2021 年 58 巻 5 号 p. 224-233
Process Analytical Technology (PAT) is played an important role in the realization of stable production in continuous pharmaceutical production. NIR, which is representative of PAT tools, is typically implemented, and knowledge or experience of spectrum preprocessing, wavelength selection, and multivariate analysis is indispensable to develope a prediction model.
Moreover, cross-validation is used to evaluation of the calibration model, however big gap between the actual measuered value and the predicted value being observed on the process. We examined the applicability of an efficient prediction model that can produce a high accuracy model at a manufacturing site without specialized knowledge or experience based on the actual process of the continuous tablet coating process, as one step of continuous production. In this study, the importance of calculating standard error of prediction and validating verification by measuring unknown samples at an early stage is concluded. In addition, in this coating process, process visualization was performed by principle component analysis of process parameters without using NIR, and process behavior between batches and equipments was confirmed.