Determinants of respirable crystalline silica exposure in Australian tunnelling

Abstract

Exposure to respirable crystalline silica (RCS) is a well-established cause of silicosis, lung cancer, and other serious health conditions. Contemporary data describing RCS exposures among tunnel construction workers in Australia has been limited, restricting an understanding of the nature and magnitude of the risk of exposure. This observational study addresses that gap by accessing, compiling, and evaluating the largest known dataset of RCS exposures in contemporary Australian tunnelling. Data were obtained through freedom of information requests and parliamentary records, resulting in 3,288 individual personal exposure measurements across 12 tunnel projects constructed between 2016 and 2024, employing a range of construction methods: via tunnel boring machines, drill and blast, cut-and-cover, and mined tunnelling. Determinants of exposure were assessed on ln-transformed data using linear mixed models (LMM), and included primary location, proportion of crystalline silica content in respirable dust, tunnelling method, or specific above-ground work location. LMMs explained between 53% and 68% of the observed variability in RCS exposures. Underground workers were predicted to have 2.3 times the RCS exposures of above-ground workers (geometric mean [GM] 0.026 vs 0.012 mg/m³); and each 1% increase in the proportion of RCS in respirable dust corresponded to a predicted 5% increase in RCS exposure. Estimated RCS exposure was lower after the reduction of the workplace exposure standard in July 2020 (GM 0.023 vs 0.013 mg/m³). Above-ground exposure varied by location (eg precast facilities, spoil sheds, or general site areas), while tunnelling method influenced RCS exposure underground (eg mined tunnelling, using tunnel boring machines or cut-and-cover methods). While limitations such as variable data quality and incomplete task attribution are acknowledged, the findings support the need for targeted control measures and transparent access to data across tunnel construction environments.

What’s Important About This Paper.

This study provides the most comprehensive analysis of respirable crystalline silica exposures among Australian tunnel workers to date. Findings reveal high levels of exposure, influenced by workers' location, tunnelling method, time period, and crystalline silica content.

Introduction

Australia's substantial investment in transport and energy infrastructure has led to a marked increase in tunnelling projects, particularly in the country's most populous state of New South Wales (NSW), where rapid urban expansion has driven major development (ITA-AITES 2024). Tunnel construction involves excavating crystalline silica-containing rock, a process that can generate substantial amounts of airborne respirable crystalline silica (RCS). This RCS is also released into the air through applying or disturbing materials such as concrete and grout, which are used to support and line tunnels, declines, and shafts. Cumulative exposure to RCS is known to increase the risk of the development of silicosis, in addition to lung cancer and other diseases (IARC 1997; Leung et al. 2012).

Within the construction sector, tunnel construction workers (TCWs) have been identified as one of the groups experiencing high levels of RCS exposure (Workplace Health and Safety Queensland 2013) with studies confirming cases of silicosis in this cohort in Australia (Seevnarain et al. 2021; Turner and Yates 2022; Muir 2024). Tunnelling projects employ thousands of workers, therefore levels of RCS exposure and the way that it is managed have the potential to significantly influence the national burden of silica-related disease. Studies published on this topic internationally from Norway (Ulvestad et al. 2001; Bakke et al. 2001a; Bakke et al. 2001b; Bakke et al. 2002; Leite et al. 2023), Sweden (Inyang 2024) and the UK (Galea 2015; Galea et al. 2015), show that high levels of RCS exposure can occur while using various tunnel construction methods. However, access to similar data in Australia has been limited to one regulatory intervention (Workplace Health and Safety Queensland 2010) and a series of smaller case studies (Henderson 2001; Cole 2016a, 2016b; Cole and Fisher 2019; Cole et al. 2021) (Table S1).

Differences in geological formations in some areas of Australia, particularly in NSW, suggest that RCS exposures observed internationally may not be directly applicable to workers in NSW tunnelling operations. For example, the host rock that is disturbed through tunnelling in Sydney NSW, contains a high crystalline silica content of greater than 90% (Zaid and Gahtani 2015; Cole et al. 2021), higher than the granite–gneiss terrain in Norway (18%) (Forsman et al. 2023), and the London clays in the UK (20% to 70%) (Zhou et al. 2017). High-silica rock has also been identified in regional NSW through geotechnical assessments (Gomes et al. 2021). During the construction of the Hawks Nest tunnel in the United States in the 1930s, the excavated rock contained an estimated crystalline silica content of over 90%. It was one of many factors implicated in the numerous cases of silicosis and related deaths following the construction of that tunnel (Jones 2007). Although tunnelling methods and safety practices have advanced considerably since the 1930s, the inherent potential risk associated with tunnelling through silica-rich geology remains.

Tunnelling activity in Australia has grown significantly over the past decade, in both frequency and scale. Alongside this growth, the size and complexity of tunnel infrastructure have also increased. During this time, Australia has constructed some of its longest railway tunnels and largest underground station caverns, with widths exceeding 30 metres (Oliveira and Peila 2019). These developments may have important implications for RCS exposure because the choice of excavation method, selected in part by the physical dimensions of the tunnels, may influence the effectiveness of dust control measures such as ventilation.

Tunnel construction in Australia employs a range of excavation methods, selected based on factors such as ground conditions, tunnel size and alignment, and other constraints (Chapman et al. 2017). These include tunnel boring machines (TBMs), mined tunnelling, cut-and-cover, and drill and blast techniques (Australian Tunnelling Society 2018). TBMs are large cylindrical machines that excavate through ground material while simultaneously installing precast concrete segments to form a continuous segmental lining or “ring” (Chapman et al. 2017). They are designed to suit specific ground conditions, with variations for hard rock, soft soil, or mixed and variable geologies (Zheng et al. 2016). Mined tunnelling involves sequential excavation using mechanical equipment such as roadheaders, with ground support systems such as sprayed concrete lining ('shotcrete’), rock bolts, and steel sets, which are applied shortly after excavation (Chapman et al. 2017). The drill and blast method, typically used in hard rock environments, involves drilling a pattern of holes into the tunnel face, charging them with explosives, and detonating them in a controlled sequence (Kolymbas 2008). In the cut-and-cover method, a trench, decline, or shaft is excavated to the desired depth, a structural roof or support system is constructed, and the surface is restored through backfilling (Australian Tunnelling Society 2018). Each method presents distinct operational challenges and varying potential for exposure to RCS.

