Comparison between NIR and/or Raman Spectroscopy based on their Predictability of Tablet Dosage

Quality Control (QC) in the pharmaceutical industry is a key activity in ensuring medicines quality, safety and efficacy for their intended use. QC departments are responsible for all release testing of final products but also all incoming raw materials. Therefore Near Infrared Spectroscopy (NIRS) and Raman Spectroscopy are important techniques for identification and qualification of samples. The goal was to use spectral information at specific wavelengths to build regression models based on Partial Least Squares Regression (PLS) to predict the dosage of tablets. The potential benefit of combining orthogonal techniques was shown in different publications. T. Nӕs et. al. (Chemometr. Intell. Lab., 124 (2013) 33-42) presented the methodology of multiblock regression based on combinations of orthogonalization and PLS using NIRS and Raman spectroscopy. L.P. Brás et. al. (Chemometr. Intell. Lab., 75 (2005) 91-99) and C.C. Felício et.al. (Chemometr. Intell. Lab., 78 (2005) 74-80) used multiblock PLS to compare and combine NIR and MIR. Therefore the combination of spectroscopic techniques was investigated and multiblock PLS (MB-PLS) has been applied.
In our work tablets containing two different active pharmaceutical ingredients (API) [bisoprolol, hydrochlorothiazide] were analyzed with Raman- and NIR Spectroscopy. These tablets are available with different potencies of both APIs, forming five groups. Raman Spectroscopy was performed with the spectrometer TruScan RM Material Verification Analyzer (now Thermo but our instrument is Analyticon, Rosbach vor der Höhe, Germany). The spectrometer was equipped with a silicon CCD detector. The measured spectral range was between 2800 and 250 cm-1 with a 8 cm-1 resolution. For the NIR measurements the multi purpose analyzer (MPA) together with the software Opus 6.5 (both: Bruker Optik GmbH, Ettlingen, Germany) was used. The samples were measured in external transmission mode. The spectrometer was equipped with a RT-InGaAs detector. The measured spectral range was between 10000 and 5800 cm-1 with a 8 cm-1 resolution and averaging 32 scans. MVDA was performed using MATLAB 7.10.0 (R2010a) (MathWorks®, Natick, MA, USA) together with PLS Toolbox 7.3.1 (Eigenvector, Wenatchee, WA, USA). Prior to the MVDA pre-processing the data was required. For MB-PLS a normalization of the intensity of both X-blocks was necessary, otherwise the calibration will give more importance to the one with the higher rank. For all PLS-based models a leave-one-out cross-validation (LOO CV) was used to choose the optimum number of latent variables.
Single PLS models using NIR spectra showed the highest predictive power for both APIs compared with the other models. The MB-PLS showed that each spectroscopy contains information not present or captured with the other spectroscopic technique, thus demonstrating that there is a potential benefit in their combined use for both discrimination and quantitation purposes. MB-PLS showed better results than single PLS for Raman spectra, but not than single PLS for NIR spectra. Future work could be directed to try other multiblock techniques like SO-PLS and to use a different reference method like HPLC to build quantitative models.