Trends in quality in the analytical laboratory. I. Traceability and measurement uncertainty of analytical results

Transcription

1 Trends in quality in the analytical laboratory. I. Traceability and measurement uncertainty of analytical results Isabel Taverniers, Erik Van Bockstaele, Marc De Loose Credibility of analytical data has never caught the public s eye more than today. The key principle for quality and reliability of results is comparability between laboratories and on a wider, international basis. In order to be comparable, analytical results must be reported with a statement of measurement uncertainty (MU) and they must be traceable to common primary references. This work focuses on traceability and uncertainty of results. We discuss different approaches to establishing traceability and evaluating MU. We place both concepts in the broader context of analytical method validation and quality assurance. We give up-to-date information in the framework of new, more exacting European and international standards, such as those from Eurachem/CITAC, IUPAC and ISO. ª 2004 Published by Elsevier B.V. Keywords: Analytical method validation; Measurement uncertainty; Quality assurance; Reliability of results; Traceability Abbreviations: AOAC, Association of Official Analytical Chemists; AQA, analytical quality assurance; CCMAS, Codex Committee on Methods of Analysis and Sampling; CITAC, Cooperation on International Traceability in Analytical Chemistry; CRM, certified reference material; EAL, European Cooperation for Accreditation; FAO, Food and Agricultural Organization; IEC, International Electrotechnical Isabel Taverniers*, Marc De Loose Department for Plant Genetics and Breeding (DvP), Centre for Agricultural Research (CLO), Ministry of the Flemish Community, Caritasstraat 21 B-9090 Melle, Belgium Erik Van Bockstaele Department for Plant Genetics and Breeding (DvP), Centre for Agricultural Research (CLO), Ministry of the Flemish Community, Caritasstraat 21 B-9090 Melle, Belgium Department for Plant Production, Ghent University, Coupure Links 653 B-9000 Gent, Belgium *Corresponding author. Tel.: ; Fax: ; i.taverniers@clo.fgov.be Commission; ILAC, International Laboratory Accreditation Cooperation; IQC, Internal quality control; ISO, International Standardization Organization; IUPAC, International Union of Pure and Applied Chemistry; LGC, Laboratory of the Government s Chemists; MU, measurement uncertainty; QA, quality assurance; QC, quality control; RM, reference material; SI, Systeme International; U(u), Expanded uncertainty (individual uncertainty factor); WHO, World Health Organization; X, concentration or amount of measurand 1. Introduction: quality of analytical results Innumerable types of analytical methods exist in the fields of analytical and bioanalytical chemistry, biochemistry, biology, clinical biology and pharmacology and related application domains, such as forensic, toxicological, environmental, agricultural and food analyses. Regardless of the type of method, the scope and the application, laboratories must be able to produce reliable data when performing analytical tests for a client or for regulatory purposes. With the fast development of analytical methodologies, great importance is nowadays attached to the quality of the measurement data. Quality of analytical measurement data encompasses two essential criteria utility and reliability (Fig. 1) [1]. Utility means that analytical results must allow reliable decision making. A key aspect of reliability or validity of results is that they are comparable, whatever their origin. Comparability between results in the strict sense is provided by traceability to appropriate standards. Traceability to common reference standards underlies the possibility of making a comparison i.e., a distinction between different results. If results are also to be compared in terms of their quantities or levels of analyte, additional information on the analytical result is needed MU. Uncertainty of results arises from the combination of all uncertainties of the reference values (to which the /$ - see front matter ª 2004 Published by Elsevier B.V. doi: /s (04)

2 Trends in Analytical Chemistry, Vol. 23, No. 7, 2004 Trends quality utility reliability (validity) comparability 1. Are 2 results comparable to, i.e. distinguishable from each other? Y1 = Y2? traceability calibration MU estimation 2. Which of the 2 results contains, with a certain level of confidence, the highest level of analyte? Y1>Y2 or Y1<Y2? Figure 1. Relationship between quality, traceability and measurement uncertainty (MU) of results [3]. results are traceable) and all additional uncertainties associated with the measurement procedure. MU and traceability are related concepts, both defining the quality of analytical data (Fig. 1) [2,3]. Quality of results reflects adequacy (or inadequacy) of a method in terms of the extent to which the method fulfils its requirements or is fit for its particular analytical purpose (see Section 2 below). Quality is always a relative notion, referring to the requirements fixed beforehand on the basis of national or international regulations or customer needs [1,4]. The need for reliability of analytical data is stressed by the fact that measurement results will be used and may form the basis for decision making. Unreliable results bring a high risk of incorrect decisions and may lead to higher costs, health risks, and illegal practices. Imagine, for example, the consequences if results are false positives or if the uncertainty is much larger than reported [1,5,6]. The process of providing an answer to a particular analytical problem is presented in Fig. 2. The analytical system which is a defined method protocol, applicable to a specified type of test material and to a defined concentration rate of the analyte must be fit for a particular analytical purpose [4]. This analytical purpose reflects the