Exposure to Thoracic Aerosol in a Prospective Lung Function Study of Cement Production Workers

Abstract

Introduction:

An exposure study was conducted as part of a multi-national longitudinal study of lung function in cement production workers.

Aim:

To examine exposure to thoracic aerosol among cement production workers during a 4-year follow-up period.

Methods:

Personal shift measurements of thoracic aerosol were conducted among the cement production workers within seven job types, 22 plants, and eight European countries (including Turkey) in 2007, 2009, and 2011. The thoracic sub-fraction was chosen as the most relevant aerosol fraction related to obstructive dynamic lung function changes. Production factors, job type, and respirator use were recorded by questionnaire. The exposure data were log-transformed before mixed models analysis and results were presented by geometric mean (GM _adj ) exposure levels adjusted for plant or job type, worker, and season as random effects.

Results:

A total of 6111 thoracic aerosol samples were collected from 2534 workers. Repeated measurements were obtained from 1690 of these workers. The GM _adj thoracic aerosol levels varied between job types from 0.20 to 1.2mg m ⁻³ . The highest exposure levels were observed for production, cleaning, and maintenance workers (0.79–1.2mg m ⁻³ ) and could reach levels where the risk of lung function loss may be increased. The lowest levels were found for administrative personnel (0.20mg m ⁻³ ) serving tasks in the production areas. Office work was not monitored. GM _adj exposure levels between plants ranged from 0.19 to 2.0mg m ⁻³ . The time of year/season contributed significantly to the total variance, but not year of sampling. Production characteristics explained 63% of the variance explained by plant. Workers in plants with the highest number of employees (212–483 per plant) were exposed at a level more than twice as high as those in plants with fewer employees. Other production factors such as cement production, bag filling, and tidiness were significant, but explained less of the exposure variability. These determinants factors can be useful in qualitative exposure assessment and exposure prevention in the cement production industry. Respirator use was minor at exposure levels <0.5mg m ⁻³ but more common at higher levels.

Conclusion:

Production, cleaning, and maintenance work were the job types with highest exposure to thoracic aerosol in cement production plants. However, plant had an even larger effect on exposure levels than job type. The number of employees was the most important factor explaining differences between plants. Exposure reached levels where the risk of lung function loss may be increased. No significant differences in exposure between sampling campaigns were observed during the 4-year study period.

INTRODUCTION

Cement is a building material used in large quantities worldwide. In 2007 cement production in Europe reached 336 million tons. By 2011, production decreased to 274 million tons (CEMBUREA, 2013). Approximately 250 European Cement Association (CEMBUREAU) members employed almost 51000 workers in 2008. By 2011, the number of workers had decreased to ~45000 ( BCG, 2013 ).

Cement production workers are exposed to airborne particles from raw materials, mainly limestone and clay, clinker and other cement components, and fuels. A wide range of different exposure levels for total and respirable dust have been reported. Total dust exposure varied from 0.21 to 550mg m ⁻³ [geometric means (GM)], and 0.04–39mg m ⁻³ [arithmetic means (AM)]. The respirable dust exposure varied from 0.24 to 15mg m ⁻³ (GM) and 0.22–43mg m ⁻³ (AM). ( HSE, 2004 ; Mwaiselage et al. , 2005 ; Mirzaee et al. , 2008 ; Zeleke et al. , 2010 , 2011 ; Fell et al. , 2011 ; Kakooei et al. , 2012a , 2012b ). One study has reported median thoracic aerosol levels of 0.60–1.8mg m ⁻³ ( Fell et al. , 2010 ). The cross-sectional analysis at baseline of the current study reported a GM thoracic aerosol exposure of 0.85mg m ⁻³ ( Nordby et al. , 2011 ).

A decline in lung function and increased airway symptoms have been linked to occupational exposure in cement production plants ( HSE, 2004 ; Mirzaee et al. , 2008 ; Fell et al. , 2011 ; Zeleke et al. , 2011 ; Kakooei et al. , 2012a ). Most previous studies have been cross-sectional and therefore provided less convincing data on exposure–response relationships between cement dust and respiratory effects.

