During formulation/process developments, experimental designs are often used to minimize the number of experiments and maximize the necessary information on the chemical and physical stability of the drug, the flow properties, the process parameters, and the process monitoring tools, etc. As a consequence, the main effects of many parameters are often coupled with minor effects and it can become difficult to isolate phenomena of interest. Using process analytical technologies (PAT) to monitor in real time blend homogeneity and tablet uniformity has become increasingly attractive in the pharmaceutical industry. Near infrared spectroscopy (NIRS) has predominantly been used since it can predict concentrations of drugs and excipients with reasonably low detection and quantification limits. NIRS can provide, during formulation/process development, valuable information to scientists regarding the chemical and physical properties of API and excipients, including powder flow behaviors, segregation issues, tablet properties (i.e. hardness), etc. However, the amount of variability present during formulation/process experimental designs (variations in excipient ratio, variation in blend time, variation in press parameters) can perturb NIRS calibration models and prevent the development of optimal models from monitoring precisely and accurately the process of interest. Variations in compression forces used to make tablets are one of the potential interferences that can limit the chemometrics model performances. The present study proposes to desensitize near infrared calibration model to compression (physical property) information present in the calibration set to ease the interpretation of the results given by the design of experiment. Using partial least squares (PLS) and classical least square (CLS) regression models, various desensitization methods were implemented. Orthogonalization methods, aiming at removing the physical interferences were compared with the augmentation of the loading matrix of CLS and increasing the number of latent variables in the PLS model. Methods based on the net analyte signal were also developed to focus the calibration model on the parameter of interest and not include interferences. These methods were implemented along with an optimization of the spectral pretreatments. Desensitizing PAT models to interfering phenomena can help process understanding and process control by focusing efforts on the parameters of interest and not implementing strategies to limit their effects. In the long run, these approaches will simplify the development of PAT methods and help the pharmaceutical industry move toward a Quality by Design framework. This study provides practitioners with a set of tools to ease the implementation of PAT methods during formulation development where the number of samples available is limited and many variables can interfere with building a precise and accurate multivariate model.
You will be able to download and print a certificate for these PDH credits once the content has been viewed.
If you have already viewed this content,
please click here