Category: Manufacturing and Bioprocessing
Purpose: The key to a successful continuous manufacturing system is the implementation of an appropriate process control strategy. Process analytical technology (PAT) provides an opportunity to detect the variability of a powder blend in real time and enables feedback or feedforward control for the manufacturing process. Near infrared Spectroscopy (NIRS) for monitoring blend potency has become one of the most implemented PAT applications in continuous manufacturing. Measuring powder in the tablet press feed frame is a promising location due to the proximity to the final tablet compression site. A common approach to establish a robust NIRS calibration model requires the measurement of dynamic samples in the feed frame across a range of active pharmaceutical ingredient (API) concentration, which consumes a large quantity of material and resource and the calibration model is usually not transferable between different scales and set-ups. In this study, calibration models for in-line quantification of powder blend potency based on NIRS was developed off-line and validated during tableting in the feed frame.
Methods: A spin-compression apparatus (off-line) was crafted to simulate the powder flow dynamic and NIRS measurement environment of a tablet feed frame (in-line). A calibration, two tests, and a validation were designed. The calibration was collected from the spin-compression apparatus incorporating both chemical (potency and excipients ratio) and physical (compression level and spin speed) variabilities. Test I was collected from the spin-compression apparatus covering a smaller space of chemical and physical variability than the calibration. Test II was collected from the feed frame encompassing the same chemical space of Test I with addition of different paddle speeds. The validation was collected from the feed frame including the same chemical space of the calibration with addition of different paddle speeds.
Results: A subset of Test I was selected to match the spectra of Test II. A specific shape was generated from the spectral difference between the subset of Test I and Test II. The calibration data was orthogonalized against the specific shape. Two NIRS models were established from the original and orthogonalized calibration data and validated using the validation set. The model of orthogonalized calibration shows a prediction error smaller than 5% of the label claim which outperforms the model of original calibration.
Conclusion: The calibration was successfully transferred from the off-line spin-compression apparatus to the in-line feed frame application with appropriate mathematical treatments: spectral matching and orthogonalization.
Shikhar Mohan– Duquesne University, Pittsburgh, Pennsylvania
MD. Nayeem Hossain– Pittsburgh, Pennsylvania
James Drennen– Assoc. Dean Research and Graduate Programs, Duquesne University, Pittsburgh, Pennsylvania
Carl Anderson– Pittsburgh, Pennsylvania