Variance function as basis for assessment of test performance: methodological studies with two assays of...

We evaluated the usefulness of a recently described procedure to assess the analytical performance of an assay. To demonstrate the advantages of this approach, we compared the performance of two analytical systems for determining prostate-specific antigen (PSA). Triplicate measurements of PSA with the IMx (Abbott) and the ACS 180 (Ciba Corning) were used to calculate the variance function. This function was the basis for the derivation of the critical limit (LC), the limit of detection (LD), the power of definition (PD), and the lower limit of the quantification interval. The standard deviation of the blank was extrapolated by means of the variance function. LC was calculated as the concentration at which the normal distribution of the blank intersects an adjacent normal distribution (with a defined overlap, e.g., 5%). The mean of the adjacent normal distribution represents the LD. The PD is a new mathematical approach to describe the analytical sensitivity of an assay in different ranges of the quantification interval. The procedure is statistically well defined and allows one to obtain the data on the test performance directly from patients' samples, without artificial zero controls. Therefore, use of the variance function could be a general model for the assessment of the analytical performance of an assay.