Annals of Occupational Hygiene Advance Access originally published online on January 24, 2005
Annals of Occupational Hygiene 2005 49(4):309-324; doi:10.1093/annhyg/meh100
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Published by Oxford University Press (2005);
Original Article |
A Comparison of Fibre Counting Across Three European National Proficiency Testing Schemes
1 Institute of Occupational Medicine, UK; 2 Instituto Nacional de Seguridad e Higiene en el Trabajo, Spain; 3 Laboratory of FPS Employment, Labour and Social Dialogue, Belgium; 4 Health and Safety Laboratory, UK
* Author to whom correspondence should be addressed. Tel: +44 (0) 131 449 8041; fax: +44 (0) 870 850 5132; e-mail: alan.jones{at}iomhg.org.uk
ABSTRACT
Proficiency testing (PT) schemes seek to improve and maintain consistency in asbestos fibre counting by circulating mounted samples to laboratories, which return counts for comparison with consensus reference values. This study compares the level of these consensus values in three (Spanish, Belgian and UK) national PT schemes. It also assesses the effect of the imminent change in the European standard rules for counting asbestos fibres, to the new rules from the World Health Organisation (WHO), which will include more fibres. Forty samples from the three schemes were circulated to six laboratories, two in each national scheme. The UK and Spanish laboratories counted using both the ERM and the WHO method. Belgian laboratories already use the WHO method, and they counted only by this method. Densities from counts in this sample exchange served as a common basis against which we compared the national reference values (R). This produced a geometric overall mean ratio of R to exchange mean density of 0.94 for the UK scheme, 1.01 for the Spanish and 0.97 for the Belgian scheme, and thus indicated remarkably similar levels. Nevertheless, non-trivial systematic inter-laboratory differences confirmed the need for regular PT and international inter-laboratory comparisons. Examination of previous data from the national PT schemes and from an international scheme (AFRICA) provided further comparisons for the same laboratories (either comparing the laboratory's data with R values or making direct comparisons between laboratories) that are consistent with those from this special exchange. The change in fibre counting rule (to include fibres apparently touching particles with diameter >3 µm) produced mean increases in reported densities that ranged among samples from 0 to 70% (highest individual estimate 170%); the effect was broadly similar for the three schemes. The laboratories gave mostly similar estimates of percentage increase; however, one laboratory occasionally produced high estimates indicating a possible need for training for using the new method.
Keywords: Asbestos fibre counting • proficiency testing • international comparision • European national PT schemes • European and international harmonisation
INTRODUCTION
Background
The measurement of airborne asbestos fibre concentrations involves the counting of fibres collected on filters and mounted on glass slides, by phase contrast microscopy following a European standard method (HSE, 1995
). Variation arises in the results because only a sample of the filter area is examined, but this variation tends to be small in relation to other factors such as the proficiency of the analyst in discerning fibres that may be only just visible, and whose visibility may depend on fine adjustment of the focus of the microscope. Such variation has been shown in exchanges of samples between laboratories (e.g. Beckett and Attfield, 1974; Crawford and Jones, 1980; Carton et al., 1981). To help control and contain operator-related variations, many laboratories that undertake counting of fibres participate in proficiency testing (PT) schemes. Laboratories assess standard slides, and their results for each slide are compared to a reference value. Typically, the reference value for each slide is a consensus value based on data generated by the laboratories within the PT scheme.
Many countries now operate such a PT scheme at a national level (e.g. as described in Crawford and Cowie, 1984; Schlect and Shulman, 1986; Arroyo, 1990, 1991; Kauffer, 1990, 1992; Crawford et al., 1992) and the reference values for each scheme are based on counts produced nationally. While these schemes have demonstrably brought the performance of laboratories closer together nationally, less attention has been paid to possible variation in counting levels between countries. There have been some limited international exchanges of samples (e.g. Walton et al., 1976; Crawford and Cowie, 1992), and there are ongoing regular international comparisons, such as the Asbestos Fibre Regular Informal Counting Arrangement (AFRICA) operated by the Institute of Occupational Medicine (IOM) and whose participants include some laboratories that organize national PT schemes (Crawford and Brown, 1992; Brown and Jones, 2001). However, the AFRICA exchange gives only a picture of comparability between a few key laboratories. The counting levels in the AFRICA exchanges have been compared with those in one participant's national PT scheme (Arroyo and Rojo, 2001).
There is an increasing awareness that fibre counting performance should be harmonized internationally. Arroyo and Rojo (2001) have suggested adjustments to the national criteria for satisfactory performance to make them more consistent in their stringency. Those suggestions relied on two assumptions:
- •that the variability in counting is essentially common to all laboratories (more specifically that there is no reason to suppose that it should differ between countries), and therefore the risk of failure to meet the criteria should be made equal in all national PT schemes; and
- •that the target counting levels, i.e. the reference values, are consistent between national schemes.
In this study, we examined the comparability of reference values for a selection of slides from three national fibre counting PT schemes, by circulating them for assessment by two representative laboratories from each scheme. The schemes that participated were the UK Regular Interlaboratory Counting Exchanges (RICE), the Spanish Programa Interlaboratorios de Control de Calidad de Fibras de Amianto (PICC-FA) and the Belgian scheme.
