vvEPA United States Environmental Protection Agency Environmental Monitoring Systems Laboratory Research Triangle Park NC 27711 Research and Development EPA-600/S4-82-021 May 1982 Project Summary Bulk Sample Analysis for Asbestos Content: Evaluation of the Tentative Method E. P. Brantly, Jr., K. W. Gold, L E. Myers, and D. E. Lentzen The U.S. Environmental Protection Agency Office of Pesticides and Toxic Substances, Washington, DC, and the USEPA Environmental Monitoring Systems Laboratory, Research Trian- gle Park, NC, jointly sponsored an effort to produce a practical and objective analytical protocol for the qualitative and quantitative analysis of asbestos in bulk materials. Draft pro- cedures were written for analysis of bulk samples by polarized light micros- copy (PLM) and X-ray powder diffrac- tion (XRD). Following review, the Tentative Method for the Determina- tion of Asbestiform Minerals in Bulk Insulation Samples (March 1980) was submitted to a performance testing program that involved multiple labora- tory analysis of prepared samples with known asbestos content. This report presents the results of the testing study and provides preliminary obser- vations and characterization of the utility and operational, parameters of the tentative Method. PLM quantitative analysis employs a point counting procedure to estimate the relative area occupied by asbestos fiber within the microscope fields of view. PLM data must be compared with the known weight of asbestos in the sample in order to characterize the accuracy of the method, Data pro- duced by the point counting procedure are also compared with those pro- duced by the typical quantisation pro- cedures used by some of the partici- pating laboratories. Accuracy and precision of the point counting pro- cedure are considered in two contexts: (1) as PLM is currently used, regarding reported data as a direct estimate of weight percent of asbestos present and (2) allowing adjustments of the data to account for bias and variance in the relationship between the relative area occupied by asbestos and the known weight percent of asbestos in the sample. Information is also pre- sented on within-laboratory variance and the frequency of false negatives and false positives. A very limited amount of data was returned for characterizing the XRD protocol. Both thin-layer and thick- layer (bulk) techniques were used for quantitative XRD analysis. Because of the small number of XRD reports and the nonequivalence of methods em- ployed, it is not possible to draw any firm conclusions on the precision and accuracy of the XRD protocol. A general comparison of bulk and thin- layer techniques with respect to pre- cision, accuracy, and sensitivity is made. This Project Summary was develop- ed by EPA's Environmental Monitoring Systems Laboratory, Research Trian- gle Park, NC, to announce key findings of the research project that is fully ------- documented in a separate report of the same title (see Project Report ordering information at back). Introduction An interlaboratory study was conduc- ted to evaluate the accuracy, precision, and general utility of the Tentative Method forthe Determination of Asbes- tiform Minerals in Bulk Insulation Samples (March 1980). Twenty-two commercial and four government labora- tories were each supplied with 11 samples. Eight of the samples were -formulated with a known weight of amosite or chrysotile and a matrix material containing primarily gypsum. Within-laboratory duplicates, blanks, and "real-worjd" samples of sprayed insulation were also included in the materials distributed to laboratories. Four laboratories (two commercial, two gov- ernment) chose not to participate in the study. Th#22 participating laboratories provided a total of 30 polarized light microscopy (PLM) reports and six X-ray powder diffraction (XRD) reports. The Tentative Method includes proce- dures for qualitative and quantitative analysis of bulk samples by PLM and XRD. Identification of asbestos fibers by PLM requires the observation of six optical properties:_morphology, color and pleochroism, refractive indices (or dis- persion staining colors), birefringence, extinction characteristics, and sign of elongation. PLM quantitative analysis uses a point counting procedure to estimate the percent area occupied by asbestos fiber within the microscope fields of view. The prepared samples distributed in this study contained a known weight percent of asbestos. Because PLM analysis produces an estimate of the relative area occupied by asbestos, the