Posted by Sten Westgard, MS
Here's a range of statistics describing the performance of a glucose method. Try to pick: Which one has acceptable performance?
- method with combined uncertainty 3.69%
- method with combined uncertainty 6.79%
- method with expanded uncertainty 7.38%
- method with expanded uncertainty 13.58%
- method with 3.78 Sigma
Which method would you pick?
If you didn't already guess, it's a trick question. All of the choices are describing the same method. The data is from the following study:
Two Evaluation Budgets for the Measurement Uncertainty of Glucose in Clinical Chemistry, Hui Chen PhD, Ling Zhang PhD, Xiaoyun Bi, Xiaoling Deng, Korean J Lab Med 2011;31:167-171.
In this study, they estimated the measurement uncertainty of their glucose method using two different budgets. The major difference is that the second MU budget took into account the within-subject biologic variation.
The problem with the measurement uncertainty estimates is that most laboratory users are unfamiliar with the concepts and have no frame of reference for judging them. Is 4% uncertainty acceptable? Is 13% uncertainty unacceptable? For all the effort put into urging laboratories to report uncertainty, very little effort has been made to establish acceptability guidelines.
With Sigma, we have a simpler scale, one that most laboratory professionals know - indeed, most healthcare professionals in and out of the laboratory know about Six Sigma. The scale is simply 3 to 6. Six Sigma is the goal, while 3 Sigma is considered the minimum acceptable performance.
Sigma calculations are also simple: You need a quality requirement, an estimate of imprecision, and an estimate of bias. In this study, they provided the latter two:
- between-day imprecision was 1.91%
- systematic error, calculated against CAP surveys, was 2.78%
The quality requirement for glucose, according to CLIA, is Target value ± 6 mg/dL or ± 10% (greater). The level of interest in this study was 6.1 mmol/L, or about 110 mg/dL. At that level, the quality requirement for CLIA is 10%.
The Sigma-metric equation is: Sigma-metric = (TEa - bias) / CV
For our method, this means:
(10 - 2.78) / 1.91
= 7.22 / 1.91
= 3.78
If that's not helpful, you can use a graphical assessment tool, the Method Decision Chart:
The "operating point" for this glucose method is on the border of good performance.
The additional advantage of using Sigma-metrics is that you can leverage the same data to determine the appropriate QC procedure. With an OPSpecs chart, you can design the QC for this method, optimizing for appropriate error detection while minimizing false rejection:
Here we see that we'll need to use 4 control measurements per run in order to adequately control this method.
With measurement uncertainty, a finding of "expanded uncertainty of 13.58%" comes without context or implication. It's hard to judge acceptability or describe the necessary operating QC. For many laboratories, adding this MU as a comment to the test report isn't helpful - it just confuses the clinician.
Sigma-metrics provide context - placing performance into perspective based on the quality required by the test. Sigma-metrics tools also offer practical guidelines on how to monitor and control the performance of the method.
Comments