Registration Dossier
Registration Dossier
Data platform availability banner - registered substances factsheets
Please be aware that this old REACH registration data factsheet is no longer maintained; it remains frozen as of 19th May 2023.
The new ECHA CHEM database has been released by ECHA, and it now contains all REACH registration data. There are more details on the transition of ECHA's published data to ECHA CHEM here.
Diss Factsheets
Use of this information is subject to copyright laws and may require the permission of the owner of the information, as described in the ECHA Legal Notice.
EC number: 236-675-5 | CAS number: 13463-67-7
- Life Cycle description
- Uses advised against
- Endpoint summary
- Appearance / physical state / colour
- Melting point / freezing point
- Boiling point
- Density
- Particle size distribution (Granulometry)
- Vapour pressure
- Partition coefficient
- Water solubility
- Solubility in organic solvents / fat solubility
- Surface tension
- Flash point
- Auto flammability
- Flammability
- Explosiveness
- Oxidising properties
- Oxidation reduction potential
- Stability in organic solvents and identity of relevant degradation products
- Storage stability and reactivity towards container material
- Stability: thermal, sunlight, metals
- pH
- Dissociation constant
- Viscosity
- Additional physico-chemical information
- Additional physico-chemical properties of nanomaterials
- Nanomaterial agglomeration / aggregation
- Nanomaterial crystalline phase
- Nanomaterial crystallite and grain size
- Nanomaterial aspect ratio / shape
- Nanomaterial specific surface area
- Nanomaterial Zeta potential
- Nanomaterial surface chemistry
- Nanomaterial dustiness
- Nanomaterial porosity
- Nanomaterial pour density
- Nanomaterial photocatalytic activity
- Nanomaterial radical formation potential
- Nanomaterial catalytic activity
- Endpoint summary
- Stability
- Biodegradation
- Bioaccumulation
- Transport and distribution
- Environmental data
- Additional information on environmental fate and behaviour
- Ecotoxicological Summary
- Aquatic toxicity
- Endpoint summary
- Short-term toxicity to fish
- Long-term toxicity to fish
- Short-term toxicity to aquatic invertebrates
- Long-term toxicity to aquatic invertebrates
- Toxicity to aquatic algae and cyanobacteria
- Toxicity to aquatic plants other than algae
- Toxicity to microorganisms
- Endocrine disrupter testing in aquatic vertebrates – in vivo
- Toxicity to other aquatic organisms
- Sediment toxicity
- Terrestrial toxicity
- Biological effects monitoring
- Biotransformation and kinetics
- Additional ecotoxological information
- Toxicological Summary
- Toxicokinetics, metabolism and distribution
- Acute Toxicity
- Irritation / corrosion
- Sensitisation
- Repeated dose toxicity
- Genetic toxicity
- Carcinogenicity
- Toxicity to reproduction
- Specific investigations
- Exposure related observations in humans
- Toxic effects on livestock and pets
- Additional toxicological data
![](https://poisoncentres.echa.europa.eu/o/diss-blank-theme/images/factsheets/A-REACH/factsheet/print_toxicological-information.png)
Endpoint summary
Administrative data
Description of key information
Additional information
- It is stated that dose-response results were reported by Boffetta et al. (2004) solely for the pooled cohort. However, some country specific information on the Cox regression analyses is included in Table 4.4 of the full report of the study (Boffetta et al., 2003). Note that the reference category in this analysis included workers with cumulative exposure < 0.73 mg/m3-year (4 lung cancer deaths). However, the analysis like that for the pooled cohort was not adjusted for smoking as noted by Guseva Canu,Gaillen-Guedy et al. (2020).
- Annual average exposure (mg/m3) was stated to have been calculated as the ratio of cumulative respirable TiO2 dust lagged by 10 years (mg/m3-year) and duration of employment in years. However, this can’t be correct as only 5 lung cancer deaths were exposed in the 10-year lag analysis, but 14 lung cancer deaths had average exposure > 0.0 mg/m3 in Tables 1 and S2.
- There were stated to be only 43 (8.4%) current smokers among the 512 workers (61.5%) for whom smoking information was available. The percentage of current smokers does not seem consistent with the information reported previously. Boffetta et al. (2004) reported that 42% of French workers were current smokers comparable to national data (41%). Boffetta et al. (2003) further reported that 37% were current smokers in assessments made after 1994 compared to a national figure of 39% (Table 5.1). These figures were calculated for all French workers with smoking assessment information. However, Table 1.15 of Boffetta et al. (2003) states that smoking assessments were made for 545 French workers, which suggests that virtually all the French workers with smoking status information were from factory 5, and that the current smoking prevalence cannot have been much different to than 42%. Annex 4 states that only a sample of 10% of the medical records for factory 12 was checked by the occupational physician, and smoking status was available for 55% of the whole sample. Boffetta et al. (2003) reported that the 545 workers with smoking assessments comprised 57.3% of the French cohort, but it appears that only the 10% sample of factory 12 was included in the calculation, and not the full cohort of 2346 workers (2255 male and 91 female workers). For factory 5 Annex 4 stated that medical records of workers present in the last 5 years were checked and information on smoking status and number of cigarettes per day was abstracted whenever available. Information on daily consumption was stated to be available for 227 French workers (page 25), and this is also consistent with 42% of 545 French workers with smoking assessments being smokers.
- The number of exposed workers is unclear. It is stated in the text that 75% of workers were ever exposed to TiO2 (consistent with the total of 631 in Table S1), and there were 14 lung cancer deaths among exposed workers according to Table 1. However, Table S3 includes only 536 exposed workers with 14 lung cancer deaths. The difference can’t be due to using a 10-year lag to calculate average exposure as only 5 lung cancer deaths were exposed in 10-year lag analyses (see Table 1).
