The Performance of Laboratories Analysing α-quartz in the Workplace Analysis Scheme for Proficiency (WASP)

Ann. occup. Hyg., Vol. 47, No. 4, pp. 69 77, 003 Crown Copyright 003. Reproduced with the permission of the Controller of Her Majesty s Stationery Office. Published by Oxford University Press DOI: 10.1093/annhyg/meg06 The Performance of Laboratories Analysing α-quartz in the Workplace Analysis Scheme for Proficiency (WASP) PETER STACEY*, BARRY TYLEE, DELPHINE BARD and RUSSELL ATKINSON The Health and Safety Laboratory, Broad Lane, Sheffield S3 7HQ, UK Received 9 September 00; in final form 8 November 00 The Workplace Analysis Scheme for Proficiency (WASP) is a proficiency testing (PT) scheme for the analysis of occupational hygiene and environmental air samples and is operated in the United Kingdom by the Health and Safety Laboratory (HSL) on behalf of the Health & Safety Executive (HSE). One of the 6 analytes available to laboratories is silica (α-quartz) on 5 mm Gelman GLA5000 filters. This paper investigates the performance of laboratories participating in the scheme since the HSL took over the production of the samples in 1998. The average relative standard deviation (RSD) of results obtained by a laboratory is 11.5%. This is reduced to 8.5% when the values from laboratories using indirect analytical methods are excluded. Laboratories using indirect analytical methods accounted for some of the most variable data. For the on-filter analytical methods the data suggested a relationship between relative standard deviation and loading that increased gradually from ±4% at high analyte levels to ±10 15% at low levels. The average precision estimate for the on-filter analytical methods was found to be 5.6% RSD for the infrared technique and 6.7% RSD for the X-ray diffraction technique. These figures compare favourably with those reported in the published HSE methods. No significant difference was found between the average result reported by laboratories using on-filter infrared (IR) analysis and the average result reported by laboratories using on-filter X-ray diffraction (XRD) analysis. An ANOVA analysis found the repeatability estimate was just as large as the between laboratory variation for both the XRD and IR on-filter analysis techniques. When a limited number of realistic samples were included in the scheme, XRD analysis was found to perform slightly better than IR analysis. The performance of laboratories in the WASP scheme compares very favourably with other published data from a PT scheme where indirect silica analytical methods are predominately used. Keywords: quartz; workplace air; X-ray diffraction analysis; infrared analysis; analytical performance; proficiency testing INTRODUCTION Silica is one of the most abundant minerals on earth. Free silica exists in a number of crystalline and amorphous forms, the most common of which is quartz, found in a range of minerals including sandstones, grit stones and clays. The quartz content of sandstone can be up to 95% and granite stone may be as high as 30%. Many work activities, such as drilling, chiseling, grinding, polishing of glass, fettling in potteries, demolition of buildings and power saw cutting of *Author to whom correspondence should be addressed. Tel: +44-114-89645; fax +44-114-89544; e-mail: peter.stacey@hsl.gov.uk stone, concrete or brick, result in the generation of respirable silica. Silicosis is commonly accepted to be one of the oldest occupational diseases and crystalline silica is now classified as a category/class 1 carcinogen by the International Agency for Research on Cancer (IARC, 1997). The lung disease is caused by inhalation and retention in the alveoli and generally ranges from nodular silicosis to fibroses of the lung tissue. The position of the Health & Safety Executive (HSE) on the classification of silica is currently under review. In the United Kingdom, the maximum permitted level for exposure to respirable crystalline silica is agreed by the UK Health and Safety Commission and is 0.3 mg/m 3 for an 8 h time- 69

70 P. Stacey et al. weighted average (HSE, 00). Health surveillance of workers is also recommended if exposure to crystalline silica is likely to exceed 0.1 mg/m 3 (HSE, 1997). The Workplace Analysis Scheme for Proficiency (WASP) was established in 1988 and is sponsored by the HSE. Where analytical measurements are required to assess the exposure of a worker, it is important that the analytical performance of the laboratories performing these measurements is assessed independently so that there is confidence in the data. WASP provides a quality assurance service for the chemical analysis of occupational and environmental air samples. It provides a range of 5 analytes to support most analytical techniques used in the analysis of occupational hygiene samples. In 1993, the scheme introduced α-quartz on 5 mm diameter Gelman DM800 filters using a contracted laboratory to prepare the standards, and in 1998 the Health and Safety Laboratory (HSL) took over the production of the filters. Although the laboratories in the scheme are free to choose whatever method they wish, the scheme is specifically intended to support the HSE analytical methods published in the Methods for the Determination of Hazardous Substances (MDHS) guidance documents MDHS 37 (HSE, 1987) for infrared (IR) analysis and MDHS 51/ (HSE, 1988) for the X-ray diffraction (XRD) technique are onfilter analytical methods for quartz where the sample is analysed without any preparation. MDHS 38 (HSE, 1984) is an indirect method where the quartz is recovered from the filter and prepared as a potassium bromide pellet before IR analysis. This paper discusses the analytical performance of laboratories in the scheme, the differences in analytical performance between the XRD and the IR techniques and the effect on analytical performance during the limited introduction into the scheme of a proficiency testing (PT) sample with added crystalline phases. PREPARATION OF α-quartz STANDARDS AND SAMPLE HOMOGENEITY α-quartz standard The standard UK reference material used is A9950 Sikron F-600. This reference material, from Quartzwerke (Frechen, Germany), has a respirable α-quartz content of 96.3 ± 1.4% (Jeyaratnam and Nagar, 1993) when sampled through a cyclone and analysed on a filter. The quantification of the crystalline proportion of this material is based on comparison with the US National Institute of Standards and Technology (NIST) standard: respirable α-quartz, SRM 1878. The mass median diameter of the A9950 Sikron F-600, determined by Coulter counter analysis, is.6 µm and the mass average diameter is.8 µm, with 71% of the distribution below 4.3 µm (equivalent to 7 µm unit-density spheres). Generation of replicate samples The α-quartz aerosol is produced by a modified fluidized bed aerosol generator, and then sampled using a multi-port sampler developed by the National Institute of Occupational Health in Norway (Bjørgo et al., 1980; Anglov et al., 1993). This sampler loads each filter with a similar amount of α-quartz dust. The multi-port sampler consists of a large cylindrical chamber, the lid of which contains 114 ports. Behind each port is a critical orifice that restricts the flow through the hole to 1.9 l/min (±%) when a high vacuum is applied. Each port is fitted with a Higgins Dewell cyclone sampler containing a sample filter. The filters are individually weighed and filter sets are chosen so that laboratories using IR are provided with blank filters that closely match the absorption background of the sample filters. The quartz aerosol is generated into a dust box and then sampled by the multiport sampler. Each experimental run provides one loading of α-quartz onto the filters within the range 60 460 µg. All samples are screened using IR analysis and the typical relative standard deviation (RSD) of the IR measurements on all the samples produced for a single loading is ±5.6%. This also includes the analysis error which is estimated to have a RSD of ± 5% (HSE, 1987). APPROACH TO THE EVALUATION OF THE VARIANCE OF DATA Several different approaches were used to investigate the variability of the data. Within-laboratory variability For each laboratory, the variation of its results from the assigned values was calculated as a percentage relative standard deviation (% RSD). This is the standard deviation of a laboratory s results and will include contributions from the difference of a laboratory s results from the assigned values and the precision of the method or technique used. This estimate was used to compare the performance of individual laboratories and their analytical methods. Total relative standard deviation (% RSD T ) The variation of results obtained by laboratories at each analyte level was calculated as a total percentage relative standard deviation (% RSD T ). This is the standard deviation (σ) of the data at each analyte level divided by the average result (excluding extreme values) reported by laboratories. This estimate contains the variability of results due to differences between laboratories and the within laboratory variability of the different methods or techniques. It can be used to indicate the confidence interval of a result obtained at a particular measurement level, as it is a close estimate of reproducibility sometimes

