| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
|
|
|
|||||||
|
|||||||||
Electronic Letters to:
|
Electronic letters published:
![[Read eLetter]](/icons/misc/sectionDOWN.gif){kind=icon}
Response to Kristiansen's Counterpoint
|
|
|||
|
Jan S. Krouwer, Consultant Krouwer Consulting
Send letter to journal:
jan.krouwer{at}comcast.net Jan S. Krouwer
|
To the Editor In his “counterpoint”, Kristiansen (1) responded to my “point” (2). Here is my reaction to his counterpoint. There are two quite different ways to arrive at uncertainty intervals. For the purpose of this discussion, they will be called NPM (nonparametric method) (3) and GUM (4). NPM – uses observed data and a nonparametric method to estimate uncertainty intervals. There are no models or assumptions in this method other than the assumption that one has representative data. GUM – which may be in part based on observed data, is mainly a modeling method, and has many assumptions. There is much to agree with in the paper by Kristiansen (1). The NPM method – while more likely to produce a correct uncertainty interval - provides little information as to error sources. Only modeling methods such as GUM can help to pinpoint error sources and quality improvement is likely when error sources have been identified and corrected. The point of my paper is not to question the use of GUM modeling, but to question using GUM to report uncertainty intervals for commercial diagnostic assays. A point that I wish to amplify deals with outliers. Outliers are still errors, albeit large errors. Kristiansen explains that one cannot include outliers in GUM uncertainty intervals because: “an expanded uncertainty interval covering, e.g., 95% or even 99% of the values would obviously not include extreme values, which almost by definition are farther away from the “true value” than 3 SD.” In the nonparametric world, there is no assumption that distributions are normal. A 95% nonparametric, observed uncertainty interval has nothing to do with standard deviations – it simply represents the values corresponding to the 2.5th and 97.5th percentiles of values from an ordered distribution. Whereas it would be nice to identify and eliminate these large errors ahead of time, this is often not the case. Moreover, HAMA interferences are not the sole cause of outliers – there can be other types of interferences or error sources unrelated to interferences, such as instrument system events. References 1. Krouwer JS. A critique of the GUM method of estimating and reporting uncertainty in diagnostic assays. Clin Chem 2003;49:1818-1821. 2. Kristiansen, J. The Guide to Expression of Uncertainty in Measurement Approach for Estimating Uncertainty An Appraisal. Clin Chem 2003;49:1822- 1829. 3. National Committee for Clinical Laboratory Standards. Estimation of total analytical error for clinical laboratory methods; approved guideline. NCCLS document E21-A. 2003 NCCLS Villanova, PA. 4. International Organization for Standardization. Guide to the expression of uncertainty in measurement 1995:101 ISO Geneva. |
|||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |