Abstract
The Ontario Asbestos Workers Registry is a regulatory exposure registry obligating employers to report the number of work hours with asbestos-containing materials for each of their workers. Currently, each worker is notified of the need for a medical examination once they have accrued 2000 reported hours of work with asbestos. We sought to evaluate the impact on disease prevention of alternative policies limiting asbestos work hours among registry participants. A cohort of 26 164 asbestos workers were followed for cancer and nonmalignant disease diagnoses between 1986 and 2019. Analyses of the association between cumulative asbestos work hours and respiratory disease incidence rates showed substantially elevated disease rates well before reaching 2000 asbestos work hours. Using a simplified application of parametric G-computation (G-POSH), limiting cumulative asbestos work hours to 100 h would have prevented 76 asbestosis, 36 pulmonary fibrosis, 27 mesothelioma, and 79 lung cancer cases at the end of follow-up compared to the observed risk in the cohort. Limiting exposure to 2000 asbestos work hours had a smaller but still substantial impact on disease prevention, particularly among workers in the construction industry. Regulatory agencies should intervene sooner to prevent respiratory disease among workers in the registry.
Keywords: asbestos, lung cancer, mesothelioma, asbestosis, fibrosis, G-computation
Introduction
The Ontario Asbestos Workers Registry (AWR) is a regulatory exposure surveillance program for individuals exposed to asbestos at work in the province of Ontario, Canada.1 Since 1986, employers engaging in construction and building repair have been required to report all workers conducting specific work operations with asbestos-containing materials to the registry, which is administered by the provincial Ministry of Labour, Immigration, Training, and Skills Development (MLITSD). Although not required, many employers in other industries where construction-related tasks with asbestos are undertaken, such as manufacturing, education, and utilities, report to the registry as well. The Ministry has a policy of notifying workers of the need for a medical examination once the worker has accumulated 2000 reported hours of work with asbestos. While this policy is intended to help prevent the occurrence of asbestos-related disease, prior analyses involving over 26 000 workers in the AWR cohort study followed between 1986 and 2019 found that workers experienced a standardized incidence rate of mesothelioma 7 times, asbestosis 11 times, and pulmonary fibrosis 14 times that of the general population of Ontario.2 The disease burden was substantial, with over 450 cases of these diseases diagnosed.
Inhalation of all forms of asbestos fibers causes mesothelioma and lung cancer in humans.3 Asbestos inhalation also causes interstitial pulmonary fibrosis, a nonmalignant restrictive lung disease that occurs due to inflammation and scarring of lung tissue.4 Pulmonary fibrosis may be classified as asbestosis when a history of asbestos exposure is established. Although the import, manufacture, and use of asbestos-containing materials has been prohibited in Canada since 2018, the widespread production and use of asbestos in the country over the 20th century makes it a prominent occupational hazard for the future.5 Contemporary occupational exposure to asbestos is common among workers involved in the maintenance and renovation of aging buildings and structures that were built with asbestos-containing materials, such as insulation, cement, plaster, roofing, siding, flooring, paint, and drywall.6 According to CAREX Canada, a national carcinogen exposure surveillance project, an estimated 77 000 workers were exposed to asbestos in the workplace in Ontario in 2016, with the large majority employed in specialty trade contracting and building construction industries.7
We sought to compare the Ministry’s current policy, which notifies asbestos workers in the AWR at 2000 cumulative work hours, to alternative interventions that would limit their work hours to equal or lower levels. Although the current policy does not force a worker to limit their work hours, we considered work hour limits to be a proxy for the elimination of asbestos fiber inhalation regardless of continued asbestos work, with the objective of informing decision-makers about the impact of hypothetical exposure reduction interventions. We used a simplified application of parametric G-computation8 to estimate intervention effects for time-varying exposure in time-to-event analyses,9,10 a common scenario in occupational cohort analysis, which we described previously and term G-POSH (G-computation for Policy Evaluation in Occupational Safety and Health).11 We also describe an extension to G-POSH that uses inverse probability of censoring weighting to account for potential healthy worker survivor bias,12 a form of time-varying confounding that the parametric G-formula was originally developed to solve,13 when assessing an exposure intervention (Appendix S1). However, for simplicity, and due to limited employment data available in the AWR, our primary analyses focus on G-POSH without inverse probability of censoring weighting or adjustment for healthy worker survivor bias, although this bias was evaluated empirically.
In our cohort study of 26 164 workers in the AWR followed for cancer and nonmalignant disease diagnoses between 1986 and 2019 in Ontario, Canada, the specific objectives of this work were to: (1) estimate exposure-response associations between cumulative asbestos work hours and the incidence of asbestosis, pulmonary fibrosis, mesothelioma, and lung cancer, and (2) estimate the impact of interventions limiting asbestos work hours on disease prevention using an application of G-POSH. Results of this study may guide revisions to the Ministry’s current 2000-work hour notification policy and help reduce the future burden of respiratory disease and premature death among workers in the registry.
Methods
Study population
The AWR includes workers reported to the provincial occupational asbestos exposure registry in Ontario, Canada.1 Reporting to the registry began on January 1, 1986. Of the 39 898 workers in the full registry in 2018, 6702 (17%) workers were excluded who had invalid or missing identifying information, were completely missing an exposure record, or had duplicate entries.2 An additional 6923 (17%) were excluded who did not link to administrative health data due to being unregistered with the province’s universal health insurance plan (eg, out-of-province and foreign workers). A total of 26 164 workers were included in the present analyses, consisting of male and female workers aged 15 years or more at the time of their first reported asbestos work between the start of the registry in 1986 and December 31, 2018.
