Development of rice reference material and its use for evaluation of analytical performance of food analysis laboratories

Journal of Food Composition and Analysis 22 (2009) 453–462

Contents lists available at ScienceDirect

Journal of Food Composition and Analysis journal homepage: www.elsevier.com/locate/jfca

Original Article

Development of rice reference material and its use for evaluation of analytical performance of food analysis laboratories Prapasri Puwastien *, Kunchit Judprasong, Naruemol Pinprapai Institute of Nutrition, Mahidol University at Salaya, Putthamonthon 4, Nakhon Pathom 73170, Thailand

A R T I C L E I N F O

A B S T R A C T

Article history: Received 16 October 2007 Received in revised form 14 April 2008 Accepted 16 January 2009

Data quality is one of the major concerns in development of food composition database and to editors of many peer-reviewed journals in accepting a scientific paper for publication. Regular use of a reference material and participation in a proficiency testing programme could improve the reliability of the analytical data. The objectives of this project were to prepare rice test material with assigned values and to use it to assess the analytical performance of laboratories which are involved in research and analysis of rice. The international guidelines, ISO Guide 35, ISO 13528 and ISO Guide 43, were followed as much as possible throughout the preparation of the reference material and the laboratory performance study. Brown rice (Jasmine variety) was ground to particle size which passed completely through a sieve with pore size 250 mm and packed in laminated aluminum foil bags under vacuum. Based on the analyses of representative nutrients – moisture, protein, iron, zinc and vitamin B1 – the samples were demonstrated homogeneous. Ten expert laboratories from various countries, 36 laboratories from Thailand, and 16 laboratories from ASEAN and Asia registered for the laboratory performance study. The samples were sent for analysis of selected proximate composition (moisture, protein, dietary fibre and ash), two minerals (iron, zinc), and one labile nutrient (vitamin B1) using routine analytical methods. The assigned values of the nutrients in the test materials, as robust mean  robust SD or predicted SD, were established with their uncertainties. For proximate composition, 67–87% of participating laboratories showed good analytical performance. However, many of them showed questionable and unsatisfactory performance on the analyses of dietary fibre (55%) and vitamin B1 (47%). The evaluation of the results of moisture, protein and iron with their uncertainties against the assigned values of the test material using En score was also demonstrated. Finally, the consensus values of nutrients in the rice sample as mean  SD were developed from the analytical results of laboratories with good performance for both within- and between-laboratories. This test material can be used as a reference material for internal and external quality control systems to improve the quality of the analytical data. ß 2009 Elsevier Inc. All rights reserved.

Keywords: Reference material Rice Nutrients Assigned value Analytical performance Food composition

1. Introduction Reliable analytical data are required by both food analysts and data users. Standardisation of the analytical methodology and development of a quality control system in a laboratory can help ensure analytical measurement validity and increase data quality and reliability. Reference materials play key roles in the development of the internal quality control system. It is well known that different methods of nutrient analysis or the same analytical methods with some modification are being used by different laboratories, resulting in some discrepancies in the analytical data. Proficiency testing is an external quality assessment designed to

* Corresponding author. Tel.: +66 2 4410217; fax: +66 2 4419344. E-mail address: [email protected] (P. Puwastien). 0889-1575/$ – see front matter ß 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.jfca.2009.01.006

assess the laboratory analytical performance which reflects the reliability of the analytical data. It assists in increasing confidence in analyst ability in the case of good performance, or in identifying laboratories with questionable and unsatisfactory results where improvement of the competency in nutrient analysis is required. Seven rounds of laboratory performance studies were conducted by the Institute of Nutrition, Mahidol University during 1989–2003 (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 1989, 1999, 2001, 2003) using different test materials with consensus assigned values. Two approaches were used to develop assigned values of food components: one from expert laboratories and another from good performance laboratories who participated in laboratory performance studies. With the collaborative study among expert laboratories in Australia, New Zealand, USA, Austria, and laboratories in ASEANFOODS member countries, nine food reference

454

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

materials with consensus values for proximate composition and some minerals were produced. These materials were rice flour (AS-FRM1) and soybean flour (AS-FRM2) (Puwastien et al., 1989); cereal-soy (AS-FRM3) and fish flour (AS-FRM4) (Puwastien and Sungpuag, 1995); weaning food (AS-FRM-5) and fish powder (ASFRM-6) (Puwastien et al., 1999a); feed (AS-FRM-7) and fish meal (AS-FRM 8) (Puwastien et al., 2001); and milk powder (AS-FRM 9) (Puwastien et al., 2003). Since 2004, the International Year of Rice, various studies on nutritive values of different varieties of rice have been conducted for research and development, generation of food composition data, and for screening and selecting plant varieties. To evaluate performance of laboratories that have analyzed rice, a laboratory proficiency study on nutrient analysis was needed. Thus two main objectives of this study were (1) to prepare a candidate reference material (RM) of rice powder and study its physical and chemical characteristics and (2) to organise a laboratory performance study using the prepared candidate RM as test material. 2. Materials and methods 2.1. Preparation of candidate reference material (RM) 2.1.1. Test material for mineral analysis Test material used was brown rice, Jasmine variety, obtained from Center of Excellence for Rice Molecular Breeding and Product Development, Kasetsart University, Thailand. Three kilograms of the test material were frozen using liquid nitrogen in order to make rice to be brittle and facilitate the grinding process and then ground in a stainless steel grinding machine (to avoid iron and zinc contamination) until the fine particles passed completely through sieve No. 60 mesh (250 mm). The sample was mixed thoroughly manually, in a humidity controlled air conditioned room and then packed under vacuum in aluminum foil bags, about 10 g each. Bags were randomly divided into 2 sets, A and B, and then labeled with sample code number. These prepared samples were used for analyses of minerals (iron, zinc) and kept in a freezer at 20 8C. 2.1.2. Test material for analyses of proximate composition and vitamin B1 Another set of the test material, 15 kg, was ground using the Cyclotec sample mill until the fine particles passed completely through sieve No. 60 mesh (250 mm). The sample was then mixed thoroughly by a rotating mixer for 5 h in a humidity controlled air conditioned room and then packed under vacuum in aluminum foil bags, about 30 g each. Bags were randomly divided into 2 sets, A and B, and then labeled with sample code number. They were used for the analysis of proximate composition and vitamin B1. The samples were kept in a freezer at 20 8C. 2.2. Characteristics of the candidate reference material 2.2.1. Particle size distribution The particle size distribution of the test material for nutrient analysis (Section 2.1.2) was manually studied by sieve analysis. One hundred gram of the test material was passed through 3 standard sieves 60, 80 and 120 mesh, with pore sizes of 250, 180 and 125 mm, respectively. Each fraction was collected, weighed and recorded. Percent distribution at each fraction was calculated. 2.2.2. Homogeneity study Ten packages (5 from set A and 5 from set B) each of the prepared sub-samples for mineral analysis and for proximate analysis were selected at random. Homogeneity of the candidate material was evaluated by analyses of selected representative nutrients, i.e. iron and zinc (representative of trace elements), moisture and protein (representatives of proximate composition)