The performance of laboratories analysing α-quartz 71 referred to as trueness. Its use applied to proficiency testing data is described by Shulman et al. (199). Reproducibility and repeatability The performance characteristics (such as reproducibility) of a particular technique or method are often obtained from duplicate results submitted by different laboratories using the same method on identical test items by using analysis of variance (ANOVA) statistical techniques as stipulated in ISO 575- (ISO, 1994). The ISO method states that the reproducibility variance or trueness is determined as: s R = s L + s r where s R is the estimate of the reproducibility variance, s L is the between-laboratory variance estimate and s r is the repeatability variance estimate. Generally, PT schemes are not able to use this method of evaluation, as they do not usually send samples with the same loading or ask laboratories to report duplicate results; consequently the repeatability variance cannot be separated from the overall variance of the data. This is because PT schemes are designed to test the ability of a laboratory to meet specific performance requirements rather than establish contributions to the overall variance. Fortunately, for α-quartz analysis in this PT scheme, each experimental run provides enough samples for more than one round. Therefore, laboratories made measurements on identical samples in different rounds. Pairing of the results from these samples will allow us to estimate the repeatability and reproducibility variances using ANOVA evaluation. However, the estimates from this approach, applied to these data, will include a between round variation because conditions between rounds (some 3 months apart) may change, e.g. due to varying calibrations, instrumental performance or analysts. Samples used in different rounds were paired if the robust averages of results for each set of samples were within 10 µg. Estimates of s R and s r were determined from an ANOVA evaluation from five pairs of loadings, chosen so they were evenly distributed across the 80 360 µg analytical range. Precision An estimate of within-laboratory precision was made for the analytical range used in the scheme by assuming precision was independent of loading. The precision estimates were calculated by determining the mean error sum of squares of the standardized ratios of the four samples, for each laboratory in each round. This estimate of within-laboratory variance, s W, is comprised of several components: Table 1. Participation in round 49 UK laboratories Non-UK laboratories Type of organization No. of participants Country No. of Participants Industry 4 Australia 1 Central 1 Denmark 1 government Consultancies 1 France 1 Germany 1 Ireland 1 Netherlands 1 Spain 1 USA 1 Total 6 8 S W = p p = 1 σ rp σ Sp ( + ) ------------------------------------- p where σ Sp is the between-sample contribution and σ rp the intra-laboratory contribution for each of the p laboratories participating in each round. Any estimate of σ rp will also contain a small contribution due to sample inhomogeneity. This precision is an estimate of the random error associated with a particular method or technique. DATA COLLECTED Data quality Since the start of the scheme, the number of laboratories analysing samples of α-quartz on filters has fallen from 3 laboratories, to 14 laboratories in round 49. At the beginning of the scheme a larger proportion of laboratories were from industry and the UK; however, by round 49 most are national institutions from Europe. The types of participants in round 49 are listed in Table 1. This work examines the 680 results reported by laboratories since the HSL undertook the production of the α-quartz standards and covers 14 rounds (36 49). A round consists of four samples, each with a different amount of silica dust, sent to laboratories every 3 months. Within-laboratory variation The ratio of each result obtained by a laboratory versus its assigned value was calculated so that the variation of a laboratory s data from the assigned values could be compared with other laboratories. The assigned value used in this scheme is a robust sample average of the results for a sample level reported by laboratories, excluding those results outside prescribed limits. The variation of laboratory

7 P. Stacey et al. sample at 500 C and then recovering the sample by filtering a suspension of the dust through a silver filter. Laboratories B and C obtained results with the largest negative bias. This is illustrated in Fig. 1 as their ratios with the assigned values are distributed well below the ideal (1). Laboratory B used a method where the filter is ashed and prepared as a potassium bromide pellet, and laboratory C used an XRD method where the filter is dissolved with hydrofluoric acid and the dust then transferred to a lowbackground substrate. Laboratories D and E obtained satisfactory WASP performance ratings. Laboratory D had just started the scheme and at the time of writing this paper and had only submitted one set of indirect results with a RSD of %. Laboratory E used a method based on MDHS 38. This laboratory submitted results for nine rounds and the RSD of its results from the assigned values was 3.6%. If the laboratories using indirect analytical methods are excluded, the average within-laboratory RSD is reduced to 8.5% and the median within-laboratory RSD to 8.3%. Fig. 1. Variation of laboratory data compared with the assigned value. results and bias from these consensus values is shown in Fig. 1. Laboratories producing results with very little bias will have their results distributed around the ideal ratio (1). The average