Robust Calibration Design in the Pharmaceutical Quantitative Measurements with Near-Infrared (NIR) Spectroscopy: Avoiding the Chemometric Pitfalls

Quantification analysis with near-infrared (NIR) spectroscopy typically requires utilizing chemometric techniques, such as partial least squares (PLS) method, to achieve the desired selectivity. This article points out a major limitation of these statistical-based calibration methods. The limitation is that the techniques suffer from the potential for chance correlation. In this article, the impact of chance correlation on the robustness of PLS model was illustrated via a pharmaceutical application with NIR to the content uniformity determination of tablets. The procedure involves evaluating the PLS models generated with two sets of calibration tablets incorporated with distinct degree of concentration correlation between the active pharmaceutical ingredient (API) and excipients. The selectivity and robustness of the two models were examined by using a series of data sets associated with placebo tablets and tablets incorporated with variations from excipient content, hardness and particle size. The result clearly revealed that the strong correlation observed in the PLS model created by the correlated design was not solely based on the API information, and there was an intrinsic difference in the variances described by the two calibration models. Diagnostic tools that enable the characterization of the chemical selectivity of the calibration model were also proposed for pharmaceutical quantitative analysis.