Estimating Trends in Quartz Exposure in Swedish Iron Foundries—Predicting Past and Present Exposures

ABSTRACT

Background:

Swedish foundries have a long tradition of legally required surveys in the workplace that, from the late 1960s onwards, included measurements of quartz. The availability of exposure data spanning almost 40 years presents a unique opportunity to study trends over that time and to evaluate the validity of exposure models based on data from shorter time spans. The aims of this study were (i) to investigate long-term trends in quartz exposure over time, (ii) using routinely collected quartz exposure measurements to develop a mathematical model that could predict both historical and current exposure patterns, and (iii) to validate this exposure model with up-to-date measurements from a targeted survey of the industry.

Methods:

Eleven foundries, representative of the Swedish iron foundry industry, were divided into three groups based on the size of the companies, i.e. the number of employees. A database containing 2333 quartz exposure measurements for 11 different job descriptions was used to create three models that covered time periods which reflected different work conditions and production processes: a historical model (1968–1989), a development model (1990–2004), and a validation model (2005–2006). A linear mixed model for repeated measurements was used to investigate trends over time. In all mixed models, time period, company size, and job title were included as fixed (categorical) determinants of exposure. The within- and between-worker variances were considered to be random effects. A linear regression analysis was performed to investigate agreement between the models. The average exposure was estimated for each combination of job title and company size.

Results:

A large reduction in exposure (51%) was seen between 1968 and 1974 and between 1975 and 1979 (28%). In later periods, quartz exposure was reduced by 8% per 5 years at best. In the first period, employees at smaller companies experienced ∼50% higher exposure levels than those at large companies, but these differences became much smaller in later years. The furnace and ladle repair job were associated with the highest exposure, with 3.9–8.0 times the average exposure compared to the lowest exposed group. Without adjusting for this autonomous trend over time, predicting early historical exposures using our development model resulted in a statistically significant regression coefficient of 2.42 (R2 = 0.81), indicating an underestimation of historical exposure levels. Similar patterns were seen for other historical time periods. Comparing our development model with our validation model resulted in a statistically significant regression coefficient of 0.31, indicating an overestimation of current exposure levels.

Conclusion:

To investigate long-term trends in quartz exposure over time, overall linear trends can be determined by using mixed model analysis. To create individual exposure measures to predict historical exposures, it is necessary to consider factors such as the time period, type of job, type of company, and company size. The mixed model analysis showed systematic changes in concentration levels, implying that extrapolation of exposure estimates outside the range of years covered by measurements may result in underestimation or overestimation of exposure.

MATERIALS AND METHODS

In total, 11 iron foundries of different sizes that use a mixture of manual and mechanized (old and new) moulding and casting production techniques were selected as a representative sample of the Swedish iron foundry industry (Andersson et al., 2009). The foundries used various types of sand, mould, and core binders and a variety of production methods. The numbers of employees at the foundries ranged between 8 and 388, with production capacities ranging from 400 to 120 000 tonnes year–1. The companies were divided into three groups: small, medium, and large as defined by the number of employees (<20, 25–120, and >120, respectively).

Exposure measurements

Recent exposures of respirable quartz were determined from 415 personal full-shift exposure measurements sampled between April 2005 and May 2006. These measurements were collected as part of a large research project, which is described in detail in another publication (Andersson et al., 2009). Personal measurement data were available from compulsory measurements performed by the 11 foundries between 1975 and 2004 and national exposure surveys between 1968 and 1974, the latter being provided by the Swedish Work Environment Authority (SWEA). The resulting database compiled from these two sources of routinely collected exposure information, along with the recent exposure survey, contained 2333 air concentration measurements of respirable quartz with sampling times ranging from 240 to 600 min (Table 2). Of the measurements, 1691 (72%) were ‘unique’, i.e. related to different workers.

The job titles used in the database (which included recent survey results and historical measurements) were caster, core maker, fettler, furnace and ladle repair, maintenance, melter, moulder, sand mixer, shake out, transportation, and others. The job title ‘others’ included cleaners, painters, and model carpenters.

Sampling and analysis

The sampling and laboratory methods used in the analysis have been reported elsewhere in detail (Andersson et al., 2009). For the measurements before 1980 (19% of the total number of measurements), total dust samples were collected and a sedimentation method was used to separate the fine fraction prior to X-ray diffraction analysis. The quartz concentrations determined by this older method were compared to the new method, and the SWEA regarded the measurements obtained by the old method were twice as high as standard cyclone respirable quartz concentration levels; however, the lack of published data on this conversion factor implies a high uncertainty level for individual converted measurement data. The quartz levels determined with the old method were consequently reduced by a factor 0.5, i.e. converted to comparable respirable quartz levels and entered into the database (Orenstein, 1965; NBOSH, 1979). Exposure measurements were performed using Swedish and international standards. However, to facilitate comparisons with historical measurements in an ongoing epidemiological study, sampling was undertaken using a cyclone system whose characteristics were defined by the Johannesburg convention (NBOSH, 1979), with a flow rate of 1.9 l min−1, as compared to the present standard based on a flow rate of 2.3 l min−1 (Swedish Standard, 1995).

