Subscriber access provided by Oregon State University
Article
Use of Combined Uncertainty of Pesticide Residue Results for Testing Compliance with MRLs Zsuzsa Farkas, Andrew Slate, Thomas B. Whitaker, Gabriella Kötelesné Suszter, and Árpád Ambrus J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/jf505512h • Publication Date (Web): 06 Feb 2015 Downloaded from http://pubs.acs.org on February 18, 2015
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Agricultural and Food Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 43
Journal of Agricultural and Food Chemistry
1
Use of Combined Uncertainty of Pesticide Residue Results for Testing Compliance
2
with MRLs.
3 4
Zsuzsa Farkas1, Andrew Slate2, Thomas B. Whitaker2, Gabriella Suszter3, Árpád Ambrus1a
5 6
1. National Food Chain Safety Office, Budapest Tábornok u 2, 1143 Hungary
7
2. North Carolina State University 124 Weaver Labs, Raleigh, NC 27695-7625, USA
8
3. Wessling Hungary Ltd, Budapest Fóti út 56, 1047 Hungary
9
1a Retired Scientific Adviser, ,
10 11
1 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
12
ABSTRACT
13 14
The uncertainty of pesticide residue levels in crops due to sampling, estimated for 106 individual crops
15
and 24 crop groups from residue data obtained from supervised trials was adjusted with a factor of 1.3 to
16
accommodate the larger variability of residues under normal field conditions. Further adjustment may be
17
necessary in case of mixed lots. The combined uncertainty of residue data including the contribution of
18
sampling is used for calculation of an action limit, which should not be exceeded when compliance with
19
maximum residue limits is certified as part of pre-marketing self control programs. On the contrary, for
20
testing compliance of marketed commodities the residues measured in composite samples should be ≥
21
the decision limit calculated only from the combined uncertainty of the laboratory phase of the residue
22
determination. The options of minimizing the combined uncertainty of measured residues are discussed.
23
The principles described are also applicable to other chemical contaminants.
24 25 26
KEY WORDS: sampling, uncertainty of measurement results, compliance with legal limits, pesticide
27
residues, self control of production, certification of compliance
28 29
2 ACS Paragon Plus Environment
Page 2 of 43
Page 3 of 43
30
Journal of Agricultural and Food Chemistry
INTRODUCTION
31 32
Distribution of pesticide residues.
33
The distribution of pesticide residues in/on treated crops is affected by many factors including, , for
34
instance, the type and spatial form of plants, cultivation and application method, and weather conditions
35
during application. The field to field variation of the magnitude of pesticide residues and their within
36
field concentration ranges can be expected to vary up to 10000 and 100 fold, respectively.1 Horváth and
37
co-workers have recently reviewed the characteristics of distribution of pesticide residues in crop units
38
or primary samples.2 Their statistical analyses of over 19000 residue data representing combinations of
39
20 crops and 46 pesticides confirmed previous assumptions3,4 that the residue sets obtained from the
40
analyses of crop units can be best described with lognormal distribution having a relative standard
41
deviation, CV, around 0.8. Even if 100-300 natural crop units are collected from one field, the CV of the
42
measured residues would provide only one estimate of the variability of residues. If the random
43
sampling operations are repeated many times the average of CV values approaches the true CV of the
44
parent populations. Consequently, the average of CV values obtained from the analyses of crop units
45
taken from different fields would provide the best estimate for the typical CV of residue values in the
46
given crop units.5 Due to the large variation of pesticide residues in crop units, the average residues in
47
composite samples containing 5-10 or larger number of primary samples6 will also substantially vary.
48
Uncertainty of measurement results.
49
The uncertainty of sampling, as one of the components of the combined uncertainty of the results of
50
pesticide residue analysis, has been estimated based on three independent databases: residues in 21000
51
primary samples5,7, residues in composite samples taken from commercially treated crops8, and residues
52
derived from 25876 samples taken from 12087 independent supervised residue trials.9
53
The contribution of sub-sampling, sample processing and homogenization to the variability of analytical
54
results has been studied by several authors.10-14 It was shown that the efficient comminuting of the
55
samples, especially those with hard peel and soft pulp, is a difficult task, and the result largely depends 3 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 4 of 43
56
on the crop material, equipment used, as well as the temperature and the length of processing. On the
57
other hand, it is worth to note that, if the residues are stable under the sample processing conditions, the
58
efficiency of sample processing does not depend on the nature of the chemical substances. In addition to
59
homogenization of the sample material, the sample size reduction15 may also contribute to the combined
60
uncertainty with a sub-sampling (CVSS) of 8-12%.
61
Maximum residue limits for chemical contaminants.