Assessment of asbestos exposure
Asbestos exposure was defined as the cumulative number of hours working with asbestos-containing materials in work operations rated as having a medium (type 2) or high (type 3) potential for exposure to asbestos fibers.14 Type 2 and 3 work operations include enclosure, repair, removal, cutting, and sanding of friable and nonfriable asbestos-containing materials, among other activities. Every worker engaged in type 2 or 3 work in construction or repair operations in Ontario is required to have their asbestos work hours reported to the AWR by their employer to be compliant with provincial regulation, although actual compliance may vary. The number of asbestos work hours was reported by employers on a designated reporting form and sent to the Ministry at least once per year. No other employment or occupational information about workers was reported to the registry and there was no systematic record of employment status. Historical exposure reports were accepted by the registry and were included in the calculation of cumulative work hours. The work report included identifying information on the employer and workers were required to receive a copy of their report.
The number of asbestos work hours was missing on 8467 (13%) of submitted work reports with 4270 (16%) individuals in the cohort having at least 1 report with a missing value. Missing values were imputed by using the worker’s geometric mean work hours in their own work history or, if unavailable, by using the geometric mean in their industry group of employment (North American Industry Classification System, NAICS; 3-digit level). Employers were classified into NAICS groups by the Ministry. A total of 13 workers with unimputable exposure due to missing work hours and employer information were excluded from analyses. A sensitivity analysis restricting the cohort to workers never missing any work hours was conducted to evaluate the impact of imputed exposure data on standard regression results (Table S1).
Follow-up and disease ascertainment
Cancer diagnoses were ascertained through follow-up in the Ontario Cancer Registry from January 1, 1986 to December 31, 2019, while nonmalignant disease diagnoses were ascertained through administrative health databases for ambulatory care, hospitalizations, and physician billing from 1999 to 2019. Health data sources included the National Ambulatory Care Reporting System (NACRS), the Discharge Abstract Database (DAD), and the Ontario Health Insurance Plan (OHIP) claims database, with ambulatory care and hospitalization data available from 2006 to 2019 only. Workers were censored if there was indication of deceased vital status or emigration out of the province as ascertained through follow-up in Ontario’s universal health insurance registration database (OHIP-Registered Persons Database) through December 31, 2019.
Mesothelioma cases were defined as individuals in the cancer registry with a first primary diagnosis of malignant neoplasm of the pleura (ICD-10 code C38.4) or mesothelioma (ICD-10 C45), which includes peritoneal disease (ICD-10 C45.1). Lung cancer cases were those with a first primary diagnosis of malignant neoplasm of the bronchus and lung (ICD-10 C34). For nonmalignant disease, asbestosis cases were defined as those with a diagnosis of asbestosis (ICD-10 J61 or ICD-9501) in at least 1 hospitalization record, 1 ambulatory care visit, or 2 physician billing records. Pulmonary fibrosis cases were those with a diagnosis of interstitial pulmonary disease with fibrosis (ICD-10 J84.1 or ICD-9515), which includes idiopathic disease, in at least 1 hospitalization record, 1 ambulatory care visit, or 2 physician billing records within 1 year. Case definitions for asbestosis and pulmonary fibrosis were based on previous surveillance studies in Canada.15-17 The cancer registry and health databases are considered complete for almost all residents of Ontario during the calendar periods used for follow-up.
Statistical analysis
First, standard Poisson regression analysis was used to estimate incidence rate ratios (IRRs) and Wald 95% CIs for the association between the cumulative number of asbestos work hours and the incidence of cancer and nonmalignant respiratory disease, with each outcome modeled independently. Follow-up started on the later of the date of first reported asbestos work or 1986 for cancer outcomes or 1999 for nonmalignant disease. Follow-up ended on the latest of the date of death, emigration from Ontario, age 85 years, disease diagnosis, or end of study in 2019. Time-varying cumulative asbestos work hours, under a 0-, 5-, and 10-year lag, were modeled using categorical indicator and polynomial terms. Models were covariate adjusted for available confounders of attained age (cubic), birth year (linear), and sex (binary). For analyses of lung cancer, given that individuals in this occupational cohort share similar socioeconomic characteristics by virtue of their employment in the same industry and job type (ie, construction trades), some additional control of confounding by lifestyle behaviors is inherent in the study design. Tobacco smoking and lifestyle behaviors were not considered confounders for asbestosis (and related fibrosis) or mesothelioma as these diseases are caused exclusively by asbestos. Model fit under various flexible covariate specifications was evaluated based on the akaike information criterion (AIC).
Next, to estimate intervention effects, we used an application of parametric G-computation8-10,13,18,19 that is simplified for Policy evaluation in Occupational Safety and Health (G-POSH), which has previously been described in detail.11 A person-year structured dataset was extended to include records for counterfactual person-years of follow-up for each worker up to the end of study in 2019 or age 85 years (whichever was earlier), hereafter referred to as “pseudo” years of follow-up. For each disease and for all-cause mortality, a logistic regression model was fit to obtain predicted values for the probability of a given diagnosis (or death) in each person-year of follow-up. However, logistic models were weighted such that they were only fit to person-years of follow-up where: (1) a worker’s cumulative work hours was at or below a specified exposure limit, and (2) the follow-up was truly observed for a given outcome (ie, not pseudo). The logistic models were fit using age, birth year, and sex as predictive covariates and specified with the same terms used in the standard Poisson models.