within-laboratory RSD was 11.5%, but several poorly performing laboratories unduly influenced this value. The median within-laboratory RSD is less influenced by extreme values, suggesting that a typical variation of results submitted by an individual laboratory was about ±9.4%. The data were examined further to investigate possible reasons for the high RSD values produced by some laboratories. The laboratory with the largest variation of results (39%) had stated it used a direct on-filter analytical method using XRD. When investigating the data it was found that, although the PT samples were analysed without any sample preparation, the amount of quartz on the filter was quantified using a calibration where the standards had been prepared by ashing in an oven at 500 C and the dust transferred as a liquid suspension onto a silver filter. Five laboratories, identified here as A, B, C, D and E, used indirect analytical methods in the rounds investigated in this paper. Three of these five laboratories had within-laboratory RSDs >9.4%. Laboratory A had the second widest spread of results (8%) and used an indirect XRD method based on ashing the Inter-laboratory variation Figure compares the RSD T for all data at each loading level, (excluding extreme results identified using a Grubbs test over the 1% significance level) with the RSD T for the direct on-filter analytical methods. Of all data (684 results), 4.86% were identified as extreme values, and for on-filter analytical methods (516 results), 1.5% were identified as extreme results. A linear trend line was drawn on the chart for all data to allow comparison. This line shows that there was no relationship for the data following the equation (% RSD T )y = 0.0065x + 11.9 (P = 0.061) at the 5% significance level where the slope was not significantly different from zero, and the average RSD T was 11.9%. However, the trend line drawn for the on-filter analytical methods is significant (P 0.01) and suggested a slope where at the highest loading value (400 µg), the RSD T is about ±4% and at the lowest loading values (80 90 g) is about ±10 to ±15%. At levels near the United Kingdom 8 h maximum exposure limit (MEL) of 0.3 mg/m 3 (equivalent to 316.8 µg on the filter assuming a flow rate of. l/min), Fig. indicates the RSD T of results reported by laboratories using on-filter analytical methods is about ±7%. The relationship where the %RSD T increases gradually as the loading decreases is expected when instrumental techniques start to approach their limit of quantification as the reduced signal and increased noise at lower analyte levels reduce the relative precision of the analysis. Any linear relationship with zero or almost zero slope shown in Fig. on the chart for all data would have suggested that the results are substantially above the limit of detection. However, there is a large scatter of values around the line for the

The performance of laboratories analysing α-quartz 73 Fig.. Relationship of %RSD T with loading for all data and for data using direct-on-filter methods. chart for all data, suggesting the trend line drawn on the chart line is not a good indication of the true relationship. The lower the scatter of points around the line, the better is the indication that the laboratories are successfully controlling the errors in analysis. Between-round inter-laboratory variation It was noted that higher %RSD T values occurred in rounds 4, 43 and 45. The variation of average %RSD T of the four samples in each round is shown in Fig. 3, where the average %RSD T figures in rounds 4, 43 and 45 were higher for all data than for the on-filter analysis method users. Until round 4 only one laboratory was using an indirect analytical method and this number gradually increased to three in round 45. The introduction into the scheme of a new laboratory using an indirect method of analysis may account for the higher average %RSD T shown for all data in Fig. 3 because these laboratories would not be familiar with the scheme and their performance is likely to be worse. However, there is another factor that may have adversely affected results from round 4. This round had two sample filters loaded with 100 µg of quartz where a larger variability is expected with methods involving a recovery process because a loss of a small amount of sample is more significant at low analyte levels. Trend in the average round %RSD T The data for laboratories using the direct on-filter methods showed a general reduction of the average RSD T from 13% in round 41 to 4.5% in round 49. The Spearman rank correlation coefficient of average %RSD T against round number, over all rounds, for laboratories using the direct on-filter analytical methods, was 0.71 and the probability of a relationship not existing was <0.01. This correlation coefficient indicates a significant trend in the improvement of performance for laboratories using on-filter methods from round 36. Within-laboratory precision The precision estimates for all results submitted from round 36 are shown in Table. This table gives an indication of the average and median laboratory