In order to further investigate the associations between aerosol exposure and respiratory effects among cement production workers, the European Cement Association (CEMBUREAU) initiated this multi-national, 4-year prospective study of obstructive lung function changes. A cohort including workers from 24 cement production plants situated in eight countries was established in 2007, with follow-up conducted in 2009 and 2011. We chose to measure the thoracic sub-fraction of the aerosol as this fraction contains particles that penetrate beyond the larynx and is considered to be more relevant regarding bronchial effects than the inhalable and respirable size fractions ( CEN, 1993 ; Vincent, 1995 ). This fraction is defined by a penetration curve of the total airborne particles with a 50% cut-off at an aerodynamic diameter of 10 µm and geometric standard deviation (GSD) of 1.5. The thoracic aerosol sub-fraction, hereafter called thoracic aerosol, is comparable to the PM ₁₀ fraction commonly measured in environmental studies. It should be noted that the definitions are not exactly the same ( Vincent, 1995 ).

Cement production

The main raw materials for cement production are limestone and clay. Other minerals containing aluminium, iron, and silicon are added depending on the desired quality. Limestone is usually quarried and crushed in the vicinity of the cement plant, and is transported to the plant production area. Clay is often obtained from local quarries. Other raw materials arrive by ship, truck, or train. All materials are ground and milled together to form a dry powder termed ‘raw meal’ or slurry in case of the wet process. The raw meal is heated to 1450°C in the kiln where clinker is formed by an endothermic reaction. The clinker is cooled, stored, and milled together with calcium sulphate to form cement. Dependent on the desired cement quality different ingredients may be added ( Taylor, 2004 ; Joint Research Centre, 2013 ). The most common cement type is Portland cement, which contains clinker with a few percentage of calcium sulphate ( Taylor, 2004 ). The final cement products are packed in bags or delivered in bulk for transport by truck, train, or ship. Fig. 1 shows the outline of a typical cement production plant based on the dry process.

The objectives of the study were to examine the exposure levels to thoracic aerosol in the European cement production industry and to create a job exposure matrix for a lung function study of a large cohort of cement production workers. This paper presents the thoracic aerosol exposure of the cohort of cement production workers by job type and explores production-related exposure determinants.

MATERIALS AND METHODS

Selection of cement plants

Approximately 250 cement-producing plants [223 plants in 2003; A. Sciamarelli (personal communication); CEMBUREAU] that were members of CEMBUREAU in 2006 were invited to participate in the study. The criteria for including plants were different geographical regions, complete personnel records, access to occupational health personnel to perform exposure monitoring and spirometry measurements, and no previous production of asbestos cement. Twenty-five plants agreed to participate in the study, but for practical reasons, one plant was excluded. The remaining plants in the study were situated in Turkey (9), Italy (5), Sweden (3), Spain (2), Norway (2), Estonia (1), Greece (1), and Switzerland (1).

Job types

Job types were defined in cooperation with national coordinators who represented the plants in their country, and selected personnel from some of the plants. Jobs were classified as production, cleaning, maintenance, laboratory workers, foremen, and administration. Their main tasks are listed in Table 1 . Production workers handled raw materials and clinker, performed process control, and packed final products. Job rotation, which was common for production workers, may lead to mixed exposure from many sources. These workers were also responsible for process supervision which was carried out in a control room and thus mainly unexposed. In some plants a sub-group of the production workers was responsible for all process control. Cleaning workers were engaged in tasks such as removing dust generated during normal conditions and from spills. Maintenance workers performed electrical or mechanical repair and other maintenance work either in a workshop or on the machinery throughout the plant. Laboratory personnel collected and analysed samples from the production line. They worked in the production areas and in the laboratory. Foremen supervised production routines contributing to both administrative and practical work in the production area. Administrative personnel were primarily situated in an office area, but some performed work in the production area for shorter periods of time.

Sampling strategy

All workers in the production departments were eligible for exposure measurements, except for administrative personnel not entering production areas, and workers employed in quarries or in out-sourced services. Quarry workers were excluded because their exposure to clinker and cement was minor, and they were likely exposed to other agents with known lung effects, such as crystalline silica. Routine measurements of silica among cement production workers indicated low exposure levels [S. Gardi (personal communication); Italcementi Group, Bergamo, Italy]. Workers employed by contractors were excluded because of anticipated difficulties with follow-up. Cleaning and maintenance work during production shut-down were especially prone to out-sourcing.

