Occupational exposure to cobalt is well recognized in the hardmetal industry, which mainly produces cutting tools that are used in the manufacturing of industrial products and parts. The term “hardmetal” describes a group of composite materials that consists predominantly of the hard tungsten carbide (WC) particulate phase using cobalt as a binder, for some grades also including nickel and chromium.1 Occupational exposure during production of hardmetal is associated with several adverse health effects that suggest cobalt being the cause. Those effects include rhinitis, sinusitis, bronchitis,2 asthma,3 dose-related decreased lung function over time,4 and hardmetal lung disease (HMLD).5 Allergic dermatitis,6 cardiomyopathy,7 as well as increased incidences of ischemic heart disease8 have been reported. An increased risk of developing lung cancer has also been seen among people working in the hardmetal production industry.9–11 The International Agency for Research on Cancer (IARC) has classified cobalt metal with tungsten as probably carcinogenic to humans (group 2A).12
An international epidemiological mortality cohort study has been undertaken, based on persons currently working or ever worked in the hardmetal industry in the USA, UK, Germany, Austria, and Sweden. Results of other country-specific cohort analysis and exposure assessment for the US plants have been described in as a series of companion papers.13–18 Exposure assessment is a key element in exposure–response analysis of epidemiology.19,20 Methods have been developed for data management and exposure assessment, and statistical models have been employed in order to estimate exposures retrospectively for periods when measurement data are sparse or nonexistent. These models are based on current and available historical data. Assessment includes data acquisition, construction of a model, and evaluation of the measurement data to be used as the basis for exposure estimates. Analysis of variance, multiple linear regression, or mixed model analysis involving variables thought to determine exposure variability can also be used.21
This paper describes occupational exposure measurement data and a quantitative exposure assessment of cobalt, nickel, and tungsten for three Swedish hardmetal plants operating from the 1930s onwards. The exposure and area air concentrations, exposure measures as well as the impact of different exposure modeling approaches on the exposure–response analysis, in particular cobalt exposure and lung cancer mortality in the Swedish cohort, are described here.
Exposure assessment is performed for cobalt, nickel, and tungsten for workers at three Swedish hardmetal plants operating from 1930s and onwards. The exposure assessment is used in a Swedish mortality study in the hardmetal industry.18 The present detailed exposure study is based on a measurement database from 1970 and onwards. Exposure and area measurement data are presented, and exposure modeling using the database will enable individual exposure assessment of cumulative and mean cobalt exposure data. Different extrapolation assumptions for the time period before 1970 and their impact of exposure measures are carried out.
Study Objects and Processes
Three major Swedish hardmetal production sites were included, with companies A and B being rurally located and company C located in urban areas. These companies currently employ approximately 1340, 1440, and 350 white and blue-collar workers, and began operating in 1931, 1951, and 1954, respectively. The main products made at sites A and B are inserts for cutting applications. Site C also produces covered inserts for cutting applications, rock drillings, and wear parts, including hot rolls.
Up to 1990, the production of hardmetal at sites A and C included treatment of Scheelite (CaWO4), with hydrochloric acid (HCl) followed by dissolution of the formed ammonium- paratungstate (APT) in ammonium hydroxide (NH4OH). The APT was then precipitated by evaporation of the liquid. After calcination at 900 ° C, tungsten oxide (WO3) or blue oxide (WOx) was formed, and subsequently reduced to tungsten (W) in a hydrogen (H2) atmosphere. This step was followed by carburization, where tungsten and carbon black is merged in a heating process at 1600°C to 2400°C forming tungsten carbide (WC) powder. Tungsten carbide powder was then mixed with cobalt and other components according to a specific recipe, and wet milled in organic liquid for several hours for mixing and reducing the grain size. Spray drying was then used to remove the liquor from the slurry and to produce powder agglomerates. The powder was then cooled and packaged in drums or bags. Spray drying has been used since the 1970s. Before that, the milling liquid was evaporated in heated bowls. The introduction of spray drying resulted in considerable reduction in cobalt exposure. Next, the material was pressed, although extrusion or powder injection molding was also used to a lesser extent. The pressed pieces were then shaped and sintered. The products were then accurately grinded or machined into the desired shapes. The products were then sand blasted and covered with a protective coating layer. The coating was either chemical vapor deposition (CVD) involving a layer of titanium carbide (TiC), or physical vapor deposition (PVD) using TiC. In the final step, the finished products were inspected for quality, stored and shipped out of the plant.
