Article Text

Download PDFPDF

Welding fumes and lung cancer: a meta-analysis of case-control and cohort studies
Free
  1. Manoj Kumar Honaryar1,2,3,
  2. Ruth M Lunn4,
  3. Danièle Luce5,
  4. Wolfgang Ahrens6,
  5. Andrea ’t Mannetje7,
  6. Johnni Hansen8,
  7. Liacine Bouaoun1,
  8. Dana Loomis1,9,
  9. Graham Byrnes1,
  10. Nadia Vilahur1,10,
  11. Leslie Stayner11,
  12. Neela Guha1,12
  1. 1 International Agency for Research on Cancer, Lyon, France
  2. 2 Service de médcine et addictologie, Centre Hospitalier des Quatre Villes (CH4V), Saint-Cloud et Sèvres, France
  3. 3 Ecole des hautes études en santé publique (EHESP), Paris, France
  4. 4 National Institute of Environmental Health Sciences, Durham, North Carolina, USA
  5. 5 Univ Rennes, Inserm, EHESP, Irset (Institut de recherche en santé, environnement et travail) - UMR_S 1085, Pointe-à-Pitre, France
  6. 6 Leibniz Institute for Prevention Research and Epidemiology - BIPS, Bremen, Germany
  7. 7 Centre for Public Health Research, Massey University, Wellington, New Zealand
  8. 8 Danish Cancer Society Research Center, København, Denmark
  9. 9 University of Nevada, Reno, Reno, Nevada, USA
  10. 10 European Commission, Italy
  11. 11 Division of Epidemiology and Biostatistics, University of Illinois, School of Public Health, Chicago, Illinois, USA
  12. 12 Office of Environmental Health Hazard Assessment, California Environmental Protection Agency, Oakland, California, USA
  1. Correspondence to Dr Neela Guha, California Environmental Protection Agency; neela{at}berkeley.edu

Abstract

Background An estimated 110 million workers are exposed to welding fumes worldwide. Welding fumes are classified by the International Agency for Research on Cancer as carcinogenic to humans (group 1), based on sufficient evidence of lung cancer from epidemiological studies.

Objective To conduct a meta-analysis of case-control and cohort studies on welding or exposure to welding fumes and risk of lung cancer, accounting for confounding by exposure to asbestos and tobacco smoking.

Methods The literature was searched comprehensively in PubMed, reference lists of relevant publications and additional databases. Overlapping populations were removed. Meta-relative risks (mRRs) were calculated using random effects models. Publication bias was assessed using funnel plot, Eggers’s test and Begg’s test.

Results Forty-five studies met the inclusion criteria (20 case-control, 25 cohort/nested case-control), which reduced to 37 when overlapping study populations were removed. For ‘ever’ compared with ‘never’ being a welder or exposed to welding fumes, mRRs and 95% CIs were 1.29 (1.20 to 1.39; I2=26.4%; 22 studies) for cohort studies, 1.87 (1.53 to 2.29; I2=44.1%; 15 studies) for case-control studies and 1.17 (1.04 to 1.38; I2=41.2%) for 8 case-control studies that adjusted for smoking and asbestos exposure. The mRRs were 1.32 (95% CI 1.20 to 1.45; I2=6.3%; 15 studies) among ‘shipyard welders’, 1.44 (95% CI 1.07 to 1.95; I2=35.8%; 3 studies) for ‘mild steel welders’ and 1.38 (95% CI 0.89 to 2.13; I2=68.1%; 5 studies) among ‘stainless steel welders’. Increased risks persisted regardless of time period, geographic location, study design, occupational setting, exposure assessment method and histological subtype.

Conclusions These results support the conclusion that exposure to welding fumes increases the risk of lung cancer, regardless of the type of steel welded, the welding method (arc vs gas welding) and independent of exposure to asbestos or tobacco smoking.

  • meta-analysis
  • meta-regression
  • lung cancer
  • welding
  • welding fumes
  • mild steel welding
  • stainless steel welding
  • arc welding
  • gas welding
  • exposure-effect analysis

Statistics from Altmetric.com

Key messages

What is already known about this subject?

  • Welders have been reported to experience an increased risk of lung cancer in numerous epidemiological studies.

