Journal of Chromatography B, 877 (2009) 2224–2234

Contents lists available at ScienceDirect

Journal of Chromatography B
journal homepage: www.elsevier.com/locate/chromb

Key aspects of analytical method validation and linearity evaluation夽

Pedro Araujo ∗
National Institute of Nutrition and Seafood Research (NIFES), PO Box 2029, Nordnes, N-5817 Bergen, Norway

a r t i c l e i n f o a b s t r a c t

Article history: Method validation may be regarded as one of the most well-known areas in analytical chemistry as is
Received 2 July 2008 reflected in the substantial number of articles submitted and published in peer review journals every
Accepted 29 September 2008 year. However, some of the relevant parameters recommended by regulatory bodies are often used inter-
Available online 2 October 2008
changeably and incorrectly or are miscalculated, due to few references to evaluate some of the terms
as well as wrong application of the mathematical and statistical approaches used in their estimation.
Keywords:
These mistakes have led to misinterpretation and ambiguity in the terminology and in some instances
Validation
to wrong scientific conclusions. In this article, the definitions of various relevant performance indicators
Terminology
Selectivity
such as selectivity, specificity, accuracy, precision, linearity, range, limit of detection, limit of quantitation,
Specificity ruggedness, and robustness are critically discussed with a view to prevent their erroneous usage and
Accuracy ensure scientific correctness and consistency among publications.
Precision © 2008 Elsevier B.V. All rights reserved.
Linear function analysis
Range
Limit of detection
Limit of quantitation
Ruggedness
Robustness

1. Introduction comparisons were carried out with other or similar methodolo-

gies in order to check for accuracy in the reviewed articles [4].
The word validation originates from the Latin validus meaning In addition, it was pointed out that statistical data analysis was a
strong, and suggests that something has been proved to be true, subject neglected by chemists developing experimental methods
useful and of an acceptable standard. The International Organiza- [3]. In the early 70’s a series of articles were published stress-
tion for Standardization defines validation as the confirmation by ing the need of implementing a consistent set of definitions for
examination and provision of objective evidence that the particu- determining the performance-characteristics of developed ana-
lar requirements for a specified intended use are fulfilled [1]. This lytical methods and comparing unambiguously the advantages
definition primarily implies that a detailed investigation has been and disadvantages of the increasing volume of reported analyti-
carried out and gives evidence that an analytical method, when cor- cal methods [5–8]. This paved the way for the implementation of
rectly applied, produces results that are fit for purpose as well as method validation in analytical laboratories since the late 70’s and
it confirms the effectiveness of the analytical method with a high the current worldwide recognition that method validation is an
degree of accuracy. important component in any laboratory engaged in the develop-
The importance of method validation has been emphasised ment and establishment of standard methods. Nowadays, there are
since the late 40’s when the American Chemical Society and Merck several international renowned organisations offering guidelines
& Co., raised the issue of how mathematics and statistics are on method validation and related topics. Basic references are the
a necessary prerequisite to successful development and adapta- Association of Official Analytical Chemists (AOAC), the American
tion of new analytical methods [2,3]. By that time a survey of Society for Testing and Material (ASTM), the Codex Committee on
papers on development of analytical methods revealed that no Methods of Analysis and Sampling (CCMAS), the European Com-
mittee for Normalization (CEN), the Cooperation on International
Traceability in Analytical Chemistry (CITAC), the European Cooper-
夽 This paper is part of a special issue entitled “Method Validation, Comparison
ation for Accreditation (EA), the Food and Agricultural Organization
and Transfer”, guest edited by Serge Rudaz and Philippe Hubert.
(FAO), the United States Food and Drug Administration (FDA),
∗ Tel.: +47 95285039; fax: +47 55905299. the International Conference on Harmonization (ICH), the Inter-
E-mail address: [email protected]. national Laboratory Accreditation Cooperation (ILAC), The World

1570-0232/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jchromb.2008.09.030
 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234 2225

Health Organization (WHO), the International Organization for 2.4. Analysis

Standardization (ISO), the International Union of Pure and Applied
Chemistry (IUPAC), the United States Pharmacopeia (USP), The ana- The analysis is related to the instrument used to extract qual-
lytical chemistry group EURACHEM, etc. itative or quantitative information from the samples with an
Articles are submitted for publication every year highlight- acceptable uncertainty level. The analysis could be visualised, in
ing the determination of the parameters for method validation a broad sense, as a system possessing three interconnected basic
enforced by any of the above-mentioned regulatory bodies. How- elements, namely input → converter → output. In general, the input
ever, in spite of the volume of articles submitted and published and output are designated by the letters x and y and they represent
misinterpretation and miscalculation still persists due to some the concentration and the experimental response respectively. The
prevailing ambiguity in the definitions of some of the validation choice of a particular analysis is based on many considerations, such
parameters, few existing guidelines to estimate some of them, and as the chemical properties of the analytical species, the concentra-
most importantly lack of attention to the mathematical and statis- tion of the analytes in the sample, the matrix of the sample, the
tical tools involved in their calculation. speed and cost, etc.
This article discusses some key aspects that should be con-
sidered when validating analytical methods, especially those 2.5. Data evaluation
concerning chromatography methods, in order to derive useful
information from experimental data and to draw robust conclu- The main purpose of data evaluation is to summarise and
sions about the validity of the method. gain insight into a particular data set by using mathematical and
Although the aspects described in this article apply to all types of statistical approaches. Data evaluation allows extracting useful
analytical methods, in some instances they may not be applicable information and drawing conclusions about the inputs and outputs,
to certain analytical procedures. For instance, some animal mod- and most importantly about the validation process in general.
els bioassays or some immunoassays have unique features which
should be considered before submitting a validation report. 3. Validation method parameters