and vitamin B1 (representative of labile nutrient). The analyses were performed in two test portions from each package, in a random order. Each analysis was performed in one setting under repeated conditions, i.e. by competent analysts, on the same day using the same set of reagents and conditions. The results were statistically evaluated. 2.2.3. Stability study Since Vitamin B1 is the most labile nutrient in rice, its stability was checked throughout the storage period. The prepared candidate reference materials were kept at 20 8C. Five packages of the prepared sub-samples were randomly selected at 1, 6 and 12 months intervals for the first year and every 6 months thereafter. At each period, single analysis of vitamin B1 by HPLC in each sample was conducted. The stability of the vitamin was evaluated by comparing the results obtained at each period with the levels analysed at 0 month (using data from homogeneity study). 2.2.4. Chemical analyses of the components in the test materials Ten expert laboratories from different countries in OCEANIA, Europe, and North America collaborated to develop assigned values of nutrients in the test material. Each laboratory registered to analyse several and none of them analysed all of the assigned components. Since the number of derived data from expert laboratories would not be a sufficient number for reliable assigned values, the better assigned values for the measurands in the test material were derived from the analytical values of both expert laboratories and the participants of the performance study according to the ISO 13528, 2005. 2.3. Laboratory performance study Following closely the ISO Guide 43 (ISO Guide 43-1, 1997), a laboratory performance study for nutrient analyses was conducted. 2.3.1. Participants Fifty-six laboratories from various countries, mainly from ASEAN, registered to participate in the performance study. However, not all laboratories registered for analyses of all assigned measurands and four of them did not submit the report. 2.3.2. Distribution of samples and documents Two packages each of 10 g test material (package A and B with random number) for analysis of minerals and 30–40 g (package A and B with random number) for analysis of other components were sent to expert laboratories and oversea participants via airmail and by post for local laboratories. Five documents – instruction to the participants, report form, uncertainty form, questionnaire for method used and questionnaire for in-house quality control system – were sent electronically as attached files with a secret laboratory code number assigned to each laboratory. 2.3.3. Analytical components and methods of analysis Participating laboratories were assigned to analyse proximate composition (moisture, protein, dietary fibre, ash), minerals (iron and zinc), and vitamin B1 using their routine test methods. They were requested to analyse vitamin B1 within two weeks of receiving the samples. Two individual values (A and B) of each component, one from each package of the test material, were requested to be reported in the report form where unit of expression and number of significant decimal places were indicated. For the first trial, the participants were requested to report values for moisture, protein and iron with their uncertainty values (expanded uncertainty with a coverage factor of k = 2).

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

2.3.4. Submission of the results Laboratories were requested to submit the report and related documents within 10 weeks after receiving the samples.

455

2.4.4. Evaluation of laboratory performance The analytical results of all components submitted by the participating laboratories were evaluated first for within- and then for between-laboratory variations as follows.

2.4. Statistical analysis 2.4.1. Homogeneity of test materials The data of duplicate values of moisture, protein, iron, zinc and vitamin B1 derived from 10 sub-samples of rice powder were evaluated for within-sample variation using Cochran’s maximum range test (ISO 5725, 1981) which indicated the analytical precision. To determine the homogeneity (between sample variations) of the test materials, the data were evaluated using one-way ANOVA (ISO 5725, 1981) without removing any values. The measurement uncertainty associated with the sample homogeneity was estimated for information following the method used by National Measurement Institute (NMI), Australia (NARL, 2004; NMI, 2004). 2.4.2. Stability of vitamin B1 in the rice powder The results of single analysis of vitamin B1 in 5 random samples at each storage period were evaluated against the levels obtained from the analysis at 0 month. If the values fell in the range of mean 2SD value of vitamin B1 at 0 month, the results indicated the stability of the component. A slope of the regression line (linear least square) of the values during 12 months storage was also evaluated. 2.4.3. Assignment of component values in test materials

A. Within-laboratory variation: For each pair of the results (A and B), the difference between the values was used to evaluate within-laboratory variation by calculation of robust z-score(within) based on the variation [median and Normalised Inter-Quartile Range (NIQR) NATA (1996)] within the group of participants. z-scoreðwithinÞ ¼

where x is the difference between the values of A and B from pffiffiffi each laboratory= 2; median is the median of difference between the value of A and B obtained from participating laboratories; NIQR (normalised inter-quartile range) is (Quartile 3Quartile 1)  0.7413. B. Between-laboratory variation: For each pair of results, betweenlaboratory variation was evaluated by calculation of robust z-score(between). Two approaches were applied. 2.4.4.1. Approach 1. An average value of A and B was used to evaluate between-laboratory variations. The z-score was calculated based on the assigned value, which is the robust mean  robust SD, estimated from the data of participants according to the ISO 13528 (2005). z-scoreðbetweenÞ ¼

