Accuracy/BiasAccuracy may be defined as the degree of agreement between a test result and a reference value or an accepted result. Accuracy is numerically expressed as bias and is estimated according to:
It must be noted that the bias is not always a constant value. Bias often changes as the level of analyte changes (i.e., the bias is a function of the concentration of the analyte.) This will be addressed in the linearity section where the relationship between a device's test results and reference values is evaluated.
Accuracy requirements are very much dependent on the area of application and may be regulated. Consider, for example, medical devices intended for measuring blood glucose. In order for such a device to receive the Conformite Europeenne (CE) mark, it must meet the standard established by the International Organization for Standardization (ISO) for blood glucose accuracy. This standard (known as ISO 15197:2003) requires that 95% of blood glucose measurements be within +/- 15 mg/dL of the reference method at blood glucose levels less than 75 mg/dL, and that 95% of blood glucose measurements be within +/- 20% of the reference method at blood glucose levels greater than or equal 75 mg/dL.
One concern about this ISO standard is that it does not specify how far off target the remaining 5% of the results need to be. In order to ensure patient safety, accuracy data need to be assessed clinically based on the accepted Allowable Total Error (ATE) and the accepted Limits of Erroneous Results (LER). The ATE plot is, therefore, a very useful complementary tool in evaluating accuracy data. In the case of blood glucose, the Clarke Error Grid is an excellent companion to the ISO 15197 accuracy standard.