Abstract for the 2017 HEI Annual Conference
Mortality and Morbidity Effects of Long-Term Exposure to Low-Level PM2.5, Black Carbon, NO2, and O3: An Analysis of European Cohorts
University of Utrecht, Utrecht, The Netherlands
Background Epidemiological cohort studies have consistently found associations between long-term exposure to outdoor air pollution and a range of morbidity and mortality endpoints. Recent evaluations by the World Health Organization and the Global Burden of Disease study have suggested that these associations may be non-linear, and persist at very low concentrations. However, uncertainty about the shape of the concentration response function exists especially for the low and high end of the concentration distribution, partly related to the scarcity of observations in particularly the low range.
Methods In this study we focus on analyses contributing to knowledge about health effects of spatially resolved air pollution concentrations at low concentrations, defined as less than current EU, EPA and WHO Limit Values or guidelines for fine particles with an aerodynamic diameter of less than 2.5μm (PM2.5), nitrogen dioxide (NO2) and Ozone (O3). Studies have focused especially on PM2.5, but increasingly associations with NO2 are reported, particularly in studies that accounted for the fine spatial scale variation of NO2. Very few studies have evaluated long-term morbidity and mortality effects of ozone. We address the issue of health effects at low air pollution levels by performing targeted analyses of all-cause and cause-specific mortality and morbidity endpoints within selected cohorts of the ESCAPE study and a Danish nurse cohort with detailed individual data (~380,000 subjects) and in 7 very large European administrative cohorts (~35 million subjects). The analysis focuses on the pollutants PM2.5, NO2, and O3, but also exploits the rich monitoring data of black carbon (BC) available from the ESCAPE study with high spatial resolution.
Results Our exposure assessment will be finished by the end of year 1 of the study (May 1, 2017). The details are shown in a companion abstract, “Air pollution exposure assessment for the ELAPSE project using hybrid LUR models”.
Currently the project partners are processing the data available in their cohorts so that the follow up for mortality, cancer and cardiovascular events is extended until 2013 as a minimum (additional five years compared to the original average 13 years of follow-up in the ESCAPE project). We are also obtaining additional residential addresses histories. This is currently in progress within the respective cohorts. We have started preparations of common statistical analysis scripts to be used by all data analysts. A number of scripts written in STATA and R have been developed and are currently being tested at Utrecht University using a dummy dataset of 10M subjects prepared by profs. Evi Samoli and Klea Katsouyanni from the Athens University. Remote secure access to the UU servers is being organized so that analysts involved in the pooled cohort can perform analyses without physically having to travel to Utrecht. Details can be found in a companion abstract, "Statistical methods for investigating the effects of long-term exposure to low air pollutant concentrations in the ELAPSE project using data from 11 pooled European cohorts and 7 administrative cohorts".
Conclusions There are no conclusions yet from this study.
Air pollution exposure assessment for the ELAPSE project using hybrid LUR models
Bert Brunekreef1, Kees de Hoogh2,3, Jei Chen1, Gerard Hoek1, John Gulliver4, Ole Hertel5
1Institute for Risk Assessment Sciences, Utrecht University, Utrecht, the Netherlands, 2Swiss Tropical and Public Health Institute, Basel, Switzerland, 3University of Basel, Basel, Switzerland, 4MRC-PHE Centre for Environment and Health, Department of Epidemiology and Biostatistics, Imperial College London, London, United Kingdom, 5Department of Environmental Science, Faculty of Science and Technology, Aarhus University, Roskilde, Denmark.
Background In order to investigate associations between air pollution and adverse health effects fine spatial air pollution surfaces are needed to provide cohorts with exposures. In the ELAPSE project we developed hybrid LUR models for multiple pollutants and linked these to 11 cohorts plus 7 administrative cohorts in 10 countries for a total of 35 million participants.
Methods Europe-wide hybrid land use regression models were developed for 2010 estimating annual mean PM2.5, NO2, O3 and BC (including cold and warm season estimates for O3). The models developed were based on AIRBASE routine monitoring data for PM2.5, NO2 and O3, and ESCAPE monitoring data for BC and incorporated land use and traffic data supplemented with satellite observations and dispersion model estimates as additional predictor variables. Universal kriging was performed on the residual spatial variation. One model was developed using all sites (100%). To evaluate the robustness of the models, five more models were developed, each built on 80% of the monitoring sites with the remaining 20% used for validation (sites selected at random, but stratified by site type and country). Models were applied to a 100*100 m grids across Europe to allow for exposure assignment for all ELAPSE cohorts. To evaluate the stability of the model’s spatial structure over time, separate models were developed for different years, depending on the number of monitoring sites (NO2 and O3; 2000 and 2005, PM2.5; 2013).
