Response to a letter to the Editor by Dr. David Egilman and Mr. John Schilling regarding the article by Donovan et al. (2011)

To the Editor,

We have reviewed the letter prepared by Dr. David Egilman and Mr. John Schilling regarding our paper on bystander exposure to asbestos (Donovan et al., 2011). We appreciate their interest in a topic that has been discussed in principle for many years in the industrial hygiene field, but, until our analysis, had not been quantitatively characterized in the literature. They have raised a number of questions to which we welcome the opportunity to respond.

As we describe in more detail below, Egilman and Schilling try to build the case that our analysis is invalid because our model did not address small fibers (i.e. those less than 5 µm in length). Throughout their letter, they cite results from models and/or studies as being contradictory to our analysis, when, in fact, these models or studies addressed much smaller fibers than those that were the focus of our analysis (i.e. fibers 5 and 10 µm in length). We recognize that smaller fibers can remain in the air longer and can travel further than larger fibers. We remind readers, however, that fibers shorter than 5 µm have not been shown to be biologically relevant, which was the reason our analysis focused on the longer fibers (Stanton, 1973; Stanton et al., 1977; Davis et al., 1978; ERG, 2003a, 2003b; Platek et al., 1985; Berman, 2010). In response to their assertions that some of our assumptions were not valid or were at odds with published data, we reran our models to evaluate a wider range of fiber widths and settling velocities (including some of the values that Egilman and Schilling state we should have used). None of the changes had an appreciable impact on our predicted bystander:worker ratios.

We also note that many of Egilman and Schilling’s comments are devoted to their concerns regarding sources of financial support. These comments have no bearing on the scientific merits of our analysis. Dr. Egilman has previously published on the topic of bias in research funded by corporate entities, often claiming that the two go hand in hand (Egilman et al., 2003; Egilman & Billings, 2005). Broad claims about the lack of credibility of research conducted by certain persons or groups due to their affiliation or source of funding, in an attempt to immediately discredit the work, are not new. If anything, such claims have become more frequent due to the amount of litigation in the United States and competition for research funding. Nearly 20 years ago, in a commentary piece in the Journal of American Medical Association (JAMA), Dr. Kenneth Rothman, a former Harvard professor, noted that the objectivity of the process of scientific communication in the published literature “depends on judging a work by its merits, rather than on the inferred mind of the reader” and stated further that “the ironic result of pointing attention away from the work and toward the author may be thus to stifle objectivity, by taking our attention away from substance and shifting it towards credentials” (Rothman, 1993, p. 2783).

We strongly disagree with the assertion that valid and carefully conducted studies can be summarily discredited simply on the basis of a perception that a particular funding source for the research introduces a level of bias that compromises the quality of the analyses. Using this logic, one would have to discard all of Egilman’s criticisms since he admits that he “has served as an expert in asbestos litigation at the request of both injured people and asbestos product manufacturers.”

Setting aside the lack of evenhandedness in Egilman and Schilling’s comments on our paper, the following are our responses regarding the issues that they raise:

1)

Egilman and Schilling claim that our conclusions were derived wholly from an indoor air dispersion model, and that “the authors give primacy to this model despite the fact that actual workplace measurements and workplace simulation studies contradict the results generated by the mathematical model.”

Egilman and Schilling assert that we relied upon the results of a model and ignored the published scientific literature that they believe are inconsistent with our model predictions. In our analysis, however, we clearly acknowledge that the available field data have inherent limitations and are inadequate, by themselves, for estimating concentrations for bystanders. It is important to note that our statement that “no discernable trends were evident when bystander:worker ratios were plotted independently against distance from the primary worker” (Donovan et al., 2011, p. 66) refers to the lack of a consistent trend with distance once we had pooled all of the data from the various studies into a single dataset.

We did not claim that there was “no relationship between distance from an asbestos source and exposure level,” in general, nor do we believe that the available published data are uninformative or contradictory to our model predictions. It is clear from both the workplace surveys and simulation studies included in our analysis that, in nearly all of the studies we evaluated, airborne asbestos concentrations were found to decrease with distance from the source, particularly when only a single source was present. This outcome is not surprising scientifically, given that the law of mass conservation requires that concentrations of a substance must decrease with distance from an emission source as a result of mixing with air (Keil et al., 2009).

As we clearly state throughout our paper, our initial approach was to establish a “rule of thumb” using only workplace and simulation study data, but we discovered that doing so was not possible for several reasons. First, the published studies from the 1970s evaluated a wide variety of workplace conditions and reported on tasks that were not directly comparable to each other (e.g. application of spray insulation from a scaffold, application of asbestos patching and taping compounds, or various tasks involving automotive/heavy truck brakes). These studies were neither designed to evaluate bystander exposures, nor was it possible to completely exclude the possibility that other asbestos-containing products were being used at the sites at the time that samples were collected. Second, more than half of the samples (worker and bystander alike) from simulation studies were found to be below analytical detection limits. At such low levels, even slight differences in analytical detection limits will influence the bystander:worker ratios; care must be taken when interpreting the meaning of ratios from samples near the LOD. Third, the bystander samples were not collected at uniform distances from a source across the various studies. In some cases, the distance categories used in this analysis (1–5 ft, >5–10 ft, >10–30 ft, and >30 ft) were dominated by a high proportion of samples from a single study or were exclusively from either simulation studies or workplace studies. For example, the bystander:worker ratios for the >5–10 ft distance group were much higher than the other three distance groupings, but nearly all of these samples were from the Mowat et al. (2005) study, which had a significant number of samples that were less than the LOD (using LOD values creates uncertainty for ratio calculations, since a value of one half the LOD is used in calculations when, in fact, the true asbestos concentration could be much less, or zero; Mowat et al., 2005).