A. According to ISO 13528 (2005): The assigned values of components in the test material were developed from the analytical data obtained from 10 expert laboratories and the participants of the performance study following the statistical process of ISO 13528 (2005). The process started by removing the known extreme values due to common errors such as using unaccepted analytical methods, misplacement of the decimal points, wrong unit of expression, etc. Then several steps were conducted to modify the extreme values, if any, until the robust mean and robust standard deviation were obtained. At the final step, the standard and expanded uncertainties of the assigned value were estimated. The obtained robust mean  robust SD and the expanded uncertainties are the assigned values of the components in the test material. They were used to evaluate the laboratory performance in this study by using z-score and En-score (as a trial), respectively. B. According to ISO 13528 and target standard deviation (SD) of Horwitz (Horwitz et al., 1992): In some cases, variation of a set of analysed data obtained from various laboratories is too large. This occurs frequently for nutrients with low concentration or nutrients with complicated analytical methods. The high variation can be demonstrated by the percentage of coefficient variation (%CV) which is higher than the target values, such as %CV more than 10 for proximate composition or more than 15 for vitamins and minerals. The robust mean derived from the process of the ISO 13528 (2005) for the particular components will be used as the assigned mean value but the assigned robust SD was replaced by the target SD of Horwitz based on the robust mean value. Horwitz0 s Predicted Relative Standard Deviation or RSDp ¼ 210:5 log C where C = fraction concentration of the mean value of the component to be evaluated.

ðx  medianÞ NIQR

ðx  robust meanÞ robust SD

where x is average value of reported A and B of a nutrient per 100 g, obtained from each participating laboratory; Robust mean is the assigned value of the nutrient per 100 g according to ISO 13528; Robust SD is the standard deviation of the robust mean value according to ISO 13528. 2.4.4.2. Approach 2. In some cases, variation of a set of analysed data obtained from various laboratories was too large. Laboratory performance on the analysis of the particular components, i.e. dietary fibre, iron and vitamin B1, with large variance was evaluated based on the assigned robust mean of the reported values and the predicted SD obtained from Horwitz equation (ISO Guide 43-1, 1997). C. Estimation of En score (NARL, 2004): In this performance study, a trial on use of En score to evaluate the laboratory performance was also conducted on analyses of moisture, protein and iron. The En score was calculated as follows. xlab  xref En ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 þ U2 Þ ðUlab ref where xlab is the mean of the results submitted by participants, xref is the robust mean of the assigned value derived from ISO 13528 (2005), Ulab is the expanded uncertainty of xlab and Uref is the expanded uncertainty of xref derived from ISO 13528 (2005). 2.5. Interpretation of laboratory performance study 2.5.1. Z-scores Results with an absolute z-score  2 were satisfactory. Values with the absolute z-score 2 < jz-scorej < 3 were identified as questionable results. Values with the absolute z-score  3 were identified as unsatisfactory values. These criteria were applied for

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

456

both within- and between-laboratory variation. Absolute z-score

Interpretation

jzj  2 2 < jzj < 3 jzj  3

Satisfactory result Questionable result Unsatisfactory result (presented as extreme values)

An absolute En score of 1 indicates that the reported results and assigned values are in agreement (within their respective uncertainties). An absolute En score of >1 indicates that the reported results are different from the assigned value and that the uncertainty associated with the results has been understated. In this study, the En score was calculated and presented for information only. The values have not yet been included in the evaluation process of the laboratory performance. Fig. 1. Pattern of changes of vitamin B1 in sub-samples of rice flour test materials during one-year storage at 20 8C.

3. Results and discussion 3.1. Rice flour preparation Since iron and zinc were included as the target measurands, the grinder should be free from these minerals. Using the Cyclotec sample mill resulted in rice flour with uneven distribution of iron. A stainless steel grinder (with limited capacity) which provided homogenous distribution of iron and zinc was therefore used. To facilitate the grinding process due to the hard core of the rice, liquid nitrogen was used. The Cyclotec sample mill which has high speed and capacity is a proper instrument for preparation of rice flour to be used as test materials for other nutrient analyses.

Between-sample variation, which used to indicate sample homogeneity, was evaluated using the duplicate results of each representative nutrient derived from 10 packages. The F-values from ANOVA for all components in the prepared test materials were lower than the critical F-value. The results indicated that the sub-samples were considered sufficiently homogeneous to be used as test materials for laboratory performance study. For each of the analysed nutrients, an estimation of uncertainty associated with homogeneity was made following the method of NMI, Australia (NARL, 2004; NMI, 2004). A summary of homogeneity test results and statistical treatment including the uncertainty values is shown in Table 1.

3.2. Characteristics of the test material (candidate RM) 3.2.1. Particle size distribution of rice flour The prepared rice powder passed completely through 60 mesh sieve (<250 mm). Particle size analysis indicated that about 20% of the powder had particle size between 180–250 mm, about 50% of the powder was between 125–180 mm, and the rest (30%) was less than 125 mm. The effects of particle size difference on the homogeneity of the flour in terms of target nutrients were investigated in the following step. 3.2.2. Homogeneity of the prepared rice flour The within-sample variation which was evaluated by Cochran’s maximum range test showed that the ratios of the maximum range divided by sum of the ranges of all representative nutrients (0.32, 0.25, 0.26, 0.32, and 0.45 for moisture, protein, vitamin B1, iron and zinc, respectively) were less than the Cochran’s critical value (0.602 for 10 sets of data, when the number of results per set (N) = 2) (ISO 5725, 1981). The results indicated good precision of the analysts who performed the testing of the representative nutrients.

