Statistics in Nuclear Cardiology

Optimizing Assessment of Agreement between Like Continuous Measurements

抄録

Clinical researchers often need to quantify the degree of agreement, the “closeness,” between two quantitative methods of measurement taken on the same subject and same variable, either when a new or revised method is considered for use in place of an established one or when measurements are separated by time. An ideal model of agreement would capture the degree of exactly equal outcomes. Analysis should be based on differences between two methods of measurement. Inappropriate statistical methods, including correlation, e.g. Pearson’s r, are often used. But correlation, r, measures the degree of linear relationship between two variables, not the extent of agreement. A more informative and appropriate tool, the Bland and Altman (B&A) plot analysis, uses a graphical approach to present and assess differences between two measures. The differences between paired measurements are plotted on the Y-axis against the mean of the paired measures on the X-axis. The B&A chart allows for assessment of bias, both in size and consistency, over the range of measurements; it does not specify whether the agreement limits are acceptable; acceptable limits must be defined before. Additional reference lines can facilitate interpretation of the scatter plot.