Morfeld, Peter; Taeger, Dirk; Mitura, Heike; Bosch, Axel; Nordone, Adrian; Vormberg, Reinhard; McCunney, Robert; Merget, Rolf
Adverse chronic respiratory health effects because of amorphous silica have been described in workers with exposure to naturally occurring silica in the diatomaceous earth industry.1,2 Such effects have not been observed with synthetic amorphous silica (SAS) in a limited study.3 Naturally occurring forms of amorphous silica are often contaminated by crystalline silica. Only a few studies have assessed exposure to SAS dust, and to our knowledge no study has reported quantitative exposure levels. This investigation estimated historical occupational exposures to two polymorphs of SAS—pyrogenic and precipitated (CAS # 7631-86-9, EINECS # 231-545-44,5).
Synthetic amorphous silica is a form of silicon dioxide (SiO2) that is intentionally manufactured. It is produced by thermal (pyrogenic) or wet (precipitated) processes. In the initial particle formation step, primary particles with dimensions less than 100 nm are formed by nucleation, coagulation, and coalescence. These primary particles by covalent bonding form indivisible units, the aggregates, that have external dimensions typically more than 100 nm. They are fused together with no physical boundaries among them. The aggregates combine to form agglomerates in the micron size range by physical attraction forces (van der Waals and H-bridges). According to the European Union (EU) Recommendation Definition for “nanomaterial,”6 “particle” means a minute piece of matter with defined physical boundaries. “Aggregate” is defined as a particle comprising of strongly bound or fused particles, and “agglomerate” means a collection of weakly bound particles or aggregates where the resulting external surface area is similar to the sum of the surface areas of the individual components. An identifying feature of SAS powder is that it is on the market as micron-sized agglomerates. Synthetic amorphous silica has been produced and marketed for decades without significant changes in its physical–chemical properties.4,5
Both forms of SAS have similar physicochemical properties and toxicological behavior4,7 and are soluble in biological media.4(p.22)
Because of its morphology, SAS is affected by the evolving discussion about nanomaterials. No harmonized definition of nanomaterials, however, exists worldwide although many regulatory industry and nongovernmental organizations have their own definitions of nanomaterials. The International Organization for Standardization (ISO) developed a taxonomy of terms and definitions for nanomaterials.8–10 The ISO defines nanomaterials as a “material with any external dimension in the nanoscale or having internal structure or surface structure in the nanoscale.” Nanoscale is defined as a size range from approximately 1 to 100 nm. Nanomaterials comprise nano-objects and nanostructured materials. The identifying feature of nanostructured materials is that their internal or surface structure is in the nanoscale, but their external dimensions are greater than the nanoscale. The definition is covered by working draft document ISO/WD TS 80004-2 (former ISO/TS 12921).11 Pyrogenic and precipitated SASs are nanostructured materials according to the ISO definition with an internal structure in the nanoscale, the primary particles. These primary particles do not exist in the products in isolation, but as aggregates/agglomerates, with external dimensions more than 100 nm.4,5 Mean aerodynamic diameters of SAS were estimated to be 200 μm on the basis of analyses of the product as delivered (see Table 12 in ECETOC).4 Further descriptions of production processes and use of SAS are given in ECETOC4 and Fruijtier-Pölloth.5
The EU Commission has issued a recommendation on the definition of nanomaterial6 under regulatory aspects as a reference for determining whether a material should be considered as a nanomaterial. The Commission recommended to adopt this as an overarching classifier for all EU-based regulations. The definition is based solely on the size of the constituent particles of the material, without regard to hazard or risk. A nanomaterial as defined in this recommendation should consist of 50% or more of particles by number size distribution, having a size between 1 and 100 nm. The definition also includes particles in agglomerates or aggregates whenever the constituent particles (ie, in the case of SAS, the primary particles) are in the size range from 1 to 100 nm. Synthetic amorphous silica is considered a nanomaterial under the current EU Commission Recommendation definition. Image analysis by transition electron microscopy confirms this conclusion.4,5
Synthetic amorphous silica needs to be distinguished from crystalline silica in terms of its solubility and potential adverse health effects. Whereas exposure and health effects from exposure to crystalline silica dust have been identified and studied for decades, occupational health data for chronic SAS dust exposure are limited. The International Agency for Research on Cancer (IARC) has classified crystalline silica as a human lung carcinogen (group 1); however, the IARC has classified amorphous silica as a group 3 substance (ie, “not classifiable as to its carcinogenicity to humans”).12 Cancer risk has not been raised in conjunction with SAS exposure; thus, the major interest related to exposure to SAS dust is the potential for nonmalignant respiratory effects. The assessment of published results on respiratory diseases in SAS dust-exposed workers, however, has been insufficient for drawing robust conclusions.4 This situation may be because the amorphous forms have never drawn attention given their low toxicity potential (no known specific toxicity, amorphous structure, and solubility of SAS). Thus, it is not surprising that the exposure to SAS was rarely reported in the scientific literature in comparison to crystalline silica, and estimates, when given, are rather crude.2