relationship between re- ported area percent and the known weight percent of asbestos was investi- gated. Identification of sample components by XRD analysis is accomplished by comparison of the sample diffraction pattern with standard reference powder diffraction patterns. Quantitative anal- ysis involves measuring the integrated areas of diagnostic peaks selected from the full XRD scan of a thin-layer sample. Quantitative analysis must include a correction for matrix absorption effects and comparison with suitable external standards. XRD affords information only on crystal lattice structure and not on crystal morphology. XRD analysis, there- fore, cannot distinguish between asbes- tos minerals and their nonasbestiform varieties. The presence of fibrous parti- cles in a sample must be determined by an optical technique such as PLM. Conclusions Linear regression in natural logarith- mic coordinates was used to study the relation between the reported data, in terms of percent asbestos by area, and the known values, in terms of percent asbestos by weight. The fact that'a considerable amount of the variation in the data was removed by regression in logarithmic coordinates is consistent with the assumption that area percent and weight percent are related toy a power function. Analysis of the regres- sion shows that variation in the area- weight relationship is attributable to differences between laboratories, differ- ences between asbestos types (chryso- tile and amosite), and interactions between laboratory and asbestos type. Reported PLM data were divided into three groups based on the quantitation procedure(s) used by the reporting labor- atory. Group P—(Point count) PLM asbes- tos area percent determi- nations by the point count procedure (Tentative '. Method). Group B—(Both) PLM asbestos area percent determinations by the laboratories' own methods for laboratories that also provided data by the point count method. Group O—(Other) PLM asbestos area percent determina- tions by the laboratories' own methods for labora- tories declining to use the point count method. Considering reported PLM results as direct esti mates of the weight percent of asbestos (i.e., ignoring the problem of relating area percent and weight per- cent), it was found that Group O is significantly more biased than Group P. Groups P and B are similarly biased. Point counting has a greater positive bias on amosite samples than on chryso- tile samples. For a sample containing 10 percent chrysotile by weight, the aver- age bias (b) of Group P is 18.5 percent; for 50 percent chrysotile, b = -24.2 percent; for 10 percent amosite, b = 118.5 percent; for 50 percent amosite, b = 12.1 percent. A regression relating standard devia- tions and means of reported PLM results, when performed'in natural logarithmic coordinates, did not establish any differ- ences among Groups P, B, and O with respect to precision. The standard devi- ation of Group P is directly related to the mean reported value. Precision may be expressed as the coefficient of variation (CV). The CV is less than 100 percent on samples containing more than approxi- mately 6 percent asbestos by area, and less than 50 percent on samples con- taining more than approximately 32 percent asbestos by area. At a mean reported value (MP) of 10 percent asbes- tos, CV & 79 percent; at M P = 20 percent, CVs61 percent; at MP = 50 percent, CV s=41 percent. Analyses were performed on trans- formed data to investigate improve- ments in data quality that might be made by adjusting (calibrating) individ- ual laboratory results. If the parameters of the area-weight relationship are used to transform reported area percent data to predicted weights, a considerable gain in accuracy is achieved, as meas- ured by the average percent absolute error. By this measure Of accuracy. Group O has a greater error than Group P on samples containing less than 20 percent asbestos by weight. Residual variance in the transformed data is measured by the mean squared error about the regression line. Bythis meas- ure of variance. Group P was foundto be significantly more precise than Groups B and O. Further analysis of adjusted data indicates that laboratories using point count analysis are. better able to distinguish samples containing more than 7 percent from those with less than 7 percent asbestos by weight than they are able to distinguish samples with more than 1 percent from those with less than 1 percent asbestos by weight. Samples from the chrysotile series were included as within-laboratory