Issues Related to Imprecision

There was significant concern that the high imprecision at the low cutoff values suggested would result in frequent analytic false positives. Accordingly, it was recommended that the 99th percentile be measured with a coefficient of variation (CV) <10%. This value, which initially led to complaints from the diagnostic industry, was predicated on several factors and is really a conservative estimate. Data to define the 10% CV were available at that time only through the diagnostic companies and were usually based on within-assay variability. Such an approach underestimates assay variability significantly, because it ignores the additional variability that occurs across different runs, on different machines, on different days, and with different lots of reagents. Thus, the true variability is likely nearly double what is claimed by within-assay variability (101). If one then doubles the results from one publication that examined within-assay results at the 99th percentile, as many as 15% ofsamples at that concentration might yield results for which one ofthe values was considered elevated and the other "normal" (102). Because elevations in troponin in patients with ischemic heart disease trigger aggressive anticoagulant therapy (103-105), low-molecular-weight heparin use (106), and consideration of invasive evaluation (107110), it is desireable to avoid analytic false positives. Thus, it was recommended that the 99th percentile value be measured with a CV < 10%. In July 2004, the first assay to meet

Cutoff Cutoff

Fig. 1. Rate of combined end point of death and MI at 30 d among patients with troponin T levels above (black bars) or below (white bars) evaluated cut points (0.01, 0.03, and 0.1 pg/L) in FRISC-II and GUSTO-IV studies. (Modified from ref. 119 with permission.)

Cutoff Cutoff

Fig. 1. Rate of combined end point of death and MI at 30 d among patients with troponin T levels above (black bars) or below (white bars) evaluated cut points (0.01, 0.03, and 0.1 pg/L) in FRISC-II and GUSTO-IV studies. (Modified from ref. 119 with permission.)

these criteria was made available (111). Until more studies are done, diagnostic companies will continue to provide the critical information that is necessary to define assay characteristics. The ESC/ACC committee strongly urged that assay validation studies be published in the peer-reviewed literature in the hope of improving the level of sophistication and accuracy of the information. Recently, the International Federation for Clinical Chemistry (IFCC) group on markers of necrosis has codified criteria to be used for "quality troponin assays" (112).

In aggregate, this call for rigorously low imprecision has driven the assays to marked improvement. However, controversy remains regarding the value that should be used if an assay fails, as most do, to meet the criteria proposed. We and others have advocated using the value at which there is 10% imprecision to avoid analytic false positives (113). It is true that in some analyses, a small number of patients with acute coronary syndromes (ACS) who have an adverse prognosis are missed by such an approach (114) (Fig. 1).

However, it was thought that the requirement for a CV <10% would reduce the potential for analytic false positives with low-level elevations in patients with a lower clinical probability of ACS. In fact, as long as the normal value study is done similarly to the precision evaluations, the percentage of false positives will not be altered by using the 10% CV value. Accordingly, it is likely that the requirement for a CV <10% will not be included in updates to the ESC/ACC guidelines.

Was this article helpful?

0 0

Post a comment