In Australia, regulatory authorities have adopted a workplace exposure standard (WES) for RCS, being a legally enforceable threshold of exposure. The WES was halved from 0.1 to 0.05 mg/m³ as an 8-hour time-weighted average on 1 July 2020. Tunnelling contractors commission RCS monitoring to assess exposure against the WES, and are required to retain the results, inform workers of the results, and reduce exposure when levels are high (Commonwealth Government 2011). While this data exists, it has largely not been accessible for analysis by researchers or policymakers. This study aims to address this gap by accessing and evaluating exposure information to describe the determinants of exposure to RCS among workers involved in tunnel construction projects.

The urgency of this type of analysis is reinforced by recent occupational health issues. The re-identification of coal workers’ pneumoconiosis in 2015 prompted widespread concern and led to parliamentary intervention and legislative reform (Business Queensland 2024). Numerous cases of engineered stone–related silicosis generated similar concern, driving further policy reform (Tefera et al. 2025). These sentinel events reflect a recurring pattern: prolonged, inadequately controlled dust exposure that is recognized only once irreversible disease has occurred. Recent reports of 14 silicosis cases linked to a single tunnelling project, including a worker aged just 32, suggest that this preventable cycle has reappeared within the tunnelling sector (Maddison 2025). Health surveillance, while important, is fundamentally reactive, detecting harm only after exposure has occurred. Effective prevention requires systematic analysis of exposure data and temporal trends to ensure controls are working as intended. Crucially, this is only possible when exposure data are collected and accessible.

Methods

Information sources

This study used historical exposure data from Australian tunnel projects from 3 sources. Firstly, tunnelling contractors had commissioned air monitoring and held reports relating to those air monitoring activities. Some of those reports were in the possession of government agencies, as they sought that information as part of contractual arrangements, audit and assurance processes, or for the purposes of regulatory oversight. Government-held information can be accessed through freedom of information (FOI) processes, subject to public interest provisions. We identified all tunnels constructed in Australia between 2013 and 2023 using industry publications (Australian Tunnelling Society 2022) and online searches, and determined the relevant health and safety regulator for each jurisdiction. FOI applications for RCS data were submitted to those regulators in 2023 under their respective legislative frameworks. Secondly, the Australian Workers Union (AWU), being a union representing tunnel workers, was approached in 2024 to request access to information they had obtained through their own FOI requests to Australian government agencies. Thirdly, an NSW parliamentary order resulted in the public release of government documents concerning RCS in air monitoring data across multiple tunnel projects in NSW constructed from 2017 to 2024 (Parliament of New South Wales 2024); this was also accessed for this study.

Data preparation

The raw data accessed approximated 110,000 pages and 14 spreadsheets. Information included contextual data of work activities along with individual and summary exposure monitoring results in hard copy paper files, electronic PDF documents, and Microsoft Excel spreadsheet formats. Data were reviewed and assessed against the following inclusion criteria, all of which needed to be met for the data to be included: (i) exposure sampling data were collected from a worker on a tunnel construction project, (ii) included the individual results of RCS or respirable dust (RD) exposure of a worker, (iii) were designated as personal exposure samples, meaning that they were collected in the worker's breathing zone. Data with any of the following criteria were excluded: (i) collected from stationary sampling, (ii) related to workers not directly involved in tunnel construction related activities (eg demolition), (iii) results from task-based (short term) samples, (iv) taken during the fit-out or permanent works stage of the tunnel, (v) identified as invalid samples based on the supporting documentation, (vi) where RCS concentrations were greater than corresponding RD levels suggesting errors in sampling or analysis, (vii) where ≥ one exposure result had been redacted from a report and only some were available (in which case the entire dataset from that report was not included), and (viii) only summary statistics were provided rather than individual measurements. Sample results that met the criteria were manually extracted into Microsoft Excel, with each row of data assigned a unique identifier. Data cleaning was performed by the first author and involved standardizing “work group” classifications, as the definitions and categories used across different tunnel projects were similar but not identical. Duplicate records were identified and removed. Information was cross-checked between spreadsheets and reports, and best attempts were made to assign an exposure measurement to the selected variables based on the information available.

Exposure data were supplemented by detailed monitoring reports in most cases, which confirmed that sampling was conducted in accordance with the relevant Australian Standard for measuring RD (SAI Global 2009) with analyses for crystalline silica (as α-quartz and sometimes also cristobalite), undertaken by nationally accredited laboratories using either X-ray diffraction or infrared spectroscopy. A subset of results (16% of the total) was obtained directly from spreadsheets or documents providing results only. For those cases, it was not possible to verify whether the data were collected and analyzed as per recognized standards. Given the nature and source of the data, the use of nonstandard methods was considered unlikely. Sensitivity analyses were conducted to assess the impact of only including data from the results-only sources.

Accompanying contextual information was used to categorize each sample into several variables (Tables 1 and 2). These included categorizing data into work groups based on the qualitative descriptors provided. Above-ground workers were further categorized by the location where they worked, and underground workers were further categorized by the tunnelling method used in the tunnel in which they were working. Only data from tunnel projects that were in the state of NSW were available. The information obtained largely excluded personal identifying information, and any instances where it appeared, it was not used.

Table 1.

Variables considered as part of this study.

