Introduction

There is heightened concern about the potential deleterious effects of ambient air pollution on health outcomes.1 A large number of studies throughout the world have evaluated the effects of short-term exposure to air pollution and have demonstrated a positive association between air pollution and health outcomes.2–4 In contrast, although the number of studies evaluating the effects of long-term exposure is increasing, most studies have been conducted in the US and Europe.5–10 These studies in Western countries have consistently demonstrated that long-term exposure to air pollutants is associated with all-cause and cardiopulmonary mortality,11 and some studies have also suggested associations with lung cancer (LC).9 12 13 However, whether there are specific individuals or subsets of patients at increased or decreased risk (effect modification) is less well documented.1 14

While many similarities exist in the constituents of air pollution around the world, the populations of Asia may differ from those of the US and Europe with respect to health status.15 Indeed, Asian populations have lower total cholesterol levels and body mass indices (BMI) than Western populations and also have lower rates of coronary heart disease mortality.16 17 These differences may result in different patterns between long-term air pollution and health outcomes. Therefore, studies that evaluate long-term exposure among Asian populations would yield significant insight into effect modification of air pollution exposure and help to direct environmental health policies in Asian countries.

Therefore, in the present study, we evaluated the health effects of long-term exposure to traffic-related pollution, indexed by NO₂ levels, including all-cause, cardiopulmonary and LC death rates in Shizuoka, Japan. We also examined whether the effect of air pollution was influenced by individual characteristics.

Methods

Participants

Individual data were extracted from participants of an ongoing cohort study: the Shizuoka elderly cohort.18 The Shizuoka prefecture is located in the approximate centre of Japan and has an area of 7780 km², a population of 3.8 million people and 1.3 million households. It has a southern coastline facing the Pacific Ocean and there are mountains exceeding 3000 m in altitude in the north.19 In December 1999, 22 200 residents were randomly chosen from all 74 municipalities in Shizuoka by stratifying sex and age groups (65–74; 75–84). Then, questionnaires were distributed to the participants, resulting in responses from 14 001 residents (response rate: 63%). The self-completed questionnaire included age, sex, smoking habit (non-smoker, ex-smoker or current smoker), body weight, height, medical history (hypertension, diabetes mellitus and so on), financial capability and other characteristics. Socioeconomic status was assessed by asking whether participants considered themselves to be financially capable, with the possible answers being non-capable or capable. Participants were followed-up in December 2002 and in March 2006 using the same questionnaire.

Because we modelled traffic-related pollution, indexed by NO₂, using the participants' baseline residential information, we excluded 271 participants whose residential information was not available and 286 participants who moved during the study period. Therefore, we targeted 13 444 participants. As shown in figure 1, 1329 participants were lost to follow-up during the period December 1999 to 2002 and 2264 participants were lost to follow-up during the period 2002 to March 2006. Finally, 8412 survivors and 1318 deaths were identified up to March 2006. In the analyses, we treated those who survived but did not return the questionnaire from December 2002 (n=121) and those who were lost to follow-up during the period 2002 to March 2006 (n=2264) as censored at 3 years. Survivors at March 2006 were treated as being censored at the end of the study.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Follow-up flow diagram of participants from 1999 to 2006.

Exposure data

To evaluate the health effects of traffic-related exposure, indexed by NO₂, we modelled individual mean NO₂ exposure during April 2000 to March 2006 using a land use regression (LUR) model, which has recently been used in several epidemiological studies.10 20–22 LUR models have been developed and used to model traffic pollutants within the framework of a geographic information system (GIS).22

