ICH Recommended parameters for method validation

One mechanism that is proposed to ameliorate this problem is to first identify the product being tested before quantifying the analyte of interest. Using a well-established algorithm for spectral matching, such as principal component analysis followed by an appropriate qualitative pattern recognition routine like the Mahalanobis distance calculation, would be a viable approach. 

A second challenge to overcome is the linearity issue. In chromatographic and simple UV/Vis (Beer s law) methods, linearity of response to analyte is shown by making serial dilutions of a concentrated standard. These solutions are either injected into the HPLC or read on a spectrometer. The concentration is plotted against the area/peak height of the eluted peak (HPLC) or the absorbance value (UV/Vis). With NIR, things are not that simple. 

Usually, the linearity of a NIR spectroscopic method is determined from the multiple correlation coefficient (R) of the NIR predicted values of either the calibration or validation set with respect to the HPLC reference values. It may be argued that this is an insufficient proof of linearity since linearity (in this example) is not an independent test of instrument signal response to the concentration of the analyte. The analyst is comparing information from two separate instrumental methods, and thus simple linearity correlation of NIR data through regression versus some primary method is largely inappropriate without other supporting statistics. 

An analyst could extract a pure signal, directly correlated to the analyte of interest (matrix effect) only with major effort it is illogical to compare the dosage form spectra to pure standards or solutions of standards at various concentrations. Calculating a correlation coefficient between two distinctly different methods is not the same as the linearity for chromatographic methods. 

Poor values for linearity are usually the result when only production samples are used for development of quantitative calibrations. The low R values are a result of the narrow range of active concentrations due to the quality that is inherent in pharmaceutical processes. A typical pharmaceutical process gives samples that vary little from the nominal value. Typical ranges for tablet assays fall within a range of 97 to 102% of label claim. This 5% range, coupled with errors in the HPLC analysis, lead to a poor correlation line. In HPLC or UV/Vis spectroscopic methods, an R value approaching unity is common the USP recommends a value no lower than 0.995. A typical NIR linearity, based solely on actual production samples, is often closer to 0.8 than 1.0 due primarily to the narrow range. 

---