Nyberg, Fredrik1; Gustavsson, Per3 4; Järup, Lars2 5; Bellander, Tom2; Berglind, Niklas2; Jakobsson, Robert3 4; Pershagen, Göran1 2
Epidemiologic studies from many countries have shown elevated risks of lung cancer in urban or industrially polluted areas, generally by up to 1.5 times, even when adjustment for smoking has been attempted. 1,2 Traffic-related air pollution is a growing concern today, but most of the available evidence relates to areas where motor vehicles were not the major source of air pollution. Nevertheless, studies on diesel-exposed occupational groups provide support for a causative role of traffic-related air pollution for lung cancer. 2 Recent population-based cohort studies with measured air pollution data have also indicated that lung cancer incidence is increased by 30–50% in areas with high ambient air pollution levels compared with areas with lower levels. 3–5
A major deficiency of many previous studies is the lack of individual long-term data on air pollution exposure. 1 It is also unclear which sources of urban air pollution may be of importance. In many instances, the lack of individual-level air pollution data is likely to have obscured much of the true range of individual exposure. The resulting limited exposure contrast has also hampered analyses of interactions with smoking and other known risk factors for lung cancer, even when such information was available.
The present study was conceived with the specific aim of exploring the possible association of lung cancer and urban air pollution by using geographical information system (GIS) techniques to assign individual exposures to ambient air pollution from oxides of nitrogen (NOx), nitrogen dioxide (NO2), and sulfur dioxide (SO2) from defined emission sources. These pollutants were chosen as suitable indicators of air pollution from road traffic and heating, which constituted the main local sources of air pollution. Individual data on smoking, occupational exposures, and some other risk factors were also collected and used for evaluation of confounding and possible interactions.
The study population comprised men 40 to 75 years of age who were residents of Stockholm County at any time between January 1, 1985 and December 31, 1990. An additional restriction was residence outside the county for at most 5 years between 1950 to 1990, a criterion approximately 70% fulfilled. In this study base, we identified 1,196 male lung cancer cases (ICD-7 code 162.1, diagnosed between January 1, 1985 and December 31, 1990) from the Stockholm County regional cancer registry, of whom 1,042 participated in the study (Table 1). Cases who were deceased and who were still alive were included. The diagnosis was based on histology for 78.3% and cytology for 20.4%.
Incidence density sampled controls were drawn in 1992 by random sampling from retrospective population registers covering Stockholm County, stratified on age (5-year categories) and calendar year of selection (1985 to 1990) of the cases. One control group (N = 1,274) was drawn from all individuals in the study population alive at the end of each selection year (“population” controls). As foreseen, more individuals in this group than among the cases were still alive at the time of data collection (Table 1). To allow an evaluation of possible bias from using proxy interviews for deceased individuals, primarily in our occupational analyses, 6 we recruited a second control group (N = 1,090), also frequency matched to the cases on vital status on December 31, 1990, using the Cause-of-Death Registry (mortality-matched controls). We excluded individuals who had died from smoking-related diagnoses 7 from this control group. 6
Data collection via postal questionnaire was performed from 1994 through 1996. The questionnaire was sent to living subjects, or to next-of-kin (primarily wives or children) for deceased study subjects. Several mail reminders, and follow-up telephone reminders and interviews ensured a high response rate (over 85%, Table 1). The questionnaire inquired about smoking and dietary intake of vegetables and fruits, as well as detailed occupational and residential histories. For the assessment of air pollution exposure, the addresses of all residences after 1950 inhabited for over 1 year were collected. When the questionnaire residence history from 1950 was not complete, parish offices and tax authorities provided additional data. For the classification of residential radon exposure, questions concerning building materials, house type, and ground contact of dwelling were also asked. 8 All collected data were truncated at the individual selection year.
