Spectral multivariate calibration without laboratory prepared or determined reference analyte values.

An essential part to calibration is establishing the analyte calibration reference samples. These samples must characterize the sample matrix and measurement conditions (chemical, physical, instrumental, and environmental) of any sample to be predicted. Calibration usually requires measuring spectra for numerous reference samples in addition to determining the corresponding analyte reference values. Both tasks are typically time-consuming and costly. This paper reports on a method named pure component Tikhonov regularization (PCTR) that does not require laboratory prepared or determined reference values. Instead, an analyte pure component spectrum is used in conjunction with nonanalyte spectra for calibration. Nonanalyte spectra can be from different sources including pure component interference samples, blanks, and constant analyte samples. The approach is also applicable to calibration maintenance when the analyte pure component spectrum is measured in one set of conditions and nonanalyte spectra are measured in new conditions. The PCTR method balances the trade-offs between calibration model shrinkage and the degree of orthogonality to the nonanalyte content (model direction) in order to obtain accurate predictions. Using visible and near-infrared (NIR) spectral data sets, the PCTR results are comparable to those obtained using ridge regression (RR) with reference calibration sets. The flexibility of PCTR also allows including reference samples if such samples are available.

Updating a near-infrared multivariate calibration model formed with lab-prepared pharmaceutical tablet types to new tablet types in full production.

Determining active pharmaceutical ingredient (API) tablet concentrations rapidly and efficiently is of great importance to the pharmaceutical industry in order to assure quality control. Using near-infrared (NIR) spectra measured on tablets in conjunction with multivariate calibration has been shown to meet these objectives. However, the calibration is typically developed under one set of conditions (primary conditions) and new tablets are produced under different measurement conditions (secondary conditions). Hence, the accuracy of multivariate calibration is limited due to differences between primary and secondary conditions such as tablet variances (composition, dosage, and production processes and precision), different instruments, and/or new environmental conditions. This study evaluates application of Tikhonov regularization (TR) to update NIR calibration models developed in a controlled primary laboratory setting to predict API tablet concentrations manufactured in full production where conditions and tablets are significantly different than in the laboratory. With just a few new tablets from full production, it is found that TR provides reduced prediction errors by as much as 64% in one situation compared to no model-updating. TR prediction errors are reduced by as much as 51% compared to local centering, another calibration maintenance method. The TR updated primary models are also found to predict as well as a full calibration model formed in the secondary conditions.