Conventional hardmetal grades for insert production typically contain 80% to 94% tungsten carbide with 6% to 12% cobalt binder and often varying proportions of carbides of other metals such as titanium, tantalum, niobium, and chromium. Chromium carbide is added to the powder, but after sintering, the chromium is found in its metallic state. The content is typically 0.3% to 0.9%. Cermet grades for insert production are based on titanium carbonitride with cobalt (7% to 11%) and/or nickel (4% to 6%) binder. They typically also contain some tungsten carbide, tantalum carbide (TaC), vanadium carbide (VC), and molybdenum carbide (Mo2C).
Personal and area air measurement data were extracted from company records, covering a period from early 1970 to 2012. In total, there were 2683 personal samples and 1116 area measurements in the database. The majority of the personal samples related to levels of cobalt, tungsten, and nickel, that is, 1210 cobalt, 313 nickel, and 342 tungsten measurements, with the number of corresponding area measurements being 354, 154, and 58. The determinants available for all samples in the Swedish database were country, company, site, sampling date, department, job title, US job class title, exposure agent, agent concentration, sampling time, sampling volume, CAS number, particle size fraction, sampling device, sampling collection media, analytical method, level of quantification (LOQ) for the laboratory used, and occupational exposure limits (OELs), with all determinants being registered for each sample. The Swedish measurement database was primarily used as a basis for the exposure assessment in the Swedish mortality study,18 and also as a part of the international measurement database, used for the exposure assessment in the international pooled analysis (Marsh 2017, unpublished data) and the US study.13 All personal and area measurements, regardless of exposure time, were included in the measurement database and shown in full in Appendix 1.
Sampling and Analytical Methods
The sampling of the total dust was carried out according to a modified version of National Institute of Occupational Safety and Health Manual of Analytical Methods 050022 using an open-faced cassette (OFC) with a 25 mm cellulose acetate filter (Millipore 3 μm pore size [Merck Millipore, Billerica, MA]) and an airflow of 2.0 L/min. Inhalable dust was collected using a GSP filter head (GSA Messgerätebau GmbH, Gut Vellbrüggen, Germany) with a 37 mm cellulose acetate filter (Sartorius Stedim 8 μm pore size [Sartorius Group, Göttingen, Germany]) connected to a pump (GSA SG4000; Messgerätebau GmbH, Gut Vellbrüggen, Germany), operated at an air flow of 3.5 L/min.23 Determination of the mass of dust on the filters was made gravimetrically.
The vast majority of the samples in the measurement database were collected as total dust levels, with the metals subsequently determined in that particle fraction. A smaller number (50 of 1210) of the personal samples were determined from inhalable dust measurements and were subsequently transformed to cobalt in the total dust fraction. A 1 to 1 ratio was used, based on parallel stationary measurements.24 The same transformation was made in the US pooled measurement database, used for the US cohort study.
The analysis of metals was carried out using different techniques. From the 1970s up to 1992, the use of atomic absorption spectroscopy (AAS), predominantly with an LOQ of 1 μg /sample25 was used. From 1992 to 2006, X-ray fluorescence (XRF), with much the same LOQ level, was used.26 Such analysis did not require any dissolving of the sample, thus allowing for multiple analyses including wet chemistry. From 2006 and up to 2011, inductive coupled plasma spectroscopy (ICP) was used; from 2011, low-resolution ICP using mass spectroscopy for identification and quantification with an LOQ 0.05 μg/sample27 was primarily used.
Exposure Assessment by Job and Department
The job or departmental class for each worker along with time period were extracted from the personnel files and compared with company classification, which varied in resolution between different time periods. In the international study, some 69 job or departmental classes were identified.14 All Swedish workers were assigned a job class according to classifications from the international study (Appendix 2). On the basis of the number of cobalt measurements and the number for each job title occurring in the personnel files, a reduction of job classes to aggregated job classes was carried out.
For the Swedish cohort, aggregated job classes were defined on the basis of similar exposure group (SEG) considerations and measurement data related to job titles occurring in the measurement database. An SEG is defined as “a group of workers having the same general exposure profile for an agent because of the similarity of the materials and processes with which they work, and the similarity of the way they perform the task(s).”28 This leaves the aggregated job classes A–I for modeling and exposure–response analysis. Job class A was defined as background (unexposed, ie, office workers), B as intermittent low (foremen, engineers, material handler, assembly, marking, inspection, and packing), C as intermittent high (laboratory R&D, maintenance), D as powder production (weighing, mixing, spray dry, packaging), E as pressing, (pressing, forming, shaping), F as slow-moving operations, G as coating, H as rolls (big pieces), and I as grinding and honing.