- It is noted that the mean duration of exposure of workers in the group with the highest average exposures was 5.48 years, which was approximately 10 years less than that of workers in the two other groups with lower average exposures. It is claimed that this suggests the presence of HWSE, but it is more likely due to the fact that mean exposures in the plant fell sharply over time, especially after 1980. Consequently, although several workers may have had an average exposure > 2.4 mg/m3 during the initial period, their average exposure would have fallen sharply after 1980 unless they had left employment. From Table S2 it can be seen that only surface treatment workers would have been likely to have had an average exposure > 2.4 mg/m3 at any point. A HWSE effect is also not suggested by the fact that the mean duration of employment of exposed workers was 15.27 years compared to 12.26 for unexposed workers (Table S3).
- The reanalysis of the French cohort was justified by the excess of lung cancer observed among French workers. However, we do not know whether there was an excess of lung cancer in factory 5 (only the combined SMR for factory 5 and factory 12 was reported).
- Boffetta et al. (2003) noted that although a number of measurements were available from factories 4, 5 and 14, they did not have sufficient information to meet our selection criteria, with most commonly the sampling period not being known. Consequently, no adjustment of the exposure estimates derived from reconstructed exposure from factory 5 could be made based on measurement data.
Statement of Overall Conclusions
The rat is uniquely sensitive to the formation of lung tumours when exposed under conditions of particle overload to titanium dioxide and other poorly soluble low-toxicity particles (Levy, 1995). Although particle overload is observed in other experimental species, such as the mouse, it is only in the rat that a sequence of events is initiated that leads to fibroproliferative disease, septal fibrosis, hyperplasia and eventually lung tumours. Similar pathological changes are not observed in other common laboratory rodents, in non-human primates, or in exposed humans. In addition, detailed epidemiological investigations have shown no causative link between titanium dioxide exposure and cancer risk in humans. At workplace exposure concentrations, no lung cancer hazard has been observed. Thus, a carcinogen rating for titanium dioxide is not warranted.
Epidemiological studies on occupationally exposed workers
Cohort studies
Chen and Fayerweather (1988) examined mortality and cancer incidence in a cohort of 1576 male workers exposed to titanium dioxide for more than a year in two titanium dioxide production plants in the United States. Mortality due to cancer was significantly lower than expected on the basis of national rates, and mortality due to lung cancer (9 deaths) was also significantly reduced (SMR=0.52; 95% CI 0.24-0.99). Nested case-control analyses found no significant associations between titanium dioxide exposure and risk of lung cancer, chronic respiratory disease, or chest roentgenogram abnormalities. In addition, a subsequent nested case control study restricted to the oldest and largest of the two plants, was able to adjust for cigarette smoking habits, and also reported no increased risk with estimated exposure to titanium dioxide (Fayerweather et al., 1992).
The study by Ellis et al. (Ellis et al, 2010: Ellis et al, 2013) is an update and extension of the Chen and Fayerweather (1988) study which included workers from two titanium dioxide processing plants. The plants are not described by Chen and Fayerweather (1988), but Fayerweather et al (1992) notes that one is a large plant on the US east coast (Edgemoor) where all the cancer deaths in the study occurred, and the other is a smaller west coast plant. The study by Ellis et al (2010) included workers from the Edgemoor plant (the data from Chen and Fayerweather 1998 which included workers employed through 1983, were used as the basis for this study with worker data updated from 1983 onward) and two other plants (New Johnsonville and De Lisle). None of the workers were included in the US multicentre study by Fryzek et al (2003). Ellis et al (2010) included over 5000 production workers and 133 lung cancer deaths of which 111 occurred among Edgemoor workers (Chen and Fayerweather et al 1988 reported 9 lung cancer deaths among Edgemoor workers). Lung cancer mortality was less than expected (SMR = 0.90; 95% CI, 0.75-1.05), and no exposure-response relationship was found between TiO2 and mortality from lung cancer and non-malignant respiratory disease.
Fryzek et al. (2003) conducted a retrospective cohort mortality study of 4241 workers (3,832 males) employed for more than 6 months at 4 production facilities in the US. Mortality from all cancers was lower than expected (SMR=0.8; 95% CI 0.7-1.0) but the number of lung cancer deaths (61) was close to expected (SMR = 1.0; 95% CI 0.8–1.3). Workers with the highest titanium dioxide exposure (packing, micronizing or internal recycle workers) had a similar mortality pattern, i.e., lower than expected deaths for all cancer with no excess for lung cancers. Internal analyses showed that relative risks of all cause mortality and mortality due to lung cancer and non-malignant respiratory disease fell with increasing cumulative exposure. The investigators concluded that the data indicate that workers at the US plants have not experienced increased risks of lung cancer or other significant adverse health effects as a result of their occupational exposures to titanium dioxide.
The largest cohort study (Boffetta et al., 2004) included 15,017 titanium dioxide workers (14,331 males) employed for more than a year in 11 plants in six European countries (Finland, France, Germany, Italy, Norway, and the United Kingdom). Deaths due to all malignant neoplasms were fewer than expected (SMR = 0.98; 95% CI 0.91–1.05), but deaths due to lung cancer (306) were significantly higher than expected on the basis of national rates (SMR = 1.23; 95% CI 1.10–1.38). However, it was noted that lung cancer mortality rates were higher than corresponding national rates in eight out of ten locations where factories are located, i.e. the SMRs would have been lower if the investigators had used regional reference rates. Hext et al. (2005) noted that regional lung cancer mortality rates were approximately one fifth higher on average for workers in the study than national rates. Boffetta et al. (2004) undertook a very detailed and reliable exposure assessment for titanium dioxide and potential occupational confounders but internal analyses showed no evidence of an exposure– response relationship between estimated exposure to titanium dioxide dust and lung cancer mortality. The investigators concluded that the results of the internal analyses point towards the lack of a carcinogenic effect on the lung of titanium dioxide dust exposure, as experienced in this industry.