The geocoding and air pollution exposure assessment methodology using GIS is described in detail elsewhere. 9 Briefly, the addresses were transformed into geographical coordinates using standard GIS computer software 10 in conjunction with a regional geographical address data base. 11 The reference point for assessment of air pollution exposure was a detailed regional emission database for 1993 12 (see also http://www.slb.mf.stockholm.se/) containing approximately 4,300 traffic-related line sources covering all roads with over 1,000 vehicles/24 hr (90% of the estimated emission from road traffic), as well as over 500 point sources (district heating facilities, industries, etc.). Limited diffuse emission sources (eg, air traffic and merchant vessels) are mapped as area sources, and population-density related sources (eg, local heating, work machines) as grid-sources (250 or 1,000 m grids).
In this study, the estimated contributions to the total ambient NOx/NO2 and SO2 levels from the relevant sources were used as markers for air pollution from road traffic and residential heating, respectively, based on source-specific emission data. These sources form the major part of ambient NOx/NO2 and SO2 levels, respectively, in Stockholm County. We assessed area-wide emissions of traffic-related air pollution (NOx and NO2 from road traffic) for three periods: the 1960s, 1970s, and 1980s. Data on the expansion of built-up areas in Stockholm County and the growth and distribution of road traffic was collected from 1960 through 1993 to reconstruct comparable historical emission databases based on the 1993 database. Similarly, for emissions of SO2 from heating sources, three corresponding historical emission databases were reconstructed using data on the sulfur content in oil and the development of district heating (point sources) and other energy plants. Available SO2 measurement data were also used to calibrate the model regarding average emission levels from grid-type sources (mainly local oil-fuelled residential heating).
Dispersion calculations for annual mean SO2 and NOx from these emission databases used a Gaussian model, 13,14 in resolutions up to 100 × 100 m. The NOx concentrations were transformed to NO2 data using a non-linear relation derived from measurements in Stockholm County in the early 1980s. The dispersion model calculations from the NOx/NO2 1980s database (extended for this purpose to include traffic-related as well as other sources to produce estimates of total NOx/NO2) was compared with actual measurements of NO2 at six roof-top or background sites. The modeled values in these points were within ±20% from observed annual means. As all available measurements for SO2 were used for model calibration, a similar validation was not possible for SO2, although the calibration as such implies that the model approaches the measured data.
Annual levels of SO2 and NOx/NO2 were computed for each year between 1950 and 1990 by linear extrapolation and interpolation from the three database values, based on historical traffic counts for NO2, and available trend data for SO2 concentrations in Stockholm. 9 For main streets in the city center, street contributions of NOx and NO2 concentrations were added to the roof concentrations. These contributions were assessed by dispersion calculations with a street canyon model 14 and summarized in a 50% addition at street level and 20% at mid-facade.
Finally, the air pollution data for relevant time periods were linked to the nearly 11,000 individual address coordinates of the study subjects, yielding exposure indices for each of the three air pollution indicators for each year.
The occupational history included information on company names and locations, occupations, and work tasks for work periods of at least 1 year. We classified occupations according to the Nordic occupational code (NYK-83). 15 Classification of overall exposure to known or suspected occupational lung carcinogens used a published job-exposure matrix and was based on an individual’s entire occupational history. 16,17 An occupational hygienist evaluated exposure to specific occupational carcinogens including diesel exhaust, other combustion products, and asbestos for each work period, case-by-case, assigning an intensity class and a probability of exposure for each work period and substance. We calculated the cumulative exposure for each factor as the product of the intensity, the probability, and the duration of exposure, summed over all work periods in the occupational history. 6 Subjects were also categorized as to predominantly blue or white collar work and approximate educational level implied by their occupational history, by matching NYK-83 job titles to Swedish socioeconomic level (SEI) codes. 18,19 The resulting socioeconomic variable represents a cross-classification of blue/white collar and low/high educational level. We estimated radon exposure for each residence from an equation predicting radon levels based on geographical radon risk level, building material and house type, obtained by regressing 9,002 measured houses from a nationwide Swedish radon study 8 on these variables. Time-weighted average radon exposure was calculated over all available residences 30 to 3 years before end of follow-up.