Exposure Modeling and Measures
Using personal cobalt air concentrations, and available exposure determinants, we carried out a log linear multiple regression analysis, comprising aggregated job class, site, and time period, with cobalt concentrations as the dependent variable. Our model initially included five different time periods (1970 to 1979, 1980 to 1989, 1990 to 1999, 2000 to 2009, and 2010+), three different sites (companies A, B, and C), and nine categories of aggregated job titles, A to I, as independent variables. To improve the exposure assessment and make the exposure assessment used in the cohort study more comparable to earlier studies, modeling was also carried out with data for blue-collar workers, complied by aggregating job class D to I. No data were available for the time period 1950 to 1969, during which a large number of the workers studied started their jobs. We used three different extrapolation techniques for the periods before 1970, in order to carry out a sensitivity analysis to investigate the impact of different assumptions on our exposure measures used in our mortality study. In a standard exposure assessment procedure, the period before 1970 would have the same modeled exposure as the period 1970 to 1979. However, we also extrapolated model data by log linear regression analysis, the regression coefficients (B) values based on the average change per 10-year period, and used them to model cobalt exposure for the two periods 1950 to 1959 and 1960 to 1969. We also used exponential extrapolation for the period 1940 to 1969 to calculate historical measurement data. Data from the modeled cobalt exposure are presented. The exposure measurements were skewed, and required a natural logarithm transformation. The model used for estimating cobalt concentrations between 1970 and 2012 was:
For the different extrapolation scenarios, the b2-values were calculated for each decade between 1940 and 1969 and subsequently formed different model data. For all individuals in the cohort, the concentration of cobalt derived from the different log models for each time period, job and site, and the duration of exposure was used to calculate cumulative exposure measure in mg/m3*years (ie, exposure level multiplied by exposure time). The cumulative exposure of cobalt for each individual was calculated using:
where CE(j) is the individual cumulative exposure expressed as mg/m3*year, Ek(jk) is the estimated level of cobalt exposure for the jth individual during the kth time period, and T(jk) is the number of years at the exposure level prevailing for the kth time period.
The corresponding mean was calculated as:
The standard exposure measures used were ever/never exposed, duration of employment, and cumulative and mean exposure.18 Measurement data for nickel and tungsten were sparse and unevenly distributed between jobs and time periods, and a different approach was used (Appendix 3). On the basis of dichotomized exposure to tungsten, either 0.1 or less and more than 0.1 mg/m3, all sites were included and analyzed by cumulative exposures as well as by means, classified in quartiles. Measurement data for nickel were only available from one plant, and the exposure measure based on the dichotomized exposure for nickel was either 0.01 or less or more than 0.01 mg/m3.
The classification of nickel exposure was dichotomized based on median; exposure levels below 0.01 mg/m3 were assigned a value of 1 and exposures above 0.01 mg/m3 were assigned a value of 2 for each aggregated job class and time period. A corresponding classification (based on median) was set up for tungsten; exposure levels below 0.1 mg/m3 were assigned a value of 1 and exposures above 0.1 mg/m3 assigned a value of 2. Cumulative and mean exposures were calculated.
The results of the log linear modeling were used to determine cobalt concentrations for different time periods, jobs, and companies, and furthermore, cumulative cobalt exposures (mg/m3•years) and means (mg/m3) were calculated for the exposure response analysis in our mortality study in the Swedish hardmetal industry.18
In our mortality study, exposure to cobalt was defined as cumulative exposure in mg/m3•year (ie, exposure level times exposure time) and mean (cumulative exposure by duration of exposure) and then categorized in two different ways. The determined cumulative and mean exposures were categorized as quartiles, for both the total cohort and for blue-collar workers. Exposure was also categorized into three exposure groups that reflected exposures relevant to a 40-year working career at the present Swedish OEL, 0.02 mg/m3, defined as the 8-hour time weighted average air concentration (8-hour TWA). The cumulative dose values for the three categories were less than 0.20 (low), 0.21 to 0.40 (medium), and more than 0.41(high) mg/m3 year. The high exposure group reflects half of the maximum allowed lifetime exposure to cobalt, 0.8 mg/m3 year, which corresponds to 40 years of exposure at the present Swedish OEL for cobalt, that is, 0.02 mg/m3.29 The corresponding classes based on means would be 0.005 or less, 0.0051 to 0.010, and more than 0.01 mg/m3. In this paper, we compare the different quantitative exposure measures using our different extrapolation scenarios and discuss them with the perspective outlined above.
