Objectives Titanium dioxide (TiO2) is widely used in construction, food, cosmetic and medical industry. The current evidence on TiO2 carcinogenicity in humans is considered inadequate. As French participants of the European cohort of TiO2 workers exhibited an increase in mortality from lung cancer, we aimed at investigating whether TiO2 exposure, co-exposures or smoking can explain this increase.
Methods We reanalysed the data of 833 French male workers (follow-up period 1968–1997) and used multiple imputation to complete their smoking status. We considered respirable TiO2 dust as primary exposure of interest, estimated as continuous cumulative (mg/m3-year) and annual average (mg/m3) concentrations and binary and 4-class categorical variables, with cut-off values of 0.3 and 2.4 mg/m3 (the German and American occupational exposure limits, respectively). For each exposure metric, we estimated HRs and associated 95% CIs, using Cox regression models adjusted for calendar period, exposure duration and smoking.
Results The fully adjusted model yielded a HR=3.7 (95% CI=0.79 to 17.95) for TiO2-exposed workers vs unexposed and a HR=27.33 (95% CI=4.35 to 171.84) for those exposed to >2.4 mg/m3 as annual average concentration. Employment duration was negatively related with lung cancer mortality, therefore cumulative exposure had a small effect on mortality (HR=1.03 (95% CI=0.99 to 1.08) per mg/m3-year).
Conclusion This study suggests a positive relationship between TiO2 exposure and lung cancer mortality in TiO2 workers, whatever the exposure variable used, despite a limited statistical power in some models. The results question the current evidence on TiO2 carcinogenicity in humans but need to be confirmed in other cohorts, using different statistical approaches.
- mortality studies
Data availability statement
Data may be obtained from a third party and are not publicly available.
Statistics from Altmetric.com
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.
What is already known about this subject?
Since 2006, titanium dioxide (TiO2) is classified as an International Agency for Research on Cancer group 2B carcinogen (possibly carcinogenic to humans). In 2018, the European Chemicals Agency classified TiO2 under all forms as a category-2 carcinogen (suspected human carcinogen) by inhalation, given the inadequate evidence of its carcinogenicity in humans. Two large cohorts of American and European TiO2 workers reported statistically increased mortality from lung cancer, but failed to observe a dose–response relationship with cumulative exposure to TiO2. Moreover, none of the analyses was adjusted for smoking.
What are the new findings?
We re-analysed the data of a French cohort of TiO2 workers, using four different TiO2 exposure variables and multiple imputations for missing data on tobacco smoking. We showed that relationship between the duration of exposure and lung cancer mortality is negative and masks the effect of TiO2 cumulative exposure, which remains of borderline significance, event after adjustment for smoking. Yet, relationship between annual average exposure and lung cancer mortality is consistent.
How might this impact on policy or clinical practice in the foreseeable future?
If confirmed, these finding should trigger reconsidering the evidence on TiO2 carcinogenicity in humans, possibly upgrading the TiO2 classification.
Titanium dioxide (TiO2) is an odourless white pigment and opacifying agent widely used since 1920.1 TiO2 is classified as possibly carcinogenic to humans by International Agency for Research on Cancer and, since 2018, as suspected human carcinogen by inhalation by the European Chemical Agency.1 Statistically increased lung cancer mortality was reported in two cohorts of American and European TiO2 workers.2 3
In the American cohort, the association between TiO2 exposure and cause-specific mortality was first investigated by Poisson regression,3 and, in accordance with the European analysis,2 using Cox regression.4 The authors reported a statistically significant increase in HR for all cancer mortality but no results for lung cancer mortality.4 In the European study pooling cohorts from six countries,2 solely French participants experienced borderline significant increases in both all cancer (Standardized Mortality Ratio(SMR)=1.21 (95% CI=1.01 to 1.44)) and lung cancer mortality (SMR=1.42 (95% CI=0.99 to 1.96)), but the dose–response results were reported solely for the pooled cohort and were not adjusted for smoking.2
This study aimed at investigating the association between TiO2 exposure and lung cancer mortality among French participants of the European cohort, accounting for smoking and other potential confounders.
