Journal of Pharmaceutical and Biomedical Analysis 102 (2015) 535–543 Contents lists available at ScienceDirect Journal of Pharmaceutical and Biomedical Analysis j o ur na l ho mepage: www.elsevier.com/locate/jpba Multivariate figures of merit (FOM) investigation on the effect of instrument parameters on a Fourier transform-near infrared spectroscopy (FT-NIRS) based content uniformity method on core tablets Greg D. Doddridge, Zhenqi Shi ∗ SMDD, Lilly Research Laboratories, Indianapolis, IN 46285, United States a r t i c l e i n f o Article history: Received 17 June 2014 Received in revised form 14 October 2014 Accepted 17 October 2014 Available online 28 October 2014 Keywords: Near infrared spectroscopy Process Analytical Technology Content uniformity Multivariate figures of merit Method characterization a b s t r a c t Since near infrared spectroscopy (NIRS) was introduced to the pharmaceutical industry, efforts have been spent to leverage the power of chemometrics to extract out the best possible signal to correlate with the analyte of the interest. In contrast, only a few studies addressed the potential impact of instrument parameters, such as resolution and co-adds (i.e., the number of averaged replicate spectra), on the method performance of error statistics. In this study, a holistic approach was used to evaluate the effect of the instrument parameters of a FT-NIR spectrometer on the performance of a content uniformity method with respect to a list of figures of merit. The figures of merit included error statistics, signal-to-noise ratio (S/N), sensitivity, analytical sensitivity, effective resolution, selectivity, limit of detection (LOD), and noise. A Bruker MPA FT-NIR spectrometer was used for the investigation of an experimental design in terms of resolution (4 cm −1 and 32 cm −1 ) and co-adds (256 and 16) plus a center point at 8 cm −1 and 32 co-adds. Given the balance among underlying chemistry, instrument parameters, chemometrics, and measurement time, 8 cm −1 and 32 co-adds in combination with appropriate 2nd derivative preprocessing was found to fit best for the intended purpose as a content uniformity method. The considerations for optimizing both instrument parameters and chemometrics were proposed and discussed in order to maximize the method performance for its intended purpose for future NIRS method development in R&D. © 2014 Elsevier B.V. All rights reserved. 1. Introduction Since the issuance of the Process Analytical Technology (PAT) guideline by FDA [1], the number of applications of near infrared spectroscopy (NIRS) in pharmaceutical analysis has increased tremendously, permeating into a variety of aspects of R&D, man- ufacturing, and supply chain. These applications include material identification, polymorphism detection, process monitoring, coun- terfeit detection, etc. [2]. Among these applications, the use of NIRS to quantify active pharmaceutical ingredient (API) content in core tablets is one of the most popular areas given its direct impact on real-time release [3]. Meantime, the fact that multi-dimensional information of the entire manufacturing process is concentrated ∗ Corresponding author at: Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, IN 46285, United States. Tel.: +1 317 276 9431. E-mail address: shi zhenqi@lilly.com (Z. Shi). within a single tablet also makes such an application challenging in nature. Such multi-dimensional impacts include the effect of parti- cle size [4], compression force [5], etc., on the method performance of a NIRS-based content uniformity method. Due to the broad and overlapping spectral bands in near infrared wavelength range, chemometrics, i.e., multivariate analy- sis is typically used to decompose the raw spectra, extract the most representative information and correlate to the analyte concentra- tion. Given such a need, a considerable amount of chemometric literature focused on harnessing state-of-the-art algorithms to enhance the signal of the analyte of interest from raw spectra, such as orthogonal signal correction [6,7], net analyte signal [8,9], Wiener filtering [10,11], etc. In comparison, limited efforts have been spent to characterize and optimize the effect of instrument parameters on method performance [12–17]. Most papers eval- uated the impact of instrument resolution on model accuracy represented by error statistics. However, instrument optimization with respect to a list of figures of merit is routinely conducted as http://dx.doi.org/10.1016/j.jpba.2014.10.019 0731-7085/© 2014 Elsevier B.V. All rights reserved.