Variable	Type	Description
Preliminary statistics
RCS	Outcome, continuous in mg/m3	RCS exposure result
Primary location	Predictor, categorical (nominal, 2 levels)	Defined as being either above-ground or underground
Above-ground location	Predictor, categorical (nominal, 3 levels)	The primary location while the sample was collected from a worker above-ground, being the precast facility, spoil shed, or other general location above-ground
Tunnelling method	Predictor, categorical (nominal, 4 levels)	The primary method employed while the sample was collected from a worker underground, being either cut-and-cover, drill-and-blast, mined tunnelling, or TBM tunnelling
Work groupa	Predictor, categorical (nominal, refer Table 2)	Primary work group (further detailed in Table 2)
Linear mixed model b
RCSln	Outcome, continuous in ln-mg/m3	Log-transformed RCS exposure result
Tunnel project	Random effect, categorical (nominal, 12 levels)	Individual tunnel project
Tunnel project site	Random effect, categorical (nominal, 63 levels)	Individual tunnel site on a tunnel project
Date	Random effect, categorical (nominal, 87 levels)	Date as month and year (mm–yyyy)
Year	Predictor, categorical (nominal, 9 levels)	Year of sampling
Season	Predictor, categorical (nominal, 4 levels)	Defined from the sampling date as Spring (Sep–Nov), Summer (Dec–Feb), Autumn (Mar–May) and Winter (Jun–Aug)
WES change	Predictor, categorical (nominal, 2 levels)	Time of the sampling date relative to the change in the WES, defined as the sample being collected either prior to 30 June 2020, or from 1 July 2020 onwards
Tunnel type	Predictor, categorical (nominal, 3 levels)	The purpose for which the tunnel was constructed, defined as either a rail tunnel, road tunnel, or for the purposes of energy
RCS proportion	Predictor, continuous as a proportion	Concentration of RCS divided by the RD concentration in mg/m3

^aDescribed in Table 2.

^bLMMs also considered the variables listed under Preliminary statistics with the exception of “RCS” (as “RCSln” was used instead).

Table 2.

Work group descriptions and sample size.

Work group	Location/methoda	Generalized description	N
Above-ground workers			1,303
Carousel personnel	Precast	Persons involved in the operational activities associated with manufacture of precast concrete segments on the carousel lines	144
Cleaner	Precast, general site	Persons involved in cleaning site offices, crib rooms, and laundry services	58
Crane operations	Precast, general site	Persons who operate crawler or mobile cranes or perform dogman or rigger duties	55
Form, reo, pour	Precast, general site	Formworkers, steelfixers, and concreters involved in form, reo or pour activities	37
Gantry crane operator	Precast, spoil shed	Operation of an overhead crane for the purpose of lifting heavy loads of materials	78
General labourer	Precast, general site, spoil shed	Persons on foot conducting a range of tasks including spotting, jackhammering, or using hand-held tools	295
Ground support	General site	Persons involved in ground support works including piling, drilling, mesh installation and shotcrete	50
Heavy plant operators (HPO)	Precast, spoil shed, general site	Operators of heavy plant such as loaders, excavators, bobcat etc	237
Management and professional services (MPS)	Precast, general site	Client representatives, engineers, occupational hygienists, laboratory technicians, archaeologists or other consultants	64
Other surface workers	Precast, general site	Other personnel not otherwise classified, including scaffolders and light plant operators, store persons and warehouse operations	19
Processing facility	Precast, general site	Persons who work or operate the batch plant, grout plant, slurry plant or water treatment plants	75
Supervisors and surveyors	Precast, spoil shed, general site	Includes area supervisors, foreman or leading hands as well as persons nominated as a surveyor	62
Trades services	Precast, spoil shed, general site	Trades including electricians, fitters, boilermakers, carpenters, plumbers or those otherwise classified as maintenance workers	59
Traffic control	General site	Persons who direct traffic	70
Underground workers			1,985
Crane operations	Cut-and-cover	Persons who operate crawler or mobile cranes or perform dogman or rigger duties	13
Form, reo, pour	Mined tunnelling, cut-and-cover, TBM tunnelling	Formworkers, steelfixers, and concreters involved in form, reo, pour activities	123
Ground support	Cut-and-cover, drill-and-blast, mined tunnelling	Persons involved in ground support works including piling, drilling, mesh installation and shotcrete	238
HPO	Cut-and-cover, drill-and-blast, mined tunnelling, TBM tunnelling	Operators of heavy plant such as articulated haul trucks, excavators, dozers, etc.	321
MPS	Precast, tunnel site(s), workshop	Client representatives, engineers, occupational hygienists, geotechnical or other consultants	37
Other mined method	Mined tunnelling	Operators of other mined-tunnelling equipment including Brokk excavators, profilers or surface miners, being medium or large size plant for the excavation of hard surfaces	32
Roadheader operators	Mined tunnelling	Operators of roadheaders. Further categorized into those with or without enclosed cabins	154
Supervisors and surveyors	Cut-and-cover, drill-and-blast, mined tunnelling, TBM tunnelling	Includes area supervisors, foreman or leading hands as well as persons nominated as a surveyor	188
TBM operators	TBM tunnelling	Persons who are primarily stationed on, and operate, the TBM from within a cabin	44
TBM ring builders	TBM tunnelling	Persons who are primarily stationed on, and work at the front of the TBM building the segmental lining (‘rings’)	45
TBM backend	TBM tunnelling	Persons who are primarily stationed on, and work on the TBM including those who extend services and conveyors, handle segments and perform grouting activities	66
Trades services	Cut-and-cover, mined tunnelling, TBM tunnelling	Trades including electricians, fitters, boilermakers, carpenters, or otherwise classified as maintenance workers	225
Tunnellers	Drill-and-blast, mined tunnelling, TBM tunnelling	Persons on foot in tunnels under construction conducting a range of tasks	438
Waterproofing	Cut-and-cover, mined tunnelling	Personnel involved in activities associated with waterproofing	61

^aLocation is used for above-ground workers while tunnel method is used for underground workers.

Preliminary analysis

Data were grouped by work group and summarized using the number of samples (N), minimum (Min) and maximum (Max) concentrations, and percentage of censored results (%Cen.), being those reported below the analytical laboratory's limit of detection. Censored data were substituted using NDExpo which uses robust regression on order statistics for the substitution. Data were approximately lognormal and were described by the geometric standard deviation (GSD), estimated arithmetic mean (AM), and associated with 95% upper confidence limit (UCL) and lower confidence limit (LCL), which were calculated using the IHStat option in the online software application BWstats (https://bwstat.bsoh.be/). Stratification by sub-group was performed only when at least 3 individual results were obtained across a minimum of 3 separate monitoring events. The same approach was applied to calculate the corresponding parameters for RD measurements. The percentage of RCS in RD samples was estimated by directly comparing the AM concentrations of RCS divided by the AM concentration of RD. The AM of RCS was plotted for each calendar year across all tunnel projects for above-ground and underground workers.