Reported SAS exposure data are typical examples of underdeveloped exposure data matrices13 as there exist no comprehensible measurement data covering the exposure periods of the workers, jobs, and tasks appropriately.1,2 This sparse data situation can lead to underestimated or missed associations.14 Uncertainties15 are usually ignored in retrospective exposure assessments, which result in overly precise and potentially biased risk estimates. It seems inadequate to develop only one exposure estimate for each worker and to accept these data as highly precise and without bias as usually assumed in epidemiological studies.16,17 Probabilistic procedures are well developed and easily implemented in standard software packages; these procedures can be used to evaluate multiple-exposure scenarios in terms of Monte Carlo–multimodel analyses or Bayesian approaches.18–20 These methods are promising ways to address the challenges associated with conducting retrospective exposure assessments in the presence of sparse exposure data.
To fill this gap about potential adverse health effects from occupational SAS dust exposures, we conducted a cross-sectional study of workers with long-term exposure to SAS dust at five German production plants. Occupational epidemiology studies often suffer from insufficient exposure assessments.16,21 The aim of the study was to assess exposure to SAS dust by two different methods and to evaluate certain medical endpoints, such as symptoms and lung function, while addressing risk factors such as smoking and atopy. In this report, we used two independent procedures to estimate cumulative exposures to inhalable SAS dust. We describe and analyze differences between the results from both procedures. In a follow-up article, we apply the varying exposure estimates in the analyses of health effects.
MATERIAL AND METHODS
Plants and Study Group
The study was performed at five German plants that produced pyrogenic or precipitated SAS. An overview is given in Table 1.
All current full-time employees in 1997, who worked for at least 1 month at the plant, were eligible to participate in the study. In total, 522 took part in the study. Because the number of female workers was small (n = 30), the study focused on exposed male workers. For eight workers, no exposure data were identified, so the cohort comprised 484 male workers.
We used two different exposure methods, leading to two different job-exposure matrices (JEMs). In procedure P1, expert ratings were used on the basis of individual work histories for each plant to estimate the cumulative inhalable SAS dust exposure in mg/m3·yrs for each of the 484 study participants. Procedure P2 used recent SAS dust measurements at the end of the cross-sectional epidemiological study to estimate the historical exposures backwards, using information on job categories as well. The recent measurements were applied to anchor the backward estimates.
Exposure Assessment Procedure P1: JEM and Cumulative Inhalable SAS Dust Exposure Estimates
To assess exposure to SAS dust, complete individual work histories of the study group were reconstructed using data from personnel files of the five plants. In cooperation with industrial hygienists and experts from these plants, an exposure categorization scheme for SAS dust was developed for each plant on the basis of estimates made by industrial hygiene experts of the participating companies. To harmonize estimates between the plants, estimated inhalable dust concentrations (in mg/m3) were used to build up four exposure categories (Table 2).
Exposures to SAS in different work areas, jobs, and periods were estimated by the industrial hygienists of the involved study plants (Table Aa to Ae, see the Supplementary Digital Content Appendix, http://links.lww.com/JOM/A146). By assignment of the exposure concentration estimates to the individual work histories, an estimate of the cumulative exposure to inhalable SAS dust (sum of the average exposure level of each category times duration of employment in this exposure category) was obtained for each worker. We describe this approach as exposure assessment procedure P1.
Assessment Procedure P2: JEM and Cumulative Inhalable SAS Dust Exposure Estimates
For the P2 exposure assessment, we used individual work histories of the study population on the basis of personnel files of the five plants. The same primary documents were used as applied in procedure P1.
On the basis of expert discussions, job categories were identified and harmonized across all five plants to derive similar exposure groups (SEGs) (Table 3).22
Changes in production, ventilation, housekeeping, etc, were described in detail at each plant on a yearly basis for each exposure group (SEG) by industrial hygiene and plant experts (coordinated/performed by authors PM, HM, AN, and AB). Some industrial hygienists were identical to those who were active in P1. Whereas the assessment in P1 was crude without documenting details of changes at the plants, the experts in P2 had to describe in detail and in a structured form what kind of changes in production, ventilation, housekeeping, etc, had occurred. For this additional effort, other knowledgeable plant personnel were consulted. The results were documented in detail for each plant.