For descriptive purposes, standard parameters [arithmetic mean (AM), standard deviation (SD), geometric mean (GM), geometric SD (GSD), range] were calculated for the log normal distribution of all the measurements. The limit of detection (LOD) is defined as 3 SD for a concentration with signal-to-noise ratio of 3. The present detection limits are less than 40 μg for dust, less than 0.01 μg for cobalt, and less than 0.02 μg for tungsten. The corresponding air concentrations are less than 0.040 mg/m3 for dust, less than 0.00001 mg/m3 for cobalt, and less than 0.00002 mg/m3 for tungsten for an 8-hour full work shift sample and a flow rate of 2 L/min. Measurements below LOD were assigned LOD/√2 before the final calculations.30 Historically, the levels of detection and quantification were much higher for the metals and we therefore report a level of quantification for all cobalt air concentration as less than 0.001 mg/m3, even though modern analytical techniques enable a much lower LOD.
Standard linear regression analysis was carried out, including upper and lower limits for the confidence interval of the regression coefficient. Data for the exponential regression were calculated on the basis of R software. For the initial linear regression analysis, we used the Statistical Package for Social Sciences (SPSS) 22.0 (IBM Corp, Armonk, NY).
The underlying cohort study was approved by the Regional Ethical ReviewBoard, Uppsala, Sweden Dnr 2012/056.
The personal and area air measurement data were extracted from company records, covering a time period from early 1970 to 2012. The majority of the personal samples represented cobalt, tungsten, and nickel, with 1210, 313, and 342 exposure measurements, respectively (Table 1). The detailed measurement database is summarized in Appendix 1. The personal measurements with sampling time more than 360 minutes were for company A 75%, B 100%, and C 72%. The personal exposure cobalt concentration levels varied between less than 0.0001 and 2.8 mg/m3, with median and AMs of 0.01 and 0.04 mg/m3, respectively. The tungsten concentration levels varied between less than 0.0006 and 5.7 mg/m3, with median AMs of 0.042 and 0.15 mg/m3, respectively. Nickel exposure levels ranged from less than 0.0005 to 2.8 mg/m3, with corresponding median and mean of 0.007 and 0.036 mg /m3, respectively. Of the cobalt concentrations, 37% exceeded 0.02 mg/m3, and 21% exceeded 0.04 mg/m3. Some 65% of these high cobalt levels cobalt were for workers involved in powder production, and 22% for those working in to pressing and shaping. On the contrary, only 6% of nickel measurements exceeded 0.1 mg/m3 (80% for workers in powder production) and only 1% exceeded 0.5 mg/m3. For chromium, 23% exceeded 0.1 mg/m3, mainly for workers in powder production. The total dust levels varied between less than 0.010 and 58 mg/m3, with AM and median of 0.89 and 0.37 mg/m3, respectively. For the area measurements, the majority of the samples represented cobalt, tungsten, and nickel, with 354, 142, and 58 exposure measurements for cobalt, nickel, and tungsten, respectively. The area cobalt concentration levels varied between less than0.000004 and 1.7 mg/m3, with AMs and median of 0.04 and 0.007 mg/m3, respectively. The tungsten concentration levels varied between 0.00002 and 0.4 mg/m3, with median and AMs of 0.004 and 0.03 mg/m3, respectively. Nickel area levels ranged from 0.00009 to 0.26 mg/m3, with corresponding median and mean of 0.0007 and 0.007 mg/m3. The total dust levels ranged from less than 0.03 to 100 mg/m3, with median and AM 0.13 and 1.4 mg/m3, respectively. For cobalt, 12% of the measurements exceeded 0.04 mg/m3. Current and historical Swedish OELs are summarized in Table 2, with the personal cobalt exposure levels, organized by time period and aggregated job titles summarized in Table 3. The total average cobalt concentration (AM 0.042 mg/m3) for 1970 to 2010 onwards ranged from 0.060 to 0.0089 mg/m3, with the corresponding levels for powder being AM 0.062 mg/m3, ranging from 0.081 to 0.014 mg/m3, and for workers in pressing AM 0.029 mg/m3, ranging from 0.066 to 0.0069 mg/m3. Workers involved in grinding showed lower cobalt levels, with an AM of 0.025 mg/m3, and ranging from 0.0037 to 0.040 mg/m3.
