Styrene is used as a monomer in the production of polystyrene and styrene copolymers, styrene-butadiene rubbers, and in the manufacture of consumer products including house wares, insulation products, tyres and a wide variety of plastic and synthetic rubber products (IARC, 2018). It was first isolated in 1831 but only became commercially important in the late 1930’s when it began to be used in the manufacturing of synthetic rubber (IARC, 1994). Its global market is estimated to be worth around USD 97.3 billion by 2032 (Precedence Research, 2024). Occupational styrene exposure has been linked to neurobehavioral effects (Benignus et al., 2005; Sliwinska-Kowalska et al., 2020), ototoxicity (ATSDR, 2010), neurological diseases (Iversen et al., 2021), autoimmune rheumatological diseases (Hjuler Boudigaard et al., 2020) and hematopoietic, sinonasal and lung cancer (Christensen et al., 2017; Nissen et al., 2018; Bertke et al., 2021; Kogevinas et al., 1994). In 2018, The International Agency for Research on Cancer’s Monographs programme classified styrene as probably carcinogenic (Group 2 A), based on limited evidence of an increased risk of lymphohematopoietic malignancies in humans and sufficient evidence of carcinogenicity in experimental animals. This was supported by strong evidence of a mechanism that also operates in humans (IARC, 2018). The highest exposures to styrene occur in the reinforced plastics industry where there is little or no co-exposure to known carcinogens (Van Rooij et al., 2008; Jensen et al., 1990), thus reducing the chance of confounding. Different manufacturing sectors fall under the glass reinforced plastics (GRP) industry with boat building, car manufacturing, and construction among the most prominent.

The main processes during manufacture of GRP products include hand lay-up, spray lay-up, continuous lamination, pultrusion, filament winding and various closed-moulding operations. The individual process steps and the order in which they usually occur are: “(1) preparation of the mould (polishing and waxing), (2) gel coating, (3) mixing of the resin, (4) preparation of the glass fibre mats, (5) application of glass fibre and resin in the mould (lamination), (6) working of the applied material to remove enclosed air (rolling), (7) curing, (8) trimming, (9) painting of the ‘raw’ side of the laminate (top coating, optional), and (10) finishing.” (Kogevinas et al., 1994). These are typical across facilities, but practices vary across industrial sub-sectors and between facilities. While in some facilities there is a clear division of labour with workers assigned to one or more of these steps, in others, workers who conduct hand- or spray lay-up, for instance, may also engage in gel-coating, rolling, cutting of fibre mats or other process-related tasks (Crandall, 1982; Kogevinas et al., 1994). Also, techniques such as spray- and hand lamination are often combined in the manufacture of a single product (Crandall, 1982; Kogevinas et al., 1994).

One of the first comprehensive investigations into a possible link between styrene exposure and cancer began in 1982, with the establishment of an European multicentre occupational cohort study comprising 41,167 workers from Denmark, Finland, Italy, Norway, Sweden, and the UK (IARC, 1994; Kogevinas et al., 1994). A job-exposure matrix (JEM) was constructed for the exposure-response analysis (Kogevinas et al., 1994). This JEM was mainly based on styrene measurements in air collected from each country included in the cohort. Exposure metrics considered were length of exposure; cumulative exposure; time since first exposure (TSFE) and average level of exposure. With styrene production continuing apace and its proven presence in human blood (Brugnone et al., 1993), clarifying whether there is a causal association with cancer is important. This requires epidemiological studies that have sufficient power and follow-up time and include accurate exposure estimates in order to evaluate risks for subtypes of leukaemia and lymphoma.

With the aim of investigating the exposure-response relationship between styrene and subtypes of lymphohematopoietic and other malignancies in Europe and the US, cohorts of reinforced plastic workers from Finland, Italy and the UK included in the original European multicentre occupational cohort study (Kogevinas et al., 1994), the Danish cohort (Christensen et al., 2018) and a US cohort (Bertke et al., 2021) have been updated with respect to cancer incidence and mortality (Christensen et al., 2021). To facilitate this investigation, Job Exposure Matrices described so far for the European cohorts only, need to be extended to the US. We aim to develop an empirical exposure model based on personal styrene measurements from all countries in the pooled cohort study and, based on this model, derive quantitative estimates of average, full-shift occupational exposure to styrene for the period between 1947 and 2020. This is the period that is covered by the pooled cohort (Christensen et al., 2021).

During this study, we created a predictive model based on personal exposure measurements to be used for estimating styrene exposure in an international cohort study on risks of lymphohematopoietic neoplasms, other cancers, and chronic non-malignant diseases in the European and the US glass reinforced plastics industry. The model may also be used in other styrene exposed reinforced plastics cohorts with information about product, process, and calendar year, where there is an indication of similarity in work contexts. Due to limitations imposed by new data sharing legislation (EU, 2016), we were unable to access all the historic styrene exposure datasets that had been used for exposure assessment in the original cohort study by Kogevinas et al. (1994). However, it was possible to include additional exposure data from the US, UK and Norway resulting in this being the most comprehensive international styrene exposure assessment based on the largest number of personal exposure measurements since the 1990s. In the absence of data to perform an external validation the use of cross validation ensured selection of the most robust model.

Like our study, Christensen et al. (2018) reported the highest exposure levels from boat manufacturing companies. Their predictions of styrene exposure intensity were based on a linear mixed effects model at a company level with similar independent determinants to our model (i.e. main production process, main product, and decade). This is not surprising, because their data, covering the period 1970–2011 in the Danish reinforced plastics industry, were a subset of the data used in this study (~26% of our dataset). Serdar et al. (2006) found that styrene exposure varied by product even when job title was restricted to laminators; we noticed the same pattern. This was true for job titles within the same industrial sub-sector as well as across industries (not shown). Consequently, depending heavily on job title might introduce error in exposure estimations.

