Multicity study of air pollution and mortality in Latin America (the ESCALA study)
Isabelle Romieu,Nelson Gouveia,Luis A Cifuentes,Antonio Ponce de Leon,Washington Junger,Jeanette Vera,Valentina Strappa,Magali Hurtado-Díaz,Victor Miranda-Soberanis,Leonora Rojas-Bracho,Luz Carbajal-Arroyo,Guadalupe Tzintzun-Cervantes,HEI Health Review Committee
Abstract:Introduction: The ESCALA* project (Estudio de Salud y Contaminación del Aire en Latinoamérica) is an HEI-funded study that aims to examine the association between exposure to outdoor air pollution and mortality in nine Latin American cities, using a common analytic framework to obtain comparable and updated information on the effects of air pollution on several causes of death in different age groups. This report summarizes the work conducted between 2006 and 2009, describes the methodologic issues addressed during project development, and presents city-specific results of meta-analyses and meta-regression analyses. Methods: The ESCALA project involved three teams of investigators responsible for collection and analysis of city-specific air pollution and mortality data from three different countries. The teams designed five different protocols to standardize the methods of data collection and analysis that would be used to evaluate the effects of air pollution on mortality (see Appendices B-F). By following the same protocols, the investigators could directly compare the results among cities. The analysis was conducted in two stages. The first stage included analyses of all-natural-cause and cause-specific mortality related to particulate matter < or = 10 pm in aerodynamic diameter (PM10) and to ozone (O3) in cities of Brazil, Chile, and México. Analyses for PM10 and O3 were also stratified by age group and O3 analyses were stratified by season. Generalized linear models (GLM) in Poisson regression were used to fit the time-series data. Time trends and seasonality were modeled using natural splines with 3, 6, 9, or 12 degrees of freedom (df) per year. Temperature and humidity were also modeled using natural splines, initially with 3 or 6 df, and then with degrees of freedom chosen on the basis of residual diagnostics (i.e., partial autocorrelation function [PACF], periodograms, and a Q-Q plot) (Appendix H, available on the HEI Web site). Indicator variables for day-of-week and holidays were used to account for short-term cyclic fluctuations. To assess the association between exposure to air pollution and risk of death, the PM10 and O3 data were fit using distributed lag models (DLMs). These models are based on findings indicating that the health effects associated with air pollutant concentrations on a given day may accumulate over several subsequent days. Each DLM measured the cumulative effect of a pollutant concentration on a given day (day 0) and that day's contribution to the effect of that pollutant on multiple subsequent (lagged) days. For this study, exposure lags of up to 3, 5, and 10 days were explored. However, only the results of the DLMs using a 3-day lag (DLM 0-3) are presented in this report because we found a decreasing association with mortality in various age-cause groups for increasing lag effects from 3 to 5 days for both PM10 and O3. The potential modifying effect of socioeconomic status (SES) on the association of PM10 or O3 concentration and mortality was also explored in four cities: Mexico City, Rio de Janeiro, São Paulo, and Santiago. The methodology for developing a common SES index is presented in the report. The second stage included meta-analyses and metaregression. During this stage, the associations between mortality and air pollution were compared among cities to evaluate the presence of heterogeneity and to explore city-level variables that might explain this heterogeneity. Meta-analyses were conducted to combine mortality effect estimates across cities and to evaluate the presence of heterogeneity among city results, whereas meta-regression models were used to explore variables that might explain the heterogeneity among cities in mortality risks associated with exposures to PM10 (but not to O3). Results: The results of the mortality analyses are presented as risk percent changes (RPC) with a 95% confidence interval (CI). RPC is the increase in mortality risk associated with an increase of 10 microg/m3 in the 24-hour average concentration of PM10 or in the daily maximum 8-hour moving average concentration of O3. Most of the results for PM10 were positive and statistically significant, showing an increased risk of mortality with increased ambient concentrations. Results for O3 also showed a statistically significant increase in mortality in the cities with available data. With the distributed lag model, DLM 0-3, PM10 ambient concentrations were associated with an increased risk of mortality in all cities except Concepci6n and Temuco. In Mexico City and Santiago the RPC and 95% CIs were 1.02% (0.87 to 1.17) and 0.48% (0.35 to 0.61), respectively. PM10 was also significantly associated with increased mortality from