dup- licates. Although a more extensive effort would be required .to adequately eval- uate the precision of the PLM protocol on repeated analysis of the same sam- ple, a gross estimate using'the present data indicates that within-analyst vari- ability accounts for less than 25 percent of the total variance. One of the important characteristics of the point count procedure to be evaluated is the likelihood of its gener- ati ng fa Ise positives and false negatives. ------- A false positive occurs when an analyst reports asbestos present in a sample that does not contain asbestos. A false negative occurs when an analyst reports no asbestos present in an asbestos- containing sample. Data produced by point counting included five false nega- tives out of a total of 19 analyses of the 1.2 percent chrysotile sample. One false negative out of 19 analyses was reported for the 4.9 percent chrysotile sample. No false negatives were reported for any amosite samples or for any samples containing more than 5 percent chryso- tile by weight. The reporting of false negatives is more likely due to the vari- ability of sample and slide preparation steps than to the point counting proce- dure perse. One false positive out of 19 analyses was reported for the series of blank samples and was probably due to contamination. The probability of a false negative on the 4.9 percent chrysotile sample was 0.05 (1/19). EPA currently recommends the analysis of at least three samples of a suspect material. The rate of false negatives is such that the analysis of three samples, if each con- tained at least 5 percent asbestos by weight, would result in three false negatives with a probability less than 0.03 and possibly as low as 0.001. The six laboratories reporting XRD results were gfouped into two general categories for p urposes of data ana lysis. These categories, thin-layer and bulk, were defined on the basis of the XRD technique used for quantitative ana lysis. Three of the laboratories performed the requested analyses using some varia- tion of the thin-layer method ofquanti- tation included in the Tentative Method. The remaining three laboratories used alternative bulk or thick-layer methods of quantitation. It should be emphasized that within categories none of the methods used were strictly equivalent. Moreover, within the thin-layer group, no laboratory followed the Tentative Method protocol exactly. Because of the small number of participating laboratories reportingXRD results, and the nonequivalence of methods employed, it is not possible to draw any firm conclusions from the reported results about the accuracy and precision of the XRD method. However, from a general comparison of bulk vs. thin-layer methodology, two observa- tions can be made. First, bulk methods appear to be at least as accurate/and precise as thin- layer methods over the range of samples included in this study and significantly more accurate for the analysis of chryso- tile. Second, there is a suggestion that thin-layer methods of analysis may be more reliable (i.e., more sensitive) than bulk methods at the 1 percent level of chrysotile in a simple matrix. Data produced by thin-layer methods of analysis included one false negative out of three analyses of the 4.9 percent chrysotile sample. The same laboratory reported chrysotile false positives for all amosite samples and for the blank sample with reported chrysotile values ranging from <1 to 8 percent. A second laboratory reported one false negative out of three analyses in the 19.4 percent 'chrysotile sample. Data produced by bulk methods of analysis included two false negatives out of three analyses of the 1.2 percent chrysotile sample. One of these labora- tories also reported a false positive amosite in the 4.9 percent chrysotile sample. Recommendations The study presented in this report is a preliminary evaluation designed to de- termine the precision and accuracy of the Tentative Method as applied to carefully prepared samples. It should be emphasized that the samples analyzed consisted of only two types of asbestos fiberand a single matrix material. Only one type of asbestos was included in any given sample. One of the main obstacles to reliable analysis of bulk samples is the variability of sample composition. Complete characterization of the method requires that several issues be addressed, as discussed be- low. The highest priority; however, should be assigned to investigations that will extend the application of the method to a range of real-world samples involving