The details are described elsewhere,19 however, we will briefly describe the modelling approach adopted. First, we constructed a model which best predicted the monitored levels of NO₂ using geographical variables. Mean exposure data (NO₂) during April 2000 to March 2006 were available from the environmental database managed by the National Institute for Environmental Studies in Japan. During the study period, 67 sampling sites for NO₂ were available. The observed annual mean NO₂ concentration across all 67 sites was 35.75 μg/m³ (SD of 11.28) and ranged from 14.66 to 68.24 μg/m³. Geographic variables of interest were listed as follows: road type (distance from major road, number of major roads within circular buffers); traffic intensity (road density of large roads (≥13 m) and of medium roads (5.5–13 m) within circular buffers, traffic counts (of cars, buses, trucks, big trucks, sum of all vehicles) on weekdays and on weekends within circular buffers); land use (building, farm, forest, water area within circular buffers); and physical component (population and housing density within circular buffers, elevation data and distance from coastline). All geographical variables were collected by the GIS software ArcGIS V.9.2 (ESRI Japan, Tokyo, Japan).

The geographical variables selected in the final model, which had the highest adjusted determination coefficient (R²), were as follows: road density of medium roads (1000), number of major roads (400), traffic counts of cars on weekends (100), distance from coastline and farmland (100). Figures in parentheses show circle buffer metres from the sampling sites. As described elsewhere,19 the validity of the exposure model was evaluated by crossvalidation, and the adjusted R² of the model was 0.54. In contrast, although we also modelled suspended particulate matter (particulate matter with an aerodynamic diameter less than 8 μm), the adjusted R² was quite low (R²=0.11).19

After the most appropriate model was constructed, each participant was assigned geographical information of the selected variables according to their geocoded residence. Then, individual NO₂ exposure was estimated using geographical information as prediction variables.

Outcome data

Vital statistics for determining the causes of death of participants were obtained from the database of the Ministry of Health, Labour and Welfare of Japan. We linked the deceased participants and the causes of death using birthday, sex and residential area. The underlying causes of death were coded according to the 10th International Classification of Disease (ICD-10). The numbers of deaths from all causes, cardiopulmonary disease (ICD-10 code: I10–70/J00–J99), LC (ICD-10 code: C33–C34) and other causes (deaths excluding cardiopulmonary and LC deaths) were determined. Furthermore, regarding cardiopulmonary disease mortality, we more finely classified the causes of death as follows: circulatory disease (I10–70), ischaemic heart disease (IHD) (I20–I25), other cardiac disease (such as dysrhythmias, heart failure, cardiac arrest) (I26–51), cerebrovascular disease (I60–69), other circulatory disease (causes of deaths excluding those specified in I10–70), pulmonary disease (J00–J99), pneumonia and influenza (J10–29), chronic obstructive pulmonary disease and allied conditions (J40–47), and other pulmonary disease (causes of deaths excluding those specified in J00–J99). Approval for this study was obtained from the Institutional Review Board of the Okayama Graduate School of Medicine, Dentistry and Pharmaceutical Sciences.

Statistical analysis

For each study participant, person-years were counted from the baseline to the date of death or to the date of censorship, whichever occurred first. First, we categorised modelled exposure into quartiles (≤20.3; 20.3–26.5; 26.5–32.1; >32.1 μg/m³) and calculated the number of deaths according to these quartiles. Then, the crude and adjusted HR for a 10 μg/m³ increase in NO₂ levels for all-cause or cause-specific mortality were estimated using the Cox proportional hazards model. At first, we adjusted for age and sex. Then, we adjusted for age, sex, smoking (non-smoker, ex-smoker or current smoker), BMI, hypertension, diabetes and financial capability. These potential confounders were decided a priori. BMI was defined as body weight (kg) divided by height squared (m²) and treated as a continuous variable.

To evaluate whether the effect of air pollution is influenced by age (65–74; 75–84), sex, smoking habit (non-smoker, ex-smoker or current smoker), BMI, hypertension, diabetes and financial capability, we separately stratified the participants and we estimated HR by mutually adjusting for other confounders in each stratum (ie, sex, smoking, BMI, hypertension, diabetes and financial capability in model for age category). A test of interaction was conducted by entering into the model multiplicative terms for interaction between the respective factors and exposure.