Results Currently we can present the 2010 NO2 and PM2.5 models. The NO2 and PM2.5 main models (100% sites) explained respectively 64% (58% LUR + 6% kriging) and 75% (59% LUR + 16% kriging) of spatial variation in the measured concentrations at 2400 NO2 and 546 PM2.5 monitoring sites. The validation R2 ranged from 0.606 to 0.657 for NO2 and 0.677 to 0.797 for PM2.5. Dispersion model estimates, road density, nature, ports and residential area were predictor variables in the NO2 main model. The PM2.5 main model consisted of satellite derived and dispersion model estimates, altitude, road density, nature, ports and residential area. Kriging proved an efficient technique to explain a part of residual spatial variation.
Conclusions We were able to develop robust NO2 and PM2.5 hybrid LUR models to provide exposure estimates for all cohort participants in the ELAPSE project. Model development of O3 and BC and evaluation of the stability of the spatial structure of the models over time are ongoing.
Poster by de Hoogh et al, 2017 HEI Annual Conference
Statistical methods for investigating the effects of long-term exposure to low air pollutant concentrations in the ELAPSE project using data from 11 pooled European cohorts and 7 administrative cohorts
Bert Brunekreef1, Klea Katsouyanni2,3, Evangelia Samoli2, Massimo Stafoggia4,5, Gerard Hoek1, Maciek Strak1, on behalf of the ELAPSE Statistical Group
1Institute for Risk Assessment Sciences, Utrecht University, Utrecht, the Netherlands; 2Department of Hygiene, Epidemiology and Medical Statistics, Medical School, University of Athens, Athens, Greece; 3Department of Primary Care & Public Health Sciences and Environmental Research Group, King's College London, London, UK; 4Department of Epidemiology, Lazio region Health Service, Rome, Italy; 5Institute of Environmental medicine, Karolinska Institutet, Stockholm, Sweden.
Background and objectives The ELAPSE project aims to investigate the association between long-term exposure to low air pollution levels and multiple health outcomes in European cohort studies. This presentation describes the methodological approaches that will be adopted within the ELAPSE project.
Statistical methods The study population consists of 18 European cohorts: 11 selected cohorts of the ESCAPE study, of medium size range (~380,000 subjects in total), with detailed information on individual level characteristics, and 7 very large European administrative cohorts (> 25 million subjects), with less detailed individual information. The statistical analyses will be conducted separately for the two groups: the ESCAPE cohorts will be pooled into one single database on which we will apply multivariate Cox proportional hazard models; the administrative cohorts will be analysed individually, and the cohort-specific hazard ratios will be pooled into random-effects meta-analysis. Several methodological aspects will be addressed:
Confounding adjustment: several degrees of adjustment for confounding will be applied, from “crude” models (only age, gender and calendar period) to “fully-adjusted” models (all available confounders). Models with area-level socio-economic status confounders will also be considered. In case of important missing confounders (e.g. smoking intensity), in the administrative cohorts’ analysis methods of indirect adjustment will be adopted, gathering information on the missing covariate(s) from ancillary surveys conducted in the same study areas;
Missing data: methods of multiple imputation accounting for between studies heterogeneity will be adopted to fill in missing observations on specific confounders in individual cohort studies in the pooled cohorts’ analysis;
Concentration-response functions: several approaches will be explored to describe the relationship between air pollution exposure and health outcomes. These include: a) natural and penalized splines; b) fractional polynomials; c) threshold models and d) analysis for subgroups of data below certain concentration levels. Statistical comparisons between more complex and more parsimonious models will be provided;
Multi-pollutant models: We will apply two and three-pollutant models to test the sensitivity of the associations that will be statistically significant at the 10% level. Multi-pollutant models will be fit after closely examining the pollutants’ correlation structure;
Latency of the effects: different lag structures of association will be explored, including: fixed exposure at baseline, fixed exposure as average across the whole follow-up; time-varying exposure averaged over different time windows before outcome/censoring;
Effect modification: we will evaluate effect modification by: age, sex, education, smoking status and BMI levels;
Measurement error correction methods: methods based on regression calibration and parameter bootstrapping will be explored to address the issue of measurement error in exposure estimate at the residential address, with the aim of adjusting for classical and Berkson measurement errors. Exposure estimates using 80% of the monitors on which the hybrid LUR models are based will also be used to assess sensitivity of effect estimates.
Discussion A large spectrum of statistical methods will be applied within the ELAPSE project. Special attention will be devoted on the possible sources of bias in the air pollution-health relationship, introduced by missing data/covariates, measurement error and ill-specified concentration-response relationships.
Poster by Stafoggia et al, 2017 HEI Annual Conference