3.2.3. Stability of vitamin B1 in the test material Brown rice is a good source of vitamin B1. Its analysis is therefore included in this study. Since the vitamin may not be stable during transportation and storage, its stability status was monitored throughout the study period and during storage. The pattern of changes in vitamin B1 during one-year storage of the samples is shown in Fig. 1. The vitamin content in the samples, kept for two weeks at room temperature (representing the transport temperature of the samples) and at 20 8C for 2 weeks and for 12 months storage at 20 8C, was within the value of mean  2SD of the levels at 0 day. The slope of the regression line is not statistically different from zero. The intercept of the regression line (0.578) is not statistically different from the initial value (0.576 mg/100 g). The results implied that vitamin B1 in the test materials was stable throughout the storage period of one year. The level of vitamin B1 in the test materials was measured every 6 months during storage at 20 8C. An isochronous approach (ISO Guide 35, 2006) would be considered for checking the stability of vitamins in future studies in order to reduce

Table 1 Summary of homogeneity testing results of the test material. Representative nutrientsa

Mean result (per 100 g)

Uncertaintyb Uhom (per 100 g)

F-valuec (Critical F-value = 3.02)d

Sse

Uhomf (per 100 g)

Result

Moisture (g) Protein (g) Fe (mg) Zn (mg) Vitamin B1 (mg)

10.41 7.92 1.42 2.11 0.57

0.39 0.09 0.12 0.06 0.07

1.64 0.70 1.21 0.83 0.23

0.094 0.015 0.018 0.010 0.010

0.187 0.031 0.035 0.021 0.020

Pass Pass Pass Pass Pass

a Moisture was analysed by oven-drying at 100 8C, until constant weight; protein by Kjeldhal method (total N  5.95); iron and zinc by wet digestion and ICP/OES; vitamin B1 by HPLC. b Expanded combined standard uncertainty associated with homogeneity (using a coverage factor of k = 2, 95%CI). c F-value obtained from ANOVA. d Critical F-value (9,10) = 3.02 (p = 0.05). F-value < critical F-value qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi = sample is homogeneous. pffiffiffi MSbetween MSwithin e ; if F-value is <1, Ss ¼ SD= 6. Ss = sampling variance. If F-value is >1, Ss ¼ 2 f

Expanded standard uncertainty associated with homogeneity (using a coverage factor of k = 2, 95%CI).

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

457

Table 2A Assigned values of nutrients in rice, estimated by ISO 13528. Measurand

Unit per 100 g

Robust mean

Robust SD

Expanded uncertainty

%RSD

Moisture: estimated from all submitted values Moisture: estimated from data with empirical method Protein Dietary fibre Ash Iron Zinc Vitamin B1

g g g g g mg mg mg

11.64 12.15 8.07 3.48 1.41 1.19 1.72 0.39

0.73 0.51 0.25 0.68 0.05 0.22 0.21 0.12

0.25 0.33 0.09 0.34 0.02 0.08 0.09 0.07

6.3 4.2 3.1 19.6* 3.6 18.5* 12.2 30.1*

*

Too high %RSD, robust mean  Horwitz predicted SD (SDp) was applied as shown in Table 2B.

Table 2B Assigned values of dietary fibre, iron and vitamin B1 in rice as robust mean  predicted SD of Horwitz. Measurand

Unit per 100 g

Robust mean

Horwitz’s Predicted SD

%RSD

Dietary fibre Iron Vitamin B1

g mg mg

3.48 1.19 0.39

0.12 (3SDp = 0.35) 0.13 0.05

10.0 11.0 13.0

measurement variation which may encounter in various sets of analysis. 3.2.4. Assigned values of nutrients in test materials The robust mean (x*) and robust standard deviation (S*) were estimated from the results submitted by participants according to ISO 13528 (2005). A summary of the assigned values of moisture, moisture by empirical method, protein, ash and zinc in the rice sample with their expanded uncertainty, according to ISO 13528 (2005), is presented in Table 2A. Several measurands – dietary fibre, iron and vitamin B1 – showed robust mean and SD with high %CV. Their assigned values were therefore estimated as robust mean  Horwitz predicted standard deviation (SDp). The robust mean and the modified robust SD as SDp for these 3 measurands are presented in Table 2B. The estimated assigned values were used to evaluate analytical performance of all laboratories by calculation of zscore. The expanded uncertainty of moisture, protein and iron was used to calculate En scores, which is an additional parameter for the evaluation of laboratory performance. 3.3. Laboratory performance on the analysis of nutrients in rice Fifty-six laboratories, governmental and non-governmental, registered for this laboratory performance study. There were 37 laboratories from Thailand and 19 laboratories from ASEAN and Asia but four registered laboratories did not submit their reports. Besides, not all laboratories had facilities to analyse all of the assigned components. Therefore, we decided to include the results obtained from ten expert laboratories with the results obtained from 52 participants and evaluate them together. Many of the participants did not submit the requested information on methods of analysis. The information was very useful for data interpretation especially when extreme values were identified. 3.3.1. Estimation of z-score In the laboratory performance study, each laboratory received 2 packages of each test material (package A and B with random number) without knowing that they were the same homogenised samples. A single value of each component from each package was requested. The objective of this study design was to evaluate the analytical precision of the participants and to minimise their bias. Individual data from participants were first evaluated for withinlaboratory variation based on the median and NIQR of the