  • Welders are exposed to fumes, which contain several carcinogens classified by the International Agency for Research on Cancer (IARC).

  • Several studies have also reported the risk of lung cancer to increase with increasing exposure to welding fumes.

What are the new findings?

  • Since the 1990 IARC evaluation of welding fumes as ‘possibly carcinogenic to humans’ (group 2B), a substantial body of literature has been published.

  • This led to welding fumes being classified by IARC as ‘carcinogenic to humans’ (group 1) in a re-evaluation in 2017.

How might this impact on policy or clinical practice in the foreseeable future?

  • Given that there are an estimated 110 million workers worldwide exposed to welding fumes, preventing exposure is of direct relevance for policy.

  • Measures to avoid or at least reduce exposure to welding fumes for improved protection of worker health can be informed by our analysis of exposure-effect data.

  • Recognising welding fumes as carcinogenic can facilitate the compensation of related occupational cancers.

  • This meta-analysis, combined with data on exposure prevalence, can inform the calculation of the burden of cancers attributed to welding fumes.

Background

Welding is the process of joining metals through coalescence. Approximately 11 million persons work worldwide with the job title of welder, and a further 110 million people are exposed to welding-related occupational exposures (eg, by doing welding activities as part of other job, or by being exposed to welding in indoor spaces as bystanders).1 In 1989, the International Agency for Research on Cancer (IARC) classified welding fumes as possibly carcinogenic to humans (group 2B) with limited evidence of carcinogenicity from epidemiological studies reporting an increased risk of lung cancer.2 3 Since then, >20 epidemiological studies have been published on the association between welding/welding fumes and lung cancer. In March 2017, IARC convened a Working Group to systematically review all of the published literature to date and classified welding fumes as carcinogenic to humans (group 1) based on sufficient evidence of lung cancer in humans.1

Welding fumes are generated when metals are heated above their melting point and then vaporise and condense into very fine solid particulates.1 Welders are exposed to a complex mixture of chemical compounds that might vary by the type of welding method used (eg, gas, arc), the type of metal being welded (mild or stainless steel) and the occupational setting where welding is performed (eg, shipyards where asbestos was historically used in ship insulation). Nickel compounds and chromium, well-established lung carcinogens in humans,4 are constituents of stainless steel whereas they exist in much lower concentrations in mild-steel. Furthermore, mild-steel welding, commonly performed with high-emission techniques, generates higher mass concentrations of particulate matter than stainless steel welding.5 6

Asbestos exposure and tobacco smoking, established risk factors for lung cancer,4 7 may also be associated with welding. Welders have been reported to smoke more than the general population.8–10 In the past, welders may have been exposed to asbestos at shipyards or through its use in different types of insulation and heat-protective material such as gloves, blankets, etc.4 11

Here, we present the results of a meta-analysis that was conducted in parallel to the IARC Monograph 118 re-evaluation of the carcinogenicity of welding fumes, with the objectives to explore sources of heterogeneity between studies, quantify the magnitude of lung cancer risk and explore exposure-effect associations.

Methods

This meta-analysis has been conducted and reported based on the Meta-analysis of Observational Studies in Epidemiology guidelines.12 The protocol has not been registered. Each study was assessed individually for its features, including strengths and limitations as well as the exposure assessment method and its effect on the results; further detail is available in the final published IARC Monograph.1

Literature search strategy and study selection

A comprehensive electronic search of the literature until 28 March 2017 was performed (by NG, RL, DL, WA, MKH) using PubMed with the following search terms: (‘welding’[MeSH Terms] OR welder[Text Word] OR ‘welding fume*'[Text Word]) AND (‘lung neoplasms’[Mesh] OR neoplasms OR cancer OR carcino* OR tumour) AND(‘epidemiology’[Mesh] OR ‘epidemiologic studies’[Mesh] OR ‘cohort studies’[Mesh] OR ‘case-control studies’[Mesh] OR case-referent OR cohort). The reference lists of relevant publications were additionally screened. All of the articles were identified through PubMed and the reference lists of relevant publications; no additional publications were identified through additional searches of Web of Science and Google Scholar databases.

