In this large population-based cohort with individual-level information on potential confounders, increased risks for lung cancer were estimated in relation to total PM2.5 and PM10 as well as in relation to PM10-traffic, PM2.5-exhaust, and PM2.5- wood burning. The findings from our study contribute to the existing knowledge of the effects of air pollution on lung cancer by estimating associations with source-specific local air pollution exposure. Although associations were non-statistically significant, the risk estimates were higher in relation to local emissions from traffic and wood burning compared with total concentrations including background concentrations.
Previous studies suggest that particles are among the most important abundant pollutants in urban air which predisposes the lungs to high inhalation of particles. In Northern Sweden, particles emitted by traffic such as generated by combustion and by road wear and brake abrasion and residential wood burning are the two most significant local contributors to PM [15]. It has been proposed that residential wood combustion, in line with has previously been observed for traffic related air pollution, have detrimental impact on respiratory and cardiovascular health but the role of different sources have not been investigated in many epidemiological studies on air pollution and lung cancer incidence previously [19].
The potential mechanisms by which inhaled particles cause lung cancer include oxidative stress, inflammation, DNA damage, epigenetic regulation, metabolism, and related signal transduction pathways [20]. Inhaled PM may induce oxidative stress by releasing reactive oxygen species which further trigger the different markers of inflammations [21]. These inflammatory responses including the release of cytokines, TNF-α (tumour necrosis factor) and IL-1 (interleukin 1) may further lead to the development of lung cancer [22]. Exposure to PM may also damage the DNA by changing nucleotide sequence during replication which may induce the tumour promoting genes (oncogenes) or inactivate the tumour supressing genes [23]. Apart from that, PM may also cause gene dysregulation through epigenetic alterations by DNA methylation, histone modifications, and non-coding RNA expression [24, 25]. In addition to that, exposure to PM may alter the signal transduction pathways which may affect the cell cycle, metabolism, and apoptosis [26]. However, there is also support for inflammation as a direct potential mechanism. Mice with a mutation in the epidermal growth factor receptor gene, which were exposed to particles had an increased risk to develop lung tumours than unexposed control mice [27]. There was however no evidence for an increase in the number of mutations in the lung cells of the mice. A sustained inflammatory response several weeks after the particle exposure however supports that the air pollution may rather cause lung cancer by creating an environment characterized by inflammation that promotes the growth of cells with pre-existing cancer-causing mutations., rather than by mutating DNA.
Several previous studies have shown an association between long-term total particle concentrations and the risk of lung cancer in the general population. Using land use regression modelled PM10 and PM2.5 within the European Study of Cohorts for Air Pollution Effects (ESCAPE), Raaschou-Nielsen et al., observed risk increases of 22% (95% CI: 3–45%) and 18% (95% CI: -4-46%) per 10 µg/m3 PM10 and 5 µg/m3 PM2.5 respectively in a meta-analysis of 17 European cohorts [28]. In a more recent meta-analysis by Citabanni et al (2021), of 15 high quality studies that were adequately adjusted for potential confounders, a higher risk was observed than in previous studies [8]. The risk increases were 16% (95% CI: 9–23%) per 5 µg/m3 PM2.5 and 23% (95% CI: 5–40%) per 10 µg/m3 PM10. A higher risk increase was found for lung cancer mortality, compared with incidence, and among studies in Asia. In a pooled analysis of seven European cohorts, within the Effects of Low-level Air Pollution: a Study in Europe (ELAPSE) study, a 13% (95% CI: 5–23%) increased risk for incident lung cancer per 5 µg/m3 PM2.5 was reported even at concentrations below WHO Air Quality Guidelines [10].
