Analytica Chimica Acta 483 (2003) 325–334

Validation of analytical methods based on mass spectrometric

detection according to the “2002/657/EC” European
decision: guideline and application
Jean-Philippe Antignac∗ , Bruno Le Bizec, Fabrice Monteau, François Andre
Laboratoire d’étude des résidus et contaminants dans les aliments (LABERCA), Ecole Nationale Vétérinaire de Nantes,
BP 50707, F-44307 Nantes Cedex 3, France
Received 27 June 2002; received in revised form 3 October 2002; accepted 22 October 2002

Abstract
The purpose of the present paper is to present an interpretation of the concepts introduced in the new 2002/657/EC European
decision and to propose a practical guideline dedicated to the validation of analytical methods based on mass spectrometry.
Considering both the statistical significance of the results and practical aspects, the minimal number of assays permitting a
satisfying validation to be achieved appeared to be 45 for qualitative methods and 55 for quantitative methods. The parameters
validated with this protocol are specificity, sensitivity, linearity, decision limit (CCα), repeatability, detection capability (CCβ)
and recovery. It is proposed to estimate these parameters on the basis of the most intense (or unique) ion for screening methods
and on the basis of the “critical ion” (less intense ion permitting the unambiguous identification of the analyte according to
the required number of identification points) for confirmatory methods. An application of this guideline is presented and
discussed, through the validation of a liquid chromatography-tandem mass spectrometric (LC–MS/MS) method dedicated to
the determination of the corticosteroid triamcinolone acetonide (Tri Acn) in meat samples.
© 2002 Elsevier Science B.V. All rights reserved.
Keywords: Validation; 2002/657/EC; Mass spectrometry; Decision limit (CCα); Detection capability (CCβ)

1. Introduction ment in identical and/or different experimental con-

ditions. Nevertheless, the diversity of the application
Validation of analytical methods dedicated to the fields, validation procedures, estimated parameters, as
measurement of a physico-chemical parameter or well as terms and concepts definitions, led to a lack of
property relative to a substance, material or system objective quality criteria and harmonization, causing
is not a recent problem. When developing a theory great difficulty in comparing methods or results.
or a technical process, scientists have always tried to Moreover, the exponential development of applied
demonstrate the validity of their work using original mathematics, statistics and informatics as well as
experimental methodology or a more conventional theoretical considerations from several expert groups
data set, for example some repetitions of the measure- has introduced much in the way of normalization, re-
searching and universal concepts. The two main work
∗ Corresponding author. Tel.: +33-2-40-68-77-66; axes have been on one side the definition of pertinent
fax: +33-2-40-68-78-78. “quality indicators” of an analytical method, and on
E-mail address: [email protected] (J.-P. Antignac). the other side their associated calculation modalities.

0003-2670/03/$ – see front matter © 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 0 0 3 - 2 6 7 0 ( 0 2 ) 0 1 3 7 9 - X
326 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334