Linear mixed effects modelling

To understand the influence of various factors on observed RCS exposure, linear mixed models (LMMs) were constructed. LMMs were selected because the data involved repeated measurements with clustering by tunnel project and site, introducing dependency structures. LMMs suit data sets with unequal group sizes, which aligned with the characteristics of our dataset. All analyses were conducted using natural log-transformed values, with RCSln as the outcome variable. Exploratory data analysis was performed through box plots, scatter plots, and contingency tables to describe individual variables and assess crude relationships between the outcome RCSln and the predictors listed in Table 1. Scatterplots were generated to assess linear associations between the outcome and predictors and to check the correlation between predictors. Visual inspection of the data distributions confirmed that RCS concentrations followed a log-normal distribution. Censored data were substituted using NDExpo prior to transformation.

For the LMMs, the Tunnel Project Site was initially nested in Tunnel Project as random effects to account for shared influences on exposure levels. However, the nested structure did not significantly improve model fit, indicating no additional variance was explained beyond the Tunnel Project Site alone. The most parsimonious model, therefore, retained only Tunnel Project Site as the random effect. Inclusion of Date as a random effect nested within Tunnel Project Site improved model fit based on the likelihood ratio test and was included. Sampling Year was also included as a crossed random effect to account for year-to-year variation. An identity covariance structure was specified for the residuals, assuming no additional within-site correlation beyond that captured by the random effects. Potential predictors were evaluated individually using univariable LMMs that included the random effects structure. Variables with a P-value of <0.20 were retained for consideration in multivariable models.

The data had a hierarchical structure, with exposure measurements nested within project sites. As Tunnel Type was constant within project sites, it was not separately identifiable from the site random effect and was not included as a fixed effect (see Table S2). Additionally, limited overlap between certain work groups and tunnel methods (eg, TBM operators are exclusive to one tunnel method) hindered the model's ability to isolate the independent effects of each work group (see Table S3). In LMMs, reliable estimation of random effects requires adequate variability and representation within each grouping level. To maintain model robustness and interpretability, models were only constructed to the level of Above-ground location (for above-ground workers) or Tunnelling method (for underground workers). Data from drill-and-blast tunnelling were limited, comprising less than 1% of the total dataset, and were therefore excluded.

A forward stepwise approach was used to construct multiple LMMs to identify predictors positively affecting model fit and stability. For each variable, improvement in model fit was assessed using the change in Akaike's information criteria (AIC) and marginal R². The first model (LMM-1) comprised Primary location (above-ground vs. underground), WES Change (and their interaction), and RCS proportion as fixed effects. Separate above-ground (LMM-2) and underground (LMM-3) models substituted Primary location with Above-ground location or Tunnelling method. Season did not improve model fit and was not retained. The interaction with WES change was retained only in LMM-1: in LMM-3, it did not improve model fit, and in LMM-2 it did, but was excluded as the before-WES Precast data came from a single project site. Collinearity and variance components were monitored throughout, and the intraclass correlation coefficient (ICC) was reported to estimate the proportion of between-site and between-date variation. Models were fitted using restricted maximum likelihood (REML), and Bonferroni-adjusted pairwise comparisons were performed to test differences between groups, with GMs presented as the exponentiated estimated marginal means (EMMs). Consistent with previous studies (Leite et al. 2023), the AM was derived from the LMM by back-transforming the fixed-effect estimates while accounting for both within- and between-group variability. Specifically, the AM was calculated as the exponent of the sum of the EMM and 0.5 × (σ²_e + σ²_β) from each model, where σ²_e represents the residual variance and σ²_β represents the random effect variance. The 95% confidence intervals (CIs) of the AMs were derived by back-transforming the CIs of the EMMs, with the variance term added. As a sensitivity analysis, CIs were also derived via parametric bootstrap (5,000 iterations).

Fixed effects tables were produced to report coefficients adjusted for both within- and between-group variation, to evaluate the influence of predictors on RCS exposure. β-coefficients are presented with 95% CI. Residual diagnostics included visual inspection of a histogram and Q–Q plot to assess normality, and a scatterplot of standardized residuals versus predicted values for homoscedasticity. Statistical significance for multivariable models was defined as P < 0.05. All analyses were run in Jamovi (v2.7.6.0), with the exception of the sensitivity analysis of the CIs of the AM, which was performed in R (version 2026.04.0 + 526) using the lme4 package, as this calculation is not available in Jamovi.

The STROBE statement for reporting observational studies was followed (von Elm et al. 2007) (see Table S4). Information on respiratory protective equipment (RPE) use was not available for all measurements across all tunnel projects. In all cases, the data are presented without correction for the use of RPE, if it was worn.

Results

The final dataset included 3,288 individual personal exposure results for RCS from persons who worked across 12 tunnel projects from 2016 to 2024. Most data originated from projects constructed in Sydney (10 projects, making up 94% of the data), with 2 projects from regional NSW. Air monitoring activities were observed to occur on an approximate monthly basis across each of the tunnelling sites. Most measurements were collected during the construction of rail tunnels (78%), with the remainder for roads (16%), or energy (6%). Around one-third of all results (34%) were censored, most of which were from above-ground workers. Sixty percent of measurements were collected from underground workers. Eleven projects involved mined tunnelling, 6 projects used TBMs, 6 used cut-and-cover construction, and 3 used drill-and-blast methods. Complete datasets were not available from all projects.

Preliminary analysis

Preliminary analysis showed that AM RCS exposures were higher to workers underground (0.10 mg/m³; CI 0.09 to 0.11) than they were above-ground (0.02 mg/m³; CI 0.02 to 0.02). The highest exposures were recorded during mined tunnelling, specifically to roadheader operators working without enclosed cabins (4.4 mg/m³; CI 2.7 to 14). The proportion of RCS in RD was higher underground compared to above-ground and varied by tunnelling method and location. Higher RCS concentrations generally coincided with higher proportions of RCS in RD. RCS exposures showed a decreasing trend from 2017 to 2022 and were relatively constant thereafter; they did not follow a linear pattern (Fig. 1). The reduced data availability observed in 2021 to 2022 probably reflects multiple factors, including COVID-19-related disruptions (eg, decreased sampling and regulator data requests) and an associated decline in tunnel construction activity. The results from preliminary analysis are provided in Table 3, with additional data and RD results provided in Tables S5 and S6.