For sensitivity analyses, instead of financial capability, we used area mean income as socioeconomic status, and estimated the multivariate HR for all-cause, cardiopulmonary and LC mortality. The area mean income was derived by dividing total taxable gain of each municipality in 1998 by the number of taxpayers in the municipality in 1998. The data were obtained from the Statistics Bureau, Japan.23 In addition, some participants lived far from the sampling sites (eg, around 50 km), possibly resulting in extrapolation too far outside the NO₂ measurement area. Therefore, we restricted participants to those who lived within 25 km from the sampling sites and examined the air pollution effect on all-cause, cardiopulmonary and LC mortality. Furthermore, among the geographic variables selected in the final model, only distance from the coastline had a negative slope (−0.78/km).19 Consequently, some participants would be assigned negative exposure concentrations. Therefore, we also restricted participants to those who were assigned positive concentrations (over 0 μg/m³ of NO₂). Finally, we dropped distance from the coastline from the prediction variables in the LUR model and constructed an alternative model using the same model building strategy. Then, we examined the air pollution effect adapting the concentration estimated from the alternative model.

All CIs were estimated at the 95% level. SPSS V.14.0J software (SPSS Japan, Tokyo, Japan) was used for the analysis.

Results

The baseline characteristics of all participants (n=13 444) are shown in table 1. As expected, the mean BMI was much lower than that of participants in studies from Western countries (eg, 25–26 in the Harvard Six Cities Study24 and around 25 in the American Cancer Study25). The estimated NO₂ concentration ranged from −19 to 75 μg/m³. There were 709 participants who were assigned negative concentrations. Table 1 also shows the baseline characteristics of participants according to the endpoint. Those who were deceased tended to be older, male, a current smoker and to have diabetes. Those who were lost to follow-up during 1999–2002 tended to be older, current smokers, have diabetes, be financially non-capable and live in a more polluted area compared to survivors.

The baseline characteristics of participants and the number of deaths according to the exposure category are shown in table 2. We could not link 86 deaths with a cause of death, thus we excluded them from the subsequent analyses. In addition, we excluded those who were lost to follow-up during 1999–2002 (n=1329). Thus, 12 029 participants were included in the subsequent analyses. The estimated NO₂ concentrations among those 12 029 participants show a normal distribution, as shown in figure 2. In table 2, the mean age was similar across exposure categories and those who lived in more polluted areas tended to be financially capable. Total person-years across exposure categories were similar. While the number of cardiopulmonary deaths increased from the lowest exposure category to the highest, the number of all-cause and other-cause deaths decreased. Among cardiopulmonary mortality, the number of IHD and pneumonia and influenza deaths increased from the lowest exposure category to the highest.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Distribution of estimated nitrogen dioxide (NO₂) concentrations among participants.

The crude and adjusted HR following a 10 μg/m³ increase in NO₂ levels for all-cause and cause-specific mortality are shown in table 3. We found a multivariate HR of 1.02 (95% CI 0.96 to 1.08) for all-cause mortality. Regarding cause-specific mortality, we found a positive association between NO₂ levels and cardiopulmonary mortality (HR 1.16, 95% CI 1.06 to 1.26). On the other hand, no association between NO₂ levels and LC mortality was observed (HR 0.95, 95% CI 0.78 to 1.17), and NO₂ levels were significantly associated with a lower risk of mortality due to other causes (HR 0.92, 95% CI 0.85 to 0.99). Among cardiopulmonary mortality, the strongest association for increased risk was observed for IHD (HR 1.27, 95% CI 1.02 to 1.58). Although less precise, positive associations with other cardiac diseases (HR 1.19, 95% CI 0.97 to 1.47), cerebrovascular disease (HR 1.09, 95% CI 0.94 to 1.27), pneumonia and influenza (HR 1.18, 95% CI 0.96 to 1.45) and other pulmonary disease (HR 1.28, 95% CI 0.94 to 1.74) were indicated.

