(a) Harvard Six Cities Study

The Harvard Six Cities Study was designed specifically to evaluate exposures to outdoor air pollution and their association with health among 8111 White adults aged 25–74 years recruited from six cities in the eastern USA representing a range of air pollution exposures: Steubenville, Ohio, and St. Louis, Missouri (high exposure); Watertown, Massachusetts, and Kingston-Harriman, Tennessee (medium exposure); and Portage (including Wyocena and Pardeeville), Wisconsin, and Topeka, Kansas (low exposure). Follow-up began between 1975 and 1977 in each city, and each participant completed a baseline questionnaire. Active follow-up, including information on vital status, continued by mail until 1991 (Dockery et al., 1993). In addition, the cohort was followed up through searches of the National Death Index, with the last update in 2009 (Lepeule et al., 2012).

Three follow-ups of the Harvard Six Cities Study have been published: through 1989 or 1991 (determined by the last date of search of the National Death Index or the ending date of the study, respectively) (Dockery et al., 1993), through 1998 (Laden et al., 2006), and through 2009 (Lepeule et al., 2012). Between 1979 and 1986–1988 (depending on the city) the researchers measured outdoor concentrations of total suspended particles (TSP), sulfur dioxide (SO₂), ozone, suspended sulfates, and particulate matter (PM) with particles of aerodynamic diameter less than 2.5 μm (PM_2.5), less than 15 μm (PM₁₅) (before 1984), and less than 10 μm (PM₁₀) (after 1984) at central site air-monitoring stations in each community (Dockery et al., 1993; Laden et al., 2006). Measurements of PM₁₀ from routine air-quality monitoring stations are available from monitors located near the original central sites in each community starting in 1985, and measurements of PM_2.5 are available starting in 1999. For follow-up after 1999, these measurements were used to obtain city-specific annual average PM_2.5 concentrations from representative monitors within 80 km of the original sampling locations (Lepeule et al., 2012). For the period between the end of the original monitoring activity and 1999, the authors predicted city-specific annual average PM_2.5 based on PM₁₀ concentrations from the monitors, extinction coefficients (humidity-corrected visibility data from the local airport), and indicators of season (Laden et al., 2006).

Publications on the first and second follow-ups of the cohort (Dockery et al., 1993; Laden et al., 2006) reported positive associations of lung cancer mortality with sulfate and PM_2.5, respectively. In the most recent follow-up of the full cohort, through 2009, 4495 deaths (7.8% from lung cancer) were identified. In adjusted models, the hazard ratio (HR) was 1.37 (95% confidence interval [CI], 1.07–1.75) for each 10 µg/m³ increase in PM_2.5 averaged over the past 1–3 years (determined as the averaging period with the best fit). Analyses of penalized splines indicated a linear relationship. (The city-specific annual average PM_2.5 ranged from approximately 40 µg/m³ to 8 µg/m³ over the course of the entire follow-up period.) Other modelling options were also considered, with similar results (Lepeule et al., 2012). In stratified analyses, relative risks were elevated for never-smokers, current smokers, and former smokers, with former smokers appearing to have the highest risk compared with never-smokers or current smokers; however, the P-value for interaction was not statistically significant. In time-period-specific analyses, the hazard ratio for the past 8 years of follow-up (2001–2009) was the strongest (HR, 2.84; 95% CI, 1.06–7.59), but again the P-value for interaction was not significant. Because of the high correlation of exposures over time, the critical exposure window could not be determined (Lepeule et al., 2012).

[A limitation of the Harvard Six Cities Study is that it relied on central site monitoring for each city. Since all participants in the same city were assigned the same exposure value, there was no ability to assess intra-city spatial variation in exposures. In Dockery et al. (1993), a city-specific average value for each pollutant was created for 3–8 years during the late 1970s or the 1980s, depending on the pollutant; therefore, there was also no assessment of variation of air pollution over time. Furthermore, in some cases the measurements may actually have occurred after the relevant death. In the subsequent analyses, an annual average for PM_2.5 for each city was created and assigned to the follow-up time by calendar year; thus, temporal variation was accounted for. Other limitations are that there is no information on whether a person moved out of the city in which they were recruited during the follow-up period, and that only baseline information on covariates was included in the final analyses. However, in a reanalysis of the Dockery et al. (1993) study, Krewski and colleagues (HEI, 2000) found that the results were robust to updates of the covariates and changes in address. The strengths of the study are the detailed exposure monitoring, the assessment of different modelling options in the most recent follow-up, the prospective follow-up, and the control for potential confounders at the individual level: age, sex, body mass index (BMI), smoking status and pack-years of smoking, and education level.]

(b) American Cancer Society study

Six analyses of the association of measures of air pollution with lung cancer mortality have relied on data collected by the American Cancer Society (ACS) as part of the Cancer Prevention Study II (CPS-II) (Pope et al., 1995, 2002; Jerrett et al., 2005, 2013; Krewski et al., 2009; Turner et al., 2011). The ACS CPS-II is an ongoing prospective mortality study of approximately 1.2 million adults aged at least 30 years at enrolment in 1982 and residing throughout the USA. Before 1989, vital status was obtained from the study volunteers who enrolled the participants. After 1989, searches of the National Death Index were conducted. Individual-level risk factor information was available through a baseline questionnaire. The original analysis included 7 years of follow-up, through 1989 (Pope et al., 1995). Subsequent publications on the nationwide cohort included follow-up through 1998 (16 years) (Pope et al., 2002), through 2000 (18 years) (Krewski et al., 2009), and through 2008 (26 years), with the most recent follow-up restricted to never-smokers (Turner et al., 2011). Jerrett et al. (2005, 2013) and Krewski et al. (2009) conducted analyses focused on Los Angeles, New York City, and the state of California.

