate a deeper understanding of the interplay between variables and inform strategies for optimizing experimental conditions or process parameters. Based on the multivariate analysis using Pearson correlation, it is evident that several process parameters and material properties significantly impact both the relative standard deviation (%RSD) and puck density, as detailed in Table 1. Specifically, puck density demonstrated significant (greater than ± 0.5) correlations with Carr’s Index, particle size, %RSD, Angle of Internal Friction (AIF), and cohesion, as denoted by positive and negative correlation coefficients. Similarly, %RSD exhibits dependencies on Carr’s Index, particle size, puck density (or bulk density), and AIF.

Conclusion: These findings shed light on the intricate connections between various factors within the manufacturing process and material characteristics, revealing key determinants of variability in %RSD and puck density. Drawing upon insights into critical process parameters, a PLS (partial least squares) model was constructed. This model achieved an impressive accuracy of approximately 96.5% in predicting lactose puck density. Looking forward, the objective is to further refine the model for lactose-based formulations. This advancement will enable us to predict formulation puck density during filling, thereby optimizing drum size selection for the target fill weight on ModuC LS drum filling.

Figure 1: Puck density and %RSD vs Carr Index

Figure 2: score plot

Table 1: Multivariate analysis – Material attributes and DrumLab process parameters