This approach permitted real progress but some diffi- this very complete and complex document is not al-
culties remained, linked both to the existence of sev- ways very easy to read for the analyst, and the need for
eral official institutions (ISO, CODEX, IUPAC, etc.) a practical “fit-for-purpose” guideline clearly appeared
each with their own specifications, and to the diffi- both for the reference and application laboratories.
culty of changing some practical habit in laboratories. In this context, the purpose of the present paper
The best illustrative example are the concepts limit was to extract from the original reference document
of detection (LOD) and limit of identification (LOI). only the necessary theoretical and practical recom-
These two parameters are probably the most widely mendations for the validation of analytical methods
applied to estimate method performance in terms of dedicated to the analysis (qualitative/quantitative,
sensitivity. However, the nature of the measured sig- screening/confirmation) of residues and contaminants
nal, the calculation mode (practical approach based (forbidden/regulated) in biological matrices. Only
on the signal-to-noise ratio or theoretical approach mass spectrometry was considered as the main de-
based on the noise amplitude of blank samples), the tection technique for use in this domain. First, the
unambiguous identification criteria, as well as the practical meaning of the decision limit and detec-
statistical significance (number of repetitions, toler- tion capability concepts will be presented. Secondly,
ance and confidence level) are not systematically well a practical validation guideline will be proposed,
defined, even though important literature has been which was built considering a discussed compro-
dedicated to these concepts [1–6]. mise between statistical significance and practical
At the European level, an important working group aspects. Finally, an application of this guideline will
was mandated to propose a decision, now referenced be presented and discussed, corresponding to the
2002/657/EC,1 to harmonize the characterization and liquid chromatography-tandem mass spectrometry
the validation procedure of analytical methods perfor- (LC–MS/MS) analysis of the corticosteroid fluoci-
mance. This decision provides rules on how methods nolone acetonide residues in meat samples.
are to be used in the testing of official samples accord-
ing to Article 15, paragraph 1, second sentence of the
Council Directive 96/23/EC, and common criteria for 2. Decision limit and detection capability
the interpretation of analytical results of official con-
trol laboratories for samples taken according to the In the 2002/657/EC European decision, the deci-
same Directive. This decision shall not be applicable sion limit (CCα) was defined as “the limit at and
where for certain substances more particular rules have above which it can be concluded with an error prob-
been laid down in Community Legislation. The appli- ability of α that a sample is non-compliant”, and the
cation field of this reference document is analysis of detection capability (CCβ) as “the smallest content of
biological matrices for residue and contaminants, in- the substance that may be detected, identified and/or
cluding organic and mineral substances, forbidden and quantified in a sample with an error probability of β”.
regulated substances, based on qualitative and quan- In fact, these concepts had already been introduced
titative methods, screening and confirmation analysis. in the ISO/11843-1 normative document [8], to pro-
One merit of such a document is to extend some gen- pose a method fixing a limit from which a system can
eral concepts to a very large panel of detection tech- be declared different from its basic state (Fig. 1). In
niques, and to propose a common backbone for the the present case, the system is a diagnostic ion chro-
validation of the corresponding analytical methods. matogram of the target analyte, and the basic state
The best example could be the notion of identification corresponds to this ion chromatogram for a blank sam-
points for confirmatory analysis [7] and the introduc- ple (forbidden substances) or for a sample containing
tion of the decision limit (CCα) and detection capabil- the analyte at the MRL concentration (regulated com-
ity (CCβ) concepts to replace LOD and LOI. However, pounds). As shown in Fig. 2, the measured value for
characterizing the system is the analyte signal ampli-
1 2002/657/EC Commission Decision of 12 August 2002. “Im- tude (noise or analyte peak height), expressed relative
plementing Council Directive 96/23/EC concerning the perfor- to the internal standard signal amplitude. In practice,
mance of analytical methods and the interpretation of results”. two main sources have to be considered to explain
 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334 327

Fig. 1. Definition of CCα and CCβ according to the ISO/11943 normative document.

the variability observed for the recorded signal; one substances). In other words, when the recorded sig-
from the noise (mainly due to the matrix effect, de- nal is lower than CCα, the sample can be declared
pending of the method specificity) and one from the compliant (analyte absent or present at a concentra-
measurement (depending of the method sensitivity tion lower than the MRL) with a confidence level of
and repeatability). CCα and CCβ permits these two (1 − α).
different factors to be characterized. CCβ is the critical real measured concentration
The signal associated with CCα corresponds to a above which it can be concluded that the analyte
maximal noise amplitude (forbidden substances) or to is unambiguously present (forbidden substances) or
a maximal signal amplitude for a sample containing present at a concentration unambiguously higher than
the analyte at the MRL concentration (regulated sub- the MRL (regulated substances). In other word, sup-
stances), i.e. the amplitude that can only be exceeded posing the analysis of a sample (1) giving a signal
in α% of the cases. This signal can be considered as higher than the decision limit and (2) leading to an
an horizontal line drawn on the ion chromatogram estimated concentration C, this sample can be de-
to materialize a limit below which the observed sig- clared non-compliant (analyte present or present at
nal can be declared as not statistically different from a concentration higher than the MRL) if C ≥ CCβ,
the noise (forbidden substances) or not statistically with a confidence level of (1 − β).
higher than the one induced by a sample contain- For samples giving a signal higher than the deci-
ing the analyte at the MRL concentration (regulated sion limit but leading to an estimated concentration
328 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334

Fig. 2. Signal to consider for blank sample (upper) and for sample containing the analyte (lower).

between CCα and CCβ, high suspicion can be gener- method. For screening methods, e.g. mass spectrom-
ated, but from a statistical point of view, the result re- etry based on the most intense ion, or any other
mains unclassified. Depending on political decisions, screening detection technique, these values can only
protection of producers, industrials or consumers, the be estimated for the unique diagnostic signal, and
sample would be declared compliant or not compliant. consequently are not subject to further identification
A second possibility should be to use an alternative criteria requirements. For confirmatory methods, i.e.
method if available and/or increase the concentration mainly mass spectrometry based on several ions per-
of the extract. mitting a minimal identification score (3 or 4 depend-
In theory, CCα and CCβ can be calculated for each ing on the substance) to be achieved, These values
different signal characterizing the analyte. However, have to be preferably estimated for the more “critical
some additional considerations are needed depend- ion” (in general the less intense one), and only after
ing on the screening or confirmatory purpose of the having verified that all the identification criteria are
 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334 329

in the correct tolerance range (relative retention time, concentration, and the definition of CCα (Eq. (2)),
different ion ratio). the combination leading to the expression given by
Eq. (3).