Figure 1.

RCS exposures per year (AM) for all tunnel projects in NSW showing the 95% UCL & LCL for a) above-ground, and b) underground workers. Sample size is shown at each year.

Table 3.

Preliminary analysis of RCS exposures (mg/m³).

Location and work group	N	%Cen.	Min–max	AM (LCL-UCL)	GSD	%RCS
Above-ground workers	1,303	61%	<0.01–2.12	0.019 (0.019–0.020)	2.2	16%
General site	854	60%	<0.01–2.12	0.017 (0.017–0.018)	2.0	17%
Cleaner	55	45%	<0.01–0.06	0.018 (0.016–0.021)	1.8	17%
Crane operations	53	64%	<0.01–0.17	0.014 (0.012–0.017)	1.9	17%
Form, reo, pour	28	57%	<0.01–0.04	0.014 (0.012–0.016)	1.6	12%
General labourer	220	46%	<0.01–2.12	0.024 (0.022–0.028)	2.4	16%
Ground support	50	52%	<0.01–1.68	0.023 (0.018–0.031)	2.6	17%
HPO	145	69%	<0.01–0.19	0.013 (0.012–0.014)	1.6	20%
MPS	51	80%	<0.01–0.07	0.013 (0.012–0.015)	1.6	22%
Other surface worker	19	47%	<0.01–0.12	0.024 (0.017–0.039)	2.3	25%
Processing facilities	75	72%	<0.01–0.16	0.017 (0.015–0.020)	2.1	18%
Supervisors and surveyors	51	63%	<0.01–0.16	0.016 (0.014–0.019)	1.9	21%
Trades services	37	62%	<0.01–1.00	0.018 (0.014–0.025)	2.4	13%
Traffic control	70	77%	<0.01–0.06	0.013 (0.012–0.014)	1.6	25%
Precast	273	86%	<0.01–0.35	0.012 (0.011–0.013)	1.7	9%
Carousel personnel	144	85%	<0.01–0.35	0.013 (0.012–0.014)	1.8	8%
Cleaner	3	100%	<0.01–0.01	0.006 (0.004–0.014)	1.0	15%
Form, reo, pour	9	67%	<0.01–0.01	0.010 (0.009–0.011)	1.2	16%
Gantry crane operator	10	100%	<0.01–0.01	0.010 (0.010–0.010)	1.1	25%
General labourer	32	91%	<0.01–0.05	0.011 (0.010–0.012)	1.4	11%
HPO	32	81%	<0.01–0.17	0.014 (0.012–0.018)	1.8	14%
MPS	13	100%	<0.01–0.01	0.009 (0.008–0.011)	1.3	16%
Supervisors	9	89%	<0.01–0.02	0.010 (0.008–0.012)	1.4	23%
Trades services	19	89%	<0.01–0.01	0.010 (0.009–0.011)	1.2	7%
Spoil shed	176	22%	<0.01–0.89	0.049 (0.042–0.060)	3.0	27%
Gantry crane operator	68	22%	<0.01–0.89	0.072 (0.056–0.104)	3.2	33%
General labourer	43	2%	<0.01–0.27	0.063 (0.049–0.089)	2.6	29%
HPO	60	35%	<0.01–0.20	0.022 (0.018–0.027)	2.3	21%
Trades services	3	67%	<0.01–0.01	0.010 (0.009–0.011)	1.1	4%
Underground workers	1,985	16%	<0.01–10.4	0.100 (0.094–0.107)	3.6	27%
Cut-and-cover	611	32%	<0.01–0.94	0.049 (0.045–0.054)	3.1	21%
Crane operations	13	31%	<0.01–0.09	0.021 (0.015–0.039)	2.2	14%
Form, reo, pour	51	37%	<0.01–0.15	0.032 (0.025–0.045)	2.7	16%
Ground support	108	31%	<0.01–0.78	0.043 (0.035–0.054)	2.9	11%
HPO	139	28%	<0.01–0.60	0.057 (0.047–0.073)	3.3	34%
MPS	10	60%	<0.01–0.09	0.020 (0.014–0.036)	2.0	16%
Supervisors and  surveyors	82	38%	<0.01–0.57	0.052 (0.040–0.073)	3.4	29%
Trades services	31	45%	<0.01–0.13	0.020 (0.016–0.026)	2.0	13%
Tunneller	167	25%	<0.01–0.94	0.066 (0.055–0.083)	3.4	26%
Waterproofing	10	60%	<0.01–0.06	0.018 (0.013–0.033)	2.0	12%
Drill and blast	25	24%	<0.01–0.14	0.030 (0.022–0.049)	2.5	14%
Ground support	5	20%	<0.01–0.09	0.028 (0.016–0.223)	5.0	14%
HPO	12	25%	<0.01–0.14	0.031 (0.020–0.083)	2.8	15%
Tunneller	7	29%	<0.01–0.11	0.028 (0.017–0.099)	2.4	16%
Mined tunnelling	1,066	7%	<0.01–10.4	0.146 (0.134–0.160)	3.7	32%
Form, reo, pour	69	7%	<0.01–0.26	0.071 (0.058–0.091)	2.5	16%
Ground support	125	8%	<0.01–1.19	0.071 (0.060–0.089)	2.9	19%
HPO	144	6%	<0.01–8.82	0.158 (0.130–0.205)	3.4	37%
MPS	18	61%	<0.01–0.21	0.025 (0.017–0.047)	2.6	21%
Other mined method	32	9%	<0.01–2.05	0.352 (0.233–0.721)	3.8	56%
Roadheader (closed  cabin)	147	2%	<0.01–9.22	0.228 (0.187–0.295)	3.5	44%
Roadheader (open cabin)	8	0%	0.65–10.43	4.45 (2.74–13.7)	2.4	82%
Supervisors and surveyors	80	10%	<0.01–1.89	0.119 (0.088–0.185)	4.1	36%
Trades services	151	7%	<0.01–3.49	0.098 (0.081–0.126)	3.4	28%
Tunneller	241	5%	<0.01–1.79	0.144 (0.123–0.175)	3.4	30%
Waterproofing	51	2%	<0.01–0.41	0.088 (0.069–0.125)	2.8	17%
TBM tunnelling	283	15%	<0.01–1.19	0.069 (0.061–0.081)	3.1	22%
HPO	25	8%	<0.01–0.13	0.041 (0.032–0.061)	2.3	17%
MPS	9	33%	<0.01–0.18	0.040 (0.022–0.187)	3.2	11%
Supervisors and surveyors	25	8%	<0.01–0.15	0.048 (0.036–0.075)	2.4	20%
TBM back end	66	11%	<0.01–0.29	0.085 (0.068–0.114)	2.7	25%
TBM operator	44	20%	<0.01–0.21	0.033 (0.026–0.046)	2.6	19%
TBM ring builder	45	18%	<0.01–0.46	0.091 (0.064–0.155)	3.6	31%
Trades services	43	23%	<0.01–0.23	0.049 (0.037–0.074)	2.9	22%
Tunneller	23	4%	<0.01–1.19	0.207 (0.130–0.495)	3.7	24%

