Long-Term Exposure to Air Pollutants and Cancer Mortality: A Meta-Analysis of Cohort Studies

The aim of this study was to examine the relationship between main air pollutants and all cancer mortality by performing a meta-analysis. We searched PubMed, EMBASE (a biomedical and pharmacological bibliographic database of published literature produced by Elsevier), and the reference lists of other reviews until April 2018. A random-effects model was employed to analyze the meta-estimates of each pollutant. A total of 30 cohort studies were included in the final analysis. Overall risk estimates of cancer mortality for 10 µg/m3 per increase of particulate matter (PM)2.5, PM10, and NO2 were 1.17 (95% confidence interval (CI): 1.11–1.24), 1.09 (95% CI: 1.04–1.14), and 1.06 (95% CI: 1.02–1.10), respectively. With respect to the type of cancer, significant hazardous influences of PM2.5 were noticed for lung cancer mortality and non-lung cancer mortality including liver cancer, colorectal cancer, bladder cancer, and kidney cancer, respectively, while PM10 had harmful effects on mortality from lung cancer, pancreas cancer, and larynx cancer. Our meta-analysis of cohort studies indicates that exposure to the main air pollutants is associated with increased mortality from all cancers.


Introduction
The global level of particulate matter <2.5 µm in size (PM 2.5 ) rose by 11.2% from 1990 (39.7 µg/m 3 ) to 2015 (44.2 µg/m 3 ), and exposure to PM 2.5 was the fifth most common cause of death in 2015 globally, resulting in the deaths of 4.2 million people [1]. Ambient air pollutants were recently classified as lung carcinogens by the International Agency for Research on Cancer of the World Health Organization (WHO) and are considered as "the most extensive environmental carcinogens" [2].
To date, three meta-analyses [3][4][5] have examined the association between air pollution and lung cancer mortality; in them, a 10 µg/m 3 increase in PM 2.5 levels increased the risk of and mortality from cancer by 9%, 9%, and 7%, respectively. However, these three meta-analyses used both incidence and mortality data of lung cancer. Importantly, there is a difference between cancer incidence and mortality, because not all patients suffering from cancer will die from the disease [6]. Recent prospective cohort data collected from 623,048 participants over 22 years showed that a 4.4 µg/m 3 increase in PM 2.5 levels increased kidney and bladder cancer mortality rates by 14% and 13%, respectively [7]. Nitrogen dioxide (NO 2 ) was also positively linked to increased mortality from colorectal cancer in this study (hazard ratio (HR) per 6.5 parts per billion (ppb): 1.06; 95% confidence interval (CI): 1.02-1.10).
At present, there are no reported quantitative meta-analyses on the association between ambient air pollution and mortality from all types of cancers. The current study addressed this gap by performing a meta-analysis of 30 cohort studies, as well as various subgroup analyses of the factors that might influence the results.

Data Sources and Searches
We searched PubMed and EMBASE (a biomedical and pharmacological bibliographic database of published literature produced by Elsevier) from October 1958 to April 2018 using common keywords related to air pollutants and cancer mortality. The keywords were "air pollution", "air pollutants", "particulate matter", "nitrogen dioxide", "sulfur dioxide", and "ozone" for exposure factors and "cancer", "malignancy", and "carcinoma" for outcome factors. Additionally, we inspected the bibliographies of related articles and reviews to identify additional pertinent data.

Study Selection and Eligibility
We included observational articles that met the following criteria: (1) a prospective or retrospective cohort study; (2) examined the association between air pollution and mortality from any type of cancer; and (3) reported outcome measures with adjusted relative risk (RR) and 95% CI. When two or more analyses contained duplicated data or used the same participants, we included the more comprehensive analysis. We excluded the following: (1) in vivo and in vitro studies; (2) case reports, review articles, and letters; (3) studies on cancer incidence but not mortality; (4) studies with inconvertible data; and (5) studies assessing indoor, occupational, or accidental exposures to pollutants.
Using the selection criteria, three authors (H.B.K., J.Y.S., and B.P.) independently assessed the eligibility of the retrieved articles. Any disagreements among the evaluators were resolved by discussion with the help of a fourth author (Y.J.L.).

Data Extraction
Two authors (H.B.K. and B.P.) independently extracted the study characteristics from the eligible articles, which were then reviewed by a third author (Y.J.L). The extracted data included the name of the first author, publication year, type of cohort study, year in which the participants were enrolled, location of the study, means of quantifying exposure (e.g., degree of exposure, mean concentration of pollutants), number of cases, type and stage of cancer, adjusted confounding variables, and adjusted RR ratios and 95% CI.

Assessment of Methodological Quality
We used the Newcastle-Ottawa Scale (NOS) [8] to estimate the methodological quality of the studies included in our meta-analysis. The NOS is comprised of three subscales (selection of studies, comparability, and exposure), and its scores range from 0-9. There is no established cut-off point for high versus low quality; hence, we rated studies with higher than average scores as high-quality and analyzed all studies despite their score.

