In this study, pharmaceutical grade sorbitol was used as a model system for comparison of Raman based library spectral correlation methods with more sophisticated methods of chemometric data analysis. Both crystallizing sorbitol (CS) and non-crystallizing sorbitol (NCS) from several manufacturers were examined. The Raman spectrum of each sample was collected and identified by correlation with a spectral library that included the CS spectrum but not the NCS spectrum. The average hit quality index (HQI) for the measured NCS spectra and the library CS spectrum was 0.966 whereas the average HQI for the measured CS spectra was 0.991. Both HQIs exceeded the 0.950 threshold that is commonly used for material verification. To enhance the discrimination between CS and NCS, a CS/NCS classification model was constructed using soft independent modeling of class analogies (SIMCA). SIMCA was able to positively identify CS and NCS solutions with no misclassifications. When CS was adulterated with low levels (0-5%) of ethylene glycol (EG) and diethylene glycol (DEG), the HQI values of the measured spectra and the CS library spectrum were still above 0.950. When the CS SIMCA model was applied to adulterated CS spectra, it determined that CS samples with adulterant levels as low as 2% were outside of the CS class. A quantitative PLS model was also applied to EG adulterated CS and resulted in a detection limit of 0.9% for EG. The results obtained from these studies highlight the importance of selecting an appropriate data analysis process for the detection of low level adulterants in pharmaceutical raw materials using Raman spectroscopic screening methods.
Published by Elsevier B.V.