74 P. Stacey et al. Fig. 3. Between-round variation of average %RSD T. Table. Precision estimates %RSD Technique All data All on-filter a IR on-filter a XRD on-filter a Average 8.9 6.7 5.6 6.7 Median 4.9 4.8 4.8 4.8 a On-filter: the figures obtained exclude data from laboratories using indirect measurement methods precision over the analytical range used in the scheme. The results obtained for the on-filter methods are comparable with the precision estimates in MHDS 37 and MDHS 51/ as both of these methods report a figure of ±5% RSD for the strongest measurement signals. These data support the figures reported in these MDHS and also the assumption that the variability of silica results between samples within a round is not greatly different from the instrumental variation on a single sample reported in the MDHS. COMPARISON BETWEEN ON-FILTER INFRARED AND ON-FILTER X-RAY DIFFRACTION ANALYTICAL METHODS Difference between the two techniques The average number of laboratories in each round using IR analysis was about four and the average number using XRD was five. The average value of each sample level for each technique in each round was calculated and compared over all 14 rounds. The difference between the two techniques, compared in Fig. 4, followed the relationship y = 0.9675x 0.1558 with a regression coefficient (r ), of 0.97, where y is the average value of XRD analysis and x is the average value of IR analysis. The average difference between the two sets of data was 6 µg and the SD was ±15 µg. A t-test at the 5% significance level indicated that the intercept of the trend line in Fig. 4 with the axis was not different from the origin and therefore a consistent difference is not present. A similar test on Fig. 4. Difference between infrared and X-ray diffraction direct on-filter methods. the slope, also at the 5% significance level, was also not significant, indicating that rotational (alternatively known as proportional) difference was not present. These data showed there was no significant difference between the average of the values reported by the IR techniques and the average of the values reported by the XRD techniques when analysing pure α-quartz standard A9950 on filters. Reproducibility estimates The values obtained from ANOVA of sets of paired data, taken from samples prepared at the same time and analysed by the same laboratories but distributed in different rounds, are shown in Fig. 5. These values suggested that the reproducibility estimates using XRD were more variable than measurements using IR analysis and its relationship with loading was less apparent than the line for IR analysis which followed the equation %RSD = 0.0118x +

The performance of laboratories analysing α-quartz 75 Fig. 5. Reproducibility and repeatability of direct-on-filter methods. 8.74 (r = 0.53). However, this relationship for reproducibility is only significant at the 10% level (P = 0.094). This trend line suggested that the reproducibility between laboratories using the IR technique at 80 µg is about ±8%, and at 50 µg the variation is about ±4%. Conclusions from these data should be regarded with caution because of the small number of laboratories involved in the ANOVA evaluation and the theoretically large uncertainty of the values obtained. However, these reproducibility estimates are comparable with the %RSD T estimates shown in Fig.. Repeatability estimates The linear trend line with the best correlation for the within-laboratory repeatability (Fig. 5) and measurement level was obtained by IR analysis. This line followed the equation y = 0.03x + 10.0 (r = 0.97), and the probability value for a relationship not existing was <0.01. These data suggested a strong relationship between loading and repeatability of about ±8% at the lowest loadings and ±4% at the highest loadings. The repeatability estimate for IR analysis is comparable with the figure reported in MDHS 37, where the precision of IR measurement on a single sample of 50 µg is given as ±5%. At the lowest level (88 µg), the repeatability estimate is ±8%. The figure reported in the MDHS is the instrumental variation of the measurements obtained on a single sample and the small additional variability found in this paper may be due to several factors. Not all laboratories may use exactly the same instruments, quartz standard or procedures. Also, conditions between rounds (some 3 months apart) may also change (e.g. instrumental and personnel) and this may introduce additional variability. The repeatability estimate for XRD measurements were between 4 and 10%, and their relationship with loading was much more variable (P = 0.31). In all cases the repeatability estimate was just as large as the estimate of between-laboratory variation. THE EFFECT ON PERFORMANCE WHEN SAMPLES WITH ADDITIONAL COMPONENTS ARE INTRODUCED Over a period of three rounds, two different types of dust were sampled onto the filters and sent to laboratories to ascertain if the performance of laboratories was affected when more realistic samples are analysed. An artificial mixture of pure calcite and quartz was sampled to produce two sets of replicate samples. Each set had a different amount of α-quartz dust. One sample from each set, was distributed to 1 laboratories in rounds 39, 40 and 41. The dust created by cutting a concrete paving slab with a powered saw was also sampled to produced another set of replicate