Analyses of air pollution effects in the ACS CPS-II took advantage of existing national monitoring networks to quantify air pollution exposures. Therefore, the study population was restricted to participants who resided in United States Metropolitan Statistical Areas (MSAs) with available pollution data (approximately 300 000–500 000 participants, depending on the follow-up and pollutants measured). Baseline average PM_2.5 levels were quantified by MSA of residence for 1979–1983 using data from the Inhalable Particle Monitoring Network. TSP and different size fractions of PM (PM₁₅, PM_15–2.5) were available for that time period as well. PM₁₀ concentrations for 1982–1998 and PM_2.5 beginning in 1999 were available from the United States Environmental Protection Agency (US EPA) AIRS database. Average outdoor PM_2.5 concentrations from 1999–2000 were used to quantify exposures towards the end of the follow-up period, and an average of the two time periods was considered as an estimate of exposure over the course of the study. Sulfate concentrations were available for 1980–1981, and SO₂, nitrogen dioxide (NO₂), carbon monoxide (CO), and ozone levels were available for 1980 and 1982–1998. Ranges and/or means of exposure for all pollutants in each follow-up are presented in Table 2.1 when available (Pope et al., 2002; Krewski et al., 2009; Turner et al., 2011).

Reports of the first two follow-ups of the ACS CPS-II cohort (Pope et al., 1995, 2002) provided data on the association of lung cancer mortality with concentrations of PM_2.5 and sulfate. Positive associations were observed for both indicators. The second report also included data for SO₂, NO₂, CO, and ozone (Pope et al., 2002).

In the most recent analysis of the cohort, with 2 more years of follow-up and additional adjustment for individual-level covariates from the baseline questionnaire, hazard ratios were 1.08 (95% CI, 1.03–1.14) and 1.11 (95% CI, 1.04–1.18) for PM_2.5 in 1979–1983 and 1999–2000, respectively, and 1.05 (95% CI, 1.02–1.11) and 1.04 (95% CI, 0.97–1.11) for sulfate in 1980 and 1990, respectively (Krewski et al., 2009). No association with lung cancer was observed for ozone, SO₂, NO₂, or CO. Additional adjustment for ecological covariates derived at the postal code and MSA level did not materially change the results. Spatial autocorrelation was also included in sensitivity analyses, but results specific to lung cancer are not presented. No clear patterns emerged from analyses of time windows of exposure.

[The nationwide assessments of the ACS CPS-II rely on MSA-specific measurements from at most two time periods (near the beginning and near the end of the follow-up). There is no assessment of intra-city variability, which could enhance contrasts in exposure but should not hamper interpretation of the results that were obtained. Furthermore, there is no accounting for whether a participant moved out of their baseline location, which could contribute to measurement error. Covariates were not updated over time, potentially reducing the ability to control confounding. However, strengths of the studies are the assessment of a linear trend, the prospective follow-up, and that individual-level covariates are controlled for in both analyses. Krewski et al. (2009) also included a large battery of ecological covariates and assessment of autocorrelation.]

In the analysis with the longest follow-up period to date (26 years), Turner et al. (2011) assessed the association of PM_2.5 with lung cancer mortality among 188 699 lifetime non-smokers in the ACS CPS-II cohort. In Cox proportional hazard models adjusting for the same covariates as in the Pope et al. (2002) study with the addition of dietary variables, additional measures of occupational exposures, and mean county-level residential radon concentrations, each 10 μg/m³ increase in PM_2.5 was associated with a 15–27% increase in lung cancer mortality, depending on the time window in which the PM_2.5 concentration was assessed. The fully adjusted hazard ratios were modestly attenuated compared with those not adjusted for residential radon. A plot of adjusted hazard ratios in relation to quartiles of PM_2.5 (1999–2000) suggests a linear trend, but a formal test was not presented. [The strengths and limitations of Turner et al. (2011) are the same as those described for Pope et al. (2002) and Krewski et al. (2009). The restriction of this study to never-smokers is a major additional strength, as is the control for other potential confounders – diet, occupation, and residential radon – as covariates in the models. Both features reduce the potential for confounding, although among the additional covariates considered, only radon had a notable effect on the association with PM_2.5.]

The ACS investigators also performed specific analyses in Los Angeles, the state of California, and New York City, where they had the ability to describe intra-city variability in exposure using land-use regression exposure models. In Los Angeles, the hazard ratio controlling for 44 individual-level covariates was 1.44 (95% CI, 0.98–2.11) per 10 μg/m³ increase in PM_2.5 estimated for each individual’s postal code centroid. This was attenuated to 1.20 (95% CI, 0.79–1.82) when statistically significant contextual covariates were included (Jerrett et al., 2005). Results using a slightly different statistical technique to predict exposures were similar (Krewski et al., 2009). In equivalent analyses in New York City, the hazard ratio adjusted for individual-level covariates was 0.96 (95% CI, 0.84–1.09) per 1.5 μg/m³ increase in PM_2.5 (the interdecile range) (Krewski et al., 2009). Jerrett et al. (2013) modelled monthly averaged PM_2.5, NO₂, and ozone at the baseline residential address for all 73 711 ACS CPS-II participants residing in California. Monitoring data for PM_2.5 and NO₂ were supplemented with data from satellites. Exposure metrics were determined as averages for 1988–2002 for NO₂ and ozone and as averages for 1998–2002 for PM_2.5. In models adjusted for 42 individual-level covariates, as well as indicators of the MSAs and 7 ecological covariates, hazard ratios for lung cancer were elevated for an interquartile range (IQR) increase in PM_2.5 (HR, 1.06; 95% CI, 0.95–1.18) and NO₂ (HR, 1.11; 95% CI, 1.02–1.21) but not for ozone (HR, 0.86; 95% CI, 0.75–0.99). Analyses representing exposure with splines showed no evidence of nonlinearity. Associations with NO₂ remained statistically significant in two pollutant models that additionally included PM_2.5 and ozone, while estimates for PM_2.5 were reduced to unity. [The authors described NO₂ as a marker for traffic-related air pollution. However, since other exposure sources (i.e. heating and industrial) were included in the models, their contributions to the modelled concentrations cannot be ruled out. The results for NO₂ are more consistent with the European studies, which also assessed intra-city variability, than with the full-country ACS analysis, which assessed variation across larger areas.]

[The strengths of the three area-specific analyses of the ACS CPS-II (Los Angeles, New York City, and California) are the detailed adjustment for baseline and ecological confounders and the ability to consider intra-city variability and temporal variability.]