The study byGuseva Canu et al. (2020)aimed at investigating the association between titanium dioxide exposure and lung cancer mortality among French participants, accounting for smoking and other potential confounders. Therefore, the authors re-analysed the French data used in the dose-response analyses of a European cohort study conducted by Boffeta P. et al. (2004, see above). The authors 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, the authors estimated HRs and associated 95% CIs, using Cox regression models adjusted for calendar period, exposure duration and smoking. According to the authors, the cohort comprised 833 male workers (17 390 person-years) among whom 75% were exposed to TiO2. 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 and therefore included in multivariate models. Smoking status was known for 61% of workers, 5% of whom smoked. 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 four-fold higher risk of lung cancer mortality, though statistically non-significant. 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 two-fold 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. 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, though of borderline statistical significance. Smoking appeared not to confound any of these associations.
The work by Guseva Canu et al. (2020) has been reviewed by the expert epidemiologist by Mr J. Tomenson. The comments and critique are given below:
Summary
[Guseva Canu et al. (2020)] reanalysed the data of 833 French male workers included in the European multi- centre study (Boffetta et al., 2004). They noted that French participants of the European cohort of TiO2 workers exhibited an increase in mortality from lung cancer, and they aimed to investigate whether TiO2 exposure, co-exposures or smoking could explain the increase. It was noted that the original study included 2255 French male workers, but that Boffetta et al. (2004) also only included 833 French workers in their Cox regression analyses as they excluded 1412 French workers with unknown or unspecified job group and 10 workers lost to follow-up (The 833 workers include a further 21 workers lost to follow-up according to Table S2. After excluding workers with unknown or unspecified exposure, Tables 1.2 and 1.4 of Boffetta et al. (2003) include 843 French workers with 17,584 person-years). However, it was not stated by Guseva Canu, Gaillen-Guedy et al. (2020) that workers from two French plants were included in the European study, but only workers from one plant could be included in exposure-response analyses as individual exposure estimates could not be derived for workers in one of the plants (The European multi- centre study included workers from two French plants (factories 5 and 12), but Boffetta et al. (2003) noted that “As individual jobs could not be identified for factory 12, it was not possible to reconstruct exposure for jobs in this plant”).
Exposure was analysed as a binary variable (yes/no), continuous cumulative respirable TiO2 dust lagged by 10 years (mg/m3-year) and continuous and categorical annual average exposure (mg/m3). The cut-off values of average annual exposure were 0.3 and 2.4 mg/m3 (the German and American occupational exposure limits, respectively). The estimates of exposure to respirable TiO2 dust derived for the original study were stated to have been used in the reanalysis, but there appear to be some slight differences (Boffetta et al. (2003) reported a maximum cumulative exposure for French workers of 68.8 mg/m3-years (table 2.8) compared to 69.1 mg/m3-years in Table S1 of Guseva Canu,Gaillen-Guedy et al. (2020)). Follow-up data for the original study were also used (follow-up period of 1968–1997), and the Cox regression analyses included 16 lung cancer deaths as did that of Boffetta et al. (2003).
Smoking information was stated to have been obtained from the medical records of 512 workers present in the last 5 years (It is unclear what this means but enrolment occurred in 2001), of which only 43 (8.4%) were current smokers. Multiple imputation (with 1000 smoking status imputations) (An ‘imputation model’ is used to randomly sample values of the missing data (‘imputed values’) from their predicted distribution based on the observed data) was performed with ‘imputed values’ for smoking status derived using a prediction model based on 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.
The fully adjusted model (adjusted for calendar period, exposure duration and smoking) gave 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 (a group of 17 workers with 4 lung cancer deaths). Employment duration was negatively related to lung cancer mortality, and cumulative exposure had a small effect on mortality (HR=1.03 [95% CI=0.99 to 1.08] per mg/m3-year).
It was concluded that the study suggests a positive relationship between TiO2 exposure and lung cancer mortality in TiO2 workers, whatever the exposure variable used. It was also stated that the findings need to be confirmed in other cohorts, using different statistical approaches.
Overall comments
Average exposure is the key exposure metric used by Guseva Canu, Gaillen-Guedy et al. (2020), but it is unclear whether it was calculated with a 10 year lag as stated, or how many workers were exposed (see below).
Adjustment for smoking is claimed, but the multiple imputation procedure appears to be badly flawed. The smoking data used by the investigators does not seem plausible (only 8.5% were current smokers), and not consistent with information reported previously. Consequently, it is of little value to predict the smoking habits distribution among those with missing information. In addition, it is almost certain that none of the workers with average exposure > 2.4 mg/m3 were present in the last 5 years, and smoking information is not missing at random. Imputation may give biased results if this is not the case, and may be worse than an analysis of subjects with complete information (Hughes et al., 2019). There is no discussion of whether imputation was appropriate, and a complete case analysis was not performed, not even as a sensitivity analysis. It is also unclear why smoking status would be expected to be related to several of the variables included in the imputation model, but no information is given about the predictive value of the imputation model or the variables with the strongest predictive value. Consequently very little reliance can be placed on results adjusted for smoking (Model 3 results in Table 1).