3. Proposed validation protocol ICCα =µN +aCCα (1)

ICCα =µN +2.33σN (2)
3.1. Qualitative and/or forbidden substances
2.33σN
CCα= (3)
Our objective was to find a compromise between a
the 2002/657/EC decision requirements, an opti- An alternative method consists of using the stan-
mized organization for the assays to use the same dard error of the fitted calibration graph interpolated
experiments to validate several parameters, and some intercept (σ e ) instead of the average of the noise
practical aspects and limitations linked to a labora- amplitude (σ N ). The advantage of this second ap-
tory activity. Finally, it was assumed that 45 assays proach would be to avoid the analysis of the 20 blank
should permit a satisfactory validation if conducted samples and to require only several (at least three)
as follows: five concentration levels for calibration, repetitions at each fortification level. Nevertheless,
2× 10 blank samples, and 2× 10 spiked samples. It the interpolated intercept appears strongly sensitive
is recommended to divide these assays into five series to the linearity, to the considered fortification levels
conducted on separate days. and to the repeatability of each three assays series.
The analysis of at least 20 blank samples, from dif- Consequently, some interpretation difficulties can
ferent origins to check the ruggedness of the method, arise with this method in the case of negative inter-
permits the specificity to be evaluated through the av- polated intercept or standard error artificially under-
erage (µN ) and standard deviation (σ N ) of the noise or over-estimated. For example, a linearity better than
amplitude, expressed relative to the internal standard 0.9999 due to the influence of one particular high
signal amplitude. The calibration should preferably fortification level and an apparent high repeatability
be realized on a pool of the 20 previously analyzed can lead to a very poor intercept standard error and
blank samples and include at least five fortification finally to a largely underestimated CCα value. This
levels, using in addition the previously estimated explains the choice made at present to consider the
noise average (µN ) as a forced intercept. The total real measured noise variability value (σ N ) and to
number of six concentration levels appears to satisfy force the fitted calibration graph by the correspond-
for many authors, and for the 2002/657/EC decision, ing average (µN ). However, it is recognized that this
to well define a calibration graph, but some dis- very critical point will merit further discussion and
cussions still remain concerning the choice of these practical testing.
levels as well as their equidistant or non-equidistant The analysis of 20 spiked samples permits the
separation. Our position was not to use systemati- repeatability to be estimated through the standard
cally equidistant levels, but on one side to verify the deviation of the signal amplitude (σ s ). Ideally, the
linearity of the concentration domain globally used fortification level used should be exactly CCβ. In
in practice (for example 0–100 ng ml−1 ) and on the practice, this concentration is estimated during the
other side to give more importance to the low concen- method development. Typically, a concentration in-
trations (for example 0–5–10–15–50–100 ng ml−1 ), ducing a signal-to-noise ratio of ca. 6 can be used. In
because this part of the graph has a strong influence order to minimize this estimation error, consideration
on the CCα and CCβ values. Finally, this calibra- of the signal relative standard deviation ((R.S.D.)s )
tion graph permits linearity to be evaluated through is preferable to the standard deviation (σ s ). Finally,
the regression coefficient (R2 ) and the sensitivity σ N , a, and (R.S.D.)s permit to the detection capa-
through the slope of the fitted graph (a). Then, a and bility (CCβ) to be calculated, considering the cal-
σ N permit the decision limit (CCα) to be calculated ibration equation (Eq. (4)), where I is the signal
considering the equation of the calibration graph amplitude and C the concentration, and the defi-
(Eq. (1)), where I is the signal amplitude and C the nition of CCβ (Eq. (5)), the combination of these
330 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334

