Scope and requisitesBesides the ana

Scope and requisites
Besides the analytes (Pb and Cd) and sample matrix
(marine water), the concentration range must be selected
in order to fully define the method’s scope (box 1b in
Fig.1). The target concentration was fixed at 10μgL
–1
for
both metals—close to the US EPA guideline values (8.1
and 8.8 μgL−1
for Pb and Cd, respectively); this
corresponds to the criterion continuous concentration
(CCC), which is an estimate of the highest concentration
in saltwater to which an aquatic community can be exposed
indefinitely without resulting in an unacceptable effect [17].
The lowest concentration level was initially established at
0.05μgL–1
, corresponding to the limit of quantification
(LQ) indicated in the application bulletin (no criterion was
provided). Therefore, the provisional concentration range
was set at 0.05–10μgL
−1
.
The accepted approach to method validation involves the
adjustment of this task to the“fit-for-purpose”concept [11,
12], i.e., the requisites of the method need to be fixed (box
1a in Fig.1). No external requisites (i.e., legislation) for
accuracy were found for marine water samples. Alterna-tively, the Horwitz equation is widely used for analytical
methods in order to estimate the expected reproducibility
relative standard deviation (RSDR)[18]. For the 10 μgL
−1
level, an RSDRof 32% is expected, and based on this, an
RSDiLIMof up to this value can be used as a requirement
[19]. Similarly, the AOAC Peer-Verified Methods program
[20] (involving 2–3 laboratories, and so close to the single-laboratory case), provides expected RSDR as well as
recovery values (e.g., for 10 µg L
−1
, recovery: 60–115%
andRSD: 21%). Assuming that RSDrLIM=RSDrunLIM=
21%,RSDiLIM=(RSDrLIM
2
+RSDrunLIM
2
)
0.5
∼29.7% (see
Appendix1), which is close to the value provided by the
Horwitz equation.RSDrLIM= 29.7 andELIM= ± 15% could
be selected as external requisites (limits) for accuracy.
Feature selection
Accuracy (trueness and precision), is fixed as a mandatory
“fit-for-purpose”feature (box 2 in Fig. 1), with respect to
the AOAC requirements in this case. In a testing laboratory,
other features could be taken into account, if additional
requisites, whether external (i.e., from clients) and/or
internal (e.g., if they appear in a technical validation
SOP), are stipulated. In the present case, since no additional
requisites are available, only the trueness and precision
(which could be validated altogether) were selected in this
step.
Realistic internal method validation implies performing
accuracy validation experiments (e.g., using a CRM) under
intermediate precision conditions; for example, designing
anXNr×Ns(Nrreplicates × Nsruns; see Appendix 1) data
matrix, the analysis (ANOVA) of which permits the
“intermediate” method accuracy to be estimated [21]. At
this point, it becomes necessary to define the run
conditionsy (box 3 in Fig.1). For the present case study,
a“run”was considered a validation session that occurred
on one day.
Intermediate accuracy study
Since no previous information on the method is available, a
“prevalidation study”[22] was performed to get a provisional estimate of the method’s accuracy (box 3 in Fig.1).
AnX3×7 experimental design was used. Accuracy preva-lidation was performed with the mixture CRM solution of
10μgL
−1
(Pb and Cd). The analysis of these data revealed
the occasional presence of atypical concentration estimates.
Signal inspection by eye showed some shifts in the peak
shape in the voltage domain between samples and succes-sive standard additions as well as between replicates. In
both cases, the infrequent situation ofRSDr>RSDrunwas
observed, indicating that repeatability is the predominant
source of variability in the data.
The intermediate accuracy validation (box 3 in Fig.1)
experimental design involved Nr=6 replicates per run,
which ensures more reliable means and facilitates outlier
detection/elimination tasks, and Ns=12 runs (which is
acceptable for the expected RSDr>RSDrunsituation).
Tables1and2show the experimental results corresponding
to lead and cadmium, respectively. An automatic outlier
detection criterion was included (thez-score approach; see
Appendix1). This is a generalized acceptability limit for
routine laboratories [13]. Moreover, it is simple and easy to
automate (e.g., in MATLAB
®
) so that it can be incorporated
into an SOP. This allows eventual estimated concentrations
(in theX6×12validation matrix) that are inconsistent due to
the intrinsic signal variability encountered under repeat-ability conditions to be eliminated. Table3shows the main
accuracy validation statistics (see Appendix1) for lead and
cadmium (after outlier elimination). The results confirm the
infrequent RSDr>RSDrunsituation for this method.
Accuracy assessment
Monte Carlo simulation has been proposed as a consistent
way of assessing accuracy [15](box4ainFig.1).
Simulatedxij
values were obtained according to the mod