(T1530-02-11) Analysis of Lubrication Process Impact on Formulation Blend and Tableting Properties with Numerical Models and Statistical Analysis

Purpose: The present investigation aimed to evaluate the impact of the lubrication process on model formulation, blend and tableting properties with numerical models and statistical analysis.

Methods: A 2² full factorial design with 3 center points was used to evaluate lubrication conditions on the formulation blend flow and tableting properties. X-variables were lubrication amount (0.5 to 1.5%) and lubrication time (5 to 15 minutes). Y-variables were the Hausner ratio (HR), Carr’s index (CI), angle of repose, ejection force, and tablet mechanical strength (TMS). A model tablet formulation was composed of ketoprofen (5.00%), Vivapur® 102 (44.00%), SuperTab^® 24AN (30.00%), Starch 1500^® (19.50% to 20.50%), and magnesium stearate[(MgSt), 0.50 - 1.50%]. A formulation was blended in a V-cone blender at 15 RPM. Tablets were made using the Natoli RD-30 rotary tablet press. The data was evaluated using qualitative and quantitative multivariate analysis techniques using Unscrambler software (The Unscrambler® 10.3, CAMOS AS, Norway). A qualitative principal component analysis (PCA) was employed to decode the hidden data patterns. Partial least squares regression (PLS-1) was used to quantify the statistically significant factors and their influences. The design and response variables were weighed with their standard deviation(1/SD) before PCA and PLS modeling. During PLS modeling all excipients were treated as split category variables to separate their impact on the respective Y-variables. The uncertainty of PLS regression coefficients was estimated with full cross-validation and Jack-Knifing (1). The details of PCA and PLS will be found in our previous publications (2). To support and complement the statistical model, a numerical model based on the Discrete Element Method (DEM) was implemented. Each material in the formulation was implemented and calibrated via standard laboratory tests. The MgSt is rendered as an additional “lumped” parameter that is carried and transported by the above-mentioned materials upon blending and is transferred within colliding particles while changing their flow properties.

Results: A qualitative PCA (Figure 1) showed an expected negative relationship between lubrication time and lubrication amount, with important flow descriptors such as HR and CI along PC1. These experimental parameters showed opposite relationships with TMS along PC1, which could be attributed to the weakening of interparticle bonding by MgSt(2). It is important to note that the relationship between lubrication time and lubrication amount with ejection force varies with applied compression pressure along PC1. Quantitative PLS-1 (Table 1) models of HR, CI, angle of repose, and TMS were developed. It is important to note that the executed ranges of lubrication time and lubricant amount were considered statistically insignificant in the present study. However, bulk and tapped density, as well as other individual excipients, showed their respective significant effects on individual models. However, RMSEC and RMSEC of HR, CI and TMS models are >0.93, which indicates their better predictability. The DEM model is shown to be a valuable tool to predict the effect of lubricant concentration within the blend, although it needs precise calibration from experimental data. When modelling known excipients and APIs, it can be a useful tool to support and complement the experimentation, especially within formulation development and lubrication optimization.

Conclusion: These studies showed that though main variables such as lubrication time and lubrication amount were found to be statistically insignificant, the optimized models of HR, CI and TMS could predict blend flow and tableting properties in the desired experimental range. Qualitative trends were successfully replicated by the numerical models.

References: 1. Martens H, Martens M. 2000. Modified jack-knife estimation of parameter uncertainty in bilinear modeling by partial least square regression (PLSR). Food Qual Prefer 11:5–16.
2. Haware RV, Shivagari R, Johnson PR, Staton S, Stagner WC, Gupta MR. Application of Multivariate Methods to Evaluate the Functionality of Bovine- and Vegetable-Derived Magnesium Stearate. Journal of Pharmaceutical Sciences. 2014;103(5):1466-77.
3. Orefice L, Khinast, JG. A novel framework for a rational, fully-automatised calibration routine for DEM models of cohesive powders. Powder Technology. 2020; 361:687-703.

Figure 1A PCA scores plot. The first 2 principal components displayed 59% variance in the data.



Figure 1B PCA loadings plot. The first 2 principal components displayed 59% variance in the data.

Table 1 Regression coefficients of design variables on HR, CI, angle of repose, and TMS obtained in a separate PLS-1 models. (The significance of regression is determined by full cross-validation and Marten’s uncertainty test and corresponds to approximately p<0.05; (-) indicates negative and (+) indicates positive significance. * Indicates statistically significant impact. RMSEC and RMSECP means root mean square error at calibration and prediction, respectively)