achievement of analytical results with an acceptable standard of accuracy. Without a statement of uncertainty, a result cannot be interpreted and, as such, has no value [8]. A result must be expressed with its expanded uncertainty, which, in general, represents a 95% confidence interval around the result. The probability that the mean measurement value is included in the expanded uncertainty is 95%, provided that it is an unbiased value that is made traceable to an internationally recognized reference or standard. In this way, the establishment of traceability and the calculation of MU are linked to each other. Before MU is estimated, it must be demonstrated that the result is traceable to a reference 2. The role of method validation in traceability and MU An analysis is a complex multistage investigation of the values of the properties of materials, i.e., the identity and the concentration of a specific component in a specific sample material [2,7]. van Zoonen et al. [1] presented chemical analysis as a cyclic process in which the final objective is the generation of chemical information. This integrated process starts with defining the basic analytical problem (specifying the analytical requirement) and ends with evaluating and reporting the analytical result. Ideally, the last step provides an answer to the initial problem, as stated by a client or based on regulatory requirements. VALIDATION ANALYTICAL SYSTEM measurement value measurand ANALYTICAL RESULT ± fitness-for-purpose uncertainty INTERPRETATION & EVALUATION VALIDATION Figure 2. Role of method validation in quality of analytical measurements. Validation is the process to demonstrate the fitnessfor-purpose of the analytical system [4,8,14,15]

3 or standard which is assumed to represent the truth [9,10]. Traceability and MU both form parts of the purpose of an analytical method. Validation plays an important role here, in the sense that it confirms the fitness-for-purpose of a particular analytical method [4]. The ISO definition of validation is confirmation by examination and provision of objective evidence that the particular requirements of a specified intended use are fulfilled [7]. Validation is the tool used to demonstrate that a specific analytical method measures what it is intended to measure, and thus is suitable for its intended purpose [2,11]. In part II of this review, the classical methodvalidation approach is described, based on evaluation of a number of method-performance parameters. Summarized, the criteria-based validation process consists of precision and bias studies, a check for specificity/ selectivity, a linearity check, robustness studies and, eventually, based on the practical requirements of the method, an assessment of the limits of detection and/or quantification. The objective of validation is to verify that the measurement conditions and the equation used to calculate the final result include all the influences that will affect the final result. Validation measures the different effects, throughout the whole analytical system, that influence the result, and ensures that there are no other effects that have to be taken into account. A specificity test ensures that the method responds to the specific analyte of interest only, and not to other interferents or contaminants. A linearity check verifies that the supposed relationship between the signal and units used for the analyte may be used. A bias study is a certified reference material (CRM) check that demonstrates that the method is not significantly biased; and, precision and robustness studies cover the effects of variability in conditions, operators, equipment and time. The role of method validation in the achievement of reliable results is: (1) to include all possible effects or factors of influence on the final result; (2) to make them traceable to stated references (reference methods, RMs or SI units); (3) to know the uncertainties associated with each of these effects and with the references. Validation is thus a tool to establish traceability to these references [2 4]. In this context, it is important to see the traceability and accuracy. A method that is accurate, in terms of true (i.e., approximating the true value ), is always traceable to what is considered to be the true value. However, the opposite is not correct. A method that is traceable to a stated reference is not necessarily true (accurate). Errors can still occur in this method, depending on the reference [12]. Analytical method validation forms the first level of quality assurance (QA) in the laboratory. Analytical QA (AQA) is the complete set of measures a laboratory must undertake to ensure that it is able to achieve high quality data continuously. Besides the use of validation and/or standardized methods, these measures are effective internal quality control (IQC) procedures (use of RMs, control charts,...), participation in proficiency testing schemes and accreditation to an international standard, normally ISO/IEC [4]. Method validation and the different aspects of QA form the subject of part II of this review (Method validation and AQA). 3. Guidelines on traceability and uncertainty of results Table 1 shows an overview of prominent institutions offering guidance and their guidelines on traceability, MU and related topics. In Europe, a leading role is played by Eurachem, a working group on analytical chemistry centralized at and originating from the UK s LGC (formerly Laboratory of the Government Chemist). Basic references are CITAC/Eurachem guides on Quality in Analytical Chemistry [2] and Traceability in chemical measurement [3], and a Eurachem guide on MU [13,14]. Eurachem has also published guides on related topics, such as RMs [7] and method validation [15]. At the international level, relevant standards are available from IUPAC, ISO and AOAC International [4,8,16,17] and from the Codex Alimentarius s working group, CCMAS [18 21]. Other helpful guides have been published by EAL [22] and ILAC [23] (see Table 1 for explanations of abbreviations). 4. The concept of traceability 4.1. Definitions Traceability is a relatively new term, gaining more and more attention in analytical measurement sciences. Traceability can be assigned to different aspects related to a measurement such as traceability of a result, method, procedure, laboratory, product, material, and equipment. As such, there is no single definition of traceability. Before exploring the different concepts of traceability, we can look to a more general, extended meaning. According to Valcarcel and Rios [24], the basic meaning of traceability integrates (1) the establishment of one or more relationships to well-stated references or standards, and (2) the documented history of a product or a system