different fiber types and mat- rices. Polarized Light Microscopy Several aspects of the PLM method .require further investigation. Briefly, future studies should be designed to determine the following: 1. The feasibility of specifying defini- tive sample preparation proce- dures to be used for quantitative PLM analysis; 2. The proportion of total variance attributable to individual proce- dures of the method, i.e., sample preparation, sub-sampling, and point counting; 3. The proportion of total variance contributed^ by within-laboratory variability; 4, The effect of specific variables within the point counting proce^ dure, including the number of points to be counted, magnification used, and the possible bias intro- duced by the use of a 25-point reticle instead of a cross-hair reticle; 5. The possibility of introducing a staged point counting process that would allow.fewer counts to be determined on samples with a high percentage of asbestos; 6. The effect of the presence of more than one type of asbestos in a bulk sample; 7. The feasibility of individually cali- brating PLM laboratories with information derived in round robiri sample analysis programs. Recommendations for specific changes to the method include elimina- tion of the confidence interval calcula- tion and revision of the rule for reporting 1 percent asbestos. It is apparent from the results of this study that some type of training would be required to achieve comparable application of the PLM protocol between laboratories. While point counting is a classical petrographic technique, it is notastandard procedure inthe majority of laboratories currently analyzing bulk samples for asbestos. Training alterna- tives might include regional courses and distribution of split samples analo- gous to the NIOSH program for. the asbestos afr sampling method. It should also be noted that the PLM method presented, although an improve- ment over subjective techniques, is still a procedure for estimating the relative area occupied by asbestos fiber and matrix material. Alternative analytical techniques that measure weight percent directly or that provide an empirically more satisfying relationship to relative weight of asbestos fiber should be sought and investigated. X-Ray Powder Diffraction There are two major areas in the application of XRD techniques to quanti- tative analysis of asbestiform minerals in bulk materials that require further investigation: identification andcharac- ------- terization of standard reference mate- rials, and further development and evaluation of thin-layer and bulk meth- ods of analysis. The most common concern of labora- tories participating in the evaluation of the XRD protocol was the lack of well- characterized, readily available refer- ence materials. A thorough, systematic investigation of asbestiform materials for use as standard materials should be undertaken. This should include identifi- cation of major sources; determination of availability and cost; and complete mineralogical characterization and de- termination of purity, particle size distri- butions, and powder diffraction patterns of materials from these sources. Since asbestos minerals vary in com- position depending on the source and exhibit different behaviors in grinding, peak positions and/or relative intensi- ties of XRD patterns may vary from sample to sample. This variability is particularly problematic for the amphi- bole minerals. A quantitative study to assess the comparability of X-ray re- sponse of asbestos minerals from dif- ferent sources should be conducted. If possible, observed differences between different samples of the same asbestos variety should be correlated with specific sample characteristics (e.g., chemical composition and particle size). The need for further development and evaluation of both thin-layer and bulk methods of XRD analysis is underscored by the following observations: few laboratories .are currently set up to routinely perform the thin-layer analysis as prescribed; the proposed thin-layer method of quantitation is considerably more time-consuming and costly than bulk or thick-layer methods; and for samples analyzed in the methods evalu- ation study, the;bulk method was at least as acc.urate and precise as the thin-layer method. In particular, a comparison of the bulk and thin-layer methods should be made over a variety of asbestos types and matrix materials, with attention given to sample preparation requirements, instru- ment requirements, sensitivity, preci- sion, accuracy, and speed and cost of