Ambient air pollution and mortality: The role of socioeconomic conditions

Background: There is a vast body of literature covering the association between air pollution exposure and nonaccidental mortality. However, the role of socioeconomic status (SES) in this relationship is still not fully understood. Objectives: We investigated if individual and contextual SES modified the relationship between short-term exposure to ozone (O3), nitrogen dioxide (NO2), and particulate matter with aerodynamic diameter <10 µm (PM10) on cardiovascular, respiratory, and all nonaccidental mortality. Methods: We conducted a time-stratified case-crossover study. Analyses were based on information on 280,685 deaths from 2011 to 2015 in the city of São Paulo. Education was used as an individual SES, and information on the district of residence was used to build a contextual SES. Exposure to PM10, NO2, and O3 was accessed from monitoring stations and linked to each case based on the date of death. Conditional logistic regression models were used to estimate the effects of air pollutants, and interaction terms were added to access the effect modification of SES. Results: Individuals with lower education had an increased chance of dying for all nonaccidental outcomes (1.54% [0.91%, 2.14%]) associated with exposure to PM10. Individuals living in lower SES areas had an increased chance of dying for nonaccidental (0.52% [0.16%, 0.88%]), cardiovascular (1.17% [0.88%, 1.46%]), and respiratory (1.70% [0.47%, 2.93%]) causes owing to NO2 exposure. Conclusion: Exposure to air pollutants increases the chance of dying by nonaccidental, cardiovascular, and respiratory causes. Lower educational levels and living on lower contextual SES increased the risk of mortality associated with air pollution exposure.


Introduction
Air pollution has been identified as a serious public health problem since the first half of the 20th century.The main sources of emission in urban areas include industrial processes and the burning of fossil fuels. 1 Age groups most affected are children 2,3 and elderly. 4,5[7][8][9][10][11] Studies have evaluated how air pollution exposure and socioeconomic status (SES) correlate with mortality and if SES modifies the effects of air pollution on the population.The main hypotheses are that either poorer individuals are more exposed to air pollution directly in their jobs or living near industrial sources (environmental injustice) or that individuals with lower SES become more vulnerable to the harmful effects of air pollution because of limited access to healthy food and water, limited access to health care services, and worse lifestyle habits such as smoking and drinking. 6,12,13ome studies have already reported that both individual and area-level SES modify the effects of air pollution on the health of the population.A case-crossover study conducted in São Paulo, Brazil, reported that the effect of short-term exposure to NO 2 , SO 2, and CO on nonaccidental mortality was modified by individual-level SES (education). 7A similar finding was reported in a time-series study in China, where low educational levels modified the short-term effect of PM 10 on all-cause mortality. 14n ecological study in Wales, UK, reported a higher chance of dying owing to all-cause and respiratory diseases owing to PM 10 , PM 2.5, and NO 2 short-term exposure on low area-level SES, 15 similar to a case-crossover study in Paris, France, that reported a higher risk of all-cause mortality owing to both short-and long-term exposure to NO 2 among individuals living in most deprived census blocks. 6In a cohort conducted in the United States, it was reported that the association between longterm exposure to PM 2.5 and all-cause mortality was modified by both individual and area-level SES, with the lowest SES levels being the most vulnerable. 16On the other hand, some studies have reported associations between air pollution and adverse health effects but with no SES modification, [17][18][19][20][21] while in others the effect was higher among those with higher SES. 22,23iven the contradictory results observed in the literature and the scarcity of studies in Latin America that jointly address individual and area-level SES, the present study aims to contribute to the body of knowledge in this area by examining the role of individual and contextual SES as effect modifiers in the association between short-term exposure to air pollution and mortality in Brazil.Our study takes place in São Paulo, Brazil, a city with large socioeconomic disparities and high levels of air pollution.São Paulo has a lot of social inequality, having a Gini coefficient of 0.62 in 2010. 24Based on previous evidence, we have chosen PM 10 , O 3, and NO 2 as the air pollutants to be analyzed in relation to mortality by all nonaccidental causes, respiratory diseases, and cardiovascular diseases.

Study area
This study was carried out in the municipality of São Paulo, Brazil's largest city, with a population of over 12 million people in 2016 and a demographic density of 7960 inhabitants/km². 24

Mortality data
Daily deaths were collected by Programa de Aprimoramento das Informações de Mortalidade (PRO-AIM), São Paulo's municipal health secretary program, from 1 January 2011 to 31 December 2015.We obtained information from death certificates on individual age, sex, date of death, education, district of residence, and main cause of death according to international classification of diseases.The outcomes analyzed were deaths by all nonaccidental causes (A00-T98, Z00-Z99), owing to respiratory (J00-J98) and cardiovascular diseases (I00-I99).We excluded individuals less than 30 years old for all analyses.