participants’ values. Then the analytical performance of each laboratory was evaluated based on the assigned values of the components in the test material. The use of assigned values as robust mean and robust SD was applied for the data of moisture, protein, ash and zinc, whereas the robust mean and Horwitz’s predicted SD were applied for dietary fibre, iron, and vitamin B1. Laboratories with satisfactory, questionable and unsatisfactory results were identified based on the z-scores of the betweenlaboratory variation. General observations and comments on the results for each nutrient, based on the robust z-score, are presented as follows. A summary of individual results, including statistical parameters, i.e. assigned values, z-score and En score chart (if any), is presented for moisture in Fig. 2A and B as an example. Discussion of results of other nutrients, i.e. protein, ash, dietary fibre, iron, zinc and vitamin B1 are included without graphical presentation. Moisture: Fifty-three laboratories submitted the results for moisture content. Distribution of data of moisture, z-scores and En score obtained was graphically presented in Fig. 2A and B. The most common method used for moisture analysis among the participants was oven-drying at 100  5 8C for several hours or until constant weight. Other methods used included drying in a hot air oven at a higher temperature for a shorter time, i.e. 130–135 8C for 1– 3 h. Drying in a vacuum oven at 100 8C for 5 h was also applied in several laboratories (Lab Nos. 4, 20, 32, 46, and 52). Lab No. 32 determined the moisture content by drying the samples in a lyophilyser for 72 h. Lab No. 1 used both hot air oven and Sartorius moisture analyser MA100H (halogen heating). According to the results presented in Fig. 2A, the differences in the analytical methods used for moisture analysis had generally no effect on the submitted results which is in accordance with the findings in previous studies (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 2001, 2003). The z-scores, evaluated based on the robust mean and robust SD obtained according to the ISO 13528 indicated only few laboratories (Lab Nos. 17, 61 and 62) with extreme values for moisture content. Lab No. 17, which weighed too small an amount of the analysed samples, 0.5 g each, and dried the samples at 105 8C for a short period of only 45–70 min, reported low results for moisture content. Lab Nos. 61 and 62 did not provide the methods of analysis. According to ISO/IEC 17025 (2005), moisture should be analysed by empirical method. Thus, the assigned value of moisture in the test material was re-estimated from 15 out of 54 laboratories which used the methods of AOAC 925.09, 2005 (AOAC, 2005) (drying sample at 98–100 8C until constant weight using vacuum oven) and AOAC 925.10, 2005 (drying the sample at 130  3 8C for 1 h) for moisture analysis in rice flour. The assigned values estimated according to the ISO 13528 as robust mean  robust SD, are 12.15  0.51 g/100 g (Table 2A, Moisture, empirical method). Laboratory performance for moisture analysis was re-evaluated using the new assigned values. The evaluation (Fig. 2B) showed higher number of laboratories with extreme values, 7 out of 52 (Lab Nos. 1, 4, 17, 46, 49, 54, and 62). Among these laboratories, Lab No. 1 performed the moisture analysis method at low temperature, 85 8C, for 5 and

458

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

25 h, and Lab No. 17 weighed too small amounts of samples and dried them for a short time. Other laboratories with extreme values dried the samples at 100  5 8C for 5 h or until constant weight. Since the methods used were not much different from those which were identified as good performance laboratories, these laboratories were

real outliers for moisture analysis. Besides, more than 20 laboratories out of 35 which were identified as good performance laboratories did the moisture analysis by drying the samples at 100  5 8C for several hours or until constant weight was obtained, without applying vacuum. They reported the results in the same range as those that

Fig. 2. (A) Moisture values of rice: data obtained from individual laboratories. (B) z-scores: moisture of rice (based on assigned value derived from empirical method).

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

459

Fig. 2. (Continued ).

applied empirical methods. This implies that drying the rice samples at 100  5 8C until constant weight is reached can be used for moisture determination in rice. A summary of the performance evaluation on moisture analysis based on z-score is presented in Table 3. After evaluation of laboratory performance on moisture analysis based on the assigned values from empirical method, the consensus value for moisture content in the rice test material was estimated from the values reported by participating laboratories with good performance for both within- and betweenlaboratory variations. As shown in Table 3, the assigned consensus value for moisture from empirical method is 11.91  0.49 g/100 g (mean  SD, N = 36, %CV = 4.1). The value agrees well with that of brown rice (11.2 g/100 g, N = 6) published in the Thai Food Composition Tables (FCTs) (Puwastien et al., 1999b). Crude protein: Rice contains protein at about 8 g/100 g. The methods used for protein analysis among the participating laboratories are based on the traditional Kjeldhal method. The methods used differed in choice of catalysts and proportion. Most laboratories used a mixture of K2SO4 and CuSO4 with or without Se, as catalysts. Types of catalysts were previously found to have no significant effect on the protein content (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 2001, 2003; Torelm, 1994). The total nitrogen results were converted to

protein by multiplying by the factor of 5.95. Fifty laboratories submitted the results for protein analysis. According to the z-score for within- and between-laboratory variation, forty laboratories out of 50 (80%) were identified as good performance (Table 3). Eight out of 50 laboratories reported unsatisfactory results either for within- or between-laboratory variation. Two laboratories submitted extremely high protein values. One of these labs followed the standard method as other laboratories which had satisfactory results, thus it can be considered a true outlier for protein determination. Another which analysed total nitrogen directly by combustion using a small amount of the samples (0.01– 0.02 g) was identified as an outlier with high values for both within- and between-laboratory variation. Direct analysis of total N by combustion could result in higher values of total N than those analysed by the traditional method which involves several steps. However, with the limited amount of the sample used, the precision of this method is lower than the traditional method. The effect of the amount of the sample on the analytical precision should be further investigated. The mean  SD of protein content in the test materials derived from participating laboratories with accepted values for both withinand between-sample variation is 8.07  0.23 g/100 g (N = 40), with %CV = 2.9 (Table 3). Although a few outliers were identified in this group of participants, the performance of protein determination can