Exposure Modeling and Exposure Measures
Log linear modeling was carried out for all aggregated jobs and blue-collar workers based on two different assumptions (Tables 4 and 5). The modeling based on all aggregated jobs showed statistically significant regression coefficients (P < 0.001) between all sites, with the regression coefficients for site C was 2.5 and 2.0 times higher for site B than the reference site A. This implies that the exposure levels between sites were different, being 2.5 times higher for site C than for site A, and B 2.0 times higher, all other factors in the model being equal. No single site could be used to estimate exposure levels in the modeling. In principle, measurement data from a single plant were not suitable for representing the other sites. Looking at trends in exposure levels, it can be seen that the level drops over time, despite there being no significant change over the 1980s and 1990s when compared with the 1970s. From 2000 and onwards, there has been a statistically significant (P < 0.001) five-fold reduction in exposure levels. For aggregated job classes, the highest regression coefficients (B-values) were determined for powder production, pressing, and wearing parts, in particular rolls (14.96; 5.66 and 13.39), implying a 15-fold statistically significant (P < 0.001) difference between the job classes B (intermittent low) and D (powder production) exposure levels, all other determinants being equal. When we carried out the modeling of data for blue-collar workers, using job class I (grinding) as reference, given the rather small number of measurements for the nonproduction job titles, almost identical figures and trends were noted for the regression coefficients (B-values) representing site (company) and year by decade. When comparing job classes, the highest B-values were seen for workers in powder production, pressing, and wearing parts, in particular rolls (5.4, 2.1, and 4.78; P < 0.001) using grinding values as the reference (Table 5).
No measurement data were available for the time period 1950 to 1969, during which a large number of the studied workers were exposed. In a standard exposure assessment procedure, exposure before 1970 would be modeled the same as the period 1970 to 1979 (status quo). However, in this study, we also extrapolated regression coefficients (B) values for our time component based on average change in each 10-year period, and used them to model cobalt exposure for the two periods 1950 to 1959 and 1960 to 1969 (linear extrapolation). The B-values calculated for periods between 1950 to 1959 and 1960 to 1969, based on our regression model, were 1.56 and 1.28, respectively, using the period 1970 to 1979 as the reference (B = 1). We also performed modeling after adapting data to an exponential extrapolation, resulting in B = 1.9 for 1940 to 1949, B = 1.8 for 1950 to 1959, and B = 1.6 for 1960 to 1969. Our linear extrapolation corresponds to an annual change of 2% to 3%, and our annual change for our exponential extrapolation ranges from 6% to 9% per year for the period before 1970.
The cumulative exposures based on modeling of aggregated job titles, assuming that the exposures before 1970 to 1979 could be used as proxies for the exposures in the period 1950 to 1969, showed an average of 0.041 mg/m3 years, ranging up to 2.5 mg/m3 years, with powder and rolls production yielding the highest cumulative exposures, being 0.10 and 0.23 mg/m3 years. When using extrapolated B-values for the period 1950 to 1969, a rather small increase from 0.041 to 0.044 mg/m3 years was seen. A corresponding increase was noted for the different aggregated job titles. When the cohort was restricted to more than 1 year of exposure, increased cumulative exposures were noted, 0.068 and 0.074 mg/m3, respectively, for the average exposure when the different extrapolation techniques were used (Table 6). Notably, almost 50% of the total cohort consisted of short-term employed workers.
The cumulative exposures based on modeling of blue-collar job titles, assuming exposures for 1970 to 1979, could be used as proxies for the period 1950 to 1969 showing an average of 0.053 mg/m3 years, ranging up to 2.4 mg/m3 years, with powder and rolls production showing the highest cumulative exposures, being 0.10 and 0.20 mg/m3 years. When using extrapolated B-values for the period 1950 to 1969, a rather small increase from 0.053 to 0.056 was seen, still with powder and rolls production yielding the highest cumulative exposures. When restricted to more than 1 year of exposure, the average cumulative exposures increased to 0.079 and 0.084 mg/m3 years, respectively, when using the different extrapolation techniques (Table 7).
The mean exposure based on modeling of all aggregated job titles, assuming exposures for 1970 to 1979, could be used as proxies for exposures between 1950 and 1969 showing an average of 0.0087 mg/m3, ranging up to 0.06 mg/m3, with powder and rolls production yielding the highest mean exposures, being 0.047 and 0.038 mg/m3, respectively. When using extrapolated B-values for the period 1950 to 1969, a rather small increase from 0.0087 to 0.010 mg/m3 was seen. A corresponding increase was noted for the different aggregated job titles. When the cohort was restricted to more than 1 year of exposure, reduced mean exposures were seen, 0.0065 and 0.0072 mg/m3, respectively (Table 8).
The mean exposures based on modeling of blue-collar job titles, assuming the exposures between 1970 and 1979, could be used as proxies for exposures in the period 1950 and 1969 showing an average of 0.012 mg/m3, ranging up to 0.06 mg/m3, with powder and rolls production yielding the highest mean exposures, being 0.040 and 0.029 mg/m3. When using extrapolated B-values for the period 1950 to 1969, a rather small increase to 0.013 was seen, still with powder and rolls production yielding the highest mean exposures. When restricted to more than 1 year of exposure, the average mean exposures decreased to 0.0078 and 0.0084 mg/m3, respectively, when using the different extrapolation techniques (Table 9).
