Article Text
Abstract
Objectives Animal bioassays have demonstrated convincing evidence of the potential carcinogenicity to humans of titanium dioxide (TiO2), but limitations in cohort studies have been identified, among which is the healthy worker survivor effect (HWSE). We aimed to address this bias in a pooled study of four cohorts of TiO2 workers.
Methods We reanalysed data on respirable TiO2 dust exposure and lung cancer mortality among 7341 male workers employed in TiO2 production in Finland, France, UK and Italy using the parametric g-formula, considering three hypothetical interventions: setting annual exposures at 2.4 (U.S. occupational exposure limit), 0.3 (German limit) and 0 mg/m3 for 25 and 35 years.
Results The HWSE was evidenced. Taking this into account, we observed a positive association between lagged cumulative exposure to TiO2 and lung cancer mortality. The estimated number of lung cancer deaths at each age group decreased across increasingly stringent intervention levels. At age 70 years, the estimated number of lung cancer deaths expected in the cohort after 35-year exposure was 293 for exposure set at 2.4 mg/m3, 235 for exposure set at 0.3 mg/m3, and 211 for exposure set at 0 mg/m3.
Conclusion This analysis shows that HWSE can hide an exposure–response relationship. It also shows that TiO2 epidemiological data could demonstrate an exposure–effects relationship if analysed appropriately. More epidemiological studies and similar reanalyses of existing cohort studies are warranted to corroborate the human carcinogenicity of TiO2. This human evidence, when combined with the animal evidence, strengthens the overall evidence of carcinogenicity of TiO2.
- Occupational Health
- Lung Diseases, Interstitial
- Statistics as Topic
- Longitudinal studies
- Dust
Data availability statement
Data sharing not applicable as no datasets generated and/or analysed for this study. Not applicable.
Statistics from Altmetric.com
Key messages
What is already known about this subject?
Titanium dioxide (TiO2) is classified by the International Agency for Research on Cancer as possibly carcinogenic to humans. In 2020, the European Chemicals Agency classified TiO2 under all forms as suspected human carcinogen by inhalation. Prior large cohorts of TiO2 workers reported increased mortality from lung cancer but failed to observe an exposure–response relationship with cumulative exposure to TiO2, except in a French cohort of TiO2 workers. A concern of potential healthy worker survivor effect (HWSE) has been raised.
What are the new findings?
We reanalysed data from the European cohort of TiO2 workers and found strong evidence of the HWSE. Taking this into account, a positive exposure–response relationship with 10 years lagged cumulative exposure to TiO2 was observed. The estimated number of lung cancer deaths at each age group decreased across increasingly stringent exposure limits.
How might this impact on policy or clinical practice in the foreseeable future?
This analysis shows that HWSE can hide exposure–response relationship. It also shows that TiO2 epidemiological data could demonstrate an exposure–effects relationship if analysed appropriately. This human evidence, when combined with the animal evidence, strengthens the overall evidence of carcinogenicity of TiO2.
Introduction
There is substantial interest in human evidence regarding the carcinogenicity of titanium dioxide (TiO2), an odourless white pigment and opacifying agent widely used since the 1920s. TiO2 is classified as possibly carcinogenic to humans by the International Agency for Research on Cancer based on sufficient evidence from cancer bioassay studies and inadequate evidence from human cancer studies. Since 2020, TiO2 is also classified as a suspected human carcinogen by inhalation in the European Union according to Regulation no 1272/2008.
Prior large cohorts of TiO2 workers reported increased mortality from lung cancer, but failed to observe an exposure–response relationship with cumulative exposure to TiO2,1 except in a French cohort of TiO2 workers.2 One of the key limitations noted in the occupational cohorts was the potential healthy worker survivor effect (HWSE), which can mask association between cumulative exposure and lung cancer mortality.3 For instance, in the pooled European study of TiO2 workers, a key study on this topic, no evidence of an association between respirable TiO2 exposure and lung cancer mortality was observed despite the excess of lung cancer mortality among male TiO2 workers as compared with the general population (standardised mortality ratio=1.23; 95% CI 1.10 to 1.38).4
In this study, we reanalysed a subset of the pooled European cohort of TiO2 workers,4 to examine the evidence of HWSE and the exposure–response relationship between cumulative exposure to TiO2 and lung cancer mortality. We implemented the g-computation algorithm formula (g-formula) recommended for statistical analysis of cohort data in the presence of time-varying confounders affected by prior exposure, typical of HWSE.5 The g-formula has been proven to be an essential method for estimating human health effects of exposures and interventions on exposures in such scenarios.5 Therefore, we applied it to assess the effect of three hypothetical interventions of TiO2 exposure limitation.