Table 3 Summary of the analytical performance—evaluated based on z-scores, and the final consensus assigned values of nutrients in rice. Nutrient

Moisture Moisturea (empirical method) Protein Dietary fibre Ash Iron Zinc Vitamin B1 a b c d

Total number of participants

54 54 50 31 47 48 39 19

Evaluation of laboratory results based on z-scoresb

Consensus values of nutrients

Satisfactory z-score  2

%

Questionable 2 < z-score < 3

%

Unsatisfactory z-score  3

%

Mean  SDc

%CVd

41 36 40 14 33 26 26 10

76 67 80 45 70 54 67 53

7 8 2 0 6 7 4 2

13 15 4 0 13 15 10 11

6 10 8 17 8 15 9 7

11 19 16 55 17 31 23 37

– 11.91  0.49 8.07  0.23 3.37  0.37 (suggested value) 1.41  0.04 1.14  0.12 1.75  0.16 0.37  0.05 (suggested value)

– 4.1 2.9 11.1 3.0 10.6 8.8 12.6

Based on the assigned values obtained from laboratories which applied empirical methods for moisture determination. For both within- and between-laboratory variations. Derived from good performance laboratories for both within- and between-laboratory variations. CV, coefficient variation.

460

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

be considered as satisfactory. This is indicated by the consensus values obtained with a low %CV. Total dietary fibre (TDF): Twenty-six laboratories submitted results for TDF in which an enzymatic gravimetric method was mainly used. Five laboratories submitted values of crude fibre, and as expected they reported extremely low values. Their values were not included in the estimation of the assigned values. Although the test materials are dry rice powders, containing low amount of lipid (2–3 g/100 g) and total sugars (less than 1 g/100 g), several laboratories removed moisture, fat and sugars before analysis of TDF. Since the submitted data of the TDF had a wide range of values, robust mean and robust SD estimated according to ISO 13528 (Table 2A) could not be accepted due to high values of % RSD (about 20%). The assigned values of TDF in rice test material obtained as robust mean (according to ISO 13528) with Horwitz’s predicted SD, were 3.48  0.12 g/100 g, with %RSD of about 3. Since several steps and analyses of different components are involved in the determination of TDF, a higher RSD than the estimated Horwitz’s predicted values can be expected as experienced in previous laboratory performance studies (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 2001, 2003). Thus, the values of robust mean with less strict criteria of SD, 3 times the Horwitz predicted SD, as 3.48  0.35 g/100 g with %RSD of about 10, was used as the assigned SD for evaluation of the data (Table 2B). Seven out of the 26 laboratories who analysed TDF reported outlier high values (3 labs) and outlier low values (4 labs). Low amounts (which might not represent the tested materials) or excess amounts of the samples (which may require higher amounts of enzymes) used for dietary analysis could be important factors involved in the variable or underestimated values of the TDF. Similar findings were experienced in previous proficiency rounds (Puwastien et al., 2003). Therefore, it is recommended that sample weight of about 1 g be used for TDF analysis in rice sample. The steps to remove moisture, sugars and lipids should be included prior to enzymatic treatment only if necessary, although it seems likely that these steps did not affect the results of this study. It was surprising that some laboratories strictly followed the standard AOAC method but their results were identified as outliers with high- and low-values. A wide range of TDF data submitted by participants was recorded in previous proficiency studies when different test materials were used (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 2001, 2003). The findings in the previous studies indicate unsatisfactory performance status of laboratories on TDF analysis. The standard AOAC procedure used among participants should be carefully reviewed and an interlaboratory study to evaluate the performance on TDF analysis should be regularly organised. The consensus value as mean  SD of TDF content in the test material of rice was estimated from good performance laboratories for both within- and between-laboratory variations. It is 3.37  0.37 g/100 g (N = 14, %CV = 11) as shown in Table 3. The levels of the TDF in the rice test materials fall between the value (4.6 g/100 g) showed in the USDA database (USDA National Nutrient Database for Standard Reference, Release 21 (USDA, 2008) and the value presented in the Thai FCTs (2.8 g/100 g). Because of the limited number of participants with satisfactory results, the consensus value of TDF is given for information as suggested value. Ash: Forty-seven laboratories submitted results of ash analysis. The most common temperature used for ashing among the participants was 500–550 8C. Eleven laboratories used a higher temperature, 550–600 8C, for a period of 2 h to overnight. Similar to the previous studies no obvious effect of different ashing temperature on the ash values was seen (Puwastien and Sungpuag, 1995; Puwastien and Raroengwichit, 2000; Puwastien et al., 2001, 2003). All the values submitted did not vary much, % robust CV was 3.4. The assigned values – robust mean and robust SD – estimated