(c) Adventist Health Study on Smog

The Adventist Health Study on Smog (AHSMOG) cohort consists of 6340 Seventh-Day Adventists who were participants in the Adventist Health Study. They were non-Hispanic Whites residing in California, currently non-smokers, who had lived for at least 10 years within 5 miles (8 km) of their baseline residence and were aged 25 years to more than 80 years when they completed the baseline questionnaire in 1977 (Abbey et al., 1991a).

Monthly outdoor concentrations of TSP, PM₁₀, SO₂, NO₂, and ozone were estimated for each participant starting in 1966 from fixed-site monitors maintained by the California Air Resources Board. Exposure predictions were restricted to postal code centroids within 50 km of a monitoring station and were not allowed to cross topographical obstructions or other barriers to air flow (Abbey et al., 1991b). Before 1987, PM₁₀ was estimated from TSP, and before 1972, total oxidants were measured as opposed to ozone. For each pollutant, both means and exceedance frequencies (the sum of hours for gaseous pollutants or of days for particulate pollutants above a specified cut-off based on federal and California standards at the time) were used as exposure metrics in the majority of the epidemiological analyses. Cancer incidence was ascertained from linkage with cancer registries and medical record review of self-reported hospitalizations, and mortality was ascertained from linkage with the California death certificate file, the National Death Index, and the Seventh-Day Adventist church records (Mills et al., 1991).

In the initial follow-up through 1982, elevated hazard ratios were reported for the associations of lung cancer with TSP and total oxidants (Abbey et al., 1991a; Mills et al., 1991).

Results for associations of lung cancer with air pollution with follow-up extended through 1992 have been presented in three separate publications (Beeson et al., 1998; Abbey et al., 1999; McDonnell et al., 2000). In addition to exceedance frequencies, cumulative annual averages from January 1973 until 3 years before the date of the case defining the risk set (i.e. a 3-year lag) were estimated for PM₁₀, SO₂, and 8-hour average ozone. Among men, positive associations were observed for incident lung cancer (n = 16) after adjustment for age, years of education, pack-years of smoking, and alcohol consumption, for each 24 µg/m³ (IQR) increase in PM₁₀ (HR, 5.21; 95% CI, 1.94–13.99), each 3.7 ppb (IQR) increase in SO₂ (HR, 2.66; 95% CI, 1.62–4.39), each 19.8 ppb (IQR) increase in NO₂ (HR, 1.45; 95% CI, 0.67–3.14), and each 12.0 ppb (IQR) increase in 8-hour average ozone (2.23; 95% CI, 0.79–6.34). Among women (cases = 20), the equivalent hazard ratios were elevated only for SO₂ (HR, 2.14; 95% CI, 1.36–3.37) (Beeson et al., 1998). Among men, the equivalent hazard ratios for lung cancer mortality (n = 17) were 3.36 (95% CI, 1.57–7.19) for PM₁₀, 1.99 (95% CI, 1.24–3.20) for SO₂, 1.82 (95% CI, 0.93–3.57) for NO₂, and 2.10 (95% CI, 0.99–4.44) for 8-hour average ozone. Among women, the hazard ratios for lung cancer mortality (n = 12) were 1.08 (95% CI, 0.55–2.13) for PM₁₀, 3.01 (95% CI, 1.88–4.84) for SO₂, 2.81 (95% CI, 1.15–6.89) for NO₂, and 0.77 (95% CI, 0.37–1.61) for 8-hour average ozone (Abbey et al., 1999). [The Working Group noted a discrepancy in the IQRs for NO₂ and ozone as reported in the papers by Abbey et al. (1999), where the IQRs are given as 19.8 ppb and 12.03 ppb, respectively, and Beeson et al. (1998), where they are reported as 1.98 ppb and 2.12 ppb, respectively. The Working Group was unable to obtain clarification from the authors. Therefore, the former figures, which are of the same magnitude as those reported in other papers based on the same population, are assumed to be the correct ones.]

PM_2.5 data were available for a subcohort residing within an airshed adjacent to one of nine airports located throughout California (n = 1347 men, 2422 women) during the study baseline period (McDonnell et al., 2000). Visibility and PM_2.5 were measured concurrently at the nine airports from 1979 to 1993. These data were used to predict daily PM_2.5 concentrations for 1966–1993 using linear regression controlling for season and relative humidity, and monthly average PM_2.5 concentrations were calculated for each airshed. Monthly average PM_2.5–10 values were calculated as the monthly mean PM₁₀ (modelled as in earlier publications) minus the monthly mean PM_2.5 value for each participant. In addition, monthly ozone, SO₂, and NO₂ for 1966–1992 and SO₄ for 1977–1992 were interpolated to the postal code centroid of the participants’ residential and work addresses. Among the 1228 men with PM_2.5 data, 13 deaths from lung cancer were identified. Hazard ratios adjusted for age, pack-years of smoking, years of education, and current alcohol consumption were elevated with wide confidence intervals for all size fractions of PM using the average for the baseline years 1973–1977. In models including PM_2.5 and PM_2.5–10 simultaneously, the magnitude of the hazard ratio for PM_2.5–10 was attenuated. Results for women were not presented but were described as weak or inverse. Numbers of deaths from lung cancer in the subcohort were too small to assess the other pollutants.

[The strengths of the AHSMOG studies are the long prospective follow-up, the detailed information on long-term work and home addresses, the control for individual covariates, and the ability to assess incidence as well as mortality. Major limitations are the very small sample size, the non-standard exposure assessment, and the lack of clarity and consistency in the published reports of the results.]