76 P. Stacey et al. Table 3. Variability of data from samples with added crystalline dusts Round 39 Round 40 Round 41 Average % RSD T Average % RSD T Average % RSD T Calcite mixture 1 1.3 µg 11.7 13.9 µg 14.5 06.7 µg 11 Calcite mixture a 65 µg 10.5 66.7 µg 10.9 7.7 µg 1.3 Concrete sample b 100 µg 13.8 110 µg 0.5 a Data excludes one laboratory with consistently low results in rounds 40 and 41. b Data excludes one laboratory from round 40 with very low results. Table 4. Comparison of data Performance on WASP test Samples Calcite mixture with 11 µg quartz Calcite mixture with 68 µg quartz Concrete dust with 105 µg quartz PAT scheme at 100 µg NIOSH intercomparison % RSD T XRD 1 8 7 11 9 16 30 15 3 IR 8 4 1 1 17 16 30 13 Precision %RSD XRD 10 6 5 7 3 7 10 IR 8 4 1 13 13 7 10 filter samples with a different type of dust. X-ray diffraction analysis of the paving slab identified calcium hydroxide (Portlandite) Ca(OH ), calcite CaCO 3, quartz SiO, and a trace of a feldspar mineral, possibly albite NaAlSi 3 O 8. A single sample of this type was sent to 1 laboratories in rounds 40 and 41. Variability of data The values of total laboratory variability (%RSD T ) in Table 3, from laboratories analysing quartz in the calcite and quartz mixtures, were comparable with the values of %RSD T obtained from laboratories when they analysed the usual WASP samples (Fig. ). The data for the analysis of quartz in the concrete dust was more variable. One of the samples with the concrete dust had a total variability (%RSD T ) of ±0%. The data exclude the only laboratory in these rounds using an indirect analysis method. Its results for the calcite and quartz sample mixtures were about half those reported by the other techniques, but the results for the samples with the concrete dust were very close to the average results reported by other laboratories. One difference between the two types of samples is that the deposit on the filters from concrete dust was slightly coloured, unlike the dust from the artificially generated calcite and silica mixtures. This may have helped the laboratory recover the sample more effectively. Table 4 shows values of RSD T and precision RSD for the laboratories using on-filter XRD and IR analysis. When the realistic samples were analysed, the five laboratories using on-filter XRD analysis obtained slightly better RSD T and precision than the four laboratories using on-filter IR analysis. The values for RSD T and precision were also better for XRD when compared with values reported from the analysis of the usual WASP samples where only quartz reference material is present. It is expected that XRD analysis performs better when real samples are analysed, because the background noise caused by the filter is reduced when other non-amorphous dusts are also present. This improves the peak-to-background noise ratio, which is important in quantification. As expected, with a realistic sample containing a trace silicate (feldspar) mineral, the measurements with IR analysis were more variable than the measurements with XRD analysis since silicate minerals interfere with IR analysis. On-filter XRD analysis obtained a very good RSD for precision with this sample. Comparison of %RSD T with the values obtained from other PT schemes The American PT scheme for occupational hygiene analysis, the Proficiency Analytical Testing Program (PAT) managed by the American Industrial Hygiene Association (AIHA), provides samples of silica dust on filters with coal dust, calcite or magnesium silicate as an added component. The most common methods used in the USA are the National Institute for Occupational Safety and Health (NIOSH) methods 7500 for XRD analysis (NIOSH, 1998) and 760 or 7603 for IR analysis (NIOSH, 1994a,b). All these are indirect methods where the filter is removed by treatment and the sample is transferred to a material that produces a lower background than the filter. Table 4 compares the RSD T and the precision RSD for XRD and IR analysis reported by the PAT scheme (Shulman et al., 199) and a NIOSH inter-comparison (Anderson, 1983) with the values obtained from laboratories using the on-filter analytical methods in

