Comparison of two measurement methods with a gold standard as applied to 20-MHz-sonography and clinical palpation for ascertaining the thickness of pigmented skin tumours

Abstract

Aim: This paper focuses on different statistical methods for comparing two measurement methods with an additionally available gold standard. A given data example is used as the basis of the calculations. Method: We provide a complementary statistical analysis of a study presented by Hoffmann et al. [1] on sonometric and palpatory measurements of the size of pigmented skin tumours in 681 patients. Results: For comparing two measurement methods with respect to a gold standard, several statistical parameters assessing one measurement method can be used. In addition, there are further descriptive and some inference-statistical methods available. Conclusion: If there is a suitable categorization of the measurements, the comparison of the methods should be performed using the positive predictive values and kappa coefficients as descriptive measures. Moreover, the McNemar test can be used for comparing the differential accuracy of allocation. When investigating continuous measurements, a comparison using mere correlation analyses can lead to false conclusions. Therefore, we recommend the direct analysis of the individual measurement errors by means of numerical and graphical representations. The absolute values of the measurement errors can be compared using the sign test for paired samples.