You are on page 1of 12

Food Anal.

Methods
DOI 10.1007/s12161-016-0432-7

Interlaboratory Validation of Modified Classical Qualitative


Methods for Detection of Adulterants in Milk: Starch,
Chloride, and Sucrose
Carina de Souza Gondim 1 & Roberto Gonçalves Junqueira 1 &
Scheilla Vitorino Carvalho de Souza 1

Received: 2 November 2015 / Accepted: 1 February 2016


# Springer Science+Business Media New York 2016

Abstract Interlaboratory validation procedures were pro- Introduction


posed and performed to confirm the effectiveness of mod-
ified classical qualitative methods for the detection of adul- Due to the growing demand of milk and the price changes that
terants in milk, including starch, chloride, and sucrose, occur because of seasonal fluctuations, milk is often adulter-
which were previously validated by a single laboratory ated to produce it in bulk quantities to meet the consumer
approach. Raw milk samples that were adulterated with demand, to reduce the cost of the final product, and to increase
150 g L−1 of water and 0.0, 0.3, 0.8, and 1.2 g L−1 of profit (Almeida et al. 2012; Rani et al. 2012). Starch, sodium
starch, 0, 1.5, 2.0, and 2.5 g L−1 of chlorides, and 0.0, chloride, and sucrose have been illegally used as intentional
2.4, 3.0, and 3.6 g L−1 of sucrose were sent to 10 collabo- adulterants in milk to restore the density to established limits
rators in Brazil that represent the government, food con- and to mask another type of fraud—the addition of water
trol, food industry, and university affiliations. Reliability (Tronco 2010; Kartheek et al. 2011; Souza et al. 2011).
rates of 93 to 100, 98 to 100, and 99 to 100 % were ob- Different techniques and analytical procedures have been
tained for the starch, chlorides, and sucrose methods, re- reported for the detection of adulterants in milk (Kartheek
spectively. The prediction intervals for the probability of et al. 2011; Ellis et al. 2012). Sensorial techniques, in which
detection proved the sensitivity and selectivity of the human senses are used to record and interpret the response
methods. Concordance values were greater than 0.85 to (Trullols et al. 2004), are represented by methods that involve
starch, 0.98 to chlorides, and 0.99 to sucrose, indicating immunological responses (Renny et al. 2005) and chemical
precision and that the procedures were properly standard- reactions (India 2005; Brasil 2006; Tronco 2010; Souza et al.
ized between the collaborators. The estimated detection 2011). The instrumental techniques have included conven-
limits and unreliability regions confirmed the fitness of tional and alternative methods, such as potentiometric
the modified methods for their respective purposes. (Veloso et al. 2002), chromatographic (Rani et al. 2012), spec-
troscopic (Almeida et al. 2012; Jawaid et al. 2014; Botelho
et al. 2015; Liu et al. 2015), amperometric (Silva et al. 2012),
Keywords Collaborative studies . Classical methods . Milk conductometric, and spectrophotometric (Lima et al. 2004)
adulteration . Thickeners . Qualitative methods impedance sensors (Das et al. 2011) and digital image analysis
(Santos et al. 2012; Santos and Pereira-Filho 2013).
In this context, the advanced association of chemometric
methods with instrumental techniques can be highlighted for
* Scheilla Vitorino Carvalho de Souza food fraud detection (Almeida et al. 2012; Ellis et al. 2012;
scheilla@bromatologiaufmg.com.br Santos et al. 2012; Santos and Pereira-Filho 2013; Jawaid
et al. 2014; Botelho et al. 2015; Scampicchio et al. 2015).
1
These methodologies have advantages as the ability to analyze
Faculty of Pharmacy (FAFAR), Department of Food Science, Federal
University of Minas Gerais (UFMG), Av. Antônio Carlos, 6627,
samples with little or no preparation, ease of use, fast data
Campus da UFMG, Pampulha, 31270-010 Belo Horizonte, collection, and the capability to be used as a fingerprint tech-
MG, Brazil nique (Karoui and De Baerdemaeker 2007; Santos and
Food Anal. Methods