analysis. For both bulk and thin-layer methods, the following areas of investigation are proposed: 1. Assessment of sample preparation requirements; 2. Assessment of preferred orienta- tion effects on quantitative.anal- ysis; 3. Assessment of the effect of the use of the step-scanning mode of analysis on the limits of detection; and 4. Assessment of absorption correc- tion requirements and techniques. Results and Evaluation Polarized Light Microscopy Eleven sample series were distributed to laboratories. Eight of the series were targeted at specific weight percents of asbestos fiber. Two species of asbestos were used, chrysotile and amosite. One matrix material, containing primarily gypsum, was used in all prepared sam- ples. Target weights were designed to cover a wide range of asbestos concen- trations approximately equally spaced on a logarithmic scale. Blanks (Series F) were provided as controls and for deter- mining the method's potential for pro- ducing false positives. The "real-world" sample (Series J) was included for comparison of between-laboratory var- iance. Duplicates (Series K) were in- cluded to estimate the average within- laboratory variance. Target weights and allowable limits for matrix and asbestos fiber in each sample series are presented in Table 1. Table 1. ' Sample Composition .Group means and standard deviations are summarized in Table 2. Note that in six of nine cases the mean (MP) of the point count group is closer to the nominal weight than the mean (MB) of Group B. This is not a significant differ- ence, and it appears that Groups P and B are comparably biased. Note also in Table 2 that estimates by Group 0 are consistently higher .than those by Groups P and B. Sign tests suffice to show that Group 0 is significantly more biased than Groups P and B. Ninety- percent confidence intervals were cal- culated for Group P data. Using the midpoints of the confidence intervals, the average percentbias was calculated at several weight percent levels. These are presented in Figure 1. The percent bias varies with weight percent -of asbestos similarly for amosite and chry- sotile samples. Point counting has a greater positive bias on amosite samples than on chrysotile samples and, in fact, underestimates asbestos content in samples containing more than about 18 percent chrysotile by weight. The standard deviation of reported PLM data increases as the mea n reported area percent of asbestos increases for all groups. Precision may be expressed as the percent-relative standard devia- tion or CV. CV is related to means (MP) for Group P in Figure 2. The CV is less than 100 percent on samples with more Series C A E 1 H G D B F J Aft- Target Wt. % 1 ; .4 16 64 2 8 16 32 0 - Varies Actual Wt. % 1.2 4.9 19.4 74.5 2.5 9.8 19.4 38.8 0 50,0* Varies Fiber Type Chrysotile Chrysotile Chrysotile Chrysotile Amosite Amosite Amosite Amosite None Chrysotile Chrysotile Wt. of Asbestos (g) 0.05 0.20 0.80 3.20 0.10 0.40 0.80 1.60 ± ± ± ± ± ± ± ± .005 .01 .01 .01 .01 .01 .01 .01 None - — Wt. of Matrix (g) .4.95 4.80 4.20 1.80 4.90 4.60 4.20 3.40 3.0 ± ± ' ± ± •±. ± ± ± — - - — .05 .05 .05 .05 .05 .05 ..05 .05 5.0 *Mean of reported areapercents, Groups P and B. •f Series K samples were provided as duplicates and included samples from series C, A, E, and I. 4 ------- Table 2. Means and Standard Deviations of Reported PLM Results, by Group (P B, 0) (percent asbestos by area) Means Standard Deviations Series Type Weight MP MB MO SP SB SO C A E 1 H G D B F J Chrysotile Chrysotile Chrysotile Chrysotile Amosite Amosite Amosite Amosite None Environmental 1.2 4.9 19.4 74.5 2.5 9.8 19.4 38.8 0.0 - 4.2 7.3 21.7 64.3 12.5 26.2 37.8 48.9 0.2 50.7 5.3 4.9 19.0 63.0 18.0 29.3 42.3 57.7 0.0 49.3 7.4 24.8 42.0 85.6 24.0 40.0 41.4 65.0 1.0 65.0 4.5 6.3 14.8 19.6 8.6 16.9 17.7 19.5 0.9 16.1 5.3 2.9 5.9 . 17-.3 13.9 17.3 17.5 17.4 0.0 14.7 7.3 27.5 24.6 4.7 18.3 12.7 12.4 21.5 2.2 11.2 than approximately 6 percent asbestos by area and less than 50 percent on samples with more than approximately 32 percent asbestos by area. At M P= 10 percent asbestos, CV = 79 percent; at MP = 20 percent, CVs 61 percent; atMP = 50 percent, CV = 41 percent. It is of interest to evaluate the accuracy of the PLM methods after adjusting for the relationship between reported area percent and th.e known weight percent of the sample,^. This allows not only a better understanding of what reported PLM data mean but also indicates what improvements might be made in data quality by adjusting PLM area percent estimates to better