Clinical decision-making as the basis for assessing agreement between measures of the International Normalized Ratio

Background: The now classic approach of Bland and Altman is often used to assess the level of agreement between International Normalized Ratio (INR) measures. However, we are concerned that this method does not define agreement in a clinically meaningful way. Agreement between measures should be characterized explicitly in terms of clinical decisions that result from INR measures. Objectives: To develop and validate an extension of the Bland-Altman method to assess agreement between INR measures, based explicitly on the way clinicians make decisions. Methods and results: We developed a clinically based graphical method to estimate the level of agreement between measures of INR. We identified clinically relevant INR ranges using epidemiologic and clinical evidence regarding risk and expected outcome at different INR ranges. Clinical decisions were expected to agree within these INR ranges and, therefore, the ranges became the basis for establishing agreement between measures. We used paired INR measures and resultant clinical decisions measured during a previous prospective study to validate and compare the accuracy of our model to those of Bland and Altman's and other published models. Our method more accurately predicts when warfarin dosing decisions differ than the Bland-Altman method (P < 0.02). Our method is also superior to other published methods, particularly at the important task of identifying when measures lead to discrepant clinical decisions. Conclusions: We introduced and validated an improvement of the Bland-Altman method to assess agreement between INR measures. Our model is superior because it is based explicitly on factors that influence clinical decision-making.