2. The general method validation steps In the early 80’s, it was pointed out that the definition of the
characteristic parameters for method validation and related top-
In a general context, method validation is a process that con- ics were different between the existing organizations [11]. In 1990,
sists of at least five distinct steps, namely: system qualifications, the International Conference on Harmonization (ICH) was created
sampling, sample preparation, analysis and data evaluation. as a unique project to bring together the regulatory authorities of
Europe, Japan and the United States with the objective of achiev-
ing greater harmonization of parameters, requirements and, to
2.1. System qualifications
some extent, also methodology for analytical method validation.
The key criteria defined by the ICH and by other industrial com-
A general evaluation of system qualifications allows to verify
mittess and regulatory agencies around the world for evaluating
that the instrument is suitable for the intended analysis, the mate-
analytical methods are: selectivity/specificity, accuracy, precision,
rials (reagents, certified references, external and internal standards,
linearity, range, limit of detection, limit of quantitation, ruggedness,
etc) are suitable for use in analytical determinations, the analysts
and robustness.
have the proper training and qualifications and previous documen-
tation such as analytical procedures, proper approved protocol with
3.1. Selectivity and specificity
pre-established acceptance criteria has been reviewed. If the gen-
eral qualifications of a system are ignored and a problem arises, the
The terms selectivity and specificity have been the subject of
source of the problem will be difficult to identify [9].
intensive critical comments essentially focusing on the ways in
which they are often defined by analysts working in method val-
2.2. Sampling idation [12–15]. By definition the selectivity refers to the extent
to which a method can determine a particular analyte in a com-
The sampling step assists in the selection of a representative plex mixture without interference from other components in the
fraction of the material which is subsequently subjected to inves- mixture [16]. This definition is often wrongly used as equivalent
tigation. The choice of an appropriate sampling method is of great to specificity, which is considered to be the ultimate in selectiv-
importance because it provides assurances that the sample selected ity; it means that no interferences are supposed to occur [12].
is truly representative of the material as a whole for the purpose of Unfortunately, an inspection of the literature on method valida-
meaningful statistical inferences. Within the statistical literature, tion revealed that both terms are still used without distinction by
there is a substantial body of work on sampling strategies, how- some authors, even though by consulting the dictionary it is clear
ever the relative costs and time involved in each strategy should be that these terms should not be used interchangeably. Selectivity
evaluated in advance. should be connected with the word ‘choose’ while specificity with
the word ‘exact’. In this context, it is incorrectly to grade the term
2.3. Sample preparation specificity (either you have it or you do not). An analyst involved
in method validation should always remember that selectivity can
Sample preparation is a key element to successful method vali- be graded as low, high, bad, partial, good, etc., in order to choose
dation. It has been pointed out that sample preparation represents the appropriate category for a particular purpose. The term speci-
60–80% of the work activity and operating costs in an analytical ficity refers always to 100% selectivity [13,17,18] or, conversely, 0%
laboratory [10]. The literature on sample preparation is ample and interferences.
well documented. However, the analyst should remember that the
selection of a specific preparation procedure depends upon the ana- 3.1.1. Experimental approaches to assess selectivity
lytes, the analytical concentrations, the sample matrix, the sample During the last decade, some researchers have expressed con-
size and the instrumental technique. cern about the lack of comprehensive recommendations from
2226 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234

Fig. 1. Liquid chromatography mass spectrometry post-column infusion system.

accredited bodies, books and published articles on how to assess 3.2. Accuracy
the selectivity of a method [14,19]. A non-exhaustive survey of
papers on chromatographic method validation published in analyt- Accuracy is the degree of agreement between the experimental
ical journals during the last 10 years revealed that the three most value, obtained by replicate measurements, and the accepted ref-
popular approaches used to measure selectivity are: erence value. It has been pointed out that the accuracy is the most
crucial aspect that any analytical method should address [23]. The
determination of this parameter allows estimating the extent to
1. Comparison of the chromatograms obtained after injection of
which systematic errors affect a particular method. The preced-
blank samples with and without the analytes.
ing definition of accuracy is in accordance with several regulatory
2. Comparison of the chromatographic response obtained after
injection of analytical solutions with and without all the possible
intereferents.
3. Analysis of certified reference materials.

Other approaches used in a lesser extend in the last 10 years are:

1. Calculation of the chromatographic response factor by discrim-

inating the analytical species from closely related structures.
2. Comparison with certified methods.
3. Comparison of the slopes obtained by the standard addition and
the external standard methods.

In addition to these previously well-known mentioned strate-

gies, an interesting liquid chromatography mass spectrometry
post-column infusion technique that enables detecting specific
endogenous sample components that affect the target analytes
has been proposed [20]. This approach uses a syringe pump and
a HPLC system simultaneously coupled to a mass spectrometer
through a tee connector (Fig. 1). The flow from the syringe pump
delivers a constant amount of the analytes while the flow from
the HPLC delivers a processed blank sample, in that way it is
possible to study dynamically the effect of the matrix on the
analytical responses over the entire chromatographic run, when
different sample treatments, columns and mobile phases are used.
An example from the literature is the evaluation of the influence
of different extraction techniques on matrix effects and conse-
quently the magnitude of these effects on the signal of sirolimus,
an immunosuppressant that under specific chromatographic and
mass spectrometry conditions elutes at approximately 6 min and
presents a characteristics transition at m/z 931.6 → 864.6 [21]. The
comparison of the various infusion chromatograms is showed in
Fig. 2 which allows concluding that the observed signal suppres-
sion is mainly due to endogenous components in the whole blood
and that the analytical signal is less prone to matrix interferences Fig. 2. Comparison of (A) mobile phase, (B) whole blood sample prepared by protein
around the elution time when the solid phase extraction method is precipitation, and (C) a whole sample prepared by solid phase extraction by the post-
used. column infusion method. The areas influenced by matrix effects are shown in B and
C. The solid lines indicate the regions of altered ionization due to matrix effects.
Readers interested in analytical procedures to achieve selectiv-
Reprinted from P.J. Taylor, Matrix Effects: The Achiles Heel of Quantitative HPLC-ESI-
ity are referred to comprehensive articles on the subject published Tandem-MS. Clinical Biochemistry, 38 (4) (2005): 328–334 with permission from
elsewhere [14,17,22]. Elsevier.
 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234 2227

Table 1
Requested conditions to measure the different precision components according to the ISO/DGuide 99999 [33].