We controlled confounding from smoking by a categorical variable (never; former smokers since >2 years; current smokers of 1–10, 11–20, and >20 cigarettes daily on average) and continuous variables for years since quitting among former smokers and average amount smoked among current smokers, respectively (set to 0 for other subjects). Missing values for seven former smokers and two smokers of 1–10 cigarettes were replaced by the average corresponding value among controls.
Geocoded air pollution information was available from 1950 to each subject’s selection year, with at most 5 years of missing values for any individual, mainly due to residency outside Stockholm county and to less complete address data in early years. Since we considered extrapolation far back from the 1960s air-pollution database to be uncertain, we used only a 30-year period before the selection year to estimate air-pollution exposure for each individual (ie, 1955–1984 to 1960–1989 for selection year 1985 to 1990, respectively). We excluded the selection year since annual exposure values were used and individual exposure in the selection year varies depending on a subject’s exact selection date. In the 30-year study period for air pollution exposure thus defined, geocoded data were missing for only 159 residential exposure years (0.16%) among all 3,406 individuals; 75 individuals with at most 4 of 30 years missing. We calculated time-weighted average exposures over the study period and specific time windows. 20,21 Exposure-response relations were very similar for NOx and NO2 and only results for NO2 are presented, since routine monitoring of this pollutant is more widespread. Furthermore, the correlation between the 30-year estimated traffic-related averages of the two pollutants was 0.98.
We estimated relative risks (RR) and 95% confidence intervals (CI) by odds ratios from multiple unconditional logistic regression, using the program Stata®. 22 The full models were adjusted for matching variables (age and selection year) and potential confounders: smoking, radon, socioeconomic grouping, work in risk occupations and occupational exposure to diesel exhaust, other combustion products and asbestos. Categorical variables were coded with indicator (dummy) variables. Results using either control group were similar and we combined the two groups to provide optimal statistical stability. We calculated attributable risks based on category-specific relative risks according to standard formulae. 21
Relative risks for lung cancer associated with some risk factors are shown in Table 2. Smoking-related RRs ranged up to 34.6 (95% CI = 23.2–51.6) for current smokers with more than 20 cigarettes per day of average consumption. With adjustment for age in narrow 5-year intervals, used in the case-control matching, exposure intensity rather than duration was more strongly related to lung cancer risk among current smokers, and among former smokers the RR decreased with longer duration since quitting smoking (detailed data not shown). The RR associated with residential radon exposure was 1.13 (95% CI = 0.83–1.55) per 100 Bq/m, 3 assigning category means (see Table 2) as individual exposure level. Relative risks for three specific occupational exposures (diesel exhaust, other combustion products, and asbestos) were in the range 1.4–1.5. After adjustment for these exposures, the remaining relative risk for employment in risk occupations was 1.15 (Table 2). Low socioeconomic status was independently associated with increased risk. Variables for vegetable and fruit consumption were strongly protective in models that were adjusted only for the matching variables age and selection year (down to RR 0.3 for highest versus lowest consumption). These effects, however, largely disappeared when other risk factors (particularly smoking and socioeconomic status) were entered into the model. The dietary variables did not further confound the relation between air pollution and lung cancer. In addition, some subjects had missing dietary data, and thus we did not include the dietary variables in the final air pollution models.
The initial air pollution analyses utilized data covering the entire defined 30-year exposure period (Table 3). After we adjusted for the potential confounders identified, we found a weak effect for the 30-year average traffic-related NO2 exposure, whereas we found no increase in risk of lung cancer associated with long-term average SO2 exposure. In models incorporating both pollutants, the estimated effect of NO2 was stronger.
