Deviations from linearity in statistical evaluation of linearity in assay validation

Linearity and linear range are the key evaluations of the accuracy in assay validation. The average deviation from linearity (ADL) and the sum of squares of deviations from linearity (SSDL) have been proposed for assessment of the linearity. However, both ADL and SSDL do no consider the variability of the assay for evaluation of linearity. Therefore, we proposed the coefficient of variation of deviations from linearity (CVDL) as an alternative measure for the linearity assessment. For the inference of evaluation of linearity, we proposed testing procedures based on generalized pivotal quantities (GPQ) of ADL and CVDL for evaluation of linearity. The simulation studies were conducted to empirically investigate the size and power between the three methods. The simulation results show that all three methods adequately control size. However, the ADL method is uniformly more powerful than the other two methods. A numeric example illustrates the proposed methods. © 2009 John Wiley & Sons, Ltd.

Duke Scholars

Author Shein-Chung Chow Biostatistics & Bioinformatics, Division of Biostatistics