You are on page 1of 5

Accred Qual Assur (2012) 17:597–601

DOI 10.1007/s00769-012-0933-z

PRACTITIONER’S REPORT

Alternative calculation of measurement uncertainty with global


approach in food microbiology
Emrah Torlak

Received: 7 July 2012 / Accepted: 5 October 2012 / Published online: 19 October 2012
Ó Springer-Verlag Berlin Heidelberg 2012

Abstract Calculation of measurement uncertainty is a component of a quality system for food microbiology labo-
requirement for all laboratories accredited to ISO/IEC ratories. Measurement uncertainty is an important factor in
17025 including those carrying out microbiological anal- comparing the results with reference values or criteria given
yses. Today, calculation of measurement uncertainty in in product standards, microbiological guidelines or legisla-
microbiological analyses using precision data according to tive limits. It has also importance in comparing the results
global approach principles is widely recognized by the obtained from different laboratories.
microbiologists due to difficulties in quantification of Calculation of measurement uncertainty has been com-
individual uncertainty sources. In food microbiology, pre- monplace in physical and chemical laboratories for many
cision data obtained from different samples usually show years, and procedures for measurement uncertainty are well
over-dispersion, and the use of over-dispersed data may established. But, it is only relatively recently that the subject
result in large variance. The current ISO standard on has been addressed by microbiologists [2]. In food micro-
measurement uncertainty in food microbiology proposes biology, protocols and formulations given in the guidance
the use of log-transformed precision data to overcome this documents for measurement uncertainty are applicable to
problem. This paper proposes an alternative procedure to enumeration methods using a colony-count technique. For
calculate the measurement uncertainty in food microbiol- most probable number (MPN) methods, the theoretical
ogy using non-logarithmic precision data. The calculations 95 % confidence intervals given in McCrady’s tables can be
used in this procedure based on relative range of duplicate used as expanded uncertainty range if contributory com-
analyses can be applied to intra-laboratory reproducibility ponents can be shown to be under control [3–5].
data obtained from microbiological analyses of which The current ISO standard on measurement uncertainty in
duplicate results show relatively low variation. food microbiology is based on the log-transformed repro-
ducibility data [5]. However, most of the microbiological
Keywords Measurement uncertainty  criteria for foods in the EU and Turkish regulations [6, 7]
Food microbiology  Global approach  Precision have been provided non-logarithmically. Due to nature of
logarithm, ISO protocol yield uncertainties with asym-
metric confidence limits in natural (back-transformed)
Introduction values. Asymmetrical confidence limits cannot be expres-
sed as plus or minus half the confidence interval width
Measurement uncertainty, one of the technical requirements divided by the estimated value of the variable (e.g. ±10 %)
of ISO/IEC 17025 [1], provides quantitative estimates of the and can be only expressed as confidence interval.
level of confidence for test results and is therefore an essential In the field of food microbiology, unlike in other testing
fields (e.g. chemical analyses), expression of uncertainty in
test reports together with the results as logarithmic values
E. Torlak (&)
or an interval in natural values may cause difficulties in the
Department of Biology, Faculty of Science,
Necmettin Erbakan University, 42050 Konya, Turkey interpretation of results for customers and authorities. For
e-mail: torlakemrah@yahoo.com this reason, food microbiology laboratories seek an