Pereira-Filho 2013; López et al. 2014). However, performance Fig. 1 Proposed scheme for the validation of qualitative methods by„
studies have shown low sensitivity and detection limits upper collaborative study: main steps, minimal design, acceptability criteria,
and estimated performance parameters
to the classical methods (Botelho et al. 2015; Gondim et al.
2015; Liu et al. 2015).
To monitor the authenticity of milk, laboratories re- Test Materials
quire economical, reliable, and reproducible techniques
(Rani et al. 2012). Qualitative methods have been ap- Samples
plied for the analysis of milk and their products, and
these methods have been used by dairy industries to ver- Three raw cow milk samples produced by a herd of 50
ify conformity to quality requirements, by authorities for animals were taken from a refrigerated (4 to 7 °C) stor-
sanitary purposes and by researchers (Brasil 2011; Souza age tank on the experimental farm of Professor Hélio
et al. 2011). Barbosa of the Veterinary School/Federal University of
Although recent reports have presented procedures for the Minas Gerais State (EV/UFMG). These samples of 15,
validation of these qualitative methods, a harmonized valida- 15, and 22 L were collected for the interlaboratory vali-
tion protocol is still required (Langton et al. 2002; Ellison and dation of the starch, chlorides, and sucrose methods, re-
Fearn 2005; Macarthur and Holst 2012; Gondim et al. 2014). spectively. The samples were kept refrigerated (4 to 7 °C)
This is evidenced in studies involving qualitative methods for in polypropylene gallons from transportation until the
detection of adulterants in milk. These studies are rare, are time of preparation of the test materials, which occurred
limited regarding the parameters established in the procedures immediately after the arrival of the samples to the
mentioned above, and are restricted to in-house processes laboratory.
(Silva 2013; Botelho et al. 2015).
Regardless of the method and the analytical technique, a Preparation
fully validated process, which includes in-house and collabo-
rative studies, is necessary both to assess or verify the quality Reliability is a property of qualitative methods related to
of the work done by a laboratory and to provide reliable data trueness and precision properties in quantitative analysis
which can support the industry and regulation agencies (Ginn and can be defined as the percentage of right results
et al. 2006; López et al. 2015) and generate improved infor- obtained in independent tests performed to sample clas-
mation on milk fraud. sification (Cardenas and Valcarcel 2005). Reliability rates
The methods reported here have been previously vali- (RLRs) obtained for the single laboratory validation pro-
dated using a single laboratory process (Gondim et al. cess of the methods (Gondim et al. 2015) were used to
2015), indicating their suitability for interlaboratory study. establish the concentration of the adulterants in the
In this study, these modified classical qualitative methods samples.
for the detection of starch, sodium chloride, and sucrose Each sample was split into four batches that were spiked
were validated using collaborative trials. Also, all aspects with ultrapure water and a thickener aqueous solution to pro-
of the organization of a collaborative study for the vali- duce a target concentration of 150 g L−1 for the water and four
dation of qualitative methods are presented in detail, in- target concentrations for the restoratives which corresponded
cluding experimental design, statistics, and criteria for ho- to the upper and lower concentrations at which 100 % RLRs
mogeneity and stability tests and method performance were obtained in the previous validation process in addition to
assessment. the intermediate concentration and zero, as shown in Table 1.
For the starch method, an aliquot of 280 mL of the raw milk
sample was packaged into 50-mL conical polypropylene tubes
Material and Methods that were each filled with 20 mL using a macropipette. These
unadulterated samples were reserved as non-blind unspiked
An interlaboratory validation process consists in many samples for distribution to collaborators.
steps beyond the analysis of the test materials by collab- The spiking solutions were prepared with analytical grade
orating laboratories. It covers the preparation of test ma- starch, sodium chloride (Vetec Química Fina Ltd., Rio de
terials, the evaluation of homogeneity and stability, the Janeiro, RJ, Brazil) and sucrose (Synth, Diadema, SP,
distribution of test materials, besides the data analysis, Brazil) to final concentrations of 15, 100, and 140 g L−1, re-
and, finally, the report of results (AOAC 2002; Langton spectively. Each batch was continuously mixed in a mechan-
et al. 2002; Ellison and Fearn 2005; Thompson et al. ical stirrer (Fisatom 713 T) at 22.6×g during the test material
2006; Macarthur and Holst 2012). In Fig. 1, a scheme of preparation. The milk was transferred using measuring cylin-
a minimum design for the validation of qualitative ders (1000 and 2000 mL) to a 10-L bowl and homogenized for
methods by collaborative study is presented. 10 min. Ultrapure water was added to the milk using a
Food Anal. Methods

Step 1 Minimum design parameters:


Sampling and test materials Laboratory collaborators = 10;
Representative samples for the test materials formulation:
preparation composed samples (for example for processed foods, samples
obtained from different brands and batches);
Test materials = 4, covering 4 concentration levels,
corresponding to the lower, an intermediate and the upper
levels at which 100 % of reliability rate were obtained in
previous single laboratory validation, plus the blank;
Replicates of blank - negative materials = 180 (10 for each
laboratory collaborator, 10 for homogeneity test, 30 for stability
tests and 40 for leftovers);
Replicates of positive materials = 110 to 250 (5 to 15 for each
laboratory collaborator, 10 for homogeneity test, 30 for stability
tests and 20 to 60 for leftovers).

Step 2
Homogeneity tests Evaluation of the variation between-bottle of test materials:
Analysis in duplicate of 10 test materials replicates of each
concentration, randomly taken, by qualitative method previously
validated, under repeatability conditions.
Acceptability criterion = reliability rate ≥ 90 %.
Note: if a quantitative method is available, proceed as Thompson,
Ellison and Wood (2006). The within-bottle homogeneity will be
assured for the minimal intake of the analytical aliquot.

Stability of test materials stored at critical transportation


Step 3 conditions:
Stability tests – isochronus desing Analysis in duplicate, by qualitative method previously validated,
under repeatability conditions, of 30 test materials replicates of
each concentration, randomly taken and kept at 3 conditions (10
at critical temperature during the minimal time for transportation ,
10 at critical temperature during the maximum time for
transportation and 10 under the ideal condition),
Acceptability criterion = reliability rate ≥ 90 %.
Note: if a quantitative method is available, proceed as
Thompson, Ellison and Wood (2006).

Step 4 Stability of test materials during the time of the collaborative


Stability tests – classical desing trial:
Analysis in duplicate, by qualitative method previously validated,
under repeatability conditions, of 10 test materials replicates of
each concentration, randomly taken, on deadline for analysis,
Acceptability criterion = reliability rate ≥ 90 %.
Note: if a quantitative method is available, proceed as Thompson,
Ellison and Wood (2006).

Step 5 For each laboratory:


Distribution of test materials Number of test materials = 10 negative test materials replicates
and from 5 to 15 positive test materials replicates of each
concentration;
Documents = analytical procedure, materials receipt and report
forms;
General instructions.

Step 6 Data analysis:


Outliers analysis, assessment of false negative rates, false
Data analysis and reporting results positive rates, RLR, accordance, concordance, probabilities of
detection and detection limits.
Publication and distribution of the report results with a certificate
of participation, preserving the identity of the participants
concerning the results.

Fitness for purpose


Food Anal. Methods

Table 1 Preparation of the test


materials Thickener final Volume (mL) Number of Volume per
concentration tubes filled tube (mL)
(g L−1) Raw milk Water Thickener Total
solution

Starch
0.0 3570 630 0 4200 190 20
0.3 3570 546 84 4200 190 20
0.8 3570 406 224 4200 190 20
1.2 3570 294 336 4200 190 20
Chlorides
0.0 4675 825 0 5500 260 20
1.5 3570 567 63 4200 190 20
2.0 3570 546 84 4200 190 20
2.5 2550 375 75 3000 120 20
Sucrose
0.0 5355 945 0 6300 190 30
2.4 5355 837 108 6300 190 30
3.0 5355 810 135 6300 190 30
3.6 5355 783 162 6300 190 30

