Effectiveness of a Multidimensional Randomized Control Intervention to Reduce Quartz Exposure Among Construction Workers

ABSTRACT

There is little evidence with respect to the effectiveness of intervention programs that focus on the reduction of occupational quartz exposure in the construction industry. This article evaluates the effectiveness of a multidimensional intervention which was aimed at reducing occupational quartz exposure among construction workers by increasing the use of technical control measures. Eight companies participating in the cluster randomized controlled trial were randomly allocated to the intervention (four companies) or control condition (four companies). The multidimensional intervention included engineering, organizational, and behavioural elements at both organizational and individual level. Full-shift personal quartz exposure measurements and detailed observations were conducted before and after the intervention among bricklayers, carpenters, concrete drillers, demolishers, and tuck pointers (n = 282). About 59% of these workers measured at baseline were reassessed during follow-up. Bayesian hierarchical models were used to evaluate the intervention effect on exposure levels. Concrete drillers in the intervention group used technical control measures, particularly water suppression, for a significantly greater proportion of the time spent on abrasive tasks during follow-up compared to baseline (93 versus 62%; P < 0.05). A similar effect, although not statistically significant, was observed among demolishers. A substantial overall reduction in quartz exposure (73 versus 40% in the intervention and control group respectively; P < 0.001) was observed for concrete drillers, demolishers, and tuck pointers. The decrease in exposure in the intervention group compared to controls was significantly larger for demolishers and tuck pointers, but not for concrete drillers. The observed effect could at least partly be explained by the introduced interventions; the statistically significant increased use of control measures among concrete drillers explains the observed effect to some extent in this job category only. Sensitivity analyses indicated that the observed decrease in exposure may also partly be attributable to changes in work location and abrasiveness of the tasks performed. Despite the difficulties in assessing the exact magnitude of the intervention, this study showed that the structured intervention approach at least partly contributed to a substantial reduction in quartz exposure among high exposed construction workers.

METHODS

Study design

A detailed description of the study design and the methods have been described elsewhere (Oude Hengel et al., 2014). The effectiveness of the intervention was assessed in a cluster randomized controlled trial (cluster RCT). Companies rather than individuals were randomized since the intervention components were mostly administered at the organizational level (i.e. company) rather than at the individual level. Moreover, randomization at the organizational level minimized the risk of intervention group contamination (Christie et al., 2009).

Randomization, blinding and sample size

Cluster randomization took place at the company level after the baseline survey. All eight companies were randomly assigned to either an intervention (n = 4) or control condition (i.e. no intervention; n = 4) using an electronic randomization tool (www.randomizer.org.). Construction workers, managers, and the research team could not be blinded to the allocation. Before the intervention took place, sample size calculations were performed assuming an 30% reduction in exposure, based on a comparable study in the wood processing industry (Lazovich et al., 2002). We assumed an alpha of 5% and a power of 80%, as well as a long-term downward trends of 3% annually for two years in both the control and intervention group (Kromhout and Vermeulen, 2000) and a loss-to-follow-up of 20%. Based on these calculations, it was estimated that 60 construction workers for both the intervention and control group were required at baseline and during follow-up, resulting in a group of 120 workers. Since we aimed to conduct repeated measurements among 25% of the workers, we aimed to collect 150 personal samples in 120 workers during both baseline and follow-up.

Study population

Details about the study population are described elsewhere (van Deurssen et al., 2014). The following job categories were included: bricklayer, carpenter, concrete driller, demolisher, and tuck pointer.

Companies were recruited through sector organizations. All construction workers within the participating companies who were permanently employed at the start (November 2011) and who had sufficient Dutch language skills, were eligible to participate. Because of the large number of workers eligible to participate, a random sample of these eligible workers per company was included in the baseline measurements (i.e. pre-intervention) (van Deurssen et al., 2014). After the baseline measurement, participating companies were randomly allocated to an intervention or control group.

Follow-up measurements (i.e. post-intervention) were aimed at reassessing exposure in individuals included in the baseline random sample. However, some workers could not be included again during follow-up for practical reasons: they were working at inaccessible worksites or they were unemployed at the time of the follow-up measurements. These workers were replaced by other workers within the company with similar job titles and performing similar tasks in order to obtain an equal number of workers and measurements as in the baseline survey. All participating construction workers signed a written informed consent. The study is not part of the judgement of the Central Committee of Research Involving Human Subjects, meaning that no medical ethic approval was required for this study. The study has been executed according to the Dutch Data Protection Law.

