In this short offering we will discuss what the readily available sources of data are within the pathology laboratory. When starting to think about constructing our uncertainty budget we need to consider two things:

  1. What are the potential sources of uncertainty in our system that can explain the variability on our assay?
  2. Can we measure it? – and if yes, what do we do to measure it now?

Fundamentally we cannot do one without the other. Without data we cannot describe uncertainty associated with a contributor; equally a contributor cannot be modelled and adequately described without the presence of some data that we can (or already do) measure.

A useful starting point before doing any analysis is to check that the two criteria above tally up.

Having already described our 10 top uncertainty contributors in a previous post we will now describe the sources of data that we have at our disposal. These data sources will be the backbone of most of the analytical work we do in our uncertainty estimations. It is clear that some of these are extensions of what we have already identified as uncertainty sources.

This post will deal with data we are accustomed to using and in most cases will provide most of what we need. The second post will suggest what else can be used if/when the below sources are not available or suitable.

Internal Quality Control (IQC)

IQC is a huge topic, and not one for this post (it may be time to build another site!) However, there is one undeniable fact. All our assays need to have some form of IQC and as a result we have a plentiful supply. I plan a series on how to handle/process/analyses IQC so look out for that in the future. A summary of the process will be published next month (November 2017) in Pathology in Practice so again keep an eye out for that.

External Quality Assurance (EQA)

As with IQC, we are required to have EQA in place for all our assays – that is of course providing there is an EQA scheme available or that EQA as a concept is possible taking into account sample stability/transferability. Our EQA gives us data for assessing how our assays perform in comparison to equivalent, or similar techniques in other centres. As such it is a very important source of information for us to use when determining our uncertainty and assay performance. Commonly, EQA data is considered to be centred around determining, and quantifying, the presence of systematic error – or bias. Uncertainty estimation often doesn’t include this, and differentiates itself from total error accordingly. However, there are aspects of the data that we can use to hep us in building our uncertainty budget.

 Inter Analyser Variability

Often laboratories will have more than one analyser or system to run the same assay. Having a back up analyser or multiple analysers in use at the same time is commonplace. Our uncertainty in such situations should be considering whether our results will differ dependent on which of the analysers the sample was run.

Methods to assess, and quantify, the degree of variance associated with using different analysers include (but are not limited to):

  • Inter analyser variability during analyser validation/verification
  • Reagent lot validation if performed across analysers
  • Patient median monitoring – reliant on a known stable population of patients (this will come in a later topic)


Cross Site Comparability


With centralisation of pathology services becoming commonplace consideration must be given regarding what the results we are reporting mean in the context of our patient population and also within our networks. Particularly, comparability of results between the sites within that network can impact on the interpretation of the results. Patients may be treated across different sites and their historical results will follow them. As such we should know what constitutes a significant difference between results between sites, considering both biological (the patient) and analytical (the assays) variability.

There are many ways that such data can be collected:

  • Sample sharing schemes
  • Monitoring of carefully selected patient medians (as above with inter analyser variability)
  • Performance in proficiency testing schemes

Whichever are used, the data produced from such exercises gives us a valuable source of data to incorporate into our uncertainty budgets. These, along with some of the less common data sources should at the least be considered before we embark on trying to construct our uncertainty budget. In the next post we will look at some of the less common, but equally important sources.