Background
Construction workers are at a high risk of exposure to various types of hazardous substances such as crystalline silica [
1,
2]. Crystalline silica is an abundant material that is commonly released in respirable form during different construction activities such as concrete work, abrasive blasting, demolition, excavation, earth moving, tunnel construction, and highway building [
3]. Reports indicate that the level of silica exposure for numerous construction workers in Ontario, Canada exceed occupational exposure limit (i.e. 0.05 mg/m
3) [
1]. This is likely the case in other jurisdictions across Canada and internationally. Meanwhile, occupational silica-related diseases such as lung cancer annually impose considerable direct costs to the healthcare system and indirect costs to industry in the form of lost output and reduced productivity, as well as high intangible costs in the form of health-related quality of life losses to afflicted workers and their families [
4].
There are several silica exposure reduction interventions applicable to construction projects [
5‐
10]. These interventions work in different ways, e.g., preventing silica dust from getting into the atmosphere; removing dust in the atmosphere; and preventing workers from inhaling the dust if present in the atmosphere. Wet method (WM) refers to the use of water with devices to reduce the release of silica dust. Local exhaust ventilation (LEV) refers to the use of local vacuum systems at the point of operation to reduce the release of free silica dust into the work environment. Personal protective equipment (PPE) refers to the use of National Institute for Occupational Safety and Health approved air-purifying or supplied-air respirators. Enclosed work areas and work hygiene practices are some other common types of intervention options, but are not considered here.
Though several studies provide evidence on the effectiveness of different silica exposure reduction interventions in the construction sector, choosing a specific intervention is best informed by an economic evaluation. Despite the importance of the issue, there are only a few economic evaluations of silica exposure reduction interventions. One of these studies by Lahiri et al. [
7] evaluates the costs and effects of different interventions for the prevention of occupationally induced silicosis. They estimate the cost-effectiveness in terms of the dollars spent to obtain an additional healthy year. Another economic evaluation study by the Occupational Safety and Health Administration (OSHA) in the United States [
5] addresses issues related to costs, technological feasibility, and the economic impacts of the proposed respirable crystalline silica rule, which attempts to reduce the permissible exposure limits from its current level of 0.1 mg/m
3 to 0.05 mg/m
3. To do so, the authors forecast the number of silica-related diseases averted as a result of the proposed rule and compare the value of averted cases with the cost of compliance to the rule in all affected industrial sectors.
Uncertainty about the magnitude of input variables of an intervention, which has often been cited as a limitation in economic evaluation studies, can affect the precision of results [
11,
12]. Input data for these studies can be provided as probabilistic or deterministic values. Deterministic values should only be applied when specific values are available from a reliable source, while it is best to use probabilistic values when the reliability of information is questionable [
13]. In the case at hand, we have large number of uncertain variables that impact an intervention’s economic evaluation results. For instance the number of silica-exposed workers and the level of exposure to silica are uncertain variables. The level of exposure is influenced by several factors such as the task, workstation characteristics (e.g. being indoor or outdoor), materials being used, phase of the construction project and other unknown variables. In many circumstances, it is not possible to collect more data on the level of exposure because of the quick pace of change on a construction project site, tasks characteristic, and/or safety requirements [
11]. The risk of getting a silica-related occupational disease for workers of different age and sex also has a high degree of uncertainty, since latent health conditions such as lung cancer are influenced by multiple factors not easily recognized as attributable to occupational silica exposures [
14]. The cost of respiratory disease treatment is also an uncertain variable as it depends on, amongst other things, the stage of the disease and the age and sex of the individual [
15,
16]. In terms of the effectiveness of a silica exposure reduction interventions, the maximum is achieved by appropriate and systematic use of an intervention, which is not always the case in practice. For example, some studies suggest that the malfunction of PPE is influenced by several environmental factors such as worker’s awareness, the nature of the hazard, climate, and occupational health and safety inspections [
17]. The overall effectiveness of WM and LEV interventions also depends on the workstation characteristics and the number of people working near silica dust sources. Because work arrangements vary within occupations and across facilities of different sizes, there is no definitive data on how many workers are likely to be protected by a given intervention [
5‐
10].