Another issue for harmonization is the adoption of uniform rules and definitions for performing the assessment. In the near future, in the UK and Spain, laboratories will change from using the current European Reference Method (ERM: see MDHS39/4, HSE, 1995) to the WHO all-fibre counting method (WHO, 1997). Fibres appearing to touch particles (with diameter >3 µm) are excluded under the current method but included under the new method, which should increase counts on slides that include particulate deposits, and may decrease variability caused by uncertainities in deciding whether a fibre appears to be touching a particle. Belgium is, in effect, already using the WHO method for asbestos fibre counting.
Objectives
The main objective of the current study was to compare the target counting levels being set in three national PT schemes for asbestos fibre counting, by circulating 40 reference slides, drawn from the three schemes, to six laboratories (two from each national scheme). By comparing their counts with each other and with the reference counts, we aimed to:
- compare the levels of national reference values on asbestos samples; and
- assess any influence of sample type or origin on international agreement.
THE NATIONAL FIBRE COUNTING PT SCHEMES
General characteristics of the schemes
The three national PT schemes for asbestos fibre counting operate under similar general principles. In each scheme, batches of samples are circulated to groups of laboratories. Laboratory performance is classified on the basis of 
30 slides: 32 slides in the UK and Spanish schemes (Arroyo and Rojo, 2001), and 30 slides in the Belgian scheme. Classification as satisfactory depends on 75% of counts being within target ranges for each sample (e.g. within the range 0.65–1.35 times the reference value for the Spanish scheme). There are differences in the organizational arrangements, but these should have little influence on the comparability of the schemes.
In the UK, the RICE scheme has been operating for almost 20 years and had (at the time when this exercise was undertaken) completed nearly 60 circulations (of batches of eight reference samples), and its membership stood at 190 laboratories. At its peak, membership had been as high as 320 laboratories, but a pattern of consolidation into larger laboratories has led to fewer laboratories but almost the same number of analysts. In the RICE scheme, reference values (expressed as a density of fibres on the sample, in fibre mm–2) are calculated as the median of the densities from at least 15 counts from RICE laboratories.
In Spain, the PICC-FA scheme has been operating since 1988. At the time of this exercise, it had completed 28 circulations (with sets of 16 to 20 samples) and its membership stood at 25 laboratories. Reference values are calculated as means of the densities from all available counts returned by the member laboratories, after exclusion of outliers; i.e. a trimmed mean. This calculation procedure produces values close to the median.
In Belgium, the national scheme has been operating since 1988 and, at the time of the exercise, had completed 42 circulations (of batches of 10 slides). Membership stood at 25 laboratories. The final reference values are medians of the densities from counts from all the laboratories in the scheme.
The number of counts used to produce a reference value is of the same order: at least 15 (for the UK and Belgian), generally 25 for the Belgian and Spanish, and sometimes many more than 15 (e.g. 60 counts) for the UK RICE samples.
Main distinguishing characteristics of the national schemes
Reference values
In the Spanish and Belgian schemes, laboratories are given immediate interim feedback based either on target values estimated for each slide or on comparison with the data for other laboratories in their group. Only at the end of each year, after all the samples have been counted by every laboratory, are the final reference values calculated and the final calculations made to assess the performance of each laboratory. In contrast, in RICE all the samples have final reference values and the immediate feedback is based on these final reference values. Information based on the final reference values may be more helpful to participants (e.g. in enabling them to take appropriate action earlier), whereas the other system enables new samples (with limited initial data) to be used as test samples.
Given the differences in sizes of the schemes, there are inevitably some differences in the way that reference values are derived from the returned data. Reference values are derived for the larger RICE scheme from subsets of the participating laboratories, whereas in the smaller schemes they are derived from counts from all the laboratories. Each subset (of 10% or more of the RICE laboratories) may have some bias in their counting level compared with the overall average, and that would be a factor to consider if we seek to establish the equivalence of the performance criteria, but the average level of reference values (over the 16 samples in this exchange) would be a good estimate of the average for the whole population of RICE laboratories.
Reference slides and target ranges
The UK (RICE) scheme has samples with a wide range of densities, from 0 to >600 fibres mm–2. The Spanish scheme has mostly samples with density >64 fibres mm–2. The Belgian scheme has mainly lower density samples; the Belgian samples in this exercise all had density <200 fibres mm–2. The samples in this exchange from the Spanish and UK schemes are also typical of the range of densities in those schemes.
Where the sample densities are high, counting stops at a fixed number of fibres (100 fibres) and then the variation (in estimated densities for a given sample) is expected to be approximately constant on a log scale and therefore the target ranges are expressed as proportions of the reference density. At lower densities (<64 fibres mm–2), the counting stops at a fixed number of fields (200) and as a consequence the variation in the estimated densities is expected to be approximately constant in the scale of square root of density, and therefore the target ranges are defined in terms of proportions of the square root of the density.
Performance criteria
The criteria, in the three national schemes, for assessing the performance of participating laboratories are similar in that all three require 75% of counts (on 32 or 30 samples) to be within defined target ranges around the reference value (R). These target ranges at present differ between the national schemes: 0.65–1.35R (Spanish PICC-FA); 0.67–1.67R (Belgian); and 0.5–2.0R (UK RICE) (Arroyo and Rojo, 2001).
METHODS
Inter-scheme sample exchange
The inter-scheme sample exchange was conducted during 2002. Forty reference samples were provided by the three national schemes, and were combined into four batches containing approximately equal mixes of samples from each scheme. The four batches of 10 samples were circulated to and counted by the six laboratories and their counts were returned to the IOM.