Main and Subgroup Analyses
The main analysis examined the association between long-term exposure to air pollutants and cancer mortality. Subgroup analyses assessed the effect of the following factors on cancer mortality: type of air pollutant, gender, geographical region, duration of cohort study, mean pollutant concentration according to WHO guidelines, type of cancer, stage of cancer, number of participants, methodological quality, and smoking status. Subgroup analyses were conducted separately for the two pollutants that most significantly impacted cancer mortality.

Statistical Analyses
Because most exposure-response meta-analyses consider the relationship between air pollution and disease mortality to be linear [9,10], our protocol also included standardized increments: a 10 µg/m 3 increase in exposure to PM 2.5 ; particulate matter <10 µm in size (PM 10 ); NO 2 , nitrogen oxides (NO x ), and sulfur dioxide (SO 2 ); and a 10 ppb increase in exposure to ozone (O 3 ). We recalculated the RR for the standardized increment for each pollutant by applying the following formula [11]: where RR is the relative risk and ln is the log to the base e. If the RR was presented on a continuous scale as an interquartile range (IQR), we used the increment in IQR instead of the increments noted above.
To evaluate the association between air pollutants and cancer mortality, a pooled RR ratio and 95% CI was calculated from the adjusted RR ratio and 95% CI in each study. To test heterogeneity across studies, we used the Higgins I 2 test to determine the percentage of total variation [12]. I 2 was computed as follows: where Q is Cochran's heterogeneity statistic and df indicates the degrees of freedom. I 2 values ranged from 0% (no observed heterogeneity) to 100% (maximal heterogeneity), with values >50% indicating substantial heterogeneity [12]. A random-effects model based on the DerSimonian and Laird method was used for calculating the overall RR and 95% CI values, because populations and methodologies differed among the studies [13].
We assessed publication bias using Begg's funnel plot and Egger's test [14]. When bias was present, the funnel plot showed asymmetry or Egger's test had a p-value <0.05. We used Stata SE software, version 13.1 (StataCorp, College Station, TX, USA) for the statistical analyses.

Eligible Studies
The abstracts of a total of 1302 articles were identified in the initial investigation of two databases and by hand-searching relevant bibliographies. After excluding 485 duplicated articles, two of the authors independently surveyed the eligibility of all studies and excluded an additional 712 articles that did not meet the predetermined inclusion criteria ( Figure 1). Finally, the full texts of the remaining 105 articles were inspected, of which 75 articles were excluded for the following reasons: no RR data (n = 31), air pollution not quantified (n = 14), insufficient exposure and outcome data (n = 8), a categorical range of air pollutants was used (n = 8), population sharing (n = 7), no mortality rates for cancer (n = 5), cancer incidence was used as an outcome measurement (n = 1), and smoking status was used as a co-exposure factor (n = 1). The remaining 30 cohort studies were included in the meta-analysis [7,. All cohort studies were prospective except the study by Ancona et al. [35], which was retrospective.  Table 1 shows the general characteristics of the 30 cohort studies included in our meta-analysis. All studies were published between 1999 and 2017 and together comprised >36,077,332 participants. In studies reporting age, the mean age of the participants was 57.3 years (range: 0-120 years). Regarding the type of cancer, most of them concerned lung cancer, while some of them involved all types. Mostly, the selected studies were conducted in the United States (n = 10), the Netherlands (n = 3), and China (n = 3). Adjusted variables of each study were presented in Table A1. Ten studies used fixed-site monitor measurements for the exposure assessment method, while 17 studies used modeling-based assessment methods such as land-use regression or air dispersion models. All studies except three [29,31,35] were funded by public/governmental organizations or independent scientific foundations. The NOS scores of the studies ranged from 5 to 9; the average score was 7.7. The number of high-quality studies (NOS score ≥ 8) was 21. Data were extracted from the general population in all studies except four, which were conducted on breast cancer patients [29], lung cancer patients [37], patients with myocardial infarction [40], and liver cancer patients [43], respectively.

Subgroup Analyses of the Association between PM 2.5 and Cancer Mortality
The significant relationship between PM 2.5 and cancer mortality was very similar in the subgroup analyses stratified by gender, geographical region, follow-up period, mean levels of pollutant concentration, stage of cancer, number of participants, methodological quality, and smoking status.

Subgroup Analyses of the Association between PM 10 and Cancer Mortality
Long-term exposure to PM 10 significantly correlated with cancer mortality in subgroup analyses stratified by mean pollutant concentration, cancer stage, methodological quality, and smoking status. As shown in Table 2, it increased the mortality rate in pancreas cancer, larynx cancer, and lung cancer. However, PM 10 was not related to mortality from cancers other than lung cancer, in contrast to PM 2.5 . Similar to PM 2.5 , PM 10 best correlated with mortality in early-stage cancer. PM 10 , unlike PM 2.5 , did not adversely affect mortality rates in men, women, patients in Europe, patients with follow-up periods <10 years, and a small study size.