There are several probabilistic modeling approaches for solving problems under different levels of uncertainty and estimation of expected value, such as decision trees [
18], Markov models [
18], and Bayesian networks (BN) [
19]. Decision tree analysis involves drawing on a tree-shaped diagram to assist with statistical probability analysis and identifying a solution to the problem. In decision trees, the probability of each possible event is explicitly identified, along with the consequences of those events. This method is frequently used in health economics, specifically for problems that are more complex in nature [
18]. Markov models are being used more often in economic evaluation and are probably the most common type of model used in the economic evaluation of healthcare interventions [
18]. The main advantage of a Markov model is the representation of recurring events. Although they are similar to decision trees, they do not allow for interaction among variables [
18]. BN (also called belief networks) approach, is a graphical structure that allows one to capture the relationships between variables. To illustrate these relationships, a diagram of nodes and arrows is often used. Nodes represent the system variables and the arrows symbolize the direct dependencies among the variables. BN are used to compute the distribution probabilities in a set of variables according to the observation of some variables and the prior knowledge of the others [
19]. Recently, the BN approach has gained popularity in different areas of health economics [
13], project cost and risk analysis [
20‐
22], cost-benefit analysis [
23], and occupational health and safety decision making [
24]. BNs are preferred for several reasons, such as the ability to integrate various types of data (i.e., qualitative and quantitative), to combine available data with expert knowledge, to explicitly consider relationships between variables, to model complex problems with many variables involving a high level of uncertainty and to easily provide graphical representations [
19‐
23]. The modeling languages of BNs have straightforward semantics, namely that of cause and effect. Furthermore, the needed probability calculation of BNs is often undertaken with the assistance of software packages such as Netica, GeNIe, BayesiaLab, Analytica, Hugin, Bayes Net Toolbox, and many others (this is not a comprehensive list, and not meant to promote any specific software). In this study, our objective is to identify the most cost-beneficial silica exposure reduction intervention for the construction sector in Ontario, Canada. To estimate the net benefit of each intervention, we apply a probabilistic modeling approach to compare the expected cost of lung cancer cases averted, with expected cost of implementation of each intervention in one calendar year. We anticipate this study provides important insights for occupational health policy makers and workplace parties in the construction sector. More broadly, this study provides a methodological framework for a more complete treatment of uncertainties in the economic evaluation of occupational health and safety interventions via BN.
Results
Expected costs and benefits
Table
1 presents the expected lung cancer cases averted and net benefit of the seven silica exposure reduction interventions. The values are calculated separately for each of the seven intervention combinations. The percentage of the silica-exposed workers assumed to be protected by each intervention, and the expected lung cancer cases averted are indicated in the first and the second rows, respectively. In the table, we illustrate the cost of lung cancer cases averted (i.e. the benefit) with a positive sign and the intervention costs with a negative sign. As indicated in Table
1, we find the highest lung cancer cases are averted with a combined use of WM, LEV and PPE, about 107 cases, resulting in a net benefit of $45.9 million. With this intervention, all the silica-exposed workers are simultaneously protected with a combined use of the three methods, which makes the cost of this intervention the highest amongst the seven interventions.