Thickener aqueous solution, starch 15 g L−1 , sodium chloride 100 g L−1 , and sucrose 140 g L−1 . Final concen-
tration of water 150 g L−1 . The chloride concentrations were estimated based on the native content of the raw milk
sample, which was determined in triplicate using the quantitative titrimetric method (Brasil 2011). The quantities
of milk, water, and the aqueous solutions of the restoratives were adjusted to obtain exact amounts for the
preparation of the test materials; however, this adjustment generated additional material that was not bottled

measuring cylinder (1000 mL), followed by homogenization Stability Tests


for 10 min. The aqueous spiking solutions of the thickeners
were incorporated and homogenized for 10 min. The short-term stability was verified for each batch of mate-
Each adulterated batch was packaged into 50-mL conical rials using a qualitative approach and considering two designs.
polypropylene tubes that were filled with 20 mL for starch and First, we used an isochronous design in which failures in
chlorides and with 30 mL for sucrose, which were measured maintaining the temperature conditions during the transport of
using a micropipette. The tubes were closed with screw caps, the test materials were simulated. Thirty replicates of the test
sealed with Parafilm® and labelled with code numbers that material were taken randomly from each adulteration level.
were randomly generated. All of the test materials were refrig- Ten were refrigerated at 4 to 7 °C, and 20 were incubated at
erated at 4 to 7 °C. The number of filled tubes was sufficient to 30 to 40 °C. After 5 h, 10 replicates were returned to the
carry out the homogeneity and stability tests and to conduct refrigeration temperature. After 10 h, another 10 materials
the collaborative trials, as well as leftovers, in case of need for were removed from the 30 to 40 °C temperature condition
replacement materials. (Fig. 2). All of the 30 materials were analyzed in duplicate
under repeatability conditions. The RLR values were estimat-
ed for the three treatments and compared, considering the
Homogeneity Tests refrigeration condition as a reference. Second, we used a clas-
sical design in which the stability was investigated during the
The homogeneity was assessed for each batch of materials time between the preparation and the deadline for the analysis
using both qualitative and quantitative approaches under re- of the test materials in the collaborative trial. Ten replicates of
peatability conditions. For each interlaboratory trial, 20 test the test materials of each adulteration concentration that
material replicates of each adulteration concentration were were refrigerated at 4 to 7 °C were taken randomly on
randomly taken (Fig. 2). Ten of them were analyzed in dupli- the deadline (48 h after milking) for analysis by the collab-
cate using the modified qualitative method that was previously orators (Fig. 2). These materials were analyzed in duplicate
validated. An RLR greater or equal to 90 % confirmed the under repeatability conditions. The RLR was estimated and
homogeneity of each batch of materials. This criterion was compared with that obtained in the homogeneity assess-
based on the possibility of obtaining a false result for each ment (zero time). The stability was considered when the
adulteration level (Gondim et al. 2014). RLR was greater or equal to 90 %.
Food Anal. Methods

Fig. 2 Organization of interlaboratory trials, including homogeneity and stability tests, for validation of the modified classical qualitative methods for
the detection of adulterants in milk: starch, chlorides, and sucrose

Organization of Interlaboratory Studies all replicates within the deadline, under repeatability condi-
tions (analyze the samples in a single analytical batch); (iii)
Ten laboratories that represented government (3), food control strictly follow the given analytical procedure; (iv) report to the
(3), food industry (3), and university affiliations (1) took part organizer any abnormality or loss of the test materials so that a
in the interlaboratory studies. new shipment of materials could be sent to replace them to
Test materials were taken randomly from each batch of ma- adhere to the deadline; and (v) report the results to the orga-
terials and kept refrigerated during transportation to the collab- nizer within 2 days after the analysis.
orators. The materials were delivered within a period that en-
abled the analysis to be conducted within 48 h of milking.
For each trial, the collaborators received 40 blind and coded Statistical Analysis
test materials for a total of 400 results. For the starch method, 10
test material replicates of each adulteration concentration, 0.0, The raw data received from the collaborators were evaluated
0.3, 0.8, and 1.2 g L−1, were provided. For the chloride method, for gross errors of transcription or typing (Ellison and Fearn
we provided 10 replicates for the concentrations of 1.5 and 2005). Once the principal aim of the study was to validate the
2.0 g L−1, 15 replicates of 2.5 g L−1, and 5 replicates of qualitative methods, the raw data were used to analyze possi-
0.0 g L−1. For the sucrose method, 10 replicates were provided ble discrepancies between the classes of laboratories repre-
for each adulteration concentration, 0.0, 2.4, 3.0, and 3.6 g L−1 sented in this study by a hierarchical cluster analysis (HCA)
(Fig. 2). with the criterion of Ward’s method (Ward 1963).
In addition to the test materials, each laboratory received the In qualitative method validation, some parameters, such as
following: (i) a non-blind unspiked sample—one replicate of trueness, precision, and measurement uncertainty, cannot be
an unadulterated sample as a reference for reporting the results applied as currently defined for quantitative measurement
(only for the starch method); (ii) a copy of the method (analyt- methods (Ellison and Fearn 2005). Trueness can be defined
ical procedure); (iii) a report form for the analytical data (anal- as a measure of how well a binary classification test correctly
ysis date, laboratory identification, brand and lot of reagents, identifies or excludes a condition represented by the propor-
manufacturer and model of equipment, calibration certificates tions of true and false positive results, i.e., the false positive
of the measuring instruments such as glassware used in the rate (FPR) and the false negative rate (FNR), and by RLR, as
preparation of solutions), results (positive or negative), obser- previously defined (Ellison and Fearn 2005; NATA 2013;
vations (observed color or reference to index color code), com- Gondim et al. 2014).
ments, and suggestions; and (iv) a material receipt form. The interlaboratory validation is considered as a complete
At the time of receiving, the following parameters were validation procedure because the two extreme conditions of
evaluated: (i) temperature of the test materials; (ii) integrity, precision can be evaluated—repeatability and reproducibility
sealing and labeling (including the code) of the tubes; (iii) the (Van der Voet et al. 1999). In the context of qualitative anal-
presence of the unspiked sample (only for the starch method); ysis, Langton et al. (2002) proposed the statistics accordance
and (iv) the documents (report form and analytical procedure). (ACC) and concordance (CON) to assume the role of repeat-
The collaborators were instructed to: (i) keep samples re- ability and reproducibility, respectively, for qualitative analy-
frigerated until the time of analysis; (ii) conduct the analysis of sis. These two statistics indicate whether the particular
Food Anal. Methods