according to ISO 13528 (Table 2A) from 47 participants were 1.41  0.05 g/100 g. Out of 47 laboratories, thirty-three (70%) were identified as good performance laboratories (z-score  2), in terms of withinand between-laboratory variations. The majority of the participants had good precision (z-score(within)  2) for ash determination. Values for several laboratories which had a lower degree of precision were identified as outliers for within laboratory variation although the differences in their duplicates compared to the mean were 2.4–3.6%, which are less than the target Horwitz’s RSD (3.8%). Thus, the criteria for the evaluation of precision can be set up at each individual laboratory. The consensus values of ash content in rice sample (Table 3) obtained from 33 participating laboratories with good performance for both within- and between-laboratories, is 1.41  0.04 g/ 100 g (mean  SD), with %CV = 3.0. Minerals—iron and Zn: Rice contains low levels (1–2 mg/100 g) of iron and zinc, variable results on these trace elements submitted by various laboratories can be expected. Forty-five laboratories submitted results for iron and 37 laboratories reported data on zinc. Not every laboratory submitted the methods used and some applied 2 different methods for mineral measurements. Three different methods of sample preparation prior to minerals measurement were used. Nineteen laboratories prepared the samples by dry ashing at 450–550 8C for different periods, with or without acid treatment after ashing. Twelve laboratories applied wet digestion with combination of several acids HNO3/HClO4/ H2SO4/HCl with or without adding H2O2. Seven laboratories used microwave ovens for acid digestion. Atomic absorption spectrophotometry was used by 26 laboratories for quantitative measurement of the minerals. Inductively Coupled Plasma Optical Emission Spectrophotometry (ICP-OES) was carried out for all minerals measured by 16 laboratories. Two of the labs used ICPMass spectrophotometer (ICP-MS) and one applied both ICP-OES and ICP-MS. The robust z-scores of the submitted iron values based on the assigned values as robust mean  SDp (Table 2B) showed that 26 values out of 48 or 54% of the submitted data were identified as satisfactory results (absolute z-score  2). For zinc, based on the robust mean and robust SD derived by ISO 13528 (Table 2A), 26 laboratories out of 39 or 67% were identified as good performance laboratories with satisfactory results. Different methods for sample preparation and measurements were conducted among those laboratories with acceptable robust z-scores. It must be noted that one of the laboratories with good performance (low z-scores, 0.5 and 0.7 for iron and zinc, respectively) applied Instrumental Neutron Activation Analysis (INAA) for both iron and zinc analyses without the step of sample preparation (non-destructive method). This method excludes the loss of minerals and the incompleteness of dissociation of minerals. Therefore, the values obtained could well contribute as assigned values. Two laboratories which applied ICP-OES and/or ICP-MS reported accepted values for iron but low values for zinc. The rest of the outliers (15 laboratories for iron and 7 laboratories for zinc analysis) were random and systematic outliers for withinlaboratory variation and/or between-laboratory variation. Various methods for sample preparation and the measurement of trace minerals were conducted among these laboratories. Thus, no specific methods for sample preparation and measurements can be recommended. In conclusion, the consensus values (as mean  SD) of iron and zinc in rice test material obtained from the laboratories with accepted results are 1.14  0.12 (N = 26, %CV = 10.6) and 1.75  0.16 (N = 26, %CV = 8.8). Vitamin B1: Participants who registered for vitamin B1 analysis were requested to analyse the vitamin within two weeks of receipt

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

of the samples. Seventeen laboratories plus 2 experts from Australia submitted results for vitamin B1. Various methods for vitamin extraction (one lab did not submit information) and determination were applied which could be the reason for a large variation of reported values. Four laboratories included extraction of the vitamin from food samples by dilute acid and one used dilute alkali prior to heat treatment, followed by enzymatic hydrolysis to obtain free thiamin. Five laboratories used acid and enzyme extraction while the other 4 laboratories used only acid extraction. Among these, the lab which extracted vitamin B1 by perchloric acid and analysed it by HPLC-UV after neutralizing the extract reported the lowest results. Three laboratories used liquid–liquid extraction, i.e. methanol and water, which may extract various components other than vitamin B1; two of them measured the vitamin by HPLC with UV detector (which is not specific for the vitamin) and another one used fluorescence spectrophotometry. All of these three laboratories reported high results for vitamin B1. One laboratory used a simple technique of extraction: warm water and protein precipitation with trichloroacetic acid. In the measurement step, a spectrofluorometric method was used by 8 laboratories. Thiamin was first oxidised to form thiochrome and the level of the product was measured. Among these laboratories, two laboratories reported low values and one reported high values of vitamin B1. For those laboratories who applied similar treatment and then measured the thiochrome product by HPLC with fluorescent detection, five out of six submitted results were accepted. There were 5 laboratories who measured the extracted thiamin directly by HPLC with UV detection which is not specific only for thiamin. Two submitted results were accepted. It was surprising that one of them was the laboratory that used simple technique of extraction (warm water and protein precipitation with trichloroacetic acid) which might not be applicable for other types of samples. In summary, the submitted values of vitamin B1 were evaluated based on robust mean (estimated according to ISO 13528)  predicted SD of Horwitz (Table 2B), which was 0.39  0.05 mg/ 100 g (Robust %CV = 13). Ten out of 19 laboratories which included 2 expert laboratories from Australia, presented accepted values of the vitamin. The differences in the methods for sample extraction, rather than the methods of measurement, could explain some of the differences of the vitamin levels. The consensus value (as mean  SD) of vitamin B1 in rice test material, calculated from laboratories with accepted values in both within- and between-laboratory variations, is 0.37  0.05 mg/100 g (N = 10, with %CV = 12.6). Being slightly high in its %CV and limited number of participants with satisfactory results, the consensus value is given for information as suggested value. 3.4. Estimation of En score Data on expanded uncertainty for moisture, protein and iron determination were collected as a trial in this study. Twenty-three participants submitted uncertainty data for moisture, whereas 21 and 15 participants submitted data for protein and iron, respectively. Wide variations of the reported uncertainties for the three components were observed which could be due to the inconsistent estimation of the uncertainty budgets among the participants. A summary of each participant’s data with uncertainty and z-score was graphically presented together with Enscore. An example is shown in Fig. 2A. Although z-score is a useful indicator of laboratory performance, it does not take into account the uncertainty associated with reported results and assigned values. En score takes measurement uncertainty into account and is complementary to z-score in assessment of laboratory performance (NARL, 2004; NMI, 2004). However, in this study, the zscore and En score have not yet been evaluated together. We would