(d) Trucking Industry Particle Study

The Trucking Industry Particle Study (TrIPS) was a retrospective occupational cohort study of men in the trucking industry across the USA, which was originally designed to examine the association of occupational exposures to vehicle exhaust with lung cancer. The investigators subsequently examined the effect of annual outdoor exposures to PM₁₀, SO₂, and NO₂ on lung cancer mortality (Hart et al., 2009, 2011). Exposures to PM₁₀ and NO₂ were estimated by modelling data obtained from the US EPA Air Quality System for 1985–2000 (Hart et al., 2009), and PM_2.5 was estimated based on the nearest Air Quality System monitor in 2000. (Mean concentrations for all pollutants are presented in Table 2.1.) In models stratified by age at entry, decade of hire, and calendar year, and adjusted for race, census region of residence, the healthy worker survivor effect, and occupational exposures, the hazard ratios for annual average (1985–2000) SO₂ and NO₂ were elevated in the full cohort. When long-haul drivers who spend days away from home were excluded, elevated hazard ratios were also observed for PM₁₀. The hazard ratio for a 4 μg/m³ (IQR) increase in PM_2.5 in 2000 was 1.02 (95% CI, 0.95–1.10) in the full cohort and 1.07 (95% CI, 0.97–1.17) in the cohort excluding long-haul drivers (Hart et al., 2011). In supplemental analyses using information from a survey administered to a sample of currently employed and recently retired workers in the industry, the authors evaluated the potential impact of not controlling for smoking and BMI. Current smoking, but not BMI, was associated with small increases in the hazard ratio for PM₁₀ and NO₂ (PM_2.5 was not evaluated). Therefore, not adjusting for smoking may have led to inflation of the effect estimates.

[The limitations of this study are the lack of control for potential confounding by individual-level risk factors such as smoking history and the reliance on the most recent address to predict exposure. The strengths are the control for occupational exposures and the availability of time-varying residence-level predictions for most of the pollutants.]

(e) California Teachers Study

The California Teachers Study is a prospective cohort study of 133 479 female current and former public school professionals participating in the California State Teachers’ Retirement System. All participants completed a baseline questionnaire in 1995. Subsequent questionnaires were mailed in 1997 and 2000, and name and residential address were updated annually. Cause-specific mortality data were obtained from record linkage with California state registries, the United States Social Security Administration, and the National Death Index. Monthly surfaces of PM_2.5, PM₁₀, ozone, nitrogen oxides (NO_x), NO₂, CO, and SO₂ in 250 m grids were estimated using inverse-distance weighting of data from monitors maintained by the state. (Ranges and means [standard deviations] for each pollutant are presented in Table 2.1.) Individual monthly exposures were estimated by linking each residential address to the gridded pollutant surface, and cumulative averages were calculated for each risk set to represent long-term exposure. The final analytical data set included 101 784 women with available air pollution data (73 489 for PM_2.5, since monitoring began later). In adjusted multivariable models, there was no evidence of a positive association of any of the pollutants considered with lung cancer mortality risk (Lipsett et al., 2011). However, in analyses restricted to never-smokers, the hazard ratio for a 10 μg/m³ increase in PM_2.5 was 1.62 (95% CI, 0.83–3.16).

[The strengths of this study are the spatially and temporally resolved air pollution measurements and the control for time-varying confounding by individual-level lung cancer risk factors. The limitations are that there were only 5.6 years of follow-up for the analyses of PM_2.5 and that PM_2.5 data were only available concurrently or for the 12 months before the start of follow-up. Therefore, there was no ability to look at truly long-term exposures, which might be necessary for assessment of an association with lung cancer risk.]

(f) Ontario Tax Cohort Study

Villeneuve et al. (2013) conducted a cohort study of intra-urban variations in volatile organic compounds (VOCs) and NO₂ and cause-specific mortality in Toronto, Canada. The population consisted of 58 760 residents of Toronto in 1982 who were part of the larger Ontario Tax Cohort Study, a cohort randomly selected from income tax filings of Canadians residing in one of 10 urban areas in the province of Ontario. Residential addresses at baseline were linked by postal code centroid to estimated exposure surfaces of benzene, n-hexane, total hydrocarbons, and NO₂ modelled from measurements (obtained in 2006 for the VOCs and in 2002 and 2004 for NO₂) and spatial covariates. Deaths were identified by linkage to the Canadian Mortality Database. Individual-level information on household income and marital status was obtained from the income tax return, and contextual measures of unemployment, average household income, and immigration were obtained from the 1981 Canadian census. Data from the 2001 Canadian Community Health Survey were used to indirectly adjust for smoking and obesity, using the spatial association between smoking, BMI, and pollution measurements.

The participants ranged in age from 35 to 85 years in 1982 (mean, 51.7 years). Through 2004, 1470 deaths from lung cancer were identified among 18 020 deaths. In Cox proportional hazards models adjusted for age, sex, the socioeconomic variables described above, and NO₂, the hazard ratios for all VOCs were elevated. After indirect adjustment for smoking and BMI, the effect estimates were attenuated. No independent association was observed of NO₂ with lung cancer mortality (Villeneuve et al., 2013).

[This study is the only cohort study to date to assess residential exposure to VOCs as well as other urban air pollutants. Individual-level information on smoking was not available, but the measurements of socioeconomic status that were obtained are a strength and could help to control indirectly for smoking.]

(a) Studies with qualitative or semiquantitative assessment of air pollutant exposure

Vena (1982) conducted a hospital-based case–control study in Erie County, New York, USA. Retrospective data on residential and employment history and on smoking were obtained from 417 White male lung cancer patients and 752 controls with non-respiratory, non-neoplastic diseases, admitted from 1957 to 1965. Two-year TSP data and a historical review of point sources of air pollution were used to define air pollution zones. Subjects were classified by duration of residence in zones with medium or high air pollution levels. The results did not show a clear association of lung cancer with air pollution alone; the relative risk for exposure of 50 years or longer in medium- or high-pollution zones went from 1.58 (95% CI, 1.09–2.29) after adjustment for age and occupation to 1.26 (CI not reported; P > 0.05) after adjustment also for smoking. There was a suggestion of effect modification for air pollution as there was increased risk from smoking and occupational exposures if there was also long-term exposure to air pollution. The risk for heavy smokers with heavy exposure to air pollution was more than 4 times that of men with none of the high-exposure conditions. [Exposure assessment was rather crude. No response rates by case–control status were provided.]

