| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
1 Washington University, School of Medicine, Department of Pathology, and Immunology, St. Louis, MO 63110, E-mail parvin{at}wustl.edu
2 Hennepin County Medical Center, Department of Laboratory, Medicine and Pathology, Minneapolis, MN 55415, E-mail apple004{at}umn.edu
To the Editor:
We thank Holmes and Buhr for their thoughtful critique of our letter (1). We offer the following comments to help clarify the assumptions used in our model and to assist the reader in critically reviewing both the original letter and Holmes and Buhr’s response. First, with respect to the mathematical model used, we based our piecewise linear precision profile on a precision profile that appeared in an article previously published in Clinical Chemistry (2). The within-run imprecision (CV) and cardiac troponin I (cTnI) concentration pairs we derived from that study produce an SD profile that decreases slightly for cTnI concentrations of 0.05 to 0.14 µg/L. However, over this relatively narrow range of cTnI concentrations that are important to determining the results we obtained, the differences in effect of the approach we used to model the precision profile vs that of Holmes and Buhr is minor. For example, if we model a constant SD over the cTnI range, the numbers for the probabilities of 1 of 3 exceeding the 99th percentile limit are 0.015 for CV = 10% and 0.022 for CV = 25% (we originally reported 0.015 and 0.020 in the text); an increase in 7 per thousand rather than 5 per thousand. On the other hand, if we model a constant CV over the cTnI range, the probabilities of 1 of 3 exceeding the 99th percentile are 0.014 for a CV of 10% and 0.018 for a CV of 25%; an increase in 4 per thousand.
Examination of Holmes and Buhr’s Fig. 1, in the letter above, shows that the differences between all the profiles are fairly small for concentrations of of 0.05 to 0.14 µg/L. A constant SD profile and a constant CV profile (linear increase in SD) appear to cover all the profiles over this range in their Fig. Regarding the question of results less than zero, when gaussian random error was added to the true concentration in our reported data, if the result was <0, the result was set to zero. All these results are included in the distribution, but are piled up at zero. Because these are undetectably low results, they will have no effect on the proportion of results that are greater than the 99th percentile.
We agree that patients who present with symptoms suggestive of acute coronary syndrome (ACS), but ruled out for ACS, do have a cTnI distribution different from that of a typical reference population. However, both cardiology and laboratory medicine groups support the use of a typical reference population for determining the 99th percentile cutoff (3)(4). Regarding assessing the rate of false negative findings, studies are currently under way to address this. However, a recent publication by Kupchak et al. (5) has shown that cardiac troponin assay imprecision did not statistically change the risk stratification of ACS patients at reference limit cutoffs used, based on ROC curve analysis. Regarding the role of the 10% CV cutoff, we did not attempt to address this cutoff, as guidelines have moved toward the exclusive use of the 99th percentile cutoff, irrespective of precision. Finally, we agree that any false positive test result may cause indecision by clinicians in emergency medicine and cardiology. However, the current evidence-based findings in the literature support the use of the 99th percentile in diagnostics and risk stratification to rule in and out myocardial infarction.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