Methods
Study population
The original pooled European cohort included workers who had been employed at least 1 month in 1 of 11 TiO2 production factories in six European countries (Finland, France, Italy, Norway, Germany and the UK).4 All female workers and male workers with missing death certificates and/or lacking quantitative exposure estimates were excluded from the exposure–response analysis of this cohort.4 We used the same criteria of worker inclusion/exclusion as in original cohort, but restricted the study to four countries (Finland, France, Italy and the UK), for which data were still available and ethical approvals obtained.
Exposure assessment
The factories produced mainly pigment-grade TiO2, although TiO2 form (eg, particle size and crystalline phase) was ignored.6 Information on demographic and employment characteristics was collected from factories’ records describing date of birth, sex, race, and dates of hire, job or department change and termination. Estimated cumulative occupational exposure to respirable TiO2 dust was derived from job title and work history.6 Exposure assessments were carried out at the level of occupational titles for each plant for discrete time periods throughout the history of plant operations. Lists of occupational titles were compiled and coded for each factory. Exposure measurement data were obtained from company files along with information such as the area(s) of the plant where measurements were made, the presence of any local ventilation, the type of the materials being handled and the purpose of sampling. These were then linked to the work history of each individual in the cohort to provide exposure estimates.
Statistical analysis
The lengths of follow-up varied between countries and ranged from 1950 to 1972 until 1997–2001. The primary outcome of interest was death for which the underlying cause was attributed to cancers of the trachea, bronchus and lung (International Classification of Diseases, 9th revision code 162).
A data tabulation of person-periods and events was constructed with one record for each person-year of observation from date of entry into the analysis until end of follow-up or administrative censoring of workers alive at age 90 years. Using the observed data, we fitted logistic regression models for the probability of the outcome of interest, for the probability of remaining at work, and for the probability of dying from a competing cause, as a function of covariates and estimated exposure. The cumulative TiO2 exposure was 10-year lagged (online supplemental material figure S1 and technical appendix 1).
Supplemental material
The g-formula was implemented by a Monte Carlo simulation based on the regression model estimates of the probability of termination of employment and death.5 7 Ten Monte Carlo samples per exposure scenario were drawn randomly from the observed cohort and the estimated parameters from the parametric models to recreate the study data for each person in the sample under specified exposure intervention. Three hypothetical interventions were considered: setting workers’ annual exposures to 2.4 mg/m3, 0.3 mg/m3 (the currently recommended TiO2 occupational exposure limits in the US8 and Germany,9 respectively) and zero exposure. For each intervention, we assumed two possible exposure durations: 25 and 35 years and estimated the expected lung cancer mortality at 60, 70, 80 and 90 years of age. The associated 95% CIs were calculated using bootstrap samples.
Results
The cohort included 7341 workers (online supplemental table S2). At the end of the follow-up, 139 lung cancer deaths were observed. The presence of the HWSE was evident (online supplemental figure S1). Being in employment reduced the risk of lung cancer mortality (OR 0.14, 95% CI 0.08 to 0.22) and the probability of leaving the employment increased as a function of TiO2 exposure (OR 3.55 95% CI 2.82 to 4.46)). The OR of lung cancer death associated with lagged cumulative exposure to TiO2 was estimated at 1.03 per 1 mg/m3-year (95% CI 0.99 to 1.07), after adjustment for the employment status in previous and current years and employment duration.
G-estimates of lung cancer mortality, derived under the three hypothetical interventions, are shown in table 1. The estimated number of lung cancer deaths at 60, 70, 80 and 90 years of age all decrease across the three interventions considered and for both exposure durations.