Linear mixed models

Three LMMs were constructed: LMM-1 using data from all locations, LMM-2 from above-ground only, and LMM-3 from underground only (Table 4). The variables associated with RCS exposure were the RCS proportion; the primary location (above-ground vs underground) in LMM-1; the above-ground location or underground tunnelling method in LMM-2 and LMM-3; and whether the measurement was collected before or after the change in WES in all LMMs.

Table 4.

LMM estimates for predicting RCS concentrations (mg/m³).

Model and parameter	Exp (β)a 95% CI	P-value
LMM-1: All locations. Outcome variable: RCSln. Fixed effects: RCS proportion, primary location, WES Change, and a primary location × WES Change interaction. Date nested within Tunnel Project Site, and Year as a crossed random effect.
Intercept	0.02 (0.01–0.02)	<0.001
RCS proportionb	105 (84–132)	<0.001
WES Change	0.57 (0.42–0.78)	<0.001
Primary location
Aboveground	Ref.
Underground	2.27 (2.06–2.49)	<0.001
WES change × primary location	1.28 (1.06–1.55)	0.009
LMM-2: Above-ground. Outcome variable: RCSln. Fixed effects: RCS proportion, above-ground location, WES Change. Date nested within Tunnel Project Site, and Year as a crossed random effect.
Intercept	0.01 (<0.01–0.01)	<0.001
RCS proportionb	116 (73–183)	<0.001
WES change	0.51 (0.33–0.80)	0.009
Aboveground location
General site	Ref.
Precast	0.81 (0.62–1.06)	0.127
Spoil shed	1.27 (1.02–1.58)	0.034
LMM-3: Underground. Outcome variable: RCSln. Fixed effects: RCS proportion, tunnelling method, WES Change. Date nested within Tunnel Project Site, and Year as a crossed random effect.
Intercept	0.03 (0.03–0.04)	<0.001
RCS proportionb	98 (76–127)	<0.001
WES change	0.63 (0.49–0.81)	0.001
Tunnelling method
Cut-and-cover	Ref.
Mined tunnelling	1.94 (1.69–2.23)	<0.001
TBM tunnelling	1.57 (1.29–1.90)	<0.001

^aExp (β) = inverse ln transformation of β-coefficients for the exposure concentrations.

^bThe exponentiated coefficient represents the multiplicative change in RCS exposure for a full-unit increase in RCS/RD (ie, from 0 to 1).

Underground workers were predicted to have 2.3 times the RCS exposure (CI 2.1 to 2.5) compared to those working above-ground; RCS exposure after July 2020 was estimated to be approximately 43% lower (CI 22 to 58%) than prior to the change in WES. For each 1% increase in the proportion of RCS to RD (ie, a 0.01 increase in the predictor variable), each of the 3 models predicted an approximate 5% increase (CI 4% to 5%) in RCS exposure (mg/m³) (Fig. S1).

Above-ground location was associated with exposure. Relative to general site locations, exposure at precast facilities was estimated to be 19% lower (CI 38% lower to 6% higher), while exposure in spoil sheds was estimated to be 27% higher (CI 2% to 58%). Only the spoil shed vs precast comparison was significant. Underground, the tunnelling method was associated with exposure. Relative to cut-and-cover work, exposure was an estimated 94% higher for mined tunnelling (CI 69% to 123%) and 57% higher for TBM tunnelling (CI 29% to 90%). Only the mined vs cut-and-cover and TBM vs. cut-and-cover comparisons were significant (Table 4).

The LMMs explained a substantial proportion of the variability in RCS exposure within each model with conditional R² values of 68%, 53% and 67% for LMM-1, LMM-2, and LMM-3, respectively (Table S7). The conditional R², which incorporates both fixed and random effects estimated from the same data, should be interpreted with caution as it may overstate explained variance, although moderate values observed suggest overfitting is unlikely to be substantial. The contribution of random effects (Tunnel Project Site, Date nested within site, and Year) varied by work location. These random effects accounted for approximately 29% of the total variance in LMM-1; 21% in LMM-2, and 32% in LMM-3. Estimated GM and AM concentrations from the model are presented in Table 5. Confidence intervals for the AM derived via parametric bootstrap were comparable to those obtained by back-transforming the CIs of the EMMs, with differences of ≤0.01 mg/m³ across all estimands (Table S8). Sensitivity analyses, only including data extracted from reports where standard methods could be verified, provided similar results (Table S9).

Table 5.

RCS exposure estimates as (mg/m³) back-transformed from EMM from the LMMs, with P-values from post-hoc Bonferroni tests.