Sampling was mainly conducted when the plant was operating at normal production capacity. More seldom tasks such as maintenance work during production shut-down were included in the sampling campaigns. Workers classified themselves on the day of measurement into pre-defined job types: production, cleaning, maintenance, foreman, administration, laboratory, and other, see Table 1 . Measurements were only performed on a subset of the administrative workers who entered the production area. The category ‘several job types’ was added to identify workers who performed more than one of the specified job types during the day of measurement.

Workers were randomly selected for exposure measurements within job types in each plant. The number of workers included during first (2007) and second phase (2009) was based on the number of workers within each job type at each plant according to Appendix,  Table A3 in Leidel (1977) . This number (2–11 workers) is based on the requirement that at least one worker will be measured among the 20% highest exposed workers in the group with 90% confidence. If the number of workers in a job type was five or less, all workers were selected. From job groups of six or more workers, 5– 11 workers were randomly selected. In addition, five repeated measurements were obtained of each job type (sampling of up to five workers in the group was conducted for at least one supplementary day). A number of five repeated measurements per job group was considered sufficient to estimate the within-worker variability and was chosen arbitrarily. In the third phase (2011), 70 personal whole-shift samples were collected from each plant regardless of the total number of employees.

Due to feasibility reasons, sampling campaigns were limited to <1 year in most plants, and categorized by season.

Organization of the sampling campaign

National coordinators were responsible for the organization of the sampling campaign in their respective countries. Filters cassettes were packed in polyurethane foam and transported back and forth together with sampling equipment by the project team from the National Institute of Occupational Health (NIOH), national coordinators, or by mail. Sampling was performed by technical personnel employed at the plants or personnel from the occupational health service, except in two countries where a national team collected all samples.

Questionnaires

An exposure questionnaire was developed in English by the NIOH project group in cooperation with the national coordinators and technical personnel from some of the plants, and translated into nine languages. The questionnaire included questions about job type and respirator use during the work shift. Different types of respirators were provided by the companies. This questionnaire was completed by technical personnel after collection of each sample. In a second questionnaire in English, production details and number of workers employed in main job types were recorded at each plant by the administration in 2007 and 2011.

Instruction, training, and site visits

All sampling teams received an illustrated manual of the technical procedures and an instruction video demonstrating all sampling details. The manual and the instruction video were developed in English and translated into the appropriate languages. Technical and occupational hygiene personnel at each plant were instructed before each sampling period by the same two researchers from NIOH during site visits.

The two principal investigators of the research team classified plants as clean or less clean. This subjective judgement of tidiness was based on observation of dust deposits, storage spills, age and maintenance of machinery, and waste and disorder in production areas. In the case of differential individual judgements consensus was reached after a discussion.

Collection of thoracic aerosol samples

Personal full-shift samples were collected using thoracic cyclones (GK 2.69; BGI Instruments, Waltham, MA, USA) operated at a flow rate of 1.6 l min ⁻¹ . PVC membrane filters were used with pore size 5 μm (Millipore, Cork, Ireland; SKC Inc., Eighty Four, PA, USA; Pall Corp., Ann Arbor, MI, USA) on cellulose support pads (Millipore Corp., Billerica, MA, USA) in 37-mm diameter aerosol cassettes (Millipore Corp.). The cyclones were mounted within the breathing zone and outside respirators. Pulsation-free battery-powered pumps owned by the sampling teams or provided by NIOH in Norway (PCXR; SKC Inc., Eighty Four, PA, USA or PS 101; NIOH) were used.

Gravimetric determination of aerosol mass

All mass determinations were performed at NIOH using a Sartorius MC 5 micro balance (Sartorius AG, Göttingen, Germany). Filters, including a field blank for every 10th filter, were conditioned for 5 days in a climate-controlled room (temperature 20±1°C and relative humidity 40±2%). E2 certified reference masses and reference filters were weighed for quality control. Static charging was minimized by microbalance ionizers (Staticmaster®; NRD, LLC, Grand Island, NY, USA). Most exposed filters were weighed by the same person. Aerosol masses were corrected by field blanks weighed on the same day. The limit of detection (LOD), defined as 3 × SD of field blanks, ranged from 0.0063 to 0.073mg.