Samet et al. (1987) conducted a population-based case–control study of lung cancer in New Mexico, USA, between 1980 and 1982, including 422 cases and 727 controls. Subjects were asked to identify all locations where they had resided for 6 months or longer and were interviewed about other personal characteristics, including smoking and occupation. The residential data were coded at the county and state levels and combined with county-level socioeconomic data from population censuses to generate indices of time lived in counties or metropolitan areas of different sizes, degrees of urbanization, and extents of employment in manufacturing industries. Residential history patterns were the same in cases and controls. There was no association of the residential history variables with lung cancer risk; relative risks were constantly close to unity. [Exposure assessment was rather crude in this study.]

Katsouyanni et al. (1991) conducted a case–control study exploring the role of smoking and outdoor air pollution in the causation of lung cancer. The study was undertaken in Athens, Greece, between 1987 and 1989 and included women only; 101 women with lung cancer and 89 comparison women with fractures or other orthopaedic conditions were included. Smoking habits were ascertained through interviews, whereas lifetime exposure to air pollution was assessed by linking lifelong residential and employment addresses with objectively estimated or presumed air pollution levels. Pollution isopleths were based on smoke and NO₂ levels measured at fixed-monitor stations in the period 1983–1985. Air pollution levels were associated with increased risk of lung cancer with an odds ratio of 1.22 (95% CI, 0.91–1.63) per quartile of exposure, which was reduced to 1.09 (CI not reported) after adjustment for tobacco smoking. The P-value for the interaction of air pollution and tobacco smoking was 0.10. There was no effect of air pollution among non-smokers [crude OR, 0.7], but the relative risk contrasting extreme quartiles of air pollution among smokers of 30 years’ duration was 2.23 (CI not reported). [Detailed information on occupational exposure and smoking was available, but the size of the study is limited.]

Jöckel et al. (1992) conducted a hospital-based case–control study in five cities in Germany, including 194 lung cancer cases, 194 hospital controls, and 194 population controls (only a sample of all hospitals was included). Subjects were interviewed for their smoking, occupational, and residential history by trained interviewers, using a standardized questionnaire. For the quantification of occupational exposure to known carcinogens of the lung, an approach was developed with exposure information obtained by supplemental questionnaires. Quantification of air pollution was based on emission data for SO₂ and a semiquantitative index. After adjustment for smoking and occupational exposures, relative risks of 1.01 (95% CI, 0.53–1.91) for the emission index and 1.16 (95% CI, 0.64–2.13) for the semiquantitative index were obtained. [Exposure assessment was rather crude, on a county basis, and the statistical power was rather low.]

Barbone et al. (1995) investigated the relationship between air pollution and lung cancer with a case–control study among men who had died in Trieste, Italy, from 1979 to 1981 and from 1985 to 1986. From an autopsy registry, 755 cases and 755 controls were identified, and information on smoking, occupation, and residence was obtained from the next of kin. Air pollution at the residence of each subject was estimated from the average value of total PM deposition at the nearest monitoring station. After adjustment for age, smoking habits, likelihood of exposure to occupational carcinogens, and social group, the risk of lung cancer increased with increasing level of air pollution for all types of lung cancer combined (P = 0.022; RR, 1.4; 95% CI, 1.1–1.8 for the highest exposure category vs the lowest), for small cell carcinoma (P = 0.016), and for large cell carcinoma (P = 0.049). Compared with inhabitants of the residential area, the relative risk was 1.5 (95% CI, 1.0–2.2) for residents of the centre of the city and 1.4 (95% CI, 1.0–2.1) for residents of the industrial area. The increased relative risk of the industrial area was mainly due to exposure to an iron foundry (RR, 1.7; 95% CI, 0.7–4.1) and an incinerator (RR, 2.6; 95% CI, 1.3–5.1).

Gupta et al. (2001) conducted a case–control study on lung cancer in Chandigarh, northern India, involving 265 lung cancer cases and 525 hospital controls matched by age and sex. Data were collected in face-to-face interviews. The exposure assessment of air pollution was based on lifetime residence in areas classified by the investigators as predominantly urban, rural, or mixed. Residence in urban areas was not associated with increased risk of lung cancer: the odds ratio for men living in urban areas was 0.82 (95% CI, 0.53–1.27) and for women living in urban areas it was 0.29 (95% CI, 0.07–1.17). [Response rates were not reported. Exposure assessment was crude, and the precision was rather low.]

Edwards et al. (2006) conducted a case–control study of lung cancer among women in the highly industrialized area of Teesside, in north-eastern England. A total of 204 women aged 80 years or younger with incident primary lung cancer and 339 age-matched community controls were recruited for a population-based case–control study. Life-course residential, occupational, and active and passive smoking histories were obtained using an interviewer-administered questionnaire. The adjusted odds ratio for lung cancer among women living near (within 0–5 km of) heavy industry in Teesside or elsewhere was 1.85 (95% CI, 0.80–4.24) for more than 25 years’ residence versus 0 years, or 1.21 (95% CI, 0.99−1.47) for each period of 10 years living near industry (latency was allowed for by disregarding residential exposures within the past 20 years). [A low response rate among controls (47.8%) is noted.]

Chiu et al. (2006) investigated the relationship between air pollution and lung cancer among women in a matched case–control study on deaths (972 cases) that occurred in Taiwan, China, from 1994 to 2003. The control group consisted of women who had died from causes other than cancer or respiratory diseases (972 controls), pair-matched to the cases by sex, year of birth, and year of death. A municipality-based aggregate index of long-term exposure to air pollution was created by dividing the annual average of the measured values for each pollutant by the National Ambient Air Quality Standard for that pollutant. The ratios for each pollutant were scaled to a 100-point scale and then averaged together to generate an index value representing the net burden of these pollutants. Women who lived in municipalities with the highest levels of the air pollution exposure index had a statistically significant increased risk of lung cancer compared with those living in municipalities with the lowest air pollution exposure index after controlling for the urbanization index (RR, 1.28; 95% CI, 1.02–1.61). [Only housewives according to the death certificate were included. Since only deceased subjects were in the controls and exposure to air pollution is linked with increased cardiorespiratory mortality, exposure to air pollution may have been overrepresented in the control group, with underestimation of the effect. In addition, controlling for urbanization, which is likely to be a predictor of pollution levels, could result in overadjustment. No smoking data were available, although smoking was rare among women in Taiwan, China.]