Initially, all the samples were counted at IOM. For each field of view, (i) the fibres meeting ERM rules were counted, and (ii) the extra fibres included under WHO rules were counted before moving on to the next field of view. This gave a direct indication of the difference in count on the same fields. These initial counts were produced on fewer fields (30) than would be normal for standard counting rules, and are referred to as ‘screening’ counts to distinguish them from counts on the full number of fields. The consistency of the data from these screening counts with normal counts from the same laboratory was examined (as described in Appendix 1) before the screening counts were included in the main analysis.
For the exchange, laboratories were asked to count the samples in the same way that they would count samples in their own national PT scheme. This was intended to produce counts that would be comparable with those produced in the national PT schemes.
The core of the design was that every laboratory would produce at least one count on every sample by its normal counting method. In addition, those laboratories that currently use ERM would produce estimates of the number of extra fibres that would be counted under WHO rules. The two laboratories (the Belgian laboratories) that already routinely use the WHO methods did not attempt to produce ERM counts. In some laboratories, more than one microscopist counted each slide, but the selection of microscopists was not specified under the study design and was decided on site (as normally occurs in commercial practice).
Table 1 summarizes the numbers of counts produced by each laboratory, by each method. The main features are noted and explained below.
- The Belgian laboratories, as they already use WHO counting rules, produced counts only by that method, either one count per sample (Belgian lab 1) or three counts per sample (Belgian lab 2).
- The UK and Spanish laboratories currently use ERM rules but expect to change over to WHO rules within the next couple of years. Therefore, they produced ERM counts on all samples, and in some cases more than one ERM count per sample. These four laboratories also produced WHO counts, but with the following variations.
- The IOM produced WHO counts as part of the screening counts, as described above.
- HSL produced ERM and WHO counts on the same fields (i.e. for each field, the fibres under ERM rules were counted and then the additional fibres countable under WHO rules were counted before moving onto the next field) for three batches (30 samples), but then produced WHO and ERM counts as separate counts for one batch (10 samples).
- Spanish lab 1 produced WHO counts on 20 samples (counting ERM fibres and extra WHO fibres on the same fields), and two ERM counts on all samples, and hence a total of 100 counts on the 40 samples.
- Spanish lab 2 produced an ERM count and a WHO count on each sample on the same fields.
- The IOM produced WHO counts as part of the screening counts, as described above.
|
The numbers of analysts producing the counts varied between laboratories:
- A single analyst produced all counts in two laboratories (HSL and Spanish lab 2), and the ‘screening’ counts at IOM.
- Counting was shared among several analysts at the other four laboratories, ranging from 14 analysts (Belgian lab 2) to four analysts at (IOM) and two analysts at each of the other two laboratories (Spanish lab 1 and Belgian lab 1).
Samples
The samples supplied from the national schemes comprised
- from the UK (RICE) scheme:
- Eight samples obtained from either asbestos removal or clearance testing, with mostly amphibole asbestos fibres being reported, and (national reference) densities between 7 and 70 fibres mm–2 for all but one sample with a value of 135 fibres mm–2;
- eight samples containing chrysotile from the asbestos textile industry and with (national reference) densities ranging from 89 to 509 fibres mm–2;
- Eight samples obtained from either asbestos removal or clearance testing, with mostly amphibole asbestos fibres being reported, and (national reference) densities between 7 and 70 fibres mm–2 for all but one sample with a value of 135 fibres mm–2;
- from the Belgian schemes:
- Eight samples supplied by participants without specification of the origin but understood to come from clearance sampling, sampling around enclosures or near entrances to enclosures, exposure measurement during maintenance work on materials containing asbestos and similar activities;
- Eight samples supplied by participants without specification of the origin but understood to come from clearance sampling, sampling around enclosures or near entrances to enclosures, exposure measurement during maintenance work on materials containing asbestos and similar activities;
- from the Spanish PICC-FA scheme,
- sixteen samples obtained by sampling from airborne dust generated by a fluidized bed aerosol generator (TSI) loaded with asbestos cement dust.
- sixteen samples obtained by sampling from airborne dust generated by a fluidized bed aerosol generator (TSI) loaded with asbestos cement dust.
Data analysis
Analysis of data from fibre-counting exercises needs to take account of the fact that the estimated densities differ in precision, for multiple reasons. If the fibres were distributed on a slide at random and there was no subjective error in the counting process, then the variation in the counts would be expected to follow the Poisson distribution. However, the rules for deciding how much of a slide to assess also affect the precision of the result. Miller (1984) noted that, where counting was stopped at a constant number of fibres, precision would be approximately constant if densities were compared on the log scale. At lower densities, where counting would continue to a fixed number of fields, the log scale would be expected to overcompensate somewhat.
In practice, densities estimated by phase-contrast microscopy tend to show more variation than those implied by the Poisson distribution, but by observation the log transform still works well to standardize variation. Therefore, we have based simple summary comparisons within this paper, particularly in graphic presentations, on comparisons of means of log values, which yield geometric mean ratios on the scale of measurement. We use graphical comparisons, with geometric means, to relate the three national scheme reference values to the density estimates from the exchange and thus infer how they compare with each other.
