Relationships between statistical measures of agreement: Sensitivity, specificity and kappa

Abstract

Objectives: Cohen's kappa coefficient is presently a standard tool for the analysis of agreement on a binary outcome between two tests. In view of the ubiquity of the use of sensitivity, specificity, raw agreement and kappa in clinical studies, clearly it is advantageous to have a useful analytic relation connecting these measures of agreement. Methods: We elaborate on previous work, comment on other results appearing in the literature and discuss analytic formulas relevant to various problems connecting specificity, sensitivity and kappa. Results: For selected values of kappa that range from good to excellent, a graph of the curves representing minimal pairs of sensitivity and specificity is provided. Conclusions: The analytic formulas and graph could be potentially useful to clinicians and biostatisticians in better interpreting the outcomes of an alternative diagnostic test whenever the measures sensitivity, specificity and kappa are employed together. © 2008 The Authors.