Air pollution data
Data for air pollution were extracted directly from the Companhia Ambiental do Estado de São Paulo Monitoring Network System from 1 January 2010 to 31 December 2015.We included the year 2010 because deaths that happened in early 2011 had air pollution exposure in 2010.We obtained the daily mean value for PM 10 and NO 2 and the highest 8-hour moving average for O 3 in µg/m³ for each monitoring station.We considered using PM 2.5 in our analysis but the data was available only for 5 monitoring stations and had more than 20% missing data, which would compromise the results.Daily air pollutant measures were considered valid when at least 16 of the 24-hourly measurements had no errors.PM 10 was measured at 13 monitoring stations while O 3 at 12 and NO 2 at 9. Because the correlation between monitoring stations for each pollutant was very high (see Supplementary Material, Tables S1-S3, http://links.lww.com/EE/A266), when data were available for more than one monitoring station, the values were averaged to create a citywide pollutant concentration.

Meteorological data
Daily average temperature and relative humidity were obtained from the Instituto de Astronomia, Geofísica, e Ciências Atmosféricas da Universidade de São Paulo from 1 January 2010 to 31 December 2015 period.

Socioeconomic status
We used education attainment as an individual SES indicator which was divided into three categories: 0-8 ("low"), 9-11 ("middle"), and 12 or more years of study ("high").Details of the approach to calculate area-level SES can be found elsewhere. 25 summary, we created a socioeconomic index using information from the 2010 Census for each of the 96 districts of São Paulo.The index was constructed by applying principal components analysis with varimax rotation to the following variables: mean income; literacy rate; one-person housing; education (mean years of education for head of household); and housing conditions (proportion of residences not connected to sewage network).The 32 districts with the highest index were allocated in the "high" category, the following 32 into the "middle" category, and the lowest 32 into the "low" category.Individuals were then allocated to the SES of the district of their residence.

Study design and statistical analysis
To study the effects of air pollution in São Paulo, we conducted a time-stratified case-crossover design where each case acts as its own control.We used a time-stratified referent selection in which time was divided into fixed strata and days in each stratum were considered as referents.The control days were matched on the same rounded daily temperature in the same month and year when a case (death) happened, excluding 7 days before the event to eliminate the 6-day autocorrelation between all observations used in the analysis as well as seasonality. 26irst, we estimated the risk of mortality for nonaccidental, cardiovascular, and respiratory diseases owing to exposure to PM 10 , NO 2 , and O 3 through a multiple conditional logistic regression adjusting by age, sex, and relative humidity.Pollutants entered the models as continuous variables on single-pollutant models.Finally, we added a multiplicative interaction term, with high SES (both individual and area-level) as the reference category, adjusting for age, sex, and relative humidity for the same outcomes of the previous model.All results are reported as percent change in the odds ratios ([OR − 1]*100%) for an increase of 10 µg/m³ of the pollutant in the case period in relation to the control period, followed by the 95% confidence interval (95% CI) for each estimate.All statistical analyses were performed using Stata version 15.0.

Results
From 1 January 2011 to 31 December 2015, we had a total of 311,574 deaths for nonaccidental causes in São Paulo municipality for individuals over 30 years of age.We excluded individuals who had no information about educational level in the death certificates (30,889 or 9.9%) from all analyses.Because all other studied variables had no missing data, our final study population was 280,685 individuals.Subjects were mostly between 75 and 109 years (47.1%), between 0 and 8 years of study (70.9%), and lived in low area-SES districts (38.4%).Similar proportions were observed when broken down by cardiovascular and respiratory outcomes (Table 1).Mean age for all nonaccidental deaths was 71.1 years (SD = 15.3),similar to cardiovascular and respiratory outcomes.Women were on average 5.7 years older than men and less educated (12.6% of men had 12 or more years of study versus 7.9% among women).There were more females than males that lived in high SES districts (29.9 versus 25.7%).Mean age among the low-education group was 72.8 years, while in the high-education group it was 68.7 years.Mean age of individuals living in low-SES districts was 66.7 years while in high-SES districts was 76.4 years (data not shown).
In models that tested for effect modification of individual and area-level SES (Tables 4-6), we observed that the effect of O 3 on mortality did not seem to be modified by either individual or area-level SES (Table 4).
On the other hand, for PM 10 , we found that individual SES modified the effect of the air pollution-nonaccidental mortality association.Individuals who had 0-8 years of study had an additional 1.54% (95% CI = 0.91%, 2.14%) chance of dying for nonexternal causes, while individuals with 9-11 years of study had an additional 0.97% (95% CI = 0.15%, 1.79%) chance of dying for the same outcome, in a dose-response relationship (Table 5).
We also found a dose-response relationship of area-SES modifying the effect of increased exposure to NO 2 and nonaccidental, cardiovascular, and respiratory outcomes (Table 6).Individuals living in districts with low area SES had an additional chance of dying for nonaccidental (0.52%), cardiovascular (1.17%), and respiratory (1.70%) causes.

