Abstract
The purpose of internal quality control (IQC) is to detect clinically important errors in the analytical process before reporting patient results. Commonly the result for an IQC sample is compared with a target range that comprises the mean plus or minus two standard deviations (SDs) obtained under optimal conditions. A problem with this approach is the high incidence of false rejections, since statistically one in 20 IQC results will fall outside of this range even if the method is in control. To overcome this problem Westgard 1 proposed a set of rules to evaluate more than a single IQC result. However, a recent audit of IQC procedures in the south-east of England has exposed wide variation in practice. 2 Of particular concern was the finding that a significant number of laboratories accepted failed IQC and allowed patient results to be reported. The commonest reason given was that the failed IQC was not deemed to be clinically significant and often this decision was taken by relatively junior staff. This begs the question ‘should IQC criteria reflect some analytical ideal or merely ensure that the results produced are fit for clinical purpose?’ If we are only concerned that results are fit for purpose then IQC procedures should be formulated to reflect this.
Current practice is to test the null hypothesis that the IQC result is not statistically different from a target value. Alternatively, if we are only concerned that there is no significant analytical bias, then it would be appropriate to test the alternative hypothesis that the IQC result belongs to a distribution with significant bias. Figure 1 shows hypothetical frequency distributions (with identical imprecision) for an IQC material in which there is no bias (Figure 1A) and maximum allowable positive and negative bias (Figure 1B and C, respectively). If the IQC result is less than B mean − 2SD then there is unlikely to be significant positive bias, and if greater than C mean + 2SD then significant negative bias is also unlikely. In other words, the conventional A mean ± 2SD control limits could be replaced by C mean + 2SD to B mean − 2SD. Provided the difference B mean − A mean and A mean − C mean is greater than 4SD units (so that there is minimal overlap between the three distributions) then the effect will be to widen IQC limits and reduce the incidence of false rejections. If B mean − A mean and A mean − C mean differ by less than 4SD units then the assay was unfit for purpose in the first place.
Hypothetical frequency distributions for an IQC material with no bias (A) and maximum allowable positive (B) and negative (C) biases (assumed identical imprecision in all cases)
DECLARATIONS