For the formal statistical analysis, the imbalance (evident from Table 1) in the data set, was accommodated by fitting a generalized linear model (McCullagh and Nelder, 1989) that was appropriately customized. Adaptation of this model to densities from fibre counting assumes that the random variation is proportional to Poisson variation, with a constant of proportionality >1.0 to be estimated from the data, and that systematic differences can be described as ratios, i.e. constant or linear on the log scale (Miller, 1984). Results are weighted in the analysis in proportion to the area of the slide examined. Fit of the model is quantified by deviance, which is a generalization of a weighted sum of squares, and the statistical significance of terms added to the model is judged by the ratios of mean deviances, which follow approximately F-distributions with the appropriate degrees of freedom. A term which explains little of the variation in the data will have a mean deviance close to that for the unexplained residual. The order of fitting terms can affect, sometimes considerably, the mean deviance accounted for by a term, since each term fitted is adjusted for all those already in the model. An analysis of deviance is thus a generalization of the analysis of variance for standard linear models in an unbalanced data set. We use the formal analysis of deviance to establish and assess the factors that may influence the observed relationships.
RESULTS
Comparison of reference values
Each national scheme has reference values assigned to its samples that are consensus values from the counts produced within that national scheme. The main purpose of this study was to compare the relative levels of reference values in the three national schemes, and we seek to do that by comparing them to the density estimates produced in this exchange. If there were any important differences between the national scheme reference values for the Spanish, Belgian and UK schemes, then that should show up as a difference in their relationship with counts made by a common group of laboratories.
As the UK and Spanish samples have national scheme reference values based on ERM counts, those national scheme reference values (R) are compared to the ERM counts made in the exchange. The Belgian scheme national reference values are based on WHO counts and therefore are compared to densities from the WHO counts in this exchange.
In Fig. 1, the national scheme reference values are plotted against the individual estimates of density from this exchange, and against the geometric means of the density estimates for each slide. Each slide has one national scheme reference value. There are between one and three counts from each laboratory on each sample, with the symbols identifying each laboratory. In accord with previous experience, the spread in density estimates is approximately constant on the log scale. The individual counts mostly lie within a factor of 2 of the diagonal line denoting equality, and the variation appears broadly similar in the three graphs.
|
The geometric means cluster very closely around the line of equality. The mean ratio of the national reference value (R) to the geometric means is plotted as a solid line, and in each case is almost coincident with the dotted equality line.
The national scheme reference values for the UK (RUK) and for Spain (RSp) relate to the geometric mean densities (DGM) from the ERM counts in the exchange:
 |
 |
5%. For the Belgian samples, there are WHO counts from all six laboratories (whereas the ERM counts were produced by just four laboratories, the Spanish and the UK laboratories). Every Belgian sample had the following WHO counts:
- one count from each of four laboratories (the two UK laboratories, Spanish lab 2 and Belgian lab 1); and
- three counts from the second Belgian laboratory.
We compared the WHO national reference values (for the Belgian samples) with geometric means of the densities from WHO counts in the exchange using primarily the data produced by the same four laboratories that had produced the ERM counts for the Spanish and UK samples. These results are shown in Fig. 1. We also examined the data with inclusion of the counts from the two laboratories in the Belgian scheme. This produced very similar estimates of the relationship between national reference value and densities from the exchange, and those results are shown in Appendix 2.
Based on the WHO counts from the four laboratories that contributed counts to the ERM comparison (i.e. the Spanish and UK laboratories), RBelgian = 0.967·DGM.
Table 2 summarizes these mean ratios relating the reference values for all three national schemes to the appropriate ERM or WHO count geometric means. All the mean ratios are very close to 1. Furthermore, the three different bases for the mean ratios for the Belgian scheme samples might have substantially affected the results, but actually made very little difference.
|
Inter-laboratory differences
Table 3 summarizes the results of an analysis of deviance carried out on the returned density estimates. ERM and WHO counts were analysed separately.
|
Statistical theory tells us that, if the counts of fibres in separate fields of view are distributed as Poisson, then, after systematic sources of variation are accounted for, the residual mean deviance has an expected value of 1.0. Typically, for fibre counting data, we observe residual mean deviances between
2 and 9 (Miller, 1984), and the results in Table 3 are in line with this expectation. This suggests some overdispersion in the counts, due partly to the choice of microscopist within laboratory, and partly to random variation e.g. daily fluctuations in a counter's performance. Although the identity of the microscopist was recorded for each assessment, which gives scope for some investigation of the operator variation, we include that as part of the normal within-laboratory variation. In Table 3, the ‘origin’ and ‘slide’ terms for the densities reflect the known range of density between individual slides and between groups of slides from the three different origins (schemes).
The ‘lab’ term is large enough to indicate that there were non-trivial differences in counting level between these laboratories, whereas the terms for the interaction between laboratory and origin, or laboratory with slide, have much smaller deviance ratios than the lab term, suggesting that the relative level of the laboratories was not greatly affected by the source of the samples. This was confirmed when the geometric mean density (across slides) for each lab was calculated, for slides grouped by origin and for all slides, giving the results in Fig. 2. The laboratories' levels, relative to each other, did not appear to depend on the origin of the slide.
|
Table 3 also shows (in the rows in italics) the mean deviance associated with systematic differences between microscopists within laboratories and the remaining residual [residual (2)]. This mean deviance for microscopists was less than twice that of the remaining residual for ERM counts, and just over four times the residual for the WHO counts, suggesting some systematic differences between counters within laboratories. However, in routine fibre counting, differences between microscopists are not adjusted for, and hence form part of the measurement error. Therefore, in what follows, microscopist differences are not identified explicitly, and other effects are compared with the residual (1) that includes microscopist effects.