When we modeled exponentially decreasing exposures for blue-collar workers with more than 1 year of exposure, the cumulative and mean exposures increased (Table 10). For powder production, the average cumulative exposure increased from 0.27 to 0.4 mg/m3 years, for pressing from 0.1 to 0.15 mg/m3 years, and for the overall from 0.079 to 0.10 mg/m3 years. The corresponding figures for means for powder production were 0.012 to 0.043 mg/m3 for powder, and 0.0078 to 0.010 mg/m3 for the overall.
For cumulative exposures for short-term exposed blue-collar workers (<1 year), the total average level was 0.004 mg/m3 years, for powder production 0.0137 mg/m3 years, ranging up to 0.015 mg/m3 years, and for grinding 0.0013, ranging up to 0.0015 mg/m3 years. The corresponding mean exposures were generally higher, the total average mean level being 0.014 mg/m3 years, for powder 0.059 mg/m3 years, ranging up to 0.09 mg/m3 years, and for grinding 0.0028, ranging up to 0.01 mg/m3 years. Data are not shown in table.
Exposure assessment was performed for cobalt, nickel, and tungsten for workers at three Swedish hardmetal plants operating from 1930s and onwards. The exposure assessment was used in a Swedish mortality study in the hardmetal industry.18 A historical database (1970 to 2012) of personal and area measurements in the Swedish hardmetal industry was created. Some 24% of the personal cobalt exposures exceeded 0.02 mg/m3, and 13% exceeded 0.04 mg/m3. Most of the high cobalt levels were associated with powder production and pressing and shaping. Some 6% of the nickel air concentrations exceeded 0.1 mg/m3, with 1% exceeding 0.5 mg/m3. Only 10% of the tungsten measurement exceeded 1 mg/m3. The log linear regression modeling using extrapolation versus standard assumptions for early time periods showed statistical differences between sites, for time periods, in particular when before and after 2000 were compared with 1970 to 1979. The job classes powder production, pressing, and rolls production showed a significant impact on the model compared with grinding as the reference. Only minor differences were noted when mean and cumulative exposures was calculated based on different log linear modeling for the period before 1970. However, when log exponential modeling was used, a 50% increase was determined for cumulative exposures and means for certain jobs and a 25% increase overall. Some 2% of the cobalt cumulative exposures for blue-collar workers exceeded 0.4 mg/m3 years, 4% 0.04 mg/m3 years.
A historical database (1970 to 2012) of personal and area measurements in the Swedish hardmetal industry was created. The personal cobalt measurements are well distributed among the jobs with significant exposure and jobs occurring frequently in the personnel register, such as powder production, pressing, and grinding among the blue-collar workers and over the period covered by the database, that is, from 1970 to after 2010. Some 37% of the personal cobalt concentration exceeded the present Swedish OEL, and 21% exceeded 0.04 mg/m3, mostly for workers in powder production and pressing. Nickel and tungsten concentrations were low. Most of the high cobalt levels were associated with powder production, pressing, and shaping. No stationary measurements have been included in our regression modeling, because they were considered being sampled either too close or too far from the source to represent personal exposure for even the more static of jobs. The sampling time for the majority of the samples reflected the 8-hour work shift well. Notably, all overall means for cobalt concentrations in the period 1970 to 1999 exceeded the historical Swedish OEL 0.05 mg/m3, in contrast to the period after 2000 when the means were almost lower than half the present Swedish TLV, 0.02 mg/m3. Although not analyzed for trend, a substantial change in exposure levels (as reflected in the regression analysis) was noted in 2000 and onwards, in fact at a time when dust elimination programs were implemented at the different sites, in particular site A. It should be noted that no measurement data were available before 1990 at site A. Different time eras could be seen, with noncompliance and compliance periods apparent, in spite of a 50% reduction of the OEL for cobalt. Very high GSDs were noted, reflecting in particular the contrast between the later analyses of cobalt with very low detection limits in contrast to rather high cobalt levels determined historically. However, even though data are presented as less than 1 μg/m3 in the tables, true values in the measurement data were used for the calculations of various parameters.
