Initial evaluation of quantitative performance of chromatographic methods using replicates at multiple concentrations

The introduction of a novel analytical method must be supported by consistent information about its quantitative potentialities; this is critical for whoever considers its utilization for an specific application. Unfortunately, literature abounds in papers proposing excellent chromatographic methods of analysis that have been subjected to comparatively poor quantitative evaluation. The methodology proposed in the present work makes use of some of the performance characteristics whose measurement is recommended in validation protocols; pertinent to this stage of method development are the detection and quantitation limits, the linear range and the repeatability. All this information can be calculated from the results of a calibration with several replicates at each analyte level. Replicates enable the calculation of reproducibility at several analyte levels and the estimation of the linear range; more important, replicates are necessary to detect changes in peak area standard deviation with analyte amount. Regression of calibration data by means of unweighted least-squares (ULSR) can only be performed in those cases in which homoscedasticity has been previously verified; heteroscedastic calibration data demand regression by means of weighted least-squares (WLSR), since ULSR results in gross overestimation of prediction limits at low analyte concentration. The proposal is used for the preliminary quantitative evaluation of a method for the determination of nine biogenic amines by means of pre-column derivatization with dabsyl chloride and separation of derivatives by RPLC. Limits of detection are calculated by a regression approach and by the classical signal-to-noise ratio method (S/N approach). No significant difference was detected for the amines limits of detection estimated by WLSR and by the S/N approach; ULSR estimated limits of detection are between 7 and 78 times larger than those obtained by the other two methods, as a consequence of the heteroscedasticity of calibration data.