Using the same design as Chiu et al. (2006), Liu et al. (2008a) investigated the relationship between air pollution and lung cancer in a matched case–control study on deaths among women (1676 cases and 1676 controls) that occurred in Taiwan, China, from 1995 to 2005. The classification of exposure was based on the measured levels of PM₁₀, SO₂, NO₂, and CO. NO₂ and CO levels were used to classify subjects’ areas of residence into tertiles of pollutant concentrations. An urbanization index and marital status were considered in the analyses in addition to the matching factors. Among individual pollutants, positive associations were observed only for NO₂ and CO. A composite index based on these two pollutants yielded adjusted odds ratios of 1.24 (95% CI, 1.03–1.50) and 1.46 (95% CI, 1.18–1.81) for groups with medium and high exposure, respectively, compared with the group with low exposure, with a statistically significant trend. [The study shares the methodology and the limitations of the initial study by Chiu et al. (2006), and there may be a partial overlap in the study populations.]

Chang et al. (2009) investigated the relationship between exposure to traffic-related air pollution and development of lung cancer in women in Taiwan, China, using the density of petrol stations as a surrogate measure of exposure. A matched case–control study was based on lung cancer deaths among women (4087 cases and 4087 controls) from 1997 to 2006, using the same design as Chiu et al. (2006). Data on the number of petrol stations in study municipalities were collected from the two major petroleum supply companies. The petrol-station density per square kilometre for each municipality was used as an indicator of exposure. There was a statistically significant exposure–response relationship between the tertile of petrol-station density and the risk of lung cancer in women after controlling for possible confounders. [The study shares the methodology and the limitations of the studies by Chiu et al. (2006) and Liu et al. (2008a), and there may be an overlap in the study populations.]

López-Cima et al. (2011) conducted a hospital-based case–control study in Asturias, Spain. The study area included a large industrial setting. A total of 626 lung cancer patients and 626 controls were recruited and matched by ethnicity, hospital, age, and sex. Distances from the respective participants’ residential locations to industrial facilities and city centres were used as a metric of exposure to urban and industrial air pollution. Odds ratios for distance to pollution sources were estimated with adjustment for sex, age, hospital area, tobacco smoking, family history of cancer, and occupation. Individuals living near industries had an excess risk of lung cancer (RR, 1.49; 95% CI, 0.93–2.39). The relative risk was higher for small cell carcinoma (RR, 2.23; 95% CI, 1.01–4.92). Residents in urban areas showed a statistically significant increased risk of adenocarcinoma (RR, 1.92; 95% CI, 1.09–3.38) compared with those in rural areas.

(b) Studies of exposure to industrial pollution

Brown et al. (1984) reported the results of a case–control study on lung cancer deaths among residents near a zinc smelter and a steel manufacturing plant in Pennsylvania, USA. Lifetime residential, occupational, and smoking histories were obtained from the next of kin of 335 White male lung cancer cases and 332 White male controls. Relative risks were estimated according to the distance of the residence from the zinc smelter and the steel plant, and according to levels of several metals (arsenic, copper, lead, manganese, zinc, and cadmium) measured in soil samples. Twofold risks of lung cancer were associated with residence in areas with heavy levels of arsenic (RR, 2.3; 95% CI, 1.0–5.4) and cadmium (RR, 2.0; 95% CI, 0.9–4.6). Usual residence near the zinc plant was associated with an increased risk (RR, 1.6; 95% CI, 0.6–4.3), although the number of individuals living in the higher exposure area was small. These increases remained after accounting for cigarette smoking and employment in the zinc or steel industry. No excess risk was associated with living near the steel plant.

Pershagen (1985) studied 212 male lung cancer cases and 424 control men who had died between 1961 and 1979 in an industrialized county in northern Sweden. Data on smoking, occupation, and residence were obtained from the next of kin. A relative risk of 2.0 (95% CI, 1.2–3.4) for lung cancer was seen among men who had lived within 20 km of a large copper smelter that emitted arsenic.

Xu et al. (1989) conducted a case–control study in Shenyang, China, with data collected in interviews with 1249 patients with lung cancer and 1345 population-based controls. After adjustment for smoking, the relative risks were twice as high among those who reported living in smoky outdoor environments (RR, 2.3; 95% CI, 1.7–2.9 in men, and RR, 2.5; 95% CI, 1.8–3.5 in women) compared with subjects living in an environment that was not smoky. There were also associations with duration of residence within 200 m of industrial factories such as chemical and rubber plants, cement, glass, and asbestos factories, and ferrous and non-ferrous smelters in men, and wood and paper plants and ferrous and non-ferrous smelters in women.

In a subsequent publication on the case–control study of Xu et al. (1989), Xu et al. (1991) performed an additional analysis on residential distance from the industrial area. Soil levels of arsenic and other metals rose with increasing proximity to the Shenyang copper smelter, and elevated risks of lung cancer were found among men, but not women, living within 1 km of the smelter (OR, 3.0; 95% CI, 1.6–6.0).

Biggeri et al. (1996) used the data collected by Barbone et al. (1995) to better investigate the relationship of lung cancer with the four sources of air pollution (shipyard, iron foundry, incinerator, and city centre). Spatial models were used to evaluate the effect of sources of pollution on lung cancer adjusted for age, smoking habits, exposure to occupational carcinogens, and levels of PM. The excess relative risk at the city centre was 2.2 (P = 0.0098; CI not reported). At the incinerator source, the excess relative risk was 6.7 (P = 0.0098; CI not reported). [This is a rather large study with improved exposure assessment relative to earlier studies and detailed information on occupational exposure and smoking.]

