Peter Vedsted

Background: Clinical research often involves study of agreement amongst observers. Agreement can be measured in different ways, and one can obtain quite different values depending on which method one uses.
Objective: We review the approaches that have been discussed to assess the agreement between continuous measures and discuss their strengths and weaknesses. Different methods are illustrated using actual data from the `Delay in diagnosis of cancer in general practice´ project in Aarhus, Denmark.
Subjects and Methods: We use weighted kappa-statistic, intraclass correlation coefficient (ICC), concordance coefficient, Bland-Altman limits of agreement and percentage of agreement to assess the agreement between patient reported delay and doctor reported delay in diagnosis of cancer in general practice.
Key messages: The correct statistical approach is not obvious. Many studies give the product-moment correlation coefficient (r) between the results of the two measurements methods as an indicator of agreement, which is wrong. There have been proposed several alternative methods, which we will describe together with preconditions for use of the methods.