We conducted sensitivity analyses to evaluate the performance of our predictive model and the potential influence of healthy worker survivor bias, a form of time-varying confounding, on our results. A validation exercise20 evaluating the performance of the predictive model showed that the predicted natural course of all 4 diseases under no exposure limit estimated almost the exact number of cases observed in the cohort across the distribution of attained age (Figure S1). To empirically assess one of the component associations of HWSB in our cohort,21 we estimated the association between cumulative asbestos work hours under a 5-year lag and the rate of leaving the registry, specified as the last date of reported asbestos work. The beta parameter for cumulative exposure was very close to null. In addition, adding a binary covariate to our predictive model for having reported asbestos work hours in a given year, as a proxy for time-varying job status in the absence of employment records in our registry-based cohort, had no meaningful impact on model fit or predicted values.
The expected number of cases under each exposure limit intervention was calculated as the cumulative hazard of the outcome at the end of follow-up for all workers in the cohort, including the pseudo years. For each disease, the cumulative hazard was equal to the sum of the product of the predicted probability of disease in each year of follow-up (ie, discrete time hazard) and the probability of surviving disease-free through that year. The probability of disease-free survival was calculated as the complement of the sum of the discrete time hazard of death plus the sum of the discreet time hazard of the disease of interest. In the first year of follow-up for each worker, the survival probability was equal to 1. The expected person-years of follow-up under each exposure intervention was calculated as the cumulative sum of the survival probability in each person-year. Risk ratios were calculated at the end of follow-up to evaluate the relative proportion of reductions in disease incidence under each exposure limit using the cohort’s observed disease experience as the referent. The observed risk was used as the reference, rather than the predicted natural course under no intervention, for simpler results translation to policymakers and because estimates were virtually identical using either approach. We estimated 95% CIs as the 2.5th and 97.5th percentiles of analyses from 200 bootstrapped resamples of the original data. Risk closely approximated the incidence rate given the rarity of the diseases (≤2% prevalence). A worked example of the described methods, as well as instructions for accessing an R package (“G-POSH”) we developed to easily conduct this computation in similar studies, is provided in Appendix S2. All analyses were conducted in R version 4.3.1.
Exposure limits were evaluated at 100, 250, 500, 1000, and 2000 cumulative asbestos work hours with levels chosen according to the distribution of work hours in the cohort and the current 2000-h policy for notification. Because reporting to the AWR is required in the construction industry, while reporting by employers from other industries (eg, manufacturing) is potentially less systematic and subject to greater measurement error, we conducted separate analyses restricted to workers in the construction industry (NAICS-2 code 23) to obtain results generalizable to this industry alone. Appendix S1 describes an extension of the G-POSH method to account for healthy worker survivor bias by inverse probability of censoring weighting.12 In those analyses, exposure limits were evaluated at various levels of annual rather than cumulative asbestos work hours.
Results
Table 1 provides characteristics of the study cohort. Workers followed for cancer were followed for a median of 22 years, had a median age of 58 years at the end of follow-up, and were almost exclusively male (97%). During over half a million person-years of follow-up, 97 mesothelioma and 528 lung cancer cases and 3515 deaths occurred. There were 592 (2%) workers lost to follow-up due to emigration from Ontario. The median number of cumulative asbestos work hours at the end of follow-up was 96 h in the full cohort and 146 in workers employed in the construction industry. The subcohort of workers followed for nonmalignant disease represented 97% of the full cohort. There were 159 asbestosis and 192 pulmonary fibrosis cases diagnosed during follow-up in this group.
Table 1.
Characteristics of workers in the Asbestos Workers Registry cohort followed for cancer and nonmalignant disease, Ontario, Canada, 1986-2019.
| Characteristic | Followed for nonmalignant disease 1999-2019 (n = 25 503) | Followed for cancer 1986-2019 (n = 26 164) | ||
|---|---|---|---|---|
| Median (IQR) | n (%) | Median (IQR) | n (%) | |
| Calendar year at start of follow-up | 1999 (1999-2002) | 1993 (1989-2002) | ||
| Year of birth | 1959 (1950-1968) | 1958 (1948-1966) | ||
| Length of follow-up, years | 20 (13-20) | 22 (14-29) | ||
| Age at start of follow-up | 41 (32-51) | 35 (26-45) | ||
| Age at end of follow-up | 58 (49-68) | 58 (48-68) | ||
| Person-years of follow-up | 422 805 (100) | 567 481 (100) | ||
| Male sex | 24 738 (97) | 25 353 (97) | ||
| Deceased | 3028 (12) | 3515 (13) | ||
| Lost to follow-up (emigration) | 433 (2) | 592 (2) | ||
| Employed in constructiona | 15 954 (63) | 16 320 (62) | ||
| Cumulative no. of work hours with asbestos-containing materialsb | 96 (24-424) | 96 (24-441) | ||
| Construction industry onlya | 146 (50-640) | 146 (50-654) | ||
| Asbestos work reports by industry group | ||||
| Construction | 38 441 (64) | 39 045 (64) | ||
| Manufacturing | 10 199 (17) | 10 341 (17) | ||
| Educational services | 4048 (7) | 4261 (7) | ||
| Utilities | 2153 (4) | 2221 (4) | ||
| Health care and social assistance | 1516 (3) | 1528 (2) | ||
| Other | 3990 (7) | 4048 (7) | ||
Restricted to workers who ever had asbestos work hours reported by an employer classified in the construction industry (North American Industry Classification System code 23).
Missing values imputed based on within-individual and within-industry geometric mean asbestos work hours. No lag.