4 Trends in Analytical Chemistry, Vol. 23, No. 7, 2004 Trends Table 1. Overview of European and international guiding institutions and regulatory bodies with their guidelines and standards on traceability, measurement uncertainty (MU) and related topics Body Full name Guidance on References Eurachem A focus for analytical chemistry in Europe Traceability [2,3] CITAC Cooperation on international traceability in analytical MU [13,14] chemistry Reference materials [7] Validation [15] IUPAC International Union of Pure and Applied Chemistry MU [4,8,16,17] ISO International Standardization Organisation AOAC International Association of Official Analytical Chemists FAO/WHO: Codex/CCMAS Food and Agricultural Organization/World Health Organisation: Codex Committee on Methods of Analysis and Sampling MU [18 21] EAL European Cooperation for Accreditation MU [22] ILAC International Laboratory Accreditation Cooperation MU [23] These two parts of the basic meaning can be found again when defining traceability as a property or a characteristic of different analytical facets. The different concepts of traceability are shown in Fig. 3. The most obvious definition of traceability is a property of the result of a measurement or the value of a standard whereby it can be related to stated references, usually national or international standards, through an unbroken chain of comparisons all having stated uncertainties. The different elements and the practical use of this basic definition will be explained in Section 4.2 below. Traceability of a result is related to traceability of a method, which in turn is linked to traceability of standards and traceability of the equipment used in the analytical procedure (Fig. 3). A method is called traceable when it produces results (with their uncertainties) that are characterized by a defined traceability to wellstated references [24]. Walsh [25] defines traceable methods as validated official or standard methods or validated methods which contain uncertainty statements and which are embedded into a quality system and anchored to a common reference point. Traceability among standards is considered as the most relevant basis for traceability of results [26], as shown in Fig. 3. Traceability of equipment is defined as the detailed, timely, and customised recording of installation, malfunctioning and repairs, periodic calibration and corrections (if needed), hours of use, samples processed, standard used, etc., in such a way that all questions (what?, how?, who?, etc.) should have a detailed answer in the pertinent documents [24]. Calibration is the set of operations used to establish the relationship between values shown by a measuring instrument and the values of measurement standards. By calibrating, the results of measurements are related to and thus made traceable to values of standards or references. In practice, calibration is performed by measuring samples with known amounts of analyte, such as CRMs, and monitoring the measurement response [2,3]. These definitions confirm the links between traceability of equipment, standards and results (Fig. 3). traceability of standards traceability of results traceability of methods traceability of equipment extended meaning of traceability 1. relationship(s) to well-stated references or standards 2. documented 'history' of a product or a system Figure 3. Different concepts and extended meaning of traceability [24]

5 4.2. Traceability in practice The practical establishment of traceability is based on a step-by-step implementation of the definition. The ISO definition of traceability, originating from metrology (see above), can be translated into three basic steps: (1) to establish one or more links to well-stated references, (2) through an unbroken chain of comparisons, and (3) to estimate all uncertainties associated with those comparisons [12,24,27]. This definition is very much in line with the more practical definition described by Eurachem/CITAC [3] in its procedure for traceability that consists of the following steps: (1) specifying the measurand, the scope of measurement and the required uncertainty; (2) choosing the method of measurement; (3) validating the method of measurement; (4) identifying/quantifying all influences that will affect the result; (5) choosing appropriate references; (6) estimating the uncertainty components associated with all influences and references [3]. The key principle in both approaches is that relationships to stated references are established and that this is done through an unbroken chain of comparisons. Practically, this means that the analytical procedure is first described as a chain or a flow diagram (step (2) in ISO definition; steps (1) and (2) in Eurachem/CITAC definition). The word unbroken means that there is no loss of information when considering the different steps in the analytical procedure leading to the measurement result. Each step in the procedure then needs to be linked to either a reference method, a RM or an SI unit (step (1) in ISO definition; step (5) in Eurachem/CITAC definition) [12,24,27]. Fig. 