A case–control study involving interviews with 117 women with lung cancer and 117 matched hospital controls was conducted in Taiwan, China, from 1992 to 1993 (Ko et al., 1997). Information on cigarette smoking and suspected risk factors for lung cancer, including residential distance from industrial plants, was collected by interview. Only a small proportion (9.4%) of female cases had smoked. Among non-smoking women, the odds ratio for the association of lung cancer with living near an industrial district for 20 years or longer was 2.7 (95% CI, 0.9–7.8) after adjustment for several covariates, including indoor air pollution from cooking. [This is a relatively small study with a rather crude exposure assessment.]

To investigate the relationship between petrochemical air pollution and lung cancer, Yang et al. (1999) conducted a matched case–control study among women who had died in Taiwan, China, from 1990 to 1994, using a similar design to Chiu et al. (2006); 399 lung cancer cases and 399 controls were matched by sex, year of birth, and year of death. The proportion of a municipality’s total population employed in the petrochemical manufacturing industry was used as an indicator of exposure to air emissions from this industry. For women who lived in municipalities with the highest level of petrochemical industry employment, the odds ratio was 1.66 (95% CI, 1.05–2.61) compared with women who lived in municipalities with the lowest petrochemical industry employment level after controlling for possible confounders. [The study shares the methodology and the limitations of the studies by Chiu et al. (2006) and Liu et al. (2008a).]

Petrauskaite et al. (2002) conducted a case–control study on lung cancer near an industry producing sulfuric acid and fertilizers in central Lithuania. Between 1967 and 1973, the levels of sulfuric acid exceeded 500 µg/m³ within 2 km of the industry and 100 µg/m³ more than 5 km away. A total of 277 men diagnosed with lung cancer during 1981–1991 and 1108 deceased controls, excluding deaths from respiratory cancer, were included. Information on residential history since 1960, smoking habits, and lifetime occupations and workplaces was obtained from questionnaires mailed to the next of kin. The relative risk of lung cancer associated with living within approximately 5 km of the plant was 1.02 (95% CI, 0.76–1.38) compared with never having lived in the area. No relationship with distance or with duration of residence was observed. [Since only deceased subjects were in the control group and exposure to air pollution is linked with increased cardiorespiratory mortality, exposure to air pollution may have been overrepresented in the control group, with underestimation of the effect. The Working Group noted a minor discrepancy in the odds ratios reported in the abstract and tables of this paper; the odds ratio from the abstract is shown here.]

Bessö et al. (2003) evaluated the association of exposure to industrial air pollution and lung cancer risk in a case–control study in the vicinity of a non-ferrous metal smelter in Sweden. The work was an extension of the study previously conducted by Pershagen (1985). The smelter started operations in 1930 and had very high emissions of arsenic and SO₂ in the early years. Among people who had died in 1961–1990 in the municipality where the smelter was located and who had not worked at the smelter, 316 lung cancer cases were identified and matched by sex and year of birth to 727 controls. Information on smoking habits, occupations, and residences was collected from questionnaires mailed to the next of kin and from registry data. Living close to the smelter was associated with a relative risk for lung cancer of 1.38 (95% CI, 0.89–2.14) among men after adjustment for smoking and occupational exposures. For women, however, no overall increased risk of lung cancer was observed (OR, 0.88; 95% CI, 0.48–1.62).

Pisani et al. (2006) conducted a case–control study in Lampang Province, Thailand, to assess the risk of lung cancer associated with exposures in the area, including power plants and coal mines, and to investigate possible interactions with genetic susceptibility. A total of 211 cases of lung cancer diagnosed in 1993–1995 among residents of the province were recruited at the provincial hospital. Community (n = 202) and hospital (n = 211) controls were frequency-matched to the cases by sex and age. Sociodemographic information, complete residential history, and characteristics of the household related to cooking and heating, occupational history, and history of tobacco smoking were obtained by interview. An air pollution exposure index was calculated for each village or township reported in residential histories, based on the linear distance from the power plants, the annual SO₂, NO₂, and TSP emissions from the power plant, and the percentage of time that wind blew from the power plant centre. For the highest category of estimated cumulative exposure to SO₂ and NO₂ emissions versus the lowest category, the overall odds ratio was 1.2 (95% CI, 0.7–2.0). The cumulative index of exposure to PM was not associated with lung cancer. [The controls included individuals hospitalized for causes related to air pollution exposure, such as cardiovascular disease, with a possibility of a bias to the null.]

(c) Studies with quantitative assessment of air pollutant exposure

Jedrychowski et al. (1990) reported the results of a case–control study of 1099 lung cancer deaths and 1073 age- and sex-matched control deaths from other, non-respiratory causes that occurred in 1980–1985 in Cracow, Poland. Information on occupation, smoking habits, and residency was collected from the next of kin. Exposure to outdoor air pollution was estimated from levels of TSP and SO₂ measured by an urban monitoring network from 1973 to 1980. In men exposed to the highest air pollution level (TSP > 150 µg/m³ and SO₂ > 104 µg/m³), the relative risk was 1.46 (95% CI, 1.06–1.99). In women exposed in the combined medium and high air pollution categories, the relative risk was 1.17 (95% CI, 0.70–1.96). The joint action of the risk factors of smoking, occupational exposure, and air pollution was found to fit a multiplicative model. [This is a large study with detailed information on occupational exposure and smoking, and improved exposure assessment relative to earlier case–control studies, which did not quantify exposure.]

Nyberg et al. (2000) conducted a population-based case–control study among men aged 40–75 years with incident lung cancer in 1985–1990 in Stockholm County, Sweden. A total of 1042 cases and 2364 controls were studied, with a response rate of more than 85%. Local annual source-specific air pollution levels were estimated by dispersion modelling of emission data for NO_x/NO₂ and SO₂ and linked to residential addresses. More details on the exposure assessment are available from Bellander et al. (2001). Average traffic-related NO₂ exposure over 10 years (lagged 20 years) was associated with a relative risk of 1.10 (95% CI, 0.97–1.23) for each 10 µg/m³ increase in NO₂. The relative risk for the top decile of NO₂ exposure was 1.44 (95% CI, 1.05–1.99). In contrast, no association was found for SO₂ from heating: the relative risk was 1.01 (0.98–1.03) for each 10 µg/m³ increase in SO₂. All the risk estimates were adjusted for age, year, tobacco smoking, socioeconomic status, residential radon, and occupational exposures. The relative risk for never-smokers exposed to NO₂ above the 90th percentile (> 29.3 µg/m³) versus below the first quartile (< 12.7 µg/m³) was 1.68 (95% CI, 0.67–4.19). [This is a large study with high-quality historical exposure assessment and detailed information on smoking.]