62
For assuring the safety of marketed food, regulatory agencies define maximum residue limits (MRL) for
63
pesticides16-19 and maximum limits (ML) for chemical contaminants.17,20,21 In addition, the Codex
64
Alimentarius Commission (CAC) establishes MRLs or MLs to facilitate international trade.22-24 The
65
MRLs and MLs are defined as the maximum legally permissible average concentration of chemical
66
contaminants in a composite sample with specified minimum mass and size in terms of the number of
67
primary samples (single sample increments). The sample sizes defined by the national authorities are
68
generally the same or very similar to those given in the relevant Codex Standards, such as for instance
69
for pesticide residues.5
70
The objective of this paper is to show that available data on the uncertainty (variability) of pesticide
71
residue measurement results may be used in the process of certification of compliance with MRLs. The
72
principles described are also applicable for other chemical contaminants.
73 74
METHODS
75 76
Estimation of combined uncertainty of the measured residues.
77
The combined relative standard uncertainty (expressed as relative standard deviation or coefficient of
78
variation) of the residues measured in composite samples (CVR) can be expressed as25
79
= + +
+
4 ACS Paragon Plus Environment
Equ. 1
Page 5 of 43
Journal of Agricultural and Food Chemistry
80
It incorporates the uncertainties of sampling (CVS), sub-sampling (CVSS), sample processing (CVSP)
81
(chopping, mincing and homogenization of analytical sample) and analysis (CVA). The methods for
82
estimation of sampling uncertainty have been described in detail in previous publications,7,8,26 therefore
83
they are not repeated here. Where the contribution of uncertainty of sub-sampling and sample
84
processing is included in the estimated uncertainty of the laboratory phase of measurement (CVL), the
85
two phases of the process can be distinguished as:
86
CV = CV + CV
Equ. 2
87
CV = CV + CV + CV
Equ. 3
88
It is to be noted that the CVL should always be determined either from retained test portions of the
89
laboratory samples which contain incurred residues,10 or after treatment of the surface of the individual
90
crop units with the test compound.14 Furthermore, the results of collaborative studies or proficiency tests
91
can only provide information on the uncertainty of the analytical procedure (CVA) from the point of
92
extraction and not for the combined uncertainty of the analytical results (CVL), as the appropriate
93
homogeneity of those samples had been carefully checked before distribution to the participants.
94
Principles of control of compliance of food commodities with the legal limits.
95
As the consequence of the definition of MRLs, there are two distinctly different situations in the control
96
of compliance of food commodities with the legal limits. Different sampling plans are required for
97
testing compliance with MRLs of pesticide residues in commodities before and after marketing..
98
Control of commodities on the market (enforcement or monitoring program).
99
As the permissible maximum concentrations of residues and chemical contaminants apply to the average
100
concentration of the substances in the bulk/laboratory sample, the combined uncertainty of the
101
laboratory phase of the determination of the residue concentration (CVL) shall be taken into account in
102
deciding on the compliance or non-compliance of the sampled product. If the samples taken satisfy the
103
minimum requirements for the number of primary samples and the mass of the laboratory/bulk sample,6
104
the average residue determined in the sample provides the basis for deciding on the acceptability of the 5 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
105
sampled lot. In order to make regulatory action in accordance with the relevant ISO and Codex
106
guidelines,27,28 among other regulatory agencies, the European Commission (EC) regulations clearly
107
indicate that a lot is considered non-compliant if the measured analyte concentration corrected for
108
recovery, where it should specifically be considered, minus two times the expanded uncertainty of the
109
results are above the legal limit. For pesticide residues the default combined relative uncertainty
110
reflecting the CVL value (equ. 3) is defined as 25%.29 This means that a lot could be rejected if the
111
residue measured in the composite sample taken from the lot would be >2×MRL. We call this
112
concentration the Decision Limit (DL). Figure 1 illustrates the situation where the MRL is 0.2 mg/kg
113
and the apple lot (mean of 0.1 mg/kg and CVR of 0.35) would be tested on the market applying the
114
decision limit of 0.4 mg/kg (0.4-2×0.25×0.4=0.2 mg/kg) as defined by the EC. Under this condition a
115
testing laboratory would reject a lot containing residue ≤0.2 mg/kg in 2.3% of the cases, because the
116
dispersion of analytical measurement results can be considered normal.30
117
Premarketing self-control.
