Capability of discrimination: application to soft calibration methods

Given a nominal concentration, in order to know the behaviour of an analyti cal procedure in samples with similar concentrations, the minimum discrimin able concentration is defined as the smallest concentration of the analyte in a sample which can be distinguished, with probability 1 - beta, from the , nominal value. This definition generalises the concept of minimum detecta ble net concentration established by ISO norm 11843 which is restricted to the case in which the nominal concentration is zero. Given an analytical pr ocedure with a well-established net detectable concentration (detection lim it), it may not be possible to discriminate this same concentration when th e procedure is used in samples with a much higher nominal concentration. Fo r this reason, the discrimination capability is a, criterion for the select ion of an analytical procedure when it is going to be used to determine con centrations well above its detection limit. The discrimination capability i s established as a hypothesis test based on the data of a calibration carri ed out in a range of concentrations which contains the nominal value. As an application, the discrimination capability has been estimated when th e concentration of a sample test is obtained by means of a partial least sq uares (PLS) calibration. In this case, the proposed procedure is composed o f three steps, the first of which consists of the soft multivariate calibra tion. The second step is the evaluation of the discrimination capability by a regression of the concentration found with the multivariate calibration versus the true concentration of a new set of reference samples. The standa rd deviation of the regression estimates the repeatability (analytical proc edure and soft calibration jointly) at the concentration range analysed in the regression. In the third step, the capability of discrimination calcula ted is applied to a new test sample if the repeatability has not changed at the level of concentration considered. The procedure developed has been applied to the determination of benzaldehy de by means of differential pulse polarography (DPP), where univariate cali bration cannot be applied and a PLS calibration is appropriate. The capabil ity of discrimination has been evaluated at two different concentration ran ges: from 0.10 to 1.05 muM, and from 0.0199 to 0.1740 mM, with an estimated repeatability of 3.1 X 10(-2) muM and 3.2 x 10(-3) mK respectively. The "D PP + PLS" capability of discrimination performance is analysed by means of the "false noncompliance" and the "false compliance" probability in the cas e that the analysis method has to discriminate differences lower than 10% o f the nominal concentration with sufficient warranty. (C) 2001 Elsevier Sci ence B.V. All rights reserved.