Intervention

More details on the development and content of the intervention have been described elsewhere (Oude Hengel et al., 2014). In short, the 6-month intervention program was developed according to the Intervention Mapping protocol (Bartholomew et al., 2006), and consisted of engineering, organizational, and behavioural elements at both organizational (managers) and individual (construction workers) level. The intervention consisted of two plenary sessions and accompanying intervention materials. All permanent employees from a company were invited for the plenary sessions. The first plenary session for all employees (managers and construction workers) at the company comprised a presentation by the principal researcher (EvD) and an occupational physician, accompanied by a documentary about health risks of quartz exposure, and PIMEX videos [i.e. method to visualize the impact of good work practices, e.g. proper use of local exhaust ventilation or water suppression techniques, and poor work practices, e.g. no use of shielding dusty locations or compressed air cleaning, on exposure]. The second individual session was organized at the worksite and aimed to teach construction workers how to use technical control measures, including the discussion about constraints and possible solutions. Simultaneously with these worksite visits, a separate meeting was organized for the managers to give them more insight in the availability of state-of-the-art technical control measures. During the last plenary session at the company, all employees discussed with the principal researcher key solutions to overcome the main constraints when using technical control measures. Additionally, a labour inspector explained the policy of the labour inspection regarding quartz exposure during this session.

Outcome measures

This study investigated the effectiveness of the intervention on personal quartz exposure levels and the use of technical control measures [e.g. tool-integrated local exhaust ventilation (LEV) and water suppression techniques]. Follow-up measurements were conducted ~24 months after the baseline survey, 6 months after the implementation of the intervention. Population characteristics were obtained from the questionnaire administered to the construction workers (van Deurssen et al., 2014).

Full-shift personal quartz samples were taken using Dewell-Higgins cyclones mounted with a PVC filter (Millipore, pore size 5.0 μm, diameter 25mm), connected to a calibrated Gillian GilAir pump with an airflow of 2 l min−1 (van Deurssen et al., 2014). Quartz content of the filters was determined by infrared spectroscopy and X-ray diffraction, according to MDHS 101 (HSE, 2005). The analytical limit of detection (LOD) of quartz was 0.01mg (HSE, 2005).

Use of technical control measures was assessed through observation of the workers throughout their shift, using a structured walk-through survey to obtain detailed information on the duration and type of technical control measures used (van Deurssen et al., 2014).

Statistical analyses

Samples below the analytical LOD of quartz were assigned a value of two thirds of the detection limit. The quartz exposure distribution was highly skewed. Therefore, exposure data were log transformed prior to statistical analyses. Potential differences in population characteristics between construction workers in the intervention and control group, such as age, education level and baseline exposure levels, were tested using unpaired t tests (continuous variables) and Pearson Chi-square tests (dichotomous variables).

The statistical analyses used to evaluate the intervention effect followed a stepwise approach. First, descriptive statistics were generated for all job categories to gain insight in differences in quartz exposure levels between different groups over time. Second, hierarchical models were used to evaluate the intervention effect, defined as the difference in change in quartz exposure (natural logarithm of the quartz concentration (mg m−3) as dependent variable) from baseline to follow-up between the intervention and control group. Bayesian models were used because these are particularly suited to cope with the unbalanced structure of the dataset (i.e. absence of both a baseline and follow-up measurement for part of the subjects). On job category level, i.e. concrete driller, demolisher and tuck pointer, an effect was estimated of occasion (pre- or post-intervention), condition (control or intervention), and the occasion*condition interaction term (the intervention effect). A random intercept for subject was included to adjust for correlations between repeated measures on the same worker. The hierarchical model was estimated using a Bayesian approach with Markov chain Monte Carlo (MCMC) methods, primarily for computational reasons (i.e. good convergence properties given the relatively few observations to estimate some of the random effects). Pearson Chi-square tests were generated to investigate the difference in duration of use of control measures by job category between baseline and follow-up among the control and intervention condition. Use of control measures was expressed as fraction of the time that abrasive tasks (e.g. drilling, sawing, jackhammering, tuck pointing) were performed, since it was observed that control measures were used only during abrasive tasks.


Bayesian estimation was performed using R (version 3.1.2; R Foundation for Statistical Computing, Vienna, Austria), while the remaining analyses were performed using SAS v9.3 (SAS Institute Inc., Cary, NC, USA). A P < 0.05 was considered as statistically significant. The code for the Bayesian analyses programmed in RStan (version 2.5.0; Stan Development Team) is presented in the Supplementary data, available at Annals of Occupational Hygiene online.
RESULTS

Participant flow and population characteristics

In total, a selection of 13 companies was approached to participate. Five companies were excluded because they employed too few permanent construction workers or because they had insufficient work supply during the period of the baseline survey. Hence, 62% (8 companies) were enrolled in the intervention study (Fig. 1). These eight companies employed in total 404 eligible construction workers (177 in the control group and 227 in the intervention group). Company size varied between 15 and 103 construction workers.