Table 2 provides estimates of the association between cumulative asbestos work hours under a 5-year lag and the incidence of respiratory disease using standard Poisson regression methods. Figure 1 presents visual plots of estimates from models reported in Table 2. The number of cumulative asbestos work hours was positively associated with the incidence rate of all 4 diseases, with categorical indicator models showing a positive and monotonic exposure-response relationship across all levels of exposure for asbestosis, pulmonary fibrosis, mesothelioma, and lung cancer, although some estimates for mesothelioma were statistically imprecise. Compared to the referent of ≥0-20 work hours, adjusted IRRs were substantially elevated in magnitude during follow-up well below 2000 cumulative asbestos work hours. Positive associations were observed for asbestosis and pulmonary fibrosis during just ≥20-100 cumulative work hours. Accumulating ≥2000 work hours was associated with high relative rates of asbestosis, pulmonary fibrosis, and mesothelioma. Under a 10-year lag, associations were similar in magnitude and exposure-response relationships remained positive for all diseases except mesothelioma, where associations were modestly attenuated in all categories (Table S2).
Table 2.
Association between cumulative asbestos work hours and incidence of respiratory disease estimated using standard regression methods, Asbestos Workers Registry cohort, Ontario, Canada, 1986-2019.
| Followed for nonmalignant disease, 1999-2019 (n = 25 503) | Followed for cancer, 1986-2019 (n = 26 164) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Asbestosis | Pulmonary fibrosis | Mesothelioma | Lung cancer | |||||||||
| No. cases | Person-years | Adj. IRR a (95% CI) | No. cases | Person-years | Adj. IRR a (95% CI) | No. cases | Person-years | Adj. IRR a (95% CI) | No. cases | Person-years | Adj. IRR a (95% CI) | |
| No. of cumulative asbestos work hours, 5-year lag | ||||||||||||
| ≥0 to <20 | 24 | 135 595 | 1 | 41 | 135 639 | 1 | 23 | 219 612 | 1 | 143 | 219 339 | 1 |
| ≥20 to <100 | 29 | 118 227 | 1.39 (0.81-2.39) | 50 | 118 208 | 1.25 (0.83-1.89) | 19 | 140 446 | 1.09 (0.59-2.02) | 119 | 140 264 | 1.08 (0.85-1.39) |
| ≥100 to <500 | 29 | 89 415 | 2.31 (1.34-3.97) | 34 | 89 474 | 1.37 (0.87-2.15) | 13 | 105 967 | 1.24 (0.62-2.46) | 96 | 105 791 | 1.43 (1.10-1.86) |
| ≥500 to <2000 | 18 | 38 464 | 3.63 (1.97-6.70) | 19 | 38 563 | 1.89 (1.10-3.27) | 6 | 45 511 | 1.38 (0.56-3.40) | 41 | 45 435 | 1.48 (1.04-2.10) |
| ≥2000 | 59 | 39 840 | 4.78 (2.95-7.74) | 48 | 40 184 | 2.26 (1.47-3.46) | 36 | 55 868 | 2.75 (1.60-4.70) | 129 | 55 705 | 1.55 (1.21-1.99) |
| P linear trend | <.001 | <.001 | <.001 | <.001 | ||||||||
| IRR at 100 asbestos work-hoursb | 158 | 417 257 | 1.11 (1.07-1.14) | 191 | 417 765 | 1.07 (1.03-1.10) | 95 | 561 609 | 1.08 (1.03-1.14) | 520 | 560 747 | 1.04 (1.02-1.07) |
Abbreviations: Adj., adjusted; IRR, incidence rate ratio.
All estimates generated from Poisson regression models adjusted for age (cubic), birth year (linear), and sex (binary). Cumulative work hours specified categorically using categorical indicator variables. Wald test for linear trend estimated from ordinal model with values equal to the median exposure level in each category.
Linear-quadratic model excluding follow-up where cumulative hours exceeded the 99th percentile to eliminate outlying values and improve model fit.
Figure 1.
Exposure-response relationship between cumulative asbestos work hours and respiratory disease incidence in the Asbestos Workers Registry cohort, Ontario, Canada, 1986-2019 (n = 26 164). *Non-malignant disease follow-up included 25 503 workers. All models adjusted for age, birth year, and sex. Squares show point estimates from indicator model and are positioned at each category’s median exposure value. Squares are jittered for visibility. Dashed vertical lines show category cut points. Regression lines are from linear-quadratic model and exclude follow-up exceeding the 99th percentile of exposure. Confidence intervals (95%) shown as vertical bars and shaded ribbons. Cases shown with tick marks along plot rug. Plot truncated at 2500 h (~95th percentile) for visibility.
Using G-POSH, Table 3 presents the estimated impact of interventions limiting cumulative asbestos work hours on the occurrence of respiratory disease. Figure 2 presents plots of the cumulative incidence of disease by age under each intervention and the observed occurrence of disease under current policy. Under an intervention limiting a worker’s exposure to 2000 cumulative asbestos work hours, a total of 37 asbestosis, 15 pulmonary fibrosis, 18 mesothelioma, and 33 lung cancer cases would have been prevented in the entire cohort with complete follow-up to 2019 (or age 85 years) compared to the observed disease experience under current policy. This corresponds to a reduction in the risk of disease at the end of follow-up of 23% for asbestosis, 8% for pulmonary fibrosis, 19% for mesothelioma, and 6% for lung cancer relative to the observed risk, and between 19 538 and 23 396 disease-free person-years of life gained in the cohort for each disease. Under each examined cumulative asbestos exposure limit, percent reductions in the risk of disease were most pronounced for asbestosis followed by mesothelioma. For all diseases, the greatest number of cases were prevented with an exposure limit of 100 cumulative asbestos work hours, with a total of 76 asbestosis, 36 pulmonary fibrosis, 27 mesothelioma, and 79 lung cancer cases prevented by the end of follow-up compared to the observed.
Table 3.