We further investigated time windows and lags for calculating the individual mean exposure. Continuous and dichotomized NO2 variables (representing linear component of trend and risk from extreme exposure, respectively) showed stronger positive associations with lung cancer risk than SO2 variables, rather consistently regardless of time window used and particularly when early exposure (ie, 3rd decade before selection) was included. With a 20-year lag (ie, using a 10-year average over 21–30 years ago), the effects for traffic-related NO2 exposure thus appeared stronger than for average exposure over the whole 30-year period, and showed a clearer dose-response (Table 4). When we examined exposures in the three different decades of the exposure period separately and jointly in regression models, 21 the results indicated that the earliest decade, 20 years before the selection year, was particularly important for lung cancer risk from traffic-related air pollution. The relative risk from traffic-related NO2 exposure 21–30 years ago appeared relatively independent of smoking habits, ie, suggestive of an almost multiplicative interaction between the risks from these exposures. Heavy smokers constituted an exception, where no risk from traffic-related air pollution was indicated, although the confidence interval is compatible with a similar RR in this group (Table 5).
Despite high exposure levels in the early years of the study period, heating-related SO2 showed little effect in any time window. The results were different from those obtained for NO2, despite the fact that the 30-year averages of estimated individual SO2 and NO2 exposure showed some correlation (Pearson’s correlation 0.64). Correlations were highest in the early years; for annual averages, they were around 0.7 for the years 1950–1968 and 0.5 for the years 1969–1990 and similar for cases and controls.
When the two different control groups were evaluated separately, results were similar. The point estimates for the 90 percentile of 10-year average traffic-related NO2 exposure 20 years before selection were 1.45 for “population” controls and 1.49 for mortality-matched controls as compared with 1.44 (95% CI = 1.05–1.99) when using both control groups. For the continuous variable, the estimates were 1.090 and 1.109, respectively, as compared with 1.096 (95% CI = 0.97–1.23). Thus, both control groups appeared to produce valid and equivalent results and were combined in the analyses.
Confounding from smoking seemed adequately controlled with the categorical variable, with only minor additional effect of adding continuous variables for average amount among current smokers and time since quitting among former smokers. A continuous variable for duration of smoking had no further effect on confounding control, probably because little correlation of air pollution exposure with smoking duration remained after stratification for age in 5-year intervals, smoking dose and subdivision of smokers into current and former smokers. A minor positive confounding effect by smoking included alone in the models tended to be balanced by minor negative confounding when adding the other risk factors.
This study suggests an increased risk of lung cancer from traffic-related air pollution, assessed by individual annual estimates of traffic-related ambient NO2 concentrations at the place of residence over a 30-year period, based on emission data and dispersion modeling. The clearest results were found for a time window covering the first of the three investigated exposure decades, ie, approximately 20 years in the past, which points to a considerable latency period. No effect was discernible for SO2 related to residential heating, neither for long-term average levels, nor for past time windows. This finding appears somewhat paradoxical, as SO2 levels were high in the past and NO2 levels low, whereas in recent years SO2 levels have decreased and NO2 levels increased appreciably. Despite these contrasting temporal trends, however, the estimated exposures to heating-related SO2 and traffic-related NO2 showed reasonably high correlation, mainly due to geographical covariation. Nonetheless, traffic-related NO2 rather than heating-related SO2 was consistently the stronger risk indicator, with a suggestion of a 20-year latency period, a pattern that would seem to argue against a spurious association.
The controls in this study were selected from population registers with complete coverage of the study base from which the cases emanated. The response rate was high, over 85% among both cases and controls. Differential misclassification of air pollution exposure between cases and controls is not likely, since residential data on street address and years are unlikely to be affected by differential reporting bias, data was collected from several sources to obtain complete residential histories for virtually all subjects, and air pollution modeling is independent of case-control status. Non-differential misclassification, on the other hand, is probable and would tend to bias estimates for continuous variables and the top category of categorical variables toward the null. 21 The stronger effect seen in the time window analysis with 20-year lag suggests the possibility of decreased misclassification of biologically relevant exposure when an appropriate time window is specified. Nonetheless, the exposure indicators used in this study are still likely to be subject to non-systematic measurement error if they do not exactly correspond to the “true” exposure but are proxies for one or several components of the complex air pollution mix. Notwithstanding, a major strength of the present study lies in the long-term air pollution exposure assessment, which was based on detailed historical emission data and was performed individually for a 30-year residence period for each subject. Misclassification of true individual exposure is thus likely to be less serious than in many previous studies with cruder, non-individual exposure assessment. Furthermore, the emission data allowed us to partition exposure according to sources and use source-specific NO2-levels as an indicator of traffic-related air pollution and source-specific SO2 as an indicator of air pollution from residential heating. The individual exposure contrast appears to have been sufficient to evaluate variations in risk - the ratio between the 90th and 25th percentiles was 2.0 to 2.3 for NO2 and 1.9 to 2.0 for SO2 (Tables 3 and 4), and the 30-year average ranged 11-fold for NO2 and almost 18-fold for SO2.
