Interlaboratory coefficient of variation

Hall, P., and Selinger, B., A Statistical Justification Relating Interlaboratory Coefficients of Variation with Concentration Levels, Anal. Chem., 61, 1989, 1465-1466. 

Figure 4.6 Interlaboratory coefficient of variation as a function of concentration (note that the filled circles are values calculated by using equation (4.4), not experimental points) [10]. Reproduced by permission of AOAC International, from Horwitz, W., J. Assoc. Off. Anal. Chem., 66, 1295-1301 (1983).

Two-site immunometric or sandwich assays that made use of two or more antibodies directed at different parts of the PRL molecule were next to be developed. As with other two-site IRMA assays, the capture antibody is attached to a solid phase separation system and the second or signal antibody is labeled with a detection molecule (e.g., radio-isotope, enzyme,fluorophor, or chemiluminescence tag ). In some assays, the capture antibody is attached to the wall of test tubes, plastic beads, microtiter plates, ferromagnetic particles, or glass-fiber paper. Other assays have used the strep-avidin approach that couples biotin to the signal antibody with avidin linked to a solid phase. Most of the current immunometric assays for PRL have been adapted to fully automated immunoassay systems. Compared with the older traditional RIA methods, these automated immunometric assays for PRL generally achieve lower detection limits (0.2 to 1.0 ig/L) and improved precision (interlaboratory coefficients of variation of <8% at all concentrations), and have superior specificity (<0.05% crossreactivity with GH). 

Use of the combination of reference methodology and materials to convert incompatible enzyme activity results for aspartate aminotransferase (146), alkaline phosphatase (147), and lactate dehydrogenase (148) to compatible values by use of a single scale, termed by the authors the International Clinical Enzyme Scale (ICES) has been suggested (149). Application of the ICES concept to the 1970 Scandinavian interlaboratory surv ey decreased the interlaboratory coefficient of variation from 38% to 16% for the enzymes tested. Similarly, in the 1971 New York State aspartate aminotransferase survev, the interlaboratory CV decreased from 41 %-44% to 2%-5%, a major improvement. The Scandinavian Committee on Enzymes has expressed serious concerns about the philosophy of the ICES approach, and they again endorsed the widely accepted approach of ongoing development of reference methodologies and proper use of reference materials (150). Since this flurry of activity in 1984 and 1985, there has not been any further application or acceptance of the ICES concept. 

Participation in intercomparison exercises is also a valid opportunity for a laboratory to assess the quality of its analytical capability. Intercomparison exercises allow the interlaboratory coefficients of variation to be estimated for that specific analysis. 

Interlaboratory coefficients of variation for PAHs in standard solutions and in sediment extracts were 18% and 24%, respectively [117]. 

Interlaboratory coefficients of variation of chlorobiphenyls and organo-chlorine pesticides in sediments are about 40%, those in cod liver oil, about 26% [134]. 

We can summarize the importance of sampling by pointing out that in the case of the analysis of peanuts for the mold metabolite aflatoxin, at the parts per billion level, 90% of the total variability is derived from sampling the commodity and preparing the laboratory sample, and only 10% is derived from the analytical operations. Based on the validating collaborative study, the interlaboratory coefficient of variation (CV) of the method of analysis alone in this case is about 30% at the 10 ppb level. 

TABLE 2. INTERLABORATORY COEFFICIENT OF VARIATION AS A FUNCTION OF CONCENTRATION... 

Figure 1. Variation of the interlaboratory coefficient of variation (relative standard deviation X 100)withconcentration...

Due to the great variability in results between different laboratories, there has been a marked lack of agreement on the absolute concentration in blood at which Se deficiency begins. It was found that the interlaboratory coefficient of variation (CV) was unsatisfactory at blood Se levels <0.5 p,mol/L and, at ca. 0.2 p-mol/L, the CV was as high as 57% [83]. In an attempt to improve this situation, Koh [82] determined the effects of using a common set of blood selenium standards to reduce the interlaboratory CV. In a cooperative survey, 43 laboratories performed the assay of Se in 3 blood samples (0.2-2.9 i,mol/L) using either a set of blood Se standards, as supplied, or their own aqueous standards. Using aqueous standards for calibration, the interlaboratory CV s were 82% and 143% at 0.5 and 0.2 xmol/L, respectively. However, CV s were reduced to ca. 29% for blood standards. Thus, the benefit of a common set of standards, preferably of similar matrix and Se concentration is demonstrated. 

The average interlaboratory coefficient of variation using the standard bottle is approximately 14%. 

Horwitz claims that irrespective of the complexity found within various analytical methods the limits of analytical variability can be expressed or summarized by plotting the calculated mean coefficient of variation (CV), expressed as powers of two [ordinate], against the analyte level measured, expressed as powers of 10 [abscissa]. In an analysis of 150 independent Association of Official Analytical Chemists (AOAC) interlaboratory collaborative studies covering numerous methods, such as chromatography, atomic absorption, molecular absorption spectroscopy, spectrophotometry, and bioassay, it appears that the relationship describing the CV of an analytical method and the absolute analyte concentration is independent of the analyte type or the method used for detection. 

