|
|
|
Dual
vision procedure for solving clinical paradoxes
|
|
|
|
Kappa, Yule and
phi are the usual approaches for analysing the agreement in binary
cases. They have failures which can be illustrated with 5 paradoxes: 1st
paradox:
It occurs when with different values of chance agreement; the values of
kappa for identical values of raw agreement can be more twofold smaller
in one instance than in the other [1] 2nd
paradox:
When unbalanced marginal totals produces higher values of kappa, Yule
and phi, than more balanced totals [1] 3rd
paradox:
Kappa, Yule and phi can be null, no matter the observed agreement in the
study [2] 4th
paradox: Kappa, Yule
and phi can be negatives, no matter the observed agreement in the study [2] 5th
paradox: When b = 0, or c = 0, Yule tends to 1 (i.e., perfect
agreement) no matter the observed agreement in the study [2]. Kappa,
Yule and Phi cannot solve the Feinstein & Cicchetti (1st and 2nd)
paradoxes. As well as the last three ones. The best performance of the
new procedure is due to the use of the clinical agreement concept,
instead of the statistical one. The
point is: The
reality is the unique truth. The observed level of the agreement in a
clinical study is the reality. Kappa, Yule and phi are only mathematical
theories for explaining the reality. It has been shown that the
mentioned approaches have failures, and therefore it is suggested that
they should be replaced by another approach that performs better, like
the dual vision procedure. The
full explanation can be viewed below. References:
1.
Feinstein, A.R. and Cicchetti, D.V., High agreement but low kappa, I:
the problems of two paradoxes, J. Clin. Epidemiol. 1990, 43:543-549 2.
Azzimonti Renzo JC, Failures of Common Measures of Agreement in Medicine
and the Need for a Better Tool: Feinstein's Paradoxes and the Dual
Vision Method
|