4 depicts the successive classes of stated references (materials or methods) in a so-called traceability chain. Establishing traceability through a traceability chain brings a certain level of uncertainty, called calibration uncertainty or traceability uncertainty (see also Section 6 below) [3,7,28]. This then brings us to the third key element when applying definitions of traceability the stated uncertainties. Each step in the traceability chain, with the uncertainty components of all the stated references, will contribute to the measurement result and thus to the uncertainty associated with it. Uncertainty components must thus be estimated at each step in the analytical process (step (3) in ISO definition; step (6) in Eurachem/ CITAC definition). As described above, validation is a tool to identify all possible effects or factors within the analytical procedure that can influence the final result. As such, steps (3) and (4) in the Eurachem/CITAC definition are additional steps that can be very helpful in establishing traceability [3]. Examples of how traceability is established in practice can be found in the literature. Recently, a Special Issue of TrAC was published on Challenges for achieving traceability of environmental measurements (TrAC, Volume 23, 2004). This issue contains a lot of up-todate information and practical examples in the particular domain of environmental analysis. Many authors reported on the most important and most difficult step in establishing traceability the selection of stated references or standards. For different stages in the traceability chain shown in Fig. 4, descriptions and examples are given by Quevauviller and Donard [27], Charlet and Marschal [29] and Segura et al. [30]. Pan [28], F orstner [31] and Theocharopoulos et al. [32] applied the ISO definition for establishing traceability in different types of higher level of traceability in-house/working RMs certified RMs primary RMs SI units in-house/working method spiking reference method primary method higher level of method bias and uncertainty Figure 4. The traceability chain and the relationship between traceability and uncertainty of measurements. The three possibilities for establishing traceability referred to in Fig. 5, are indicated in bold [25,28]

6 Trends in Analytical Chemistry, Vol. 23, No. 7, 2004 Trends environmental methods of analysis. A similar approach was followed by Sabe and Gauret [33] and Drolc et al. [34]; however, they based it upon the Eurachem/CITAC Guide on Traceability [3]. In their examples, all the authors took into account specific steps or influences in the analytical procedure that can lead to a broken chain of comparisons, such as sampling and sample treatment or preparation steps. Some authors reported on uncertainties associated in particular with sampling [35,36]. 5. The concept of MU MU is the most important criterion in both method validation and IQC. It is defined as a parameter, associated with the result of a measurement, that characterizes the dispersion of the values that could reasonably be attributed to the measurand [11,14]. The measurand refers to the particular quantity or the concentration of the analyte being measured. The parameter can be a standard deviation or the width of a confidence interval [14,37]. This confidence interval represents the interval on the measurement scale within which the true value lies with a specified probability, given that all sources of error have been taken into account [37]. Within this interval, the result is regarded as being accurate, i.e., precise and true [11]. It cannot be overemphasized that MU is different from error. The error of an individual analytical result, the the result and the true value of the measurand, is always a single value [38]. Part of the value of a known error, the systematic error, can be used to correct a result. This means that, after correction, the result of an analysis may be very close to the true value. However, the uncertainty of the measurement may still be very large, because there is doubt or limited knowledge about how close the result is to the value. Uncertainty is expressed as a range and applies to an analytical procedure and a specific sample type, but to different determinations and thus measurement results. The value of the uncertainty cannot be used to correct a measurement result. The error of an analytical result is related to the (in)accuracy of an analytical method and consists of a systematic component and a random component (Fig. 5) [14]. Precision and bias studies form the basis for evaluation of the accuracy of an analytical method [18]. The accuracy of results relates only to the fitness-for-purpose of an analytical system, assessed by method validation. analytical result true value analytical result and true value error inaccuracy expected value (limiting mean) expected value and true value systematic error bias analytical result and expeted mean value random error imprecision persistent bias run effect reproducibility intermediate precision repeatability variations within the whole analytical system, over longer periods variations during a particular run inter-laboratory variation, tested by collaborative studies within-lab variation due to inter-assay precision= random effects= variability variability over a short over a longer period of time, time interval, under the under different conditions same conditions matrix variation effect method bias laboratory bias run bias random bias indicator for expected value and true value TRUENESS indicator for result and expeced value PRECISION analysis of CRMs + statistical control duplicate analysis single-laboratory validation minimally needed in method validation Figure 5. Composition of the error of an analytical result related to trueness and precision [4,8]