Expected relative risks for lung cancer were found for smoking 7,23 and radon, 8,24 and increased RRs were obtained for some well-known and suspected occupational risk factors, suggesting that questionnaire data were of good quality. Detailed results regarding occupational exposure are published elsewhere. 6 In crude analyses, protective effect estimates were obtained for vegetable and fruit consumption, but were no longer clearly apparent after detailed adjustment for other known risk factors. This confounding may partly reflect inadequate dietary reporting from proxies, leading to misclassification of these variables. The dietary variables did not confound the relation between air pollution and smoking. It is possible that overall dietary differences in our data, and possible confounding of air pollution associations, was described better by the socioeconomic and occupational variables. For the effect associated with traffic-related NO2, minor positive confounding from smoking tended to be balanced mainly by negative confounding when adding the other exposures. The degree of confounding was modest. Thus, although imprecision in measuring confounders may limit confounding control, residual confounding of importance seems unlikely in this study.
Not many studies of ambient air pollution and lung cancer risk have investigated several pollutant measures and few have considered both NO2 and SO2. Consistent with our results, two ecological studies have suggested that NO2 rather than SO2 is associated with regional differences in lung cancer mortality or incidence. 25,26 Similarly, a case-control study suggested that nitrogen oxides and carbon monoxide (city center, largely traffic-related), or ozone and particulates (incinerator area) were more likely to be responsible for the increased risk found in that study than SO2 (iron foundry area). 27,28 In a U.S. cohort study conducted among Seventh-Day Adventists in California, a strong relation for lung cancer incidence and mortality to 20-year averages of respirable particles (PM10) was observed among men; among women it was weaker. 29,30 Associations were similar also for ozone and SO2 among men and appeared stronger for SO2 among women. The gender differences appeared to be partially due to differences in exposure, mainly that males spent more time outdoors, particularly in the summer. 29,30 For NO2 exposure, a weak relation to lung cancer incidence was observed in one-pollutant models (eg, RR 1.5, 95% CI = 0.7–3.1 per 1.98 ppb NO2 among men), and slightly stronger effects on lung cancer mortality (RR 1.8, 95% CI = 0.9–3.6 among men and 2.8, 1.1–6.9 among women, per 1.98 ppb NO2). These estimates weakened further when other pollutants, including SO2, were introduced into the models. In the U.S. Six Cities study, the risk gradient across the six cities was more strongly associated with fine and sulfate particulate levels than with either SO2 and NO2 levels; the two latter were similarly correlated with risk. 4
Earlier studies used quantitative or semi-quantitative data on measured total ambient air pollution levels, whereas our study uses source-specific contributions from road traffic and residential heating emissions to population NO2 and SO2 exposure, respectively. If other emission sources are important in other localities, total NO2 and SO2 are likely to have a different interpretation as proxies for air pollution exposures. Furthermore, the use of fixed site monitors, as in the two cohort studies mentioned above, is likely to entail important non-differential misclassification of exposure, in particular for gaseous pollutants, such as SO2 and NO2, where local variation in emissions may produce sizeable variations in exposure levels.
