Regressions for Quantitative Diagnostic Testing

Abstract

Linear regressions are applied for validating quantitative diagnostic tests, but traditional analysis of variance is not good enough for the purpose. Also traditional linear regression assumes, that, either dependent or independent variable comes with an amount of residual uncertainty. In clinical research, particularly, clinical chemistry and laboratory medicine, there is no gold standard. Not only the x- but also the y-variables have been measured with a (same) amount of uncertainty. Traditional linear regression is, then, not appropriate, and regression analyses have to include the uncertainty of both the y- and x-variable. Two methods for the purpose are Deming regression and Passing-Bablok regression. Particularly, The latter of the two has many advantages. It is non-parametric, which means, that nonnormal distributions of the data are no problem. It also is robust, because medians rather than means are applied, and, so, outliers are no problem either, and overdispersion is taken into account.