123
598 Accred Qual Assur (2012) 17:597–601

alternative calculation protocol for measurement uncer- uncertainty estimation in food microbiology. Laboratories
tainty using non-logarithmic data. This contribution can only assess their bias by reviewing of inter-laboratory
describes a procedure for the estimation of measurement studies and proficiency trials results or testing certified
uncertainty in food microbiology using non-logarithmic reference materials. Proficiency trials can also provide
precision data. useful confirmation of realistic uncertainty estimates
[4, 5].
Measurement uncertainty associated with the precision
Different approaches to measurement uncertainty can be realistically evaluated with reproducibility data
[13]. Estimates of reproducibility may be derived from
Currently, there are two approaches to estimate uncer- inter-laboratory studies and proficiency trials, which have
tainty. They are referred to as the bottom-up approach and the benefit of providing estimates of laboratory bias, or
the top-down or global approach [3, 5, 8]. The bottom-up through the determination of intra-laboratory reproduc-
approach of which the basic principles were defined in the ibility. There are some drawbacks of using precision values
JCGM reference document entitled ‘‘Guide to the Expres- from inter-laboratory studies and proficiency trials, because
sion of Uncertainty in Measurement’’ requires the these precision values are generally obtained under limited
identification and estimation of each separate source of precisely defined combinations of matrix, strain of micro-
uncertainty and their combination to arrive at an overall organism, contamination level, and background microflora.
estimate of the uncertainty [9, 10]. This approach may be Additionally, natural variations in samples and variations
considered to be unsuitable for microbiological testing from sub-sampling have not been taken into account since
methods, because they have many sources of uncertainty the analyses are performed on certain amount (e.g. 25 g) of
and some of these sources of uncertainty (e.g. incubation homogenized samples. Thus, use of precision values from
temperature fluctuations and media quality) are not effec- inter-laboratory studies and proficiency trials causes an
tively quantifiable for any individual experiment [11]. For under-estimation of the uncertainty. For these reasons,
such reasons, international consensus supports the use of laboratories should always give priority to intra-laboratory
the global approach to evaluate microbiological measure- reproducibility data for measurement uncertainty in food
ment uncertainty as opposed to calculation of the combined microbiology although these data do not reflect laboratory
contributions of individual components of uncertainty [3]. bias [5].
The bottom-up approach can always be used in calibration
laboratories, but in testing laboratories, it should only be
used in cases where there is insufficient data to establish a Intra-laboratory reproducibility
reliable estimate of its long-term reproducibility.
Precision is closeness of the test results obtained from the
same sample and can be evaluated under three different
Global approach conditions: repeatability, intra-laboratory reproducibility and
reproducibility. Repeatability condition refers to stable fac-
The global approach which is based upon a statistical tors such as same laboratory, same equipment, same analyst,
evaluation of the test results from samples that have and short intervals of time. Reproducibility condition refers
undergone the complete measurement procedure [3, 5, 10]. to different laboratories [14]. When the term reproducibility
This approach has the advantage that most testing labora- is adopted under changed factors in a single laboratory, it is
tories are already familiar with precision experiments and a named as intra-laboratory reproducibility (within lab repro-
suitably devised reproducibility experiment will include ducibility, Rw). According to ISO 5725-3 [15], the intra-
the effects of all of the major contributors to uncertainty. laboratory reproducibility condition is achieved by changing
The global approach has two major uncertainty con- one or more of the factors from the repeatability conditions.
tributions: trueness and precision [8, 12]. Trueness is Among these factors, considerable change of time cannot be
expressed in terms of the bias, which is difference applied to food microbiology due to its nature.
between the mean test results and true value. However, Intra-lab reproducibility, sometimes called as interme-
due to empirical nature of quantitative microbiological diate precision, can be satisfactorily evaluated for
methods, it is not possible determine a true value which is quantitative analyses in food microbiology by duplicate
required to determine bias. Results of microbial enumer- analyses performed within an acceptable time interval.
ations are dependent on method-defined conditions such as Duplicate samples are tested by different analysts to allow
the strain and condition of the test organism, the tem- for monitoring of the effect of different staff as well as the
perature, incubation period and culture media. Therefore, variability in different portions of the same sample. In
in practice, only precision is taken into account for the addition to changing analysts, duplicate analyses should be