The investigators emphasised the categorical analysis result for annual average exposure. However, the lead author was also the lead author of a review paper (Guseva Canu, Fraize-Frontier et al., 2020) which when discussing Boffetta et al. (2004), stated that “since an analysis using cumulative exposure as a continuous variable was not used, it is not possible to conclude on the absence of risk of mortality from lung cancer in this cohort”. Hence, it is unclear why they did not give more weight to the analysis of continuous average mean exposure or cumulative exposure. Given the problems with the adjustment for smoking, the best summary statistic for average mean exposure is that from Model 2 i.e. the just significant HR of 1.70 ((95%CI= 1.03 to 2.79) per mg/m3.
It is concluded that “the study suggests a positive relationship between TiO2 exposure and lung cancer mortality in TiO2 workers, whatever the exposure variable used”. However, there was no statistically significant relationship with cumulative exposure whatever lag period was used. The cumulative exposure metric is usually regarded as a better exposure metric than average exposure, but the cumulative exposure findings were dismissed because the exposure duration was negatively related to the outcome. This is not a sufficient reason to dismiss the cumulative exposure findings. Boffetta et al. (2004) noted the high lung cancer SMR among workers employed for less than five years (SMR = 1.61) in the European cohort, but cited a paper by Kolstad and Olsen (1999) which discusses reasons why short term workers often have high mortality. A similar excess of lung cancers was observed among the French workers employed for less than five years (13 observed versus 6.2 expected) (Boffetta et al., 2003). In addition, there was no obvious trend with average exposure in the Model 2 results shown in Table 1, and an implausibly high HR of 5.94 (95% CI 1.07 to 32.99) in the lowest average exposure group (>0 to 0.3 mg/m3).
It is reasonable to suggest that results need to be confirmed in other cohorts (the investigators presumably have access to the other sub cohorts in the multi-centre study), but it is unclear why they suggest that this needs to be done using different statistical approaches.
Overall, the results hinge on 4 lung cancers among a small group of 17 workers with average exposure > 2.4 mg/m3, and an average duration of exposure of 5.48 years. The short duration is claimed to be evidence of a healthy worker survivor effect (HWSE), but is actually a consequence of the fall of exposure levels over time which meant that a long duration worker could not achieve an average exposure > 2.4 mg/m3 (see below). The investigators in reality have no idea what the smoking habits of early workers were, and little information about other possible exposures (there were 2 pleural cancer deaths among French workers employed during the same period, although the factory where the workers were employed was not stated). Furthermore there was no measured exposure data to allow adjustment of French exposure estimates based on reconstructed exposure (see below). Given the implausibility of the available smoking data, and the inability to adjust adequately for smoking using multiple imputations, the study adds little to the findings reported for the French workers by Boffetta et al. (2003) which suggested a small but insignificant association with cumulative exposure.
Specific comments
References
Boffetta P, Soutar A, Cherrie JW, et al. Mortality among workers employed in the titanium dioxide production industry in Europe. Cancer Causes Control. 2004;15:697–706.
Boffetta P, Soutar A,Weiderpass E, et al. Historical Cohort Study of Workers Employed in the Titanium Dioxide Production Industry in Europe. Results of mortality follow-up. Final Report. Stockholm, Sweden: Department of Medical Epidemiology, Karolinska Institute; 2003.
Guseva Canu I, Fraize-Frontier S, Michel C, Charles S. Weight of epidemiological evidence for titanium dioxide risk assessment: current state and further needs. J Expo Sci Environ Epidemiol 2020;30: 430-435.
Hughes RA, Heron J, Sterne JAC, Tilling K. Accounting for missing data in statistical analyses: multiple imputation is not always the answer. Int J Epidemiol. 2019 ;48:1294-1304.
Kolstad HA, Olsen J. Why do short term workers have high mortality? Am J Epidemiol 1999;149: 347–352.
Furthermore, the epidemiology expert, Mr. J. Tomenson issued a critique dealing with another paper published by Guseva Cani et al. (2020). This critique also highlighted the significant weaknesses and methodological flaws in the paper as follows:
Comments on “Weight of epidemiological evidence for titanium dioxide risk assessment: current state and further needs”. Guseva Canu I, Fraize-Frontier S, Michel C, Charles S. J Expo Sci Environ Epidemiol. 2020;30:430-435.
Comments in this document are organised according to the sections of the paper and the supplementary table.
Abstract
The abstract claims that the paper addresses the importance of epidemiological evidence in risk assessment and decision-making in Europe using TiO2 to illustrate this. However, the paper does almost exact the opposite as it is clearly an attempt by the authors to restate the argument that they previously made in the CLP proposal which is that the epidemiological evidence for TiO2 is inadequate. The abstract also states that “a recent systematic review assessing the weight of evidence on the relationship between exposure to TiO2 (all forms) and cancer in humans questions the assumptions that TiO2 is an inert material of low toxicity”. It isn’t stated who performed this review, but it is stated that France submitted a proposal to classify TiO2 as a possible human carcinogen under the European regulation based on this “new data”. It is further claimed that no consideration was given to TiO2 particle size (presumably by ECHA although not stated), which may affect human health effects and consequently, further epidemiological studies are needed to assess possible associations between different physical–chemical characteristics of TiO2 exposures and their impact on human health. However, the abstract says nothing about the weight of evidence assessment that the title suggests was the purpose of the paper, and the conclusions from that assessment.
TiO2: the rationale for the risk assessment in the European framework
Much of this section is about the rationale for risk assessment in Europe as stated. However, the bias of the authors is clear when they describe as “controversial” the views of several stakeholders who questioned the epidemiological conclusions in the French classification proposal and concluded that epidemiological data were actually adequate.
Weight of evidence for TiO2 carcinogenicity in human/Table S1
The literature search only included documents published up to 31st of August 2015, which is unusual in a paper submitted for publication in March 2019, but permits them to ignore the meta-analysis paper by Le et al. (2018).