On the basis of these documents, the experts assessed SAS dust exposure levels back to the start of SAS production at each plant and for each SEG (and in some instances forward up to 2000) by relative scoring.23 These relative estimates were anchored with the current inhalable SAS measurement data (1997 to 2000) to derive inhalable SAS concentration estimates for the whole period of interest.15
Personal inhalable SAS dust concentrations were measured by one team at plants 1, 3, and 4 (GSP, flow 3.5 I/min, quartz filter).24 Study teams at plants 2 and 5 were trained by the crew performing measurements at plants 1, 3, and 4 (first team, headed by author RV). Teams at plants 2 and 5 used the same devices that were used by the first team. Current personal dust exposures of all study participants were measured repeatedly (full shift, if possible). Each measurement was individually documented. Every study participant was measured twice. If the study participant was unable to work, his substitute was measured. Depending on the results of the two measurements, a third measurement was performed (eg, if measurements for a subject seemed to be unstable, a third or fourth of even fifth measurement was made). All available measurements that passed this quality check were then included. Measurements were carried out between 1997 and 2000.
To account for uncertainty in the exposure estimates, we performed multiple SAS dust exposure assessments in P2. First, we analyzed measurement data of inhalable SAS dust concentrations by the plant and SEG and calculated five statistical measures used as anchor values for current (1997 to 2000) concentration levels: (1) p25 (first quartile), (2) median, (3) p75 (third quartile), (4) the geometric, and (5) arithmetic means. Second, we assessed uncertainties in scoring by the experts by estimating not only a medium dust level (1) change but also high (2) and low (3) relative SAS dust levels. This approach resulted in 15 exposure scenarios on the basis of measurement data and on expert assessments (Table 4). Thus, 15 basic JEMs were constructed.22 The 15 scenarios are denoted according to the combination of statistical measure and the type of background extrapolation (eg, p25 low, median medium, or p75 high).
Jobs and tasks were allocated individually to the SEGs across time on a yearly basis. These data were combined with the JEM exposure levels per plant, SEG, and across time to derive the estimates of cumulative exposure to inhalable SAS dust in mg/m3·yrs for every worker in the study. Thus, every study participant received 15 different basic cumulative exposure assessments (Table 7).
The median medium scenario was less affected by outliers in measurements than the mean medium scenario and performed better in representing the SEGs in the expected order (Tables Ba to Be, see the Supplementary Digital Content Appendix, http://links.lww.com/JOM/A146). Thus, the median medium scenario was chosen to serve as the leading P2 exposure scenario when comparing with P1. This leading P2 scenario is anchored at the median of the inhalable SAS measurements per SEG and plant and is based on the medium backward extrapolation estimate of the experts. In the following text, we refer to it by “leading P2 scenario” or “median medium P2 scenario.”
Distributions of exposure variables were plotted and described by calculating appropriate statistics. Bland–Altman plots were applied to explore differences between exposure estimates from P1 and P2.25,26 Distributions were compared with Kolmogorov–Smirnov tests.27 We chose a significance level of 5%. All analyses were done with Stata 1028 and SAS 9.3 (SAS Institute Inc, Cary, NC).
The study group comprised 484 exposed male workers (participation rate 95%). Table 5 gives an overview of subject characteristics. The distribution across plants was as follows: 168 workers at plant 1, 31 at plant 2, 172 at plant 3, 39 at plant 4, and 74 at plant 5.
The median age was nearly 40 years at the examination date; median duration of employment was 12 years. The range of date of hire spans more than four decades, but the date of termination over one decade.
For exposure assessment procedure P2, 1375 inhalable SAS dust concentration measurements were performed and used for anchoring the exposure assessment (Table 6).
The current median dust concentration was less than 2 mg/m3 even in the job category “bagging”. Tables Ba to Be (see the Supplementary Digital Content Appendix, http://links.lww.com/JOM/A146) report on relevant statistics of the distribution of the measurement values by plants and SEGs.