Discussion
Our meta-analysis of 30 cohort studies involved >1.0 million cases in 14 countries and hence provided sufficient statistical power. It showed that ambient air pollution significantly correlated with cancer mortality in analyses including all participants, as well as those stratified for various factors. Among the pollutants examined, PM 2.5 , PM 10 , or NO 2 were most strongly associated with cancer mortality, whereas O 3 was not significantly associated.
The deleterious effects of air pollution on survival were not limited to the lungs, but also included non-lung organs, especially in cancer patients exposed to PM 2.5 . Evidence from several in vivo studies suggests that particulate pollutants can travel to the liver, kidneys, and brain [44][45][46]. Our study indicates that air pollution is more strongly linked to cancer mortality in early-stage patients than those in later stages. Although many clinicians presume that the opposite is true, current research shows that patients in earlier stages of cancer may require more education regarding air pollution exposure prevention.
How air pollution increases cancer mortality rates is unclear, but two mechanisms have been proposed. The first mechanism involves DNA damage due to oxidative stress. Reactive oxygen species cause oxidative stress and are generated in response to PM [47]. Nitrogen pollutants can exacerbate the effects of oxidative stress on the progression of breast, prostate, colorectal, cervical, and other cancers [48]. Exposure to SO 2 is extremely harmful, as it induces oxidative stress in many organs [49]. Undue oxidative stress in cancer cells may seriously affect survival outcomes by promoting cell proliferation, genetic instability, and mutations [50]. In a prospective cohort study from the United States that included 30,239 Caucasian and African-American participants, there was a significant association between an oxidative stress and cancer mortality [51].
The second mechanism involves inflammation. In an in vitro study, inhaled gaseous and particulate pollutants increased the production of proinflammatory cytokines such as interleukin (IL)-6 and IL-8 [52]. In a cohort panel study conducted in the United States, exposure to NO x and PM increased plasma IL-6 levels over a 12-week period [53]. The poor prognosis of gastric cancer and non-Hodgkin's lymphoma has been linked to excessive amounts of the proinflammatory cytokines tumor necrosis factor and IL-1, respectively [54,55]. Furthermore, the production of tumor-associated macrophages, which occurs during the inflammatory reaction, is a sign of an exacerbated cancer state [56]. Thus, inflammation caused by exposure to air pollution may result in cancer mortality.
Unlike the other pollutants in our study, O 3 did not significantly impact lung cancer and brain cancer survival. Similarly, in the meta-analysis conducted by Atkinson et al. on lung cancer only [57], there was no association between long-term exposure to O 3 and lung cancer mortality (RR: 0.95; 95% CI: 0.83-1.08; I 2: : 55%). Nonetheless, evaluating this relationship is challenging. O 3 is comprised of a combination of noxious air elements termed the "photochemical cocktail", and its mechanisms of formation and destruction differ from those of other pollutants [58].
The key strengths of our meta-analysis are its inclusion of all cancer types, its separation of cancer mortality from cancer incidence, and its coverage of more countries and cases than previous studies. It also included more factors in its subgroup analyses than did the three previous meta-analyses that assessed the association between air pollution and lung cancer risk [3][4][5]. Furthermore, unlike previous studies, it examined the impact of air pollution on non-lung cancer mortality as well as lung cancer mortality. Overall, it provided the most comprehensive information to date on the mortality risk of cancer patients exposed to the main air pollutants.
The limitations of the current study include (1) no distinction between urban and rural areas; (2) considerable heterogeneity as indicated by the Higgins I 2 values; (3) no information about indoor air pollution caused by heating, cooking, and passive smoking; (4) inclusion of only one or two studies in most cancer subgroups (lung and breast cancers were the exceptions); and (5) no data on confounding factors such as physical activity, X-ray testing, and radon exposure [59].

Conclusions
Our data showing a robust association between air pollution and all-cancer mortality have important implications for public health. This association applied to almost all of the pollutants examined in the study and was strongest for particulate pollutants in the regions wherein their mean concentrations were below standard levels. Similarly, a recent cohort study in the United States with >60 million participants found that exposure to PM 2.5 increased all-cause mortality rates at concentrations below the present national limits [60]. Hence, rigorous environmental health policies are needed to keep air pollution levels, and consequently cancer mortality rates, as low as possible. Additionally, our results show that different types of PM increase the mortality rates for different types of non-lung cancers (PM 2.5 : liver, colorectal, bladder, and kidney; PM 10 : pancreas and larynx); hence, they may act via different mechanisms. Future research should focus on the association between certain types of pollutants and mortality from organ-and type-specific cancers.