* One or more conditions should be changed.

bodies (ICH, FDA, and USP) and the IUPAC. However, it is important Another reported strategy to assess the accuracy of a method has
to mention that this definition is acknowledged by other accred- been taking part in proficiency test schemes in order to compare the
ited organisations as trueness or bias and the term accuracy as the results of a particular method under validation with the consensus
combination of trueness and precision (ISO, EURACHEM, AMC). The value obtained by the participating laboratories [28,29]. Unfortu-
concept of trueness as is stated in ISO-5725-1 has been invented nately, the seconders of this proposal have not taken into account
in order to address some philosophical objections among medi- that proficiency testing is aimed at monitoring performance and
cal and legal practitioners regarding the concept of statistical bias competence of individual accredited laboratories [30] rather than
[24,25]. In this article, the term accuracy as defined by ICH, FAD, assessing the accuracy of newly developed methods. It is irrational,
USP and IUPAC is used because it seems to be the preferred term and contrary to the general requirements for participating in profi-
in scientific journals. A ScienceDirect search using the keywords ciency testing, to take part in such schemes without a fully validated
“validation trueness” and “validation accuracy” showed that only method.
55 articles used the former keywords while 3876 the latter. The The guidance for validation of analytical procedures issued
difference becomes more dramatic when the search is performed by the ICH recommends checking the accuracy by performing a
without using the word “validation”. minimum of nine determinations over a minimum of three con-
Several approaches have been suggested to evaluate the accu- centration levels (low, medium and high) corresponding to the
racy of a method. The main strategies currently used to assess the whole analytical range investigated (3 levels × 3 replicates per
accuracy are: level = 9 determinations) [31]. Although, this minimum require-
ment is suitable in general for chromatographic or spectroscopy
1. Measuring the analyte in a particular reference material and methods, the analyst should follow the recommendations suit-
comparing the result with the certified value. able for his/her particular method. For instance, the FDA guidance
2. Measuring the analyte in blank matrix samples spiked with for validation of bioanalytical methods suggests evaluating the
known analytical concentrations and determining the percent- accuracy by measuring a minimum of three concentration levels
age of recovery. prepared in pentaplicate in the range of expected concentrations
3. Comparing the results from the method under validation with (3 levels × 5 replicates per level = 15 determinations) [32].
those from a reference method.
4. Determining the analytical concentration in the sample by 3.3. Precision
means of the standard addition technique.
The term precision is defined by the ISO International Vocab-
The first strategy should be used as long as the reference material ulary of Basic and General Terms in Metrology (ISO-VIM) as the
closely resembles the analytical concentration and the matrix of the closeness of agreement between quantity values obtained by repli-
sample under investigation. cate measurements of a quantity under specified conditions [33].
The second strategy, despite its popularity, it has the disadvan- The determination of this parameter is one of the basic steps in the
tage that the accuracy can be misestimated if the spiked analyte process of achieving repeatability and reproducibility in method
and the analyte contained in the sample behave differently due to validation. Assessing the precision implies expressing numerically
dissimilarities in their chemical form and reactivity. For speciation the random error or the degree of dispersion of a set of individual
purposes this approach is not recommended [26]. The IUPAC Har- measurements by means of the standard deviation, the variance or
monised Guidelines for In-House Validation of Methods of Analysis the coefficient of variation.
Technical Report recommends the use of the second approach only The regulatory bodies emphasize that the terms precision and
in cases where the method under validation is intended either for accuracy should not be used as synonyms. A literature review of the
liquid samples or samples subjected to total destruction or disso- last decade demonstrated that analysts engaged in validating meth-
lution [27]. ods are well aware of the difference between these terms. However,
A ‘reference method’ in the context of the third strategy refers the review revealed that in many instances, the terms repeatability,
to a nationally or an internationally fully validated method with reproducibility and intermediate precision are used interchange-
different or similar measurement principles and sources of errors. ably. The reader should keep in mind that the word repeatability in
The fourth strategy is generally used in cases where blank sam- the present context, refers to obtaining the magnitude of a partic-
ples are unavailable. ular property of a sample more than once by keeping constant the
2228 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234

global factors (human, preparation, instrumental and geographi- The importance of estimating a total error in validation studies
cal) over a short period of time; the term reproducibility refers to is that it provides a measure of quality that can be compared to
reproduce the magnitude of an already measured property by the intended analytical quality of a test, which can be described in
changing one or more of the global factors over a short or an terms of an allowable total error [35] which sets a nominal limit
extended period of time; and the term intermediate precision refers for the total error tolerable in a single measurement or single test
to obtaining the magnitude of a particular property of a sample result. It is advisable, when determining the total error of a partic-
more than once by using the same preparation, instrumental and ular assay, to follow the recommended criteria. For instance, recent
geographical condition over an extended period of time. The differ- reports addressing the issue of best practices for ligand binding
ences between the three above mentioned precision components assays have established a nominal total error of 30% for this kind of
showed in Table 1 allow to derive the following conclusions: assays based on their inherent variability [36,37]. In addition, it is
compulsory to include all assay runs in the calculation of the total
• It is wrong to report a so-called “inter-day repeatability” term. error. Rejection of particular assay runs in the calculation of the total
Such a term should never be used in method validation. error is allowed only in cases, where obvious and well-documented
• The term inter-day variation should be connected with interme- errors arise [36].
diate precision or in some circumstances with reproducibility. An important consideration is that total error is a quality param-
• The repeatability always ensures that the variability of the results eter derived from two performance parameters (precision and
remains constant under identical conditions and also that factors accuracy) which contribute mutually to the quality of a test result.
such as analyst, procedures, instrumental conditions, laborato- It is desirable to set goals for the allowable total error, rather than
ries, and time have a negligible contribution to the variability of set individual goals for the allowable standard deviation and the
the results. allowable bias.
• The reproducibility always makes certain that the variability of
the results remains constant under different conditions and that 3.3.2. Uncertainty
the variation of one or more factors does not contribute signifi- The uncertainty of measurement is a parameter, associated with
cantly to the variability of the results. the result of a measurement that characterises the dispersion of
the values that could reasonably be attributed to measurand (the
For validation of chromatographic methods, it is generally rec- quantity being measured) [33]. This parameter is estimated in order
ommended checking the precision component by measuring a to judge the adequacy of a result for its intended purpose and to
minimum of three concentration levels (low, medium and high) verify its consistency with other similar results.
prepared in triplicate and covering the whole analytical range Countless studies have reported the determination of the uncer-
under study (3 levels × 3 replicates per level = 9 determinations) tainty of measurement results by using different approaches.
[30]. However, as was mentioned before, the readers must consider However, the lack of consensus among the various approaches
that minimum criteria vary according to the nature of the analytical along with the absence of a worldwide-accepted criterion for the
procedures. For instance, the minimum number determinations to estimation and notation of measurement uncertainty were the
check the precision of a biological or a biotechnological method may main driving factors behind the development in 1980 of the general
be different from the minimum established in the present article. recommendation INC-1 for the expression of experimental uncer-
tainties [33]. INC-1 states that the uncertainty in the results of a
3.3.1. Total error measurement generally consists of several components which may
Total error, a concept introduced 34 years ago, describes the be grouped into two categories according to the way in which their
overall error that may occur in a test result due to the simulta- numerical value is estimated.
neous contribution of random and systematic error components of
• Uncertainty Type A is the evaluation of uncertainty by statisti-
the measurement procedure [34]. In general terms, the total error
(Etotal ) could be defined by the expression: cal methods. The components of this category are determined by
means of the variances or standard deviations. The evaluation of
Etotal = Esystematic + Erandom this uncertainty applies to random and systematic errors.
• Uncertainty Type B is the evaluation of uncertainty by other
where the term E represents the error and the associated system- means. The components of this category are usually based on
atic and random subscripts define the individual error contribution. scientific judgement using all the relevant information available.
From the previous expression, it is evident that the mentioned For instance, uncertainty assigned to reference data taken from
differences between the regulatory bodies regarding the defini- handbooks, reference standards certificates, etc.
tion of the parameter that allows to estimate the systematic errors • Combined uncertainty is characterised by the summation of the
(Esystematic ), termed for instance “accuracy” or “trueness”, make Type A and Type B variances and expressed as standard devia-
mandatory the use of appropriate documentation of the particular tions.
definition used in the calculation of the total error. By document-
ing the specific definition used to express the systematic errors The task of determining the uncertainty of a particular measure-
it is possible to derive alternative expressions for total error and ment method requires the participation of experts familiar with the
consequently avoid ambiguities between definitions. For instance, various aspects involved in the recommendations issued by accred-
the following alternative expressions can be found in the current ited bodies. For instance, a laboratory interested in applying ISO
literature on validation: 5725-3 recommendation [38] in the evaluation of the uncertainty of
Etotal = Accuracy + Precision (using ICH terminology); a chromatographic method will require the assistance of an expert
familiar with the measurement method and its application and also
an expert with experience in the implementation of the guidance
Etotal = Trueness + Precision
recommended fully nested design and its statistical analysis.
or It should be noted that even though the author of the present
article has reported several studies on the determination of the
Etotal = Accuracy (using ISO terminology). uncertainty in linear calibration, central composite designs and
 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234 2229

