Final Report Summary - STRPM (Assessing Temporally and Spatially Resolved PM2.5 and AirTemperature Exposures For Epidemiological Studies Using SatelliteBased Methods)
1. FINAL PUBLISHABLE SUMMARY REPORT
We aimed to develop better statistical air pollution (PM 2.5,10) and temperature exposure assessment methods across Italy and France to handle contemporary exposure datasets, building on our previous work. We validated computationally efficient models, allowing us to more accurately estimate PM and Ta at very high spatial (1×1 km) and temporal (daily) resolutions across Italy and France for 2000-2014. We then used these generated PM and Ta estimations for a nationwide study in France to estimate the association between exposure to PM2.5 Ta and birth outcomes.
The grant started with aim 1 (“Create, validate and generate PM2.5 estimations”). This aim was very successful and resulted in models that far exceeded the initial aims by not only providing daily PM2.5 predictions but PM10 estimation across France and Italy as well. The result of this work was published in the high impact scientific Journal “Environment International” (Estimation of daily PM10 concentrations in Italy (2006–2012) using finely resolved satellite data, land use variables and meteorology) and 2 other papers (one for France and one for Italy) are being drafted currently.
We combined finely resolved data on Aerosol Optical Depth (AOD) from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm, ground-level PM10/2.5 measurements, land-use variables and meteorological parameters into a four-stage mixed model framework to derive estimates of daily PM10/2.5 concentrations at 1-km2 grid over France and Italy, for the years 2000–2014. We checked performance of our models by applying 10-fold cross-validation (CV) for each year. Our models displayed very good fitting, for example across Italy a mean CV-R2 of 0.65 and 0.83 (for PM 10 and 2.5 respectively) and little bias (average slope of predicted VS observed PM10 = 0.99). Our model performance was excellent for both days with and without available PM observations showing that PM can be reliably predicted using daily MODIS AOD data across France and Italy.
Aim 1 was successfully followed by Aim 2 (“Create, validate and generate Ta estimations”). The result of the France section was published in the high impact scientific Journal “International journal of climatology” (“Modelling spatio-temporally resolved air temperature across the complex geo-climate area of France using satellite-derived land surface temperature data”) while the Italian air temperature (Ta) model is now being submitted to a high impact journal. Similar to the air pollution models in aim 1 we developed spatiotemporally resolved models which allowed us to predict Ta on a fine 1 km grid across both France and Italy. We use a daily calibration approach using a series of processes to generate daily Ta for every day across the entire study area and period. First, we started by calibrating MODIS (Moderate Resolution Imaging Spectroradiometer) satellite-gridded surface temperature (Ts) data against Ta collected within 1 km of the Ts centroid. The calibration stage adjusted for spatio-temporal predictors, as done in environmental exposure assessment methods such as land use regressions. Second, to estimate Ta when no Ts data are available we fit a second model which uses the association of predicted grid cells Ta values (based on satellite Ts) with surrounding Ta monitors and the association with values in neighboring grid cells. Out-of-sample tenfold cross-validation was used to quantify the accuracy of our predictions. Our model performance was excellent for both days with available Ts and days without Ts observations (overall mean out-of-sample R2=0.95 in France and 0.96 in Italy). We again demonstrated how Ts can be used reliably to predict daily Ta at high-resolution across France for use in studies looking at the effects of fine resolution Ta exposure on various health outcomes.
Completing these daily PM2.5,PM10 and Ta predictions datasets at high resolution which are now distributed and disseminated to our Italian and French research collaborators and community has advanced environmental epidemiological analyses of adverse birth outcomes and susceptibility. We made use of these novel estimation models to address aim 3 (“Conduct a nationwide study in France to estimate the association between exposure to PM2.5 and air temperature and fetal growth”).