cardiopulmonary, respiratory, cardiovascular, cerebrovascular-stroke, and chronic obstructive lung diseases (COPD) in most cities. The few nonsignificant effects generally were observed in the smallest cities (Concepción, Temuco, and Toluca). The percentage increases in mortality associated with ambient O3 concentrations were smaller than for those associated with PM10. All-natural-cause mortality was significantly related to O3 in Mexico City, Monterrey, São Paulo and Rio de Janeiro. Increased mortality risks for some specific causes were also observed in these cities and in Santiago. In the analyses stratified by season, different patterns in mortality and O3 were observed for cold and warm seasons. Risk estimates for the warm season were larger and significant for several causes of death in São Paulo and Rio de Janeiro. Risk estimates for the cold season were larger and significant for some causes of death in Mexico City, Monterrey, and Toluca. In an analysis stratified by SES, the all-natural-cause mortality risk in Mexico City was larger for people with a medium SES; however we observed that the risk of mortality related to respiratory causes was larger among people with a low SES, while the risk of mortality related to cardiovascular and cerebrovascular-stroke causes was larger among people with medium or high SES. In São Paulo, the all-natural-cause mortality risk was larger in people with a high SES, while in Rio de Janeiro the all-natural-cause mortality risk was larger in people with a low SES. In both Brazilian cities, the risks of mortality were larger for respiratory causes, especially for the low- and high-SES groups. In Santiago, all-natural-cause mortality risk did not vary with level of SES; however, people with a low SES had a higher respiratory mortality risk, particularly for COPD. People with a medium SES had larger risks of mortality from cardiovascular and cerebrovascular-stroke disease. The effect of ambient PM10 concentrations on infant and child mortality from respiratory causes and lower respiratory infection (LRI) was studied only for Mexico City, Santiago, and São Paulo. Significant increased mortality risk from these causes was observed in both Santiago (in infants and older children) and Mexico City (only in infants). For O3, an increased mortality risk was observed in Mexico City (in infants and older children) and in São Paulo (only in infants during the warm season). The results of the meta-analyses confirmed the positive and statistically significant association between PM10 and all-natural-cause mortality (RPC = 0.77% [95% CI: 0.60 to 1.00]) using the random-effects model. For mortality from specific causes, the percentage increase in mortality ranged from 0.72% (0.54 to 0.89) for cardiovascular disease to 2.44% (1.36 to 3.59) for COPD, also using the random-effects model. For O3, significant positive associations were observed using the random-effects model for some causes, but not for all natural causes or for respiratory diseases in people 65 years or older (> or = 65 years), and not for COPD and cerebrovascular-stroke in the all-age and the > or = 65 age groups. The percentage increase in all-natural-cause mortality was 0.16% (-0.02 to 0.33). In the meta-regression analyses, variables that best explained heterogeneity in mortality risks among cities were the mean average of temperature in the warm season, population percentage of infants (< 1 year), population percentage of children at least 1 year old but < 5 years (i.e., 1-4 years), population percentage of people > or = 65 years, geographic density of PM10 monitors, annual average concentrations of PM10, and mortality rates for lung cancer. Conclusions: The ESCALA project was undertaken to obtain information for assessing the effects of air pollutants on mortality in Latin America, where large populations are exposed to relatively high levels of ambient air pollution. An important goal was to provide evidence that could inform policies for controlling air pollution in Latin America. This project included the development of standardized protocols for data collection and for statistical analyses as well as statistical analytic programs (routines developed in R by the ESCALA team) to insure comparability of results. The analytic approach and statistical programming developed within this project should be of value for researchers carrying out single-city analyses and should facilitate the inclusion of additional Latin American cities within the ESCALA multicity project. Our analyses confirm what has been observed in other parts of the world regarding the effects of ambient PM10 and 03 concentrations on daily mortality. They also suggest that SES plays a role in the susceptibility of a population to air pollution; people with a lower SES appeared to have an increased risk of death from respiratory causes, particularly COPD. Compared with the general population, infants and young children appeared to be more susceptible to both PM10 and O3, although an increased risk of mortality was not observed in these age groups in all cities. (ABSTRACT TRUNCATED)