The results of stratified analyses for cardiopulmonary and LC mortality are shown in figures 3 and 4, respectively. The effect estimate for LC among non-smokers (HR 1.30, 95% CI 0.85 to 1.99) was higher and qualitatively different compared to former and current smokers (HR 0.87, 95% CI 0.69 to 1.09) and the p value (two-sided test) for interaction was 0.10 (figure 4). There were no meaningful differences among other stratifications.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Associations between estimated nitrogen dioxide (NO₂) levels and cardiopulmonary mortality in each subgroup of the Shizuoka elderly cohort (age category (<75; 75≤), smoking habit (non-smoker, ex-smoker or current smoker), sex, body mass index (BMI) (<21.8; 21.8≤), hypertension (HTN), diabetes (DM) and financial capability). The interactions with exposure were examined and p values (two-sided test) for statistical interaction are shown.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Associations between estimated NO₂ exposure and lung cancer mortality in each subgroup of the Shizuoka elderly cohort (age category (<75; 75≤), smoking habit (non-smoker, ex-smoker or current smoker), sex, body mass index (BMI) (<21.8; 21.8≤), hypertension (HTN), diabetes (DM) and financial capability). The interactions with exposure were examined and p values (two-sided test) for statistical interaction are shown.

The results of sensitivity analyses for cardiopulmonary mortality only are shown in table 4. When we used area mean income as a socioeconomic factor, no substantial changes were observed. Furthermore, when we restricted participants to those living within 25 km of sampling sites, the number of participants was reduced from 12 209 to 10 708. In consequence, the effect estimate was also almost the same. When we restricted participants to those who were assigned positive concentrations and the number of participants was reduced from 12 209 to 11 500. As a consequence, the effect estimate was again almost the same. Finally, we dropped distance from the coastline and adapted an alternative model. Although the adjusted R² declined (R²=0.45) and some HRs slightly fluctuated, the positive relationships between air pollution and mortality were not changed.

Discussion

We used data from an ongoing cohort study to evaluate the health effects of long-term exposure to traffic-related pollution, indexed by NO₂, in Shizuoka, Japan. We found an adverse effect of traffic-related pollution on cardiopulmonary mortality, especially on IHD mortality. No associations of traffic-related pollution with all-cause and LC mortality were observed, however, in the stratified analyses, non-smokers were indicated to have higher risks of LC mortality associated with air pollution exposure. This is the first cohort study to have evaluated the long-term effects of exposure to traffic-related pollution in Asia.

Even in a study population with a relatively low BMI, long-term air pollution was shown to increase the risk of cardiopulmonary mortality, especially of IHD, and, although estimates were less precise, of other cardiac disease (dysrhythmias, heart failure, cardiac arrest, etc), cerebrovascular disease, and pneumonia and influenza. These results are consistent with previous findings.8 26 The mean BMI in the present study was much lower than that in earlier studies from Western countries. Compared to previous studies, which used a 10 μg/m³ increase in NO₂ concentration as a traffic-related exposure indicator, the effect estimates of the present study were of similar magnitude and precision. For example, a study in Norway showed that NO₂ was related to myocardial infarction mortality (HR 1.08, 95% CI 1.03 to 1.12), respiratory mortality (HR 1.16, 95% CI 1.06 to 1.26) and cerebrovascular mortality (HR 1.04, 95% CI 0.94 to 1.15),8 and a study in France also demonstrated that NO₂ was associated with cardiopulmonary mortality (HR 1.27, 95% CI 1.04 to 1.56).12 Recently, a study in The Netherlands also showed that NO₂ was associated with respiratory mortality (HR 1.37, 95% CI 1.00 to 1.87).6 Together, these data suggest that air pollution causes the same adverse effects with the same magnitude, even in a low BMI population.