two formula leading to the final expression given by this concentration is estimated during method devel-
Eq. (6). opment (concentration inducing a signal-to-noise ratio
>6) and considering the relative standard deviation of
ICCβ = µN + 2.33σN + 1.64σCCβ the signal ((R.S.D.)s ) rather than its standard deviation
= µN + 2.33σN + 1.64(R.S.D.)s ICCβ (4) (σ s ) to minimize the estimation error on the concentra-
tion used for fortification. Finally, a, µN , µMRL , σ MRL
2.33σN + 1.64µN (R.S.D.)s and (R.S.D.)s permit the detection capability (CCβ)
CCβ = (5)
a[1 − 1.64(R.S.D.)s ] to be calculated, considering the calibration equation
(Eq. (9)), were I is the signal amplitude and C the con-
3.2. Quantitative methods and/or regulated centration, and the definition of CCβ (Eq. (10)), the
substances combination of these two formula leading to the final
expression given by the Eq. (11).
It was assumed that 55 assays should permit a sat-
isfying validation if conducted as follows. Calibration ICCβ = µN + aCCβ (9)
should be realized using real extracted samples when
possible and include blank samples and at least five ICCβ = µMRL + 1.64σMRL + 1.64σCCβ
fortification levels around the critical value (MRL) and = µMRL + 1.64σMRL + 1.64(R.S.D.)s ICCβ
blank samples. The minimal number of assays should
be 20 replicates at the MRL concentration, 10 repli- (10)
cates at 0.5MRL and 1.5MRL, and 5 replicates un-
der 0.5MRL and above 1.5MRL as well as for blank µMRL − µN + 1.64σMRL + 1.64µN (R.S.D.)s
CCβ=
samples. These assays authorize the evaluation of the a[1 − 1.64(R.S.D.)s ]
linearity around the critical value through the regres- (11)
sion coefficient (R2 ), the sensitivity through the slope
of the fitted graph (a), and the intercept corresponding
to the noise amplitude average obtained on the basis 4. Application example
of the five blank sample analyses (µN ).
The 20 spiked samples at the MRL concentration The analytical method presented to illustrate the
give access to the repeatability at this concentration proposed validation procedure concerns the analysis
level through the average (µMRL ) and standard de- for the corticosteroid triamcinolone acetonide (Tri
viation (σ MRL ) of the signal amplitude. Finally, a, Acn) in meat samples. The extraction and purification
µN , µMRL and σ MRL permit the decision limit CCα method was presented elsewhere [9]. The technique
to be calculated considering the calibration equation was liquid chromatography-tandem mass spectrome-
(Eq. (6)), were I is the signal amplitude and C the try using negative electrospray ionization and MRM
concentration, and the definition of CCα (Eq. (7)), the acquisition using two transitions. The analyte response
combination of these two formula leading to the final was always related to the internal standard response
expression given by Eq. (8). (IS: triamcinolone acétonide-d6, added at 5 ng ml−1 in
ICCα = µN + aCCα (6) each sample). The results are summarized in Table 1.
The specificity, in terms of absence of interfer-
ICCα = µMRL + 1.64σMRL (7) ing compounds appearing on the ion chromatograms
(Fig. 3) and noise variability, was greatly improved by
µMRL −µN + 1.64σMRL
CCα = (8) the use of MS/MS but clearly also by the efficient pu-
a rification process. The response linearity was found to
The 10 spiked samples at the 1.5MRL concentra- be satisfactory with a R2 value >0.99 for the two tran-
tion to the estimation of the repeatability at this con- sitions (Fig. 4). The sensitivity ratio between the two
centration level through the standard deviation of the calibration slopes (0.82) perfectly reflected the inten-
signal amplitude (σ s ). Ideally, the fortification level sity ratio between the two product ions (0.85 ± 0.04,
used should be exactly CCβ. As previously discussed, n = 20). For the spiked samples, all the required
 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334
Fig. 3. Triamcinolone acetonide diagnostic ion chromatograms for a blank meat sample (left) and for a 150 ng ml−1 spiked meat sample (right).

331
332 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334

Fig. 4. Calibration graphs determined on the basis of triamcinolone acetonide meat hair samples for the two diagnostic MRM transitions.