The inclusion and exclusion criteria were defined before the search was initiated and agreed to in full consensus by the IARC Monograph 118 Working Group. Studies were included if they were: i) peer-reviewed publications of observational epidemiological studies of case-control, nested case-control or cohort designs and ii) reported a relative risk estimate (risk ratio [RR], OR, HR, SIR or SMR) with corresponding 95% CIs for the association between welding or exposure to welding fumes and lung cancer or provided enough data to calculate this.

Studies were excluded if: i) they did not meet the inclusion criteria, ii) only reported risk estimates for welding-related occupational categories too broad to attribute cancer risks to welding fumes like pipefitters, plumbers or metal-workers, iii) used no reference group (eg, case-case studies that did not report a comparable estimate of relative risk between welders and non-welders), iv) overlapped subsequent publications using the same data, including individual studies with analyses that were superseded by a pooled analysis.11 13

Details of the literature search strategy are shown in online supplementary figure 1 and the excluded studies are shown in online supplementary table 1.

Supplemental material

Data extraction

Literature searches and data extraction were organised using specialised software to ensure transparency and reproducibility.14 The included studies were organised using tags for keywords in Health Assessment Workplace Collaborative, a comprehensive tool for classifying research articles, for conducting systematic reviews (by NG, MKH). Study characteristics, detailed population description (eg, sample size, definition of cases and controls, matching factors for control groups, duration of follow-up for cohort studies, exposure assessment methods), risk estimates with 95% CIs, information on covariates adjusted for and the number of exposed cases were extracted into the IARC Table Builder software (by RL, DL, MKH, WA).

Statistical analysis

Data from the publications were formatted before conducting the meta-analysis. If 95% CIs were not reported, they were calculated using the openEpi online software15 and the mid-p Exact method.16 Similarly, risk estimates were calculated from the raw data by the Working Group (identified in square brackets), when not provided in the original publication.1

If a study did not report a single estimate of overall association, multiple estimates from that study were combined using fixed effects (FE) models (which assumes one true effect is shared and common to all strata considered in a single study).17 18 For example, a single FE estimate was calculated by Gustavsson et al,19 who reported adjusted risk estimates for welding exposure in quartiles and Pukkala et al,20 who reported separate risk estimates for men and women.

In cohort studies, we compared the observed versus expected morbidity from lung cancer and from mesothelioma to indirectly assess potential confounding from tobacco smoking and asbestos exposure, respectively.

STATA software (StataCorp, College Station, Texas, USA) V.14.2 was used for statistical analyses. Summary meta-relative risk (mRR) estimates with their respective 95% CIs were computed using a random effects (RE) meta-analysis model comparing ‘welders’ with ‘never’ being a welder or exposed to welding fumes across studies. The weight assigned to each study was inversely proportional to the within-study sampling variances in the estimates.21 Forest plots were used to present the results graphically.22

As the OR calculated from case-control studies approximates the relative risk for low-incidence (ie, rare) diseases,23 summary mRR estimates were computed for the results of case-control and cohort studies combined, as well as for each study design separately (table 2).

Table 2

Results of the meta-analysis and sensitivity analyses for the risk of lung cancer and welders or exposure to welding fumes

The I2 statistic was calculated to quantify the amount of inconsistency between studies; it estimates the percentage of total variation across studies that is due to heterogeneity rather than chance. The I2 statistic is used to test the hypothesis of heterogeneity across studies where the null hypothesis (H0) of no statistical heterogeneity of point estimates (~observed effects) among studies is investigated against the alternative hypothesis (H1) that the effect estimates (OR or RR) vary across studies, having adjusted for other factors included within the model. I2 values ranging from 0% to 25% were considered to represent low heterogeneity, from 26% to 50% as moderate heterogeneity and above 50% as substantial heterogeneity.24

To investigate publication bias,25 funnel plots,26 Egger’s and Begg’s tests (at a 10% significance level) were used.27 It should be noted that the Egger’s test may yield false positive results if fewer than 10 studies are included.27 28

In each analysis, priority was given to the pooled analyses by Simonato et al 13 and Kendzia et al,11 which included several studies that were not published separately (a description of these studies is included in tables 2.4 and 2.6 of IARC Monograph 118). For the primary analysis, studies reporting risk estimates for ‘ever’ compared with ‘never’ being a welder or exposed to welding fumes were included, and when multiple estimates were provided in a publication, RRs from the most informative adjusted models (such as those including tobacco smoking and asbestos exposure) were selected. Several sensitivity analyses were conducted to explore which factors contribute to heterogeneity in the mRR estimate (eg, type of material welded, region, others).