Data analysis

Collected samples with obvious errors including broken and double-exposed filters, incorrect weight determination, and flow rate outside the range of ±20% were excluded. Samples with negative masses <−3 SD of the field blanks were also excluded because of very low probability of obtaining such values by random analytical and sampling errors. These results are more likely to be due to unidentified errors such as minor breakage of the filter. Positive masses below the detection limit were used as observed ( Porter and Ward, 1991 ; Cressie, 1994 ). Air concentrations were calculated using thoracic aerosol mass, a fixed flow rate of 1.6 l min ⁻¹ ( NIOSH, 1998 ) and sampling time. Measurements above 150mg m ⁻³ were excluded as light microscopic inspection of the filters revealed larger particles than expected from the thoracic convention (>50 μm) which probably reflects that the cyclone was not operated as intended. Plots of the exposure data indicated that the distribution was skewed to the right, and the data were log ₁₀ -transformed before statistical analysis. Measurements with standardized residuals outside ±3.29 calculated from a linear regression model including indicator variables for job type, plant, and year as predictors, were considered as outliers and were excluded in order to obtain more robust exposure estimates.

Exposure data were described by the measures of central tendency and dispersion: AM, GM, GSD, and median.

The exposure data have a multilevel dependency structure. There are repeated measurements for workers who are furthermore nested within job types and plants, while the observations of workers are distributed over years and seasons. The contribution of these components to the total variance of the exposure data was estimated by means of a random effects model.

The random effects model can be represented formally as:

Where,

$ypjwysk$ = exposure for plant p , job type j , worker w , year y , season s , and replication k

μ = the intercept.

$up(1),...,upjwys(5)$ , = independent random effects with mean 0 and variances.

$σ(1)2,...,σ(5)2$ .… representing variation within plants, job types, workers, years, and seasons, respectively.

$εpjwysk$ = independent, and identically and normally distributed random residuals variables with mean 0 and variance $σε2$ .

The year component did not contribute to the total variance and was not included in the linear mixed models described below.

In order to analyse the effect of job type, plant, and selected plant determinants on exposure levels and adjusted exposure means were estimated using linear mixed models. Adjusted exposure means for job types were estimated by a model with job type as fixed effect and season, worker, and plant as nested random effects. Adjusted exposure means for plants were estimated using a model with plant as fixed effect and season, worker, and job type as nested random effects. The effects of plant-specific determinants were analysed by including these as fixed effects, while season, worker, and job type were included as nested random effects. Plant was not included in this model in order to estimate the extent that plant-specific determinants explained the observed variance. Models with different selections of plant-specific covariates were constructed by a forward selection procedure. The significance of added fixed effect terms were evaluated by likelihood ratio tests using maximum likelihood estimation. All models were finally estimated by restricted maximum likelihood.

The mixed models with fixed effects can be represented mathematically as described below for Model 3 n Table 5 :

Where, for plant p:

N_p = number employed;

MaxCem _p = maximum cement production;

MaxCem/Empl _p = maximum cement production per employee

Tidi _p = tidiness

BagFill _p = bag filling

$uj(1),ujw(2) and ujws(3)$ = random effects representing variation within job types, workers and seasons, respectively.

$εpjwsk$ = independent, and identically and normally distributed random residuals variables with mean 0 and variance $σε2$ .

GM exposure levels by job type and plant adjusted for other variables (GM _adj ) were obtained by back-transformation by antilog ₁₀ of the respective regression coefficient with the intercept added.

The model fit was evaluated by inspecting standardized residual plots.

Statistical significance was set at a P value of 0.05.

Access 2007 (Microsoft Corporation, Redmond, WA, USA) and STATA 13 (StataCorp LP, College Station, TX, USA) were used for data management and statistical analysis, respectively.

RESULTS

Data selection

A total of 7085 measurements of thoracic aerosol exposure with completed questionnaires from 24 plants were obtained. Approximately 14% of the measurements including all measurements from two plants were excluded before statistical analysis because of errors as shown in Appendix,  Table A1 . The measurements from two plants were excluded because it was uncertain whether correct procedures had been followed. It was likely that communication problems lead to misunderstandings regarding the operation of the cyclones.

