74 / 258
Thompson et al.:
J
ournal of
AOAC I
nternational
V
ol.
98, N
o.
6, 2015
1715
the standard’s nominal concentration). The inclusion of a PB
(run as a sample; its measured concentration must be <1/2 of
the lowest calibration standard), a duplicate sample (relative
difference within 10% for Cr, 7% for Se, and 5% for all other
elements), and known reference materials serving as control
samples (recovery check within control or certified limits) are
mandatory for good method performance. If any of these QC
checks fails, results should be considered invalid.
(c)
The order of analysis should be calibration standards,
followed by rinse, blank check (PB run as a sample), check
standard, control sample, sample, sample duplicate (up to
10 samples), and finally a repeated check standard.
G. Calculations
Sample concentrations in ng/g are automatically calculated
by the software using a nonweighted least-squares linear
regression calibration analysis to produce a best-fit line:
= +
a blank
Y x
Note that for the Agilent software used in this work, the
sample blank is identical to the Cal Blk and is essentially zero
because high purity reagents are used.
The analyte concentration in the sample is then calculated:
= − ×
x y blank
a
DF
where
x
= analyte concentration (ng/g);
y
= analyte to ISTD
intensity ratio, which is the measured count of each analyte’s
standard solution data point in the calibration curve divided by
the counts of the ISTD at the same level; similarly, the blank =
analyte to ISTD intensity ratio, which is the measured count of
the blank standard solution data point in the calibration curve
divided by the counts of the ISTD at the same level as the blank
standard solution;
a
= slope of the calibration curve (mL/ng);
and DF = volume of the sample solution (mL) divided by
sample weight (g).
H. Method Validation
This method has undergone a thorough single-laboratory
validation (SLV) using AOAC guidelines to probe its linearity,
LOQ, specificity, precision, accuracy, and ruggedness/
robustness. Accuracy has also been affirmed by comparison
to ICP-atomic emission spectrometry (AES) results generated
in the authors’ own laboratory. In addition, reproducibility
was estimated during a limited multilaboratory testing (MLT)
study employing six laboratories and four different ICP/MS
instruments. Both the SLV and MLT results are summarized in
a concurrent publication (6).
Results and Discussion
Specificity
The specificity of the method was determined using a single
element standard at 50 mg/L for each analyte and checking for
apparent signal from the other analytes. None of the standards
produced a response above the PLOQ for any of the other 11
analytes (data not shown), demonstrating that each response is
specific for that analyte. The ISTDs were not tested since they
are used at a low concentration of 50 μg/L.
Linearity
Linearity was demonstrated by analyzing various independent
standards (made from the same stock) as samples against
the normal calibration curve. Linearity standards at nine
concentrations of each analyte spanning the range from 50%
of the lowest calibration standard to 50% above the highest
calibration standard were analyzed twice on each of 3 days using
freshly made standards each day. The means of all six analyses
are reported in Table 2. At the lowest level, 50% of the lowest
calibration standard, all analytes demonstrated acceptable
agreement (95–105%, with rounding) with the nominal value.
Therefore, 50% of the lowest calibration standard concentration
is set as the PLOQ. Overall, the recoveries varied from 91 to
107%, and RSDs varied from 0.3 to 9.3%. The recoveries were
nearly all within a desired 95–105% range, though there are no
specific criteria in the SMPR for linearity. The only elements that
presented any linearity issues were P and Fe, which were routinely
under-recovered (P) or over-recovered (Fe) by about 5–6% across
the calibration curve. Possibly, the linearity could be improved
by adjusting some factors for the analysis of these elements, as
they both have relatively low mass with significant background
interferences that must be handled by the CRC. In practice, no
accuracy issues were observed except for some apparent bias in
P results relative to SRM 1849a (
see
below). Typical correlation
coefficients were 0.9995 or better for all analytes.
LOQ
The PLOQ values from the linearity experiment were
converted from a solution concentration (mg/L) to a weight
basis (mg/100 g for a typical dilution of 1.0 g RTF to 50 mL)
and compared to the SMPR (
see
Table 3). The PLOQs meet
the SMPR for all elements except Fe, Cu, and Mn. In these
cases, the test portion size could be increased to 2–3 g RTF to
improve the PLOQ 2–3-fold lower. The lowest concentrations
of Mn, Cu, and Fe found in the SPIFAN matrixes were
150 ng/g (0.015 mg/100 g), 580 ng/g (0.058 mg/100 g), and
14000 ng/g (1.4 mg/100 g), respectively, all in the SPIFAN
control milk. SMPR for LOQ for Mn, Cu, and Fe are 0.001,
0.001, and 0.01 mg/100 g, respectively, at least 10-fold lower
than observed values.
Precision
SPIFANmatrixes were tested on 8 days (including two analysts
and two instruments) in duplicate, and the results are summarized
in Table 4. The SMPRs require RSD
r
to be ≤5% in all 11 matrixes.
All analytes in all matrixes meet this criterion for the within-day
duplicates (data not shown), typically in the 1–2% range. This
requirement is built into the method due to the criterion that
duplicate results must agree to within 5%. When considering
intermediate reproducibility precision (among days/analysts/
instruments, but in a single laboratory), of the 12 elements and
11 matrixes, there are 11 instances of RSD
iR
>5%. Ten of these
are for the ultratrace elements, Mo and Cr, and there is one
instance for Ca in Adult RTF with high fat. The Adult RTF with
high fat matrix has since been shown to be unstable and perhaps
74