procedure used is sufficiently standardized (Langton et al. chromate with a yellow color, indicating a positive result
2002; Ellison and Fearn 2005; Gondim et al. 2014). (Brasil 2006).
Thus, the results reported by each laboratory were assessed
based on the characteristic parameters of the qualitative Analytical Procedure Allow the test material to warm to
methods: FPR, FNR, RLR, ACC, and CON. The rates that room temperature. Measure 10 mL and transfer it to a test tube
were estimated were designated as satisfactory if the RLR was with a volumetric pipette. Add 4.5 mL of silver nitrate solution
greater than or equal to 90 %. The ACC and CON were de- (0.1 mol L−1) with a graduated pipette. Shake manually with
termined to be acceptable when the estimated values were 30 pendular moves. Add, with a graduated pipette, 0.5 mL of
greater than or equal to 0.80 (for the materials that were dis- potassium chromate solution (50 g L−1). Shake manually
tributed in at least 10 replicates) or 0.60 (for materials that again with 30 pendular moves. Report a positive result when
were distributed in 5 replicates). These criteria were based the color of the test material is yellow and a negative result
on the possibility of obtaining a false result for each adultera- when the color is brick-red or intermediate.
tion concentration (Gondim et al. 2014). Note 1: Index color references for reporting the results: (i)
Additionally, the results were used to estimate the detection negative result, brick-red color (color index 200r 80g 60b) and
limits (DLs) and the prediction intervals for the probability of intermediate colors (color indexes 250r 168g 95b and 254r
detection when the methods were applied in a new laboratory 207g 140b); or (ii) positive result, yellow color (color index
at each concentration, according to Gondim et al. (2014) and 255r 246g 133b) (Brasil 2006; Tronco 2010; Gondim et al.
Macarthur and Holst (2012), respectively. 2015).

Methods Sucrose

Starch Principle The detection of sucrose was indicated by the de-


velopment of an orange-to-deep-red color due to the conden-
The milk sample heating results in the opening of the starch sation of ketoses with resorcinol in an acid medium (India
helical structure. Then, the iodine/potassium iodide solution 2005; Nigam and Ayyagari 2007).
(Lugol’s solution) is adsorbed, producing a blue characteristic
color upon cooling (Brasil 2006). Analytical Procedure Allow the test material to warm to
room temperature. Measure 15 mL with a graduated pipette
and transfer it to a test tube. With a volumetric pipette, add
Analytical Procedure Allow the test materials to warm to 1 mL of concentrated hydrochloric acid. Allow to stand for
room temperature. Measure 10 mL of test material and trans- 10 min. Filter through qualitative filter paper. Take 1 mL of the
fer it to a test tube using a volumetric pipette. Heat in a boiling filtrate and transfer to another test tube with a volumetric
water bath for 5 min. Cool under running water. Add two pipette. Add 5 mL of resorcinol solution (10 g L−1 in hydro-
drops of Lugol’s solution with a dropper and shake manually. chloric acid solution—1:1.5, v/v) with a volumetric pipette.
Follow the same procedure for the non-blind unspiked sam- Heat in a boiling water bath for 5 min. Report a positive result
ple. Immediately, observe the formed color. Report a positive when the color of the test material changes from orange to
result when the color of the test material is different from the deep-red and a negative result when an intermediate color or
one obtained for the non-blind unspiked sample (grayish a colorless solution is observed.
green to blue) and a negative result when it is similar (yellow Note 1: Index color references for reporting the results:
color). (i) negative result, colorless liquid and intermediate color
Note 1: If the readings of the test materials take more than (color index 142r 107g 75b); or (ii) positive result, inter-
10 min after the addition of Lugol’s solution to the non-blind mediate color (color index 117r 73g 48b) and deep-red
unspiked sample, repeat the addition to the control sample. colors (color index 95r 46g 32b and 100r 35g 17b)
Note 2: Index color references for reporting the following (Gondim et al. 2015).
results: (i) negative result, yellow color (color index 255r 246g
133b); or (ii) positive result, grayish green color (color index
208r 197g 151b or 230r 217g 165b) and blue color (color Results and Discussion
index 107r 120g 164b) (Gondim et al. 2015).
Homogeneity and Stability Tests
Sodium Chloride
The homogeneity was observed for all test materials that were
Principle The detection of sodium chloride was based on its prepared for the validation of the starch, chlorides, and sucrose
reaction with silver nitrate in the presence of potassium methods. For the adulterated materials without the thickeners,
Food Anal. Methods

