Estimation of Active Pharmaceutical Ingredients Content in Blending Process for Drug Products Manufacturing

摘要

This study focuses on the content estimation of Active Pharmaceutical Ingredients (API)in the blending process from Near-Infrared spectrum (NIRS) data. Particle size data areused as inputs in addition to spectrum data to take account of the influence of particlesize distribution on NIRS, and locally weighted partial least squares (LW-PLS) is appliedto estimate API content. LW-PLS builds a local linear regression model based on thesimilarity between a query and sample data stored in the database when output estimationis required. LW-PLS can be adaptive and cope with nonlinearity in spite of its simplicity.In addition, a statistical wavelength selection method is proposed. This method can select asuitable set of wavelengths for the estimation; the selected wavelengths are found to includethose corresponding to the spectrum peaks specific for API. LW-PLS and the proposedwavelength selection method were applied to real process data, and the estimation accuracywas improved by 18.7 % in Root Mean Square Error of Prediction (RMSEP) compared withthe conventional PLS and wavelength selection method base on spectrum peak positions.The results clearly show that the proposed method is useful for the API content estimationand is superior to the conventional method.