Estimated impact of interventions limiting cumulative asbestos work hours on the occurrence of respiratory disease using G-POSH, Asbestos Workers Registry cohort, Ontario, Canada, 1986-2019.
| Exposure limit, no. of cumulative asbestos work hours | No. of observed cases | No. of expected cases | No. of prevented cases | No. of disease-free person-years of life gained | Risk ratio a | 95% CI b |
|---|---|---|---|---|---|---|
| Full cohort (n = 26 164) | ||||||
| Asbestosisc | ||||||
| Current policy (observed) | 159 | 1 | ||||
| 2000 | 122 | 37 | 19 538 | 0.77 | 0.70-0.86 | |
| 500 | 105 | 54 | 19 752 | 0.66 | 0.59-0.74 | |
| 250 | 99 | 60 | 19 990 | 0.62 | 0.54-0.70 | |
| 100 | 83 | 76 | 19 907 | 0.52 | 0.43-0.63 | |
| Pulmonary fibrosisc | ||||||
| Current policy (observed) | 192 | 1 | ||||
| 2000 | 177 | 15 | 19 099 | 0.92 | 0.87-0.97 | |
| 500 | 164 | 28 | 19 174 | 0.85 | 0.79-0.92 | |
| 250 | 169 | 23 | 19 312 | 0.88 | 0.81-0.96 | |
| 100 | 156 | 36 | 19 161 | 0.81 | 0.72-0.90 | |
| Mesothelioma | ||||||
| Current policy (observed) | 97 | 1 | ||||
| 2000 | 79 | 18 | 23 396 | 0.81 | 0.73-0.90 | |
| 500 | 75 | 22 | 23 732 | 0.77 | 0.66-0.89 | |
| 250 | 75 | 22 | 23 990 | 0.77 | 0.64-0.89 | |
| 100 | 70 | 27 | 23 610 | 0.72 | 0.55-0.86 | |
| Lung cancer | ||||||
| Current policy (observed) | 528 | 1 | ||||
| 2000 | 495 | 33 | 20 618 | 0.94 | 0.91-0.96 | |
| 500 | 484 | 44 | 21 002 | 0.92 | 0.86-0.96 | |
| 250 | 474 | 54 | 21 320 | 0.90 | 0.83-0.95 | |
| 100 | 449 | 79 | 21 072 | 0.85 | 0.78-0.92 | |
| Construction workers (n = 16 320) | ||||||
| Asbestosisc | ||||||
| Current policy (observed) | 103 | 1 | ||||
| 2000 | 72 | 31 | 12 984 | 0.70 | 0.61-0.81 | |
| 500 | 63 | 40 | 12 946 | 0.61 | 0.48-0.73 | |
| 250 | 65 | 38 | 13 084 | 0.63 | 0.51-0.77 | |
| 100 | 53 | 50 | 12 804 | 0.51 | 0.36-0.68 | |
| Pulmonary fibrosisc | ||||||
| Current policy (observed) | 97 | 1 | ||||
| 2000 | 82 | 15 | 12 724 | 0.85 | 0.76-0.93 | |
| 500 | 79 | 18 | 12 622 | 0.81 | 0.69-0.96 | |
| 250 | 87 | 10 | 12 746 | 0.90 | 0.77-1.03 | |
| 100 | 76 | 21 | 12 404 | 0.78 | 0.60-0.97 | |
| Mesothelioma | ||||||
| Current policy (observed) | 59 | 1 | ||||
| 2000 | 45 | 14 | 15 709 | 0.76 | 0.63-0.94 | |
| 500 | 43 | 16 | 15 635 | 0.73 | 0.54-0.92 | |
| 250 | 43 | 16 | 15 819 | 0.73 | 0.50-0.91 | |
| 100 | 42 | 17 | 14 977 | 0.71 | 0.48-0.97 | |
| Lung cancer | ||||||
| Current policy (observed) | 279 | 1 | ||||
| 2000 | 260 | 19 | 14 172 | 0.93 | 0.88-0.99 | |
| 500 | 253 | 26 | 14 150 | 0.91 | 0.82-0.97 | |
| 250 | 248 | 31 | 14 383 | 0.89 | 0.81-0.98 | |
| 100 | 224 | 55 | 13 666 | 0.80 | 0.68-0.90 |
Risk ratio calculated at end of follow-up and closely approximates the IRR.
CI estimated from analyses of 200 bootstrapped resamples of original data.
Nonmalignant disease follow-up included 15 962 workers in construction and 25 514 workers in the full cohort with follow-up from 1999 to 2019.
Figure 2.
Cumulative incidence of respiratory disease by age under interventions limiting cumulative asbestos work hours compared to current policy, Asbestos Workers Registry cohort, Ontario, Canada, 1986-2019 (n = 26 164). *Follow-up for asbestosis and pulmonary fibrosis included 25 514 workers from 1999 to 2019. †The predicted natural course under no work hour intervention is nearly equivalent to the observed disease occurrence in the cohort. See Figure S1 for model validation results.
Analyses restricted to workers employed in the construction industry (Table 3) showed greater percent reductions in risk under higher exposure limits. Under the 2000-h limit, the risk ratio for both asbestosis and pulmonary fibrosis was lower among construction workers compared to that among the overall cohort. The number of prevented cases of these diseases was the same or similar among construction workers compared to the overall cohort despite them consisting of only two-thirds of the study population. With decreasing exposure limits, the relative benefit in the impact of reduced exposure for construction workers versus the overall cohort attenuated.