The following STATA commands were used: 1) metan to produce mRR with the random, fixed and graph options to consider the RE model, FE model and obtain forest plots, respectively29; 2) metabias to assess publication bias, with egger and begg options30 and 3) metafunnel command for producing funnel plots.31

Meta regression

RE weighted meta-regression was performed using the metareg STATA package32 to examine the simultaneous association between effect sizes and participant and study characteristics, as well as adjustments for covariates in a multivariable model. The explanatory variables included study design, exposure assessment methods, region, adjustment for tobacco smoking and exposure to asbestos as well as an interaction term between adjustment for smoking and asbestos.

Exposure-effect analysis

We performed an exposure-effect meta-analysis to investigate the association between exposure to welding fumes and lung cancer risk. Analyses were conducted using the dosresmeta R package developed by Orsini et al. 33 This method accounts for the correlation between RR estimates for exposure categories from the same study.

Nine studies reported exposure-effect data through duration of employment as welders (N=6),11 34–38 time since first exposure (n=2)13 39 and cumulative exposure which is based on both the duration and the intensity (n=1).40 The exposure-effect meta-analysis focused on the duration of employment as welders, due to the availability of data.

Among the six studies using duration of employment, only three11 37 38 reported on either the amount of person-time or the total number of subjects available for each exposure category. When we included the study by Danielsen et al,37 the analysis failed to work (probably due to an estimate of zero found for one exposure category) and therefore the exposure-effect meta-analysis was only based on the two remaining studies, which reported multiple exposure levels.

To model the exposure-effect relationship of welding fumes and the risk of lung cancer using data from the only two studies available for this meta-analysis, the choice of models for the exposure on the outcome were: linear, log-linear and a flexible cubic spline. The model that assumed a log-linear exposure-effect association between exposure and lung cancer risk and the model that assumed a more flexible exposure-effect using cubic splines of exposure were compared using a Wald-type test of the hypothesis of deviation from the linear trend.

Results

Twenty-four cohorts, 1 nested case-control study and 20 case-control studies were eligible for the meta-analysis. This represented a total of 16 485 328 participants from the cohort studies, and 137 624 cases and 364 555 controls from the case-control studies. The characteristics of these studies and how they were included in sensitivity analyses are captured in table 1.

Table 1

Description of the studies included in the meta-analysis of welding fumes and risk of lung cancer

The mRR estimate for lung cancer for ‘ever’ compared with ‘never’ being a welder or exposed to welding fumes was 1.43 (95% CI 1.31 to 1.55; I2=54.6%; n=37). The mRR estimate was 1.17 (95% CI 1.04 to 1.38; I2=41.2%; n=8) for studies that adjusted for smoking and asbestos exposure simultaneously (figure 1, table 2). Mild steel welders (mRR, 1.44; 95% CI 1.07 to 1.95; I2=63.8%; n=3) had approximately the same magnitude of lung cancer risk as stainless steel welders (mRR, 1.38; 95% CI 0.89 to 2.13; I2=68.1%). Risk estimates for exclusively gas welding (mRR, 1.71; 95% CI 1.10 to 2.66; I2=52.3% n=3) were higher than for exclusively arc welding (mRR, 1.36; 95% CI 0.70 to 2.67; I2=83.0%; n=2). The magnitude of the mRRs did not vary significantly by the histological subtype of lung cancer, nor by exposure assessment method (table 2).

Figure 1

Forest plot of primary (overall) analysis for risk of lung cancer comparing ’ever' vs ’never' welders or exposed to welding fumes.