The final data set of 6111 measurements contained no negative values. Samples below the level of detection (LOD) amounted to 2.4% in total and varied between 0 and 9.7% across plants and between 0.8 and 8.7% in all job types except office workers (24%).

Exposure by plant and job type

Exposure data were obtained from 2534 workers in 22 plants (89–483 workers per plant) within 8 job types (103–1184 workers per job type). Repeated measurements (maximum 14) were obtained from 1690 workers with a mean of 2.4 measurements per worker. Sampling time varied from 135 to 749min (median 432min). The proportion of samples with shorter sampling time than 360min was 7.5%. However, the mean exposure of these samples (AM = 2.3mg m ⁻³ ) was similar to the mean exposure of all 6111 samples (AM = 2.2mg m ⁻³ ) and no major bias was expected from including these samples.

The measured exposure levels of thoracic aerosol by plant and year are shown in Appendix,  Table A2 . The AM exposure in the plants for all sampling years combined ranged from 0.50 to 7.5mg m ⁻³ and GM exposures ranged from 0.22 to 2.6mg m ⁻³ . The AM exposure levels of all plants were similar between sampling years, ranging from 2.1mg m ⁻³ in 2007 to 2.3mg m ⁻³ in 2011. However, in 44% of the plants the AM exposure was higher in 2009 compared to 2007. In 2011, AM exposure was increased in 50% of the plants compared to 2007. The exposure levels by job type for all plants combined are shown in Fig. 2 . The highest AM exposure levels were found among production (2.9mg m ⁻³ ) and cleaning workers (2.5mg m ⁻³ ). The lowest was among administrative personnel (0.21mg m ⁻³ ). The group of administrative personnel included in the measurements was a subset of the office staff because measurements were only performed on workers who entered the production area. These measurements are therefore not representative for all administrative personnel at the cement plants. The thoracic aerosol exposure levels by job type within plants are shown in Appendix,  Table A3 .

Plant-specific determinants

Because measurements aggregated at the plant level do not explain how plant characteristics influence exposure levels, information on production variables was explored in order to explain the effect of plant in statistical models.

The maximum of cement production from 2007 to 2011 was used as an estimate of maximum production capacity. For four plants, this information was not available from the questionnaire but was obtained from websites of the respective plants. The number of employees, highest in 2007, was used to estimate the total workforce. The ratio of the maximum cement production to the number of employees in 2007 was computed and served as a proxy of automation level. Data were available for all plants and were categorized by tertiles as we did not expect a priori that these variables had a linear relationship to the log ₁₀ -transformed exposure. The number of employees per plant ranged from 69 to 483, the maximum cement production from 0.43 × 106 to 4.0 × 10 ⁶ ton year ⁻¹ and from 2.7 × 106 to 12 × 10 ³ ton year ⁻¹ per employee.

Information on the proportion of the cement production that was packed and shipped in bags (bag filling) was obtained from 13 plants and represented by 3624 measurements. This proportion did not change substantially between the sampling campaigns and the highest recorded proportion was used in the models (range 0–40%). Cement production data for 2007, 2009, and 2011 were available from 18 plants, including 4949 measurements. Cement production differed between the sampling campaigns. Annual cement production was computed relative to the maximum cement production ranging from 35 to 100%.

In addition, 58% of the measurements had been carried out in 13 plants classified as clean by subjective judgement of tidiness.

All but one plant (Plant 1) in this study used the dry process technology. The plant using the wet process had a higher mean exposure than the mean of all plants, (Appendix,  Table A2 ). This factor was not further explored.

Statistical modelling of exposure to thoracic aerosol

The main predictors of thoracic aerosol exposure: plant, job type, worker, year, and season were explored in a random intercept model in order to study the distribution of the variance between these variables, Table 2 . These predictors explained 21, 16, 19, 0, and 7% of the total variance of this model, respectively. The variance component of year did not contribute to the total variance, and was omitted from further models.

The exposure of workers with different job types was estimated by a mixed model with job type as fixed effect, and plant, worker, and season as random effects, Table 3 . Job type was highly significant ( P < 0.001). The GM exposure levels adjusted for all random effects varied from 0.20 to 1.2mg m ⁻³ between job types and differed from the unadjusted data by −15% to +100%. The largest differences were observed for the groups with fewest measurements; foremen (+55%, N = 148) and administration (+100%, N = 226).

