Introduction

Stability testing for EQA samples is important to proof that the analyte under investigation has the same assigned value during the analysis period.  Therefore it is required by ISO 17043, that refers for details to ISO 13528.  The algorithm proposed by ISO 13528 is vulnerable to wrong conclusions, though.  It can be easily proven that, if the minimum required samples for stability testing are taken, there is a very low chance of accepting the sample for stability, even when the parameter value does not change during the analysis period. A simple test for stability testing is proposed.  It is based on equivalence testing and minimising the number of wrong EQA evaluations.

Materials an methods

Stability testing requires the proof that the difference in mean analyte concentration between two distinct time points is lower than a certain limit. 

 In a first case, the maximum allowed difference between the means of the two distinct time points is derived by assessing the increase in probability of falsely flagging a result from a well performing laboratory if the results are analysed without distinction between analysis date.  False flagging means that the laboratory is flagged for its Z-scores or for its fixed deviation with respect to the assigned value. 

Equivalence testing is performed by comparing the confidence interval around the difference between the two means with the limit of the maximum allowed difference. 

The method was applied on the results obtained from the Belgian external quality assessment scheme for hematology.  Investigated parameters were reticulocytes, hematocrit, hemoglobin, thrombocytes, MCV, red blood cells and white blood cells.  Only results analysed on a Sysmex system were taken into consideration.

Results

For both samples, lack of stability was observed for hematocrit and MCV. For all other parameters, proof of sufficient stability was found.

Discussion

The current approach mentioned in ISO 13528 is prone to wrong conclusions.  A simple t-test with a higher limit of samples is not an option neither because this test does not respond to the question of proving that the difference between two measurements is sufficiently low. Equivalence testing is an elegant solution, but needs a limit that defines when the means of the measurement times are close enough to each other. For this reason, an algorithm was developed to draw the link between the maximum difference between the two means and an allowable, very slight, increase of the risk of wrongly evaluating EQA results.  The approach can be applied to every scheme for quantitative results, and, if laboratories report the time of analysis, can even be used based on the reported EQA results and hence, doesn’t need any extra laboratory analysis.