Sources of heterogeneity in the overall mRR of 1.43 were identified, notably study design and control source in case-control studies, occupational setting and geographical region (table 2). The overall mRR for the cohort studies was 1.29 (95% CI 1.20 to 1.39; I2=26.4%; n=22) and 1.10 (95% CI 1.06 to 1.14; I2=0%; n=6) in the cohorts with adjustment for smoking. There was less heterogeneity in the industrial cohorts (I2=6.7%; n=16) compared with the population-based cohorts (I2=59.7%; n=6). The mRR for case-control studies was 1.87 (95% CI 1.53 to 2.29; I2=44.1%; n=15), for hospital-based 1.84 (95% CI 1.36 to 2.49; I2=0.0%; n=8) and 2.03 (95% CI 1.61 to 2.57; I2=13.9%; n=5) for population-based case-control studies. The mRR was similar for European studies (n=18) and for North American studies (n=16), although the heterogeneity was higher in the later (I²=73.7%). Too few studies were available from other regions to present a combined estimate.

A meta-regression of exposure to welding fumes and risk of lung cancer was also conducted to further explore sources of heterogeneity (table 3). Without any covariates in the meta-regression model (that represented the adjustments made in individual studies), a high amount of inconsistency between studies was observed (I²=55%). Inclusion of study design and smoking adjustment reduced this to 14%, while additional covariates (asbestos, region, exposure assessment method) or their interactions did not reduce the I² any further.

Table 3

Pooled random-effects model-based summary relative risk estimates and 95% CIs of risk of lung cancer associated with exposure to welding fumes

Exposure-effect analyses (figure 2) using both non-linear (spline) and non-parametric fits for duration of employment as a welder showed statistically significant associations with lung cancer risk (p=0.001 and 0.009, respectively).

Figure 2

Dose-response association between duration of employment as a welder and lung cancer risk. Duration of employment was introduced using restricted cubic splines in the random-effect model. The solid line represents the smooth curve while the dotted line represents the linear trend. Dashed lines represent the 95% CIs of the spline curve. Circles and triangles below or above the curve represent the positions of the study-specific relative risks. The value of 1 year (of employment duration) served as referent. The relative risks are plotted on the log-scale.

Publication bias was assessed using a funnel plot, which was relatively symmetric but suggested an enrichment with smaller studies with positive findings (online supplementary figure 2). The result for the Egger’s test was 0.58 (95% CI −0.037 to 1.19; p=0.064) and for the Begg’s test (adjusted Kendall score=141; p=0.017), both suggesting asymmetry of the funnel plot.

Discussion

This meta-analysis complements the recent qualitative systematic review that was conducted by the IARC Monograph 118 Working Group of the association between occupation as a welder (a proxy for exposure to welding fumes) and risk of lung cancer.1 Results from this meta-analysis, in line with a previously published meta-analysis,41 contribute to strengthen the rationale for classifying welding fumes as ‘carcinogenic to humans’ with regard to lung cancer, help identify sources of heterogeneity in the published literature and allow quantification of the magnitude of this association, which is of public health relevance for decision-making purposes.

Welders presented, on average, a 43% increased risk of lung cancer when compared with those classified as ‘never’ being welders or exposed to welding fumes. Additionally, the risk increased with years of employment as a welder. Results remained significant when restricting to a subset of 8 occupational cohort studies that adjusted for both tobacco smoking and asbestos exposure, as well as in the 20 studies (both cohort and case-control) that had adjusted for smoking only. Thus, tobacco smoking and asbestos exposure are unlikely to account for all the observed excess risk of lung cancer observed among welders. Excess lung cancer was observed among welders in cohorts with low or minimal asbestos exposure36 42 and in almost all of the studies that adjusted for asbestos exposure, including those with detailed and high-quality exposure assessment for asbestos. Similarly, internal analyses which found positive associations within groups of welders may have indirectly adjusted for exposure to asbestos.

