Choosing the calibration model in assay validation.

摘要

Data transformations and weighting schemes are normally used to obtain the best-fit of standard curves in bioanalysis and the calibration model is usually selected during prevalidation. In the present study, a comparison has been made between unweighted and weighted (1/x, 1/x2, and 1/square root of x) regression models with or without an intercept in achieving the best-fit for the standard curve of CDRI compound 81/470, a new anthelmintic agent, in cow milk. Validation samples in milk at the LLOQ, medium, and high concentrations were also analysed by each of the calibration models. An unweighted regression equation with an intercept overestimated the concentrations at the LLOQ. An unweighted equation without intercept and weighted equations with or without an intercept significantly minimized the bias at the LLOQ without distorting the results at higher concentrations. Hence, an unweighted equation for a straight line passing through the origin was found to be the best model for a standard curve of 81/470 in milk. Similar results were obtained for 81/470 and UMF-078 in serum and plasma, respectively. Bioanalysts should routinely test these models to obtain the best fit model for their calibration curves as part of their assay validation not during prevalidation.