Discussion
In the Ontario AWR, the current policy of notifying workers about the need for medical screening after 2000 reported hours of work with asbestos is too late to prevent many respiratory disease cases. Our estimated exposure-response relationships showed that the incidence rate of all 4 examined asbestos-related diseases increased while workers had accrued only 100-500 asbestos work hours. After accruing 2000 work hours, observed increases were substantial, especially when considering that the referent group consisted of workers who were also exposed to asbestos work, with no one in the cohort having zero exposure. Under hypothetical interventions limiting asbestos work hours at various thresholds, we estimated that a substantial number of cases of asbestosis, pulmonary fibrosis, mesothelioma, and lung cancer could be prevented and disease-free years of life gained compared to the observed disease risk. The estimated risk of asbestosis was reduced by nearly half under an intervention limiting asbestos work to 100 h. Even limiting work to 2000 h led to estimates that imply substantial impacts on prevention of all 4 diseases, with a stronger benefit among construction workers. Although the interventions we evaluated assume the cessation of asbestos work, which is not part of the current notification policy, our results indicate that workers must be warned about their exposure and medically monitored for disease much sooner than the current 2000-h threshold for notification.
The Ministry’s current policy of notifying workers of the need for a medical examination at 2000 h has been in place since at least 1997 and has not been updated in nearly 30 years.22 While the notification policy does not force a worker to cease asbestos work operations as implied by the exposure interventions we evaluated, it may promote earlier medical screening, the uptake of improved exposure controls in the workplace, or the worker’s self-imposed or medically recommended removal from hazardous work. Less than half (47%) of the cohort reached 100 h of cumulative asbestos work hours, and only 10% reached 2000 h, suggesting that a lower notification threshold is plausible to implement. In all our analyses, we used asbestos work hours in type 2 and 3 operations as a proxy for inhalation of asbestos fibers, and our positive exposure-response results indicate that this proxy was valid. For many workers in the cohort, asbestos work hours entailed the inhalation of unsafe levels of asbestos fibers during work operations, suggesting that employers may have failed to ensure adequate controls were in place to protect workers from exposure.
Considering the complete AWR of approximately 40 000 workers is larger than the cohort enumerated for inclusion in the current analysis, the number of disease cases preventable through earlier intervention in the target population is even greater. For example, our hypothetical intervention limiting asbestos work hours at 100 was estimated to lead to 76 fewer cases of asbestosis, and 79 fewer cases of lung cancer, among the 26 164 workers in the cohort (Table 3). This implies an estimate of approximately 3 asbestosis cases and 3 lung cancer cases prevented for every 1000 workers under a 100-h limit compared to the observed occurrence of disease in the AWR.
Nearly two-thirds of the AWR cohort consists of workers in the construction industry, and our findings of a strongly positive association between asbestos work and respiratory disease is consistent with other epidemiologic studies in this population. Results from the Occupational Disease Surveillance System (ODSS) cohort in Ontario, which included 2.18 million workers in diverse industries and over 300 000 workers in construction trades occupations followed over a 33-year period, showed that construction trades workers had adjusted hazard rates that were 3.6 times the reference rate for asbestosis and 2.4 times that for mesothelioma, with a substantial disease burden of nearly 500 cases diagnosed.17 Similar findings documenting the disproportionate burden of asbestos-related disease among construction workers have been reported in Sweden.23 Workers employed by specialty trade contractors are estimated to have the highest prevalence of occupational asbestos exposure, where exposure to asbestos is common through the removal and remediation of asbestos-containing materials and the maintenance and renovation of aging buildings and structures. In our study, those employed in the construction industry had a greater distribution of reported asbestos work hours than other industries, which may explain the greater benefit of interventions limiting exposure at 2000-hours in this group.
In occupational health, decision-makers are often required to compare the expected benefit of a policy limiting exposure with the cost of its implementation. G-methods can be used to generate evidence expressed in terms of cumulative risks under well-defined exposure scenarios rather than modeled effect estimates, with such evidence informing decision-making by estimating the impact of interventions on the prevention of disease and prolonging of life. A similar approach to G-POSH has been used in other substantive areas of epidemiology to evaluate the effect of fixed or dynamic baseline exposure interventions.8 The G-POSH approach can be applied to any time-varying exposure in time-to-event analyses, which are common in occupational cohort studies, and emphasizes simple implementation, graphical illustrations, and intuitive results for ease of communication and uptake by policy decision-makers. The approach can readily be implemented in R using our G-POSH package on GitHub (Appendix S2).
The use of G-methods for evaluating hypothetical exposure interventions is not new, particularly within occupational epidemiology where the parametric G-formula has been used in analyses of several cohorts. One appeal of using the parametric G-formula is the ability to control for time-varying confounding affected by prior exposure (and specifically healthy worker survivor bias in occupational cohort analyses, which the method was first described to address),13 with examples for acrylonitrile,24 silica,25 arsenic,26 radon,27 diesel,28 metalworking fluids,9 and asbestos.10 Although controlling healthy worker survivor bias is an important strength of such analyses, it is possible to use G-methods to obtain estimates of intervention effects even in settings where there is little or no healthy worker survivor bias.21 In the current study, both standard regression analyses and our G-POSH analysis found strong positive associations between asbestos work hours and asbestosis and mesothelioma incidence (Table 2), and in Table S3 we further examined G-POSH results incorporating inverse probability of censoring weights to address healthy worker survivor bias.12 In settings where healthy worker survivor bias is a concern, an extension of the G-POSH approach allows an investigator to address such bias when considering policy interventions that can be expressed in terms of an annual limit on exposure (eg, a worker never exceeds 50 asbestos work hours in a year). In Appendix S2 we provide illustrative code and a sample calculation to extend G-POSH to use inverse probability of censoring weighting. Like any inverse probability weighting method, attention to model specification and weight distribution is important, as extreme weights can signal subsets of data where positivity assumptions are challenging, and results rely heavily on model extrapolation.