The exposure levels in plants were estimated by a mixed model with plant as fixed effect, and job type, worker, and season as random effects (Appendix,  Table A4 ). Plant was highly significant ( P < 0.001). The GM exposure levels adjusted for all random effects varied from 0.19 to 2.0mg m ⁻³ between plants and differed from the unadjusted data by −34% to +4%. In this model plant explained 23% of the total variance in the ‘empty’ Model 0, see Table 4 .

The production variables with complete data: number of employees, maximum cement production and maximum cement production per employee, and tidiness were explored as fixed effects in Model 2, Table 4 . All variables were significant ( P = 0.02 to <0.001). Back-transformation of the regression coefficients by antilog ₁₀ showed that factories with higher number of employees ( N = 212–483) had 2.2–2.7 times higher exposure than the factories with fewer employees. Exposure levels in plants with the highest production volumes (≥1.9 × 10 ⁶ ton year ⁻¹ ) was 20% lower than in plants producing ≤1.1 × 10 ⁶ ton year ⁻¹ , while the middle category showed 18% higher exposure than the lowest category. Maximum cement production per employee showed a 14% higher exposure in the middle category (6.4–8.5 × 10 ³ ton year ⁻¹ and employee) while the exposure in the highest category was similar to the lowest category. Tidiness was also important, indicating a 52% higher exposure in plants classified as less clean.

The determinants in Model 2 explained 17% of the total variance in the model with job type, worker, and season as random effects but without fixed effects (Model 0). Inclusion of plant as a random effect (Model 1) reduced the sum of the variance components of job type, worker, and season by 27% compared to Model 0. The production determinants could therefore explain the greater part of the exposure differences across plants.

Due to incomplete data on bag filling and annual cement production, models with these determinants had to be based on measurements from fewer plants.

The effect of bag filling was significant ( P = 0.01). The highest categories of bag filling (6–40% of the cement production) predicted increased exposure to thoracic dust by 37–38% compared to the lowest category with only 0–4% filled in bags, Model 4 in Table 5 .

The annual cement production was significantly ( P = 0.05) predicting higher exposure to thoracic dust by 9% during lower production rate (86–98%) compared to production at full capacity. In the lowest production category (38–85%) the exposure level decreased compared to the full capacity, but only by 5%, Model 6 in Table 5 .

Inspection of residuals indicated good fit of the models.

Respirator use in relation to the thoracic aerosol level

All plants provided respirators for their workers. It should be noted that there was no information on whether mandatory use of respirators in the plants was instructed, or on type of respirators. Respirators were used most often by cleaning, production, and maintenance personnel, while other workers used respirators less often. When the categories of respirator use most of the time and during the whole shift are combined, respirator use ranged from 9% in the lowest exposed group (below 0.5mg m ⁻³ ) to 53% in the highest exposed group (above 4mg m ⁻³ ), Table 6 .

DISCUSSION

Cement production workers are exposed to a varying level of thoracic aerosol. The most important sources of exposure variability were plant, job type, and worker. These explained 56% of the total variance in a random effects model. Season also contributed significantly, but the effect of sampling year was minor. Adjusted GM exposure levels varied from 0.20 to 1.2mg m ⁻³ between job types, showing the highest exposure levels for production, cleaning, and maintenance personnel, 0.79 to 1.2mg m ⁻³ . Laboratory workers, foremen, and workers with other job types were lower exposed, 0.42–0.45mg m ⁻³ , and administrative personnel the lowest, 0.20mg m ⁻³ . The differences between plants were larger as the GM _adj exposure level of all workers combined ranged from 0.19 to 2.0mg m ⁻³ . Production characteristics explained 63% of the variance explained by plant. The most important characteristic was the number of workers per plant as workers in plants with the highest number of employees were more than two times higher exposed than in plants with fewer employees.

To our knowledge the size of this study is unique for cement production workers. Exposure to the thoracic aerosol fraction has not been assessed in the cement industry previously. The studies by Fell et al. (2010) and Nordby et al. (2011) have reported thoracic aerosol exposure levels, and their measurements are included in the present study. It is surprising that thoracic aerosol exposure has not been measured previously since the thoracic aerosol fraction is considered to be the more health-relevant particle size fraction for bronchial effects ( CEN, 1993 ).