The 5 main cohort studies are included in the assessment (Chen and Fayerweather, 1988; Boffetta et al., 2004, Ellis et al., 2010; Ellis et al., 2013; Fryzek et al., 2003), but the two population case control studies (Boffetta et al., 2001; Ramanakumar et al., 2008) were excluded because they “included a broad array of workers not specifically exposed to TiO2”. This does not seem a valid reason to exclude studies from a systematic review.
It is stated that “the weight of evidence of TiO2 carcinogenicity in humans was documented and assessed according to the guidelines prescribed by the “Risk Assessment Methodology” by the work group of ANSES (sic)”, and that a standardised evaluation form was used to document critical aspects of each study under consideration such as the design, population exposure etc.The reference cited for the methodology is a critical review of the literature relating to weight of evidence for hazard Identification (Martin et al., 2018) which rated 24 approaches according to their prescriptive nature, relevance, and feasibility for screening of their potential for application within ANSES. However, Martin et al. (2018) did not state what approaches ANSES adopted, or describe a standardised form to be used.
Irrespective of where the standardised form came from, Table S1 wasn’t completed in a systematic manner, and the main results section is poorly described and has important omissions. Table S1 doesn’t state what the effect estimates are that are presented (presumably SMRs), or the numbers of deaths on which they are based. More importantly, the main results section gives no results at all from the internal analyses which were reported by 4 of the 5 studies summarised in Table S1 (Chen and Fayerweather, 1988; Boffetta et al., 2004, Ellis et al., 2013; Fryzek et al., 2003), most of which provide little support for the authors’ thesis. The casual reader might be unaware that any internal analyses were done, although Ellis et al. (2010) is criticised in Table S1 for having no analysis stratified by exposure level, nor any internal analyses for examining the exposure-effect and dose-response relationships. In addition, the study results are not presented in a systematic way. For example, the authors highlighted the results for the 3 countries in the European multicentre study by Boffetta et al. (2004) with lung cancer SMRs > 1, but did not give the SMRs for the other 3 countries. This was compounded by incorrectly claiming that the UK lung cancer SMR of 1.09 (95% 0.90-1.31) was of borderline significance (The p value is 0.36 (mid P exact test)
A risk of bias assessment is contained in the section entitled “Internal Validity” in Table S1 (excluding the Conclusions section). However, the authors’ don’t seem to understand many of the different types of bias listed in the OHAT assessment, and there is often little justification for their assessments of risk of bias. There are too many misunderstandings of the OHAT approach and bias in Table S1 to list them all, but the assessments of selection and reporting bias are particularly poor. Selection is discussed below, and the assessment of reporting bias focused almost entirely on funding and barely addressed selective reporting. Several criticisms relate to exposure misclassification and how it might have contributed to mask any increased risk from exposure, even though there is little indication in the paper that any exposure-response analyses were performed.
It is stated that the analysis of the risk for bias according to the approach proposed by the OHAT (2015) completed their evaluation. However, the data extraction in the first part of Table S1 isn’t an evaluation, and assessing the risk of bias of individual studies does not constitute a weight of evidence assessment which they claim to have done. The reviewers did not attempt to rate the total body of epidemiological evidence rather than simply assess risk of bias at a study level. Hence, they cannot claim to have performed a weight of evidence assessment as claimed. Weight of evidence approaches like GRADE include an assessment of the risk of bias of individual studies, but further require reviewers to make a judgement about whether the risk of bias in the individual studies is sufficiently large that their confidence in the estimated exposure effect is lower (Guyatt et al., 2011).
Specific comments
i. “Statistically significant increase of mortality for lung cancer was reported in two independent populations (one US and one European) among the included cohort studies”. The significantly elevated SMR reported by the US study (Ellis et al. 2013) is based on DuPont reference rates which inflated SMRs because the company mortality registry only includes active employees and pensioners prior to 1979 (Le et al., 2018). The SMR for lung cancer is < 1 when US reference rates are used.
ii. “All studies suffered from selection and exposure misclassification bias,…” The authors don’t seem to understand selection bias as described in the OHAT protocol. They incorrectly claim that a high risk of bias results from excluding subjects with limited exposure such as female workers and salary role employees. They also criticise the exclusion of short duration workers, but the possible exposure of short duration workers to other hazardous agents, and lifestyle factors, may dominate an effect of exposure to TiO2 (Kolstad and Olsen, 1999). Restricting the study group to workers who were employed on or after a particular date as in the study by Fryzek et al. (2003) is often necessary to avoid bias when it is not possible to identify all leavers before that date. Furthermore, exposure misclassification is usually not a serious limitation when it is non-differential. However, the authors didn’t assess whether exposure misclassification was likely to be differential when assessing the risk of bias.
iii. “….along with confounding effect of smoking and occupational exposures other than TiO2”. It is correct that none of the studies were able to adjust for smoking. However, it is more likely to have biased SMRs upwards. Boffetta et al. (2004) noted that in only one country was the prevalence of current smokers among cohort members higher than the national figures. In addition, plants were likely to be situated in areas with higher local rates than the national rates.
iv. “…there were weakness and inconsistencies in exposure assessment in all studies available…” The limited availability of personal sampling measurements is noted, and the reliance on area sampling measurements. However, there is little discussion of this in terms of the risk of bias. There is also no acknowledgement that studies like the European multicentre study used state of the art methodology to reconstruct exposure and took careful steps to ensure the consistency of estimates from different plants.