The residual mean deviances were only slightly different between ERM and WHO methods.
Laboratory counting level relative to reference densities
Since the study's primary aim concerns the relationship of the national reference density to the densities produced by the laboratories in the exchange, it is natural to analyse the ratio of the laboratories reported density to the reference density. The counts in the exchange need to be compared to reference values for the same counting method (WHO for the Belgian samples, ERM for the Spanish and UK samples).
The mean ratios are summarized in Table 4. For the WHO counts, there are results for six laboratories, and here the range between highest and lowest ratios is greatest (0.86–1.44), with the highest value being for a laboratory that normally uses the ERM method. This high value (1.44) compares with ratios of 1.06 and 1.10 for the laboratory that counted by ERM, so it might be associated with the change rather than the normal laboratory performance. For the ERM counts, the spread in ratios between laboratories was very similar for the Spanish and UK samples.
|
Fibre-counting method
The counts, by ERM and WHO methods, were compared to assess the difference attributable to the counting method. We examined the extent to which the effect:
- was consistent between laboratories;
- varied between samples, and in particular:
- depended on sample density;
- depended on either the sample source (Spanish, Belgian or UK) or sample type (asbestos cement, textile or clearance/removal).
- depended on sample density;
Table 1 summarizes the number of counts made by each laboratory; and the details are described in the Methods section. The ratios of WHO density to ERM density for each of the 40 samples are shown in Fig. 3, as geometric means, individual counts and range bars.
|
For many samples, the results for different laboratories are so closely grouped as to be difficult to distinguish in Fig. 3. However, Spanish lab 2 appeared to produce a higher share of the high ratios. This was confirmed by taking geometric means over all samples counted by each laboratory (Table 5). The mean ratios for samples from each scheme (1.28, 1.20 and 1.14) suggest that if these samples are representative of the overall stock of each of the schemes, then the change in rule would have a similar average effect (on counts) for each scheme.
|
In Fig. 3, there is no strong dependence on density of fibres on the filter (on average or for any individual laboratory); except that a low density sample (at 9 fibres mm–2) showed the highest estimate of percentage increase, and a wide range of results. Such variability might arise from the few fibres being counted at that density (
15 fibres in 200 fields).
The ratio is expected to vary from sample to sample because (i) a change can occur only if a sample has particles with diameter >3 µm, (ii) chance overlap of fibre and particles becomes increasingly likely with more particulate on a sample (as described by Iles and Johnston, 1983) and (iii) the particulate content was variable among samples. The geometric standard deviations (in Table 5) reflect the differences in average ratios between samples. Figure 3 shows the range of individual estimates and the mean estimate of the ratio for every sample.
Level of the participating laboratories in their own national PT scheme
Having compared the counts made by each laboratory with the national reference values (Table 4), we examined the extent to which the level of counts in this exchange was consistent with each laboratory's own counts in the recent data from the national schemes. (Each national scheme provided data for its two laboratories.) The number of samples counted by each laboratory per year in the national scheme data ranged from 10 to 16 for the Spanish laboratories, 25 for the Belgian laboratories and 30 for the UK laboratories. In this exchange, the laboratories counted 16 Spanish PICC-FA samples, eight Belgian samples and 16 UK RICE samples.
Figure 4 plots the geometric mean of the ratio of the laboratory's counts to national scheme reference values in previous years of each national scheme, and also the corresponding value for the samples (from its own national scheme) in this exchange. The results show that the counts in this exchange are generally consistent with the level of counts made previously in the national scheme exchanges. The within-laboratory variation from year to year is of similar magnitude to the variation from the level of the counts in this special exchange. For example, the Belgian laboratories, counting just by their normal method, show some variation from year to year, such that their levels in the exchange appear to be within that variation. The Spanish laboratories, counting by the modified procedure (of concurrent ERM and WHO counts) in the exchange, showed relatively little variation from year to year, and were remarkably consistent with their levels in the exchange counts. Thus, in most cases, it appears that the use of the special counting procedure (to produce both WHO and ERM counts) did not influence the level of the counts substantially.
|
Consistency with data from the international (AFRICA) scheme
Three of the laboratories in the exchange participate regularly in an international fibre counting comparison scheme operated by IOM, AFRICA. The participants in that scheme are asked to count the samples by their normal methods, and the guidelines for ‘satisfactory’ performance are 75% of counts within the range 0.5–2.0R (for samples with density above 63.5 fibres mm–2), where R is the reference value derived as the median of (a minimum of 15) previous counts on the sample. For lower density samples, the limits are constant on the scale of square root of density; and this gives a wider range on the original scale at the lower densities. All the counts (in Fig. 5) are within these bounds. The densities from the Belgian laboratory, being produced by the WHO counting method since 1999, appear somewhat higher than those from the other two laboratories. Figure 5 shows all the counts made in the AFRICA scheme by these three laboratories over the past 4 years.
|
DISCUSSION AND CONCLUSIONS
Comparability of the national reference counts
The main aim of this study was to compare the relative levels of the consensus reference values of the national fibre counting proficiency testing schemes in Spain, Belgium and the UK. In principle, consensus reference values produced by separate groups of laboratories could be very different. These groups have not been completely isolated due to the existence of regular international counting comparisons involving some of these laboratories in the AFRICA scheme, and therefore we started this exchange with the expectation that huge differences would be unlikely but important differences could be possible. If the levels of the reference values are consistent, then, since they are consensus values, it implies that the counting levels are broadly consistent across national schemes.