negative results of 100 % (or 100 % of RLR and 0 % of FPR) Interlaboratory Study Results
were obtained. For the materials containing the thickeners,
positive results of 100 % (or 100 % of RLR and 0 % of Starch
FNR) were achieved. According to Ellison and Fearn
(2005), the inhomogeneity is difficult to detect when a quali- Considering each collaborator separately, for the adulterated
tative method is used. However, for positive samples, if the materials without the thickeners, the FPR ranged from 0 to
lack of homogeneity does not significantly influence the re- 10 %, resulting in a RLR from 100 to 90 %. Three false
sults, the homogeneity can be assumed. positive results were observed by laboratories 1, 2, and 7.
The short-term stability of the test materials was demon- Similar results of the RLR were obtained for materials with
strated. For the isochronous design, the results indicated that 0.3 g L−1 of starch for laboratories 2 and 7, in which each
failures in maintaining the temperature conditions during the reported one negative result. These results implicated in
transport did not affect the results. RLRs of 100 % were ob- FNR and RLR of 10 and 90 %, respectively. However, for
served for all test materials submitted to the three treatments laboratory 1, the estimated RLR was 40 %. A FNR and a RLR
with refrigeration and incubation at 30 to 40 °C and for 5 and of 0 and 100 %, respectively, were observed for the materials
10 h. containing 0.8 and 1.2 g L−1 of starch by all laboratories.
The same results were achieved for the classical design Considering all of the data, the RLR ranged from 93 to
because a 100 % RLR was estimated for all materials that 100 % for the adulterated materials that were also spiked with
were analyzed within the deadline established for the analysis thickeners, and the RLR was 97 % for the adulterated material
by the collaborators (48 h after milking), as observed at the without thickeners, indicating that the modified method ex-
moment of preparation of the materials (zero time). hibits both sensitivity and selectivity, respectively (Table 2).
Satisfactory values of the ACC and CON were estimated
Hierarchical Cluster Analysis for all laboratories and concentrations of starch, indicating an
adequate standardization of the analytical procedure under the
The results of HCA hierarchical clustering are presented in the repeatability and reproducibility conditions, respectively, ex-
dendrogram (Fig. 3). In the first group observed, the four cept for the ACC obtained by laboratory 1 for the 0.3 g L−1 of
segments included in the study (government, food control, starch (Table 3).
food industry, and university) were represented by seven lab- The estimates of the statistical parameters that were used to
oratories (3, 4, 5, 6, 8, 9, and 10). The second group has two describe the prediction intervals for the probability of detec-
laboratories (2 and 7) representing the government and food tion are presented in Table 4 (Macarthur and Holst 2012). The
industry. Laboratory 1, industry representative, presented the probability to detect starch in materials containing from 0.3 to
unequal performance in relation to the others. In general, HCA 1.2 g L−1 ranged from 0.92 to 1.00, while the probability of
showed that the affiliation of the laboratory did not influence reporting false positive results was 0.03. These findings con-
the results obtained by each laboratory and, consequently, the firmed the sensitivity and selectivity of the method.
process of validation of the qualitative methods. The limits of unreliability regions (URs) and the DLs esti-
mated via non-linear regression (Gondim et al. 2014) for both
4 the single laboratory (Gondim et al. 2015) and interlaboratory
processes are shown in Table 5. The DL estimated for starch in
8
the collaborative trial was 0.32 g L−1. This value was close to
6 the DL of 0.2 g L−1, which was calculated in the previous
10
single laboratory process. The prediction interval estimated
for the DL according to the Macarthur and Holst (2012) pro-
9
cedure included these limits (Fig. 4).
5 As mentioned, interlaboratory validation studies were not
reported in the literature for these scopes. In-house validation
3
processes of qualitative methods for starch detection include,
7 largely, the use of instrumental techniques as infrared spec-
2 troscopy (IR) methodologies. For these methods, the reported
DLs ranged between 1.5 and 5 g L−1 (Zang et al. 2014;
1
Botelho et al. 2015), lower than that obtained in both single
0 10 20 30 40 50 60 70 80
and interlaboratory validation processes of the modified meth-
Variance Weighted Distance Between Cluster Centers od. Milk powder adulterated with starch was also analyzed by
Fig. 3 Hierarchical clustering analysis (HCA) dendrogram with the IR, and a complete discrimination from samples adulterated
distances between the laboratories with 2 % of starch was reported (Capuano et al. 2015). In a
Table 2 False positive (FPR), false negative (FNR), and reliability (RLR) rates that were obtained in the interlaboratory validation of the modified classical qualitative methods for detection of starch,
chlorides, and sucrose in milk

Laboratory Starch (g L−1) Chlorides (g L−1) Sucrose (g L−1)


number
0.0 0.3 0.8 1.2 0.9 (native) 1.5 2.0 2.5 0.0 2.4 3.0 3.6

FPR RLR FNR RLR FNR RLR FNR RLR FPR RLR FNR RLR FNR RLR FNR RLR FPR RLR FNR RLR FNR RLR FNR RLR

1 10 90 60 40 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
2 10 90 10 90 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
3 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
4 0 100 0 100 0 100 0 100 13 87 0 100 0 100 0 100 0 100 0 100 0 100 0 100
5 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
6 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
7 10 90 10 90 0 100 0 100 7 93 0 100 0 100 0 100 0 100 0 100 0 100 0 100
8 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 10 90 0 100 0 100
9 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
10 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100 0 100
Total 3 97 7 93 0 100 0 100 2 98 0 100 0 100 0 100 0 100 1 99 0 100 0 100
n 100 100 100 100 150 100 100 50 100 100 100 100

The values of the RLR highlighted in bold italics indicated results that were lower than 90 %, which were unsatisfactory
n total of test materials distributed among the collaborating laboratories
Food Anal. Methods
Food Anal. Methods

Table 3 Accordance (ACC) and


concordance (CON) values that Laboratory ACC
were obtained in the number
interlaboratory validation of the Starch (g L−1) Chlorides (g L−1) Sucrose (g L−1)
modified classical qualitative tests
for detection of starch, chlorides, 0.0 0.3 0.8 1.2 0.9 1.5 2.0 2.5 0.0 2.4 3.0 3.6
and sucrose in milk
1 0.80 0.47 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
2 0.80 0.80 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
3 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
4 1.00 1.00 1.00 1.00 0.75 1.00 1.00 1.00 1.00 1.00 1.00 1.00
5 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
6 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
7 0.80 0.80 1.00 1.00 0.87 1.00 1.00 1.00 1.00 1.00 1.00 1.00
8 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.80 1.00 1.00
9 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
10 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
CON 0.94 0.85 1.00 1.00 0.96 1.00 1.00 1.00 1.00 0.98 1.00 1.00

Values of ACC and CON highlighted in bold italics indicated results that were lower than 0.80 (for the materials
that were distributed in at least 10 replicates) or 0.60 (for the materials that were distributed in 5 replicates), which
were unsatisfactory

performance study of the official method for starch detection Generally, the results obtained in the interlaboratory vali-
in pasteurized milk carried out by Silva (2013), the detection dation process for the modified qualitative method for the
of starch was observed in 100 % of the samples adulterated detection of starch in milk were influenced by the false results
with concentrations between 0.1 and 25 g L−1 of starch, in reported by laboratory 1 for the concentration of 0.3 g L−1.
accordance with this interlaboratory study. The intermediate color formed at this concentration was dif-
ferent from the yellow color observed for the non-blind
unspiked sample, and it was also different from the blue color
Table 4 Statistical parameters and lower (5 %) and upper (95 %)
estimated limits for the detection probability of the thickeners for the
established in classical tests as the criterion for positive results,
modified classical qualitative methods for the detection of starch, which was only observed for samples adulterated with large
chlorides, and sucrose in milk concentrations of starch. Thus, these results could have oc-
curred because the modified procedure (criterion) was not
Thickener N X P sR Lower Upper
(g L−1) limit limit understood by this specific collaborator.