v. “We noted several inconsistencies in sampling and statistical methods. The TiO2 exposure was either assessed as an aerosol, i.e., use of measurement data based on inhalable fraction (comprising coarse, fine and ultrafine particles, such as total dust) or as a respirable fraction (comprising only fine and ultrafine particles). Statistical treatment of the measurement data reported inconsistent choices of exposure cutoffs”. There were no inconsistencies in the characterisation of exposure within studies and Boffetta et al. (2004) took great care to standardise all measurements, whatever protocol was used, to respirable dust. Furthermore, whether TiO2 exposure is assessed as total dust or respirable fraction, it makes no difference when using internal exposure response analyses to assess hazard. It is also not surprising that the studies used different cutoffs for exposure categories. It is a positive feature that the studies that performed dose-response analyses, used cutoffs determined by the exposure data for their study and not arbitrary cutoffs chosen by the investigators. Ellis et al. (2013) used time-dependent cumulative exposure with the cutoffs chosen with the aim of distributing person-years similarly among them. Fryzek et al. (2003) used tertiles of cumulative exposure, and Boffetta et al. (2004) selected cutoffs to give equal numbers of deaths in each category.
vi. “These inconsistencies in exposure assessments could affect the strength of the observed exposure-response effect by lowering the risk estimates toward the null while overestimating the exposure, and finding statistically non-significant estimates arising from high uncertainty and errors in exposure variables”. It is difficult to see how inconsistencies in exposure assessment could affect the strength of the observed exposure-response. It is also unclear why it would only result in overestimation of exposure, but as noted earlier, the level of exposure that an effect occurs is not important when assessing hazard.
vii. “The young age of the workers (around 30-years-old) at study entry and a follow-up duration that might be shorter than the latency-time needed between TiO2 exposure and the occurrence of lung cancer were additional drawbacks”. Age at entry is not relevant. Mean follow up in the studies ranged from 21 to 29 years which is not short, but the relevant parameter is time since first exposure, not follow-up. Many workers in the studies had been employed in the industry for several years before follow-up started and there was a lot of power to detect an effect with a latent period of 20+ years. For instance, in the European cohort study there were 181.3 expected lung cancer deaths among workers during the period 20+ years since first exposure (Table 3.9 of Boffetta et al. [2003]). It is clearly the same for the other cohorts, but the number of expected lung cancer deaths among workers during the period 20+ years since first exposure cannot be calculated. For instance, approximately a third of the cohort studied by Ellis et al. (2010) had worked 6 months in a process area by 1960 and were followed for at least 46 years. A further third had worked 6 months in a process area by 1980 and were followed for between 26 and 46 years.
viii. “The main issue in all studies reviewed was the presence of the healthy worker effect and in particular, the healthy worker survivor effect (HWSE).” These are two distinct forms of bias and shouldn’t be lumped together or used interchangeably as in the Performance section of Table S1. All the studies are described as having inclusion and exclusion criteria whose restrictiveness may have resulted to a loss of power and/or contributed to a “healthy worker” effect (HWE), but they are clearly referring to HWSE in this instance. They correctly define HWE as a bias faced when comparing occupational cohorts and national populations due to the selection of healthy individuals into the workforce. This may affect overall mortality, but probably had little effect on lung cancer among workers whose age at study entry ranged from 26 to 31 years and who when recruited would have been unlikely to have shown signs of chronic diseases linked to lung cancer through smoking. Furthermore, it is unlikely to be a problem when interpreting internal analyses. HWSE is a bias that occurs in occupational studies when less healthy workers are more likely to reduce their workplace exposures, and bias can occur in estimates of cumulative exposure-mortality associations. However, Guseva Canu, Fraize-Frontier et al. give no reasons why this would be expected to be important for low levels of exposure to an inert dust such as TiO2. However, the issue was addressed by Boffetta et al. (2004) who concluded that “the results of the analysis on the inception cohort, composed of workers whose employment is entirely covered by the follow-up, are remarkably similar to the results of the whole cohort, arguing against survival bias.” However, this analysis wasn’t mentioned by the reviewers.
ix. “…temporal variation in mortality rates has not been addressed in TiO2 worker cohorts, however. Such an effect seems very likely to have masked or underestimated the association between TiO2 exposure and mortality”. It is noted that Richardson et al. (2004) concluded that HWSE can lead to temporal variation in mortality rates that is correlated with cumulative exposure. However, Guseva Canu, Fraize-Frontier et al. have provided no evidence that HWSE has occurred, or that it has resulted in the temporal variation that Richardson et al. (2004) said can occur. In any case, Richardson et al. (2004) were only able to demonstrate a modest effect in simulations. Consequently it is complete speculation to conclude that HWSE seems very likely to have masked an association between TiO2 exposure and mortality.
x. The authors also listed their conclusions about the 5 studies in Table S1. These are often not linked to their assessment of risk of bias and some simply reflect their determination to show that the epidemiology is inadequate. For instance, they conclude that it is not possible to conclude on the absence of risk of mortality from lung cancer in the large European cohort study of Boffetta et al. (2004) because “an analysis using cumulative exposure as a continuous variable was not used”. This is nonsense, but the authors will be well aware that such an analysis was reported by Le. et al. (2018).
Scientific advances to strengthen the epidemiological evidence in TiO2 risk assessment
The authors claim that their systematic review raised the need to characterize the HWSE and reassess the exposure-mortality association for lung cancer in a large TiO2 occupational cohort with adequate control for this bias. However, as noted earlier they have provided no evidence whatsoever that HWSE is an issue in studies of TiO2, and don’t mention that the largest study (Boffetta et al., 2004) reported that their results “argued against survival bias”. However, some of the same group claim to have demonstrated a HWSE effect in their reanalysis of French workers included in the European multicentre study Guseva, Canu, Gaillen-Guedy et al. 2020), although it is clear that the short duration of employment of the workers with the highest average exposure in that study is not due to HWSE.