When a restriction to NOx/NO2 from road traffic is made, as in this study, it is likely to represent not only traffic-related NOx/NO2 emissions but also may be a good proxy for other components of vehicle exhausts, including components of diesel exhaust and possibly fine or ultrafine particles, which have been suggested to be particularly important for mortality. For example, a study from Finland in an area where traffic is a main source of pollutants found correlation coefficients of 0.55–0.94 between NO2 and various particulate measures including PM10, black smoke and number concentrations of fine and ultra-fine particles. 31 We were not able to make direct analyses of particulate air pollution in this study because of lack of historical measurements, past emission data and validated dispersion models for particulates.
Interestingly, our study gives evidence for lung cancer risk related to several combustion sources, smoking being by far the strongest risk factor. In addition, we found an increased risk for occupational diesel exposure and occupational exposure to other combustion products, 6 providing some support for the relation with traffic-related air pollution reported here.
Lag or induction times for an effect of air pollution on lung cancer risk have not often been considered. An ecological study in an area with very low smoking rates investigated the effect of opening a steel mill that became the major air pollution source and found increased lung cancer mortality rates within 15 years. 32 Two case-control studies found increased risks associated with air pollution indices at the last place of residence, but since the average duration of residence was 30 years or more, these indices may represent both recent and long-term exposure. 27,33 Another case-control study suggested a stronger effect by ambient air pollution when allowing for a latency period of 20 years than when lifetime exposure was considered. 34 Most other case-control studies did not investigate this aspect of exposure in detail. 1
Of the approximately 10 cohort studies on ambient air pollution and lung cancer, the majority are older studies using an urban/rural exposure contrast. 1 Individual estimates of air pollution exposure were only made in one study, based on interpolation from fixed site monitoring stations. 3,29,30,35 Most cohort studies observed increased risk of lung cancer in the order of 1.5, surprisingly consistent and similar to the case-control studies. 1 Two recent U.S. studies with aggregate measured air pollution data suggest that the risk may be associated with fine or sulfate particulates.45 In the third study, with individual exposure estimates, an effect of particulates, as well as ozone, was seen mainly in males, whereas a strong effect of SO2 was seen in both genders. 29,30 NO2 showed less of an association. One often-emphasized advantage of cohort studies is that because exposure information is collected before disease occurrence, differential bias in the exposure assessment is very unlikely. This advantage does not really apply in relation to our case-control study, however. Detailed exposure assessment using the methodology we have employed is unlikely to be affected by case-control status and represents a substantial improvement over most previous attempts to estimate long-term exposure to air pollution for individuals.
Some previous studies have suggested a multiplicative interaction between air pollution exposure and smoking, while others have been more consistent with an additive relation. 36 Our results are more compatible with a multiplicative interaction, except among heavy smokers, where no clear effect of traffic-related air pollution was evident. Similar weaker effects among heavy smokers have been observed for occupational arsenic 37 and residential radon 38,39 exposure. Possible explanations include a thickening of the bronchial mucosa, 40 a selection bias similar to the “healthy worker survivor effect” for maintaining high tobacco consumption, or chance.
Since exposure is widespread, the public health impact of a 50% increase in lung cancer risk among heavily exposed in the general population from traffic-related air pollution, as suggested by this study, may be important, and lower risk increases at more common moderate exposures potentially play a large role, too. An attributable risk calculation based on exposure above the 25th percentile suggests that the proportion of lung cancer among smoking and non-smoking males 40–75 years old in Stockholm County related to traffic-related air pollution exposure 20 years earlier could be as high as 10%.
We thank Tage Jonsson and Malin Pettersson for development of the retrospective emission databases, and Camilla Bengtsson, Anna Boberg, Eva-Britt Gustavsson, Marcus Hugosson, Elin Junghahn, Ulla Klinga, Cecilia Rudengren, Patrik Schéele, Ewa Skarke, Agneta Wahlbom and Kerstin Åström for assistance with data collection and analyses.