Despite major advances in the knowledge on the association between air pollution and lung cancer, some of the issues related to the source and type of air pollutants remain unanswered. The results were somewhat inconsistent among a handful of previous studies estimating the risk of lung cancer in relation to source specific particle concentrations. For example, a meta-analysis of cohort, case-control and nested case-control studies conducted largely in Europe and North America reported an increase in the risk of lung cancer in relation to NO2, a marker of traffic generated air pollution, with an odds ratio of 1.06 (95% CI: 0.99–1.13). For traffic-related PM2.5, the odds ratio in the same study was 1.11 (95% CI: 1.00–1.22). Among studies investigating the association between PM2.5 and the risk of lung cancer, some focussed on mortality. In an analysis of the US Medicare cohort of elderly participants investigating the effect of long term PM2.5 components and their resources, a positive but statistically non-significant association with lung cancer mortality was found in relation to coal-related PM2.5 with an HR of 1.04 (95% CI: 1.00–1.08), and traffic-related PM2.5 with an HR of 1.03 (95% CI: 0.99–1.06), per IQR increase, whereas no association was observed with PM2.5 from biomass burning [29]. The large scale American Cancer Society Cancer Prevention Study reported an increased risk of lung cancer mortality by 6% per 10 µg/m3 of total PM2.5 as well as 4% increased risk per 1.6 µg/m3 of near-source (primarily traffic-related) PM2.5 whereas a decreased risk by 4% was observed in relation to NO2 [30]. Although NO2 is not an established carcinogenic factor, several studies use it as a surrogate for combustion emissions including traffic-related emissions [31–35]. Several studies employ proxy measures such as proximity to the nearest major road to depict traffic-related exposure. For example, in an analysis of 110,000 participants within the Netherlands Cohort Study on Diet and Cancer, a non-statistically significant increased risk of lung cancer was found in relation to traffic intensity on the nearest road and living near a major road [36]. Although the findings in the present study were non-statistically significant, the magnitude of our estimates was larger than in most previous studies. This discrepancy can be explained by differences in population characteristics, differences in pollutants sources as well as differences in the assessments of air pollution exposure. For instance, even though in ESCAPE cohort air pollution was assessed at residential addresses of participants with reasonable precision, the exposure assessments did not account for the change in addresses which would introduce exposure misclassification. Furthermore, the meta-analysis by Ciabattini et al included studies that measured average concentrations of PM2.5 and PM10 exposure at an ecological level that may introduce non-differential misclassification resulting in underestimation of risk [8]. In our study, besides the use of residential address histories, exposure models were updated during the years of follow-up to account for temporal variability in exposure. Moreover, we estimated air pollution with high spatial resolution using dispersion modelling which capture the variation in long-term air pollution present in urban environments, and with focus on locally emitted PM2.5.
Our study has several strengths worth mentioning. Our study benefits from a large sample size and long follow-up period. Lung cancer cases were identified using well validated patient and cause of death registers, thus reducing the risk of classification bias. The strengths of our study also include the use of state-of-the-art air pollution modelling for exposure assessment. We assessed ambient total and source-specific particle concentrations with a high spatial resolution. The model further took into consideration many factors affecting emission and dispersion, such as meteorological conditions, amount of traffic (including vehicle types and speed, which affect exhaust and wear emissions), the width of the street, and the height of neighbouring buildings that affect the local dispersion of air pollutants. In addition, participants’ change in residential address was accounted for. Moreover, our study is strengthened by the adjustment of a large set of confounders at both the individual and area level.
The present study also has several limitations that should be highlighted. Firstly, exposure measurement at participant’s residential addresses is prone to exposure misclassification because, in reality, true exposure for individuals is not restricted to ambient concentrations at the home address but also include indoor particle concentrations at the place of residence, at the workplace, and during commuting. Secondly, we cannot rule out that non-participation of individuals with low socioeconomic status at recruitment might lead to the underestimation of the true effect of air pollution on the risk of lung cancer. This would be the fact if low socioeconomic status was assiocaiated with higher levels of air pollution, since this group also has a higher incidence of lung cancer. In the study area however, the correlation between air pollution and socioeconomy is quite low. Thirdly, information on noise and green space is missing in the VIP cohort which may confound the association between air pollution and incident of lung cancer. Evidence of these environmental exposures as independent risk factors for lung cancer is emerging but still somewhat inconsistent and limited [37, 38]. Occupational exposures may also cause lung cancer [39], but the association between occupational exposure and ambient air pollution exposure in this study area is not yet investigated. Fourth, VIP participants lack information on exposure during the life course. Furthermore, information on potential confounders was available only at recruitment which may not fully reflect the true association throughout follow-up. Different factors such as disease incidence, injury, or a health screening intervention may influence these baseline variables.