Comparison of several strategies for the deployment of a multivariate regression model on several handheld NIR instruments. Application to the quality control of medicines

Abstract

Chemometrics applied to spectroscopic measurements such as near-infrared are gaining more and more importance for quality control of pharmaceutical products. Handheld near-infrared devices show great promise as a medicines quality screening technique for post-marketing surveillance. These devices are able to detect substandard and falsified medicines in pharmaceutical supply chains and enable rapid action before these medicines reach patients. The instrumental and environmental changes, expected or not, can adversely affect the analytical performances of prediction models developed for routine applications. Based on a previous study, PLS prediction models were developed and validated on three similar handheld NIR transmission spectrophotometers of the same model and from same company. These models have shown to be effective in analyzing metformin tablet samples, but significant spectral differences between handheld systems complicated their deployment for routine analysis. In this study, different strategies have been applied and compared to correct the instrumental variations, including global modelling (GM) and calibration transfer methods (Direct Standardization, DS; Spectral Space Transformation, SST and Slope/Bias correction, SBC), considering the RMSEP and the accuracy profile as assessment criteria. The transfer methods showed good capabilities to maintain the predictive performances comparable to that of the global modelling approach, except for a remaining slight bias. This approach is interesting since very few standardization samples are required to develop an adequate transfer model. GM, SST and SBC were able to correct/handle drifts in the spectral responses of different handheld instruments and thus may help to avoid the need for a long, laborious, and costly full recalibration process due to inter-instrument variations.