Table 1Expected Costs and Benefits of Silica Exposure Reduction Interventions
Protected workersa | 100% | 100% | 100% | 60% | 100% | 40% | 100% |
Lung cancer cases avertedb | 107 | 95 | 102 | 55 | 101 | 40 | 96 |
Averted costs (benefits) |
Direct | $9.5 M | $8.4 M | $9.0 M | $4.9 M | $8.9 M | $3.5 M | $8.6 M |
Indirect | $41.2 M | $36.6 M | $39.3 M | $21.1 M | $38.8 M | $15.3 M | $37.1 M |
Intangible | $133.9 M | $119.1 M | $127.6 M | $68.6 M | $126.0 M | $49.7 M | $120.7 M |
Total | $184.5 M | $164.2 M | $175.9 M | $94.5 M | $173.8 M | $68.5 M | $166.4 M |
Intervention costs |
WMc | -$42.0 M | -$42.0 M | -$42.0 M | -$42.0 M | $0 | $0 | $0 |
LEVd | -$15.5 M | -$15.5 M | $0 | $0 | -$15.5 M | -$15.5 M | $0 |
PPEe | -$81.1 M | $0 | -$81.1 M | $0 | -$81.1 M | $0 | -$81.1 M |
Total | -$138.6 M | -$57.6 M | -$123.1 M | -$42.0 M | -$96.6 M | -$15.5 M | -$81.1 M |
Net benefitf | $45.9 M | $106.6 M | $52.8 M | $52.5 M | $77.2 | $53.0 M | $85.3 M |
Benefit to cost ratiog | 1.3 | 2.9 | 1.4 | 2.2 | 1.8 | 4.4 | 2.1 |
With simultaneous use of WM and LEV, about 95 lung cancer cases are expected to be averted. With this intervention, all silica-exposed workers are protected via WM or LEV. The net benefit of this intervention is $106.6 million, which is the highest among the seven interventions. The implementation cost of this intervention is much less than the cost of the combined use of all three methods, which makes it a more desirable intervention in the case of budget restrictions.
In the case of WM-PPE or LEV-PPE use, we expect a similar number of lung cancer cases averted, about 102 and 101 cases, respectively. With these interventions all silica-exposed workers are protected by PPE, but only a percentage of them are protected by WM or LEV. For example, in WM-PPE, 60% of the silica-exposed workers are protected by both WM and PPE and the remainder are protected with PPE, while for LEV-PPE only 40% of all silica-exposed workers are protected by both LEV and PPE. The net benefit of WM-PPE is estimated at $52.8 million, which is much lower than LEV-PPE, at about $77.2 million, due to its higher intervention cost.
With PPE use alone, we expect 96 lung cancer cases averted and estimate a net benefit of $85.3 million. The results indicate that lung cancer cases averted with PPE are relatively higher than WM and PPE on their own. However, the total benefit of this intervention is lower than WM and PPE, due to a higher implementation cost.
The lung cancer cases averted with WM and LEV on their own are estimated at 57 and 42 cases, respectively, which is relatively lower in comparison to other intervention options, as they only protect a percentage of the silica-exposed workers (i.e., 60% in WM and 40% in LEV). The net benefit of WM is estimated at $52.8 million, which is slightly lower than LEV, at $53 million, due to its higher intervention cost.
The benefit-cost ratio of all seven interventions are positive. The highest benefit-cost ratio is achieved with LEV (4.4), followed by combined use of WM and LEV (2.9), WM (2.2), PPE (2.1), LEV-PPE (1.8), WM-PPE (1.4), and WM-LEV-PPE (1.3). The general rule of thumb is that if the benefit is higher than the cost, the project is a good investment (i.e., a benefit-cost ratio greater than 1). Although it is important to note this fact, WM and LEV on their own protect only a percentage of silica-exposed workers.
Sensitivity analysis
Table
2 shows how the number of silica-exposed workers and the level of exposure affect the net benefit of each of silica exposure reduction interventions. For this part, we only evaluated interventions that protect the entire silica-exposed workers, namely WM-LEV-PPE, WM-LEV and PPE. The first column represents outcomes, when all variables are in their default distribution. When none of the interventions are implemented, we expect 110 lung cancer cases, which results in an economic burden of $189 million. We set the level of exposure to low, medium, and high and estimate the net benefit of the interventions, for the lower and upper bound values of silica-exposed workers in the construction sector. With a combined use of the three types of prevention activities, we expect a net benefit of $4 million when we set silica-exposed workers and level of exposure at the lower bound value, while we expect net benefit of $107 million when set at the upper bound value. With WM and LEV combined and PPE on its own, we expect a net benefit of $60 million and $45 million, respectively, when we set the silica-exposed workers and level of exposure at the lower bound. We expect a net benefit of $101 million and $94 million respectively when we set it at the upper bound. Note that WM and LEV combined and PPE on their own both protect 100% of silica-exposed workers.
