(M1430-08-46) The Importance of Design of Experiment Model Selection in Pharmaceutical Product Development - A Case Study of Tablet Compression

Purpose: For pharmaceutical product development, proper selection of a design of experiments (DOE) model is critical. At the early stages of the development, minimal materials are available, yet the design and response spaces must be appropriately characterized to develop a robust manufacturing process. The purpose of this study was to develop an empirical model to characterize and optimize the design and response spaces of a tablet compression process, as well as to determine specification limits for the tablet thickness based on the DOE results.

Methods: The I-optimal DOE method was developed by minimizing the correlations among the independent variables such that the results would be harvested with minimal noise without increasing the size of the DOE model. The impact of three independent variables, including (a) Turret speed (TS), (b) pre-compression force (PCF), and (c) main-compression force (MCF), on the response variables including (a) tablet weight, (b) hardness, and (c) thickness, were studied. The model contained 8 trials, including one center-point condition. The DOE trials were executed on a two-sided commercial-scale tablet press using ~40 kg of materials. The response variables were analyzed using Neural Network (NN) predictive modeling to explore the design and response spaces. Optimal conditions of the independent variables were proposed to achieve the target tablet weight of 220 mg and target tablet hardness of 12 kP. Also, the associated target tablet thickness was predicted by utilizing NN simulations. Three validation batches were then manufactured at the proposed optimal conditions, and the results were compared against the values predicted by the DOE analyses.

Results: Table 1 shows the measurements of the DOE response variables. Using NN predictive modeling, an empirical model was developed to determine the optimal values of Turret speed (2400 tablets/min/side), pre-compression force (1.5 kN), and main-compression force (9.1 kN) to achieve a tablet target weight of W = 220 mg and a tablet target hardness of H = 12 kp.
NN simulation was also utilized to predict the target tablet thickness associated with the target tablet weight and hardness. Simulated results predicted a tablet thickness of T_(Predicted) = 3.809 mm (Figure 1).
Using the aforementioned optimal process parameters, three validation compression batches (550 kg each) were then manufactured. The average results for the tablet weight (W_(Average) ), hardness ( H_(Average)), and thickness ( T_(Average)) from the three validation batches are presented in Figure 2.
The results of the validation batches were significantly compatible with the predicted ones as follow; tablet weight ( W_(Predicted) =220 mg vs. W_(Average) =220.2 mg), tablet hardness ( H_(Predicted) = 12.0 kp vs. H_(Average) = 12.01 kp), and tablet thickness ( T_(Predicted) = 3.809 mm vs. T_(Average) = 3.808 mm).

Conclusion: A comprehensive yet small I-optimal DOE model (8 trials) was executed using a minimal amount of materials (~40 kg) on a double-sided commercial-scale tablet press to study the design and response spaces of a tablet compression process. Using Neural Network modeling, the optimal values of the independent variables were determined. Validation compression batches (~550 kg each) were then manufactured using the aforementioned optimal process parameters, and the results showed significant compatibility with the predicted values. This study showed the importance of properly executed DOE models in decreasing the research and development time and cost. Furthermore, optimal DOE models are crucial to harvesting significant results with minimum noise without increasing the DOE model size.


Table 1: Core Tablets DOE Summary


Figure 1: NN Model, T=f(W,H), Core Tablets


Figure 2: Tablet Characteristics, Validation Batches