One important limitation of our study is the reliance on employer-reported asbestos work hours in assessing exposure. Employers are suspected to have underreported the number of asbestos work hours among their workers, which could have introduced bias in our results due to exposure misclassification. This misclassification is likely nondifferential by disease status and is expected to have created a mostly downward bias toward the null in exposure-response analyses, as workers misclassified in the lowest exposure category used as the reference group would have a greater disease rate than their correctly classified counterparts. This misclassification is in addition to any created by the substantial proportion of missing work hours (13% of work reports) that we chose to impute, although results from a sensitivity analysis excluding 4270 workers with any missing asbestos work hours showed that associations remained positive and largely unchanged (Table S1). Underreporting of asbestos work hours could be an important factor explaining the low work hour limits needed to prevent substantial proportions of disease if the true work hours are in fact much greater than those reported to the registry.
Another limitation of our study is our inability to directly control for potential confounding by tobacco smoking on results for lung cancer. Smoking could be a confounder of estimates for lung cancer if smoking is associated with cumulative asbestos work hours. The direction of this potential confounding is unknown, as the distribution of smoking by asbestos work hours in the study population is also unknown although not expected to substantially vary. Given that joint exposure to smoking and asbestos has a very strong synergistic, multiplicative causal effect on lung cancer risk,29,30 and we did not observe an exceptionally elevated rate of lung cancer in the highest exposure group, the likelihood of strong positive confounding in our study is low. Tobacco smoking does not cause mesothelioma or asbestosis and is not a significant cause of pulmonary fibrosis. Therefore, smoking is not expected to confound results for these other 3 diseases.
Other potential sources of bias in our study pertain to medical surveillance and selection into the cohort. Increased medical surveillance was possible for workers who received notification recommending screening upon reaching 2000 reported work hours. As shown in the plot rug in Figure 1, there appeared to be a clustering of cases diagnosed with all 4 diseases after accumulating work hours at and above the notification threshold. If workers were more likely to be diagnosed with disease after being notified, it may have introduced an upward bias in effect estimates in the ≥2000-h exposure category. However, such bias would only be significant if the increased medical surveillance led to respiratory disease diagnoses that would otherwise have gone undetected until death. With respect to selection bias, 34% of individual records in the AWR were excluded from the analytic cohort due to missing or erroneous data or failing to probabilistically link to a registered health insurance number in Ontario. The potential impact on bias of these exclusions is difficult to estimate without additional information, although excluding out-of-province workers may have reduced bias in our results if this group would have had a greater probability of being lost to follow-up due to emigration to their jurisdiction of origin.
In conclusion, we found compelling evidence of positive exposure-response associations between cumulative asbestos work hours and incidence of all 4 examined asbestos-related diseases. We derived estimates of the potential impact of asbestos work restriction and exposure elimination on disease prevention. Despite the recent federal ban in Canada, asbestos will remain in older products, buildings, and structures in Ontario for the long term. Regulatory agencies should intervene sooner to prevent a substantial number of diseases cases among workers in the AWR.
Supplementary Material
Acknowledgments
We thank Dr. Stephen Bertke, Dr. Alex Keil, Dr. Sharef Danho, and Dr. Leon Genesove for their contributions to the work.
Contributor Information
Nathan L DeBono, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada; Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada.
Louis Everest, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada.
David B Richardson, Department of Environmental and Occupational Health, Wen School of Population and Public Health, University of California, Irvine, Irvine, CA, United States.
Colin Berriault, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada.
Ryann E Yeo, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada.
Maya A Meeds, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada; Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada.
Victoria Arrandale, Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada.
Paul A Demers, Occupational Cancer Research Centre, Ontario Health, Toronto, ON, Canada; Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada.
Supplementary material
Supplementary material is available at American Journal of Epidemiology online.
Funding
This work was funded by a grant from the Ontario Ministry of Labour, Immigration, Training, and Skills Development through the Research Opportunities Program.
Conflict of interest
The authors declare no conflicts of interest.
References
- 1. Pefoyo AJK, Genesove L, Moore K, et al. Exploring the usefulness of occupational exposure registries for surveillance the case of the Ontario Asbestos Workers Registry (1986-2012). J Occup Environ Med. 2014;56(10):1100-1110. 10.1097/JOM.0000000000000235 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Arrandale VH, Berriault C, Song C, et al. Surveillance of asbestos related disease among workers enrolled in an exposure registry. Am J Ind Med. 2024;67(12):1077-1088. 10.1002/ajim.23668 [DOI] [PubMed] [Google Scholar]
- 3. International Agency for Research on Cancer . IARC Monographs on the Evaluation of Carcinogenic Hazards to Humans: Arsenic, Metals, Fibres, and Dusts (Vol 100C); 2012. https://publications.iarc.fr/120 [PMC free article] [PubMed] [Google Scholar]
- 4. American Thoracic Society . Diagnosis and initial management of nonmalignant diseases related to asbestos. Am J Respir Crit Care Med. 2004;170(6):691-715. 10.1164/rccm.200310-1436ST [DOI] [PubMed] [Google Scholar]
