Variations in predicted values from near-infrared spectra of samples in vials by using a calibration model developed from spectra of samples in vials: causes of the variations and compensation methods

The causes and compensation methods of variations in the predicted values from near-infrared (NIR) spectra of samples in vials by using a calibration model developed from spectra of samples in vials were studied. From the viewpoint of the path length and the peak height of spectra in the NIR region, vials with 1 mL volume (external diameter: 8 mm, inside diameter: about 6 mm, length: 40 mm) were selected as disposable vials for the samples. To investigate the causes of the variations in the predicted values, the optical path lengths and transmittance spectra of empty vials in five lots produced during the last four years were examined. The results showed that there were some differences in the path lengths of the vials and in the intensities of peaks near 7130 and 4515 cm-1, which were attributed to the hydroxyl groups in the spectra of the vials among the five lots. The calibration models for the concentration of isopropyl alcohol (IPA) in toluene contained in vials of the same lot were developed. To search the pretreatment methods for the best calibration model with small variations in the predictive values of the samples in the vials, six kinds of pretreatments (no pretreatment, baseline correction, first derivative, second derivative, multiplicative scattering correction (MSC), and standard normal variate (SNV)) were compared. The results indicated that MSC is a very effective method with small variation in the predicted values from the spectra of samples in vials. It was found that the variations in the predicted values are caused mainly by variations in the path lengths of vials. In real quality control, the NIR-chemometrics method is requested for long-term traceability and good reproducibility of the predictive value. Therefore, it is crucial to watch changes in the path length under a fixed path length as a reference. From this standpoint, modified MSC, which uses a particular spectrum as a reference instead of the average spectrum of a sample set, was proposed in the present study, and it was demonstrated that the variations due to the individual differences in the vials can not only be compensated, but the path lengths of vials could also be estimated by the modified MSC. Moreover, a calibration model for the toluene concentration in silicone oil was developed to confirm the effects of modified MSC by using another sample set. The results showed that the compensation of vial path lengths using the modified MSC is also useful for other samples, similar to the case of the sample of IPA in toluene.