(T1130-08-55) Monolayer Tablet Compression Process Optimization Using a Response Surface Method Design of Experiments

Purpose: Execute response surface method design of experiments (DOE) to optimize an immediate release monolayer tablet compression process using a Fette 3090i double-sided commercial scale rotary tablet press to identify significant and reliable models relating turret speed, feeder speed, and main compression force settings to routine in-process controls/parameters.

Methods: The formulation contained 12% drug substance. Thirty randomized runs with varying combinations of turret speed, feeder speed and main compression force were executed as a response surface method DOE. The factors were turret speed (20-50 rpm), feeder speed (10-50 rpm), and main compression force (4-13 kN). The responses were weight, thickness, and hardness mean and relative standard deviation, friability, mean disintegration, and main compression force relative standard deviation (srel). Since the tablet press is double-sided, each response had two sets of data (identified as sides A and B). The DOE was executed with the tablet press in manual mode. After each combination of press speed, feeder speed, and main compression force parameters were entered into the press human-machine interface, the operator adjusted fill depth to achieve tablets within weight limits and let approximately three minutes elapse before taking tablet samples from both sides of the press. Weight, thickness, and hardness data were generated using a Sotax ST50 automated tablet tester. Friability and disintegration data were generated using a friabulator and Sotax DT2 tester, respectively. JMP 16 software was used for the DOE analysis.

Results: The analysis comprised three steps: Model significance, model fit, and diagnostic check; Model Significance. The models were optimized automatically using the stepwise personality, both minimum Akaike’s Information Criterion and Bayesian Information Criterion stopping rules, and the forward direction. Common model terms using both stopping rules were taken forward into the analysis to ensure model parsimony. From these optimized models, a p-value limit (alpha) of 0.05 was chosen for model term and overall model significance due to the high number of DOE runs. Terms with p-values not more than 0.05 remained in the model as well as main effects contained by significant two-way interactions (see Table 1). Significant p-values in all models except for mean weight (both sides) and hardness RSD (side B) were identified. Model Fit; The model fit evaluation was performed next to identify reliable models that can be safely used to describe the relationship between the DOE inputs and outputs. For adequate model fit, a limit of NLT 0.70 was chosen for the adjusted R² term. This signified that the model must explain at least 70% of the variation in the data. Additionally, the difference between raw and adjusted R² values, as well as the difference between adjusted and predicted R² values, needed to be NMT 0.2. The lack-of-fit p-value was another measure of how well the model fits the data and needed to be NLT 0.05. Mean hardness (both sides), mean disintegration (both sides), friability (Side A), and both srel models met these acceptance criteria, indicating adequate fit of the models to the data. The srel data required an inverse square root transformation to improve model fit. No other responses need a transformation. No mean thickness lack-of-fit p-values were < 0.001, indicating a potential problem with model fit for that response. However, since the R² values met the limits, the mean thickness model was taken forward into the diagnostic check. R² and lack-of-fit p-values are given in Table 2, with values in red font indicating model fit problems. See Table 3 for a graphical representation of the mean hardness model. Diagnostic Check; Adequately fitted models 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. No red flags were observed for these models.

Conclusion: The models generated adequately describe the relationship between main compression force, turret speed, feeder speed factors and mean hardness, disintegration, friability (Side A only), and srel responses. For the significant well-fitted models, main compression force had the largest effect on the outputs which was consistent with compression theory and knowledge of comparable direct compression products. For main compression force relative standard deviation, turret speed also had a significant effect.

Table 1. Significant Model Term P-Values

Table 2. Model Fit Summary

Table 3: Graphical Representation of Mean Hardness Model