Development, validation, and implementation of NIR calibration models in a commercial continuous manufacturing process

Abstract

Near infrared spectroscopy is a very promising non-invasive technique since it allows the use of calibration models to monitor physical and chemical properties of raw materials, intermediate products (blend), and the end product (tablets) without sample preparation. In this investigation, near infrared (NIR) spectroscopy and chemometric models were used as in-line techniques integrated to a closed-loop control system that provides drug concentration results in real time. The use of NIR chemometric modelling for continuous manufacturing (CM) processes, such as the one discussed in this study, allows the manufacture of large quantities of product in a short time while maintaining all necessary controls to ensure high quality of the end product. The first investigation presented in this dissertation focuses on the use of chemometric models and variographic analysis to evaluate the analytical and sampling errors of the predicted API concentration of blends produced during a pharmaceutical CM process. An NIR calibration model was developed using blends prepared in lab scale equipment. The model was validated with blends prepared using lab scale, pilot plant, and CM processes. Variographic analysis was performed to blends and tablets prepared using the CM process. The second investigation presented in this dissertation focuses on the integration of PAT and CM in a CGMP regulated pharmaceutical plant. This study shows the application of CM and chemometric modelling for the commercial manufacturing of a pharmaceutical product. Two NIR chemometric models were developed, validated, and implemented for the identification and quantification of blends produced during a commercial CM process. The calibration and validation sets were prepared using the CM process, thus including sample and process variations into the models. All blend spectra were collected in-line, during the manufacturing process. An original approach is suggested for the calculation of the standard error of prediction (SEP) acceptance criteria. Variographic analysis of a 28-hour commercial run was performed. The third investigation presented in this dissertation focuses on testing the robustness of an NIR calibration model for the prediction of drug concentration in core tablets during a CM process. The robustness evaluation was performed by exploring how the NIR spectra and predictions were affected when tablets were: (1) exposed to the environment for prolonged times; (2) protected from the environment; and (3) experiencing their “relaxation” phase (elastic recovery). An NIR calibration model was developed with tablets prepared using lab-scale equipment. Two iii optimizations were performed to the NIR calibration model based on: (1) spectral range and (2) calibration sample set. The inclusion of tablets representative of the CM process to the NIR calibration model proved to be an efficient way of including inherent process variations, thus increasing the robustness of the model.