Chance-corrected measures of the validity of a binary diagnostic test

Chance-corrected measures of the validity of diagnostic tests can be a valuable supplement to traditional measures which do not take chance-agreement between diagnostic test and true disease status into account. Recently, several authors have propagated chance-corrected measures of sensitivity and specificity, the most commonly employed indices of validity of binary diagnostic tests. A major limitation of these measures is their dependence on disease prevalence. In this paper, we propose simple alternative chance-corrected measures of sensitivity and specificity which do not share this limitation. The chance-corrected measure of sensitivity can be expressed as 1--negative likelihood ratio, and the chance-corrected measure of specificity can be expressed as 1--inverse of the positive likelihood ratio. The properties of the proposed measures are described and graphically illustrated, and formulas are given for point and interval estimation of the measures.