Sampling and Analysis
Up to 2005, the sampling was carried out using 37 or 25 mm open-faced filter cassettes collecting total dust, after that the Swedish OEL for cobalt has been defined by sampling of the inhalable fraction. For the inhalable dust measurements in the database, all except 4% were sampled in parallel with total dust. We have used the same cobalt concentration levels determined in the inhalable fraction for the total dust cobalt levels used in our measurement database. The rationale for that transformation comes from evaluation of parallel stationary sampling data at one of the sites in the study, which showed almost unity when the cobalt concentrations were compared.24 The majority of the samples had sampling time above 360 and 240 minutes, implying our measurements well covered a major part of the 8-hour shift sampling. Although gravimetric analysis was used over the whole study period for total dust concentrations, the analytical methods for determining the metals cobalt, nickel, chromium, and tungsten changed and subsequently, the levels of detection. Until the 1990s, atomic absorption spectrophotometry (AAS) was used, and then there was a gradual change from low to high-resolution ICP-MS, producing a corresponding increase in detection limits from 1 μg per sample to 0.05 and 0.001 μg/sample. For reasons of comparability, we have set the detection limit for all metal air concentration analysis to 0.001 mg/m3, and subsequently as summarized in Tables 1 and 3 and in Appendices1, 2, and 3. When calculating parameters such as means and SDs, the true value or less than (<) values have been used. Accredited laboratories with a vast experience in metal analysis have carried out the analysis of metals from the 1970s; in fact, two well-known Swedish laboratories, ALS Scandinavia, and our own laboratory at the Department of Occupational and Environmental Medicine, are responsible for the bulk of analysis.
Exposure Assessment and Exposure Measures
We have used standard exposure measures to evaluate our cohort (ever/never exposed, duration of exposure including latency, and cumulative and mean cobalt exposure) for the exposure–response analysis.31 This paper describes the assessment for quantitative exposure measures for cobalt in particular. It is important to note that, when the cumulative dose is strongly related to the duration of exposure and long-term effects, other mechanisms related to high exposure at shorter time periods could be valid. Therefore, it is also useful to include mean as an exposure measure.32 Our aggregated job classes reflect the job titles occurring in the company data files and registers accurately, as do data in our historical measurement database. This could be further justified by looking at the distribution of aggregated jobs, especially blue collar, and the occurrence of job titles and measurement data over time (Appendix 2). The necessity of this aggregation based on SEGs is obvious for the Swedish cohort study; the background and nonproduction departments only contributed to 69 cobalt measurements, with the corresponding job titles occurring 11,910 times in the personnel files. In contrast, for powder production, there were 414 measurements in 713 job classes. Of particular interest is the low resolution of job titles in the Swedish measurement database in relation to powder production, with 206 cobalt measurements in 632 job classes, all in the unspecified category “hardmetal production.” Tungsten carbide production is not included in the epidemiological study. A similar pattern was noted for pressing, forming, shaping, and grinding (Appendix 2).
Our quantitative cobalt exposure measures are based on log linear modeling, including company, job class, and period by decade, from 1970 to 2012.33 However, as a large group of our cohort was included in the national mortality registers starting in the early 1950s, we have used our three different modeling approaches to estimate cobalt exposure for the periods 1940 to 1949, 1950 to 1959, and 1960 to 1969, rather than using standard approaches and applying data from 1970 to 1979 for these earlier years. Our log linear model for extrapolation shows decreasing exposure concentration levels at a rate of 2% to 3% per year, in line with trends reported from several other industrial cohorts.34,35 Using log exponential extrapolation showed yearly decreased exposure levels ranging from 6% to 9%, data that could also be found in the literature.36 However, we have no measurement data at all for the early time periods. In our mortality paper, standardized mortality ratios using the different extrapolation techniques are presented for all three extrapolation models.18
To consider both the various exposure measures and to avoid the strong influence of short-term employed workers in our cohort,18 we defined two different exposure groupings. Exposure to cobalt was defined as cumulative exposure with units mg/m3•year (ie, exposure level multiplied by exposure time) and mean (cumulative exposure by duration of exposure), then and categorized in two different ways. The determined cumulative and mean exposures were categorized as quartiles, for both the total cohort and for blue-collar workers. Exposure was also categorized into three exposure groups that reflected exposures relevant to a 40-year working career with current Swedish OEL for cobalt, that is, 0.02 mg/m3, defined as the 8-hour time weighted average air concentration (8-hour TWA). The cumulative dose values for the three categories were 0.20 or less (low), 0.21 to 0.40 (medium), and more than 0.41 (high) mg/m3 •year. The high exposure group reflects half of the maximum allowed lifetime exposure to cobalt, 0.8 mg/m3• year, which corresponds to 40 years of exposure at the present Swedish OEL for cobalt, 0.02 mg/m3 (SWEA 2015). The corresponding classes based on means would be 0.005 or less, 0.0051 to 0.010, and more than 0.01 mg/m3.