Model and predictor	GM (95% CI)	AM (95% CI)	P-value
LMM-1: All locations
Overall predicted meana	0.017 (0.013–0.023)	0.036 (0.027–0.048)	N/A
WES Change
Prior to 30 June 2020	0.023 (0.016–0.032)	0.047 (0.034–0.066)	Ref.
Post 1 July 2020	0.013 (0.009–0.018)	0.027 (0.020–0.037)	<0.001
Primary location
Above-ground	0.012 (0.009–0.014)	0.024 (0.018–0.032)	Ref.
Underground	0.026 (0.021–0.032)	0.054 (0.040–0.072)	<0.001
Primary location * WES change
Above-ground
Prior to 30 June 2020	0.016 (0.011–0.023)	0.033 (0.023–0.047)	Ref.
Post 1 July 2020	0.008 (0.006–0.011)	0.017 (0.012–0.023)	<0.001
Underground
Prior to 30 June 2020	0.032 (0.023–0.045)	0.067 (0.048–0.093)	Ref.
Post 1 July 2020	0.021 (0.015–0.029)	0.043 (0.031–0.059)	0.043
LMM-2: Above-ground workers
Overall predicted meana	0.008 (0.006–0.010)	0.016 (0.011–0.021)	N/A
WES Change
Prior to 30 June 2020	0.011 (0.007–0.016)	0.022 (0.014–0.033)	Ref.
Post 1 July 2020	0.005 (0.004–0.008)	0.011 (0.008–0.016)	0.009
Above-ground locationb
General site	0.008 (0.005–0.010)	0.015 (0.011–0.021)	Ref.
Precast	0.006 (0.004–0.009)	0.012 (0.009–0.018)	0.380
Spoil shed	0.010 (0.007–0.013)	0.020 (0.014–0.027)	0.102
LMM-3: Underground workers
Overall predicted meana	0.034 (0.029–0.041)	0.062 (0.052–0.074)	N/A
WES change
Prior to 30 June 2020	0.043 (0.034–0.054)	0.078 (0.062–0.098)	Ref.
Post 1 July 2020	0.027 (0.022–0.034)	0.049 (0.040–0.061)	0.001
Tunnelling methodc
Cut-and-cover	0.024 (0.020–0.029)	0.043 (0.036–0.052)	Ref.
Mined tunnelling	0.046 (0.038–0.055)	0.083 (0.069–0.099)	<0.001
TBM tunnelling	0.037 (0.030–0.046)	0.067 (0.054–0.084)	<0.001

^aOverall predicted mean, with the RCS proportion held at its mean value.

^b P-value when comparing spoil shed with Precast was 0.028.

^c P-value when comparing TBM tunnelling with mined tunnelling was 0.060.

Discussion

This observational study found that determinants of RCS exposure included a worker's primary location (above-ground vs. underground), proportion of RCS in RD, and the tunnelling method or specific above-ground location. The highest exposed were mined tunnelling workers underground, while the lowest were those in precast facilities above-ground. Estimated RCS exposure was lower after the reduction of the WES in July 2020. This difference may reflect several coinciding factors, such as the introduction of new control measures, shifts in work practices, or broader industry developments.

Our findings support previous industry reports (Australian Tunnelling Society 2018) that RCS exposure underground varies by tunnelling method. In mined tunnelling, the highest RCS exposures were observed among roadheader operators; however, workers on foot not directly involved in RCS-generating activities such as those involved with form, reo, pour activities; supervision and surveying; and waterproofing also had elevated exposures. This suggests that source control measures were insufficient to prevent ambient RCS exposure throughout the tunnel environment. Australian data from 2001 suggested typical levels of RCS ranging from 0.5 to 10 mg/m³ for roadheaders, 0.2 to 1.5 mg/m³ for truck drivers, and 0.1 to 0.8 mg/m³ in the outbye tunnel sections (Henderson 2001). Our findings show lower levels, suggesting some improvement over the past 2 decades. A Queensland regulatory intervention in 2010 reported “excessive” RCS concentrations exceeding 1 mg/m³ for roadheader operators (Workplace Health and Safety Queensland 2010). Our findings suggest that RCS exposures have not markedly improved since then. While operators in enclosed cabins experienced lower exposures, higher levels persisted among roadheader operators without cabins and HPOs. In TBM environments, tunnellers experienced the highest exposures. These workers performed various tasks, including using hand-held tools, installing invert lining, and patching. On the TBM, those who constructed segmental linings (“rings”) near the cutter head had the highest exposures, with lower exposures reported at the rear or “back end” of the TBM, and TBM operators with the lowest exposures. This is consistent with Norwegian data (Leite et al. 2023), which similarly reported the highest exposures among “shield workers” (equivalent to ring builders) and lower levels among TBM operators and pipe workers. Other international studies have reported higher exposures (Bakke et al. 2001a; Bakke et al. 2002; Inyang 2024), while more recent smaller-scale investigations of TBM projects in Sydney (Cole 2016b; Cole et al. 2021) have reported broadly comparable or higher results. The lowest exposures underground were observed during cut-and-cover tunnelling. Within this group, tunnellers, HPO, supervisors and surveyors, reported the highest exposures. Whilst high, these levels were lower than those previously reported for labourers in cut-and-cover environments internationally (Blute et al. 1999). Above-ground, the highest exposures were observed among workers within spoil sheds where excavated material is transferred for off-site disposal. The findings demonstrate that high RCS exposure is not constrained to underground environments, highlighting the need for continued focus on effective dust control strategies above-ground.

Multiple sources of crystalline silica are present in the tunnelling industry, including high-silica sources such as the host rock, as well as lower yet still important contributors such as concrete products used for ground support. The proportion of RCS in RD varied across locations, tunnelling methods, and work groups, reflecting differences in the relative contribution of these sources to overall exposure. This was illustrated in the lower RCS proportions observed above-ground (16%), particularly in the precast yard (9%), compared to underground environments (27%), where dust generation is more closely associated with excavation of the host rock and higher proportions were observed (eg 82% for roadheader operators). Despite this variability, the LMMs demonstrated a consistent positive relationship between RCS proportion and RCSln in both above-ground and underground settings. The results suggest that RCS exposures are influenced by the relative contribution of multiple dust sources that vary by work location, task, and construction method. Notably, the RCS proportions reported in this study are higher than those documented in other Australian and international settings (Table S1), probably reflecting the comparatively higher silica content of host rock in NSW.