The authors also suggest performing a joint international cohort study based on rigorous standards of data harmonization where the exposure assessment is emphasised. However, that is precisely what was done in US and European studies and those studies included all TiO2 workers that it was possible to study at that time as established in comprehensive feasibility studies. They also suggest that the analysis of the exposure-mortality association for lung cancer with respect to TiO2 exposure should be performed in a nested case–control study. However, it is unlikely that better exposure and lifestyle data would be obtained for cases and controls than is already available in the cohort studies, and a nested case-control study would result in a loss of power.
They also note that none of the epidemiological studies on TiO2 attempted to characterise the size, crystal structure and surface coating/treatment of particles. They claim this should be possible citing a pilot study of French nuclear workers as support. Clearly this is an important issue, but something that the investigators of the US and European studies were well aware of. It may be feasible to characterize particles in this way in the nuclear industry where better records have to be kept, but previous attempts by the industry to assess the feasibility of doing this suggest that it is not realistic in the TiO2 industry.
Conclusion
It is stated that “the new epidemiological evidence questions the assumptions that TiO2 is an inert material of low toxicity” They are presumably referring to this poor quality review paper (or the CLP review) which has no synthesis of evidence from different studies, and provides no new data. New data is presented by Le et al. (2018), but that is overlooked by the authors.
References
Boffetta P, Gaborieau V, Nadon L, Parent MF, Weiderpass E, Siemiatycki J. Exposure to titanium dioxide and risk of lung cancer in a population-based study from Montreal. Scand J Work Environ Health. 2001;27:227–32.
Ellis ED, Watkins J, Tankersley W, Phillips J, Girardi D. Mortality among titanium dioxide workers at three DuPont plants. J Occup Environ Med. 2010;52:303–9.
Boffetta P, Soutar A, Cherrie JW, Granath F, Andersen A, Anttila A, et al. Mortality among workers employed in the titanium dioxide production industry in Europe. Cancer Causes Control. 2004;15:697–706.
Boffetta P, Soutar A,Weiderpass E, et al. Historical Cohort Study of Workers Employed in the Titanium Dioxide Production Industry in Europe. Results of mortality follow-up. Final Report. Stockholm, Sweden: Department of Medical Epidemiology, Karolinska Institute; 2003.
Guseva Canu I, Gaillen-Guedy A, Wild P, Straif K, Luce D. Lung cancer mortality in the French cohort of titanium dioxide workers: some aetiological insights. Occupational and Environmental Medicine Published Online First: 31 July 2020. doi: 10.1136/oemed-2020-106522
Chen JL, Fayerweather WE. Epidemiologic study of workers exposed to titanium dioxide. J Occup Med. 1988;30:937–42.
Ellis ED, Watkins JP, Tankersley WG, Phillips JA, Girardi DJ. Occupational exposure and mortality among workers at three titanium dioxide plants. Am J Ind Med. 2013;56:282–91.
Fryzek JP, Chadda B, Marano D, White K, Schweitzer S, McLaughlin JK, et al. A cohort mortality study among titanium dioxide manufacturing workers in the United States. J Occup Environ Med. 2003;45:400–9.
Guyatt GH, Oxman AD, Vist G, Kunz R, Brozek J, Alonso-Coello P, et al. GRADE guidelines: 4. Rating the quality of evidence–study limitations (risk of bias). Journal of clinical epidemiology. 2011;64(4):407-15.
Kolstad HA, Olsen J. Why do short term workers have high mortality?Am J Epidemiol 1999;149: 347–352.
Le HQ, Tomenson JA, Warheit DB, Fryzek JP, Golden AP, Ellis ED.A Review and Meta-Analysis of Occupational Titanium Dioxide Exposure and Lung Cancer Mortality. J Occup Environ Med. 2018 Jul;60(7):e356-e367.
Martin P, Bladier C, Meek B, Bruyere O, Feinblatt E, Touvier M, et al. Weight of evidence for hazard identification: a critical review of the literature. Environ Health Perspect. 2018;127:1–15.
OHAT (Office of Health Assessment and Translation) Handbook for conducting a literature-based health assessment using OHAT approach for systematic review and evidence integration. Research Triangle Park, NC: OHAT; 2015.
Ramanakumar AV, Parent ME, Latreille B, Siemiatycki J. Risk of lung cancer following exposure to carbon black, titanium dioxide and talc: results from two case-control studies in Montreal. Int J
Cancer. 2008;122:183–9.
Richardson D, Wing S, Steenland K, McKelvey W. Time-related aspects of the healthy worker survivor effect. Ann Epidemiol. 2004;14:633–9.
Case-control studies
Siemiatycki (1991) conducted a hypothesis-generating case-control study in Montreal that included male patients with 20 different types of cancer and assessed exposure to 293 substances including titanium dioxide. A more refined analysis of the relationship between titanium dioxide and lung cancer in the Montreal study was later performed by Boffetta et al. (2001) and which incorporated an improved exposure assessment, which resulted in several changes to the ascribed exposure status of subjects. Boffetta et al. (2001) included all 857 cases of lung cancer of the original study, but constructed a new group of 1066 controls comprising all 533 population controls and a random sample of 533 of the 1349 cancer controls (subjects with cancers of other organs) included by Siemiatycki (1991). Odds ratios (OR) were adjusted for confounders including smoking history and were not elevated for ever exposure to titanium dioxide (OR=0.9; 95% CI 0.5-1.5; 33 cases) or substantial exposure (OR=1.0; 95% CI 0.3-2.7; 8 cases).