nested designs for proficiency testing, it is not his intention to 3.4.1. Linear calibration function
present a detailed coverage of the subject. The interested reader The expression y = f(x) + ı can be rewritten as
is referred to the comprehensive work on uncertainty by Kimothi
[39]. ŷi = xi + ı (1)

where ŷi and xi represent the estimated experimental response

3.4. Linear function analysis
and the analytical concentration respectively, both at a concentra-
tion level i. The coefficients  and ı represent the sensitivity of the
Linear function analysis is an area familiar to everyone involved
analysis and the intercept respectively.
in calibration experiments and perhaps this familiarity has led
The calibration function described by Eq. (1) must have an r close
to the introduction of some fatal flaws when the linearity of a
to 1 but the condition given by “if linear, then r = 1” is not equivalent
calibration is assessed. Nowadays, such erroneous procedures are
to “if r = 1, then linear”. In logical analysis terms, the best strategy to
described in oral presentations, laboratory reports and papers and
show that a condition is not equivalent to its converse is to provide
accepted as correct.
a counterexample to it. To find a counterexample to “if linear, then
Linearity, in the context of the previously described analysis sys-
r = 1” it suffices to show that there is something that has r = 1 but it is
tem consisting of the basic elements input → converter → output,
not linear. Appendix A provides a comprehensive counterexample
is the assumption that there is a straight line relationship between
based on statistical analysis of the various error sum squares and
the input (x) and output (y) variables that can be written mathemat-
variance components from a linear calibration data set reported
ically by the expression y = f(x) if the straight line crosses through
in the literature [45]. Before studying Appendix A, some impor-
the origin or by the expression y = f(x) + ı if the straight line does
tant aspects such as replication, error sum squares and degrees of
not cross through the origin.
freedom should be discussed in advance.
It is common practice to check the linearity of a calibration
curve by inspection of the correlation coefficient r. A correlation
coefficient close to unity (r = 1) is considered sufficient evidence 3.4.1.1. Replication. Replication is an important aspect that must
to conclude that the experimenter has a perfect linear calibration. be considered when the experimenter wants to test if a particular
Although the Analytical Methods Committee and some articles on experimental calibration model, for instance Eq. (1), is linear. The
analytical validation discouraged using the correlation coefficient experimenter must have a reasonable number of standard solutions
in the context of testing for linearity [40–43], many laboratories and instrumental replicates. It has been pointed out that the best
around the world base the linearity of their instrumental meth- calibration strategies are those with standard solution replicates
ods on a so-called (by the author of the present article) “r-test”. higher than instrumental replicates [42,46]. A preparation error
Countless published papers reinforce the idea, perhaps indirectly, of 26 times higher than the instrumental error has been reported
using the r-test to check for linearity by reporting in their abstracts, in calibration experiments of triacylglycerols by LCMS [47]. This
discussions and conclusions statements such as implies that the most serious problems are related to prepara-
tion and not to instrumental stability. Calibration experiments with
“This method has a linear calibration range of 1.00–1000 ng/ml only one standard per concentration level are a poor calibration
with a correlation coefficient of 0.9999.” strategy and must be avoided unless the standard solutions are
“The calibration graphs were linear with correlation coefficients effectively error-free. For the establishing of linearity the Analytical
greater than 0.999 for all compounds.” Methods Committee suggests preparing a minimum of six concen-
“It was clear that the calibration was linear as a result of a corre- tration levels in duplicates [42]. Even though duplication at each
lation coefficient close to 1 (rexperimental = 0.9996).” concentration level is considered an optimal design strategy by this
accredited organisation, it is a poor approach that should not be fol-
The author of the present article is not stating that the previ- lowed. A study of the behaviour of the uncertainty as a function of
ous quotations are incorrect, however it must be said that in some the number of replicates for the model described by Eq. (1) demon-
respect they are misleading in the context of linearity evaluation. In strated that performing between four and six replicates at each
addition, they fail to indicate which statistical methods were used experimental level decreases the uncertainty significantly along
to evaluate their linear relationship. the experimental range and produces a uniform confidence pre-
The FDA guidance for validation of analytical procedures [31] diction region around the centre of the calibration graph which is
which is based on the Text on Validation of Analytical Procedures an important feature for quantification experiments [48]. Based on
issued by the ICH [44], recommends that the correlation coeffi- the reported behaviour of the uncertainty as a function of the repli-
cient should be submitted when evaluating a linear relationship cation and considering that the minimum number of concentration
and that the linearity should be evaluated by appropriate statis- levels proposed by various guidelines and articles on analytical val-
tical methods. This guidance does not suggest that the numerical idation varies between five and six, it is reasonable to measure the
value of r can be interpreted in terms of degrees of deviation from linearity of a calibration function by preparing a minimum of five
linearity. Hence, it is extremely important to emphasise that an concentration levels in triplicates [31,43].
“r-test” to check for linearity does not exist. We cannot say that
r = 0.999 is more linear than r = 0.997. It is surprising that despite the 3.4.1.2. Error sum squares. After selecting a sensible number of
recommendations of the accredited bodies, few published articles concentration levels (I) and replicating every concentration level
have reported the use of statistical methods for linearity evalu- J-times in a particular calibration experiment, the summation of
ation. This observation seems to indicate that the issue, pointed three squared differences, namely the residual error sum of squares
out over half a century ago, about statistical data analysis being a (SSr ) pure experimental error sum of squares (SSε ) and lack-of-fit
neglected subject by practitioners validating methods [3] is still a error sum of squares (SSlof ), must be calculated according to the
topic of contemporary relevance that needs an imperative atten- following equations:
tion by those currently engaged in validation of analytical methods
and by those responsible for educating and training people to be 
I 
Ji
2
SSr = (yij − ŷi ) (2)
embarked upon the various aspects of this important area of ana-
lytical chemistry. i=1 j=1
2230 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234