Coefficient of variation of interlaboratory results as a function of sample concentration (expressed as g analyte/g sample). The shaded region has been referred to as the Horwitz trumpet" because Of the way it flares open. [From W. Horwitz, Evaluation of Analytical Methods Used for Regulation of Foods and Drugs." Anal. Chem. 1982, 54.67A]... 

In reviewing more than 150 interlaboratory studies with different analytes measured by different techniques, it was observed that the coefficient of variation of mean values reported by different laboratories increased as analyte concentration decreased. At best, the coefficient of variation never seemed to be better than5... 

An interlaboratory test program was used to evaluate the ASTM standard method for measuring Koc (ASTM, 1987b). Conducted at four laboratories using trifluralin and 13 soils types, it was found that the Koc values varied from 2400 to 14990. The coefficient of variation was about 50%. In general, one can expect a variation of 20 to 50% in Koc, (ASTM, 1987b). 

The results are reported to the nearest mg T and the coefficient of variation of test results is about 6-14%, depending on the level of COD present, based on interlaboratory studies (Standard Methods, 1998). 

Previous exercises have shown the difficulties of determining Hg in seawater. As an example, coefficients of variation of 26.7 and 11.6% between eleven laboratories (CV of the mean of laboratory means) were found for Hg levels of respectively 6.0 and 24.2 ng L [31]. Therefore, it was chosen to consider seawater samples containing much higher Hg concentrations (coastal seawater samples spiked with mercury) in a first method performance study and to use natural coastal seawater, along with a spiked sample, in the second interlaboratory study. 

In the first interlaboratory study, the examination of the raw data (14 sets of results of which 12 involved CVAAS, one RNAA and one MIP-AES) revealed a high spread of results due to two outliers. The mean obtained was 12.6 pg L of Hg with a coefficient of variation (CV) between laboratories of 33%. The two high results were attributed to a laboratory contamination. The accepted values showed a picture which was found more acceptable, i.e. the mean obtained was 10.8 pg L with a CV between laboratories of 6.6"/n [8]. At this stage, the agreement between the laboratories was found to be satisfactory however, the Hg content in this (spiked) sample was considered much too high for being representative of natural samples which justified the organization of a second interlaboratory exercise for which results are described below. 

RESULTS OF THE FIRST AND SECOND INTERLABORATORY EXERCISES (STEPS 1, 2 AND 3). THE TABLE LISTS THE MEAN OF THE LABORATORY MEANS ALONG WITH THE STANDARD DEVIATION (SD) AND THE COEFFICIENT OF VARIATION (CV) OBTAINED. THE TRACE METAL CONTENTS ARE GIVEN IN... 

Some estimates of the precision of various typical modem procedures are given in Table 3. General clinical chemists may well be surprised at some of the values when compared with objective experience in other areas. Thus it is well known that the estimation of serum bilimbin has a coefficient of variation (C.V.) of around 15% in most laboratories, and that with creatinine a value of 8-10% is fairly normal (W3). In a recent blind (but not double-blind) interlaboratory trial of a standardized method for 17-ketogenic steroids and one for 17-ketosteroids, Gray et al. (GIO) obtained C.V. s varying between 4% and 14% for both methods. Six of the ten laboratories cooperating in the trial had special steroid experience and were asked to obtain duplicate estimates of any thirty routine specimens of urine. These results are for two well-established and relatively simple procedures. 

A number of studies have investigated interferences with the LAL assay and have attempted to optimize the assay (Douwes et al., 1995 Hollander et al., 1993 Thorne et al., 1997). These studies have demonstrated that results may vary depending upon the sample matrix, the extraction method and the assay method. Other constituents present in the sample may interfere with the LAL assay and cause inhibition or enhancement of the test or aggregation and adsorption of endotoxins, resulting in under- or over-estimation of the concentration. Techniques such as spiking with known quantities of purified endotoxin and analysis of dilution series of the same sample have been described to deal with these interferences (Hollander et al., 1993 Milton et al., 1990 Milton et al., 1992 Whitakker, 1988). Studies in the laboratories of the authors of this chapter have demonstrated within-Iaboratory coefficients of variation between 15 % and 20 % in routine assay work. When extra care is taken to optimize precision this can be reduced to under 5 % in the endpoint chromogenic assay (Thorne, unpublished data). Several interlaboratory comparison studies have been performed and demonstrate much greater variability. One in-depth comparison of two laboratories experienced in the LAL assay. 

Results of this interlaboratory study appeared satisfactory and it was decided to continue the evaluation with synthetic solutions before starting the exercises on real extracts. For a concentration of DMA of 5 pmol kg , the mean of the mean values was very close to the target value [(5.05 0.39) jjmol kg ] and the coefficient of variation of the mean of means was only 7.7%. 

Table 11.8 Coefficients of variation of the mean of laboratory means of EDTA and acetic acid extractable contents as obtained in the first interlaboratory study...

---