To address these concerns, we believed that it was important to identify mathematical models that could be used to predict concentrations of airborne fibers in the workplace. We focused on a simple scenario with a single point source. We selected two indoor air eddy diffusion models for particulate emissions because a model for fiber emissions was not available (Drivas et al., 1996; Keil & Ten Berge, 2000). These models can account for the difference in shape between a fiber and a particle by adjusting the settling velocity for particle geometry. Such approaches have been used for more than 50 years (Drinker & Hatch, 1954). The primary distinction between these two models is that the Drivas et al. (1996) model, which we considered more representative of a work environment, accounts for surface deposition and reflection off of room walls, while the Keil and Ten Berge (2000) model does not consider fiber removal from the air from these transport and loss mechanisms. We used results from the Keil and Ten Berge (2000) model to predict upper bound fiber concentrations. Both of these models are discussed by the American Industrial Hygiene Association (AIHA) in their guidance handbook on mathematical modeling for exposure assessment (Keil et al., 2009).

Table 1 below presents the values that were considered in our “rule of thumb.” As shown, the Drivas et al. model predicts that airborne asbestos concentrations will be 54% that of the source at distances between one and five feet. This value was based on the average predicted reduction for each one foot interval, ranging from 88.4% of the initial concentration at one foot to 13.8% of the initial concentration at five feet. For this same distance interval, the upper bound estimate, based on the Keil and Ten Berge (2000) model, is that airborne asbestos concentrations at one to five feet from the source would be, on average, 87% that of the source. In contrast, available field data for this distance category suggest a much more significant reduction, with concentrations at one to five feet being only 8% that of the source. Thus, because the values predicted by the Drivas et al. model were the midpoint between the upper bound and average values based on the available published data, 50% was used as the predicted reduction value for this distance category.

Table 1.  Summary of average percent reductions in airborne asbestos concentrations with distance from the source that were the basis for our recommended “rule of thumb.”

Distance	Proposed rule of thumb from Donovan et al. (2011)	Average airborne concentrations predicted as a percent of the concentration at the source; based on Drivas et al. (1996)	Average airborne concentrations predicted as a percent of the concentration at the source; based on Keil and Ten Berge (2000)	Average airborne concentrations predicted as a percent of the concentration at the source; based on published workplace and simulation study data
1–5 ft	50%	54%	87%	8%
>5–10 ft	35%	7%	48%	61%
>10–30 ft	10%	0.4%	14%	11%
>30 ft	1%	<0.001%	0%	10%

A similar approach was taken for the other distance categories. For >5 to 10 feet, the model predicted that airborne asbestos concentrations would be reduced to an average of only 7% of the airborne concentration at the source. Upper bound predicted values ranged from 60.8% at six feet to 36.2% at 10 feet (average value was 47.9%), whereas the average value from the published data was 61%. Because most of the published samples were from Mowat et al. (2005), which has a large percentage of samples less than the LOD, this value was interpreted with caution. The remaining samples from the published data in this distance category averaged 37.9%, so a value of 35% overall was selected for this category. At distances greater than 10 feet, the values predicted by the Drivas et al. model are less than 1% the concentration at the source. Our proposed rule of thumb, however, has a value of 10% to account for the fact that the average value for the workplace and simulation studies was approximately 11%. Similarly, at distances of 30 feet or greater, a value of 1% was used in the rule of thumb, although estimates from both models were less than 0.1% and, again, many of the simulation study data were dominated by results near or below the LOD.

Using the Drivas et al. model as a starting point, we made adjustments to incorporate the available workplace and simulation study data in order to avoid relying solely on the model results. Because of the differences in the available field data for each distance category, it was not possible to devise a single adjustment that could be applied consistently across all categories. The uncertainties associated with our approach of developing a single rule of thumb using several types of data are discussed in our paper. It is important to note that incorporating the field data, particularly at distances of 10 feet or greater, led to higher predictions of airborne asbestos concentrations within 50 feet than the Drivas et al. (1996) model alone would have.

2)	Egilman and Schilling believe that “all the evidence indicates that exposure concentration depends on many complex variables, such as ventilation conditions and task. Distance from a point source or primary worker is an unreliable predictor of bystander exposure.”

We agree that it is not possible to state authoritatively that, for all workplaces, we can predict the exposure of bystanders with a high level of precision. We also agree that ventilation is an important factor in terms of predicting airborne concentrations of a fiber released into the work environment. As we clearly acknowledge in our paper “the models in this analysis are not able to account for situations in which there may be local exhaust ventilation, open windows, large body fans, or high dilution ventilation (such as in some hot workplaces, for example ship boiler rooms or steel mills)” (p. 71).

However, the relationship between airborne concentrations in the breathing zone of workers performing a task, airborne concentrations at some distance from the source, and the influence of dilution ventilation, can be reasonably estimated for many open industrial environments, accounting for a number of basic properties relating to particle physics and mass transfer. Even considering the unique characteristics of a particular workplace, the universe of plausible answers to the relationship between distance from a source and concentration is knowable, which is one of the key points that we wanted to convey in our analysis. Egilman and Schilling are mistaken in their claim that models cannot be used to predict airborne asbestos concentrations experienced by bystanders in “real life” workplace conditions. The key is to ensure that the models are applied correctly and are based on realistic assumptions.