I 
Ji Analytical Method Committee suggests using the F-test as a reli-
SSε = (yij − ȳi )2 (3) able approach to check the linearity of any calibration function.
i=1 j=1 The procedure is as follows:


I • The purely experimental variance and lack-of-fit variance desig-
2
SSlof = SSr − SSε = (ȳi − ŷi ) (4) nated by ε2 and lof 2 are estimated by computing the quotients

i=1 SSε /(IJ − I) and SSlof /(I − 2) respectively.

• The calculated ε2 and  2 variance terms are used to calculate
The term yij represents the experimental response, ŷi is the esti- lof
the Fisher variance ratio or F-test by the expression:
mated response obtained by using Eq. (1), and ȳi is the average
response at every concentration level. 2
lof
F(I−2)/(IJ−I) = (9)
ε2
3.4.1.3. Degrees of freedom. The degrees of freedom (DF) associated
• The value of F(I–2)/(IJ–I) calculated experimentally is compared
to Eqs. (2)–(4) are respectively:
against the critical value of F found in statistical tables, generally
DFr = (IJ − 2) (5) at the 95% confidence level for I − 2 and IJ − J degrees of freedom in
the numerator and denominator respectively. If the experimental
DFε = (IJ − I) (6) data set describes a genuine linear calibration of the form given
by Eq. (1) then the condition Ftabulated > F(I–2)/(IJ–I) must be fulfilled.
DFlof = (I − 2) (7)
Otherwise there are grounds to suspect that a different model to
The bracketed number 2 in Eqs. (5) and (7) and associated with the described by Eq. (1) must be proposed.
Eqs. (2) and (4) respectively, represents the number of parameters
described by Eq. (1) (the  slope + the ı intercept = 2 parameters). If The estimation of the various error sum squares and the Fisher
a model with a different number of parameters to those described ratio for testing the acceptability of a linear model proposed in the
by Eq. (1) were studied, for instance: literature is discussed in Appendix A.

ŷi = xi + ϕxi2 + ı (8) 3.5. Range

The degrees of freedom associated with Eqs. (2) and (4) would In general, the range of an analytical procedure can be defined
be (IJ – 3) and (I – 3) respectively. The bracketed number 3 in this as the interval between the upper and lower concentration of ana-
case, represents the three parameters ( + ϕ + ı) of Eq. (8). lyte for which suitable precision, accuracy and linearity have been
By using Eq. (1) and the minimum criteria of five concentration demonstrated. The literature on method validation describes differ-
levels (I = 5) in triplicates (J = 3) established in the previous section, ent ranges. For instance, linear range, analytical range, calibration
it is possible to estimate 13, 10 and 3 degrees of freedom for SSr , SSε range, dynamic range, working range. However, they can be sum-
and SSlof respectively. marised as working (or analytical) range and linear (or calibration,
or dynamic) range. The former range which is wider than the latter,
3.4.1.4. Acceptability of linearity data. A ScienceDirect search was describes the interval between the lowest (limit of detection) and
performed using the keywords “linearity test” and revealed that the highest concentration where the signal can be related to the
several articles used these two words as a true measure of linearity concentration for the evaluation of random and systematic errors.
in chromatography method validation. It has been reported repeat- The linear range corresponds to the valid interval of functional
edly that an analyst engaged in any analytical validation should be dependence of the signal on concentration or mass [52] which is
aware that there is no test for linearity as such [42,43,49–51]. The usually determined by using the method of least squares, which

Scheme 1. Main approaches proposed in the literature for determining the limit of detection.
 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234 2231

Fig. 3. Visual determination of the limit of detection for PGE2 determined visually by diluting successively replicate solutions of PGE2 at an initial concentration of 0.4 ng/ml.