Discussion
This reanalysis provides the first evidence of an exposure–response relationship between TiO2 cumulative exposure and lung cancer mortality using the parametric g-formula. Adjustment of a standard regression model for employment status or exposure duration is not sufficient for complete HWSE correction.3 However, rank ordering of lung cancer deaths across levels of the intervention estimated by g-formula is consistent with a positive exposure–response association between TiO2 and lung cancer (3% per 1 mg/m3-year of respiratory TiO2).
A limitation of the g-formula is the g-null paradox. The g-formula may be guaranteed some degree of model misspecification if there is treatment-confounder feedback and the sharp causal null hypothesis (ie, the intervention has no effect on any individual’s outcome at any time) holds.10 To address concern that model misspecification may lead to bias, we followed recent guidance11 to avoid overly parsimonious models for the components of the g-formula in this analysis.
A causal interpretation of the findings requires certain statistical assumptions, including consistency, positivity, exchangeability.12 The consistency assumption may be challenging given the complexity of historical exposure conditions and the various TiO2forms, the latter being insufficiently documented. We nevertheless believe that exposure contrasts are defined here well enough to support meaningful inference regarding TiO2’s effect. The positivity hypothesis (ie, observations on exposed and unexposed workers through covariate levels) was difficult to confirm as 82% of workers were exposed. However, the positive exposure–response relationship based on continuous cumulative lagged exposure to TiO2 supports these assumptions.
The exchangeability assumption (ie, no unmeasured confounding) is challenging given the limitations of available data on smoking. The most complete smoking data were available for the French cohort and showed no effect of adjustment for smoking on estimates of TiO2-lung cancer mortality associations.2 Moreover, we were able to assess some other occupational coexposures that are suspected or known lung carcinogens, including asbestos, welding fumes and other mineral dusts; adjustment for these exposures had no effect on the association with TiO2.
A confounder typically of concern in occupational cohort mortality studies is the HWSE.3 In the current study, this source of confounding was addressed by the g-formula. In prior analyses of these data using standard multivariable regression method, the association was likely masked by the HWSE and potentially the exposure misclassification in the early years of follow-up. Prior to undertaking the g-formula analysis, we assessed the most important HWSE components in these data and confirmed their presence. Moreover, prior reports of SMR analyses indicated patterns consistent with the HWSE3 and in the analysis of the French cohort, duration of employment was negatively associated with lung cancer mortality.2 These underscore the relevance of g-methods in these cohorts and future investigations of TiO2’s effects on human health.
The estimates with overlapping CI are likely due to the limited statistical power in this study, which might be addressed by additional follow-up of these cohorts. More epidemiological studies and similar reanalyses of updated existing cohort studies are warranted to corroborate the human carcinogenicity of TiO2. This human evidence, when combined with the animal evidence, strengthens the overall evidence of carcinogenicity of TiO2.
Data availability statement
Data sharing not applicable as no datasets generated and/or analysed for this study. Not applicable.
Ethics statements
Patient consent for publication
Ethics approval
The study was authorised by the IARC ethics Committee (IEC project number 18-32). The use of the data for the French cohort was approved by the French Data Protection Authority (CNIL), Authorisation No 999250. The use of the data for the UK cohort was approved by the EPS ethics committee of Heriot-Watt University (approval number 19/EA/JC/1). The use of the data for the Italian cohort was approved by the Comitato Etico Interaziendale A.S.L Citta di Torino (approval number CS2/1250). The use of the data for the Finnish cohort was approved by Statistics Finland (approval number TK-53-988-19). Retrospective mortality study.
Acknowledgments
The authors acknowledge Statistics Finland as a source of mortality data for the Finnish cohort.
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Twitter @DamienMcElvenny
Contributors IGC designed and conducted this study and drafted the manuscript, AG-G, DR and PW conducted statistical analyses. KS and MS-B centralised the data, facilitated data access, and obtained IARC ethical approval. SC, SF-F and CM contributed to study coordination and French ethical approval. All authors discussed the study methodology, read the manuscript, critically reviewed it and agreed on the final version.
Funding This work was supported by the ANSES, Grant number ANSES/IST 2017-CRD-18.
Disclaimer Where authors are identified as personnel of the International Agency for Research on Cancer/WHO, the authors alone are responsible for the views expressed in this article and they do not necessarily represent the decisions, policy or views of the International Agency for Research on Cancer/WHO.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.