Estimates of average SAS concentrations by calendar year from assessment procedures P1 and P2 (median medium scenario) are reported in Fig. 1. Only periods with exposure to inhalable SAS concentrations of more than 0 mg/m3 were evaluated to produce this figure. Figure 1 describes average concentrations across all job categories/SEGs and plants under observation during the respective calendar year and, thus, is affected by varying combinations of job categories/SEGs and plants in time. The P1 scenario yielded concentrations less than 9 mg/m3 throughout and was rather constant across calendar time when compared with the leading P2 exposure scenario (median medium). Estimates as high as 25 mg/m3 (or even higher) were reported by the leading P2 scenario in the early 1960s. A pronounced decrease in exposure after 1966 is described such that both exposure scenarios cross in the mid-1970s with median medium P2 concentrations remaining lower than P1 estimates thereafter. At the end of the study period, P1 and median medium P2 concentrations were less than 1 mg/m3.
A comparison of cumulative exposures to inhalable SAS dust calculated from both exposure scenarios for the 484 study members up to the medical investigations is given in a Bland–Altman plot (Fig. 2). P1 and median medium P2 cumulative exposure estimates differed considerably. There is substantial heteroscedasticity (ie, higher mean values of both procedures have higher variance).
Table 7 notes the distribution of total cumulative exposures to inhalable SAS dust for all exposure scenarios considered. P1 cumulative exposure estimates had a mean of 56.9 mg/m3·yrs (range, 0.1 to 419); the leading P2 (median medium) estimates were on average 31.8 mg/m3·yrs (range, 0.4 to 480). A Kolmogorov–Smirnov test proved the difference of these distributions as clearly significant (P < 0.0001). A cumulative exposure of 80 mg/m3·yrs corresponds to an exposure to inhalable SAS dust of 2 mg/m3 over 40 years, and this is well covered by the distribution of the P1 exposure scenario and the leading P2 scenario (median medium). Table 7 reports also on all estimates returned by the multiple exposure assessment procedure P2. Mean estimates varied between the 15 P2-scenarios from 12.6 (p25 low) to 109.6 mg/m3·yrs (p75 high). Even the medians varied from 5.8 (p25-low) to 19.3 mg/m3·yrs (mean high).
This study estimated cumulative exposures to inhalable SAS dust in 484 exposed male workers from five German (pyrogenic and precipitated) SAS-producing plants by two independent procedures. P1 was based on expert estimates of current and historical exposures in job categories that differed between plants. To harmonize data between plants, assignments to estimated dust concentrations were made and agreed by industrial hygienists of all plants. P2 was a multiple exposure assessment (15 exposure scenarios) anchored by a recent measurement series and expert assessments of historical exposure levels after harmonizing the job categories across plants. In contrast to P1, the P2 historical estimates were based on detailed descriptions of changes in production, ventilation, housekeeping, and other control measures specific to plants and SEGs.
In this article, we compared two retrospective exposure assessment methods for SAS developed for the same cross-sectional morbidity study. Such comparisons of different exposure assessment methods were indicated and performed in multicenter pooling projects of lung cancer case-control studies to tackle with inherent uncertainties in exposure assessment because of the different source populations and research teams involved (eg, Mannetje et al29 and Peters et al30). We used this approach here to identify, estimate, and discuss the uncertainties in a single study. In a subsequent article, we will report epidemiological analyses of symptoms, lung function, and chest radiographs in association with exposure to inhalable SAS dust, and we will evaluate the impact of the different exposure assessment procedures on the epidemiological findings.
The P1 exposure assessment procedure is often applied in occupational epidemiology, if exposure data are lacking. In contrast, the more complex P2 procedure covers multiple exposure assessments on the basis of a large number of recent personal dust exposure measurements. These cross-sectional measurements anchor the P2 backward extrapolation necessary for historical exposure reconstruction. Anchoring of retrospective exposure estimates by actual measurements is suggested by exposure scientists, particularly in a study covering different plants.31 Moreover, P2 included detailed documentation of relevant changes in production, ventilation, housekeeping, etc, described in detail for each plant and exposure group (SEG). These changes were noted on a yearly basis by industrial hygiene and plant experts.
Thus, the major difference between P1 and P2 is that P2 uses recent exposure measurements as a starting base and then assesses various changes at the plants on the basis of detailed documentations to eventually lead to multiple exposure categories.
Considerable differences between both JEMs were observed between P1 and the leading P2 procedure (median-medium). Between the mid-1970s and at the start of this cross-sectional study, differences between P1 and P2 were negligible; before 1975, however, differences were substantial. This difference may reflect that with passage of time it is much harder to be accurate about past exposures. Usually, it is inappropriate and against evidence to assume that exposure concentrations were invariant across time and to simply allocate recent exposure concentration data to historical exposure periods.32 P2 addressed this challenge by a detailed assessment of important changes at the plants that may have affected SAS inhalable dust concentrations. P1 relied on experts' educated guesses only and their knowledge of few available historical measurements. Nevertheless, challenges of assessing past exposures apply to both procedures.