Increased risks of lung cancer were observed in both case-control and cohort studies, regardless of the material welded or the welding method. Therefore, exposure to chromium and nickel compounds, which are well-established lung carcinogens present in much higher concentrations in stainless steel compared with mild steel, did not completely explain the total increased lung cancer risk found in welders. The mRR for mild steel welders was slightly higher than that for stainless steel welders, although the CIs overlapped. Mild steel is commonly welded with high emission techniques that generate higher mass concentrations of particulate matter than welding stainless steel, which could explain the difference in risk estimates.6 Exposure misclassification could be another possible explanation for the difference in risk estimates, as some of the mild steel welders could have been exposed to stainless steel welding fumes from another worksite; however, the misclassification would be expected to be non-differential and bias risk estimates towards the null. Several studies with detailed assessment of welding tasks tended to report lower risk estimates for exclusively arc welding43 44 as compared with exclusively gas welding.43–45

Study design was a significant source of heterogeneity, with the hospital-based case-control studies (n=8) and the industrial cohorts (n=16) representing the most homogenous subset of results (I2=0.0% and I2=6.7%, respectively). Stratifying by study design accounted for much of the heterogeneity when the meta-analyses were restricted to studies that adjusted for smoking, method of exposure assessment and material welded.

A limitation of this meta-analysis was the inability to provide estimates for lung cancer risk by quantitative level of exposure to welding fumes—a useful measure for informing policies for protecting workers’ health. Due to lack of data, we calculated an exposure-effect meta-analysis based only on two studies; the wide CIs limit the interpretation of the data but are shown in order to inform health impact assessments. Due to limited information, we were also not able to distinguish welding processes in greater detail (eg, flux-core arc welding, gas metal arc welding and gas tungsten arc welding). Most studies reported the job title or job task with no detailed information on welding methods. Ever ‘welder’ was a crude exposure category and defined differently across studies by duration of employment (eg, worked at least for 3 months). In some studies, the reference group was not precisely defined and therefore could affect comparability between studies for obtaining a mRR in the meta-analysis.

Some studies used a specific occupational group as reference (eg, other metal workers who may share some exposures causing lung cancer with welders) rather than a population-based reference group; this would result in an underestimation of the risk. Publication bias could have also resulted in overestimation of mRR for ‘ever’ welders.30 46

This review has several strengths. A large number of studies from 1954 to 2017 were included: in total, 16 485 328 participants from the cohort studies and 137 624 cases and 364 555 controls from case-control studies. Factors that could explain heterogeneity in the literature were examined in depth through stratified meta-analyses and meta-regression, and the direction and significance of the results remained robust to multiple stratified analyses. To overcome search and selection biases the complete set of relevant and standard keywords were selected47 and the inclusion criteria were clearly defined before initiating the bibliographic search.

Gas and arc welding inherently produce fumes. Welding fumes were often not measured directly in the reviewed studies but assessed indirectly through indicator variables such as welding process, welding materials, branch of industry, job title, expert assessment or self-report. Since the association with lung cancer was observed for both gas and arc welding and could not be explained by other exposures that occur with these two predominant welding procedures, the IARC Monograph 118 Working Group concluded that increased risk of lung cancer among welders can be causally attributed to exposure to welding fumes. Given that welding is often performed as part of jobs not classified as welder or that bystanders in the same work environment are also exposed, the IARC classification of welding fumes has the potential to inform public health actions for a larger number of workers than those classified occupationally as welders.

Conclusion

In conclusion, this meta-analysis, in addition to the comprehensive qualitative assessment of the IARC Working Group, supports the rationale for the IARC Monograph 118 Working Group’s classification that welding fumes are ‘carcinogenic to humans’ (group 1). There was a significantly increased risk of lung cancer, independent of main confounders (tobacco smoking and exposure to asbestos) and regardless of the welding method (arc vs gas) and the type of metal welded, which increased with duration of welding. The risk estimates from this meta-analysis combined with country-specific estimates of the number of welders worldwide could be used to inform on the burden of lung cancer cases attributed to exposure to welding fumes and design evidence-based preventive public health measures worldwide.

Acknowledgments

The authors would like to thank Dr Nilmara Oliveira Alves Brito for her contributions to this project.

References

Footnotes

  • Contributors NG designed the analysis plan in collaboration with all other coauthors. MKH, RL, DL, WA extracted data from original manuscripts. MKH, LB, GB conducted the data analysis. All authors made substantial contributions to the conception or design of the work; or the acquisition, analysis or interpretation of data for the work; AND drafting the work or revising it critically for important intellectual content; AND final approval of the version to be published; AND agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Patient consent for publication Not required.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.