Bland-Altman Analysis A Paradigm To Understand Correlation And Agreement

The distortion of -27.2 units is represented by the distance between the X axis, which corresponds to zero differences, and the line parallel to the X axis at -27.2 units. This negative distortion appears to be due to measurements greater than 200 units, while at lower concentrations, the data are closer. A negative trend seems to appear along the graph, as figure 3 best shows. Drawing a regression line of differences could help identify a proportional difference (10 – 12). The visual examination of the action allows us to assess the overall agreement between the two measures. In our example, we can summarize the lack of agreement by calculating the distortion estimated by the average difference (d) and the standard difference of differences (s). We expect that most of the differences will be between d-2s and d-2s, or more accurately 95% of the differences between d -1.96s and d -1.96s if the differences are distributed normally (Gaussian). The normal distribution of differences should always be checked, for example. B by drawing a histogram. If this is distorted or has very long tails, the acceptance of normality may not be valid. According to the example in Table 1, the measurements of the two methods are not normally distributed, but on the other hand, there appear to be differences (Figure 4). Statistical tests should always be used to determine whether the distribution is normal, because in some cases normality cannot be determined simply by observing the histogram diagram. If statistical software is available, a normal distribution test (such as the Shapiro-Wilk test (13), the Agostino-Pearson test (14), the Kolmogorov-Smirnov test (15) can be performed, for the hypothesis that the distribution of observations in the sample is normal (if P < 0.05 then reject normality).

If the differences are not normally distributed, a logarithmic transformation of the original data may be attempted. Keywords: biostatistics; Bland-Altman analysis; Correlation analysis Match limits. Correlation is a statistical technique that determines whether and to what extent pairs of variables are related.