118
When the product is tested before placing it on the market, it should be certified that at least a specified
119
proportion of the product in terms of the minimum size and mass of bulk/laboratory sample complies
120
with the legal limit. In this case the combined uncertainty (CVR) including sampling uncertainty shall be
121
taken into account, and the measured value should not be directly compared to the MRL as it may allow
122
placing a product on the market with a substantial proportion containing the contaminant above the
123
permitted limit. Let’s assume that the MRL is 0.1 mg/kg, and the average concentration of a pesticide
124
residue is also 0.1 mg/kg in the sampled lot. If the measured residue in the sample taken from the given
125
lot is < 0.1 mg/kg the sampled lot would be declared to be compliant. However the decision would only
126
be correct in about 56% and would be wrong in 44% of the cases as indicated by the cumulative
127
frequency distribution curve in Figure 2. Therefore a lower concentration must be selected as Action
128
Limit (AL) which should not be exceeded by the residue measured in the sample. The AL can be
129
calculated by taking into account the combined relative uncertainty of the measured value (CVR) (equ. 1)
130
and the targeted level of compliance. 6 ACS Paragon Plus Environment
Page 6 of 43
Page 7 of 43
131 132
Journal of Agricultural and Food Chemistry
The relationship of AL and MRL can be described with the following equations: AL + k × CV × AL = MRL AL =
133
Equ. 4 Equ. 5
134
The value of k depends on the agreed acceptable violation rate (Pv) or the compliant proportion of the
135
lot, Pc, relative to the minimum sample size specified in the corresponding regulation (1-Pv=Pc). For
136
normal distribution, which may approximately be assumed with samples of size ≥25 for fruits and
137
vegetables, the k is equal to the corresponding standard normal variate, Z. A normal distribution may
138
also be assumed for liquid materials and processed products, such as for instance tomato puree. It is
139
pointed out that in other cases the relative frequency distribution of pesticide residues can be best
140
described with a lognormal distribution.1
141
Calculation of AL for given acceptance probability at MRL.
142
If the distribution of xi can be described by the lognormal function, then the transformed variable yi =
143
Ln(xi) is normally distributed. If the distribution of xi has a mean (m) and standard deviation (s), then the
144
distribution of the transformed variable yi has a mean (M) and standard deviation (S). M and S can be
145
calculated from m and s using the equations: #$
146
M = Ln "
(
Equ. 6
147
) = *+ ,.$ + 10
Equ. 7
%&$ #$ ' -$
148
Since yi is normally distributed, the probability Pi (0.0 to 1.0) for any value yi occurring from a
149
distribution with mean M and standard deviation S can be calculated from normal distribution tables
150
using the standardized equation:
151 152
1=
%23 45'
67 = 8 + 1 × )
Equ. 8 Equ. 9
153
In the sampling domain, m takes on the value of the legal limit MRL and xi takes on the value of the AL.
154
Typically xi ≤ m (AL≤MRL), therefore Z will vary from zero to a negative value usually between 0.0
155
and -3.5 based upon the specified acceptance probability (acceptable violation rate), Pv. Once m and s 7 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 8 of 43
156
are transformed into M and S and the acceptance probability (that will define Z) is specified, yi can be
157
determined from Equation 9. Then xi can be calculated by transforming yi back to xi. 97 = exp %67 '
158 159
Equ. 10
where xi is the AL in the linear scale.
160 161
RESULTS AND DISCUSSION
162 163
Comparison of sampling uncertainties estimated from supervised trials and commercial field
164
samples.
165
Based on the limited data available, the average sampling uncertainty for primary samples, CVSprim of
166
0.78, 0.81, 0.74 and 0.45, were estimated for residues in crop units of a small, medium and large size
167
crops and Brassica leafy vegetables, respectively.7 Additional residue data measured in carrot and
168
parsley enabled the estimation of CVSprim values of 0.59 and 0.60, respectively.8 Duplicate samples taken
169
from 12087 supervised trials carried out in 106 different crops treated with 66 different pesticides
170
provided the best available database for estimation of uncertainty of sampling.31 The supervised trials
171
are carried out usually on small scale with strictly controlled conditions for the application of pesticides.
172
They aim to provide information on the maximum residues which can occur on treated crops under the
173
critical use conditions [shortest interval between last application and harvest (pre-harvest interval),
174
highest permitted dose and application frequency] specified in the registration or use permit
175
documents.32 Consequently, these trials may not reflect the within field residue distribution which occurs
176
where the pesticides are applied according to the general agricultural practice. Table 1 summarizes the
177
estimated sampling uncertainties obtained from the databases consisting of residues in/on primary
178
samples, which were taken from fields treated according to normal farming practice and from supervised
179
trials. The CVS values were calculated for the minimum size of composite samples (12) specified by the
180
Codex Standard for sampling plant products.5 In table 2 the sampling uncertainties calculated from
181
residues in samples taken from commercial fields and from supervised trials are compared. The relative 8 ACS Paragon Plus Environment
Page 9 of 43
Journal of Agricultural and Food Chemistry
182
sampling uncertainty of residues in field composite samples (CVcs) was calculated from 4 replicate
183
samples and for supervised trials mostly from duplicate samples with range statistics.8 The same
184
tendency was observed in both cases. As the number of datasets from which the CV values were
185
calculated greatly varies and the sample sizes in supervised trials had only sometimes been reported,
186
precise comparison of the variability of residues in supervised trial and field samples could not be done.