While the P1 procedure is common, the multiple exposure scenario in the P2 approach is rarely used although well suited for situations with sparse exposure information. Possible reasons not to apply multiple exposure assessments more regularly are the assumed additional efforts required in data collection and epidemiological analyses. Nevertheless, data collection is only marginally changed if experts not only give point data but also report on uncertainty of their exposure estimates. Measurement data can easily be summarized into different statistics. Probabilistic procedures can evaluate epidemiological studies with multiple exposure scenarios in terms of Monte Carlo–multimodel analyses or Bayesian approaches.18–20
When statistically anchoring the SEGs by measurement data in P2, we did not take account of repeated measurements within individuals. Thus, the data had some dependencies (clusters) that stabilized the estimates within the subjects, but were ignored in the estimation procedure. First, some workers contributed information to different SEGs when repeating the measurement. Second, it is unclear what kind of statistic should summarize the measurement data within the individual when calculating overall statistics other than an arithmetic mean. Third, the difference in mean values for plant–SEG combinations, with or without collapsing the data into one average for each worker in a first step, was rather small in comparison with the variance of the anchor values because of the different statistics used in anchoring. Fourth, statistically significant differences in average inhalable dust concentrations between and within the workers remained even after taking plants and SEGs as covariates into account. This motivated the multiple anchoring in P2 as performed.
This group-based procedure of anchoring may be refined by taking ordering information from the SEGs into account, but the group-based procedure as applied in this study seems to be appropriate because the measurement data sets are not too small.33
We did not use the sparse historical measurement data of inhalable SAS concentrations (data not shown) as these measurements represent very unusual and rare working conditions (worst-case scenarios; not representative for the working conditions) and were performed with different methods and sampling strategies. There was no obvious association between historical measurements and the present exposure assessments reported in this article (P1 and the leading P2 median medium scenario). Anyhow, such data seem to be compatible to our p75 high scenario and, thus, is potentially covered by P2 (Table 7). The backward extrapolation due to the p75 high scenario approximated these historical worst-case measurements rather well (data not shown). This indicates how a multiple exposure assessment can cover nonrepresentative measurements.
P2 seems prima facie to be preferable. It is anchored by a comprehensive set of representative SAS concentration measurements that characterized the exposure situation during the health survey. The backward extrapolation was performed on the basis of internal official protocols, which described in detail changes in production, ventilation, housekeeping, etc. On the basis of these data, relative concentration changes were estimated by engineers with an experience of decades in SAS production. The measurement data and the backward extrapolation were done in multiple ways covering uncertainties (multiple exposure assessment). On the contrary, P1 ignored the representative exposure measurements. A detailed evaluation of changes in the production etc was not performed, and just a single assessment was derived. P1 was much simpler to generate and of low cost in comparison with P2. Nevertheless, it is unclear whether P2 will lead to different results when applied in epidemiological evaluations.
Importantly, the different exposure assessments returned inhalable dust concentration and cumulative exposure estimates that varied considerably between the assessment procedures P1 and P2 median medium scenario, and varied significantly within the multiple P2 scenarios. Backward extrapolation over many decades introduced substantial differences between exposure estimates of both JEMs. In a following article, we will examine how differences between JEMs will affect the effect estimates of the health outcomes.
Synthetic amorphous silicas are nanostructured polymorphs of silicon dioxide. We compared two different exposure assessments for 484 male workers from five German SAS-producing plants. Both approaches suffer from considerable uncertainties that need to be considered in epidemiological studies by appropriate modeling. This is the first occupational study that reports quantitative exposure levels in workers exposed to synthetic amorphous silica. These data provide insight into workplace exposures of a nanostructured material.
We acknowledge the assistance of both technical and medical staffs of the plants and the support of the workforces. We thank Rolf Breitstadt, Evonik Industries, Thomas Brüning, IPA, and Kurt Straif, IARC, for their organizational support of this investigation. The study was sponsored by the Association of Synthetic Amorphous Silica Producers, a sector group of CEFIC.
1. McLaughlin JK, Chow W-H, Levy LS. Amorphous silica: a review of health effects from inhalation exposure with particular reference to cancer. J Toxicol Environ Health. 1997;50:553–566.