Starch
Chlorides
0.0 100 3 0.03 0.005 0.022 0.038
0.3 100 92 0.92 0.019 0.887 0.950
The FPR ranged from 0 to 13 %, resulting in a RLR from 100
0.8 100 100 1.00 0.00 0.971 1.00
to 87 % for the data from each collaborator for the adulterated
1.2 100 100 1.00 0.00 0.971 1.00
Chlorides
0.0 150 3 0.03 0.007 0.007 0.046 Table 5 Limits of unreliability regions and detection limits (DL) that
were estimated for the modified classical qualitative methods for the
1.5 100 99 0.99 0.003 0.984 0.998 detection of starch, chlorides, and sucrose in milk via the single
2.0 100 100 1.00 0.00 0.971 1.00 laboratory and interlaboratory processes
2.5 50 50 1.00 0.00 0.942 1.00
Method Limits of unreliability regions (g L−1)
Sucrose
0.0 100 1 0.01 0.003 0.002 0.038 Single laboratory validationa Interlaboratorial validation
2.4 100 99 0.99 0.003 0.962 0.998
3.0 100 99 0.99 0.003 0.971 0.998 Lower Upper (or DL) Lower limit Upper (or DL)
3.6 100 100 1.00 0.00 0.971 1.00
Starch 0.02 0.20 0.03 0.32
N total number of analyses, X total number of positive results, p estimated Chlorides 0.80 1.42 1.00 1.62
mean probability of detection of the thickener, sR standard deviation of Sucrose 0.30 1.90 0.43 1.99
the estimates of the probability of detection from the individual laborato-
a
ries (Macarthur and Holst 2012) Gondim et al. 2015.
Food Anal. Methods

1
with chlorides and it was 98 % for the material adulterated
0.9 with water that was not spiked with the thickeners (Table 2).
0.8 An ACC value of 0.75 was observed for laboratory 4, 0.87
0.7 for laboratory 7, and 1.00 for all of the other laboratories for
Probability

0.6 the adulterated test materials without the addition of sodium


0.5 chloride. Only the first disagreed with the established criteria.
0.4 However, this did not result in a CON value lower than 0.80.
0.3 For the other adulteration chloride concentrations, the estimat-
0.2 ed ACC and CON values were satisfactory. These results in-
0.1 dicated precision, i.e., that the validated procedure was suffi-
0 ciently standardized (Table 3).
0 0.2 0.4 0.6 0.8 1 1.2 The DL estimated by non-linear regression for this
Starch (g.L-1) interlaboratory trial was 1.38 g L−1 (Gondim et al. 2015),
1 similar to the calculated DL from the single laboratory process
0.9 (1.42 g L−1, Table 5). Considering the probability of detection
0.8 of 0.95 and the respective prediction interval (Macarthur and
0.7 Holst 2012), the DL was close to these values (Fig. 4).
Probability

0.6 The probability of reporting false positive results for the


0.5 modified method for the detection of chlorides was 0.03,
0.4 and the probability of detection was 0.99 to 1.00 for the test
0.3 materials containing 1.5 to 2.5 g L−1 of chlorides (Table 4),
0.2 proving the selectivity and sensitivity of the method.
0.1 Silva (2013) assessed the official method for chloride de-
0 tection in pasteurized milk and obtained 100 % of positive
0.9 1.4 1.9 2.4 results in samples added with commercial common salt at
Chlorides (g.L-1) concentrations from 0.27 to 1 g L−1. In this collaborative
1 study, upper concentrations were evaluated, although 100 %
0.9 of positive results were also observed.
0.8
0.7 Sucrose
Probability

0.6
0.5 All of the estimated parameters for the sucrose method were
0.4 satisfactory. For laboratory 8, for the test materials that
0.3 contained 2.4 g L−1 of sucrose, the estimated FPR, RLR, and
0.2 ACC parameters were not the maximum values due to one false
0.1 result. The probability to detect sucrose in materials containing
0 from 2.4 to 3.6 g L−1 ranged from 0.99 to 1.00, while the
0 1.2 2.4 3.6 probability of reporting false positive results was 0.01.
Sucrose (g.L-1) The DL calculated by the non-linear model in this
Fig. 4 Estimates of lower and upper limits for the detection probability interlaboratory trial was similar to that reported in the previous
and prediction interval for the limit of detection for the modified classical single laboratory study (1.90 and 1.99 g L−1, respectively),
qualitative methods for the detection of starch, chlorides, and sucrose. indicating the agreement in the results (Table 5). A little higher
Dashed lines, projection from the probability of detection 0.95 to give
lower (5 %) and upper (95 %) limits of a prediction interval for the limit of DL value and respective interval was achieved by the
detection (Macarthur and Holst 2012) Macarthur and Holst (2012) procedure (Fig. 4).
The method of detection of sucrose was evaluated during
the final round of the interlaboratory trial. Thus, the best re-
materials that did not contain sodium chloride, and these re- sults can be justified by the understanding of the modified
sults indicated the selectivity of the method. Laboratories 4 procedure that was adopted in this study and by an improved
and 7 reported 13 and 7 % false positive results, respectively. familiarization by the collaborators of the process of the
The sensitivity was demonstrated by 0 % of FNR, i.e., an RLR interlaboratory validation.
of 100 %, for all of the adulterated test materials that were All samples of pasteurized milk adulterated with sucrose in
spiked with sodium chloride. Based on all of the data, the RLR concentrations from 0.45 to 9 g L−1 were detected as positive
was 100 % for the adulterated materials that were also spiked in an evaluation of the official method of sucrose detection
Food Anal. Methods