123
Accred Qual Assur (2012) 17:597–601 599

performed under conditions as different as possible and directly on the natural values for calculation of the relative
should ideally include as many variables as may be standard deviation of intra-laboratory reproducibility.
encountered within a laboratory, in terms of batches of In colony-count technique, unpredictable variation
culture media and reagents, equipment and time of analy- increases rapidly as the number of colonies on plate
sis. Thus, results of duplicate analyses can realistically decreases and global design is not suitable for low counts.
reflect the intra-laboratory imprecision [3, 16]. Variation of Enumeration results based on less than 10 counted colonies
influence quantities, such as equipment for homogenization should be excluded from calculations [18, 19].
of the initial suspension, position and elapsed time inside The relative standard deviation of intra-laboratory
the incubator, and batches of nutrient media and reagents, reproducibility (RSDRw) is calculated by [19–21]:
can only be approximated by a random choice of variables.
R
Samples used in duplicate analyses should represent the RSDRw ¼
d2
scope of the method. The routine sample composition of
laboratories and matrix heterogeneity caused by the het- where R is the mean relative range and d2 is the conversion
erogeneous distribution of microorganisms should be taken factor. For duplicate counts, d2 is taken as 1.128 [22].
into account for the selection of samples. Ideally, duplicate The mean relative range (R) for n duplicate determina-
analyses should be performed starting from the sub-sam- tions is calculated by [19]:
pling stage with naturally contaminated samples. P
Ri
Physiological conditions of microorganisms may vary R ¼
n
considerably for different samples depending on their
processing and storage conditions. Artificially contami- where Ri is the relative range.
nated samples may be insufficient to represent these Relative range (Ri) for per duplicate determination is
variations for certain types of food [5]. As well as opera- calculated by [19]:
tional variability, results of duplicate analyses starting from j Ai  Bi j
Ri ¼
sub-sampling stage represent unpredictable variation that is mi
due to distribution of microorganisms [17].
based on the results of two analysts (Ai and Bi) and their
According to the technical requirements of ISO/IEC
arithmetic mean (mi = (Ai ? Bi)/2).
17025, the precision of test results must be investigated by
quality control programs. Duplicate testing of a certain
number of samples is a part of internal quality program
performed by most of food testing laboratories. Therefore, Expanded uncertainty
the use of data obtained from an internal quality control
program offers a practical solution for laboratories to The intra-laboratory reproducibility standard deviation can
determine measurement uncertainty without the need for be accepted as the combined standard uncertainty (u) for
additional data. microbiological analyses according to global approach
principles. An interval expected to include a large fraction
of the distribution of values reasonably attributable to the
Standard deviation of intra-laboratory reproducibility measurand is covered by expanded uncertainty (U) which
is calculated as U = ku with an appropriate coverage factor
Analyses results may vary considerably for different sam- k [9]. Routinely, when 30 or more sets of duplicate anal-
ples depending on their contamination levels in food yses are performed, a coverage factor of 2 is used to give
microbiology. The use of data of duplicate analyses to distribution limits (confidence interval) with an approxi-
calculate reproducibility standard deviation over a wide mation of 95 % around the normally distributed mean
data range may give a large standard deviation value. To value [19].
overcome this difficulty, the current practice in food Suppose that the relative intra-laboratory reproducibility
microbiology is to transform the results logarithmically or standard deviation (RSDRw) is calculated as 30 %; then
classify the results with respect to their contamination expanded uncertainty (U) is obtained as 60 %. If testing
levels calculating the reproducibility standard deviation for result of a sample is 2000 cfu/g, then result with uncer-
each contamination level [5]. Alternatively, if relative tainty can be expressed as (2000 ± 1200) cfu/g in test
differences between results of duplicate analyses (Relative report.
ranges) are used instead of absolute differences, distribu- An example of the uncertainty calculations for enu-
tion of range of data can be decreased considerably. Thus, meration of Staphylococcus aureus in cheese by the
obtained data are approximately normally distributed, relative range-based protocol proposed in this paper and
which allow the use of the average of relative ranges based the ISO protocol are given in Table 1. In order to compare

123
600 Accred Qual Assur (2012) 17:597–601

Table 1 Calculations of measurement uncertainty by relative range-based and ISO protocols using results of duplicate analyses
Sample Results R Logarithmic results
A (cfu/g) B (cfu/g) log10{A} log10{B} (log10{A} - log10{B})2

1 210 280 0.29 2.32 2.45 0.016


2 3100 2200 0.34 3.49 3.34 0.022
3 18000 24000 0.29 4.26 4.38 0.016
4 960 1300 0.30 2.98 3.11 0.017
5 16000 13000 0.21 4.20 4.11 0.008
6 1600 2100 0.27 3.20 3.32 0.014
7 650 310 0.71 2.81 2.49 0.103
8 460 580 0.23 2.66 2.76 0.010
9 6700 8100 0.19 3.83 3.91 0.007
10 190 110 0.53 2.28 2.04 0.056
11 540 790 0.38 2.73 2.90 0.027
12 1200 870 0.32 3.08 2.94 0.020
13 1300 1600 0.21 3.11 3.20 0.008
14 540 350 0.43 2.73 2.54 0.035
15 6600 4400 0.40 3.82 3.64 0.031
Relative range-based protocol ISO protocol
P 
R 0.34 ðlog10 fAg  log10 fBgÞ2 =2n 0.013
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
P
R=d2 ¼ RSDRw 0.30 ðlog10 fAg  log10 fBgÞ2 =2n ¼ SDRw 0.114

U (%) 60.0 U (log10) 0.23


Data were kindly provided by Konya Food Control Laboratory, Turkey
R, SDRw denote the relative range and standard deviation of intra-laboratory reproducibility, respectively
{A}, {B} denote the numerical values of the results A and B, respectively

the confidence intervals of relative range-based protocol or higher than 100 % by relative range-based protocol for
and the ISO protocol, some examples based on the uncer- microbiological quantifications, especially for the ones
tainty values from Table 1 are given in Table 2. These requiring additional confirmation steps and these uncer-
examples show that using the relative range-based protocol tainty values produces unacceptable confidence intervals
instead of ISO protocol for uncertainty calculation in which overlap zero. Therefore, ISO protocol is more suit-
quantitative food microbiology does not make considerable able for microbiological analyses which can yield results
difference in the confidence intervals. However, differ- with the high variation, due to its asymmetrical upper and
ences between confidence intervals obtained by two lower limits of the confidence interval (e.g. -45 %,
protocol increase as the uncertainty values increase. ?70 %) in natural values. It can be proposed that the
Moreover, expanded uncertainties may be calculated equal microbiological data yielded uncertainty values up to 75 %