Indeed, the stimulus for our analysis was to bring clarity to the question of whether airborne asbestos fibers, measured at various distances from a point source, could pose a health hazard for individuals not directly working with asbestos products. In doing so, we hoped to advance the discussion of bystander exposure from the qualitative realm of “they may have been exposed to relatively high airborne concentrations” to more quantitative estimates that might allow better characterization of the actual health risk, consistent with the principles set forth by the National Academies of Sciences for conducting a health risk assessment (NRC, 1983, 1994, 2009). Our analysis includes a review of approximately 100 published relevant scientific papers, and evaluates various models that have been widely accepted within the industrial hygiene and exposure assessment communities to estimate those concentrations of fibers at various distances from an individual working with an asbestos-containing product (Bullock & Ignacio, 2006; Keil et al., 2009).

3)	Egilman and Schilling state that “capping the [bystander:worker] ratios at 1 skews the mean bystander:worker ratios shown in Figure 1 towards lower levels and misrepresents the actual data.”

By definition, for a single point source, bystander: worker ratios greater than one are impossible. However, such ratios were occasionally calculated in our analysis, which led to us to further examine why this occurred. We found a total of 15 paired samples that reported bystander:worker ratios greater than one (range = 1.14 to 2.67). These ratios are clearly depicted in Figure 4 of our paper. As expected, all of these samples were at or below the LOD, creating a mathematical artifact that Egilman and Schilling state has material significance and “discredit[s] [the] entire analysis.” A review of the underlying data, however, reveals that their claim is unfounded.

Ten of the 15 samples with ratios greater than one were from the Mowat et al. (2005) study. Worker samples in this study were based on an average of two or more replicates per task, and included results that were less than the LOD. Because a value of one half the LOD was used to generate these average values, the corresponding bystander:worker ratios could be greater than one. Egilman and Schilling acknowledge this fact, but then go on to cite two results at approximately 40 feet from the source in which the bystander concentrations are nearly twice those of the worker. What they fail to note, however, is that the airborne asbestos concentrations for the samples in question are extremely low, hovering around the analytical detection limits, thereby introducing uncertainty about the precision of the ratios calculated from these results.

Of the remaining bystander-worker pairs with ratios greater than one, all reported results (worker and bystander alike) were at or below 0.01 f/cc (Blake et al., 2003; Paustenbach et al., 2006). At such low concentrations, even slight differences in analytical detection limits can impact the reported values that are, in turn, used to calculate the bystander:worker ratios. In the realm of air sampling, and from a statistical standpoint, such differences are meaningless.

In addition to the samples from the various simulation studies, there were several samples in the Reitze et al. (1972) study that had higher airborne asbestos concentrations relative to the source (i.e. bystander:worker ratios would be greater than one). As is discussed elsewhere in this response, the Reitze et al. scenario was highly unusual and we did not consider these data reliable for our analysis because of the acknowledged variations in the ventilation conditions at the site during sampling.

Additionally, Egilman and Schilling assert that we attempted to “explain away” results where ratios were greater than one, but they only partially cited a sentence in our paper to support their claim. They quoted that “the authors argue that all the ratios greater than 1 are due to bystander samples that are <LOD.” We do not make the claim that all ratios greater than one were due to bystander samples that were below the LOD. The full quotation from page 56, not 54, of our paper is as follows:

Secondary box plots and x-y scatter plots were also generated using a bystander:worker ratio cutoff of 1 (i.e., for any bystander:worker ratios greater than 1, a value of 1 was used) in an attempt to reduce some of the variability that was due to the high number of ratios based on bystander samples with <LOD results; this phenomenon occurred almost exclusively in the simulation studies.” (Donovan et al., 2011, p. 56)

This sentence actually refers to setting a cutoff of one for the values for the whiskers of the box plots, representing upper percentiles of the data (97.5th percentile in our case). These values are estimated based on the standard deviation of the data set, which can be skewed if the estimate of standard deviation is biased, a phenomenon that is known to occur in data sets with a high frequency of values below the LOD (Singh & Nocerino, 2002; Lubin et al., 2004; Helsel, 2005, 1990; Baccarelli et al., 2005; Singh, 2006; Hewett & Ganser, 2007). Therefore, we elected to cut off the whiskers of the box plot at a bystander:worker ratio of one because the estimated upper bound values may have been skewed by the biased estimate of standard deviation caused by the high frequency of values less than the LOD.

4)	Egilman and Schilling state that “this cutoff level of 50 feet was not supported by their field study analysis, it was an assumption the authors made in this analysis.”

This statement mischaracterizes the logic behind our approach. First, a cutoff value of 50 feet was not an unsupported assumption. At this distance, both models predicted bystander:worker ratios of 0.0001 or less. In fact, as is shown in Table 5 of our paper, the predicted ratio based on Drivas et al. (1996), which we considered more representative of workplace conditions, is <0.00001 at 25 feet. At 50 feet from the source, upper bound predicted bystander:worker ratios based on Keil and Ten Berge (2000) are essentially the same (0.00001). Thus, both models clearly predict that there should be virtually no impact of asbestos emitted from a point source, for OSHA fibers, at distances greater than 50 feet. We do not deny that very small fibers could be present at longer distances, but such fibers, if they were present, would not be biologically significant.

Second, the statement that this cutoff level is not supported by field studies is incorrect. Egilman and Schilling misrepresented our statement that “no discernable trends were evident when bystander:worker ratios were plotted against distance.” Again, this statement only applies to the analysis that was conducted on a pooled dataset of all the available worker and bystander studies across all the workplace and simulation studies.

