(M1530-02-08) Compression Process Optimization Using a 2^3 Full Factorial Design of Experiments

Purpose: Execute and analyze a 2³ full factorial design of experiments (DOE) + 3 center points to optimize the compression process for an immediate-release tablet drug product using a Fette 1200i pilot-scale rotary tablet press.  The study was performed to identify significant and reliable models relating turret speed, feeder speed, and main compression force settings to routine compression in-process controls/parameters for process development purposes.

Methods: The formulation contained 3% drug substance.  Eleven randomized runs were executed with turret speed (20-60 rpm), feeder speed (20-60 rpm), and main compression force (10-40 kN) as factors. The responses were weight, thickness, and hardness means and relative standard deviations (RSD), friability, mean disintegration, upper and lower punch stiffness, main compression force relative standard deviation (srel), and dissolution at the 5-, 10-, 15-, 30-, 75-minute time points. The DOE was executed with the tablet press in manual mode.  Weight, thickness, and hardness data were generated using a Sotax ST50 automated tablet tester.  Friability and disintegration data were generated using a friabilator and Sotax DT2 tester, respectively.  JMP 17 software was used for the DOE analysis.

Results: The analysis comprised four steps: Model significance, model fit, diagnostic check, and model interpretation. Model Significance: The data were analyzed for potential models and the models were automatically optimized using the stepwise personality, the minimum Bayesian Information Criterion stopping rule, and the forward direction. This stopping rule was selected to ensure model parsimony which simplifies interpretation. A p-value limit (alpha) of 0.1 (equating to 10% statistical uncertainty) was chosen for model term and overall model significance. The p-values for significant models are in Table 1. Weight RSD, mean thickness, mean hardness, mean disintegration, friability, srel, upper punch stiffness, and mean dissolution (5-, 10-, 15-, 30-, and 75-minute time points) responses had p-values that met this limit, signifying they were appropriate to evaluate model fit. The srel response data required a natural log transformation to improve model fit. Model Fit: Model fit evaluation was performed on the significant models to identify models that could reliably be used to describe the relationship between the DOE factors and responses.  For adequate model fit, first, a limit of NLT 0.70 was chosen for the adjusted R² term, which signified that the model must explain at least 70% of the variation in the data.  Second, the difference between raw and adjusted R² values, as well as the difference between adjusted and predicted R² values, needed to be NMT 0.20.  Third, if available, the lack-of-fit p-value needed to be non-significant ( >0.10).  R² and lack-of-fit p-values are in Table 2.  Values in red font indicate results that did not meet model fit requirements. The mean thickness, mean hardness, disintegration, srel, and upper punch stiffness models met these criteria and were thus classified as well-fitted models.  The weight RSD and mean dissolution at 15-, 30-, and -75 minutes models were not well-fitted, so further analysis was not performed on those models. No red flags were found in the other models. Diagnostic Check: Each well-fitted model’s diagnostic plots were checked to ensure the linear regression assumptions were satisfied for this study.  This was done by examining the actual by predicted, residual by predicted, residual normal quantile, and residual by row plots.  For the mean dissolution @ 5- and 10-minute models, red flags were observed in the residual by predicted plot (residuals increased throughout the study).  This signified that these dissolution models were unsuitable to describe the relationship between the studied factors and these responses. Model Interpretation: Classical relationships between factors and responses signified relationships that are commonly observed in other tablet products. Mean thickness was inversely proportional to main compression force which is the classical relationship between this in-process control and parameter (see Figure 1). The model showed that a mean thickness minimum of approximately 4.05 mm was obtained as main compression force approached 40 kN. Mean hardness was directly proportional to main compression force and inversely proportional to turret speed (see Figure 2).  The relationship with force was a classical relationship.  The inverse proportional relationship with main compression force demonstrated strain rate sensitivity for this formulation.  Disintegration was directly proportional to main compression force which is the classical relationship for many tablet products (see Figure 3).  All disintegration data points were lower than 60 seconds, which is very low. The other responses are not discussed in this abstract due to lack of space.

Conclusion: For the compression process DOE, with factors of turret speed, feeder speed, and main compression force, significant and well-fitted models were identified for mean thickness, mean hardness, mean disintegration, srel, and upper punch stiffness.  For this batch, proven acceptable ranges for turret speed, feeder speed, and main compression force were 20-53 rpm, 26-60 rpm, and 10-40 kN, respectively.

Mean Thickness Model Representation

Mean Hardness Model Representation

Mean Disintegration Model Representation