7 However, reliability of results has to do with more than method validation alone. MU is more than just a single figure expression of accuracy. It covers all sources of errors that are relevant for all analyte concentration levels. MU is a key indicator of both fitness-for-purpose and reliability of results, binding together the ideas of fitness-for-purpose, QC and thus covering the whole QA system [4,37]. The MU of an analytical procedure is thus derived from, but differs from, the error of a single analytical result. The deviation of the measurement result from the true value comprises a number of systematic and random errors as shown in Fig. 5. Each of these error components adds its own uncertainty to the total uncertainty budget of the analytical procedure. The different error components are therefore referred to as sources of uncertainty. Depending on the sources of uncertainty taken into account and thus the conditions of the measurement, the overall MU will be different and another definition of MU will apply. This means that there is no single, straightforward definition of MU. It is rather a concept, the interpretation of which changes according to the measurement conditions and to the reference to which the result is traceable [10]. The different definitions of MU are subjects of the following section. 6. Different operational definitions of MU As illustrated in Fig. 6, the error of an analytical result for a specified analyte concentration comprises different error components, forming together the ladder of errors : (1) the method bias, a systematic error associated with the method as such; (2) the laboratory bias, which is either a systematic error if the laboratory is considered on its own, or a random error if the laboratory is considered as one of a group, as is the case in interlaboratory studies; (3) the run error, seen as a systematic error for one run and as a random variation over several runs performed intralaboratory; (4) the repeatability error, which is a random error from the replicate measurements performed within a single run [10]. As the error being considered applies for only a specified concentration of analyte isolated from a specified type of sample or matrix, sampling errors and matrix variation effects are not included here [4]. The more traditional distinction between error components is between random errors and systematic errors (Fig. 5). In this classical approach, random errors are generally referred to as precision (repeatability, intermediate precision and reproducibility), while systematic errors are typically attributed to the uncertainty on the bias-estimate and calibration uncertainty. To this classification, other uncertainty contributions are added, such as sampling effects, matrix effects and uncertainties associated with certain assumptions that underlie the measurement method and/or the calculation equation [2]. Result = true value + method bias + lab bias + run error + repeatability error Result = true value + traceability U + U on estimated bias + lab bias + run error + repeatability error systematic error random error Uncertainty = random error random error intermediate precision within-laboratory uncertainty reproducibility precision Bias is estimated with: 1. reference method 2. spiking U (reference) = 0 U (spike) ~ 0 two methods two measured results intermediate precision reproducibility uncertainty bias-included uncertainty 3. certified reference material (CRM) U (CRM) = 0 measured & certified value intermediate precision absolute uncertainty Figure 6. Composition of the error of an analytical result related to measurement uncertainty. Different operational definitions of measurement uncertainty [10]. Below on the left are the three possibilities for establishing traceability (see also Fig. 4)

8 Trends in Analytical Chemistry, Vol. 23, No. 7, 2004 Trends As mentioned above, each of these error components is a potential source of uncertainty. Depending on the conditions under which the analysis is performed, different sources of uncertainty contribute to the overall uncertainty. Hund et al. [10] introduced different operational definitions of uncertainty, according to the number and type of uncertainty sources considered (Fig. 6): (1) within-laboratory uncertainty, derived from intermediate precision and including only the repeatability error and the run error; (2) reproducibility uncertainty, derived from reproducibility precision (interlaboratory tests) and accounting for the repeatability error, run and laboratory effects; (3) bias-included uncertainty and absolute uncertainty additionally take into account the method bias, which is the most important source of uncertainty because it refers to a reference or a standard, to which the method is considered to be traceable. If the working method is not a primary method which is traceable to SI units (see Fig. 4) the method is always compared to another, reference method or is applied using appropriate CRMs. This reference or standard needs to be considered when the uncertainty associated with the method bias is estimated. In addition to the uncertainty associated with this reference or standard, there is the uncertainty on the estimated bias (Fig. 6). The different possibilities of bias estimation and thus of traceability are depicted in Fig. 5. If the method is compared to a reference method, the uncertainty associated with this reference method is considered negligible and the bias is estimated as the the two methods (case 1 in Fig. 6). If there is no method to compare with, bias can be estimated by spiking samples and assessing the difference between the spiked sample and the measured sample. In this case also, the uncertainty on the spike will approximate to zero (case 2) and the only method bias is the the measured sample and the spiked sample. Absolute uncertainty can be estimated only if CRMs are used (case 3). Only in this case can full traceability to SI units be guaranteed [10]. 