Table 2Sensitivity Analysis of Interventions for Different Numbers of Silica-Exposed Workers and Different Levels of Exposure
Silica-exposed workersa | | Lower bound | Upper bound | Lower bound | Upper bound | Lower bound | Upper bound |
91 | 46 | 118 | 46 | 118 | 46 | 118 |
No Intervention |
Expected LC casesb | 110 | 60 | 111 | 80 | 140 | 95 | 180 |
Total LC costsc | $189 M | $103 M | $191 M | $138 M | $241 M | $164 M | $310 M |
WM-LEV-PPE |
LC cases avertedd | 107 | 60 | 111 | 79 | 138 | 84 | 162 |
Total LC costs avertede | $185 M | $103 M | $191 M | $136 M | $238 M | $145 M | $280 M |
Total intervention costsf | $151 M | $99 M | $173 M | $99 M | $173 M | $99 M | $173 M |
Net benefitg | $46 M | $4 M | $19 M | $37 M | $65 M | $46 M | $107 M |
Net benefit change (%) | – | 10% | 41% | 80% | 142% | 100% | 233% |
WM-LEV |
LC cases avertedd | 95 | 60 | 111 | 71 | 124 | 46 | 100 |
Total LC costs avertede | $164 M | $103 M | $191 M | $122 M | $214 M | $79 M | $172 M |
Total intervention costsf | $63 M | $43 M | $71 M | $43 M | $71 M | $43 M | $71 M |
Net benefitg | $107 M | $60 M | $120 M | $79 M | $143 M | $36 M | $101 M |
Net benefit change (%) | – | 57% | 113% | 74% | 134% | 34% | 95% |
PPE |
LC cases avertedd | 96 | 59 | 109 | 73 | 129 | 57 | 113 |
Total LC costs avertede | $166 M | $101 M | $187 M | $126 M | $222 M | $97 M | $195 M |
Total intervention costsf | $81 M | $56 M | $101 M | $56 M | $101 M | $56 M | $101 M |
Net benefitg | $85 M | $45 M | $86 M | $70 M | $120 M | $42 M | $94 M |
Net benefit change (%) | – | 53% | 101% | 83% | 141% | 49% | 110% |
Discussion
Among the seven silica exposure reduction interventions considered in this study, we estimate the highest number of lung cancer cases are averted with a combined use of WM-LEV-PPE (107 cases). Despite this fact, the highest net benefit is achieved with WM and LEV, about $106.6 million, due to their lower implementation costs. The lowest number of lung cancer cases are averted with WM or LEV (55 and 40 cases), as these interventions protect only a fraction of the silica-exposed workers. With a low or medium level of silica exposure, a combined use of WM and LEV are expected to produce the highest net benefit, while with a high level of exposure, the combined use of WM-LEV-PPE is expected to result in the highest net benefit.
In terms on future uses of BN in the area of Occupational Health and Safety (OHS) economic evaluation, one potential use is trade-off analysis between expected costs and benefits of an intervention when there is a budget constraint, or when one is interested in identifying the required budget to avert a specific number of lung cancer cases. For example, as shown in Table
1, we can consider a situation in which the budget is constrained to $70 million. In such a situation, using WM-LEV is the only intervention that will protect 100% of silica-exposed workers without the total intervention cost exceeding the pre-set amount. Trade-off analysis provides an opportunity for decision makers to define their targets regarding the prevention of a specific number of occupational lung cancer cases in the context of a predetermined budget.