In Aim 3, collaborating with the INSERM team we used the France temperature model to assess the temperature exposure of pregnant women in 4 French mother-child cohorts: EDEN (http://eden.vjf.inserm.fr/index.php/fr/ Heude et al, 2015), PELAGIE (http://www.pelagie-inserm.fr Petit et al, 2012), SEPAGES (http://sepages.inserm.fr/en/home/) and ELFE (https://www.elfe-france.fr/en/ Vandentorren et al. 2009). Using our models strengths such as the large spatial (i.e. national) and temporal (i.e. 2002 to 2011) coverage and its fine spatio-temporal resolution (1 x 1 km, daily), the large spatio-temporal coverage provided us with a homogenous exposure tool for the 4 cohorts, which were recruited during different years from 2002 to 2011. It allowed us to study pooled data of the 4 cohorts therefore increasing the statistical power to detect an effect of temperature on pregnancy outcomes. The fine spatio-temporal resolution of the model allowed us to estimate the exposure of each woman at her home addresses for each day of the pregnancy and to then study the effects of temperature exposure during long-term time-windows (i.e. whole pregnancy) but also short-term time windows of a few days before the delivery.
In a pooled analysis of the EDEN and PELAGIE cohorts, we showed negative associations between temperature exposure in the 3rd trimester of pregnancy with the birth weight of the baby as well as between temperature exposure higher than 16°C during the whole pregnancy and gestational duration. In this study, we observed consistent results when we used the stations of the national weather monitoring system (Abraham et al, in revision) and when we used the developed temperature model (Jakpor et al, in preparation) to estimate temperature exposure. Birth weight and gestational duration are predictors of health later in childhood and adulthood. In an air pollution study conducted the ELFE cohort, we were able to adjust our results for the exposure to temperature (Ouidir et al, In Review). Future work include the study of the associations between temperature exposure and pregnancy outcomes in the ELFE, EDEN, SEPAGES and PELAGIE cohorts using a refined version of the temperature model currently developed by Ian Hough, a PhD student jointly supervised by I. Kloog and J. Lepeule.
The implications for environment and health studies in France and italy are many. For one we made available through many current and future collaborations the entire temperature and air pollution datasets to be used as robust and validated exposure assessment input in future studies. In addition, the results from our epidemiology studies have added valuable body of evidence looking at the effects of air pollution and climate change on adverse pregnancy outcomes.
These findings may assist in the development of proactive public health policies.
We aimed to develop better statistical air pollution (PM 2.5,10) and temperature exposure assessment methods across Italy and France to handle contemporary exposure datasets, building on our previous work. We validated computationally efficient models, allowing us to more accurately estimate PM and Ta at very high spatial (1×1 km) and temporal (daily) resolutions across Italy and France for 2000-2014. We then used these generated PM and Ta estimations for a nationwide study in France to estimate the association between exposure to PM2.5 Ta and birth outcomes.
The grant started with aim 1 (“Create, validate and generate PM2.5 estimations”). This aim was very successful and resulted in models that far exceeded the initial aims by not only providing daily PM2.5 predictions but PM10 estimation across France and Italy as well. The result of this work was published in the high impact scientific Journal “Environment International” (Estimation of daily PM10 concentrations in Italy (2006–2012) using finely resolved satellite data, land use variables and meteorology) and 2 other papers (one for France and one for Italy) are being drafted currently.
We combined finely resolved data on Aerosol Optical Depth (AOD) from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm, ground-level PM10/2.5 measurements, land-use variables and meteorological parameters into a four-stage mixed model framework to derive estimates of daily PM10/2.5 concentrations at 1-km2 grid over France and Italy, for the years 2000–2014. We checked performance of our models by applying 10-fold cross-validation (CV) for each year. Our models displayed very good fitting, for example across Italy a mean CV-R2 of 0.65 and 0.83 (for PM 10 and 2.5 respectively) and little bias (average slope of predicted VS observed PM10 = 0.99). Our model performance was excellent for both days with and without available PM observations showing that PM can be reliably predicted using daily MODIS AOD data across France and Italy.