in turns assumes homoscedasticy of the measurements over the lytical concentration that produces a signal equivalent to three or
linear range. two times the standard deviation of the blank sample (3 ×  blank or
To demonstrate an acceptable linear range, it is generally sug- 2 ×  blank ).
gested to prepare five different standard solutions from 50 to 150% The implications of using the previous criteria are illustrated in
of the target analytical concentration [43]. However, it is important Fig. 4 where a limit of decision derived from the blank distribution
to remember that the specified ranges often differ depending on is established at +2 ×  blank . This limit of decision, at the 95% confi-
the intended application of the procedure. For instance, the ICH rec- dence level of the blank distribution is a probability that indicates
ommends preparing five different standard solutions (plus a blank) whether or not a signal could be due to the sample (>+2 ×  blank )
from 80 to 120% for the assay of a drug or a finished product [31]. An or to the blank (<+2 ×  blank ). When 3 ×  blank is used as a criterion
analyst interested in validating a particular method should consult for estimating the LOD (Fig. 4A), the probability ␤ of obtaining a
the guidelines to encompass the recommended minimum specified false negative is 16%. Conversely, it is possible to conclude that
range of the intended method. the concentration or amount of analyte in the analysed sample
is larger than in a blank matrix with a statistical power (1 − ˇ) of
3.6. Limit of detection 84%. The statistical power measures the confidence with which it
is possible to detect a particular difference if one exists [54]. There
The limit of detection (LOD) is commonly defined as the lowest is not a conventional criterion to determine what is a suitable
amount of analyte in a sample that can be reliably detected but not statistical power, however a value of 80% is generally considered
necessarily quantitated by a particular analytical method. A decade the minimum desirable. When the criterion LOD = 2 ×  blank is used
ago, the International Organization for Standardization proposed to (Fig. 4B) the probability ␤ of reporting a false LOD (ˇ = 50%) is equal
use the term “minimum detectable net concentration” [53] defined to the probability of finding a true LOD (1 − ˇ = 50%). Although
as the confidence with which it is possible to detect an amount
of analyte in the sample larger than that in a blank sample with a
statistical power of (1 − ˇ).
Different criteria are used for evaluating the LOD. A summary
of the main approaches proposed for determining the LOD in the
literature on validation is presented in Scheme 1. Visual determi-
nation of the LOD is performed by preparing samples with known
concentrations of the analyte and by establishing the level at which
the analyte can be reliably detected. Fig. 3 shows the chromatog-
raphy LOD for PGE2 determined visually by diluting successively
replicate solutions of PGE2 at an initial concentration of 0.4 ng/ml.
The replicate solutions are diluted up to a concentration level
where the analyte is not longer detected visually. According to
Fig. 3 the visual LOD corresponds to a concentration of 0.1 ng/ml
PGE2 . The guidelines for validation of analytical procedures [31]
recommend the presentation of relevant chromatograms when
submitting a validation report where the LOD determination is
based on a visual criteria. Another approach used for estimating
the LOD is the calculation of the signal/noise relationship under
the assumption that data normality, homoscedasticity and inde-
pendency of residuals are met. An inspection of the literature on
method validation revealed that this is the most popular approach
among analysts performing validation studies. The signal to noise
relationship is determined by comparing the analytical signals at Fig. 4. Statistical power when (A) 3 ×  blank and (B) 2 ×  blank are used as criteria
known low concentrations with those of blank sample up to an ana- for estimating the limit of detection.
2232 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234

the minimum number of independent determinations required to 3.8. Ruggedness

establish the LOD is 10, it is advisable to increase the number of
replicates when the LOD is defined as 2 ×  blank to avoid reaching This parameter evaluates the constancy of the results when
wrong conclusions. Both LOD criteria (3 ×  blank or 2 ×  blank ) external factors such as analyst, instruments, laboratories, reagents,
should be justified by presenting the relevant chromatograms. days are varied deliberately. By considering these critical external
The final approach described in Scheme 1 is based on the quo- factors and inspecting Table 1, it is evident that ruggedness is a
tient of two analytical parameters, namely the standard deviation measure of reproducibility of test results under normal, expected
() and the slope of a regression curve ( as in Eq. (1)). The former operational conditions from laboratory to laboratory and from ana-
parameter could be expressed as the standard deviation of the blank lyst to analyst [58]. Ruggedness cannot be erroneously used as a
( blank ), as the residual standard deviation of the calibration curve synonymous of robustness as it is going to be explained in the next
or as standard deviation of the intercept of the calibration curve. section.
The guidelines on method validation do not express any particular
preference for the approaches described in Scheme 1, however they 3.9. Robustness
recommend that when reporting the LOD the definition used in its
evaluation should be stated. This parameter evaluates the constancy of the results when
internal factors (no external factors as in ruggedness) such as flow
3.7. Limit of quantitation rate, column temperature, injection volume, mobile phase compo-
sition or any other variable inherent to the method of analysis are
The limit of quantification (LOQ) is defined as the lowest con- varied deliberately. It is generally not considered in most validation
centration or amount of analyte that can be determined with an guidelines [46].
acceptable level of precision and accuracy [9]. Similarly to LOD, LOQ Although robustness and ruggedness aim at testing the repro-
is evaluated by using different approaches, that is: ducibility of the test results regardless of internal or external
influences respectively, the literature on method validation bears
1. Visual evaluation: samples with known analytical concentration evidence that both terms are used interchangeably. The analyst
are prepared and the minimum level at which the analyte can be performing a method validation should distinguish the similari-
quantified with an acceptable level of uncertainty is established. ties and differences between these validation parameters and avoid
2. Signal/noise ratio: the signals of samples with known analytical misconstruing ruggedness as robustness.
concentrations are compared with those of blank samples up to Classical and multivariate methodologies such as the one-factor-
an analytical concentration that produces a signal equivalent to at-the time approach or a factorial design have been proposed to
10 times the standard deviation of the blank sample (10 ×  blank ). evaluate both ruggedness and robustness. However for more com-
3. Standard-deviation/slope ratio (LOQ = 10 × /): the parameters  prehensive studies on robustness and ruggedness evaluation the
and  are calculated in the same fashion as LOD. reader is referred to [59–62].

The second approach, which is the most used in the litera- 4. Final remarks
ture, could be defined in a more general context as the lowest
amount of analyte that can be reproducibly quantified above the Method validation has evolved rapidly over the last half a cen-
LOD (LOQ = n × LOD). tury from being a neglected area of many scientific disciplines into a
Other definitions to express the LOQ different from those widely recognized process used to confirm that an analytical proce-
described above could be used, provided that the definition used dure employed for a specific test is appropriate for its intended use.
in its evaluation is stated. However, despite this rapid evolution and the efforts of different
It is important to note that the discussed approaches for the regulatory bodies to reach greater harmonization of the relevant
evaluation of the LOQ do not demonstrate that at the LOQ concen- performance indicators commonly evaluated in method valida-
tration there is an adequate accuracy and precision. The different tion, some relentless inconsistencies in relation to their definitions,
approaches proposed in the literature have been critically revised calculation and interpretation are repeatedly used. Practitioners
in an article which advocates using the accuracy profile approach should be vigilant over the herein described key aspects of method
in order to estimate an LOQ more in accordance with its contex- validation and bear in mind that their misapplication goes much
tual definition [55]. The reader interested in applying the accuracy further than a simple rejection of a submitted report or a waste
profile approach is referred to the articles of Hubert et al. [56] of money, time and resources. The acceptance and application of
and Boulanger et al. [57] who were the first to introduce this erroneous terminology can have serious implications in results reli-
concept. ability, laboratory performance and institution credibility.
 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234 2233

Appendix A. Demonstration that a correlation coefficient close to 1 is not a reliable indicator of linearity

The data printed in bold type have been reported elsewhere [45].

