Figure 15.2. A Bland-Altman diagram from Clinic 104 in relation to the DB13 laboratory data. This page also shows the standard deviation (SD) of the differences between the two test methods (called SD bias) and the 95% chord limits, calculated as the average difference (Bias) plus or minus 1.96 times sD. Based on a normal distribution of differences and sufficient sample size, the sample average and the sample SD are fairly close to their population, 95% of the differences between the two test methods would likely be within the range described by the 95% agreement limits. For each methodological validation, a comparison of methods is required to estimate inaccuracy or systematic error. CT tests in our laboratory are usually verified using a combination of comparative patient samples with a reference laboratory and samples of PT (PT) or commercial standards (certified reference material), if available. Figure 8.2 shows the comparison of the busulfan method (n-69) by LC-MS/MS with GC-MS (reference method) with the differential diagram (Bland-Altman-Plot). The results were compared to a standard linear regression. The acceptance criteria were that the average difference between the results 20% and the slope – 1.0 ± 0.1, 95% of the Confidence Interval Intercepte 0 and r2 > 0.98. If it is broad (as clinically defined), the results are ambiguous. If the boundaries are narrow (and the distortion is tiny), then the two methods are substantially equivalent.
Compliance limitations include both systematic errors (bias) and random errors (precision) and provide a useful measure for comparing likely differences between different results measured using two methods. If one method is a reference method, compliance limits can be used as a measure of the total error of a measurement method (Krouwer, 2002). The pearson values r from 0 to 1.0 are a positive correlation; 1.0 is a perfect correlation. Laboratory experts use the Pearson formula to evaluate the range of two values such as tests or to compare assay results with the standard or potassium results previously assigned. Most operators set a r value of 0.975 (or r2 of 0.95) as a lower correlation limit; A pearson value r less than 0.975 is considered zero because it indicates unacceptable variability in the reference method. One of the main applications of the Bland-Altman plot is to compare two clinical measurements, each of which has produced an error in its measurements.  It can also be used to compare a new technique or measurement method with a gold standard, because even a gold standard does not imply it without error – and should not involve it.  Software that provides Bland Altman plots is available on Analysis-it, MedCalc, NCSS, GraphPad Prism, R or StatsDirect.