When studies are considered individually, only one simulation study included samples collected at 50 feet (Jiang et al., 2008). In this study, 23 of 24 samples collected at this distance were below the LOD (consistent with our model predictions), and the remaining sample had an airborne asbestos concentration of 0.003 f/cc. Egilman and Schilling note, however, that two of the workplace studies include samples collected at distances greater than 50 feet and that these samples deserve consideration (Reitze et al., 1972; Lorimer et al., 1976). As noted previously, the Reitze et al. (1972) paper described an atypical source (involving insulation sprayed from a high pressure hose). Samples collected at 15, 35, and 75 feet had airborne asbestos concentrations of 17 f/cc, 10 f/cc, and 46 f/cc, respectively [only one sample was collected]. The authors stated, however, that there were “changes in on-site ventilation” during the sampling, making interpretation of these data difficult (p.182).

Egilman and Schilling also cite Lorimer et al. (1976) as evidence that airborne asbestos from a point source can travel farther than 50 feet. They note that, in this study, the airborne asbestos concentration at 60 feet is 5.3% that of the worker. This percent reduction, however, was calculated by dividing 0.2 f/cc [the value at 60 feet] by the average value of 3.75 f/cc that was based on ten separate short-term worker samples, making it difficult to relate the individual concentrations. In citing specific data points from Lorimer et al. (1976) and Reitze et al. (1972), Egilman and Schilling formulated an argument based on one or two sample results and chose to ignore the rest of the available sample data, an approach that is both inappropriate and misleading.

5)

Egilman and Schilling “are also concerned by the authors use of ‘professional judgment’ to select key data and ‘adjust’ results,” and state that “none of the authors are old enough to have developed any ‘professional judgment’ based on work performed before 1972 and they do not cite any communication with anyone who might have had such experience.”

Egilman and Schilling assert that because the majority of the authors on our paper do not have workplace experience prior to 1972, we are not qualified to use professional judgment in our analysis. While it is true that not all of the authors were working or training in the field prior to 1972, we have considerable experience conducting retrospective exposure reconstructions, and have published extensively on this topic (Proctor et al., 2004; Williams & Paustenbach, 2005; Williams et al., 2007a; Williams et al., 2007b; McAtee et al., 2009; Gaffney et al., 2010; Sahmel et al., 2010; Widner et al., 2010). Such analyses require a detailed evaluation of historical work practices, which typically entails in-depth review of historical industrial hygiene data, evaluation of historical work practices and how they evolved over time, as well as interviews with historical workers, plant managers, and health and safety personnel. As a firm, we have probably reviewed hundreds of thousands of industrial hygiene samples across a wide variety of industrial settings, including workplaces where asbestos was used. Two of the authors on this paper are certified industrial hygienists with considerable field experience (JS and DJP), and one of the authors (DJP) has taught both undergraduate and graduate courses on industrial ventilation and the practice of industrial hygiene. Taken together, the collective experience of the authors of this paper is more than adequate for providing insight on various aspects of particle transport in workplace air.

Nonetheless, the first example that Egilman and Schilling cited concerned our estimates of distance between the bystander and source. Only two distances were estimated in our analysis (as noted in Table 3 of our paper). The first estimated distance was from the engine gasket on a car to the tire area [estimated at 2 feet]. The other sample was taken on an “observer” of engine disassembly work [estimated at 3 feet]. In both cases, we were comfortable estimating that someone who is observing engine work would not do so from more than several feet away (especially since these samples were grouped into the 1–5 ft distance category); we do not see how this assumption requires much professional judgment.

The second comment pertaining to our use of “professional judgment” relates to the fact that we chose to use the Drivas et al. (1996) model as a starting point for our “rule of thumb,” and that we made adjustments to the predicted bystander:worker ratios using professional judgment. We concluded that the Drivas et al. (1996) model was appropriate to use, primarily because it accounts for surface deposition and reflection off of room walls, phenomena that are known to occur in real workplaces. As described above, we clearly described our rationale for making adjustments for each distance category, based on the characteristics of the available field data for each category. We also discussed the limitations of this approach in the final section of our paper.

6)

Egilman and Schilling claim that the fiber deposition rate of 33.1 ft/hr is “[contrary to] EPA estimates (which were cited and discussed by Donovan et al.) [that] state that fibers that are 5, 2, and 1µm in length ‘require 4, 20, and 80 hours, respectively, to settle out of still air at a height of 9 feet.”

Egilman and Schilling cite EPA (1978) as evidence that we overestimated the settling velocity for fibers. However, this alleged discrepancy is due only to fiber size differences between our model assumptions and the model assumptions used by EPA. The EPA guidance document cited by Egilman and Schilling used fiber lengths of 1, 2, and 5 µm and an aspect ratio of 5:1, whereas we present results for fiber lengths of 5 and 10 µm, with aspect ratios of 12.5 to 25 for 5 micron fiber lengths and 25 to 50 for 10 micron fiber lengths. Fiber lengths of 5 µm and greater are considered the fibers that are biologically significant rather than those in the 1–2 µm range (Stanton, 1973; Stanton et al., 1977; Davis et al., 1978; Platek et al., 1985; ERG, 2003a, 2003b; Berman, 2010).

Using the modeled fiber deposition rates that EPA presented in their 1978 guidance document, fiber lengths of 1, 2, and 5 µm, with a 5:1 aspect ratio will have corresponding settling velocities of 0.00001, 0.000038, and 0.00019 m/sec, respectively. All of these settling velocities are slower than what we used in the Drivas et al. model (0.0028 m/sec). So, Egilman and Schilling are correct that we used a faster settling velocity than what is discussed in the EPA guidance document.