- 5. Ruff K. How Canada changed from exporting asbestos to banning asbestos: the challenges that had to be overcome. Int J Environ Res Public Health. 2017;14(10):1-9. 10.3390/ijerph14101135 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Health Canada . Asbestos and Your Health. 2023. Accessed June 27, 2024. https://www.canada.ca/en/health-canada/services/air-quality/indoor-air-contaminants/health-risks-asbestos.html [Google Scholar]
- 7. CAREX . Asbestos occupational exposures. 2024. Accessed June 17, 2024. https://carexcanada.ca/profile/asbestos-occupational-exposures/
- 8. Breskin A, Edmonds A, Cole SR, et al. G-computation for policy-relevant effects of interventions on time-to-event outcomes. Int J Epidemiol. 2021;49(6):2021-2029. 10.1093/ije/dyaa156 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Picciotto S, Chevrier J, Balmes J, et al. Hypothetical interventions to limit metalworking fluid exposures and their effects on COPD mortality: G-estimation within a public health framework. Epidemiology. 2014;25(3):436-443. 10.1097/EDE.0000000000000082 [DOI] [PubMed] [Google Scholar]
- 10. Cole SR, Richardson DB, Chu H, et al. Analysis of occupational asbestos exposure and lung cancer mortality using the G formula. Am J Epidemiol. 2013;177(9):989-996. 10.1093/aje/kws343 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Richardson DB, DeBono NL, Berriault C, et al. Innovations in applied decision theory for health and safety. Occup Environ Med. 2020;77(8):520-526. 10.1136/oemed-2019-106303 [DOI] [PubMed] [Google Scholar]
- 12. Keil AP, Li Y, Lan Q, et al. Inverse probability weighting to estimate impacts of hypothetical occupational limits on radon exposure to reduce lung cancer. Am J Epidemiol. 2024;194(5). 10.1093/aje/kwae299 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Robins J. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Math Model. 1986;7(9):1393-1512. 10.1016/0270-0255(86)90088-6 [DOI] [Google Scholar]
- 14. Ministry of Labour Immigration Training and Skills Development . A guide to the regulation respecting asbestos on construction projects and in buildings and repair operations: classification of work. 2022. Accessed May 30, 2024. https://www.ontario.ca/document/guide-regulation-respecting-asbestos-construction-projects-and-buildings-and-repair-8
- 15. Zhang Y, Rajaram N, Lau A, et al. Silicosis, asbestosis, and pulmonary fibrosis in Ontario, Canada from 1996 to 2019. Am J Ind Med. 2023;66(8):670-678. 10.1002/ajim.23504 [DOI] [PubMed] [Google Scholar]
- 16. Gan WQ, Demers PA, McLeod CB, et al. Population-based asbestosis surveillance in British Columbia. Occup Environ Med. 2009;66(11):766-771. 10.1136/oem.2008.045211 [DOI] [PubMed] [Google Scholar]
- 17. DeBono NL, Warden H, Logar-Henderson C, et al. Incidence of mesothelioma and asbestosis by occupation in a diverse workforce. Am J Ind Med. 2021;64(6):476-487. 10.1002/ajim.23245 [DOI] [PubMed] [Google Scholar]
- 18. Keil AP, Edwards JK, Richardson DB, et al. The parametric g-formula for time-to-event data: intuition and a worked example. Epidemiology. 2014;25(6):889-897. 10.1097/EDE.0000000000000160 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Chevrier J, Picciotto S, Eisen EA. A comparison of standard methods with g-estimation of accelerated failure-time models to address the healthy-worker survivor effect: application in a cohort of autoworkers exposed to metalworking fluids. Epidemiology. 2012;23(2):212-219. 10.1097/EDE.0b013e318245fc06 [DOI] [PubMed] [Google Scholar]
- 20. Rudolph JE, Cartus A, Bodnar LM, et al. The role of the natural course in causal analysis. Am J Epidemiol. 2022;191(2):341-348. 10.1093/aje/kwab248 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Naimi AI, Cole SR, Hudgens MG, et al. Assessing the component associations of the healthy worker survivor bias: occupational asbestos exposure and lung cancer mortality. Ann Epidemiol. 2013;23(6):334-341. 10.1016/j.annepidem.2013.03.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Ontario Ministry of Labour . Designated substances in the workplace: a guide to the regulation respecting asbestos on construction projects and in buildings and repair operations. 1997. Accessed June 25, 2024. https://archive.org/details/31761116533407
- 23. Järvholm B, Englund A. The impact of asbestos exposure in Swedish construction workers. Am J Ind Med. 2014;57(1):49-55. 10.1002/ajim.22264 [DOI] [PubMed] [Google Scholar]
- 24. Keil A, Haber G, Graubard B, et al. Estimating impacts of reducing acrylonitrile exposure on lung cancer mortality in an occupational cohort with the parametric g-formula. Occup Environ Med. 2024;81(5):225-231. 10.1136/oemed-2024-109532 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Keil AP, Richardson DB, Westreich D, et al. Estimating the impact of changes to occupational standards for silica exposure on lung cancer mortality. Epidemiology. 2018;29(5):658-665. 10.1097/EDE.0000000000000867 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Keil AP, Richardson DB. Reassessing the link between airborne arsenic exposure among Anaconda copper smelter workers and multiple causes of death using the parametric g-formula. Environ Health Perspect. 2017;125(4):608-614. 10.1289/EHP438 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Edwards JK, McGrath LJ, Buckley JP, et al. Occupational radon exposure and lung cancer mortality: estimating intervention effects using the parametric g-formula. Epidemiology. 2014;25(6):829-834. 10.1097/EDE.0000000000000164 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Neophytou AM, Picciotto S, Costello S, et al. Occupational diesel exposure, duration of employment, and lung cancer: an application of the parametric G-formula. Epidemiology. 2016;27(1):21-28. 10.1097/EDE.0000000000000389 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Klebe S, Leigh J, Henderson DW, et al. Asbestos, smoking and lung cancer: an update. Int J Environ Res Public Health. 2019;17(1):1-23. 10.3390/ijerph17010258 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Olsson AC, Vermeulen R, Schüz J, et al. Exposure-response analyses of asbestos and lung cancer subtypes in a pooled analysis of case-control studies. Epidemiology. 2017;28(2):288-299. 10.1097/EDE.0000000000000604 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.