461

like only to show that the uncertainty values must be estimated with good understanding. Over- or under-estimated uncertainty values can directly affect the En score and the interpretation of the laboratory performance. 4. Conclusion A candidate reference material of rice was prepared as fine powder with particle size less than 250 mm. It was proved to be sufficiently homogeneous for representative nutrients – moisture, protein, iron, zinc and vitamin B1 – and stable in terms of vitamin B1 (which is the most labile nutrient in rice) to be used as test material for laboratory performance study. The assigned values of components in the prepared test materials were established following the ISO 13528. The test materials with the assigned values were used to evaluate analytical performance of 62 laboratories. The analyses included proximate composition (except lipid which is a minor component in rice), trace minerals (iron and zinc), and vitamin B1. For proximate composition, good performance on moisture, protein and ash analyses was found among the majority of the participating laboratories (70–80%). Many laboratories still analyse crude fibre instead of TDF, and those data were identified as outlier low values. Conversely, several laboratories reported outliers of high values of TDF, thus an improvement on TDF analysis is still required among these laboratories. Since the levels of iron and zinc in rice are low, several laboratories reported outlier low and high values. Differences in methods used for sample extraction and for measurement of vitamin B1 resulted in discrepancies of results and contributed some problems in estimating the assigned and consensus values. The performance of vitamin B1 analysis among the participating laboratories must be improved. Laboratories whose results were identified as ‘‘unsatisfactory or questionable’’ for any nutrients should determine the possible causes for the disagreeable results and do some corrective action. After evaluation of laboratory performance for nutrients analysed, the consensus values of moisture, protein, TDF, iron, zinc and vitamin B1 in the test material were developed from participating laboratories with accepted z-score values for both within- and between sample variations. The values as mean  SD are presented. Values with high relative standard variation – TDF, vitamin B1 – are given as suggested values. The test materials with consensus values of nutrients become reference materials which can be used as test materials for future laboratory performance study, as quality control samples for internal quality control system, and as test materials for method validation. Acknowledgements The authors would like to acknowledge the National Science and Technology Development Agency (NSTDA), Thailand, for the financial support and the Institute of Nutrition, Mahidol University, for providing facilities to conduct this research. We would also like to acknowledge Dr. Harold Furr, Institute of Nutrition, Mahidol University, for his assistance in editing the manuscript. References AOAC, 2005. Official Methods of Analysis of AOAC International, 18th. Ed., 2005 32.1.02 (method 925.09); 32.1.03 (method 925.10). AOAC International, Maryland, USA. Horwitz, W., Albert, R., Deutsch, M.J., 1992. Precision parameters of methods of analysis required for nutrition labelling. Part II. Macro elements. Journal of Association Official of Analytical Chemistry, International 75, 227–239. ISO 13528, 2005. Statistical Methods for Use in Proficiency Testing by Interlaboratory Comparisons. ISO, Geneva, Switzerland. ISO Guide 35, 2006. Reference Materials—General and Statistical Principles for Certification. ISO, Geneva, Switzerland.

462

P. Puwastien et al. / Journal of Food Composition and Analysis 22 (2009) 453–462

ISO Guide 43-1, 1997. Proficiency Testing by Inter-Laboratory Comparisons—Part 1: Development and Operation of Proficiency Testing Schemes, 2nd edition. ISO, Geneva, Switzerland. ISO/IEC 17025, 2005. General Requirement for the Competence of Testing and Calibration Laboratories. ISO, Geneva, Switzerland. ISO 5725, 1981. Precision of Test Methods—Determination of Repeatability and Reproducibility by Inter-Laboratory Tests, 1st edition. ISO, Geneva, Switzerland. National Analytical Reference Laboratory (NARL), 2004. NARL CRM M1 Pureed Tomato, Australia. National Association of testing Authorities (NATA), 1996. New statistics for NATA’s proficiency testing programmes, Australia. National Measurement Institute (NMI), 2004. Proficiency Study No. 04-04, Australia. Puwastien, P., Sungpuag, P., Chitchunroonchokchai, C., 1989. Report of the ASEANFOODS interlaboratory trial on nutrient analysis. In: Proceedings ASEAN Workshop on food data system. Institute of Nutrition, Mahidol University, Thailand. Puwastien, P., Sungpuag, P., 1995. Interlaboratory study 1993–1994: development of food reference materials. In: Proceedings of the fourth OCEANIAFOODS Conference, Suva, Fiji, 10–12 April, pp. 167–180.

Puwastien, P., Sungpuag, P., Judprasong, K., 1999. External analytical quality control programme for nutrition labeling. Final Report. Institute of Nutrition, Mahidol University at Salaya, Phutthamonthon 4, Nakorn Pathom 73170, Thailand. Puwastien, P., Raroengwichit, M., Sungpuag, P., Judprasong, K., 1999b. Thai Food Composition Tables, first ed. Institute of Nutrition, Mahidol University, Thailand. Puwastien, P., Raroengwichit, M., 2000. Proficiency testing IV. Final Report. Institute of Nutrition, Mahidol University at Salaya, Phutthamonthon 4, Nakorn Pathom 73170, Thailand. Puwastien, P., Pinprapai, N., Judprasong, K., 2001. Laboratory performance study V: main nutrients analysis. Final Report. Institute of Nutrition, Mahidol University at Salaya, Phutthamonthon 4, Nakorn Pathom 73170, Thailand. Puwastien, P., Pinprapai, N., Judprasong, K., Sungpuag, P., 2003. Laboratory performance study VII: analysis of mandatory nutrients for nutrition labelling. Final Report. Institute of Nutrition, Mahidol University at Salaya, Phutthamonthon 4, Nakorn Pathom 73170, Thailand. Torelm, I., 1994. Interlaboratory variance in analysis of major nutrients in foods. Journal of Food Composition and Analysis 7 (1–2), 2–22. USDAdatabase, 2008. USDA National Nutrient Database for Standard Reference— Release 21. .