It is important to note, however, that the difference between EPA’s chosen velocities and the value that we used has virtually no impact on the ultimate output of our model. To evaluate this question, we put all of the settling velocities reported by EPA into the Drivas et al. (1996) model and the resulting predicted decrease in airborne asbestos concentration with distance from the source was essentially the same, regardless of the settling velocity that was used. This trend was consistent across all the distance categories considered in this analysis. The reason that settling velocity does not have a significant impact on predicted bystander concentrations is that, for our simplified scenario, these settling velocities are not high enough for deposition to have a greater impact on the predicted air concentrations compared to reflection. Table 2 illustrates the bystander:worker ratios that were calculated using both EPA and our selected settling velocities.

Table 2.  Comparison of bystander:worker ratios generated using the Drivas et al. model with fiber lengths of 1, 2, and 5 µm, with a 5:1 aspect ratio.

Scenario	Settling velocity (m/s)	Bystander:worker ratio
		1–5 ft	>5–10 ft	>10–30 ft	>30 ft
5 µm fiber length (Environmental Protection Agency [EPA], 1978)	0.00019	0.55	0.08	0.005	<0.00001
2 µm fiber length (Environmental Protection Agency [EPA], 1978)	0.000038	0.55	0.08	0.005	<0.00001
1 µm fiber length (Environmental Protection Agency [EPA], 1978)	0.00001	0.55	0.08	0.005	<0.00001
Original Donovan et al. 2011 5 µm and 10 µm fiber lengths	0.0028	0.54	0.07	0.004	<0.00001

Table 2 clearly indicates that for settling velocities of 0.003 m/s or less, there is virtually no difference in the predicted bystander:worker ratios. As a point of comparison, we also conducted an additional analysis where a significantly higher settling velocity was put into the Drivas et al. model (0.16 m/s for a fiber 10 µm long and 3 µm wide). Using this much higher settling velocity, there is a bigger difference in the predicted bystander:worker ratios, especially at distances greater than 10 feet from the source. For example, the calculated bystander:worker ratio at 1–5 ft was 0.46 for the highest settling velocity (0.16 m/s) compared to 0.54 based on our settling velocity of 0.0028 m/s. Differences in the ratios at >5–10 ft were 0.004 compared to 0.08 (0.16 m/s and 0.0028 m/s, respectively), and at >10–30 ft were 0.000002 compared to 0.004 (0.16 m/s and 0.0028 m/s, respectively).

Egilman and Schilling attempt to discredit our analysis by suggesting that our estimated airborne concentrations were inaccurate because we did not use the settling velocities suggested by EPA, but they failed to recognize that settling velocities below 0.003 m/s actually have almost no impact whatsoever on our estimated bystander:worker ratios. If we had used the settling velocities described in EPA (1978), our results would have been no different.

7)

Egilman and Schilling claim that “Moorcroft and Duggan (1984) did not formally study this [issue of falling rates], rather their results were ‘obtained in the course of other measurements [they had] been conducting” and that “these results are not applicable to the question of bystander exposures to a large point source emission of asbestos fibers.”

Egilman and Schilling make the argument that the results of the Moorcroft and Duggan (1984) study “are not applicable to the question of bystander exposures,” citing reports that they believe show that fibers remain airborne for a long period of time.

These reports are primarily nonpeer-reviewed guidance documents, all of which we also cite and discuss in our paper (EPA, 1978; Sawyer & Spooner, 1978; Committee on Indoor Pollutants Board on Toxicology and Environmental Health Hazards, 1981). Egilman and Schilling also claim that field studies exist that “are consistent with EPA’s conclusion that asbestos fibers remain airborne for long periods of time” but only cite a single unpublished study as support for this statement (Ewing, 2002). No acknowledgement of funding source is provided in this study, which should have been a concern to Egilman and Schilling given that the author (Ewing) has previously testified as an expert in asbestos-related litigation.

More importantly, Egilman and Schilling fail to acknowledge that the abovementioned models describe fiber sizes that are well below what is measurable by PCM analysis and, more importantly, do not have any biological significance. Our paper clearly states that we only evaluated “OSHA fibers” (i.e., greater than 5 µm in length, with a 3:1 aspect ratio), which are the fibers that are considered biologically relevant (Stanton, 1973; Stanton et al., 1977; ERG, 2003a; ERG, 2003b; Berman, 2010). In trying to use EPA’s modeled fiber settling velocities to refute our analysis (and ignoring the published field data), Egilman and Schilling focus their criticisms on the fibers too small to be important to human health.

In our paper, we cite two published studies that address the issue of fiber settling rates, both of which have shown that asbestos fibers appear to remain airborne for shorter time periods than would be expected based on model predictions (Corn & Stein, 1966; Moorcroft & Duggan, 1984). Egilman and Schilling attempt to discredit the conclusions drawn by Moorcroft and Duggan (1984) on the basis that their study was not specifically designed to study fiber settling rates, and they also claimed that the airborne asbestos concentrations measured in this study were too low to be reliable. They do not provide any commentary on the results discussed in Corn and Stein (1966). The results of both papers, however, warrant further discussion.