It is difficult to ascribe the reasons for the regional differences that we observed. However, similar country and regional differences have been reported for exposure other than styrene (Basinas et al., 2023; de Vocht et al., 2006; Liu et al., 2011; Peters et al., 2011). Safety culture, workplace practices, workplace layouts and equipment vary across companies, countries, and regions (Deadman et al., 1996; Johansen et al., 2002). The GRP industry is no exception. As previously mentioned, distinct steps in processes could be assigned exclusively to specific workers in some facilities while in others, several steps might be performed by the same worker. For instance, the job title ‘gel-coater’ or process ‘gel-coating’ implies that this job title is associated with the distinct tasks of ‘gel-coating’. However, this is only true for some facilities. In others, the gel-coater also engaged in other steps in the lamination process. This held to a lesser extent for the job title and process ‘moulder’ and ‘moulding’. Regarding differences in control measures some workplaces partitioned work areas to limit the dispersion of resin to areas not directly engaged in the process (Crandall and Hartle, 1985). Work practices like these required by policies of regional authorities, such as is the case across the European Union, can result in regional differences in exposure. Our results suggest that differences in exposure among European countries and regions should be considered when estimating styrene exposure in epidemiological studies, noting that based on our model the differences were more pronounced in the past, certainly before 2000 (details not shown).

We identified repeated measurements from the US and Norway but cannot rule out that there are other repeated measures in our dataset that have not been tagged as such and would have contributed to the residual (within-worker) variance rather than being decomposed into between vs within worker variances. The unidentified repeated measures would also violate the assumption of the residual independence of measurements, but it is difficult to anticipate what bias this may have produced in our predictions. It is likely that a portion of the unexplained variance in our model comes from the between-worker variance that has been attributed to personal factors such as seniority/experience, age, and casual short-term work in the GRP industry (Serdar et al., 2006). Another contributing source would have been the between-facility variation resulting from different work practices and control measures.

The declining time-trend that we observed is consistent with what has been observed across industries using similar modelling approaches (Creely et al., 2007; Basinas et al., 2023; Kromhout and Vermeulen, 2000; Symanski et al., 1998). The accuracy of model predictions improved over the years. This may have been due to the limited availability of data in the earlier time periods along with less consistency in operations, including regulations around control measures, across the industry. However, although in the earlier years (<1975) the model predicts moderately high in terms of absolute exposure values, it reliably identifies relative differences in exposure across all year. Relative differences in exposure between worker/cohort members, which is important for epidemiological analyses, would therefore been correctly assessed. Furthermore, the sensitivity analysis where those measurements were removed showed that this did not appreciably change the model outcome.

The assignment of exposure estimates from 1962 to the years prior to 1962 that were outside the applicable range of the model is a considerable gap, but in the absence of data for that time it is a plausible approach. When faced with a similar challenge of extrapolation back to times where measurements were not available, de Vocht et al. (2009) considered a variety of extrapolations to determine pre-data time--trends. A similar approach can also be implemented in exposure assessment that relies on our models as part of sensitivity analyses (de Vocht et al., 2009). There is less of an issue with the predictions for the recent period of 2019–2020; the 2018 levels were used since there was stability in exposure trends across all countries for the second half of the 2000’s indicating that these were the best predictions achievable. Moreover, this becomes irrelevant when applied in the cohort study since with the appropriate lagging of exposure for cancer analyses and, due to the strong decreasing trends over time, the contribution of these 2 years to the overall cumulative exposure is negligible.

Products small enough to be manufactured using machine moulds and closed processes result in lower exposures since workers do not need to be close to the source of styrene (Lenvik et al., 1999). When small items are manually handled in an open process, the exposure potential depends on the number of small parts being made (Crandall, 1982) but is likely to be low relative to the potential exposure when producing large items. Larger items such as (parts of) boats, vehicles, and panels used in construction usually involve manual processes with long periods of exposure close to the source. This, in part, explains the predictive power of product and process over job title. While there is added value in including job title where such information is available it may not appreciably increase the explained variation in multiple regression mixed models for this industry. This lends support for an approach to exposure assessment for the GRP industry that prioritises product and process over job title as predictors of exposure.

Exposure metrics that can be calculated based on the exposure matrix include, but are not limited to, cumulative exposure (mg/m³-years); average exposure (cumulative exposure/total exposure time (years)); number of (consecutive) years of exposure above thresholds. Our explicit nature of assumptions about exposure assessment captured in our mixed effects models and equations that describe how we assessed exposures created a reproducible method that enables sensitivity analyses exemplified by the work of de Vocht et al. (2009). Our models also provide information about some of the imprecision entailed in such retrospective estimates within the exposure matrix and exposure metrics that result from its application to occupational histories. Understanding uncertainty in our prediction model can inform analysis of the bias that application of our exposure assessment approach can produce in epidemiologic analyses of exposure-response associations, as was done by Sallmén et al. (2024). Addressing systematic sources of bias that arise from complex measurement error structures should aid interpretation of risks attributed to styrene, and increase confidence in the conclusions of the epidemiologic analysis.

The styrene exposure predictions based on the statistical model described in this paper were used to construct an exposure matrix that can be used to retrospectively assess the exposure of workers in the reinforced plastics industry for the period 1947 to 2020. This may help to enhance the quality of epidemiological studies on styrene exposure and health effects.