1. Katsouyanni K, Pershagen G. Ambient air pollution exposure and cancer. Cancer Causes Control 1997; 8: 284–291.
2. Cohen AJ, Pope CA. Lung cancer and air pollution. Environ Health Perspect 1995; 103 (suppl 8): 219–224.
3. Abbey DE, Lebowitz MD, Mills PK, Petersen FF, Beeson WL, Burchette RJ. Long-term ambient concentrations of particulates and oxidants and development of chronic disease in a cohort of nonsmoking California residents. Inhal Toxicol 1995; 7: 19–34.
4. Dockery DW, Pope CA 3d, Xu X, Spengler JD, Ware JH, Fay ME, Ferris BG Jr, Speizer FE. An association between air pollution and mortality in six U. S. cities. N Engl J Med 1993; 329: 1753–1759.
5. Pope CA 3d, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, Heath CW Jr. Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am J Respir Crit Care Med 1995; 151: 669–674.
6. Gustavsson P, Jakobsson R, Nyberg F, Pershagen G, Järup L, Schéele P. Occupational exposure and lung cancer risk - a population-based case-referent study in Sweden. Am J Epidemiol (in press).
7. Carstensen JM, Pershagen G, Eklund G. Mortality in relation to cigarette and pipe smoking: 16 years’ observation of 25,000 Swedish men. J Epidemiol Community Health 1987; 41: 166–172.
8. Pershagen G, Åkerblom G, Axelson O, Clavensjö B, Damber L, Desai G, Enflo A, Lagarde F, Mellander H, Svartengren M, Swedjemark GA. Residential radon exposure and lung cancer in Sweden. N Engl J Med 1994; 330: 159–164.
9. Bellander T, Järup L, Berglind N, Gustavsson P, Jonson T, Nyberg F, Pershagen G. Using Geographic Information System techniques to assess individual historical exposure to air pollution in a case-control study of lung cancer in Stockholm. Stockholm, Sweden: Miljömedicinska enheten, Stockholms läns landsting [Department of Environmental Health, Stockholm County Council], in press.
10. MapInfo Professional (TM) User’s Guide. Troy, New York: MapInfo Corporation, 1995.
11. Tätort 2000 Computerized geographical database of Swedish urban areas. Stockholm, Sweden: Kartcentrum, Lantmäteriverket, National Land Survey of Sweden, 1996.
12. Ekström A, Hadenius A, Jonson T. Emissionsdatabas 93 - en dokumentation (Emission database 93 - a documentation) (Swedish). Report 2:95. Stockholm, Sweden: Stockholm Air Quality and Noise Analysis, 1995; see also http://www.slb.mf.stockholm.se/.
13. Swedish Meteorological and Hydrological Institute. Airviro user documentation. Norrköping, Sweden: SMHI, 1997.
14. Swedish Meteorological and Hydrological Institute. Technical description of the dispersion models, Indic Airviro. Norrköping, Sweden: SMHI, 1993.
15. Nordisk yrkesklassificering (Nordic standard occupational classification) (Swedish). Stockholm, Sweden: Arbetsmarknadsstyrelsen (Swedish National Labour Market Board), 1983.
16. Boffetta P, Kogevinas M, Simonato L, Wilbourn J, Saracci R. Current perspectives on occupational cancer. Int J Occup Environ Health 1995; 1: 315–325.
17. Ahrens W, Merletti F. A standard tool for the analysis of occupational lung cancer in epidemiological studies. Int J Occup Environ Health 1998; 4: 236–240.
18. Socioekonomisk indelning (Swedish socioeconomic classification) (Swedish). Stockholm, Sweden: SCB (Statistics Sweden), 1983.
19. Yrkesklassificeringar i FoB 85 enligt Nordisk yrkesklassificering (NYK) och Socioekonomisk indelning (SEI): alfabetisk version [Occupations in population and housing census 1985 (FoB 85) according to Nordic standard occupational classification (NYK) and Swedish socio-economic classification (SEI): alphabetical version] (Swedish). Stockholm, Sweden: SCB (Statistics Sweden), 1989.
20. Checkoway H, Pearce N, Hickey JL, Dement JM. Latency analysis in occupational epidemiology. Arch Environ Health 1990; 45: 95–100.