Fig. 5. Test of the identification criteria requirements for the spiked samples used for the method validation.
 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334 333

Table 1
Validation results for the method dedicated to the LC–MS/MS determinations of the corticosteroid triamcinolone acetonide (Tri Acn) in
meat samples (15 g)
Process Parametera Signal 1 (more intense) Signal 2 (less intense)

(0) Diagnostic signals Analyte (Tri Can) 493 > 413 493 > 375
Internal standard (Tri Can-d6) 499 > 419
(1) Blank samples Specificity (µN ; σ N ) 0.0122; 0.006 0.0124; 0.007
(2) Calibration curve Concentration levels (ng ml−1 ) 0.15; 0.3; 1.5; 3; 7.5
Linearity (R2 ) 0.9985 0.9984
Sensitivity (a) 0.2141 0.1748
(3) Spiked samples Concentration level (ng ml−1 ) 1.50
Repeatability (R.S.D.)
Signal (%) 5.7 5.8
RRT (%) 0.4
Ratio (%) 4.6
Recovery (%) 96.6 100
(4) Critical limits CCα (ng ml−1 ) 0.07 0.10
CCβ (ng ml−1 ) 0.08 0.11
a R2 , coefficient of determination; a, slope of the fitted curve; (µ ; σ ), average and standard deviation of the noise; R.S.D., relative
N N
standard deviation; RRT, relative retention time.

identification criteria were successfully tested using its application in the field of drugs and contaminant
a self-made automatic diagnostic file (Fig. 5) and residue analysis by mass spectrometry. It appeared that
the repeatability of the two signals appeared very the minimal number of assays to achieve this require-
satisfying (5.7 and 5.8%, respectively). The decision ments is 45 for qualitative methods and 55 for quanti-
limit and the detection capability were calculated for tative methods, if organized and applied as described.
each signal (Table 1). However, it was assumed that Of course, additional assays can be included especially
the values with main practical interest were the deci- when recovery yield performances are needed. More-
sion limit estimated on the basis of the more intense over, a continuous survey using quality test systems as
transition (the objective at this level is to detect the control charts remains as a necessary tool to prevent
analyte) and the detection capability estimated on the any deviation of the method performance. This practi-
basis of the less intense transition (the objective at cal methodology was established taking into account
this level is to identify unambiguously the analyte). a good compromise between statistical significance
The obtained corresponding values (CCα = 70 ppt) of the results and practical aspects, with the objective
and detection capability (CCβ = 110 ppt) appeared to give a more “fit-for-purpose” protocol as possible.
very satisfying and coherent with the observation of However, some important points remains to be de-
the ion chromatograms in term of signal-to-noise ratio fined and harmonized, such as an official position on
as shown in Fig. 3. the decision to give a quantitative result regarding the
method CCα and CCβ. In conclusion, we think that
a large diffusion and utilization of the detection ca-
5. Conclusion pability concept should be encouraged. Indeed, it can
represent an elegant alternative to highly complex un-
The main objective for the presentation of this val- certainty calculations. The counterpart is the necessity
idation guideline was to underline the necessary re- of highly improved analytical methods, particularly
quirements of the 2002/657/EC European decision for in term of specificity, for comfortable application.
334 J.-P. Antignac et al. / Analytica Chimica Acta 483 (2003) 325–334

References [6] H. Van der Voet, J.A.H. Van Rhijn, H.J. Van de Wiel, Anal.
Chim. Acta 391 (1999) 159–171.
[1] L.A. Currie, Anal. Chim. Acta 391 (1999) 127–134. [7] F. André, K. De Wasch, H.F. De Brabander, S. Impens, L.
[2] L.A. Currie, Anal. Chim. Acta 391 (1999) 105–126. Stolker, L. van Ginkel, R. Stephany, R. Schilt, D. Courtheyn,
[3] M. Feinberg, N. Raguènès, Anal. Chim. Acta 391 (1999) 239– Y. Bonnaire, P. Fürst, P. Gowik, G. Kennedy, T. Kuhn, J.P.
252. Moretain, M. Sauer, Trends Anal. Chem. 20 (2001) 435–
[4] P. Hubert, P. Chiap, J. Crommen, B. Boulanger, E. Chapuzet, 445.
N. Mercier, S. Bervoas-Martin, P. Chevalier, D. Grandjean, P. [8] ISO/11843-1, Capability of Detection (Part 1): Terms and
Lagorce, M. Lallier, M.C. Laparra, M. Laurentie, J.C. Nivet, Definitions, International Organization for Standardization,
Anal. Chim. Acta 391 (1999) 135–148. 1997.
[5] A. Maroto, J. Riu, R. Boqué, F.X. Rius, Anal. Chim. Acta 391 [9] J.P. Antignac, B. Le Bizec, F. Monteau, F. Poulain, F. André,
(1999) 173–185. J. Chromatogr. B 757 (2001) 11–19.