5.2 Determining bias in practice

The following practical aspects are important for determining trueness:

1.    Obtaining the reference value (see also section 6.2)

Reliable reference values are essential for determining bias. There are essentially 4 approaches for obtaining reference values:

(a)   Certified reference materials (CRMs) with sufficiently small uncertainty of the reference value.

(b)  Reference materials without a certificate. This includes laboratory reference materials and samples of proficiency testings or interlaboratory comparisons.

(c)   Reference values obtained from spiking studies. Spiking means adding a known amount of the analyte to a known amount of the matrix, which may or may not contain the analyte. If the amount of the added analyte and the amount of the matrix are accurately measured and if the analyte is carefully dispersed into the matrix, then the increase of the analyte content in the matrix will be accurately known and can be used as a reference value.

(d)  Reference value for a sample can be obtained from the analysis with a reference method. A reference method in this context means a method that is more reliable and provides lower uncertainty than the method currently under validation.

Reference values provided by these four approaches have different qualities. Reference values from CRMs are generally considered the most reliable, but an important prerequisite is a matrix match between the CRM and the samples actually analyzed. Reference values of the reference materials without a certificate are usually less reliable and a matrix match is required as well as in the case of CRMs. Spiking can give reliable reference values if done carefully and correctly. The main issue is dispersing the spiked analyte in the matrix the same way as the native analyte. The possibility of this varies widely between matrices (easily doable with water and blood plasma, but very difficult with soil or minerals). If spiking is feasible, then the ideal matrix match can be achieved, because then the routine samples can be spiked. Lastly, measuring samples with a reference method can in principle give very reliable reference values, but this approach is seldom used because a suitable reference method is usually not available.

Some further aspects of reference values are addressed in section 6.2.

2.    Number of replicates: this should be higher than 2 and it should be proportionally larger if the intermediate precision standard deviation is higher. Replicate measurements are needed for separating systematic effects from random effects. The number of replicates is just as important in determining bias as it is in determining precision.

3.    Time range of analyzing the replicates: this time range should be at least few weeks (preferably few months). This way the obtained bias will be a long-term bias (see section 5.1).

 

The following video presents a practical example of estimating trueness/bias:

 

Trueness (CRM)

http://www.uttv.ee/naita?id=24826

https://www.youtube.com/watch?v=htHmkmbTcMI

The trueness/bias estimate obtained in this video is a bias according to Equations 1 and 2 in section 5.1. It embraces all four constituents as explained in 5.1 but this way of estimating bias does not enable “dissecting” the obtained bias into its constituents.