The CLSI document EP21, which is about total error is being revised. First, some background.
Clinical laboratory evaluations have been hampered for many years by a recommended modeling of errors, which unfortunately leaves some errors out of the picture. The typical model adds two times the standard deviation of a candidate method (evaluated from a precision experiment) to the average bias (evaluated from regression between a candidate and comparative method). This gives an estimate of the location of 95% of the differences between the candidate and comparative method (given that certain assumptions are met). This location can be compared to a set of “acceptable limits.” Because one is estimating parameters (average bias and imprecision) it is perfectly acceptable to discard outlier results.
The main problem with this method is one would also like to know the location of the remaining 5% of the differences. If these differences are large enough (e.g. beyond the set of acceptable limits), then they might cause patient harm. Since many clinical laboratories report a million results a year, up to 50,000 results could potentially harm patients for an assay that might have been judged acceptable according to the modeling method just discussed.
EP21 changed this by dropping modeling. It simply observes differences between the candidate and comparative method. Particularly useful is the “mountain plot” which shows probability vs. differences. Important to this analysis is that no data can be discarded.
The EP21 analysis highlights outlying observations, whereas the modeling method suppresses them. But it is precisely these outlying observations which have the potential to harm patients.
An EP21 revision now includes in additional to analytical errors, pre-analytical and post-analytical errors. Since there is no change to the data collection protocol, it is probably better stated that the analysis no longer excludes pre-analytical and post-analytical errors. Examples of pre-analytical errors are incorrect sample preparation or mixing up two patient samples. Errors such as mixing up patient samples are not specific to an assay being evaluated. Another type of error is when an assayed sample fails to produce a result. This can happen when an instrument detects a fault in the analytical process and suppresses the result. A non result does not generate a difference but delayed results are a potential source of patient harm and their frequency can be measured.
This has raised an objection since traditional evaluations focus on analytical error only. EP21 does allow a subset of errors due to the analytical process only to be reported, but it is time to estimate errors from all sources.
Krouwer JS. Estimating Total Analytical Error and Its Sources: Techniques to Improve Method Evaluation. (1992) Arch Pathol Lab Med, 116;726:731.