The performance of laboratories analysing α-quartz 77 the WASP scheme. The RSD T values for the WASP scheme are better than those reported by PAT (Eller et al., 1999) or the NISOH laboratory inter-comparison. However, the PAT scheme involves 80 laboratories while the WASP scheme has a much smaller number. In the WASP scheme only five laboratories were using on-filter XRD and four were using onfilter IR analysis. The limited number of on-filter XRD analysis data was also more precise than the data reported by the inter-comparison organized by NIOSH to test their methods with samples simulated to represent real matrices. However, the precision for the on-filter IR analysis on the realistic WASP samples was slightly poorer. The limited amount of WASP data suggested that over the analytical range used in the WASP scheme there is little advantage in using a NIOSH type method. CONCLUSIONS The average within-laboratory RSD for laboratories using direct on-filter analytical methods was ±8.5% Some laboratories using indirect analytical methods with a sample preparation step produced the most variable data. The data suggests that for on-filter methods of analysis, there is a relationship between loading and the total variation of results between laboratories (RSD T ) that is about ±4% at analyte levels of 400 µg and about ±10 to ±15% at analyte levels of 100 µg. The total variability (RSD T ) of results between laboratories involved in the WASP scheme, using on-filter analytical methods, analysing quartz and producing results at the MEL for crystalline silica is about ±7%. Estimates of precision compared very favourably with those published in the MDHS guidance documents 51/ and 37. There is no significant difference between the average result of laboratories using direct on-filter IR analysis and the average result of those laboratories using direct on-filter XRD analysis when analysing WASP proficiency test samples. However, the data suggested that on-filter XRD analysis produced slightly more variable data than on-filter IR analysis. The performance of laboratories analysing samples loaded with a mixture of calcite and α-quartz was similar to the performance of laboratories analysing the usual WASP samples, where only α- quartz is present. The data suggests that the variation of results between laboratories (RSD T ) and precision RSD of on-filter XRD analysis is slightly better than on-filter IR analysis when real samples are analysed. The filter sample loaded with concrete dust containing a trace feldspar mineral showed that (in these circumstances) direct on filter IR results are more variable than on-filter XRD results. The data from the on-filter analytical methods compared very favourably with other published data from indirect analytical methods. REFERENCES Anderson C. (1983) Collaborative tests for determining free silica in airborne dust. Cincinnati, OH: US Department of Health and Human Services (NIOSH) 83-14. Anglov J, Holst E, Dyg S, Christensen J. (1993) Iron, manganese, copper, and titanium in welding fume dust on filters for internal and external quality assurance. Fresenius J Anal Chem; 345: 335 9. Bjørgo R, Fjeldstad P, Nagy K, Thommassen Y. (1980) Parallelfiltre fra arbeidsatmosfæren i aluminiumsindustrien for interlaboratoriekontroll av analyse på PAH og flurid. Oslo: Institute of Occupational Health. Eller P, Feng A, Song R, Key-Schwartz R, Esche C, Groff J. (1999) Proficiency Analytical Testing (PAT) silica variability 1990 1998. Am Ind Hyg Assoc J; 60: 533 9. HSE. (1984) Method for the determination of hazardous substances (MDHS) 38, Quartz in respirable airborne dusts: laboratory method using infra-red spectroscopy (KBr disc technique). London: HSE Books. ISBN 011885694. HSE. (1987) Method for the determination of hazardous substances (MDHS) 37, Quartz in respirable airborne dusts: laboratory method using infra-red spectroscopy (direct method). London: HSE Books. ISBN 071760087. HSE. (1988) Method for the determination of hazardous substances (MDHS) 51/, Quartz in respirable airborne dusts: laboratory method using X-ray diffraction (direct method). London: HSE Books. ISBN 0118856413. HSE. (1997) Respirable crystalline silica. Guidance Note EH59, nd edn. London: HSE Books. ISBN 071761438. HSE. (00) EH40, Occupational exposure limits. Guidance Note. London: HSE Books. ISBN 07176083. IARC. (1997) Monographs on the evaluation of carcinogenic risks to humans and their supplements. Lyon: IARC. ISBN 9831681. Vol. 68, pp. 149 4. ISO. (1994) Accuracy (trueness and precision) of measurement methods and results, ISO 575: Part. Geneva: ISO. Jeyaratnam M, Nagar N. (1993) Comparison of α-quartz standard Sikron F600 (HSE A9950) with the NIST respirable α- quartz standard SRM 1878 for bulk and on-filter analysis. Ann Occup Hyg; 37: 167 79. NIOSH. (1994) Silica crystalline by IR, method 760. Manual of Analytical Methods. Issue. Cincinnati, OH: NIOSH. NIOSH. (1994) Silica in coal mine dust, method 7603. Manual of Analytical Methods. Issue. Cincinnati, OH: NIOSH. NIOSH. (1998) Silica, crystalline by XRD, method 7500. Manual of Analytical Methods. Issue 3. Cincinnati, OH: NIOSH. Shulman S, Groff J. Abell M, (199) Performance of laboratories measuring silica in the proficiency analytical testing program, Am Ind Hyg Assoc J; 53: 49 56.