This large sampling effort would not have been possible without the assistance of the personnel employed or contracted by the plants. Sampling by different teams and challenges represented by providing and obtaining information in nine different languages may have affected the quality of the data. Repeated training and follow-up visits were conducted by the research team in order to supervise the quality of the sampling campaigns. In spite of these efforts 14% of the samples (including all samples from two plants) were excluded from the statistical analyses, mainly due to technical errors. Sampling was mainly conducted when plants operated under normal conditions, but also included tasks performed during shorter periods of maintenance work during production shut-down. A detailed analysis of the influence of specific tasks is planned for a future publication.

Risk evaluation of the thoracic exposure levels in this study is limited as no studies of cement-exposed workers have assessed exposure to the thoracic fraction. Only the cross-sectional analysis of the present study provides some guidance ( Nordby et al. , 2011 ). Workers exposed to median thoracic aerosol levels of 0.49–1.1mg m ⁻³ showed decreased lung function (FEV ₁ ) compared to cement production workers exposed to levels <0.49mg m ⁻³ . The observed GM exposure of 88 of 157 job groups within the plants was at or above this level. Therefore a substantial proportion of the cement production workers seems at risk of accelerated loss of lung function. This must be considered as a preliminary evaluation as the longitudinal analysis of the respiratory effects is expected to provide more valid criteria for evaluation of health risks from thoracic aerosol exposure in cement production workers.

Differences between plants explained most of the exposure variability. Plant characteristics were therefore explored in order to study the factors behind this variation. Approximately 63% of the variability due to plants could be explained by production determinants that were available for all plants. The greatest effect stems from the number of employees as the exposure level of workers in plants with the greatest number of workers ( N = 212–483), was more than twice as high as in plants with fewer employees. The number of employees is probably a proxy of the level of automation with more manual work being done in plants with many employees. This effect is likely to be associated with production capacity as well. Cement production capacity may also indicate automation level as exposure levels in plants with the largest production, 1.9 × 106–4.0 × 10 ⁶ ton year ⁻¹ , were 20% lower than in plants producing 0.43 × 106–1.1 × 10 ⁶ ton year ⁻¹ . The ratio of these determinants, and production capacity per employee, showed a 14% lower exposure in the middle category compared to the lowest and the highest category. Tidiness was also important, as the exposure was 52% higher in less clean plants compared to those judged as clean. This indicates that re-suspension of dust deposits may contribute to increased exposure levels. Tidiness is especially interesting as it indicates a potential for exposure prevention by dust removal. The production determinants indirectly indicate the effect of automation and are valuable for qualitative assessments of the exposure.

Statistical modelling of annual cement production in 2007, 2009, and 2011 relative to maximum production capacity was limited to fewer plants, as data were not available from four plants. The annual production varied between sampling campaigns and was significantly associated with exposure levels. The highest exposure was found in the medium category of 86–98% of maximum production. Exposure was lower at 100% of maximum capacity and lowest at 38–86% capacity. Exposure differences were small as the largest difference between categories was 15%.

For evaluation of bag filling we only had data available from 13 plants. This determinant expectedly showed 38% higher exposure in plants where a significant proportion of the cement production (6–40%) was filled in bags compared to plants shipping mostly in bulk. Bagging departments thus seem to have a potential for exposure prevention.

