470147 Systematic Method for Finding Material Surrogates with Matching Flow Properties
Flow properties are the physical attributes of API and excipients that have the strongest impact on the process in continuous direct compression of solid dosage forms. In order to select an appropriate surrogate material, data obtained from a library of previously characterized surrogate candidates can be analyzed using a statistical toolbox to identify the most “similar” materials. A big part of the problem in doing this effectively is that the practitioner often must select among too many, partially redundant characterization methods of unknown reliability.
Principal Component Analysis (PCA) is a powerful linear projection method used to examine over-specified data sets. It projects multidimensional data into a few orthogonal features, called principal components (PCs). The PCs are constructed as a linear combination of the original variables to maximize the preservation of the data variance. PCA projects a high dimensional dataset into a new coordinate system. The new variable value, based on the observation’s component loading and the standardized value of the original variable, is called the component score. Examining the component scores of each sample helps to find similarity between materials and to identify redundant measurements.
The specific example discussed in this talk focuses on the application of PCA to identify placebo materials exhibiting similar physical properties to that of an API that is a controlled drug substance and hence difficult to use in large quantities. Materials with similar flow characteristics are expected to exhibit similar behavior during manufacturing. Here, PCA is applied to a data set with physical properties spanning diverse physical characteristics from a library of pharmaceutical grade materials, which would enable the identification of materials that are mostly likely to behave similarly to the API of interest. The current dataset includes 21 materials with 33 variables derived from different material property characterization test results. The measurements include particle size distribution, bulk density, FT4 test (bulk, dynamic and shear tests) and electrostatics tests. Unscrambler X software from Camo was used to conduct PCA analysis and the scatter plots of loadings and scores were obtained to examine the relationship between samples by looking at their relative positions in the scatter plot. The relative position of each original variable in the loadings plot can give information about how they relate to one another. For a better understanding of which measurements are driving the trends/groupings seen in the scores plot, the scores and loadings plot need to be examined in conjunction and hence the scores and loadings plots were superimposed in a bi-plot. Materials that are projected close together suggest similarity in terms of all the flow property parameters measured. The scores of each material on the latent variables can be used to calculate statistical distance between the target API and the materials in the library. The statistical distance can be ranked, and the most similar material to the target API is the one that has smallest distance to the target API.
Based on present PCA; the target API was found to close to 3 other API’s. This was further verified experimentally by conducting feeder studies on all of the four identified similar materials. Sticking to the feeder was observed for all the API’s under study. Moreover, the relative standard deviation and deviation around the set point was found comparable to the target API. Thus, the PCA based systematic approach can be potentially validated by using similarity scoring, and be further expanded by including more common pharmaceutical materials and other material properties.
See more of this Group/Topical: Particle Technology Forum