This study is, we believe, the first time that national fibre-counting PT schemes have provided a pool of their reference samples for circulation around a representative group of laboratories from the schemes, and that attempts have been made to compare counting levels in this way.
We have used the counts produced in this exchange on 40 representative slides to provide a common basis against which to compare the national PT schemes' reference values. There are, of course, uncertainties in using this data as the basis for comparison, including the inherent variation in all fibre counts and the systematic differences between the laboratories participating in this exchange. Fortunately, the systematic inter-laboratory bias remained fairly constant across slides of different density and origin (Spanish, Belgian, UK).
It was necessary to look at the subset that comprised data from the four same laboratories for all three schemes, as only four of the laboratories (the Spanish and the UK laboratories) currently use the ERM method. We used these four as a common basis for comparing the reference values for all three schemes. However, for the Belgian samples, with reference counts produced by the WHO method, we also examined whether the picture would change substantially if the analysis included counts by all six laboratories, in order to include the Belgian laboratories' counts. This produced only a relatively minor effect on the comparability.
The counts in this exchange were produced by a special method by four of the labs, with ERM and WHO counts being produced concurrently on the same fields. It is well recognized that the use of special counting methods may affect variability or counting level (e.g. Kenny et al., 1983). Therefore it was very useful to confirm (Fig. 4) that the level of the ERM counts generally appeared consistent with the level of previous counts in the national PT schemes over recent years. However, the normal level of the future counting by WHO method for these laboratories can only be known after they routinely use the WHO method. Nevertheless, the consistency in the ERM counting level is a reasonable ground for assuming that their future WHO counting is likely to be consistent with the present WHO count data. However, there will be a need to monitor the comparability after the WHO method has been adopted as the routine method in the UK and Spain, to check that assumption.
The present results are of course based on data from only a limited number of samples from each scheme (e.g. 16 from the RICE scheme, compared with a stock of 900 samples). Nevertheless, they were chosen to represent the schemes and we have no reason to believe that they are other than a typical selection of samples. Therefore, the results are indicative of the relative counting level in these three schemes.
The geometric overall mean ratios between the national reference values and the mean counts made in this special exchange were all remarkably close to 1, with values ranging from 0.94 to 1.01. This implies that the national reference values for these three national schemes are very consistent, with the difference in average levels being only 5%. This is a remarkably small average difference, being much less than the systematic bias between some of the laboratories participating in this special exchange. Since the reference values are consensus values for the counts within each scheme, the agreement in consensus values implies that the counting levels are broadly consistent between these three national schemes.
The UK scheme appears to have reference values that are, on average, slightly lower than those in the Belgian or Spanish schemes. The observed difference between national schemes, of 5%, is small compared to the width of the target range for acceptable counts in each of the national schemes; nevertheless, it may be sensible to take account of that information in any future revision of the target ranges being set for counts within each national PT scheme.
Sample origin
Sample origin did not appear to have much effect on the inter-laboratory differences in this exchange (in the analysis of deviance, the interaction between laboratory and sample origin was a minor term). Therefore, it appears that performance within these three national schemes is unlikely to be substantially affected by differences in their samples.
Inter-laboratory comparisons
The inter-laboratory differences were large enough to confirm the need for regular inter-laboratory comparisons in national proficiency testing schemes to maintain consistency between laboratories. Similarly, it implies that there is a need to continue to monitor comparability internationally.
Three of the participating laboratories already participate in regular international counting comparisons (in the AFRICA scheme), and those previous comparisons may have helped to produce the good comparability seen in the national reference values. We suggest that continuation of the international comparisons may help to maintain this comparability.
The present study was conducted within a period of a few months during 2002, and with slides that are only a small subset of the slides in the national schemes. Ongoing linkage between the schemes is needed to confirm the findings and to monitor the stability of the apparently good consistency in mean counting levels. The quality assurance for fibre counting would thus involve very regular internal QA, then frequent participation in a national PT scheme for every laboratory, and then participation by some key laboratories in international comparisons.
Fibre-counting method
The comparisons of counts produced by the WHO method and by the ERM method showed that the imminent change in rules will probably cause counts to increase by 20% on many samples in all three schemes. The extent of the change is expected to vary from sample to sample; larger changes are expected on samples with lots of particles, and no change on samples that have little or no particulate dust. So the results reflect the particulate loading on this representative set of samples from the schemes.
Because these samples originated from different sources (asbestos textile, asbestos cement manufacturing and asbestos removal), the results are also indicative of the order of difference to be expected in counts on samples from those types of dust cloud, and it is useful to see if the ratios obtained here are consistent with previous estimates. Iles (1989) reported on the ratio of counts obtained by two rules, where one rule was the current ERM and the other was essentially the new WHO rule. His data were based on separate counts by the two methods on a total of 200 samples, comprising 50 samples from asbestos textile manufacturing; 50 samples from asbestos friction product manufacturing; 50 samples from asbestos cement manufacturing; and 50 ‘low density’ samples from asbestos clearance type samples. He found that the use of separate counts (i.e. ERM and WHO counts produced on different fields of view) caused approximately half of the variation in estimates of the ratio of one count to the other, as seen by comparing repeat ERM counts by the analyst. This is the main reason why (in Table 6) the standard deviations obtained in this study are consistently smaller than those reported by Iles (1989). It is also possible that there may have been more variation in the particulate content on the samples used by Iles (1989). The geometric mean ratios reported by Iles are generally higher than those found here; that may reflect a difference between the samples kept in the PT schemes and those obtained by Iles from these different types of source. The RICE scheme may have excluded some textile samples with excessive levels of particulate; the asbestos cement samples in the Spanish scheme were generated from a laboratory dust disperser; and the activities sampled in clearance/removal work were not well defined in either case.
|
These combined results confirm that the change in method can make substantial differences to the level of counts on some samples. Clearly the magnitude of the difference will depend on the particulate content of the samples, and each laboratory or analyst's interpretation of the subjective rule (in the current ERM method) regarding fibres that appear to touch particles.