187
Therefore the averages of CV ratios (1.15 and 1.19) are considered to be the best indicator of the
188
difference in the variability of residues. Exceptions are papaya, strawberry, cabbage, carrot and parsley
189
for which the ratios of relative sampling uncertainty of composite samples taken from supervised trials
190
and relative sampling uncertainty of residues in field composite samples (CVcs/CVStcs) are within the
191
range of 0.44 and 0.70. The low variability in field trials cannot be explained based on the available
192
information. As it is quite unlikely that supervised trials lead to higher variability of residues than field
193
trials, these values were disregarded.
194
Taking into account the uncertainties of the estimations a rounded correction factor of 1.2 is used to
195
account for the larger variability of residues in field samples.
196
Effect of number of residue values below the LOQ on the estimated sampling uncertainty.
197
The range of residues in primary samples taken from a single field is typically hundred fold. Depending
198
on the average residue and the limit of quantification (LOQ) of the analytical method various
199
proportions of the primary samples may not contain detectable residues. The proportion of residues
200
below LOQ is not known. The effect of the LOQ values in relation to the average of residue data
201
population was studied with experimental supervised trial data1 and synthetic lognormal distributions, as
202
it was found that the within field distribution of residues in crop units (primary samples) can be best
203
described with lognormal distribution.2 Of the supervised trial residue datasets, compiled by the FAO
204
WHO Joint Meeting on Pesticide Residues for estimation of MRLs, those were selected which contained
205
relatively large number of data points. The descriptive statistical parameters of the selected datasets are
206
summarized in Table 3.The selected datasets contained no residue below the LOQ value, except dataset
207
8 where the 27 values below the varying LOQs were replaced with the maximum likelihood estimation 9 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 10 of 43
208
(MLE) procedure assuming lognormal distribution provided by Villanueva.33 The residue values were
209
also normalized by dividing them with the corresponding average residues of their dataset in order to
210
assist comparison of the spread of residues in the datasets with different mean residues.
211
For studying the effect of LOQ [mg/kg] values on the CV of the data population, the selected data sets
212
were arranged in increasing order. The lowest number of residue values corresponding to the selected
213
percentage of the original data population (10%-90%) were replaced with various LOQ values ranging
214
from 2× reported LOQ to 0.7 average residues in the dataset. These datasets are called adjusted datasets.
215
For instance, in case of dataset 1 the lowest 11 values (0.03, 0.06, 0.07, 0.09, 0.09, 0.1, 0.1, 0.11,
216
0.12,0.12, 0.12) were replaced with 0.04, 0.2, ...1.16) and then the descriptive statistical parameters were
217
recalculated for each adjusted data set. The process was repeated with 20% of the residue values in the
218
original data set, and so on.
219 220
The percentage of residue values below the LOQ and the value of LOQ affect the average and the
221
relative standard deviation of the original residue population (CVo). The relationships of the
222
corresponding CV and average residue of the adjusted datasets are complex, and they are best illustrated
223
graphically. Figure 3 shows the ratios of the CV-s of adjusted dataset (CVAdj) and the original dataset
224
(CVo) as the function of the ratio of LOQ and the average residue (=>' of adjusted dataset (LOQ/=> '.
225
Figure 4 illustrates the CVAdj/CVo relationship as the function of the LOQ/=> residues and the percentage
226
of residues at or below the LOQ obtained from the dataset 8 (Table 3. MLE fitted prochloraz residues).
227
As similar LOQ/=> residues could be obtained from various combinations of LOQ values (Figure 4) and
228
their percentage proportion in the adjusted datasets, the CVAdj values corresponding to LOQ/=> ranges
229
from ≤0.03 to ≤0.7 were grouped and the averages of LOQ/=> and the CVAdj/CVo values being in the
230
given ranges were calculated. The results obtained from the supervised trials and synthetic datasets are
231
shown in Table 4. The results indicate certain variation among datasets which is attributed to the
232
distribution of residues within the individual datasets and sampling uncertainty. As the typical sampling
233
uncertainties were obtained from datasets with detectable residues, they do not represent the situations 10 ACS Paragon Plus Environment
Page 11 of 43
Journal of Agricultural and Food Chemistry
234
where the various proportions of the treated crops may contain residues below the LOQ. The probability
235
of finding primary samples with non-detectable residues increases as the LOQ/=> increases. However,
236
no general tendency in the relationship between the LOQ/=> and CVAdj/CVo could be established. The
237
grand averages of CVAdj/CVo up to LOQ/=>