2. Merget R, Bauer T, Kupper HU, et al. Health hazards due to the inhalation of amorphous silica. Arch Toxicol. 2002;75:625–634.
3. Volk H. The health of workers in a plant making highly dispersed silica. Arch Environ Health. 1960;1:125–128.
4. ECETOC. JACC Report No 51. Synthetic amorphous silica (CAS No. 7631-86-9). Brussels: ECETOC; 2006:1–237.
5. Fruijtier-Pölloth C. The toxicological mode of action and the safety of synthetic amorphous silica—a nanostructured material. Toxicol. 2012;294:61–79.
8. International Organization for Standardization. ISO/TS 27687:2008. Nanotechnologies—terminology and definitions for nano-objects. Nanoparticle, nanofibre and nanoplate. 2008.
12. IARC. Silica, Some Silicates, Coal Dust and Para-Aramid Fibrils. Lyon: International Agency for Research on Cancer; 1997.
13. Seixas NS, Checkoway H. Exposure assessment in industry specific retrospective occupational epidemiology studies. Occup Environ Med. 1995;52:625–633.
14. Kriebel D, Checkoway H, Pearce N. Exposure and dose modelling in occupational epidemiology. Occup Environ Med. 2007;64:492–498.
15. Dahmann D, Taeger D, Kappler M, et al. Assessment of exposure in epidemiological studies: the example of silica dust. J Expo Sci Environ Epidemiol. 2008;18:452–461.
16. Checkoway H, Pearce N, Kriebel D. Research Methods in Occupational Epidemiology. New York, NY: Oxford University Press; 2004.
17. Loomis D, Kromhout H. Exposure variability: concepts and applications in occupational epidemiology. Am J Ind Med. 2004;45:113–122.
18. Burnham KP, Anderson DR. Model Selection and Multimodel Inference. A Practical Information-Theoretic Approach. New York, NY: Springer Science + Business Media, LLC; 2002.
19. Greenland S. Sensitivity analysis, Monte Carlo risk analysis, and Bayesian uncertainty assessment. Risk Anal. 2001;21:579–583.
20. Morfeld P, McCunney RJ. Bayesian bias adjustments of the lung cancer SMR in a cohort of German carbon black production workers. J Occup Med Toxicol. 2010;5:23. doi:10.1186/1745-6673-1185-1123.
21. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008.
22. Ignacio JS, Bullock WH. A Strategy for Assessing and Managing Occupational Exposures. Fairfax: AIHA; 2006.
23. Harber P, Muranko H, Shvartsblat S, Solis S, Torossian A, Oren T. A triangulation approach to historical exposure assessment for the carbon black industry. J Occup Environ Med. 2003;45:131–143.
24. Institut für Arbeitsschutz der Deutschen Gesetzlichen Unfallversicherung (IFA). Messung von Gefahrstoffen—IFA-Arbeitsmappe (Expositionsermittlung bei chemischen und biologischen Einwirkungen). Berlin: Erich Schmidt Verlag; 2013.
25. Altman DG, Bland AE. Measurement in medicine: the analysis of method comparison studies. Statistician. 1983;32:307–317.
26. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1:307–310.
27. Conover WJ. Practical Nonparametric Statistics. New York, NY: Wiley; 1999.
28. StataCorp. Stata Statistical Software: Release 10. College Station, TX: Stata-Corp LP; 2008.
29. Mannetje A, Fevotte J, Fletcher T, et al. Assessing exposure misclassification by expert assessment in multicenter occupational studies. Epidemiology. 2003;14:585–592.
30. Peters S, Vermeulen R, Cassidy A, et al. Comparison of exposure assessment methods for occupational carcinogens in a multi-centre lung cancer case-control study. Occup Environ Med. 2011;68:148–153.
31. Kromhout H, Oostendorp Y, Heederik D, Boleij JS. Agreement between qualitative exposure estimates and quantitative exposure measurements. Am J Ind Med. 1987;12:551–562.
32. Creely KS, Cowie H, Van Tongeren M, Kromhout H, Tickner J, Cherrie JW. Trends in inhalation exposure—a review of the data in the published scientific literature. Ann Occup Hyg. 2007;51:665–678.
33. Kim H-M, Park CG, van Tongeren M, Burstyn I. Estimation methods with ordered exposure subject to measurement error and missingness in semi-ecological design. BMC Med Res Methodol. 2012;12:135.
Copyright © 2014 by the American College of Occupational and Environmental Medicine