7. Approaches to establishing MU In general, to estimate the overall uncertainty on a particular result, it is necessary to know: (1) all uncertainties arising from the measurement procedure itself; (2) all uncertainties associated with the references or standards, the analytical results are made traceable to [3]. Different approaches exist for the estimation of overall MU, as reviewed by several authors [9,10,39] and summarized in Table 2. The most well-known, traditional approach is based on identifying, quantifying and combining all individual contributions to uncertainty. In this bottom-up approach, the overall uncertainty is derived from the uncertainties of the individual components. The component-by-component assessment of MU was originally developed for physical measurements and adopted by Eurachem for chemical measurements [13]. However, because of its complexity, this methodology has significant costs in time and effort and has never found widespread applications. A simplified approach to assessing MU is the fitness-for-purpose approach, defining a single parameter called the fitness function. This fitness function has the form of an algebraic expression u ¼ f ðcþ and describes the relationship between the MU and the concentration of the analyte. For example u ¼ 0:05c means that the MU is 5% of the concentration. Calculation of the MU will hereby rely on data obtained by evaluating individual methodperformance characteristics, mainly repeatability and reproducibility precision, and preferably also bias [21,40,41]. This approach can more or less be seen as a simplification of the step-by-step protocol for testing the MU, as described by Eurachem [14]. Although MU comprises more than systematic and random errors, it can be estimated from method-validation data. Data from method-performance studies can give all or nearly all the information required to evaluate the uncertainty [2,4,18,37]. This includes the use of data from in-house and collaborative validation studies (typically precision data), proficiency-testing schemes (typically bias data) or QA data, relevant for uncertainty. If such data are available and used to estimate the uncertainty, it is not necessary to estimate MU using the component-by-component approach [18,19]. In particular, validation studies and QC measures are considered highly relevant sources for estimating MU [42,43]. Table 2 describes three methodologies for assessing MU based on validation data. In the Analytical Methods Committee s top-down approach [37], the laboratory is seen from a higher level, as a member of a population of groups. As a consequence, systematic errors within one laboratory become random errors and the estimated uncertainty is the reproducibility uncertainty (Fig. 6). Examples of MU-estimation studies using data from collaborative ring trials are works by Dehouck et al. [44] and Maroto et al. [45,46]

9 Table 2. Different approaches for estimating measurement uncertainty (MU) Reference Name of approach Basic principle Strengths Weaknesses Eurachem [13] and ISO [16] Bottom-up, error-budget, error-propagation or component-bycomponent Identification, quantification and combination of all sources of uncertainty Holistic ¼ all important sources of error should be included Complex, Expensive, Time-consuming Codex Alimentarius/ CCMAS [18 21] Fitness-for-purpose Establishment of a fitness-function, u ¼ f(c), based mainly on precision and bias studies Simple MU can be assessed for different concentrations Some sources of uncertainty may be overlooked Analytical Methods Committee [37] Top-down Based on data obtained from inter-laboratory studies (precision) MU can be assessed for different concentrations Some sources of uncertainty may be overlooked Only if data on collaborative studies are available Eurachem [14] Barwick & Ellison [47] Validation-based Based on inter- or intra-laboratory validation studies (precision, trueness, robustness) Extension of validation work, so no extra work is needed Some sources of uncertainty may be overlooked Hund et al. [39] Robustness-based Based on robustness tests as intra-laboratory simulations of inter-laboratory studies Simple, time-efficient Some sources of uncertainty may be overlooked Method must first show to be robust The other two approaches mentioned in Table 2 [14,39,47] make use of different method-performance parameters. All three validation-based methodologies can be seen as simpler, and more time- and cost-efficient extensions of validation. However, it is important to note that not all sources of uncertainty are covered by method-performance data. Some sources that may need particular consideration in addition to the available data are sampling, pre-treatment, method bias, variation in conditions and changes in the sample matrix [14,18,41]. Many of those principles for estimating MU were applied in a case study for toluene in ground water performed by Armishaw [48]. His idea was that, for a routine method that has been validated previously and performed in a laboratory where QC measures are in place, it is possible to estimate MU in a working afternoon. The key is to extract all relevant information from already available data: validation studies (bias/recovery and precision data); instrument calibration data (RM uncertainties); and, information from regularly performed QC measurements (replicate analyses and control samples) and from the method procedure itself (sampling, homogeneity of samples,...). After identifying the components contributing to the uncertainty budget and assigning the relevant sources of information, standard uncertainties were quantified and combined to form the combined standard uncertainty. The results for toluene in water were reported as x U, where U was the expanded uncertainty obtained by multiplying the combined standard uncertainty, u c, with a coverage factor of 2. In addition to this bottom-up MU estimation (Approach 1), Armishaw reported the expanded uncertainty as a function of the concentration of toluene (Approach 2). Finally, the authors compared this experimentally assessed MU with calculated MU values, based on: (1) a within-laboratory reproducibility estimate; (2) proficiency test data; (3) the models of Horwitz [49] and Thompson and Lowthian [50]. These models allowed calculation of % RSD values as a function of the analyte concentration. To obtain expanded uncertainty values, Armishaw multiplied the predicted SD values from the models by 2 [48]. All three MU calculations are variants of validation-based approaches (Approaches 3 and 4, Table 2). 