68 / 71
S
chneider
&
A
ndersen
:
J
ournal of
AOAC I
nternational
V
ol
.
98, N
o
. 3, 2015
667
Repeatability SDs (s
r
), reproducibility SDs (s
R
), repeatability
RSDs (RSD
r
), reproducibility RSDs (RSD
R
), and number
of statistical outliers are presented in Table 3. HorRat values
are also presented in this table and are calculated as RSD
R
(observed)/RSD
R
(predicted), where the RSD
R
(predicted) is
calculated using the equation RSD
R
= 2C
–0.1505
, where C is the
measured analyte concentration in decimal mass units. Cochran,
Grubbs, and double Grubbs tests were used to remove statistical
outliers where appropriate. When one data point was deemed to
be an outlier, both replicates for that concentration level for that
laboratory were excluded from the data set. The
n
= 14 number
of participating laboratories permitted data from up to three
laboratories to be excluded at each concentration level.
Analyte Quantification
Overall, the analytical results in all matrixes were excellent
for the test samples fortified at 0.42, 0.90 and 1.75 µg/kg, as can
be seen in Table 3. Trueness ranged from 88 to 108% recovery
for analytes in all matrixes except for MG in catfish, which
yielded lower recoveries (78–79%). RSD
r
values were generally
≤10%, except in the case of low level incurred samples (CV
in salmon and catfish). HorRat values were uniformly very
low (<1). The sole exception was CV incurred salmon, with a
HorRat of 2.1. The exposure of salmon to a low concentration
mixture of the analytes for 1 h was not sufficient to provide a
significant concentration of CV residue in the salmon muscle.
The higher HorRat for CV incurred salmon is a clear indication
that the mean measured concentration of 0.03 µg/kg of CV is
below the LOQ for this method.
In the single-laboratory validation of the First Action method
in trout matrix, the authors determined the decision limit
(CC
α
) and the detection capability (CC
β
) for each analyte (12).
CC
α
ranged from 0.13 to 0.42 µg/kg for the five analytes in
trout matrix, and CC
β
ranged from 0.17 to 0.54 µg/kg. In
trout matrix, the method was determined to have the greatest
sensitivity for CV and the least for LCV. From collaborative
study data, CC
α
and CC
β
were determined from the quantitative
product ion transition
1
for each analyte in each matrix from
the extracted calibration curves according to ISO 11843-2 (16)
and Commission Decision 2002/657/EC (10). The medians
of individual CC
α
and CC
β
values determined for each of the
14 laboratories are reported in Table 4a, as recommended in
ISO 11843-2 for a multilaboratory validation (16). Individual
values of CC
α
were >1 µg/kg only for one salmon analysis by
one laboratory, where calibration data for MG and BG yielded
CC
α
values of 1.23 and 1.16 µg/kg, respectively. For the overall
median data, CC
α
ranged from 0.14 to 0.42 µg/kg for the five
analytes in salmon, catfish, and shrimp, and CC
β
ranged from
0.16 to 0.47 µg/kg. Collaborative study results for salmon,
catfish, and shrimp were consistent with those reported in the
First Action method validation for trout (12).
In addition to CC
α
and CC
β
, the method detection level
(MDL) and LOQ were calculated from the 0.42 μg/kg fortified
sample data at the 99% confidence level (17). The MDL was
calculated as the SD of the 0.42 µg/kg sample results (
n
= 28,
14 laboratories with duplicate samples) multiplied by the
Students
t
-value at the 99 % confidence interval (one tailed)
for that number of samples. The LOQ was determined as
10 times the SD of the 0.42 μg/kg sample results. Results for
the MDL and LOQ for the different analytes and matrixes are
summarized in Table 4b. Samples were excluded from the MDL
and LOQ determinations if they had been identified as statistical
outliers or excluded for cause, and the total number of samples
(degrees of freedom) was adjusted accordingly. All MDLs
were less than the lowest concentration level for the fortified
samples, and the majority of the LOQs were determined to be
below the 1.0 µg/kg level of concern. Considering these data
are compared across 14 different laboratories using different
analytical instrumentation, the low MDLs and LOQs highlight
the sensitivity and robustness of this method. MDLs and LOQs
calculated from validation data produced by a single laboratory
would be expected to be significantly lower; however, for the
collaborative study, each laboratory only generated results for
two samples/matrix at the 0.42 μg/kg concentration, which
1
For CC
α
calculations, quantitative product ion transition data was
calculated as the peak area ratio relative to the internal standard. At
the zero calibration level, most participating laboratories reported
numerical peak area data for small peaks or noise detected at the
retention time of the analyte, while other labs reported the value
“0”. It was beyond the scope of the study to obtain non-zero noise
measurements from each laboratory. Of the 210 individual CC
α
calculations, 30% were based on calibration data sets that included
the value of “0” for the peak area ratio at the zero calibration level.
Table 7. Analytical screening results for samples
compared only to a single 0.5 μg/kg extracted matrix
calibrant (
n
= 28
a
; 14 laboratories with duplicate samples at
each concentration level)
Samples with peak area response >0.5 µg/kg
calibrant, %
MG LMG CV LCV BG
Salmon
Negative control
0
0
0
0
0
Spike level 0.42 µg/kg
7
0
7
4
18
Spike level 0.90 µg/kg
100 100 100 100 100
Spike level 1.75 µg/kg
100 100 100 100 100
Incurred
100 100
0
b
0
b
100
Catfish
Negative control
0
0
0
0
0
Spike level 0.42 µg/kg
14
11
5
32
25
Spike level 0.90 µg/kg
79
100 100 100
96
Spike level 1.75 µg/kg
100 100 100 100 100
Incurred
100 100
4
b
100 100
Shrimp
Negative control
0
0
0
0
0
Spike level 0.42 µg/kg
31
0
21
4
23
Spike level 0.90 µg/kg
96
100 100 100 100
Spike level 1.75 µg/kg
100 100 100 100 100
Incurred
96
c
100 100 100 100
a
Data excluded in the case of reported cause and for calibration curve
nonlinearity; statistical outliers were not excluded.
b
Mean concentrations of incurred samples were <0.4 µg/kg (0.03 for
CV in salmon, 0.15 for CV in catfish, and 0.39 for LCV in salmon).
c
Mean concentration found for MG incurred shrimp was 0.71 µg/kg.