In their paper, Moorcroft and Duggan summarize airborne asbestos concentrations that were measured in four separate classrooms in a North London school, from which sprayed amosite ceilings had recently been stripped. The original intent of the study was to determine whether the classrooms could be reoccupied, and the testing was conducted in three periods: (i) baseline testing after the rooms had been thoroughly cleaned, no dust disturbance; (ii) continuous and vigorous dust disturbance; and (iii) no dust disturbance (similar to period 1, but immediately following the dust disturbance). While the authors explicitly state the study was not originally designed to evaluate settling rates, they used the data obtained in the second and third testing periods to characterize the removal of fibers from the air following the disturbance and to compare this value to theoretical estimates. Table 2 in this study provides calculated settling velocities (i.e. model estimates) for fiber diameters ranging from 0.5 to 3 µm. In order to compare their measured results to model estimates, Moorcroft and Duggan characterized the air samples collected in Period 2 of the study: “very few fibers [were] attached to dust particles and [there were] hardly any fibre clumps. Almost all of the fibers had an aspect ratio greater than 10:1 and there was a predominance of fine fibers; typically, less than 25% of the fibres counted a diameter greater than 1µm” (p. 456). So, while it is true that the original intent of the study was to determine whether classrooms could be reoccupied, the data collected during Periods 2 and 3 were sufficiently robust to allow them to study the issue of fiber deposition.

In their comparison of field data to modeled estimates, Moorcroft and Duggan noted that for fibers with a diameter of 1 µm or less, the factor by which the concentration would fall as a result of gravitational settling during a period of between 30 and 90 min was predicted to be less than 1.7. However, during post-settling phase, the authors noted that “fibre concentrations were seen to decline fairly rapidly, in some cases by a factor of more than ten” and that “the residence time for the majority of the fibres made airborne during our measurements was less than 1 h [one hour]” (p. 457). Based on the results of their measurements, Moorcroft and Duggan concluded that “theoretical estimates of the reduction expected from gravitational settling and ventilation suggest that these two mechanisms of fibre loss from the air were not sufficient to account for all of the observed reductions in concentrations (p. 457).” In short, the authors note that fibers settled out of the air more quickly than models predicted.

Regarding Egilman and Schilling’s claim that these data are unreliable because the concentrations were too low, we note that five out of eight dust disturbance tests resulted in airborne asbestos concentrations above the PCM sensitivity limit of 0.01 f/cc. These measured fiber concentrations decreased by an order of magnitude in the hour following the disturbance phase. A statistical t-test of the differences in reported airborne concentrations between the disturbance phase and the fiber settling phase indicates a statistically significant decrease in fiber concentrations between these two phases for four of the eight tests conducted (p = 0.007 to 0.05). Although the PCM method may not be sufficiently sensitive at the concentrations reported in this study to reliably detect small changes in airborne asbestos concentration, differences of the magnitude reported should not be discounted.

Egilman and Schilling did not acknowledge in their letter that we also cited another well-respected paper that presented field measurements indicating that particulate matter is unlikely to remain airborne for as long as some simple models estimate (Corn and Stein, 1966). Specifically, Corn and Stein state that “after the particles are airborne, room air currents could, theoretically, keep unit density particles as large a 90 µ [microns] aloft. However, we find very few airborne particles as large as 5 µ. The mechanisms responsible for this discrepancy are not fully understood” (p. 53). Since the publication of that paper, several textbooks have discussed the movement of particles in air and the many mechanisms of particle removal from air, including coagulation/agglomeration, condensation, impaction, centrifugation, Brownian motion, and diffusion (Reist, 1984; Hinds, 1999); all of which can help to account for why theoretical predictions of settling velocity can be much higher than what is actually observed in the workplace. Thus, it seems apparent that our current, relatively simplistic models of particulate movement in air do not yet take into account some of these more complex methods of particle removal from the air.

Obviously, there are many factors to consider when evaluating the distance that OSHA fibers will travel in air and the duration of time that these fibers will remain airborne. A carefully executed field study would be helpful. Nonetheless, it is apparent from the literature that many factors play a part in removing OSHA fibers from the air. Thus, models that consider only settling velocity will usually significantly overestimate the time that fibers of this size will remain airborne. Although many factors will influence settling time and travel distance, we believe that our “rule of thumb” remains a reasonable approximation of what can be expected in most workplaces.

8)	Egilman and Schilling state that “Donovan et al.’s analysis does not address fiber widths [<0.1 µm] that can cause mesothelioma and therefore does not assess mesothelioma risk from bystander exposure to asbestos.”

Egilman and Schilling are surely aware that fiber dimension (length and width) has long been recognized as an important factor in defining the potency of asbestos, regardless of fiber type. We know of no reliable source, however, that states that fibers must be less than 0.1 µm in width to pose a mesothelioma hazard. Since the late 1940s, a substantial amount of research has been conducted in which fiber dimension has been either directly or indirectly evaluated with respect to it biological significance. Specifically, beginning in the early 1970s, Stanton and colleagues conducted a series of animal studies with asbestos and nonasbestos fibers of varying widths and lengths (Stanton, 1973; Stanton et al., 1977, 1981), and ultimately concluded that “carcinogenicity correlates best with increasing numbers of fibers having both diameters of 0.25 µm or less and lengths of more than 8 µm and that the correlation diminishes with fibers of greater diameter or lesser length” (Stanton et al., 1981, p. 965). In 1978, Davis et al. conducted a study where rats were exposed to commercially used amosite, chyrsotile, and crocidolite. The experiment was designed to evaluate the difference in pathological effects of the three fiber types when exposed to either equal fiber mass or equal fiber number. The administered doses were further evaluated using scanning electron microscope in order to better characterize fiber size and it was shown that the chrysotile doses, which were associated with the highest rate of lung fibrosis, also contained “many more fibres over 20 µm long than either of the amphibole clouds. The results, therefore, support previous suggestions that long asbestos fibres are more dangerous than short” (p. 673). Later, Platek et al. (1985) conducted a study in monkeys and rats to determine the biological effects after inhalation of chrysotile fibers less than 5 µm and reported that “no significant differences in histochemical data were seen between the exposed and control groups” (p. 327).