8. Importance of traceability and MU The underlying motivation to establish traceability and MU is the need to make decisions based on the analytical results obtained or to be in compliance with regulatory limits (for quantitative determinations) [51]. MU is an essential feature of analytical results, for three different reasons: (1) Customers want to have an idea about the range of results and about the comparability of results between different laboratories [43]. Any result must be accompanied by a statement of MU, so that the user of the result knows the level of confidence associated with it [52]. The concepts of compara

10 Trends in Analytical Chemistry, Vol. 23, No. 7, 2004 Trends bility and reliability of results have been discussed briefly in Section 1 and are presented in Fig. 1. (2) It demonstrates traceability. Before MU can be evaluated, traceability to stated references or standards must be established. Moser et al. [43] claim that traceability is proved by the appropriate use of RMs or standards and by a full uncertainty budget. (3) There is a requirement to know the method, to understand the underlying principles and mechanisms of the measurement procedure. Lack of profound knowledge on the method itself will lead to certain unknown uncertainty contributions not being taken into account and thus to gaps in the uncertainty budget. MU can be estimated only if the method is well understood [43,53]. MU is increasingly gaining attention, in particular within the framework of accreditation. The new accreditation standard ISO/IEC [17], which has been in force from December 2002, contains clear requirements about estimating MU and when and how it should be stated in test reports. ISO/IEC requires MU to be reported when required by the client and when relevant to the application and the interpretation of the measurement results, within the framework of certain specifications or decision limits. The MU should be readily available and reported together with the result as X U, where U is the expanded uncertainty [17,47,51,54]. Eurachem and CCMAS within the Codex Alimentarius deal with MU as a separate issue [14,18 20]. The Analytical Methods Committee even claims that MU will become the main unifying principle of analytical data quality [37]. 9. Summary Rather than focusing on the techniques and methodologies being used, attention is nowadays paid to the quality and the reliability of the final results of analysis. This is influenced by greater demand for regulatory compliance and greater awareness of the customer the client wants to know the level of confidence of the reported result. In order for results to be comparable, they must be reported with a statement of MU and they must be traceable to common primary references. Methods must be validated to show that they actually measure what they are intended to measure; that they are fit for exacting European and international standards, such as the ISO/IEC norm for laboratory accreditation. On the basis of quality and reliability of analytical data rests the comparability of results for their specific purpose. An analytical method is a complex, multi-step process, starting with sampling and ending with the generation of a result. Although every method has its specific scope, application and analytical requirement, the basic principles of QA are the same, regardless the type of method or the sector of application. The information in this article is taken mainly from the analytical chemistry, but it also applies to other sectors. The validation of analytical methods, the establishment of traceability of results and the assessment of MU should be done in a uniform, harmonized way, conforming with internationally recognized standards from institutions such as Eurachem, IUPAC or ISO. This update on analytical quality provides a common understanding for the topics of method validation, traceability and MU of measurements. It has elucidated the interrelationships between method validation and traceability and MU of results. From all the guidelines and standards, we selected and summarized the most relevant information. We discussed different approaches to establishing traceability and assessing MU of analytical methods in general. We highlighted the importance of both concepts and the link with method validation and analytical QA. Acknowledgements We wish to thank Andrew Damant for giving suggestions and Friedle Vanhee for reading and assistance. References [1] P. van Zoonen, R. Hoogerbrugge, S.M. Gort, H.J. van de Wiel, H.A. van t Klooster, Trends Anal. Chem. 18 (1999) 584. [2] CITAC/Eurachem Guide: Guide to Quality in Analytical Chemistry An Aid to Accreditation, Available from: < [3] Eurachem/CITAC Guide: traceability in chemical measurement, A guide to achieving comparable results in chemical measurement, Joint Eurachem/CITAC Working Group on Measurement Uncertainty and Traceability, Available from: < [4] M. Thompson, S. Ellison, R. Wood, Pure Appl. Chem. 74 (2002) 835. [5] R. Battaglia, Accred. Qual. Assur. 1 (1996) 256. [6] R.J. Mesley, W.D. Pocklington, R.F. Walker, Analyst (Cambridge, UK) 116 (1991) 975. [7] Eurachem Guide EEE/RM/062rev3, The Selection and Use of Reference Materials, A Basic Guide for Laboratories and Accreditation Bodies, Available from: < bam.de>. [8] M. Thompson, R. Wood, Pure Appl. Chem. 67 (1995) 649. [9] A. Maroto, R. Boque, J. Riu, F.X. Rius, Trends Anal. Chem. 18 (1999) 577. [10] E. Hund, D.L. Massar, J. Smeyers-Verbeke, Trends Anal. Chem. 20 (2001) 394. [11] J. Fleming, H. Albus, B. Neidhart, W. Wegschieder, Accred. Qual. Assur. 1 (1996) 87. [12] Ph. Quevauviller, Trends Anal. Chem. 23 (2004) 171. [13] S.L.R. Ellison, M. Rosslein, A. Williams (Eds.), Eurachem/CITAC Guide: Quantifying Uncertainty in Analytical Measurement, first ed., Available from: < [14] S.L.R. Ellison, M. Rosslein, A. Williams (Eds.), Eurachem/ CITAC Guide: Quantifying Uncertainty in Analytical 489