To our knowledge, this is the first study to use the cost of silica-related occupational lung cancer cases averted for the benefit component in the economic evaluation of an intervention. Therefore, it is difficult to compare our findings with those of other studies. For example, Lahiri et al. [
7] consider the averted cost of occupationally induced silicosis as a benefit and the cost of different interventions. They estimate the cost-effectiveness of interventions with a ratio (i.e., dollars per healthy years gained), and find they vary between $132.3 ($105.9 in 2005 US dollars) and $136.2 ($109 in 2005 US dollars) for different geographic sub-regions. However, they do not include cost items such as healthcare, informal caregiving, out-of-pocket, and home production losses in their analysis. Despite difference in economic evaluation methodologies and the inconsistencies of considered outcomes, our results are in line with Lahiri et al. [
7], as we also identify the net benefit of WM-LEV as the highest among seven interventions. However, as they neither report the average per-case cost for interventions nor the number of silica-exposed workers affected, so we are unable to estimate a per-case value for their study.
In another study in the United States, OSHA estimates the net benefit of compliance with a new silica rule in terms of reduction of cost of silica-related diseases (i.e., fatal cases of lung cancer, non-malignant respiratory diseases, renal diseases and nonfatal cases of silicosis) [
5]. They estimate the net annualized benefit of a reduction in the acceptable limit of exposure to be between $2.4 billion and $9.9 billion ($1.8 billion and $7.5 billion in 2009 US dollars), with a midpoint value of $6.1 billion ($4.6 billion in 2009 US dollars). Annually, the lowering of the exposure limit prevents 688 fatalities (567 fatalities in the construction sector) and 1585 moderate-to-severe silicosis cases (1080 cases in the construction sector).
In both of the studies referred to above, researchers assume all the variables are deterministic. However, uncertainty of variables is relevant for most of occupational health interventions. In an economic evaluation of an intervention one generally includes many variables with different levels of uncertainty, yet this uncertainty across data inputs has not been substantively addressed in OHS studies in the past. Ours is one of the first economic evaluation studies in OHS field to use BN. We also provide an overview of some of the potential benefits of using this approach and guidance on how to do so. Specifically, we explained the main steps of developing a BN in OHS setting, how to parametrize the variables, define the variable distribution, and incorporate them into a model. We capture the uncertainty of each variable and integrate the dependencies between them using a BN, and then estimate the expected net benefit of various interventions.
While the BN model developed in this study can support decision making, in its current form there is room for improvement of the approach. Future work in this area ought to include further research on the expansion of the model contents, including consideration of a broader set of variables. For example, in our study, the benefit side of our economic evaluation is limited to occupational lung cancer cases averted, despite the fact there are several other silica-related occupational diseases such as silicosis and silicosis-related diseases [
4,
5,
7]. Additionally, to estimate averted productivity losses we focus only on absenteeism, not presenteeism, primarily because there is a lack of evidence to draw on for the magnitude of productivity losses associated with lung cancer cases upon return to work. Furthermore, our model structure can be improved upon by considering a greater number of relationships between the key variables, since we ignored some interactions because of limitations in background knowledge. For instance, interventions may adversely influence labour and/or equipment productivity [
5], and under certain circumstances, the health-related quality of life of workers may be affected by the intervention [
17]. Another example in this regard is the dependencies that might exist between age and sex in terms of the survival rates of occupational silica-related lung cancer cases. A more comprehensive analysis would consider other variable interactions that are caused by implementing an intervention. Moreover, in this study we do not investigate the time needed to implement silica reduction interventions and the duration of their effectiveness. Undoubtedly, because of the relatively long latency period of lung cancer, ultimate effect of silica reduction interventions will only be realized after several years. Further research is also needed on how to incorporate the uncertainty of the timeline of interventions into an economic evaluation, for example to account for variability in how many years after the introduction of an intervention it takes before the reduction of lung cancer cases reaches a steady state. Lastly, implementation of sensitivity analysis and sorting model variables by level of uncertainty is also one of the abilities wherein BN can provide invaluable analytic insight for policy makers, particularly for the purpose of developing data gathering strategies. In this regard, we also recommend implementation of Value of Information Analyses in future research, as it also enables one to identify parts of a model where additional data (reduction of uncertainty) is most useful.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.