Kenny et al. (1983) reported on an exercise to train analysts in changing from one counting method (an HSE in-house method) to the current ERM method (when it was first introduced). They found that microscopists' counting levels fluctuated over a period of months, and they suggested that this variation might be due to the effect of the ERM rule requiring a subjective decision as to whether fibres touched particles. If their explanation is correct, then a change to the new WHO rules from the ERM rules may be less prone to temporal fluctuation in the level of counts. Nevertheless, such temporal variation after changing to a new method is another reason in favour of using the special concurrent counting procedures applied by four laboratories in this study.
In the data from the present study, there was reasonably good agreement between the laboratories in the estimate of the percentage change in count due to the change in rules for most samples. However, one (out of the four laboratories) occasionally produced much higher estimates of the percentage change. That suggests that some (perhaps temporary) increases in inter-laboratory differences might arise during the change over from ERM to WHO counting rules in the Spanish and UK schemes. Other trials, with more laboratories, are underway in the UK to examine how variable the effect of the transition may be from laboratory to laboratory.
Harmonizing fibre-counting levels
Regulatory standards for controlling exposure to asbestos fibres are defined in terms of airborne fibre concentrations that are measured by the fibre-counting method. Considerable effort has been devoted to establishing European regulatory standards, but their application relies on consistency in the application of the measurement process, and for fibre counting that consistency relies on controlling inter-laboratory variation by quality assurance and proficiency testing. The present study indicates that average counting levels are, at present, consistent among these three European national PT schemes, so that would appear to be a positive base from which to build towards a wider European consistency.
Extension to establish equivalence
The current study focused on the comparability of national schemes' consensus reference values. However, we also noted the need to address the comparability of their target ranges (the range around the consensus target level that is deemed satisfactory) in a way that takes account of differences between national schemes in their maturity, size and level of reference values. Progress towards tighter limits becomes more feasible now that the consistency of the counting levels has been established (at least for the schemes participating in this study).
Arroyo and Rojo (2001) recommended new limits that would produce an equal chance of counts falling outside the limits for each of these national schemes, and that implied tighter limits for the UK RICE scheme. Recently, the UK RICE scheme has started sub-dividing the list of laboratories published as satisfactory into two categories, where category 1 includes only those laboratories that meet limits that are closer to those recommended by Arroyo and Rojo.
Since it is important to be able to compare fibre-counting measurements internationally, it is important to have consistency in the target levels being set by fibre-counting PT schemes. The consistency between the three schemes (in this study) would be a good platform for further comparisons to assess and develop comparability with other national PT schemes, perhaps as either special exercises such as this or within the process of regular international exchanges such as those in the AFRICA scheme. Extension is needed to find how comparable fibre-counting levels are across all of Europe, and wider.
The AFRICA scheme includes one or two laboratories in 10 European countries, as well as laboratories in North and South America, Africa, India and the Far East. Participation provides members with information to assess their own comparability with other laboratories internationally. We recommend the advantages of developing an existing scheme for international comparisons. However, special exercises (such as in this study) may be needed to achieve the consistency across all of Europe. Then the regular international exchanges are needed to maintain consistency between countries and their national schemes.
APPENDIX 1: SCREENING COUNTS
Introduction
The 40 samples were counted initially by a single analyst at IOM, tallying fibres, ERM fields and extra WHO fibres, but examining only 30 fields on each sample. We call these counts on a reduced number of fields, ‘screening counts’. Counting fewer than the normal 100 fields should theoretically make the counts subject to greater statistical variation. Therefore the ERM counts produced from these screening counts were compared with the subsequent normal ERM counts from the same laboratory to check whether the screening counts should be included with or without restriction in the main analysis.
Results
The screening counts (30-field counts) are compared with counts on the usual number of fields in Fig. A1.1. For those samples where the screening counts and the subsequent full counts were by the same analyst, there is remarkably little difference between the estimated densities. Where the subsequent count was by another analyst, the differences are still acceptably small compared with the variation expected and acceptable between repeat counts. Where the density estimate exceeds 380 fibres mm–2, a count of 100 fibres would be reached in 30 fields and so the ‘full counts’ and ‘screening counts’ would examine the same number of fields on samples with this or higher density. The geometric mean ratio of screening count to full count was 1.09 over 40 samples.
|
|
Conclusions
The consistency between the ‘screening counts’ by one analyst and subsequent full counts either by the same analyst or other analysts is sufficiently good that the ‘screening count’ data can be included without restriction in the main analysis.
APPENDIX 2
The data for the graphs plotted in this appendix for the Belgian samples comprised:
- all WHO counts from the exchange; and
- all WHO counts, but giving equal weight to laboratories (by averaging over multiple counts from the same laboratory, so that each laboratory contributed one value to the geometric mean).
