(308a) Accelerated Multicomponent Tablet Tensile Strength Predictions Using a Structured Workflow: A Case of Four Active Pharmaceutical Ingredients

The prediction of multicomponent tablet properties such as tensile strength, hardness, thickness and porosity is challenging to achieve. Despite several models previously studied in a direct compression process, compaction mechanisms change with respect to the tablet constituents' compositions and properties. This study aims to implement a structured workflow for rapid assessment and selection of an appropriate model and mixing rule suitable for tablet property predictions of a directly compressed formulation. Often, the property predictions are largely dependent on the functionality with the dominant property within the formulation. As a result, a trial and error approach is often used for selecting the appropriate model as no singular model is generically sufficient to predict across different active pharmaceutical ingredients (API).

A lean experimental approach was adopted for the five-component formulation investigated by regrouping the five to three-component formulation based on fixing some constituents compositions. The API investigated include Ibuprofen, Paracetamol, Mefenamic acid and Calcium carbonate, while the formulation excipients include Fastflo 316 Lactose, Avicel ph102 (microcrystalline cellulose), Magnesium stearate and Croscarmellose sodium. A rapid end-to-end three-step workflow was developed for the API considered as case studies and formulated into low (5%), medium (20%) and high (40%) drug loadings. Tablet formulation and direct compression to tablet formed Step 1 of the workflow while algorithm-driven parameters estimation by fitting experimental data to the models, extrapolation for non-compactible API, tensile strength predictions based on selecting an appropriate mixing rule from a group of five, and experimental verification of predictions constituted the Step 2 of the workflow. Step 3 of the workflow explores a classification model previously reported to inform new approaches that may be implemented should no mixing rule predicts the tensile strength sufficiently. Good predictions within a defined threshold of 80 – 100% parity regressions and root mean square errors are expected at the end of step 2 and 3.

Following the workflow implementation, tensile strength prediction achieved 80 – 99% accuracy across the three drug loadings for all the API investigated. To be specific, Ibuprofen, Mefenamic acid, and Calcium carbonate's tensile strengths were predicted by selecting an appropriate mixing rule at Step 2 of the workflow. On the contrary, Paracetamol tensile strength in either powder or granular form could not be sufficiently predicted across the drug loadings. Hence a mixture model-based approach adopted in Step 3 of the workflow resulted in good predictions. The predictive platform presented here offers potential time savings by linking predictive and experimental capability together. A shorter time may be achieved if the API is readily compactible and there is an existing property database for the excipients. This workflow provides a rapid, bespoke problem-solving approach to identifying appropriate model suites very early during the drug product formulation stage and accelerates the development phase by reducing cost, resources and time.