Tunnelling is often compared to mining, but a comparison with the coal industry in NSW suggests that tunnelling involves a higher risk of exposure. Between 2019 and 2024, the average RCS exposures of NSW coal miners were ≤0.02 mg/m³ for high-dust tasks such as cement spraying and continuous mining (Coal Services 2024, 2025). In contrast, tunnelling ground support workers experienced average exposures ranging from 0.03 to 0.07 mg/m³, while values for roadheader operations were between 0.23 and 4.4 mg/m³. Part of this difference between industries may be explained by the high proportion of crystalline silica present during tunnelling in NSW. Our study found that the proportion of RCS in NSW exceeded those reported elsewhere, including in cut-and-cover tunnelling in the USA (Blute et al. 1999), drill-and-blast (Bakke et al. 2001b; Bakke et al. 2002), and TBM tunnelling in Norway (Bakke et al. 2002; Leite et al. 2023), during the application of sprayed concrete lining in the UK (Galea 2015; Galea et al. 2015), and during mined tunnelling in Queensland (Workplace Health and Safety Queensland 2010). As one example, the proportion of RCS measured in RD in NSW during roadheader use was 2 to 3 times higher than in Queensland. This indicates that RCS exposure concentrations in other regions probably underestimate the true exposure risk in NSW tunnelling environments.

Determinants of exposure in tunnel construction have previously been shown to include job group, job site, certain tasks, the presence of operator cabins, tunnel breakthrough (Bakke et al. 2002), work location, type of work (production, maintenance etc), and date of sampling (Leite et al. 2023). Although the observational nature of our study limited our ability to assess all these factors, the full model including all variables explained a substantial proportion of variability in exposure levels (R² = 54% to 68%).

In Australia, air monitoring results are evaluated in relation to the WES under health and safety legislation; however, no specific method for determining compliance is prescribed in this industry. Air monitoring reports indicated that occupational hygienists commonly applied the minimum variance unbiased estimator as the AM with a 95% UCL (Jahn et al. 2015), aligning with local guidance favouring the AM for assessing exposures related to chronic health effects when the WES is expressed as a time-weighted average (Australian Institute of Occupational Hygienists 2014). More recent work by occupational hygienists adopted EN689 procedures (European Committee for Standardization 2018). Internationally, metrics vary, with studies presenting GM, AM, or both. Caution is advised against using GM-based metrics to assess compliance, as reliance on the GM can underestimate inhaled dose (Seixas et al. 1988; Crump 1998; Rando et al. 2018). Geometric and AMs are presented for comparability with other studies, but AMs are the preferred basis for characterizing exposure.

Consistent information on RPE use was not available for all tunnel projects, and is therefore not discussed within this paper. However, a separate analysis of 6 of the 12 tunnel projects included in this study found that, on average, in 17% of cases where RCS exposure exceeded the WES, RPE was considered ineffective by occupational hygienists in adequately reducing exposure (Cole et al. 2026). Accordingly, it would be inappropriate to assume consistent or effective RPE use across all exposures.

While descriptive statistics are presented in this paper, aggregating observations from multiple projects into broad work categories introduces important limitations. Grouping data on individuals that are hierarchically structured ignores higher-level contexts. When nested structures exist, the assumption of independence is violated, making traditional single-level statistical techniques (eg, linear regression) inappropriate. Consequently, applying single-level analyses to nested data increases the risk of inaccurate results and erroneous conclusions (Bash et al. 2021). For these reasons, LMMs are preferred as they account for the multilevel nature of the data through the inclusion of random effects, improving the ability to distinguish within- and between-group variability. Nevertheless, the application of LMMs in this study was constrained by characteristics of the available data. Structural confounding or collinearity among predictors, as well as small sample sizes at one or more hierarchical levels, limited our ability to disentangle effects and may have reduced model stability. While LMMs offer a more rigorous approach for modelling hierarchical exposure data, their successful application depends on having sufficient data structure to support the model. In this study, limitations in the available data precluded reliable model fitting at the work group level.

Another limitation of this study is that workers often performed multiple tasks within a shift, and although substantial effort was made to assign exposure measurements to specific activities, workers were not always exclusively engaged in those tasks. For example, HPOs sometimes worked outside enclosed cabins, potentially being exposed beyond what cabin operation alone would suggest. Most monitoring occurred during day shifts, and the representativeness of these results for night-shift conditions is unclear. Exposure monitoring covered selected project periods, ranging from brief snapshots to extended construction phases. We note the variation within work groups as indicated by the presence of some wide GSDs from the preliminary analysis. The GSDs for RD samples were different than those for RCS samples, suggesting that variability was not solely attributable to work group definition, but may have also related to variability in RCS proportion. Due to privacy constraints, individual worker names were not accessible, so it was not possible to assess within-worker variability in exposure levels. Although efforts were made to align exposure measurements with the variables analyzed in this study, some reports provided only basic information regarding the specific tasks undertaken by the workers, which limited the precision of work-group-based exposure categorization. While every effort was made to transcribe data accurately, minor errors cannot be excluded.

Notwithstanding these limitations, this study highlights the importance of accessible RCS exposure data in tunnel construction for identifying exposure determinants and informing prevention, emphasizing that effective analysis and control evaluation rely on consistent data collection and availability.

Conclusion

Tunnelling is recognized as a high-risk industry because of the potential for high exposure to RCS and the resultant development of silicosis and lung cancer, yet little research has examined RCS exposure among TCWs, and reliable public data remain scarce. This study provides a comprehensive evaluation of RCS exposure across tunnel construction projects in Australia between 2016 and 2024. Results show significantly higher exposures among underground workers, particularly during mined tunnelling, where the absence of enclosed cabins was associated with the highest levels. Elevated exposures were also observed during TBM and cut-and-cover operations. Above-ground exposures were lower overall but remained notable. Despite tunnelling being recognized as a high-risk industry due to RCS exposure, research and publicly accessible data on RCS exposures among TCWs has been limited. Our findings support the need for targeted control measures and transparent access to data across tunnel construction environments.

Supplementary material

Supplementary material is available at Annals of Work Exposures and Health online.

Funding

Research being undertaken by the primary author is supported by the University of Sydney Postgraduate Award.

Conflicts of interests

KC and TD serve on the NSW Tunnelling Dust Safety Taskforce in a volunteer capacity. The primary author (KC) is a member of the NSW Dust Diseases Board; the Asbestos and Silica Safety and Eradication Council; has acted as an expert witness for law firms in litigation concerning workers in tunnelling; and may in future serve as an expert witness in court proceedings related to silica-related diseases in tunnel workers. TD has prepared reports for Government, and as an expert witness, in relation to occupational exposure to silica.

Data availability

The data underlying this article will be shared for research purposes on reasonable request to the corresponding author.