ACKNOWLEDGEMENTS
We wish to acknowledge the funding provided by HSE to support the IOM's work in this study. The authors wish to thank the staff in the six laboratories who produced the data. These laboratories were: from the UK: The IOM, and the Health and Safety Laboratory (HSL); from Belgium: Laboratorium voor Industriele Toxicologie (LIT), Laboratory of FPS Employment, Labour and Social Dialogue and Fibrecount NV; from Spain: Centro Nacional de Verificación de Maquinaria (CNVM), Instituto Nacional de Seguridad e Higiene en el Trabajo (INSHT), Centro Nacional de Medios de Protección (CNMP), Instituto Nacional de Seguridad e Higiene en el Trabajo (INSHT).
Received October 2, 2003; in final form October 21, 2004
REFERENCES
Arroyo MC. (1990) Programma Interlaboratorios de control de calidad de fibres de amianto. Salud y Trabjo; 80: 4–10.
Arroyo MC. (1991) Quality control of airborne asbestos fibre counts in Spain. In: Clean air at work. Luxembourg 9–13 September 1991. Royal Society of Chemistry. Special Publication no 108: 440–4.
Arroyo MC, Rojo JM. (2001) A proposal for harmonising laboratory performance criteria in national fibre counting schemes. Ann Occup Hyg; 45: 447–55.
Beckett ST, Attfield MD. (1974) Inter-laboratory comparisons of the counting of asbestos fibres. Ann Occup Hyg; 17: 85–96.
Brown PW, Jones AD. (2001) Interlaboratory comparison schemes for fibre counting: AFRICA and the WHO/EURO reference scheme. Gefahrstoffe Reinhaltung der Luft; 61: 101–4.
Carton B, Kauffer E, Derzko G et al. (1981) Results of an asbestos fibre counting trial in France in 1979. Ann Occup Hyg; 24: 9–21.
Crawford NP, Brown P. (1992) The RICE and AFRICA schemes for asbestos fibre counting. Ann Occup Hyg; 36: 59–69.
Crawford NP, Cowie AJ. (1984) Quality control of airborne asbestos fibre counts in the United Kingdom—the present position. Ann Occup Hyg; 28: 391–8.
Crawford NP, Jones AD. (1980) Central Reference Scheme for asbestos fibre counting. In: Asbestos International Association. Third International Colloquium on Dust Measuring Technique and Strategy, Cannes, June 1980. Paris: Asbestos International Association. pp. 97–9.
Crawford NP, Brown PW, Cowie AJ. (1992) The RICE and AFRICA schemes for asbestos fibre counting. Ann Occup Hyg; 36: 59–69.
Health and Safety Executive. (1995) Asbestos fibres in air: sampling and evaluation by phase contrast microscopy (PCM) under the control of asbestos at work regulations MDHS 39/4 Sudbury: HSE Books.
Iles PJ. (1989) Comparison of asbestos and MMMF counting rules when applied to asbestos. In: Proceedings of the AIA 6th International Colloquium on Dust Measurement and Strategy, Jersey, November 1989, pp. 171–84. Asbestos International Association, London.
Iles PJ, Johnston AM. (1983) Problems of asbestos fibre counting in the presence of fibre–fibre and particle–fibre overlap. Ann Occup Hyg; 27: 389–403.
Kauffer E. (1990) The French asbestos counting quality control scheme. In: Proceedings of the AIA 6th international colloquium on dust measurement technique and strategy, 7 November 1989 Channel Islands, Paris: Asbestos International Association. pp. 156–63.
Kauffer E. (1992) The asbestos fibre inter-laboratory counting exchange in France. Cahiers de notes documentaires- Sécurité et hygiene du travail. Note No 1883-147-92. pp. 249–54.
Kenny LC, Shenton-Taylor T, Ogden TL. (1983) Training of HSE's manual asbestos counters to the European reference method. HSE Labs report IR/L/FD/83/16.
McCullagh P, Nelder JA. (1989) Generalized linear models. London: Chapman and Hall.
Miller BG. (1984) Statistical method for analysis of membrane filter samples of airborne asbestos. Ann Occup Hyg; 28: 217–27.
Schlect PC, Shulman SA. (1986) Performance of asbestos fibre counting laboratories in the NISOH Proficiency Analytical Testing (PAT) program. Am Ind Hyg Assoc J; 47: 259–69.[Web of Science][Medline]
Walton WH, Attfield MD, Beckett ST. (1976) An international comparison of counts of airborne asbestos fibres sampled on membrane filters. Ann Occup Hyg; 19: 215–24.
WHO (1997) Determination of airborne fibre number concentrations. A recommended method, by phase contrast optical microscopy (membrane filter method). World Health Organisation, Geneva. ISBN 92 4 154496 1.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  T. OGDEN Annals of Occupational Hygiene at Volume 50: Many Achievements, a Few Mistakes, and an Interesting Future Ann. Hyg., November 1, 2006; 50(8): 751 - 764. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. GRZEBYK, E. KAUFFER, and L. FREVILLE Proposal to Adapt the WASP Programme to Fibre Counting Tests Ann. Hyg., June 1, 2006; 50(4): 427 - 427. [Full Text] [PDF]
|
|
|
|
|
|
B. PREAT Comments About the Proposal to Adapt the WASP Programme to Fibre Counting Tests Ann. Hyg., January 1, 2006; 50(1): 105 - 105. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||