performed by Silva (2013), which corroborates the results (LAFQ/LQA/IMA); Laboratório de Química Bromatológica da
Fundação Ezequiel Dias (FUNED); Laboratório Nacional Agropecuário
observed in this interlaboratory validation. Studies of IR and
de Minas Gerais do Ministério da Agricultura, Pecuária e Abastecimento
high-performance thin layer chromatography (HPTLC) (LANAGRO-MG/MAPA); Laboratório de Bromatologia—Unidade de
methods for, respectively, detection and quantification of su- Pesquisa Análise de Alimentos da Faculdade de Farmácia da
crose in milk were reported. For the qualitative method (IR), Universidade Federal de Minas Gerais (BRO-UPAA/FAFAR/UFMG);
NUGAP—Núcleo Global de Análise e Pesquisa Ltda.; and Trevo
samples were detected as positive in concentrations as low as Alimentos—Nogueira e Rezende Indústria de Laticínio Ltda.
5 g L−1. The DL and quantification limit of the HPTLC meth-
od were calculated as 3.58 and 10 ng L−1 (Rani et al. 2012; Liu Compliance with Ethical Standards
et al. 2015), respectively. The performance of these method-
ologies was lower and upper than that observed for the mod- Conflict of Interest Carina de Souza Gondim declares that she has no
ified method validated here. These results highlight that in- conflict of interest. Roberto Gonçalves Junqueira declares that he has no
conflict of interest. Scheilla Vitorino Carvalho de Souza declares that she
strumental techniques do not always have better performance has no conflict of interest.
than simple and low-cost qualitative techniques as used in this
work. Ethical Approval This article does not contain any studies with human
participants or animals performed by any of the authors.
Collaborator’s Comments Informed Consent Not applicable.

Starch

Laboratory 9—The reading was performed immediately after


the addition of Lugol’s solution, and the total reading time of References
the 20 samples was an average of 6 min. After reading the first
20 samples, we added Lugol’s solution again to the non-blind Almeida MR, Oliveira KS, Stephani R, Oliveira LFC (2012) Application
unspiked sample to monitor the remaining test materials. of FT-raman spectroscopy and chemometric analysis for determina-
Conducting the tests in this way facilitated the analysis be- tion of adulteration in milk powder. Anal Lett 45:2589–2602
cause the color that formed disappeared after some time. AOAC (Association of Official Analytical Chemists) (2002) AOAC
Official methods of analysis. Appendix D: guidelines for collabora-
Later, all samples that tested positive for starch showed a blue tive study procedures to validate characteristics of a method of anal-
precipitate, and some of them presented a bluish ring on the ysis. Washington: AOAC, 2002
surface. Botelho BG, Reis N, Oliveira LR, Sena MM (2015) Development and
analytical validation of a screening method for simultaneous detec-
tion of five adulterants in raw milk using mid-infrared spectroscopy
and PLS-DA. Food Chem 181:31–37
Conclusions Brasil – Ministério da Agricultura, Pecuária e Abastecimento (2006)
Instrução Normativa n° 68 de 12 de dezembro de 2006. Accessed
This interlaboratory study provided a complementary and fi- on April 2015. Available in www.agricultura.gov.br
nal assessment of the fitness for purpose of the modified clas- Brasil – Ministério da Agricultura, Pecuária e Abastecimento (2011)
Instrução Normativa n° 62 de 29 de dezembro de 2011. Accessed
sical methods for detection of thickeners in milk. These on April 2015. Available in www.agricultura.gov.br
methods may be an important and reliable tool for researchers, Capuano E, Boerrigter-Eenling R, Koot A, Ruth S (2015) Targeted and
industries, and regulation agencies for the monitoring and untargeted detection of skim milk powder adulteration by near-
control of milk adulteration. The detailed and systematized infrared spectroscopy. Food Anal Methods 8:2125–2134
Cardenas S, Valcarcel M (2005) Analytical features in qualitative analy-
scheme for the organization of a collaborative study for vali-
sis. Trends Anal Chem 24:477–87
dating qualitative methods was successfully applied. Das S, Sivaramakrishnaa M, Biswasb K, Goswam B (2011) Performance
study of a ‘constant phase angle based’ impedance sensor to detect
Acknowledgments The authors acknowledge the experimental farm of milk adulteration. Sens Actuators A 167:273–278
BProfessor Hélio Barbosa^ of the Veterinary School/Federal University of Ellis DI, Brewster VL, Dunn WB, Allwood JW, Golovanov AP,
Minas Gerais State (EV/UFMG) for allowing the use of their facilities and Goodacre R (2012) Fingerprinting food: current technologies for
animals during this project and for providing the raw cow milk samples. the detection of food adulteration and contamination. Chem Soc
We acknowledge Pedro Paulo Borges Santos for his analytical assistance Rev 41:5706–5727
and the Brazilian agency BCoordenação de Aperfeiçoamento de Pessoal Ellison SLR, Fearn T (2005) Characterising the performance of qualita-
de Nível Superior (CAPES)^ for financial support. The authors also ex- tive analytical methods: statistics and terminology. Trends Anal
press their appreciation to the collaborators for their participation in the Chem 24:468–476
study: GMO Centro de Pesquisas e Controle de Qualidade Ltda.; Ginn R, Wilson L, Souza SVC, Calle MB, Barbosa J, Berendsen B,
Hidrocepe Servicos de Qualidade Ltda. Epp; Ita Alimentos—Laticínios Bockborn I, Brandtner M, Delahaut P, Doering T, Fuerst P, Griffin
Ita Indústria e Comércio de Alimentos Ltda.; Itambé Alimentos S.A.; C, Gude T, Janosi A, Jaus A, Kennedy G, Mandix M, Hilla EM,
Laboratório de Análise Físico-Química em Alimentos do Laboratório Plonevez S, Posyniak A, Saari L, Bruijnsvoort M, Verdon E,
de Química Agropecuária do Instituto Mineiro de Agropecuária Wohlfarth R (2006) Determination of semicarbazide in baby food
Food Anal. Methods