Ramanakumar et al. (2008) has reported on combined results from the case-control study conducted by Siemiatycki (1991) and another large population case control study conducted in Montreal (Ramanakumar et al., 2006) which included 1236 cases of lung cancer (765 males) and 1512 population controls (899 males). OR adjusted for a number of possible confounders including smoking, were calculated using population and cancer controls in the former study, for male and female subjects in the latter study, and for the combined group of 2093 lung cancer cases and 3394 controls. In both studies there was no evidence of excess risk among subjects who had been exposed to titanium dioxide and also in pooled analyses, OR were close to unity for any exposure to titanium dioxide (OR=1.0; 95% CI 0.6-1.7; 76 cases) and substantial exposure (OR=1.2; 95% CI 0.4-3.6; 8 cases). The authors concluded that occupational exposure to titanium dioxide did not produce an excess risk of cancer, consistent with the evaluations of the IARC working group.
Conclusion
The epidemiological studies are well conducted, and the findings are robust and replicated in all three large cohort studies and two large case-control series. Statistical power is not a limitation of the epidemiological studies as the three key cohort studies (Fryzek et al, 2003; Boffetta et al, 2004; Ellis et al, 2010) include over 24,000 production workers in 18 manufacturing plants in 7 countries, with a total of 457 expected lung cancer deaths of which a high proportion were expected after a sufficient latency period. The Canadian case-control studies included 2093 lung cancer cases (Boffetta et al, 2001; Ramanakumar et al, 2008).
In summary, no causative link between titanium dioxide exposure and cancer risk in humans has been demonstrated.
Respiratory irritation
During literature search, a reference was identified presenting an investigation on respiratory irritation. The investigation was carried out with occupationally exposed workers to titanium dioxide. The reference was considered of very limited relevance for hazard assessment purposes. The criteria for quality, reliability and adequacy of experimental data under REACH and for hazard assessment purposes (ECHA guidance R4 in conjunction with regulation (EC) 1907/2006, Annexes VII-X) are not fulfilled. This reference is discussed below, and highlighting its findings:
Elo et al. (1972) wanted to determine the biological effects of occupational TiO2 exposure in a case control study. Three TiO2-exposed factory workers were analysed regarding pulmonary metal deposition and symptoms of pulmonary irritation. TiO2 deposition was analysed via x-ray fluorescence spectrometry, and quantitative optic emission spectrography. Pathological changes of lung tissue were analysed using light microscopy, and electron microscopy. Specimens derived in two cases from biopsies and in one case from an autopsy. According to the authors, the findings suggest that industrially processed TiO2, either alone or with other compounds such as silica, behaves as a mild irritant in pulmonary interstitium. The irritation mechanism remains unknown. Lastly, it cannot be ruled out that the clinicopathologic findings may have been due to the toxic effects of other metals, e.g. silica, rather than titanium.
Summary entry – respiratory irritation
Another two references were identified during a literature search, representing investigations on respiratory irritation. These investigations were conducted with occupationally exposed workers to nano-form titanium dioxide. The study designs are not in accordance with accepted guidelines and are therefore of limited relevance for chemicals hazard assessment. The references usually lack significance due to, e.g., insufficient description of method, uncertainty in number workers analysed and/or confounding factors cannot be excluded. It is therefore concluded that all references do not fulfil the criteria for quality, reliability and adequacy of experimental data for the fulfilment of data requirements under REACH and hazard assessment purposes (ECHA guidance R4 in conjunction with regulation (EC) 1907/2006, Annexes VII-X). The studies given below were included in the IUCLID for information purposes only:
Pelclova, D. et al (2016): method description is insufficient. Leukotriene analyses provide only limited data on potential adverse effects. In this study, workers with elevated levels of leukotrienes have no manifested respiratory diseases. In Pelclova et al. (2016), where the same cohort is used, it is stated that 34 workers were used for analyses. However, here it is stated that 30 workers were analysed. Thus, it is questionable why different numbers of workers were analysed without a statement. Furthermore, smokers were included and other confounders could not been excluded.
Pelclova, D. et al (2016): method description is insufficient. Lipid markers for oxidative stress provide only limited data on potential pathological effects. They only reflect temporally changes in physiology but are not linked directly to diseases. Implausible statement on the TiO2 exposed cohort. In Pelclova et al. (2016), where the same cohort is used, it is stated that 30 workers were used for analyses. However, here it is stated that 34 workers were analysed. Thus, it is questionable why different numbers of workers were analysed without a statement. Furthermore, smokers were included and other confounders could not been excluded.
Conclusion
Three references were identified, representing investigations on respiratory irritation in humans. The study design of the reference is not in accordance with accepted guidelines and are therefore of limited relevance for chemicals hazard assessment. The references lack significance due to, e.g. , insufficient method description and/or confounding factors cannot be excluded. It is therefore concluded that all references do not fulfil the criteria for quality, reliability and adequacy of experimental data for the fulfilment of data requirements under REACH and hazard assessment purposes (ECHA guidance R4 in conjunction with regulation (EC) 1907/2006, Annexes VII-X). The information contained therein were included for information purposes only.
Information on Registered Substances comes from registration dossiers which have been assigned a registration number. The assignment of a registration number does however not guarantee that the information in the dossier is correct or that the dossier is compliant with Regulation (EC) No 1907/2006 (the REACH Regulation). This information has not been reviewed or verified by the Agency or any other authority. The content is subject to change without prior notice.
Reproduction or further distribution of this information may be subject to copyright protection. Use of the information without obtaining the permission from the owner(s) of the respective information might violate the rights of the owner.
![ECHA](/o/diss-blank-theme/images/factsheets/A-REACH/factsheet/echa_logo.png)