Personal full-shift exposure measurements were collected from 116 construction workers (n = 149 measurements) during the baseline survey (van Deurssen et al., 2014), and 104 construction workers (n = 133 measurements) during follow-up (Fig. 1). In total, 68 construction workers had at least one measurement on both occasions. At baseline a higher percentage of the intervention group only followed secondary school (P < 0.05), while a higher percentage of the control group followed medium or high education (P < 0.05) (Table 1). Workers lost-to-follow-up had a lower level of education (P < 0.05) than the remaining workers measured at baseline, whereas no differences were observed between new entrants and workers measured at baseline.

Intervention effects and quartz exposure

The study demonstrated an overall reduction in quartz exposure in both the control and intervention group. This reduction was larger in the intervention group (73% compared to 40% in the control group; P < 0.001). The intervention effect could only be estimated for concrete drillers, demolishers, and tuck pointers, as the model provided unreliable estimates when all job categories were included. This was due to the absence of carpenters in the intervention group and the low exposure levels at baseline for both bricklayers and carpenters, which left very little potential for improvement (van Deurssen et al., 2014). The difference in reduction in exposure was significant for demolishers and tuck pointers (P = 0.005 and P = 0.008, respectively), but not for concrete drillers (P = 0.15) (Fig. 2).

The reduction in exposure was also reflected in the number of measurements above the OEL. In the three high exposed job categories, 75 and 86% of the baseline measurements exceeded the Dutch OEL for quartz (0.075mg m−3) in the intervention group and control group, respectively. During follow-up, this was reduced to 40% of the measurements in the intervention group versus 60% of the measurements in the control group (Table 2).

The intervention aimed to establish an increase in the use of technical control measures in order to reduce occupational quartz exposure. Such an increased use of control measures was observed for concrete drillers in particular, even though there was no statistically significant effect of the intervention on exposure within this job category. Concrete drillers in the intervention group used control measures for a significant greater proportion of time spent on abrasive tasks during follow-up, compared to baseline (P < 0.05; Table 3). This increase in use of control measures was attributable to an increase in the use of water suppression techniques. Although not statistically significant, demolishers and tuck pointers in the intervention group also tended to use water suppression techniques for a greater proportion of time spent on abrasive tasks during follow-up compared to baseline. No clear differences in use of LEV could be observed between the control and intervention group.

To test if the observed intervention effects for the various job categories could be explained by the increased use of control measures, we adjusted the intervention effect for the change in usage of control measures by job category by adding control measure as an explanatory variable into the model. This resulted in a diminished change in exposure for concrete drillers (not statistically significant). Although the effect was not statistically significant prior to adding control measures as an explanatory variable, this may indicate that the increased use of control measures among concrete drillers is at least partially responsible for the decrease in exposure observed in this job category. The change in exposure remained similar with the addition of the control use variable for demolishers and tuck pointers although the significance level decreased (Table 4). Since the use of control measures was only slightly increased for demolishers and even slightly decreased for tuck pointers in the intervention group during follow-up (both not statistically significant), it is not likely that the change in exposure in these groups was caused by an increased use of control measures among these two job categories.

Several other variables potentially influencing exposure, which were not directly the part of intervention program, changed over time. These variables were selected if they changed over time, if they were not related to the primary intervention, and if they were associated with exposure. Due to colinearity and limited statistical power, some of these variables representing (almost) similar determinants were merged into composite variables. As these composite variables may have confounded the estimated intervention effect (in subgroups), sensitivity analyses were performed by job category. Work location and time spent on abrasive tasks were selected as composite variables, since it was observed that construction workers in the intervention group performed less abrasive tasks and were more often working outside during the follow-up measurements compared with the baseline measurements. The results of the sensitivity analyses showed that the intervention effects differed by job category (Table 4). Changes in work location attenuated the intervention effect for tuck pointers, although this was not statistically significant. However, for concrete drillers and demolishers the intervention effect disappeared or even was reversed when adjusting for changes in work location. A similar analysis with adjustment for time spent on abrasive tasks showed that in general the intervention effect remained visible for each of the job categories although the effect was almost halved for the demolishers.