Aim 1 was successfully followed by Aim 2 (“Create, validate and generate Ta estimations”). The result of the France section was published in the high impact scientific Journal “International journal of climatology” (“Modelling spatio-temporally resolved air temperature across the complex geo-climate area of France using satellite-derived land surface temperature data”) while the Italian air temperature (Ta) model is now being submitted to a high impact journal. Similar to the air pollution models in aim 1 we developed spatiotemporally resolved models which allowed us to predict Ta on a fine 1 km grid across both France and Italy. We use a daily calibration approach using a series of processes to generate daily Ta for every day across the entire study area and period. First, we started by calibrating MODIS (Moderate Resolution Imaging Spectroradiometer) satellite-gridded surface temperature (Ts) data against Ta collected within 1 km of the Ts centroid. The calibration stage adjusted for spatio-temporal predictors, as done in environmental exposure assessment methods such as land use regressions. Second, to estimate Ta when no Ts data are available we fit a second model which uses the association of predicted grid cells Ta values (based on satellite Ts) with surrounding Ta monitors and the association with values in neighboring grid cells. Out-of-sample tenfold cross-validation was used to quantify the accuracy of our predictions. Our model performance was excellent for both days with available Ts and days without Ts observations (overall mean out-of-sample R2=0.95 in France and 0.96 in Italy). We again demonstrated how Ts can be used reliably to predict daily Ta at high-resolution across France for use in studies looking at the effects of fine resolution Ta exposure on various health outcomes.
Completing these daily PM2.5,PM10 and Ta predictions datasets at high resolution which are now distributed and disseminated to our Italian and French research collaborators and community has advanced environmental epidemiological analyses of adverse birth outcomes and susceptibility. We made use of these novel estimation models to address aim 3 (“Conduct a nationwide study in France to estimate the association between exposure to PM2.5 and air temperature and fetal growth”).
In Aim 3, collaborating with the INSERM team we used the France temperature model to assess the temperature exposure of pregnant women in 4 French mother-child cohorts: EDEN (http://eden.vjf.inserm.fr/index.php/fr/ Heude et al, 2015), PELAGIE (http://www.pelagie-inserm.fr Petit et al, 2012), SEPAGES (http://sepages.inserm.fr/en/home/) and ELFE (https://www.elfe-france.fr/en/ Vandentorren et al. 2009). Using our models strengths such as the large spatial (i.e. national) and temporal (i.e. 2002 to 2011) coverage and its fine spatio-temporal resolution (1 x 1 km, daily), the large spatio-temporal coverage provided us with a homogenous exposure tool for the 4 cohorts, which were recruited during different years from 2002 to 2011. It allowed us to study pooled data of the 4 cohorts therefore increasing the statistical power to detect an effect of temperature on pregnancy outcomes. The fine spatio-temporal resolution of the model allowed us to estimate the exposure of each woman at her home addresses for each day of the pregnancy and to then study the effects of temperature exposure during long-term time-windows (i.e. whole pregnancy) but also short-term time windows of a few days before the delivery.
In a pooled analysis of the EDEN and PELAGIE cohorts, we showed negative associations between temperature exposure in the 3rd trimester of pregnancy with the birth weight of the baby as well as between temperature exposure higher than 16°C during the whole pregnancy and gestational duration. In this study, we observed consistent results when we used the stations of the national weather monitoring system (Abraham et al, in revision) and when we used the developed temperature model (Jakpor et al, in preparation) to estimate temperature exposure. Birth weight and gestational duration are predictors of health later in childhood and adulthood. In an air pollution study conducted the ELFE cohort, we were able to adjust our results for the exposure to temperature (Ouidir et al, In Review). Future work include the study of the associations between temperature exposure and pregnancy outcomes in the ELFE, EDEN, SEPAGES and PELAGIE cohorts using a refined version of the temperature model currently developed by Ian Hough, a PhD student jointly supervised by I. Kloog and J. Lepeule.
The implications for environment and health studies in France and italy are many. For one we made available through many current and future collaborations the entire temperature and air pollution datasets to be used as robust and validated exposure assessment input in future studies. In addition, the results from our epidemiology studies have added valuable body of evidence looking at the effects of air pollution and climate change on adverse pregnancy outcomes.
These findings may assist in the development of proactive public health policies.