Study sample, follow-up and main outcome
We re-analysed the French data used in the dose–response analyses of the European cohort study.2 The included workers were men employed before 1990, for at least 1 year in a French TiO2 factory and had sufficiently complete data on demographic characteristics and on occupational history. Workers employed only in non-production jobs, and those for whom 25% of occupational history or more than 5 years were missing were excluded. Vital status and causes of death were assessed by the National Natural Persons Identification Index and the National Cause of Death Registry, respectively. The latter contains anonymised records of all deaths in France since 1968 and their causes. Death records were matched to cohort members by date of birth, gender, and date and place of death. The causes of death were coded according to the ninth revision of the International Classification of Diseases (ICD-9). As primary outcome, we considered lung cancer deaths specified as underlying cause on the death certificate (ICD-9 code 162) that occurred over the period 1968–1997.
We re-used the exposure data collected in the context of the European study.5 An occupational hygienist visited each factory, including the French one and collected information on exposure circumstances and files with relevant measurement data. All data were entered into a computer database and converted to a common exposure metric using appropriate conversion factors.5 Exposure levels were then estimated for respirable TiO2 dust, sulphuric acid mist, hydrochloric acid, asbestos and welding fumes at the level of occupational titles for discrete time-periods throughout the factory’s operational history. Estimated exposure levels were then linked to the occupational history of each workers to provide estimates of cumulative occupational exposure.
Initially, the factory nurse checked the medical records of workers present in the last 5 years and abstracted information on smoking status whenever available.2 To expand the usage of available data, we used multiple imputation method for the missing smoking data. We created a multinomial logistic regression imputation model with a possible outcome of one for non-smoker, two for ex-smoker and three for smoker using the following variables: the year of birth, the years of employment, the age at the start of employment, the age at the end of follow-up, the total duration of employment, the different types of work performed and their respective durations. Using this model and the existing data on smoking, we run 1000 smoking status imputations for workers with missing data. The different Cox models (see below), adjusted on the smoking categories were run for each of the 1000 imputations. The estimates were then combined using Royston’s method.6
We considered respirable TiO2 dust as a binary variable (yes/no), then by cumulative exposure lagged by 10 years (mg/m3-year) and by duration of employment in years; we computed the ratio of these two variables to calculate the annual average exposure (mg/m3). We also considered the annual average exposure as a 4-class categorical variables (unexposed, 0–0.3, 0.3–2.4 and >2.4 mg/m3), with cut-off values determined by the currently recommended occupational exposure limits in Germany and the USA.7 8 All other exposures were treated as co-exposures. We implemented several Cox regression models (with age as the main time variable). Models were first adjusted for calendar period (before 1975 then by 5-year periods until 1997), then for employment duration and if relevant, for the co-exposure. To control for potential confounding by smoking, we first ran all models without adjustment on smoking, second with the adjustment using the multiple imputed smoking categories. Sensitivity analysis consisted of testing 0, 5 and 15-year lag on the cumulative TiO2 exposure to account for a latency period. Data were analysed using Stata V.15.
The cohort comprised 833 male workers (17 390 person-years) among whom 75% were exposed to TiO2 (online supplementary tables S1 and S2). Other dusts and sulfuric acid were the most common co-exposures. However, none of the co-exposure was associated with TiO2 exposure and the outcome (results not shown) and therefore included in multivariate models. Smoking status was known for 61% of workers, 5% of whom smoked (online supplementary table S1). At the end of follow-up, 13% of workers were deceased, with 16 lung cancer deaths in total. Compared to unexposed, TiO2-exposed workers exhibited an approximately fourfold higher risk of lung cancer mortality, though statistically non-significant (table 1). The analysis according to the annual average exposure showed a significant increase in lung cancer mortality per mg/m3 of respirable TiO2 dust exposure. The adjustment for exposure duration decreased the HRs, while adjustment for smoking slightly increased them. The fully adjusted model resulted in a HR=2.07 (95% CI=1.34 to 3.20), that is, an approximately twofold increased risk of lung cancer mortality per increment of one mg/m3 of respirable TiO2 dust exposure as annual average concentration. In contrast, the exposure duration was negatively related to the outcome (data not shown). Therefore, the analysis according to cumulative exposure to respirable TiO2 dust resulted only in a small increase of lung cancer mortality (2% to 4% per mg/m3-year of respirable TiO2 dust exposure, depending on the lag duration (table 1), though of borderline statistical significance. Smoking appeared not to confound any of these associations.