More recent work provides additional support to what was learned in prior animal studies. For example, in 1995, Berman et al. evaluated data from 13 rat inhalation bioassays in which the animals were exposed to nine different types of asbestos dusts (Berman et al., 1995). The authors concluded that structures contributing to lung tumor risk appeared to be long (≥5 µm) and thin (0.4 µm) fibers, and further noted that potency appeared to increase with increasing length, with structures longer than 40 µm approximately 500 times more potent than those between five and 40 µm in length. In addition, modeling results reported by Miller et al. (1999) indicated that the concentration of fibers longer than 20 µm and thinner than 1 µm in diameter is most influential for determining the tumorogenic potential of fibers (Miller et al., 1999).

In their letter, Egilman and Schilling cite Lippmann (1990) to support the claim that fibers must be less than 0.1 µm in width in order to cause mesothelioma. The exact sentence in the Lippman paper reads: “In my recent review on asbestos exposure indices, I showed that asbestosis was most closely related to the surface area of fibers longer than about 2 µm and thicker than about 0.15 µm; mesothelioma to the number of fibers longer than about 5 µm and thinner than about 0.1 µm …” (Lippman, 1990, p. 311). Lippmann then cited his 1988 review, which Egilman and Schilling did not identify. In Table 3 from that 1988 publication, Lippmann reported that mesothelioma induction is most closely associated with the number of fibers longer than ~5 µm and thinner than ~0.1 µm (Lippmann, 1988). Lippman went on to state that “… if there is negligible potential for exposure to fibers longer than 5 μm, there would be virtually no risk of either mesothelioma or lung cancer” (p. 103). In a later paper that evaluated fiber length distribution data from rat inhalation studies using amosite, brucite, chrysotile, crocidolite, erionite and tremolite, Lippmann concluded that the concentration of fibers longer than either 10 or 20 µm in length was a better predictor of tumor yield than is the concentration of fibers longer than 5 µm (Lippmann, 1994).

Egilman and Schilling’s implication, then, that fiber diameter was the only critical feature identified by Lippmann is both erroneous and misleading. Clearly, like many others who have studied this topic, Lippman recognized the importance of fiber length in determining potency (Stanton, 1973; Stanton et al., 1977; Wagner et al., 1980; Lippmann, 1988; ATSDR, 2002; ERG, 2003a, 2003b; Berman, 2010). Egilman and Schilling appear to try to focus on fiber diameter in order to buttress their claim that very small fibers, which do tend to remain airborne for longer periods of time, should have been the focus of our analysis. The size fractions of fibers about which they seem to be concerned are considered too small to be biologically important.

Although Egilman and Schilling may wish readers to believe that our estimates of concentration with distance would underestimate the health hazards associated with these exposures, in fact, the opposite is true. By focusing on fibers that are 5 µm and 10 µm in length, our analysis considers the fibers that dictate the carcinogenic risk (and would also have a larger aerodynamic diameter). This conclusion is supported by a recent paper by Berman that discusses more than a dozen studies, and concludes that “carcinogenicity is mediated by a structure’s size. These studies suggest, for example, that potency increases with increasing length and thinner structures may contribute more to potency than thicker structures” (Berman, 2010, p. 154). In short, there has been a growing belief that mesothelima is caused by asbestos fibers that are longer than 10–20 µm. Width appears to be secondary importance compared to length and aspect ratio.

9)	Egilman and Schilling note that “Gibbs et al. (1990) studied the type and size of asbestos fibers found in the lungs of mesothelioma cases. The range of mean fiber widths they found for chrysotile, crocidolite and amosite was 0.06-0.07, 0.08-0.11, and 0.16-0.43 µm, respectively.”

This anecdotal observation by Egilman and Schilling is apparently intended to suggest that our model is unable to address mesothelioma hazard, since the particles that we discussed typically had widths greater than those reported by Gibbs et al. (1990). This statement gives the impression that we erred when, in fact, the foundation of Egilman and Schilling’s claim is without substance. First, Gibbs et al. (1990) never claimed that the fibers they found caused mesothelioma (Gibbs et al., 1990), as it not clear whether the fibers that were detected were actually the same that caused the disease or if they were biologically unimportant remnants. Indeed, most chrysotile fibers only have a biological half-life of less than 30 days in the lung (Bernstein et al., 2005, 2010). Second, no mention is made by Egilman and Schilling about fiber length, the parameter generally considered equally, if not more, important than fiber width in determining fiber potency.

Regardless of what fiber width is most predictive of mesothelioma risk, in the context of this analysis, fiber width only matters in one of the two models that we used. As described previously, fiber width is used to estimate the settling velocity used in the Drivas et al. (1996) model. The Keil and Ten Berge (2000) model does not consider fiber removal from the air from deposition or reflection, therefore fiber width has no impact on the results. As a part of this response, however, we re-ran the Drivas et al. (1996) model to account for a much wider range of fiber widths (0.05–3 µm). For fibers 5 µm long, and with widths ranging from 0.05 to 0.1 µm, corresponding settling velocities ranged from 0.00011 to 0.0004 m/s. As stated previously, deposition does not have a significant impact on predicted air concentrations for our simplified scenario for settling velocities below 0.003 m/s. Therefore, even if we had evaluated smaller fiber widths, our estimated bystander:worker ratios would have been the same.