Statistical methods

The concentrations of respirable quartz were calculated as time-weighted averages (TWAs). Concentrations below the detection limits were estimated by reducing the detection limit by formula (Hornung and Reed, 1990). Since the exposure measurement data were skewed, the quartz concentrations were log-normally transformed before devising a linear mixed-effect model for repeated measurements. The measurements taken in the period from 1968 to 2004 by both the SWEA and the surveyed companies were used to build two of the three different mixed models described in this paper. The first model (the historical model) was based on measurements taken between 1968 and 1989 (22 years), reflecting the period where there were substantial improvements in working conditions (Table 1). The second model (the development model) was based on the measurements conducted between 1990 and 2004 (15 years), a period of relatively minor changes in the production process. This model was used to predict historical exposure patterns. This may reflect a commonly occurring situation, where exposure information is available for more recent times but is not available for older periods. The third model (the validation model) was derived from the measurements taken by the research group between 2005 and 2006. This validation model allowed us to evaluate the validity of the development model, thus comparing routinely collected data by a company with a structured measurement programme for research purpose a few years later.

Mixed model

The mixed model used the following equation:

where Xghkij = the quartz concentration measured for the ith worker on the jth day at the gth time period in the hth foundry with kth job title; Yghkij=ln(Xghkij),

μ = the overall average quartz concentration on a log scale, αg = the fixed effect of the gth time period when g = 1,…,4, βh = the fixed effect of the hth foundry when h = 1,…,11, φk = the fixed effect of the kth job title when k = 1,…,11, δi = the random effect of the ith worker, and εghkij = the random within-worker variation.

The model assumed that δi and εghkij are normally distributed with means equal to zero and variances of σ2BW and σ2WW, respectively, representing the between-worker and within-worker variance components. Furthermore,δi and εghkij were assumed to be statistically independent of each other.

A linear mixed-effect model for repeated measurements was used to describe trends over time. In all mixed-effect models, the time period, company size, and job title were included as fixed (categorical) determinants of exposure. Approximately 56% of the 2333 quartz measurements were repeated measurements within workers. These variance components were pooled across all workers and approximated as equal across all time periods, job titles, and company sizes. This approximation, though eliminating information such as the dependence of variances on time period, was chosen because of the relatively few measurements available for some determinants, which limited the number of parameters that could be estimated in the model (Burdorf, 2005). The Akaike information criterion was used as a measure of the overall fit of the mixed model. The contribution of the determinants of exposure was evaluated by their influence on the estimated mean exposure as well as their influence on the reduction of the between-worker variance. It has to be noted that the fixed effects were collected at the individual worker level and, thus, could not have any impact on the within-worker variance.

A linear regression analysis was performed to investigate agreement between the historical model and the development model and also between the development model and the validation model. For each combination of job title and company size, the average exposure was estimated in each model, resulting in 33 comparisons. For example, using a linear regression model, the intercept reflects the systematic difference in quartz exposure estimates, and the regression coefficient represents the change in predicted concentration between 2005 and 2006 due to a one-unit change of the estimated concentration in the development model based on measurements from 1990 to 2004. All analyses were conducted using the Proc Mixed code in SAS version 6.12 software (SAS Institute, Cary, NC, USA).
RESULTS

The distribution of measurements across time periods, company size, and job titles is presented in Table 1. The total number of measurements for each 5-year time period range from 208 to 415, generating a large database divided into a historical model with 1109 measurements, a development model with 809 measurements, and a validation model with 415 measurements. The number of measurements produced from the four largest companies was considerably higher than the number of available measurements from the four medium and three small companies. Some job titles had relatively few measurements associated with them, most notably transportation and caster. The overall mean respirable quartz exposure (Table 2) ranges from 0.0018 to 6.4 mg m−3, where the highest exposures are seen in furnace and ladle repair workers [arithmetic mean (AM) = 0.45 mg m−3] and the lowest exposures in core makers (AM = 0.024 mg m−3).

Trends in quartz exposure are presented as regression coefficients for the different mixed models (Table 3). The exposure levels in the periods 1968–1974 and 1975–1979 were 51 and 28% higher, respectively, than the exposure levels between 1985 and 1989. In later periods, quartz exposure was reduced by 8% per 5 years at best. In the first period, smaller companies had ∼50% higher exposure levels than large companies, but these differences became much smaller in later years. The data used for the model fit are presented in Table 3 as the percent reduction of the between-worker variability, being 27, 48, and 19% for the historical, development, and validation models, respectively. Figure 1 depicts the long-term trend in exposure for three typical jobs, illustrating the sharp decline between 1968 and 1984 and the slow decrease after 1985. The furnace and ladle repair jobs consistently had a high exposure, with a 3.9–8.0 times higher average exposure than the core maker.

The comparison of the three mixed models for different time periods showed the same relative ranking of exposure levels across jobs in the three models with lowest exposures for core makers and highest exposures for furnace and ladle repair personnel (Table 4). However, profound differences were observed in actual exposure levels, both in absolute and in relative terms.

The agreement between the estimated exposure in the first time period by the historical model and the last time period by the development model showed a regression coefficient (B) = 2.42, 95% confidence interval (CI) 1.99–2.84, an intercept value of 0.018 and an R2 value of 0.81 (Fig. 2), showing that the development model underestimates the historical exposure levels between 1968 and 1974 by a factor of 2.4. Comparison with other time periods in the historical model showed that the agreement tends to improve with the later periods, as can be seen from the decreasing B values (2.42, 1.66, 1.20, and 1.19) and intercept values. The corresponding data for the agreement between estimated exposure in the last time period of the development model and the validation model produced regression values of B = 0.31, 95% CI 0.15–0.47, an intercept value of 0.019, and an R2 value of 0.33 (Fig. 3), indicating that the development model overestimates current exposure levels.