Comments x y ȳ

The amount of analyte (ng), the chromatographic 0.35 0.0956 0.0948 0.0934 0.0946
analyte/internal-standard peak area ratio and its 0.50 0.1356 0.1393 0.1361 0.1370
average are designated as x, y and ȳ respectively. 1.00 0.2575 0.2551 0.2535 0.2554
2.00 0.5028 0.4962 0.4940 0.4977
5.00 1.2605 1.2534 1.2516 1.2552
The calibration graph was obtained by plotting y vs x. 7.00 1.6706 1.6950 1.6928 1.6861
Proposed linear model by using the reported data. ŷ = 0.242x + 0.015
Reported squared correlation coefficient (r2 ). 2
r = 0.9995

Testing the acceptability of the above linear model by using the principles and equations described in Section 3.4.
2 2 2
x (y − ŷ) (y − ŷ) (y − ŷ)

0.35 1.68 × 10−5 2.40 × 10−5 3.97 × 10−5 1.00 × 10−6 4.00 × 10−8 1.44 × 10−6 2.60 × 10−5 2.60 × 10−5 2.60 × 10−5
0.50 1.60 × 10−7 1.09 × 10−5 1.00 × 10−8 1.96 × 10−6 5.29 × 10−6 8.10 × 10−7 1.00 × 10−6 1.00 × 10−6 1.00 × 10−6
1.00 2.50 × 10−7 3.61 × 10−6 1.23 × 10−5 4.55 × 10−6 7.11 × 10−8 3.48 × 10−6 2.67 × 10−6 2.67 × 10−6 2.67 × 10−6
2.00 1.44 × 10−5 7.84 × 10−6 2.50 × 10−5 2.64 × 10−5 2.15 × 10−6 1.34 × 10−5 1.78 × 10−6 1.78 × 10−6 1.78 × 10−6
5.00 1.26 × 10−3 8.07 × 10−4 7.08 × 10−4 2.84 × 10−5 3.12 × 10−6 1.27 × 10−5 9.10 × 10−4 9.10 × 10−4 9.10 × 10−4
7.00 1.47 × 10−3 1.96 × 10−4 2.62 × 10−4 2.41 × 10−4 7.86 × 10−5 4.44 × 10−5 5.23 × 10−4 5.23 × 10−4 5.23 × 10−4
Residual error sum SSr = SSε = SSlof =
squares (Eq. (2)) 
I

Ji

I

Ji

I
2
(yij − ŷi ) (yij − ȳi )2 (ȳi − ŷi )
2
Pure error sum squares
i=1 j=1 i=1 j=1 i=1
(Eq. (3))
(Eq. 2) (Eq. 3) (Eq. 4)
Lack-of-fit error sum
squares (Eq. (4))
Results after applying SSr = 4.86 × 10−3 SSε = 4.69 × 10−4 SSlof =4.40 × 10−3
Eqs. (2)–(4) →
Degrees of freedom (DF) 18 − 2 = 16 18 − 6 = 12 6−2=4
Associated variances r2 = 3.04 × 10−4 ε2 = 3.91 × 10−5  2 = 1.10 × 10−3
lof
( 2 = SS/DF)
Fisher ratio (F =  2 /ε2 ) 28.151 (calculated)  3.259 (tabulated at the 95% with 4 and 12 degrees of freedom)
lof
calculated (if
Fcalculated < Ftabulated
then Linear)
Conclusions Although r2 = 0.9995 there is a significant non-linearity in the reported data. The proposed linear model ŷ = 0.242x + 0.015 is not
adequate to describe the observed data due to some significant lack of fit.