by liquid chromatography/tandem mass spectrometry: Rani R, Medhe S, Raj KR, Srivastava MM (2012) High performance thin
interlaboratory validation study. J AOAC Int 89:728–734 layer chromatography for routine monitoring of adulterants in milk.
Gondim CS, Coelho OAM, Alvarenga RL, Junqueira RG, Souza SVC Natl Acad Sci Lett 35:309–313
(2014) An appropriate and systematized procedure for validating Renny EF, Daniel DK, Krastanov AI, Zachariah CA, Elizabeth R (2005)
qualitative methods: Its application in the detection of sulfonamide Enzyme based sensor for detection of urea in milk. Biotechnol
residues in raw milk. Anal Chim Acta 830:11–22 Biotechnol Equip 19:198–201
Gondim CS, Souza RCS, Palhares MPP, Junqueira RG, Souza SVC Santos PM, Wentzell PD, Pereira-Filho ER (2012) Scanner digital images
(2015) Performance improvement and single laboratory valida- combined with color parameters: a case study to detect adulterations
tion of classical qualitative methods for the detection of adul- in liquid cow’s milk. Food Anal Method 5:89–95
terants in milk: starch, chlorides and sucrose. Anal Methods 7: Santos PM, Pereira-Filho ER (2013) Digital image analysis—an alterna-
9692–9701 tive tool for monitoring milk authenticity. Anal Methods 5:3669–
India - Ministry of Health and Family Welfare of India (2005) Manual of 3674
Methods of Analysis of Foods - Milk and Milk Products Scampicchio M, Eisenstecken D, De Benedictis L, Capici C, Ballabio D,
Jawaid S, Talpur FN, Afridi HI, Nizamani SM, Khaskheli AA, Naz S Mimmo T, Robatscher P, Kerschbaumer L, Oberhuber M, Kaser A,
(2014) Quick determination of melamine in infant powder and liq- Huck C, Cesco S (2015) Multimethod approach to trace the geo-
uid milk by Fourier transform infrared spectroscopy. Anal Methods graphical origin of alpine milk: a case study of Tyrol region. Food
6:5269–5273 Anal Chem. doi:10.1007/s12161-015-0308-2
Karoui R, De Baerdemaeker J (2007) A review of the analytical methods Silva RAB, Montes RHO, Richter EM, Munoz RAA (2012) Rapid and
coupled with chemometric tools for the determination of the quality selective determination of hydrogen peroxide residues in milk by
and identity of dairy products. Food Chem 102:621–640 batch injection analysis with amperometric detection. Food Chem
133:200–204
Kartheek M, Smith AA, Muthu AK, Manavalan R (2011) Determination
Silva LCCD (2013) Capacidade de detecção de adulterantes e suficiência
of adulterants in food: a review. J Chem Pharm Res 3:629–636
das provas oficiais para assegurar a qualidade do leite pasteurizado.
Langton SD, Chevennement R, Nagelkeke N, Lombard B (2002)
PhD Thesis, Universidade Estadual de Londrina. Accessed on
Analysing collaborative trials for qualitative microbiological
January 2016. Available in https://www.lume.ufrgs.br/bitstream/
methods: accordance and concordance. Int J Food Microbiol 79:
handle/10183/81561/000905445.pdf?sequence=1
171–181
Souza SS, Cruz AG, Walter EHM, Faria JJF, Celeghini RMS, Ferreira
Liu J, Ren J, Liu ZM, Guo BH (2015) A new comprehensive index for
MMC, Granato D, Santana AS (2011) Monitoring the authenticity
discriminating adulteration in bovine raw milk. Food Chem 172:
of Brazilian UHT milk: a chemometric approach. Food Chem 124:
251–256
692–695
Lima RMJ, Fernandes SMV, Rangel AOSS (2004) Enzymatic determi- Thompson M, Ellison SLR, Wood R (2006) The international harmo-
nation of urea in milk by sequential injection with spectrophotomet- nized protocol for the proficiency testing of analytical chemistry
ric and conductometric detection. J Agric Food Chem 52:6887– laboratories. Pure Appl Chem 78:145–196
6890 Tronco VM (2010) Manual para inspeção da qualidade do leite. Editora
López MI, Colomer N, Ruisánchez I, Callao MP (2014) Validation of da UFSM, Santa Maria
multivariate screening methodology. Case study: detection of food Trullols E, Ruisánchez I, Rius FX (2004) Validation of qualitative
fraud. Anal Chim Acta 827:28–33 methods of analysis that use control samples. Trend Anal Chem
López MI, Callao MP, Ruisánchez I (2015) A tutorial on the validation of 23:137–145
qualitative methods: from the univariate to the multivariate ap- Van der Voet H, Van Rhijn JA, Van de Wiel HJ (1999) Inter-laboratory,
proach. Anal Chim Acta 891:62–72 time, and fitness-for-purpose aspects of effective validation. Anal
Macarthur R, Holst C (2012) A protocol for the validation of qualitative Chim Acta 391:159–171
methods of detection. Anal Methods 4:2744–2754 Veloso ACA, Teixeira N, Ferreira IMPLVO, Ferreira MA (2002)
NATA (National Association of Testing Authorities – Australia) (2013) Detecção de adulterações em produtos alimentares contendo leite
Technical Note 17 – October 2013 - Guidelines for the validation e/ou proteínas lácteas. Quim Nov. 25:609–615
and verification of quantitative and qualitative test methods. Ward JH (1963) Hierarchical grouping to optimize an objective function.
Accessed on January 2016. Available in http://www.nata.com.au/ J Am Stat Assoc 58:236–244
nata/phocadownload/publications/Guidance_information/tech- Zang LG, Zang X, Ni LJ, Xue ZB, Gu X, Huang SX (2014) Rapid
notes-information-papers/technical_note_17.pdf identification of adulterated cow milk by non-linear pattern recog-
Nigam A, Ayyagari A (2007) Laboratory manual in biochemistry: immu- nition methods based on near infrared spectroscopy. Food Chem
nology and biotechnology. Tata McGraw-Hill, New Delhi 145:342–348

You might also like