Discussion
This study found some evidence that individual SES may modify the effect of short-term exposure to PM 10 on nonaccidental mortality, and area-level SES modifies the effect of NO 2 on daily mortality from nonaccidental, cardiovascular, and respiratory outcomes.On the other hand, we did not observe the effect modification of individual or area-level SES in the association between O 3 and mortality.
Our findings agree with previous results reported in different places.A cohort study in Canada reported that individuals with  education lower than school had a higher risk of dying from nonaccidental causes associated with long-term exposure to PM 2.5 than those with more years of study (hazard ratio [HR] = 1.060; 95% CI = 1.042, 1.079). 27A time-series study in the United States reported a higher risk of death for cardiorespiratory causes in individuals with 0-12 years of schooling than those with more than 12 years associated with an increase of 10 µg/m³ in both short-and long-term PM 2.5 exposure (OR = 1.90; 95% CI = 1.60, 2.10 in lower education versus OR = 1.40; 95% CI = 1.20, 1.60 in higher education). 28ffect modification by SES in the association between shortterm NO 2 exposure and all-cause mortality has been documented in a case-crossover study in France, where individuals living in most deprived census blocks had statistically higher risks of dying compared with mid and less deprived blocks (excess risk % = 3.14; 95% CI = 1.41, 4.90). 6ther studies reported effect modification in the opposite direction, such as the Chinese cohort that found a higher risk ratio for cardiovascular mortality associated with an increase of 10 µg/m³ in long-term PM 10 exposure in the high education group (risk ratio = 1.52; 95% CI = 1.41, 1.65 in relation to risk ratio = 1.19; 95% CI = 1.15, 1.22 in the low education category). 2329,30 Comparison between estimates of health impacts from other studies may not be completely viable owing to differences in air pollution mixture, health care systems, climate, and definition of SES. 7 The plausibility of individuals with lower SES being more susceptible to air pollutants has been suggested previously and it was proposed that low SES may indicate worse housing conditions, which could also indicate higher indoor pollution, less access to healthy food and clean water, and unhealthy lifestyle such as smoking, leading to a predisposition to obesity, diabetes, and other comorbidities. 31ur study had some strengths, such as the use of both individual data collected from death certificates and contextual information based on an index constructed from principal components analysis that was able to socioeconomically divide the entire population of the municipality.Our study had also some limitations.It was not possible to measure each individual exposure to pollution because we could not access occupational   exposure or smoking habits.Our area-level SES index did not take into account the geographic extension or the population size in each area, which could have led to a misclassification in SES.

Conclusions
In conclusion, we report further evidence for the association between O 3 and NO 2 short-term exposure and nonaccidental, cardiovascular, and respiratory mortality.We could not find any effect modification of SES in the association between shortterm exposure to O 3 and mortality but we were able to detect effect modification by individual-level SES in the association between short-term exposure to PM 10 and nonaccidental mortality, where individuals with lower schooling had increased chance of dying.Area-level SES modified the association between short-term exposure to NO 2 and mortality by nonaccidental, cardiovascular, and respiratory outcomes, where individuals living on lower area-level SES had an increased chance of dying.

Table 3 .
Percentage change (95% confidence interval) of mortality associated with an increase of 10 µg/m³ in air pollutant concentration a a Adjusted by age, sex, and relative humidity.NO 2 , nitrogen dioxide; O 3 , ozone; PM 10 , particulate matter with aerodynamic diameter <10 µm.

Table 4 .
Percentage change (95% confidence intervals) of interaction between O 3 exposure and individual and area-level SES, for an increase of 10 μg/m

Table 5 .
Percentage change (95% confidence intervals) of interaction between PM 10 exposure and individual and area-level SES, for an increase of 10 μg/m³ SES, socioeconomic status.

Table 6 .
Percentage change (95% confidence intervals) of interaction between NO 2 exposure and individual and area-level SES, for an increase of 10 μg/m³ SES, socioeconomic status.