Hystad et al. (2013) investigated lung cancer incidence in relation to long-term exposure to outdoor air pollutants and proximity to major roads in a population-based case–control study. Annual residential exposure to fine PM (PM_2.5), NO₂, and ozone over a 20-year period was compared among 2390 incident lung cancer cases and 3507 population controls in eight Canadian provinces from 1994 to 1997. Residential exposure to air pollutants was estimated using self-reported residential histories from 1975 to 1994 and national spatial surfaces of outdoor air pollution compiled from satellite-based estimates (for PM_2.5 and NO₂) and a chemical transport model (for ozone) and then adjusted with historical annual air pollution monitoring data. Details of the exposure assessment are presented by Hystad et al. (2012). Hierarchical logistic regression models incorporated a comprehensive set of individual and geographical covariates. There was an increase in lung cancer incidence, with relative risks of 1.29 (95% CI, 0.95–1.76) with a 10-unit increase in PM_2.5 (µg/m³), 1.11 (95% CI, 1.00–1.24) with a 10-unit increase in NO₂ (ppb), and 1.09 (95% CI, 0.85–1.39) with a 10-unit increase in ozone (ppb). A subanalysis conducted in urban centres using exposures derived from fixed-site air pollution monitors supported the national results, with larger associations for NO₂ (RR, 1.34; 95% CI, 1.07–1.69) and PM_2.5 (RR, 1.33; 95% CI, 0.82–2.15) per 10-unit increase. An elevated relative risk was found among those living within 50 m of highways (RR, 1.23; 95% CI, 0.76–1.98) but not among those living near major roads. There was an increased risk of adenocarcinoma with an increase of 10 ppb in NO₂ exposure (OR, 1.17; 95% CI, 1.01–1.35) and an increased but non-significant risk with an increase of 10 µg/m³ in PM_2.5 exposure (OR, 1.27; 95% CI, 0.84–1.90). The odds ratio for PM_2.5 among never-smokers was 0.95 (95% CI, 0.38–2.34) based on 120 cases. [This is a large study, and notable strengths are the historical exposure estimation at an individual level, data for lung cancer subtypes, and adjustment for an extensive list of potential confounders, including known lung cancer risk factors. Despite the large size, the study had limited power to examine associations among never-smokers.]

(a) Cohort studies

(b) Case–control studies

Several case–control studies of lung cancer have evaluated risks among professional drivers and other outdoor occupations potentially exposed to air pollution. The case–control studies of truck drivers exposed to diesel exhaust have been reviewed in the IARC Monograph on diesel and gasoline engine exhausts (IARC, 2013), whereas the studies on other drivers (including broad groupings of drivers that sometimes included truck drivers) and other outdoor workers are briefly reviewed here.

Hansen et al. (1998) conducted a nationwide case–control study (1970–1989) based on employees, including 28 744 men with primary lung cancer and incidence density sampled matched controls (1:1 match). Employment histories were reconstructed back to 1964 for each study subject from the records of a nationwide pension scheme with compulsory membership, and socioeconomic status was derived from the individual job title taken from the national population registry. The adjusted odds ratio for lung cancer was 1.6 (95% CI, 1.2−2.2; 277 cases) for taxi drivers (considered to be the most highly exposed to outdoor air pollution), 1.3 (95% CI, 1.2–1.5; 972 cases) for bus and lorry drivers, and 1.4 (95% CI, 1.3–1.5; 1002 cases) for unspecified drivers. The risk of lung cancer increased significantly with increasing duration of employment as a driver.

Brüske-Hohlfeld et al. (1999) conducted a pooled analysis of two case–control studies of lung cancer in Germany on 3498 male cases with histologically or cytologically ascertained lung cancer and 3541 male population controls. Information about lifelong occupational and smoking history was obtained by interview. The group of professional drivers (e.g. trucks, buses, and taxis) showed an increased risk in western Germany (OR, 1.44; 95% CI, 1.18–1.76) but not in eastern Germany (OR, 0.83; 95% CI, 0.60–1.14) after adjustment for smoking and asbestos exposure.

Menvielle et al. (2003) investigated all lung cancer cases diagnosed between January 1993 and December 1995 (228 lung cancers) in New Caledonia and 305 population controls. Information on lifetime job history, smoking, and other potential risk factors was collected by interview. Among men, an excess risk of lung cancer was found for bus, lorry, and van drivers (OR, 2.7; 95% CI, 1.1–7.0; 13 exposed cases) after adjustment for age, ethnicity, and smoking.

Consonni et al. (2010) examined the relationship between occupation and lung cancer in a case–control study (2002–2005) in the Lombardy region of northern Italy, including 2100 incident lung cancer cases and 2120 randomly selected population controls. The odds ratio for bus and truck drivers was 1.23 (95% CI, 0.90–1.68) after adjustment for area of residence, age, smoking, and number of jobs held.

(c) Meta-analyses

Tsoi & Tse (2012) conducted a systematic review on the association between professional drivers and lung cancer, taking into consideration the potential confounding effect of cigarette smoking. They systematically searched all published cohort and case–control studies in English from January 1996 to January 2011. A total of 19 studies were included in the meta-analysis (8 cohort studies and 11 case–control studies), and a significantly increased risk of lung cancer (pooled smoking-adjusted RR, 1.18; 95% CI, 1.05–1.33) among professional drivers was observed after combining 4 cohort studies and 9 case–control studies. A higher pooled relative risk was observed among smoking-adjusted studies reporting 10 years or longer of employment (RR, 1.19; 95% CI, 1.06–1.34) compared with the study reporting shorter duration of employment (6 years; RR, 1.00; 95% CI, 0.92–1.09). [There was no information on never-smokers or non-smokers.]