References [24] International Organization for Standardization. Guide 5725-1: Accuracy (True-
ness and Precision) of Measurement Methods and Results, Part 1: General
[1] International Organization for Standardization. Guide 8402: Quality Vocabu- Principles and Definitions. ISO, Geneva, 1994.
lary. ISO, Geneva, 1994. [25] W.W. Hauck, W. Koch, D. Abernethy, R.L. Williams, Pharmacol. Forum 34 (2008)
[2] W.J. Murphy, Anal. Chem. 19 (1947) 821. 838.
[3] B.L. Clarke, Anal. Chem. 19 (1947) 943. [26] L. Ebdon, L. Pitts, R. Corneils, H. Crews, O.F.X. Donard, P. Quevauviller (Eds.),
[4] F.C. Strong, Anal. Chem. 19 (1947) 968. Trace Elements Speciation for Environment Food and Health, Royal Society of
[5] A.L. Wilson, Talanta 17 (1970) 21. Chemistry, Cambridge, 2001.
[6] A.L. Wilson, Talanta 17 (1970) 31. [27] International Union of Pure Applied Chemistry. Harmonised Guidelines for In-
[7] A.L. Wilson, Talanta 20 (1973) 725. House Validation of Methods of Analysis (Technical Report). IUPAC, Budapest,
[8] A.L. Wilson, Talanta 21 (1974) 1109. 1999. <http://old.iupac.org/divisions/V/501/draftoct19.pdf>.
[9] M.E. Swartz, I.S. Krull, Analytical Method Development and Validation, Marcel [28] R. Boqué, A. Maroto, J. Riu, F.X. Rius, Grasas y Aceites 53 (2002) 128.
Dekker Inc., New York, 1997. [29] P. Van Zoonen, R. Hoogerbrugge, S.M. Gort, H.J. Van de Wiel, H.A. Van’t Klooster,
[10] http://fshn.ifas.ufl.edu/faculty/STTalcott/www/Food%20Analysis/Anthocyanin- Trends Anal. Chem. 18 (1999) 584.
SPE.doc. [30] International Organization for Standardization/International Electrotech-
[11] L. Huber, LC–GC Int. 11 (1998) 96, www.geocities.com/HotSprings/Spa/6896/ nical Commission. R103: General Requirements: Proficiency Testing for
methval.pdf. ISO/IEC 17025 Laboratories. ISO/IEC, 2008. <http://a2la.org/requirements/
[12] G.D. Boef, A. Hulanicki, Pure Appl. Chem. 55 (1983) 553. A2LA General Requirements for Proficiency Testing.pdf>.
[13] B. Persson, J. Vessman, Trends Anal. Chem. 17 (1998) 117. [31] International Conference on Harmonization. Guidance for Industry. Q2B:
[14] M. Valcárcel, A. Gómez-Hens, S. Rubio, Trends Anal. Chem. 20 (2001) 386. Validation of Analytical Procedures: Methodology. ICH, 1996. <http://www.
[15] B. Persson, J. Vessman, Trends Anal. Chem. 20 (2001) 526. fda.gov/Cder/Guidance/1320fnl.pdf>.
[16] Western European Laboratory Accreditation Cooperation. Guidance Document [32] US Food and Drug Administration. Guidance for Industry. Bioanalytical method
No. WG D2. EURACHEM/WELAC Chemistry, Teddington, 1993. Validation. FDA, 2001. <http://www.fda.gov/CDER/GUIDANCE/4252fnl.pdf>.
[17] I. Taverniers, M. De Loose, E. Van, Bockstaele, Trends Anal. Chem. 23 (2004) 535. [33] International Organization for Standarization. DGuide 99999: International
[18] A.G. González, M.A. Herrador, Trends Anal. Chem. 26 (2007) 227. Vocabulary of basic and General Terms in Metrology (VIM). ISO, 2004.
[19] R.D. McDowall, Anal. Chim. Acta 391 (1999) 149. <http://www.ntmdt.ru/download/vim.pdf>.
[20] R. Bonfiglio, R.C. King, T.V. Olah, K. Merkle, Rapid Commun. Mass Spectrom. 13 [34] J.O. Westgard, R.N. Carey, S. Wold, Clin. Chem. 20 (1974) 825.
(1999) 1175. [35] J.O. Westgard, R.W. Burnett, Clin. Chem. 36 (1990) 1629.
[21] P.J. Taylor, Clin. Biochem. 38 (2005) 328. [36] C.T. Viswanathan, S. Bansal, B. Booth, A.J. DeStefano, M.J. Rose, J. Sailstad, V.P.
[22] D. Dadgar, P.E. Burnett, J. Pharm. Biomed. Anal. 14 (1995) 23. Shah, J.P. Skelly, P.G. Swann, R. Weiner, AAPS J. 9 (2007) E30.
[23] J.N. Miller, J.C. Miller, Statistics and Chemometrics for Analytical Chemistry, [37] R. Bischoff, G. Hopfgartner, H.T. Karnes, D.K. Lloyd, T.M. Phillips, D. Tsikas, G.
Prentice-Hall, Harlow, England, 2000. Xu, J. Chromatogr. B 860 (2007) 1.
2234 P. Araujo / J. Chromatogr. B 877 (2009) 2224–2234

[38] International Organization for Standardization. Guide 5725-3: Accuracy (True- [51] Analytical Methods Committee. AMC Technical Brief 3, 2000, 1. <http://www.
ness and Precision) of Measurement Methods and Results, Part 3: Intermediate rsc.org/images/brief3 tcm18-25924.pdf>.
Measures of the Precision of a Standard Measurement Method. ISO, Geneva, [52] O. Mathias, Chemometrics: Statistics and Computer Application in Analytical
1994. Chemistry, Wiley–VCH, Weinheim, 1999.
[39] S.K. Kimothi, The Uncertainty of Measurements, Physical and Chemical Impact [53] International Organization for Standarization. ISO 11843-1: Capability of Detec-
and Analysis, ASQ Quality Press Publication, Milwaukee, USA, 2002. tion. Part 1: Terms and Definitions. ISO, Geneva, 1997.
[40] Analytical Methods Committee, Analyst 113 (1988) 1469. [54] P. Araujo, L. Frøyland, J. Chromatogr. B 847 (2007) 305.
[41] M. Thompson, Anal. Proc. 27 (1990) 142. [55] E. Rozet, A. Ceccato, C. Hubert, E. Ziemons, R. Oprean, S. Rudaz, B. Boulanger,
[42] Analytical Methods Committee, Analyst 119 (1994) 2363. Ph. Hubert, J. Chromatogr. A 1158 (2007) 111.
[43] J.M. Green, Anal. Chem. 68 (1996) 305A. [56] Ph. Hubert, P. Chiap, J. Crommen, B. Boulanger, E. Chapuzet, N. Mercier, S.
[44] International Conference on Harmonization. Guideline for Industry. Q2A: Bervoas-Martin, P. Chevalier, D. Grandjean, P. Lagorce, M. Lallier, M.C. Laparra,
Text on Validation of Analytical Procedures. ICH, 1995. <http://www.fda.gov/ M. Laurentie, J.C. Nivet, Anal. Chim. Acta 391 (1999) 135.
cder/Guidance/ichq2a.pdf>. [57] B. Boulanger, P. Chiap, W. Dewé, J. Crommen, Ph. Hubert, J. Pharm. Biomed. Anal.
[45] S. Umeda, G.W. Stagliano, M.R. Borenstein, R.B. Raffa, J. Pharmacol. Toxicol. 32 (2003) 753.
Methods 51 (2005) 73. [58] http://www.labcompliance.com/tutorial/methods/default.aspx.
[46] C. Hartmann, J. Smeyers-Verbeke, D.L. Massart, R.D. McDowall, J. Pharm. [59] M. Mulholland, Trends Anal. Chem. 7 (1988) 383.
Biomed. Anal. 17 (1998) 193. [60] E. Aguilera, R. Lucena, S. Cárdenas, M. Valcárcel, E. Trullols, I. Ruisánchez, Trends
[47] P. Araujo, T. Nguyen, L. Frøyland, J. Am. Soc. Mass Spectrom. 16 (2005) Anal. Chem. 25 (2006) 621.
388. [61] J.J.B. Nevado, M.J.V. Llerena, C.G. Cabanillas, V.R. Robledo, J. Chromatogr. A 1123
[48] P. Araujo, L. Frøyland, Accred. Qual. Assur. 10 (2005) 185. (2006) 130.
[49] R. Cassidy, M. Janoski, LC–GC 10 (1992) 692. [62] B. Dejaegher, Y.V. Heyden, J. Chromatogr. A 1158 (2007) 138.
[50] C.A. Dorschel, J.L. Ekmanis, J.E. Oberholtzer, F.V. Warren, B.A. Bidlingmeyer,
Anal. Chem. 61 (1989) 951A.