Although job type explained 16% of the total variance, differences in exposure levels were not major as the GM _adj exposure levels varied from 0.42 to 1.2mg m ⁻³ except administration personnel that were exposed to 0.20mg m ⁻³ . The latter can probably be explained by administrative personnel staying only for shorter periods in exposed areas and not being directly involved in work operations. These exposure data are not representative of typical exposure levels for this job type as measurements on administrative personnel were only performed during working days when they were entering exposed areas and not on days spent in the office buildings. Workers performing production, maintenance, and cleaning operations were the highest exposed workers with GM _adj exposure levels of 0.79–1.2mg m ⁻³ . These workers spend most of the shift in the production areas performing process control, cleaning, and repairing. They work close to sources of exposure such as mills and bag-filling machines. Laboratory personnel, foremen, and workers with other job types were exposed to GM _adj levels of 0.42–0.45mg m ⁻³ . Laboratory workers stayed most of the time in the laboratory analysing samples which may lead to exposure for example from milling and sieving of samples. They also enter the production areas to collect samples. Cleaning jobs were completely out-sourced in 5 plants and partly out-sourced in 10 plants. As measurements only were performed on cleaning workers employed at the plants, it is uncertain whether these data are applicable to out-sourced personnel. The number of measurements was relatively small ( N = 217) leading to lower precision of the estimated exposure of these workers as well as foremen and administration, N = 148 and 226, respectively ( Table 3 ). As cleaning workers had the highest exposure workers in out-sourced companies should be included in occupational health surveillance programs.

The thoracic aerosol levels varied significantly between seasons showing the highest levels in the autumn and lowest in the winter. As measurements were performed in different seasons, partly also in the same plant, this was an important factor to adjust for. Differences between sampling campaigns were expected as changes in production volumes occurred during the study. Small differences between exposure levels during sampling years in all plants combined were found in the observed data, ranging from 2.1mg m ⁻³ in 2007 to 2.3mg m ⁻³ in 2011. Differences between sampling campaigns within each plant were larger. However, the effect of year of sampling was negligible in a random effects model with plant, job type, worker, year, and season.

The use of respirators increased with increasing aerosol levels, indicating that the measured aerosol levels overestimate the actual exposure. For thoracic aerosol measurements <0.5mg m ⁻³ only 9% of the workers reported using a respirator most or all of the time. These measurements approximate the actual internal exposure of the workers more closely than measurements at higher aerosol levels, where respirators are used more frequently.

The statistical models showed good fit. Although the present study included 22 cement plants from eight countries generalization of the findings is probably restricted to plants with a relatively good occupational hygienic standard because the companies were members of CEMBUREAU and had access to occupational health personnel. The estimated exposure levels found in this study (0.21–2.0mg m ⁻³ ) also seem at the lower range of levels in other studies reporting GM and AM exposure levels for total dust of 0.04–550mg m ⁻³ and respirable dust of 0.22–43mg m ⁻³ (review HSE, 2004 ; Mwaiselage et al. , 2005 ; Mirzaee et al. , 2008 ; Zeleke et al. , 2010 , 2011 ; Fell et al. , 2011 ; Kakooei et al. , 2012a , 2012b ). Two studies have reported thoracic aerosol levels, and their measurements are included in the present study ( Fell et al. , 2010 ; Nordby et al. , 2011 ). Without published data on thoracic aerosol levels, it is difficult to compare our results to previous studies. The relationships between the thoracic, ‘total’ dust, inhalable and respirable fractions will therefore be subject for further study. Construction of a job exposure matrix is planned as a basis for the epidemiological analysis study of the longitudinal lung function data as well as a detailed analysis of tasks and other determinants recorded by the measurement questionnaires.

Conclusions

The thoracic aerosol levels in the cement production industry varied between job types and even more between plants. Significant differences between job types were observed with production, cleaning, and maintenance work as the highest exposed work tasks. The number of employees at the plants was the most important determinant explaining the differences in exposure between plants. Exposure reached levels where the risk of lung function loss may be increased. No differences in exposure during the 4-year study period were observed. Respirator use was minor at levels <0.5mg m ⁻³ but was more common at higher levels.

FUNDING

European Cement Association (CEMBUREAU), Brussels, Belgium. No indirect sources of support have been received by the authors.

APPENDIX

ACKNOWLEDGMENTS

Without the extensive efforts and contributions from the national coordinators, technical and health personnel, and management and employees of the plants, this project would not have been accomplished. We greatly acknowledge Grete Friisk, Carl Fredrik Halgard, Thea Halgard, Kristin Hovland, Susanne Nordby, Heidi Notø, Ingrid Notø, Elianne Jørgen Seberg, and Marte Skjønsberg for practical assistance, Asbjørn Skogstad for microscopic analysis, Terje Nilsen and Ole Synnes for technical support, Charlotte Wu Homme for linguistic help, and Berit Bakke for critically reviewing the manuscript.

REFERENCES