10)	Egilman and Schilling assert that we “should have used utilized [ranges]” for the “rule of thumb.”

As we clearly state in our paper, we considered offering up a range of estimates for each distance category. We chose to not to do so for several reasons. First, the purpose of the “rule of thumb” is to help an exposure or risk assessor to quickly estimate the likely reductions in airborne asbestos concentrations that will occur at varying distances from a source. For the stated purpose, we believed that simplicity was important. Second, it is misleading to attempt to imply that one can characterize the uncertainty with that much precision when the vagaries of air movement and deposition (and agglomeration) inherent in a workplace can make it difficult to identify concentrations to within a factor of two in the breathing zone of a person, at low concentrations (as evidenced by the fact that at such concentrations, the difference between a left and right lapel sample often varies by this much or more). Third, similar differences in concentrations can occur based on inter-individual approaches to performing certain tasks.

Our motivation for undertaking this analysis was simple. Over the 80-year history of the practice of industrial hygiene, no one has attempted to offer any quantitative guidance about estimating fiber concentration for bystanders. As we stated in our paper, “given that there is currently no guidance offered for estimating historical exposures of workers in the typical ‘open industrial environment,’ we believe that providing this analysis of both field data and current dispersion models will be helpful to those involved in estimating asbestos concentrations in workplaces for which no quantitative exposure data exist, such as in historical exposure reconstruction efforts” (p. 72).

11)

Egilman and Schilling state that they “question the validity of a model which was specifically developed for defending an asbestos lawsuit, then used as the foundation for a review which was also conducted for and supported by defendants in asbestos litigation, and omitted entirely from the review’s references.”

We mistakenly did not include the Drivas et al. (1996) citation in our list of references. The full citation is as follows: Drivas PJ, Valberg PA, Murphy BL, and Wilson R (1996). Modeling indoor air exposure from short-term point source releases. Indoor Air 6(4):221–227.

Egilman and Schilling’s suggestion that the model described in Drivas et al. (1996) is invalid simply because it may have been developed for or used in litigation is ridiculous. The Drivas et al. (1996) model is an extension of the basic model for an instantaneous release of a pollutant that includes a term for deposition and correction factors for reflection from room walls, which can be readily seen by comparing Equation 1 from page 57 of our paper, the model without reflection and deposition, to Equation 2, the model with reflection and deposition. In addition, Drivas et al. (1996) compare the results of their model to measured data from Cooper and Horowitz (1986), and found that their model estimates were close to the measured results (Cooper & Horowitz, 1986; Drivas et al., 1996).

The Drivas et al. paper is cited in the chapter on eddy diffusion models in Mathematical Models for Estimating Occupational Exposure to Chemicals, second edition (Keil et al., 2009), a book published by AIHA that has become the main toolkit used by most industrial hygienists and risk assessors looking for methods for estimating occupational exposures. In addition, Dr. Mark Nicas, an acknowledged expert in indoor air modeling from the University of California, Berkeley, has used the deposition term from the Drivas model as part of the random walk models that he has developed and discussed in several peer reviewed publications (Nicas, 2000, 2001; Nicas & Jayjock, 2002; Nicas & Armstrong, 2003). The Drivas et al. model has also been used in a more recent paper (Cheng et al., 2011), in which the authors compared model estimates of CO concentration throughout a room to measured values as part of a model validation study.

12)

Concerns regarding funding: We would like to address the fact that Egilman and Schilling printed only half of our funding acknowledgment, and did so in a misleading way. By stating that our review was “done in preparation for litigation and was supported by a number of companies that have been, and are, involved in asbestos-related litigation,” they seem to want to convey that our entire analysis was done in preparation for litigation and, therefore, is biased. In fact, much of the work of analyzing all of the published scientific literature was not funded by any outside source, including our subsequent modeling and derivation of a proposed “rule of thumb.”

Our full acknowledgment, as printed in the journal (p. 72), reads: “A small portion of the research associated with this work was originally done in preparation for litigation and was supported by a number of companies that have been, and are, involved in asbestos litigation. The authors were not compensated for most of the work of compiling the information for this article, or for preparing it. One of the authors (D.J.P.) has served as an expert in asbestos-related litigation.”

As far as the claim that this paper was created with the sole purpose to “solicit business for ChemRisk employees,” Egilman and Schilling can hardly speak to our intent for undertaking this analysis. Our firm has long been committed to sharing information and ideas in the literature because we believe it is in the best interests of the industrial hygiene, exposure assessment, toxicology, epidemiology, and medical communities to have access to papers such as ours that discuss and analyze topics relevant to these fields. We are proud of our firm’s long history of supporting research performed by our staff, which is often conducted at our own expense. Numerous experts, both for the defense and the plaintiffs, as well as academics, can rely upon this paper, as well as hundreds of others published by our staff on a variety of topics.

We readily welcome a scholarly exchange about the movement of particles in the workplace and we plan to continue to study this topic. We recognize that more information is always helpful for building confidence in any modeling approach. As such, we stand behind the final recommendation presented in our paper: “We suggest that future studies incorporate air sample collection at numerous fixed distances from the source (north, south, east, west), along with consideration of fiber length and aspect ratio” (Donovan et al., 2011, p. 72).

Declaration of interest

Two of the authors (DP and JS) have served as experts in asbestos-related litigation. ED and PS are health scientists at ChemRisk and have not previously served as experts in asbestos-related litigation.