Pathology laboratories generate substantial amounts of data in the on-going quality management of their processes/assays. If sufficient data (of adequate quality) is available the Single Laboratory Validation (SLV) approach is optimal for the determination of MU. Acting as an assessment of the entire resulting system, the SLV takes into account all available information (and therefore potential influences of uncertainty) on the final result. This reduces the likelihood of underestimation of MU in the final budget.
Using this approach the calculation of uncertainty is not applied to any particular sample. Rather the uncertainty is calculated for the procedure, and it is this uncertainty of the procedure that is then assigned to the sample. To be valid the value of the measured quantity must not be too different (within 2-3 x) the value(s) used to determine the uncertainty. Generally, for all medical laboratory assays, this criterion will be comfortably met as all data used for statistical analysis will be within the clinically relevant range for the given assay.
Uncertainty contributors appropriate for use of the SLV approach
From the categories above the following contributory categories should be used for this approach.
- Repeatability
- Reproducibility
- Bias
Utilising data from the following sources:
- IQC
- EQA
- Inter analyser variability
- Cross site comparability
- Result calculations
The SLV approach allows uncertainty from both random and systematic sources to be captured within the calculation. Each is calculated separately then combined to give an overall uncertainty according to the equation
Combined uncertainty = (random uncertainty2 + systematic uncertainty2)1/2
Random uncertainty
The random uncertainty contribution is derived from the within laboratory reproducibility, generally Internal Quality Control, recorded over a prolonged period of time, incorporating changes in:
- Operator
- Days
- Reagent batches
But performed by the same analytical method and covering the entire analytical process. As such, multiple levels of IQC should be used with results combined. Repeatability is measured as either the Standard deviation (SD) or Coefficient of Variation of the QC level depending whether a relative (CV) or absolute (SD) uncertainty is required. In the event the QC levels are near to the limit of detection of the assay (low levels) absolute uncertainties are more appropriate, whereas higher concentrations require relative (CV) uncertainties to be computed to reflect any proportional uncertainty influence.
Results are combined using the root of the sum of squares for each QC level i.e.
Random uncertainty = ((*QC Level 1)2 + (*QC Level 2)2)1/2
Where * refers to either the SD or CV of the QC level at that point depending on whether relative or absolute uncertainty is required.
Systematic uncertainty
The systematic uncertainty originates from the uncertainty associated with the deviation of the labs’ results from the true value and may be measured by any of the following methods:
- Repeat the same samples using a reference procedure and assess the “bias”
- Repeated analysis of certified reference materials (if available)
- Inter laboratory comparisons
- Spiking experiments
Repeatability data sourced from EQA results can be used for assessment of bias uncertainty.
EQA schemes regularly asses the performance of a site in comparison to others across the country. The data from these schemes allows a result to be compared across multiple hospitals and clinically significant differences or changes in this setting can be assessed. As a measure of proficiency testing, EQA in accordance with ISO/IEC Guide 17043 (Clause 3.7) is considered to include all EQA performed within the department.
Some assays are only assessed using EQA and have no associated IQC. In this scenario EQA is used as the sole determinant of assay performance.
To use EQA two components must be assessed.
- The Bias of the laboratory from the consensus result (in as well matched a group as the statistical analysis from the EQA scheme allows)
- The uncertainty of the consensus value derived from the same reagent and method matched data.
Both influences are calculated differently.
Laboratory Bias from consensus result
This is determined from the Root Mean Square (RMS) of results of successive EQA surveys by comparing the laboratory performance against the consensus mean/median as follows:
Bias = lab result – consensus result
RMS = ((Sum of all biases)2 / number of surveys)1/2
Consensus value uncertainty
U = Sample SD / (number of respondents)1/2
UCref = (((Sum of U)2) / n)1/2
When each of these components is calculated they are combined as the root of the sum of squares of each component as below to determine the uncertainty due to bias.
Systematic uncertainty = (RMS2 + UCref2)1/2
Combined uncertainty
Using the equation above the overall uncertainty for the entire process can be calculated by:
Combined uncertainty = (random uncertainty2 + systematic uncertainty2)1/2
Pathology Uncertainty 