This study is based on a small cohort of French TiO2-exposed workers with available TiO2 exposure data, and only a few lung cancer deaths, especially among non-exposed workers. We included 833 workers and not the 2255 French workers reported in the Boffeta et al study.2 It should be noted that Boffeta et al actually included the same number of French workers in their Cox regression analysis as us (n=833). They excluded 1412 French male workers with unknown or unspecified job group and ten workers lost to follow-up. This calls for a careful result interpretation. However, a positive exposure–response relationship based on continuous annual average exposure found in this study, seems quite precise and robust to adjustment for confounders. This result is consistent with results based on all other TiO2 exposure variables used, despite imprecision of some HRs and questions the previously suggested absence of a relationship between TiO2 exposure and lung cancer in humans.
Adjustment for smoking had no effect on these results. However, the prevalence of smoking in our cohort seems unusually low. Insufficient quality of the smoking data due to potential under-reporting of smoking by workers during the 90s is likely to explain this, although we cannot verify this assumption.
Previous studies focused on the cumulative exposure as continuous or categorical variable, and observed no statistically significant trend with lung cancer incidence and/or mortality. However, here, we observed that the exposure duration is negatively associated with the outcome. As shown in online supplementary table S3, the most heavily exposed workers left their employment much earlier than their less exposed colleagues did, suggesting the presence of a healthy worker survivor effect (HWSE). HWSE impedes dose–response analysis with cumulative exposure variables using Cox regression and therefore undermines the relevance of the results.9 Therefore, the HRs should be interpreted with caution.
Today, only experimental studies documented the carcinogenic potential of TiO2.1 10 The latter was systematically observed in highly exposed rats, as result of ‘particle lung overload’ that exceeds the particle clearance ability of alveolar macrophages.11 Given the anatomic differences between rats and humans, and lack of evidence from previous epidemiological studies, the relevance of the carcinogenic effect of TiO2 at high doses for humans remains controversial. Recent experimental studies elucidated TiO2 inflammogenic potential and capacity to elicit pulmonary toxicity with contribution from neutrophils and macrophages independent of the overload phenomenon.11 A secondary genotoxic mechanism of TiO2, through oxidative stress, was recently evidenced for TiO2 nanoparticles.12 However, the nuclear transfer of TiO2 nanoparticles suggests the likelihood of primary genotoxicity, through direct interaction with DNA.12
Although, no physiochemical characterisation of TiO2 was possible in this study, we assume the workers have handled essentially pigment-grade microparticulated TiO2. Our results should be confirmed in other national cohorts of TiO2 workers, using different statistical approaches. In particular, G-methods, enabling correction for the HWSE, should be implemented as a next analytical step for the European cohort of TiO2 workers along with a better physical-chemical characterisation of TiO2 in future studies.
Data availability statement
Data may be obtained from a third party and are not publicly available.
The use of the individual data was approved by the French Data Protection Authority (CNIL), Authorisation No 999250.
The authors thank S. Charles and C. Michel from the French Agency for Food, Environmental and Occupational Health & Safety (ANSES) for their help in accessing the data.
Contributors IGC designed and conducted this study and drafted the manuscript, AG-G and PW conducted statistical analyses; KS and DL provided access to the data. 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 University of Lausanne.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.