Job Tasks as Determinants of Thoracic Aerosol Exposure in the Cement Production Industry

Introduction

Cross-shift studies have been performed in the cement industry indicating that there might be some possible health risks concerning lung function among their workers (Fell et al., 2017) and with the implementation of REACH The European Cement Association (CEMBUREAU) decided to investigate this. The present study is part of this initiative to improve knowledge and we were appointed to organize a longitudinal study of exposure and lung function in 2006 with planned sampling start in 2007 and follow-ups in 2009 and 2011. Calculations of power indicated that a follow-up of 2000 persons were needed to detect a putative small loss in lung function due to exposure during this period. We decided to measure the thoracic aerosol fraction as the best suited particle size fraction for comparison with lung function (CEN 1993).

In a previous paper, plant-specific characteristics together with job types were analyzed as predictors of exposure among cement production workers (Notø et al., 2015) and in the present paper we will explore information on tasks as determinants of exposure.

Numerous determinants can be recorded and evaluated on plant, department, job type, and individual level in cement plants. Identifying influential determinants is a crucial part of the exposure characterization. Relevant determinants of exposure may differ according to the purpose, regarding epidemiological studies versus exposure control (Burstyn et al., 1999). In the early 1990’s Rappaport and Kromhout published comprehensive evaluations of within- and between-worker components of variance in occupational exposure studies (Rappaport 1991a, 1991b, 1993, 1995; Kromhout et al., 1993) and since then many studies have used mixed models to estimate and predict workers exposure (Burdorf 2005; Symanski et al., 2006).

This is the first study to our knowledge identifying determinants of thoracic aerosol exposure and analyzing variance components based on mixed models in the cement production industry. In this paper, we focus on job tasks that influence the exposure for the main job groups. We also studied the influence of these determinants on the between-plant (BP) (plant), between-worker-within-plant and within-worker (WW) variance components.

Materials and Methods

A total of 6111 personal thoracic aerosol measurements of 2534 workers in 22 cement production plants situated in seven European countries and Turkey were available from the previous study. The data collection took place between 2007 and 2012. Repeated measurements are included in the data with 28, 17, and 14% of the persons having 2, 3, or >3 repeated measurements, respectively (Notø et al., 2015).

Questionnaires

A questionnaire (Supplementary Material A1, available at Annals of Occupational Hygiene online) was developed to collect information about the workers’ job types and tasks and was mainly filled out by personnel employed at each plant (hygienists, health, and safety workers). A few measurements from 2009 were also conducted by the investigators themselves. These were also used in the paper (Notø et al., 2016). In agreement with the national coordinators and employees from some of plants, we selected determinants suspecting to influence the exposure but still keeping the questionnaire as short and simple as possible. We decided to group the workers into seven job types: production, cleaning, maintenance, foreman administration, laboratory, and other. The workers indicated one or more job types performed at the day of sampling.

More detailed information about time spent on different tasks within the jobs was recorded in the questionnaire (questions 8 in Supplementary Material A1, available at Annals of Occupational Hygiene online). Time spent on a specific task was recorded in six categories, from <15 min to a full shift. Job situations regarded by the research team as being high-exposed, especially cleaning tasks, were recorded by a second set of questions (questions 9 in Supplementary Material A1, available at Annals of Occupational Hygiene online) where workers reported hours and minutes spent on each task.

Data analysis

Measurements from workers that checked for more than one job type (the category several jobs from Notø et al., 2015) or those that indicated time absent from work during the shift were excluded from the original data set and a total of 4765 measurements from 2373 persons were now available for the exposure modeling (Table 2).

For the job types production, cleaning, maintenance, administration, and laboratory, we had predefined the tasks that were typical for each group. When workers spent less than half a shift (<210 min) on these tasks, measurements were also excluded from the analysis. The work of foreman and workers with other jobs could change from shift to shift depending on situations in the production process or temporary lack of personnel in the other job types. All determinants from the main questionnaire and the significant plant determinants from the technical questionnaire (Notø et al., 2015) were therefore included in the analysis of the job categories foreman and other. The determinants included in exposure models for each job type are listed in Table 1. The job types have been described in more detail previously (Notø et al., 2015).

The exposure data were log-normal distributed and were log₁₀ transformed before statistical analysis. Different categorizations of the determinants were explored and an example representing the production job type can be found in Supplementary Table B1 (available at Annals of Occupational Hygiene online). The alternative with three categories was selected for simplicity. The categories were no time spent, between zero and 210 min (up to half a shift), and 210 or more minutes (more than half a shift) spent during the work shift. If the number of measurements in one or both of the two categories for spent time (less and longer than half a shift) was <5, these measurements were combined into one category. If this combined category still had <5 measurements, the determinant was not explored. As women contributed with only 1.5% of the measurements the effect of gender was not explored.

Correlations between the determinants belonging to each job type were evaluated by Spearman correlation coefficients (r_S). If pairs of determinants had r_S values >0.8 both determinants were tested in separate mixed models. The determinant from the model with the highest Akaike information criterion (AIC) was excluded from the model.

Modeling was performed separately for all seven job types to study the contribution of the determinants to the WW, between-worker within-plant (BW), BP, and total variance. All models were specified with plant and worker as independent nested random intercepts estimated by maximum likelihood. Analysis of variance type tests of the significance of each determinant in the models were calculated by the contrast command in STATA.

Mixed models with no fixed effects (null model), and models including all plant determinants (plant), with all job determinants (job) and with both plant and job determinants (plant&job) as fixed effects were tested. The number and type of job determinants varied with the job type while the plant determinants were the same for all job types. The questionnaire included only one job determinant for laboratory workers. In order to select models containing only the most important determinants, we chose for each job type the model with the lowest AIC value (hereafter referred to as “best model”). This model selection procedure compared all possible alternative mixed models (2^{number of determinants}) and was developed and implemented in the statistical program R (Supplementary Material A2, available at Annals of Occupational Hygiene online).