21. Rothman KJ, Greenland S. Modern Epidemiology, 2nd ed. Philadelphia: Lippincott-Raven Publishers, 1998.
22. StataCorp. Stata Statistical Software: Release 5.0. College Station, TX: Stata Corporation, 1997.
23. International Agency for Research on Cancer. IARC Monographs on the Evaluation of the Carcinogenic Risk of Chemicals to Humans. vol 38. Tobacco Smoking. Lyon: International Agency for Research on Cancer, 1986.
24. Lubin JH, Boice JDJ. Lung cancer risk from residential radon: meta-analysis of eight epidemiologic studies. J Natl Cancer Inst 1997; 89: 49–57.
25. Choi KS, Inoue S, Shinozaki R. Air pollution, temperature, and regional differences in lung cancer mortality in Japan. Arch Environ Health 1997; 52: 160–168.
26. Tango T. Effect of air pollution on lung cancer: a Poisson regression model based on vital statistics. Environ Health Perspect 1994; 102 (suppl 8): 41–45.
27. Barbone F, Bovenzi M, Cavallieri F, Stanta G. Air pollution and lung cancer in Trieste, Italy. Am J Epidemiol 1995; 141: 1161–1169.
28. Biggeri A, Barbone F, Lagazio C, Bovenzi M, Stanta G. Air pollution and lung cancer in Trieste, Italy: spatial analysis of risk as a function of distance from sources. Environ Health Perspect 1996; 104: 750–754.
29. Beeson WL, Abbey DE, Knutsen SF. Long-term concentrations of ambient air pollutants and incident lung cancer in California adults: Results from the AHSMOG study. Review. Environ Health Perspect 1998; 106: 813–822.
30. Abbey DE, Nishino N, McDonnell WF, Burchette RJ, Knutsen SF, Beeson WL, Yang JX. Long-term inhalable particles and other air pollutants related to mortality in nonsmokers. Am J Respir Crit Care Med 1999; 159: 373–382.
31. Pekkanen J, Timonen KL, Ruuskanen J, Reponen A, Mirme A. Effects of ultrafine and fine particles in urban air on peak expiratory flow among children with asthmatic symptoms. Environ Res 1997; 74: 24–33.
32. Archer VE. Air pollution and fatal lung disease in three Utah counties. Arch Environ Health 1990; 45: 325–334.
33. Jedrychowski W, Becher H, Wahrendorf J, Basa-Cierpialek Z. A case-control study of lung cancer with special reference to the effect of air pollution in Poland. J Epidemiol Community Health 1990; 44: 114–120.
34. Jockel KH, Ahrens W, Wichmann HE, Becher H, Bolm-Audorff U, Jahn I, Molik B, Greiser E, Timm J. Occupational and environmental hazards associated with lung cancer. Int J Epidemiol 1992; 21: 202–213.
35. Mills PK, Abbey DE, Beeson WL, Petersen F. Ambient air pollution and cancer in California Seventh-Day Adventists. Arch Environ Health 1991; 46: 271–280.
36. Pershagen G, Simonato L. Epidemiological evidence on outdoor air pollution and cancer. In: Tomatis L, ed. Indoor and Outdoor Air Pollution and Human Cancer. Berlin: Springer-Verlag, 1993; 135–148.
37. Järup L, Pershagen G. Arsenic exposure, smoking, and lung cancer in smelter workers—a case-control study [published erratum appears in Am J Epidemiol 1992;136: 1174]. Am J Epidemiol 1991; 134: 545–551.
38. Svensson C, Pershagen G, Klominek J. Lung cancer in women and type of dwelling in relation to radon exposure. Cancer Res 1989; 49: 1861–1865.
39. Axelson O, Sundell L. Mining, lung cancer and smoking. Scand J Work Environ Health 1978; 4: 46–52.
40. Walsh PJ. Radiation dose to the respiratory tract of uranium miners—a review of the literature. Environ Res 1970; 3: 14–36.
© 2000 Lippincott Williams & Wilkins, Inc.