11 Measurement, second ed., Available from: < bam.de>. [15] Eurachem Guide: The Fitness for Purpose of Analytical Methods, A Laboratory Guide to Method Validation and Related Topics, LGC, Teddington, UK, Available from: < [16] International Standards Organization, GUM: Guide to the Expression of Uncertainty in Measurement, ISO, Geneva, Switzerland, [17] ISO/IEC on General Requirements for the Competence of Calibration and Testing Laboratories, ISO, Geneva, Switzerland, [18] CX/MAS 01/8, Codex Alimentarius Commission, Codex Committee on Methods of Analysis and Sampling (FAO/WHO), Measurement uncertainty, Relationship between the analytical result, the measurement uncertainty and the specification in Codex standards, Agenda Item 4a of the 23rd Session, Budapest, Hungary, 26 February 2 March [19] CX/MAS 02/6, Codex Alimentarius Commission, Codex Committee on Methods of Analysis and Sampling (FAO/WHO), Proposed draft guidelines on measurement uncertainty, Agenda Item 5 of the 24th Session, Budapest, Hungary, November [20] CX/MAS 02/13, Codex Alimentarius Commission, Codex Committee on Methods of Analysis and Sampling (FAO/ WHO), The use of analytical results: sampling, relationship between the analytical results, the measurement uncertainty, recovery factors and the provisions in Codex standards, Agenda Item 9 of the 24th Session, Budapest, Hungary, November [21] CX/MAS 02/4, Codex Alimentarius Commission, Codex Committee on Methods of Analysis and Sampling (FAO/WHO), Proposed draft guidelines for evaluating acceptable methods of analysis, Agenda Item 4a of the 24th Session, Budapest, Hungary, November CX/MAS 02/4-Add 2 Dispute situations. [22] EAL-G23, The Expression of Uncertainty in Quantitative Testing, EAL, 1996, 9 pp. [23] ILAC-G17:2002, Introducing the concept of uncertainty of measurement in testing in association with the application of the standard ISO/IEC 17025, ILAC Technical Accreditation Issues Committee, 2002, 7 pp. Available from: < [24] M. Valcarcel, A. Rios, Trends Anal. Chem. 18 (1999) 570. [25] M.C. Walsh, Trends Anal. Chem. 18 (1999) 616. [26] M. Valcarcel, A. Rios, Fresenius J. Anal. Chem. 359 (1997) 473. [27] Ph. Quevauviller, O.F.X. Donard, Trends Anal. Chem. 20 (2001) 600. [28] X.R. Pan, Accred. Qual. Assur. 1 (1996) 181. [29] P. Charlet, A. Marschal, Trends Anal. Chem. 23 (2004) 178. [30] M. Segura, C. Camara, Y. Madrid, C. Rebollo, J. Azcarate, G.N. Kramer, B.M. Gawlik, A. Lamberty, Ph. Quevauviller, Trends Anal. Chem. 23 (2004) 194. [31] U. F orstner, Trends Anal. Chem. 23 (2004) 217. [32] S.P. Theocharopoulos, I.K. Mitsios, J. Arvanitoyannis, Trends Anal. Chem. 23 (2004) 237. [33] R. Sabe, G. Rauret, Trends Anal. Chem. 23 (2004) 273. [34] A. Drolc, M. Ros, M. Cotman, Anal. Bioanal. Chem. 378 (2004) [35] M. Thompson, Accred. Qual. Assur. 3 (1998) 117. [36] S. Roy, A.-M. Fouillac, Trends Anal. Chem. 23 (2004) 185. [37] Analytical Methods Committee, Analyst (Cambridge, UK) 120 (1995) [38] J. Fleming, H. Albus, B. Neidhart, W. Wegschieder, Accred. Qual. Assur. 2 (1997) 160. [39] E. Hund, D.L. Massart, J. Smeyers-Verbeke, Anal. Chim. Acta 480 (2003) 39. [40] Eurachem/EA Guide 04/10, Accreditation for Microbiological Laboratories, Available from: < bam.de>. [41] S. K uppers, Accred. Qual. Assur. 3 (1998) 412. [42] S.L.R. Ellison, V.J. Barwick, Analyst (Cambridge, UK) 123 (1998) [43] J. Moser, W. Wegscheider, C. Sperka-Gottlieb, Fresenius J. Anal. Chem. 370 (2001) 679. [44] P. Dehouck, Y. Vander Heyden, J. Smeyers-Verbeke, D.L. Massart, P.H. Crommen, R.D. Marini, O.S. Smeets, G. Decristoforo, W. Van de Wauw, J. De Beer, M.G. Quaglia, C. Stella, J.L. Veuthey, O. Estevenon, A. Van Schepdael, E. Roets, J. Hoogmartens, Anal. Chim. Acta 481 (2003) 261. [45] A. Maroto, J. Riu, R. Boque, F.X. Rius, Anal. Chim. Acta 391 (2003) 173. [46] A. Maroto, R. Boque, J. Riu, F.X. Rius, Anal. Chim. Acta 446 (2001) 133. [47] V.J. Barwick, S.L.R. Ellison, VAM Project 3.2.1, Development and Harmonisation of Measurement Uncertainty Principles, Part d, Protocol for Uncertainty Evaluation from Validation Data, Version 5.1, January 2000, p. 9. Available from: < ca/vam%20uncertainty.pdf>. [48] P. Armishaw, Accred. Qual. Assur. 8 (2003) 218. [49] W. Horwitz, Anal. Chem. 54 (1982) 67A. [50] M. Thompson, P.J. Lowthian, J. AOAC Int. 80 (1997) 676. [51] B. King, Fresenius J. Anal. Chem. 371 (2001) 714. [52] I. Mueller-Harvey, Food Agric. Environ. 1 (2003) 9. [53] M. R osslein, Accred. Qual. Assur. 5 (2000) 88. [54] N. Mueller, Accred. Qual. Assur. 7 (2002) 79. Isabel Taverniers graduated in Agricultural and Applied Biological Sciences from the University of Gent, Belgium, in Until April 2001, she worked at AgriFing, a joint spin-off laboratory of Gent University, Hogeschool Gent and the Department for Plant Genetics and Breeding (DvP), where she specialized in DNA fingerprinting technologies. She is now preparing a Ph.D. thesis in the Laboratory of Applied Plant Biotechnology of the Department for Plant Genetics and Breeding (CLO, Flemish Community). Erik Van Bockstaele is Head of the Department for Plant Genetics and Breeding and Professor at the Faculty of Agricultural and Applied Biological Sciences of the University of Gent. Marc De Loose is Head of the Section Applied Plant Biotechnology at the Department for Plant Genetics and Breeding