represent weight percent. Area percent data were ad- justed for laboratory and asbestos-type effects to yield predicted weight percent values for each individual result. The most obvious and expected result in comparingthe average percent absolute errors of treated and untreated data is the considerable gainin accuracy (reduc- tion of error) that results from the trans- formation. After transformation, the average Group P inaccuracy is only one- fifth of the original. Adjusted Group P data are significantly more precise than those of Groups B and 0. If laboratories had access to information with which they could calibrate their results (accord- ing to the area-weight relationship for each laboratory and asbestos type), considerable gains in accuracy and precision of results could be achieved. The gains in precision would be greater for laboratories using the point counting quantitation procedure than for labora- tories using alternative procedures. X-Ray Powder Diffraction Means and standard deviations of all reported XRD results are shown in Table 3. Average reported values for XRD are shown for bulk methods, thin-layer methods, and both methods together. Except for Series G, the means of the bulk method are closer to the reference values than those of thethin-layer meth- od. Estimates of precision/given by the coefficient of variation, showed no sig- nificant difference between bulk and thin-layer methods. Considering individ- ual CVs, those for bulk are all less than or equal to those for thin-layer, except for Series C and B, further suggesting that bulk methods are at least as precise as thin-layer'methods, as applied by laboratories in this study. Comparison of the bulk and thin-layer methods by asbestos type i ndicates that for analysis of Chrysotile, bulk methods are significantly less biased than thin- layered methods. No significant differ- ence in slopes (bias) was observed between bulk and thin-layer methods for amosite. The results do give evidence that XRD is capable of detecting chrysotile at the 1 percent level in a simple matrix and suggest that at this level the thin-layer method may be more reliable. Further investigation is required to determine reliable detection limits over a variety of sample materials for both procedures. Bulk methods appear to be at least as accurate and precise as thin-layer meth- ods over the range of samples included in this study and significantly more accurate for the analysis of chrysotile. Since chrysotile is the most commonly occurring asbestos mineral jn bulk insulation materials, and since most laboratories routinely performing quan- titative analysis of asbestos in insulation samples use bulk methods of analysis, use of bulk methods of XRD analysis (ancillary to PLM) should be given further consideration. ------- 100 80 60 40 I 20 -20 -40 amosite chrysolite <20 40 60 -»„.. 80 Asbestos we.fght percent 100 Figure 1. Average percent bias of Group P (point count} data. ------- 7.7 7.0 .5 I 1 .7 .6 .5 .4 chrysotlle amosite 20 40 60 Group P mean 80 100 Figure 2. Coefficient of variation of Group P (point count) data. Table 3. Means and Standard Deviations of Reported XRD Results (percent asbestos) Weight Thin-Layer Bulk Pooled Series C A E 1 H G D B F Type Chrysotile Chrysotile Chrysotile Chrysotile Amosite Amosite Amosite Amosite None % 1.2 4.9 19.4 74.5 2.5 9.8 19.4 38.8 0 M 3.0 3.3 3.7 50.0 1.5 7.0 28.0 61.0 0.2 S 2.0 3.5 3.5 7.1 0.7 5.7 12.7 11.3 0 M 1.0 4.3 18.0- 74.5 3.0 21.7 24.0 52.0 0 S 1.7 4.2 11.4 0.7 1.4 4.2 6.9 22.5 0 M 2.5 4.3 10.8 62.2 2.8 15.8 25.6 55.6 0 S 1.8 3.3 10.9 14.7 1.0 9.0 8.3 17.6 0 *USGPO: 1982—559-092/3407 i ------- E. P. Brantly, Jr., K. W. Gold. L E. Myers, and D. E. Lentzen are with the Research Triangle Institute. Research Triangle Park, NC 27709. J. J. Breen and M. E. Beard are the EPA Project Officers (see below). The complete report, entitled "Bulk Sample Analysis for Asbestos Content: Evaluation of the Tentative Method." (Order No. PB 82-196 841; Cost: $13.50, subject to change) will be available only from: National Technical Information Service 5285 Port Royal Road Springfield. VA 22161 Telephone: 703-487-4650 The EPA Project Officers can be contacted at: Environmental Monitoring Systems Laboratory U.S. Environmental Protection Agency Research Triangle Park, NC 27711 , United States Environmental Protection Agency Center, for Environmental .Research Information Cincinnati OH 45268 Postage and Fees Paid Environmental Protection Agency EPA 335 Official Business Penalty for Private Use $300 -I ------- |