By David Plaut and Nathalie Lepage
For the Tech
We are all familiar with repeats. For this discussion "repeat" indicates the replication of one or more samples (i.e., controls or patients) either at the same time or over a period. The period is related to the goal of repeating samples (e.g., method validation is different from delta checks). Below are some of the more common instances when one or both patient samples and controls are repeated.
When validating a new method: Two sources of error need to be quantified when validating a new method - bias and imprecision. The best approach to imprecision in one's own lab is to assay not only controls, both within and across runs (an n of 10 should suffice), but patients as well (an n of 10 should suffice in both cases). The argument for determining the within run imprecision is to determine whether the across runs is really the contributing factor to imprecision. It will make it easier to discuss variation with a clinician. The statistic for measuring imprecision is the standard deviation (SD) from which it is easy to determine the % CV.
When recalibrating: In addition to running the controls 3-5 times (should you have found the calibrations do yield some variation in the controls), it is wise to re-run patient samples (4-6) that you have saved for the occasion (supning widely across the analytical range). The statistic that is best in this case is the unpaired student-test.
When changing reagents (this may not always be necessary): If you have found that changing reagents has a measureable effect on the values for controls and patients, follow the same procedure above when changing reagents.
After major changes to the hardware and sometimes when software has been modified: Follow the same procedure above when changing reagents.
After trouble shooting an 'out of control' situation: Follow the same procedure above when changing reagents.
As a part of the quality control program: At the very least, follow the guidelines from CLIA/CAP. It is normal practice to run all the controls before the start of the patient samples. This is a good idea, especially if the instrument has been resting for a period of several hours (e.g., at night). There are good arguments for repeating one or more controls and/or patient samples within a run, especially when a run may be comprised of 30 or more patient samples. The argument of this is that it is better to detect an error that may have affected 30 samples than detecting the error after 50 or more samples. Parvin has discussed this elegantly in two articles (both of which are available to download at no charge).1 If your laboratory decided to include a patient sample periodically in a run, we suggest that a pool be made and aliquots frozen so that more data can be collected. It is wise to analyze the pool 4-8 times when the pool is first used to get a useful measure of the CV of the pool. If the pool is used as a monitor of the stability of the analyzers, it makes good sense to plot it and treat it as another control.
For delta checks: The delta check methods are used to detect random errors in clinical laboratory tests, including specimen abnormalities, specimen mix-up, problems in analysis processes, and clerical errors. With the availability of computers (LIS or HIS) to calculate the difference between a patient's value on the current run and an earlier one (which could have been hours or days or even longer apart), it is important to include delta checks for analytes where delta checks are useful.
For example, delta checks are not of much value for drugs on people who are not taking them or are just starting as they will not be at steady state; thus, increases in the results are expected. On the other hand, Rheim and Lee found that "the multivariate delta check methods are .superior to the univariate delta check methods."2 They suggested that 4 analytes - total cholesterol, albumin and total protein and either AST or ALT - in chemistry would reduce the number of delta checks to validate while yielding a high rate to true errors in identification or handling of patient samples. Straseski and Strathmann also pointed out that delta checks could detect errors at all areas of processing a sample as well as detecting changes that would result in a change in treatment of patients.3
To determine the biological variation and the critical value: Biological variation (BV) is a metric that can aid in the interpretation of laboratory data as well as designing a better quality control system. BV is simply the variation within an individual apart (i.e., intra-individual) from any analytical variation.
Repeating patient samples or quality control should be regarded as an essential component of any quality assurance program for clinical laboratories.
David Plaut is a chemist and statistician in Plano, TX. Nathalie Lepage is a clinical biochemist and a biochemical geneticist at the Children's Hospital of Eastern Ontario and an associate professor in the Department of Pathology and Laboratory Medicine at the University of Ottawa, Ontario, Canada.
©Copyright 2013 Merion Matters. All rights reserved.
1. Should I repeat my 1:2s QC rejection? Parvin CA, Kuchipudi L, Yundt-Pacheco JC.Clin Chem. 2012 May;58(5):925-9 See also Validating the performance of QC procedures.Yundt-Pacheco J, Parvin CA. Clin Lab Med. 2013 Mar;33(1):75-88.
2. Rheem I, Lee KN. The multi-item univariate delta check method: a new approach. Stud Health Technol Inform. 1998;52 Pt 2:859-63
3. Straseski JA, Strathmann FG. Patient data algorithms. Clin Lab Med. 2013 Mar;33(1):147-60.