Overall, our study did not provide evidence that air pollution was related with LC mortality, in contrast to previous studies.9 12 13 This might be due to the short follow-up (6 years) of the present study compared to other studies (eg, 27 years in Nafstad et al,13 25 years in Filleu et al,12 and 16 years in Pope et al9). However, in the stratified analyses, although less precise, a positive association between air pollution and LC mortality was suggested among non-smokers. This finding is consistent with previous studies in The Netherlands and US, which indicated stronger association between air pollution and LC mortality in non-smokers compared to former and current smokers.6 9 This finding might reflect greater sensitivity of non-smokers to air pollution. The effects of air pollution on LC mortality in all participants will be evaluated in a future follow-up study.

In the present study, air pollution was significantly associated with a lower risk of mortality from other causes. This might be because participants in areas with higher air pollution were also of higher socioeconomic status (table 2).

Individual exposure modelling by LUR is an important area of this study, which made it possible to conduct a precise exposure assessment. Former studies used exposure data from monitoring stations instead of from individual measurement, which may cause underestimation of mortality risks by up to threefold.27 The present modelling has been properly validated and the R² value was moderate (R²=0.54),19 comparable to values found in certain LUR models for traffic-related pollution exposure,20 27–29 but lower compared to those in others.10 28 30 31

A limitation of our exposure modelling is that there were 709 participants who were assigned negative concentrations due to extrapolation far outside the NO₂ measurement area. Because the adopted model included distance from coastline as a prediction variable, the participants who lived in inland regions were assigned such extrapolated negative concentrations. However, since we considered the estimated concentration as an index for traffic-related air pollution and as a relative indicator, we analysed all participants. When we restricted participants to those living within 25 km of the sampling sites, we obtained similar effect estimates. Even when we restricted participants to those who were assigned positive concentrations, we still obtained similar effect estimates.

Another potentially confounding limitation of the present study was the inability to obtain rigorous socioeconomic factors (eg, occupation, education and income). Since our study participants were older, the potential confound of occupation is likely to be negligible. Indeed, when we used area mean income as a socioeconomic factor, the obtained HR values were almost the same. Thus, the potential confound of unknown socioeconomic status is unlikely to fully explain the present results. In addition, since the information about smoking pack-years (1 pack-year is defined as one pack smoked per day for 1 year) is not available, the possibility of a residual confounding factor remains. However, in the stratified analyses, the effect estimates for cardiopulmonary mortality in non-smokers were also significantly elevated. Thus, although the possibility of a residual confound cannot be completely ruled out, the present results seem robust.

In the present study, selection bias was non-negligible, because almost 10% of participants (1329 out of 13 444 participants) were lost to follow-up during 1999–2002 and did not contribute to person-years. According to table 1, participants who were lost to follow-up during that period were older, more likely to smoke, more likely to be diabetic and more likely to be financially non-capable, and they tended to live in more polluted areas compared to survivors, which might have caused underestimation of effect estimates.

When we linked deceased participants and the causes of death using birthday, sex and residential area, we could not link 83 deaths. This could be due to participants moving away or to coding error. However, it would probably be non-differential disease misclassification, leading to an underestimation of the elevated effects.

This is the first cohort study to evaluate the long-term exposure to traffic-related air pollution in a country other than in Europe or the US. In the present study, we used NO₂ as an indicator of traffic-related air pollution. Since fine particles <2.5 μm in diameter (PM_2.5) are considered to be a more sensitive indicator of air pollution than NO₂, further studies adopting PM_2.5 are expected to provide more evidence of adverse health effects of air pollution in Japan.

In conclusion, the present study suggests that long-term exposure to traffic-related air pollution increases the risk of cardiopulmonary mortality, especially of IHD mortality, even in a population with a relatively low BMI. The present findings also provide additional evidence that non-smokers have higher risks of LC mortality due to air pollution. Further research is also needed to examine the long-term effects of air pollution in various Asian populations.