6.2. Obtaining a reference value for bias calculation

Three types of samples can be used for obtaining a reference value:

1. Certified reference materials (CRMs) with sufficiently small uncertainty of the reference value. CRMs should match both the matrix of the samples routinely analyzed by the validated method as well as the range of the expected concentrations of the analytes in routine samples.
2. Reference materials (RM) that do not have to have a certified uncertainty estimate. These can be materials characterized by a reference material producer, but whose values are not accompanied by an uncertainty statement or are otherwise qualified like materials characterized by a manufacturer of the material, materials characterized in the laboratory for use as in-house reference materials and materials subjected to a restricted proficiency testing. If there is no better possibility then materials distributed in a proficiency test and characterized by consensus values obtained from the participant results can also be used, but such consensus values generally have low reliability, leading to unreliable bias estimates. Again, the type of matrix and analyte content range should match that of routinely analyzed samples.
3. When no reference materials are available, bias can be investigated by spiking studies. Then the sample is split into two aliquots – one is analyzed in its original state and the other one is analyzed after a known amount of the analyte has been added. It is important that the spiked analyte must be dispersed into the sample in such a way that its molecules experience the same molecular environment as that of the native analyte. With many matrices (soil, fruits-vegetables, meat, plant leaves, …) this is difficult to achieve. In such cases the molecular interactions experienced by the analyte molecules originating from the spike differ from those experienced by the native analyte. As a result, the spiked analyte molecules behave somewhat differently from the native analyte and the obtained bias value may not accurately reflect the bias operating on the native analyte. In most cases, the native analyte is more strongly bound by the matrix than the spiked analyte, resulting in somewhat optimistic bias estimates from spiking experiments [ref 3, ref 12]. Therefore, bias (recovery)  studies by spiking are strongly subject to the observation that while small bias (good recovery) is not a guarantee of good trueness, large bias (poor recovery), however, is certainly an indication of poor trueness. Strictly speaking, trueness studies of this type only assess bias due to the effects influencing the added analyte. The smaller the recovery, i.e. the larger the bias affecting the method, the lower is the trueness of the method.

 Some important considerations:

1.       It is acceptable to use a particular reference material for only one purpose during a validation study: either for calibration or for evaluation of trueness, but not for both at the same time.

2.        How good is the matrix match between a CRM and a routine sample and at which point is it good enough? It is not possible to give a universal answer to this question. Depending on the matrix and the analyte(s), the sufficient similarity of matrices may be different and the decision has to be based on the physical and chemical properties of the matrix and the analyte. In the case of food commodity matrices some general directions can be found in SANTE/SANCO [ref 4], which relate to grouping of matrices according to their properties, with the assumption that extraction recovery of an analyte from the same group of matrices should be similar. A comprehensive treatment of the topic of matrix match and usage of surrogate/artificial matrices, together with a large number of examples from the area of bioanalysis (analysis of endogenous compounds) is available in [ref 55].

It is important, however, to note that in the case of LC-MS the situation is additionally complicated by the existence of the matrix effect as a bias component. This means that the question is not only about extracting the analyte from the sample, but also about possible co-extracting and co-eluting interferences. The following table intends to give some examples of reasoning in such cases. The table is by no means a complete and the recommendations given there should be taken with caution, as the table lists only matrices, not analyte-matrix combinations.

 

Table 1. Example situations of incomplete matrix match between samples and CRMs with comments.

Sample matrix	CRM or RM matrix	Comments	Recommendation
Drinking water	Sea water	Both are homogeneous matrices and drinking water can be regarded as the “limiting case” of sea water by salt content and matrix simplicity.	Sea water CRM can be used for validating a method of drinking water analysis and if only a small bias is detected then the method can be considered good from the bias perspective. If significant bias is detected then the method may in fact still be OK and bias correction certainly should not be applied.
Sea water	Drinking water	Sea water contains more components, especially salts, than drinking water and drinking water cannot be used for approximating sea water.	Drinking water CRM should not be used for validation of a method intended for sea water analysis.
A citrus fruit	Another citrus fruit	Citrus fruits are similar by almost all chemical constituents and acidity.	In general, citrus fruits can be used interchangeably. However, consider that the co-eluting matrix compounds may still be different and the matrix effects may vary from one citrus fruit to another. It may be a good idea to determine the matrix effects separately.
Apple	Another variety of apple	It turns out that apple varieties can differ significantly between themselves. The differences are not large from the point of view of analyte extraction, but can be large from the point of view of matrix effects (co-eluting compounds) [ref 27].	It is of course logical that if an apple CRM is available then it is used for bias determination. Care should be taken however, and it may be a good idea to determine matrix effects separately, as is explained in section 5.4.
Blood plasma from an ill patient	Blood plasma from a healthy volunteer	Although blood plasma properties of all humans should in principle be similar, it has been found that the blood plasma matrix can differ substantially between individuals [ref 39].	Again, it is logical that if a blood plasma RM or CRM is available then it is used for a bias determination. In addition, since generic blood plasma is easily available for bioanalytical groups and is reasonably homogeneous RMs can conveniently be prepared in the lab. Care should be taken however, as certain specific interferents can be different and therefore the success is greatly dependent on the specific analytes that are determined. It may be a good idea to determine matrix effects separately, as is explained in section 5.4.
Crude rapeseed oil	Refined rapeseed oil	Analyte extraction properties of these two matrices should be very similar, but an important constituent of crude rapeseed oil are phospholipids, which are known to be a major cause of LC-MS matrix effect.	Because of phospholipids in the sample matrix, which are absent in the RM/CRM, every care should be taken, that they will not co-elute with the analyte. Matrix effect profile, as explained in 5.3 can be very useful.

3.        It is strongly recommended to determine bias using at least two different reference samples. They can be of the same type or of different types.

 

Obtaining a reference value 

http://www.uttv.ee/naita?id=23478 

https://www.youtube.com/watch?v=kURtbQ7ACYs&t=10s