Table 2 Examples of confidence intervals obtained by relative range-based and ISO protocols
Results/(cfu/g) Lower/upper limits of confidence intervala
Relative range-based protocol (U = 60 %) ISO protocol (U(log10) = 0.23)
Linear Logarithmic cfu/g Logarithmic values cfu/g

200 2.30 80b 320c 2.07b 2.53c 120b 340c


2000 3.30 800 3200 3.07 3.53 1200 3400
20000 4.30 8000 32000 4.07 4.53 12000 34000
200000 5.30 80000 320000 5.07 5.53 120000 340000
a
Limit values in non-logarithmic form were rounded to two significant numbers
b
Lower limit
c
Upper limit

123
Accred Qual Assur (2012) 17:597–601 601

can be treated by relative range-based protocol. The ISO 8. EUROLAB Technical Report No. 1/2007 (2007) Measurement
protocol should be followed when the calculated uncer- uncertainty revisited: alternative approaches to uncertainty test
results and the uncertainty of evaluation. www.eurolab.org
tainty value is higher than 75 %. 9. EURACHEM (2000) Quantifying uncertainty in analytical mea-
surement. www.eurachem.org
10. BIPM, IEC, IFCC, ILAC, IUPAC, IUPAP, ISO, OIML (2008)
Conclusions Evaluation of measurement data—guide for the expression of
uncertainty in measurement. JCGM 100, http://www.bipm.org/en/
publications/guides/gum.html
The calculation protocol described in this paper is based on 11. Corry JEL, Jarvis B, Passmore SM, Hedges AJ (2007) A critical
relative differences between results of duplicate analyses; it review of measurement uncertainty in the enumeration of food
is therefore not necessary to have logarithmic results or to microorganisms. Food Microbiol 24:230–253
12. Barwick VJ, Ellison SLR (2000) Protocol for uncertainty evalu-
calculate the reproducibility standard deviations for dif- ation from validation data. LGC/VAM/1998/088
ferent contamination levels. This alternative protocol may 13. ISO 21748 (2010) Guidance for the use of repeatability, reproduc-
be useful especially for small scale food laboratories of ibility and trueness estimates in measurement uncertainty
which uncertainties for many microbiological analyses estimation. International Organization for Standardization, Geneva
14. BIPM, IEC, IFCC, ILAC, IUPAC, IUPAP, ISO, OIML (2008)
may be calculated as lower than 75 %, due to their rela- International vocabulary of metrology—basic and general con-
tively low operational variations. Thus, measurement cepts and associated terms (VIM) 3rd edn, JCGM 200;
uncertainty for results of these microbiological analyses http://www.bipm.org/vim
can be expressed the same as chemical test results 15. ISO 5725-3 (1994) Accuracy (trueness and precision) of mea-
surement methods and results—part 3: intermediate measures of
expressed in test reports. the precision of a standard measurement method. International
Organization for Standardization, Geneva
16. HPA QSOP 18 (2005) Recommended minimum internal quality
References control in food microbiology testing laboratories. National
Standard Methods, Health Protection Agency, London
1. ISO/IEC 17025 (2005) General requirements for the competence 17. ISO 29201 (2012) Water quality—the variability of enumeration
of testing and calibration laboratories. International Organization methods measurement of microbiological enumeration methods.
for Standardization, Geneva International Organization for Standardization, Geneva
2. Torlak E (2008) Measurement uncertainty in testing for antimi- 18. ISO 7218 (2007) Microbiology of food and animal feeding
crobial activity on textile materials. Accred Qual Assur 13:563–566 stuffs—general requirements and guidance for microbiological
3. HPA QSOP 4 (2005) Uncertainty of measurement in testing. examinations. International Organization for Standardization,
National Standard Methods. Health Protection Agency, London Geneva
4. IANZ AS TG5 (2004) Uncertainty of measurement, precision and 19. FDA Ora-Lab.5.9 (2003) Assuring quality of test results. Food
limits of detection in chemical and microbiological testing lab- and Drug Administration, Silver Spring
oratories. International Accreditation New Zealand, Auckland 20. ISO/TS 13530 (2009) Water quality—guidance on analytical
5. ISO/TS 19036/Amd 1 (2009) Microbiology of food and animal quality control for chemical and physicochemical water analysis.
feeding stuffs—guide on measurement of uncertainty for quantita- International Organization for Standardization, Geneva
tive determinations. International Organization for Standardization, 21. NORDTEST Report TR 569 (2011) Internal quality control
Geneva handbook for chemical laboratories. www.nordtest.info
6. EC Commission Regulation 2073/2005 (2005) Microbiological 22. ASTM MNL 7A (2002) Manual on presentation of data and
criteria for foodstuffs. European Commission, Brussels control chart analysis. American Society for Testing and Mate-
7. TFC Regulation (2011) Microbiological criteria for foodstuffs. rials, West Conshohocken
Turkish Food Codex, Ankara

123

You might also like