Scattering Orthogonalization of Near-Infrared Spectra for Analysis of Pharmaceutical Tablets Zhenqi Shi and Carl A. Anderson* Graduate school of Pharmaceutical Sciences, Duquesne University, Pittsburgh, Pennsylvania 15282, and Duquesne University Center for Pharmaceutical Technology, Pittsburgh, Pennsylvania 15282 The paper explores scattering orthogonalization as a preprocessing technique to reduce physical interference and maintain chemical information in near-infrared (NIR) spectra of pharmaceutical tablets. Samples used in this study were tablets compressed at five compression forces; they were composed of theophylline, lactose, and micro- crystalline cellulose (PH200). The NIR spectra were orthogonalized against the reduced scattering coefficients (representative of physical interference of scattering), and concentrations of all constituents were predicted. The robustness of predictions was compared to the widely employed standard normal variate (SNV) for the specificity of removing interference representative of physical pa- rameter (such as tablet density). Group-wise cross-valida- tion (groups were based upon similar chemical compo- sition) and prediction demonstrated the enhanced robustness on prediction of chemical information via scattering or- thogonalization in comparison to SNV. When compared to the SNV, scattering orthogonalization demonstrated an improved capacity to reduce physical interference while maintaining spectral variance attributable to chemical information. The improved capacity is expected to be useful for spectroscopy-based multivariate model calibra- tion and continuous model update. The number of near-infrared (NIR) spectroscopic applications in pharmaceutical analysis has grown significantly in the past decade. 1 Much of its appeal is due to the fact that little to no sample preparation is required, and non-invasive detection makes NIR spectroscopy a suitable tool for online process monitoring. NIR spectroscopy contains both chemical and physical information which are typically extracted using multivariate modeling tech- niques; these models can then be used to monitor process variations during routine pharmaceutical unit operations. In most cases, data pretreatments are performed to suppress physical interference prior to calculation of multivariate models for predict- ing chemical information. Potential sources of physical interfer- ence include particle size of a powder mixture and density of a pharmaceutical compact, and so forth. Pretreatments are typically employed to reduce error induced by physical interference and enhance signal-to-noise in multivariate models designed to predict chemical information. Since physical interference is typically observed as a baseline shift or slope in an NIR absorbance spectrum, the most widely used data pretreatments are additive and multiplicative scattering corrections, for example, standard normal variate (SNV) and multiplicative scattering correction (MSC). It is understood that scattering is both wavelength and absorption (constituent) de- pendent, and thus, single slope or intercept correction is often insufficient to remove the effects of physical interference. 2-4 It was recently reported that traditional scattering correction, such as MSC, removed chemical information from NIR spectra as a result of reducing baseline shift (attributed to scattering caused by physical interference) when NIR transmittance data (log 1/T) were calibrated against gluten concentration. 4 Thus, an improved data pretreatment method is desirable, ideally one which selec- tively removes from spectra information attributable only to physical interference (i.e., noise for a chemical calibration). There are two fundamental events that occur as NIR light impinges upon a sample, absorption and scattering. 5,6 Absorption reduces the intensity of photons of specific energy because of the altered molecular dipole of a bond. The parameter used to describe absorption is the absorption coefficient, μ a , which is defined as the probability of absorption per unit path-length. Scattering is caused by variations of refractive index at an interface; therefore, physical parameters such as sample density and porosity can have a dramatic effect on scattering events. Examples of these in pharmaceutical products (such as tablets) include air-solid and solid-solid interfaces. Since diffuse reflectance is an elastic scattering event, there is no energy change or photon intensity loss as a result of scattering. Using the diffusion approximation to the radiative transfer equation, scattering can be described by the reduced scattering coef- ficient, μ s ′, which is the probability for scattering per unit path- length. Absorption and reduced scattering coefficients together are referred to as the optical properties or optical coefficients of a material. Although absorption and scattering can be understood as independent interactions between photons and * To whom correspondence should be addressed. E-mail: andersonca@ duq.edu. Phone: 412-396-1102. Fax: 412-396-4660. (1) Ciurczak, E. W.; Drennen, J. K. In Handbook of Near-Infrared Analysis; Burns, D. A.; Ciurczak, E. W.; Eds.; Marcel Dekker: New York, 2001; pp 609. (2) Burger, T.; Kuhn, J.; Caps, R.; Fricke, J. Appl. Spectrosc. 1997, 51, 309– 317. (3) Burger, T.; Ploss, H. J.; Ebel, S.; Fricke, J. Appl. Spectrosc. 1997, 51, 1323– 1329. (4) Martens, H.; Nielsen, J. P.; Engelsen, S. B. Anal. Chem. 2003, 75, 394– 404. (5) Groenhuis, R. A. J.; Ferwerda, H. A.; Ten Bosch, J. J. Appl. Opt. 1983, 22, 2456–2462. (6) Keijzer, M.; Star, W. M.; Storchi, R. M. Appl. Opt. 1988, 27, 1820–1824. Anal. Chem. 2009, 81, 1389–1396 10.1021/ac802105v CCC: $40.75  2009 American Chemical Society 1389 Analytical Chemistry, Vol. 81, No. 4, February 15, 2009 Published on Web 01/22/2009