The data were analyzed using STATA 14.2 (StataCorp LP, College Station, TX, USA) and R version 3.2.5, (The R Foundation for Statistical Computing, Vienna. Austria).

Results

The applied restriction of measurements (see footnote to Table 2) resulted in small changes of the arithmetic mean (AM) exposure and geometric standard deviations (GSD) for most job types, except for cleaners where AM exposure increased by 0.8 mg m⁻³ and GSD decreased from 4.0 to 3.2 on the restriction of samples, compared to AM (GSD) among all the available samples from the original data set.

The null, plant, job, and plant&job models obtained for all job types are shown in the online Supplementary Tables B5a–B5g (available at Annals of Occupational Hygiene online).

Production work

The correlations between job determinants belonging to this job type were low with r_S between −0.29 and 0.23.

Variance components

The best model is shown in Table 3 and the other models in Supplementary Table B5a (available at Annals of Occupational Hygiene online). In the null model, the BP variance, the BW variance, and the WW variance accounted for 20%, 27%, and 53%, respectively, of the total variance. The effect has been calculated by using effect size = 10^(β1−1) * 100 and compared to the corresponding reference value of the covariate (100).

Introducing plant determinants to the null model decreased the BP by 55% while the BW and the WW were unchanged. Job determinants explained 30% of the BW while the BP and also the WW showed small reductions (4–5%). When all plant and job determinants were in the model BP was reduced by 61% compared to the null model while the BW and WW were reduced by 30% and 4%, respectively. The best model explained 21% of the total variance in the null model and the variance components were close to the plant&job model.

Effects of determinants

Work in the control room and handling of fuel was associated with lower exposure. For those spending <210 min with these tasks, the thoracic aerosol exposure was 24% and 16% lower, respectively, compared to work without these tasks. For those working more than half a shift in the control room or with handling of fuel, the exposures decreased with 54% and 52%, respectively. Packing and shipping was associated with increased exposure (12% and 33%) when working less than and more than half a shift, respectively.

The other tasks showed more complicated associations. Less than half a shift work with the tasks cleaning in raw meal, cement, and filter areas, and de-clogging of the cyclones increased the exposure by 52, 83, 104, and 36%, respectively. Those that spent more than half a shift with these tasks also had increased exposure by 11, 37, 31, and 27%, respectively, but less than the previous category. Plant determinants showed significant contributions to the models for production workers. If the production workers were employed in plants that had been subjectively assessed as less clean their exposure were 80% higher than for those employed in plants regarded as cleaner. Production workers from plants with a medium-size workforce (144–204 workers) were less exposed (21%) than those from plants with fewer employees. Production workers from plants with >212 workers had 43% higher exposure than those from plants with <138 workers.

Example of the estimation of the thoracic aerosol exposure for a production worker

Exposure of production workers can be estimated from the regression equation of the best model of the production job type shown in Table 2:

Log₁₀Thoracic aerosol_{Production worker} = −0.17 − 0.12 * (Control room 1–209 min) − 0.34 * (Control room ≥ 210 min + 0.048 * (Packing &shipping 1–209 min) + 0.13 * (Packing &shipping ≥ 210 min) + 0.18 * (Cleaning in raw meal areas 1–209 min) + 0.046 * (Cleaning in raw meal areas ≥ 210 min) + 0.26 * (Cleaning in cement areas 1–209 min) + 0.14*(Cleaning in cement areas ≥ 210 min) + 0.31 * (Cleaning of filters 1–209 min) + 0.12 * (Cleaning of filters ≥ 210 min) + 0.13 * (De-clogging of cyclones 1–209 min) + 0.10 * (De-clogging of cyclones ≥ 210 min) − 0.08 * (Handling of fuel 1–209 min) − 0.32 * (Handling of fuel ≥ 210 min) − 0.010 * (Number employed 144–204) + 0.16 * (Number employed 212–483) + 0.26 * (Tidiness Less Clean).

Example 1: A production worker working less than half a day with packing or shipping, and less than half a day with cleaning of filters and employed at a plant with a large work force and assessed as clean will have an estimated exposure of:

Log₁₀ Thoracic aerosol_Example1 = −0.17 + 0.048 * (Packing &shipping 1–209 min) + 0.31 * (Cleaning of filters 1–209 min) + 0.16 * (Number employed 212–483) = −0.17 + 0.048 + 0.31 + 0.16 = 0.348.

The GM exposure estimate is the antilog of log₁₀ Thoracic aerosol_Example1 = 10^0.348 = 2.2 mg m⁻³.

Example 2: Another production worker from a plant of the same size category and tidiness who has been working less than half a shift in the control room, and less than half a shift with handling of fuel will have an estimated exposure of:

Log₁₀Thoracic aerosol_Example2 = −0.17 − 0.12 * (Control room ≥ 210 min) − 0.08 * (Handling of fuel 1–209 min) + 0.16 * (Number employed 212–483) = −0.21.

The estimated GM exposure for this worker is: 10^−0.21 = 0.62 mg m⁻³.

Maintenance

The job determinants electrical and mechanical maintenance were correlated (r_S −0.77), and also the plant determinants number of employees and maximum cement production (r_S 0.91). The other correlation coefficients for determinants in the plant&job model varied between −0.29 and 0.26. The correlation coefficients between determinants in the best model varied between −0.28 and 0.09. The best model is shown in Table 4 and the other models in Supplementary Table B5c (available at Annals of Occupational Hygiene online). The determinants of dismantling before repair work and tidiness are included in the best model because of lower AIC although P contrast values did not reach statistical significance (P = 0.059 and 0.088).