| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
Krouwer Consulting, 26 Parks Drive, Sherborn, MA 01770, Fax 1-508-647-9380, E-mail jan.krouwer{at}comcast.net
To the Editor:
Myers et al. (1) discuss the importance of creatinine analytical performance in the estimation of the glomerular filtration rate. They correctly specify a model of assay performance but subsequently do not seem to use that model. I suggest a variation of their model that is less subject to misinterpretation. The Myers et al. (1) model is:
 | (1) |
Although the Myers et al. (1) model is correct, the simulation carried out by Myers et al. is subject to misinterpretation. It is confusing to present the 2 error sources, total imprecision, and random interferences as 1 combined error source, because these 2 error sources are quite different. This confusion seems to have taken place during the preparation of the report, in which the result of a proficiency survey was compared to the simulation. Because neither controls nor pooled samples are used in a proficiency survey, random patient interferences cannot be estimated. Thus, conclusions drawn from this comparison are suspect. Moreover, their alternative way to describe creatinine assay performance goals clearly leaves out random patient interferences [Table 1 in Myers et al. (1)].
The following is suggested as a more useful way to model assay performance:
 | (2) |
and Eq. 1
is that total imprecision and random patient interferences have been separated. In a previous letter published in Clinical Chemistry (2), we added a 4th term, random biases unrelated to interferences from a specific patient sample. It is simpler, however, to lump this term into total imprecision, which gives the same model described by Lawton et al. (3).
Assay performance goals should account for all 3 terms in Eq. 2
. In setting these goals, average bias can be set at a low amount because, as stated by Myers et al. (1), manufacturers have a way of achieving low average bias through standardization. Thus, the majority of error can be allocated between imprecision and random patient interferences. Random patient interferences are also a known factor—their expected value would be zero for an assay with perfect analytical specificity—because Myers et al. (1) discuss analytical nonspecificity problems for several types of creatinine assays and recommend improvement, yet do not specify the magnitude of improvement needed. The specific magnitude of improvement needed could be calculated with the 3-term model described above.
Estimates for each of the 3 error sources can be generated by comparing field and reference creatinine measurements with a series of patient samples. Analysis is simplified because the concentration range of interest for glomerular filtration rate estimation is narrow. The average difference and the SD of differences (for which the difference is between the field and reference method for each patient sample) gives the 2 quantities in Eq. 1
, with total imprecision limited to the time interval of the method comparison experiment. (To ensure that differences are largely due to the field method, the reference method should be replicated to minimize imprecision). If an independent estimate of total imprecision for the field creatinine assay for this time interval is available, the imprecision term from Eq. 1
can be separated into the 2 components expressed in Eq. 2
.
Acknowledgments
Grant/funding support: None declared.
Financial disclosures: None declared.
References
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  W. G. Miller, G. L. Myers, and J. H. Eckfeldt The authors of the article cited above respond: Clin. Chem., September 1, 2007; 53(9): 1716 - 1717. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