Nickel, Chromium, and Tungsten
Metallic nickel and chromium could be present in certain cemented carbide qualities during the production of hardmetal products. Nickel was used at all plants, although measurement data were limited to company C. At company A, nickel was introduced in the early 1990s and used for some 20 years, accounting for 2% to 2.5% of total production and a content ranging from 4% to 6%. At company B, nickel has been used since mid-1980s, accounting for only 0.25% of the total hardmetal production, and at company C, it was introduced in the 1960s in about 1% of the production. Chromium carbide (Cr3C2) was introduced in the early 1990s as a grain growth inhibitor and currently over 40% of powder production contains the substance, at a content of 0.3% to 0.9%. Our historical measurement database showed that only 6% exceeded 0.1 mg/m3 and just 1% exceeded 0.5 mg/m3. Given the low levels of nickel and chromium air measurements, reflecting minor addition to the different grades and poor distribution of these between time periods and sites, quantitative exposure assessment by modeling was not performed. The Swedish OEL29 for nickel in the metal state is currently 0.5 mg/m3, and is not classified carcinogenic. Within the EU, insoluble and soluble compounds are considered carcinogenic and have low recommended OELs; however, metallic nickel is excluded, as neither animal experiment nor human epidemiological studies have indicated any carcinogenicity.37 IARC classifies metallic nickel as possibly carcinogenic to humans (group 1), based on “sufficient evidence in humans for carcinogenicity of mixtures that include nickel and nickel compounds.”38 Tungsten exposure measurement data were highly correlated to cobalt, but no agencies consider the metal to be potentially carcinogenic.
Exposure Assessment in Other Studies
Exposure assessment was carried out performed in an original mortality cohort of Swedish hardmetal workers (3163 men), evaluated for the period 1940 to 1982.8 This study identified five exposure categories, unexposed (0), temporary presence in an area of hardmetal production (1), permanent presence in areas where cobalt products were machined (2), production of hardmetal products (3), and hardmetal powder production (4). A quantitative assessment was also carried out for the different categories by decade from 1940 to 1980. For the whole period, exposure in category 1 ranged from 1 to 2 μg/m3, in category 2 from 2 to 5 μg/m3, in category 3 from 10 to 30 μg/m3, and in category 4 from 10 to 1100 μg/m3 for the whole period, decreasing by each decade. The greatest changes by time were noted for the high exposure group, powder production, with substantial improvement when the 1940s was compared with the 1950s. Our extrapolated data from the period 1950 to 1959 were used to assess exposure levels for the 1940s. However, only a small number of workers were employed at that time period.
A French cohort of 709 hardmetal workers, employed between 1956 and 1989, was analyzed for mortality and exposure–response patterns.9 Exposure categories 1 to 4 were using the range of cobalt air concentrations, where 1 was considered nonexposed, 2 less than 10 μg/m3, 3 15 to 40 μg/m3, and 4 more than 50 μg/m3. No quantitative cumulative or mean exposure measures were used in the analysis.
In a French multicenter study, a cohort was formed using data from 10 different plants, with an inclusion criteria of greater than 3 months exposure and a mortality follow-up from 1968 to 1991.11 The cohort included 7459 men and women. The high-dose group in the analysis consisted of more than 164 or more than 299 level months; according to the given job exposure levels in the study, this would, for example, correspond to 40 months (3.3 years) at level 4, 0.06 mg/m3, that is, equivalent to 0.20 mg/ m3 •years.
Another cohort study from a French plant producing hardmetal and other cobalt products analyzed data from 2860 workers.12 The company started in1940, and the follow-up period was 1968 to 1992. The exposure assessment was based on a 1 to 9 exposure-level scale, including frequency of exposure, but no quantitative cumulative or mean exposure measures were used in the analysis. Our data, used in the Swedish cohort analysis, used cumulative dose values for the three categories were less than 0.20 (low), 0.21 to 0.40 (medium), and more than 0.41 (high) mg/m3 •year.
Exposure assessments for cobalt, nickel, and tungsten were carried out on a historical cohort of Swedish hardmetal industry workers. The assessments were based on a historical measurement database, using data from 1970 to 2012. Some 37% of the personal cobalt measurements exceeded 0.02 mg /m3, mostly for workers involved in powder production, pressing, and shaping. The nickel and tungsten levels were low. The log linear regression for the time period 1970 to 2012 showed statistically significant differences between sites, time periods, and jobs. Applying different extrapolation techniques for the time period before 1970 revealed differences between linear and exponential regressions. For the log linear extrapolation model used, 1.6% of the cobalt cumulative exposures for blue-collar workers exceeded 0.4 mg/m3 years determined as cumulative exposures. Both cumulative exposures and means were used in our exposure–response analysis in the cohort, including a sensitivity analysis of the outcome for our cohort when different exposure modeling was used.
We would like to acknowledge the cooperation and assistance of the representatives from companies and their Swedish sites. In particular, we would like to thank all the employees who provided assistance throughout the study.