Limits...
A unified spatiotemporal modeling approach for predicting concentrations of multiple air pollutants in the multi-ethnic study of atherosclerosis and air pollution.

Keller JP, Olives C, Kim SY, Sheppard L, Sampson PD, Szpiro AA, Oron AP, Lindström J, Vedal S, Kaufman JD - Environ. Health Perspect. (2014)

Bottom Line: The mean structure was derived from a large set of geographic covariates that was reduced using partial least-squares regression.This novel spatiotemporal modeling approach provides accurate fine-scale predictions in multiple regions for four pollutants.These successes highlight modeling advances that can be adopted more widely in modern cohort studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics, and Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, USA.

ABSTRACT

Background: Cohort studies of the relationship between air pollution exposure and chronic health effects require predictions of exposure over long periods of time.

Objectives: We developed a unified modeling approach for predicting fine particulate matter, nitrogen dioxide, oxides of nitrogen, and black carbon (as measured by light absorption coefficient) in six U.S. metropolitan regions from 1999 through early 2012 as part of the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air).

Methods: We obtained monitoring data from regulatory networks and supplemented those data with study-specific measurements collected from MESA Air community locations and participants' homes. In each region, we applied a spatiotemporal model that included a long-term spatial mean, time trends with spatially varying coefficients, and a spatiotemporal residual. The mean structure was derived from a large set of geographic covariates that was reduced using partial least-squares regression. We estimated time trends from observed time series and used spatial smoothing methods to borrow strength between observations.

Results: Prediction accuracy was high for most models, with cross-validation R2 (R2CV) > 0.80 at regulatory and fixed sites for most regions and pollutants. At home sites, overall R2CV ranged from 0.45 to 0.92, and temporally adjusted R2CV ranged from 0.23 to 0.92.

Conclusions: This novel spatiotemporal modeling approach provides accurate fine-scale predictions in multiple regions for four pollutants. We have generated participant-specific predictions for MESA Air to investigate health effects of long-term air pollution exposures. These successes highlight modeling advances that can be adopted more widely in modern cohort studies.

No MeSH data available.


Related in: MedlinePlus

Maps of the modeling areas (denoted by dashed black line) in the six metropolitan regions, including monitor and subject locations. Abbreviations: Fixed, MESA Air fixed monitoring sites; Home, MESA Air home monitoring sites; Snapshot, MESA Air snapshot monitoring sites; Participant, MESA Air participant residence location (moved slightly to protect confidentiality).
© Copyright Policy - public-domain
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4384200&req=5

f1: Maps of the modeling areas (denoted by dashed black line) in the six metropolitan regions, including monitor and subject locations. Abbreviations: Fixed, MESA Air fixed monitoring sites; Home, MESA Air home monitoring sites; Snapshot, MESA Air snapshot monitoring sites; Participant, MESA Air participant residence location (moved slightly to protect confidentiality).

Mentions: In each metropolitan region, we defined the modeling area to be locations within approximately 75 km of each metropolitan center (Figure 1). AQS monitors within the modeling regions were considered for inclusion in the model, and predictions at participant residences were restricted to locations within these modeling regions. In New York, MESA Air participants were recruited from both New York City and Rockland County, so the modeling region included locations near both areas. In Winston-Salem, only one AQS monitoring location for NO2 and NOx met inclusion criteria. To have a complete time series for the 14-year modeling period, an AQS monitor in Charlotte, North Carolina, was included for estimating time trends. In Chicago, the modeling region was further restricted to locations west of –87.5°W longitude because some covariates were unavailable east of that meridian. In Los Angeles, only locations south and/or west of the San Gabriel Mountains were included.


A unified spatiotemporal modeling approach for predicting concentrations of multiple air pollutants in the multi-ethnic study of atherosclerosis and air pollution.

Keller JP, Olives C, Kim SY, Sheppard L, Sampson PD, Szpiro AA, Oron AP, Lindström J, Vedal S, Kaufman JD - Environ. Health Perspect. (2014)

Maps of the modeling areas (denoted by dashed black line) in the six metropolitan regions, including monitor and subject locations. Abbreviations: Fixed, MESA Air fixed monitoring sites; Home, MESA Air home monitoring sites; Snapshot, MESA Air snapshot monitoring sites; Participant, MESA Air participant residence location (moved slightly to protect confidentiality).
© Copyright Policy - public-domain
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4384200&req=5

f1: Maps of the modeling areas (denoted by dashed black line) in the six metropolitan regions, including monitor and subject locations. Abbreviations: Fixed, MESA Air fixed monitoring sites; Home, MESA Air home monitoring sites; Snapshot, MESA Air snapshot monitoring sites; Participant, MESA Air participant residence location (moved slightly to protect confidentiality).
Mentions: In each metropolitan region, we defined the modeling area to be locations within approximately 75 km of each metropolitan center (Figure 1). AQS monitors within the modeling regions were considered for inclusion in the model, and predictions at participant residences were restricted to locations within these modeling regions. In New York, MESA Air participants were recruited from both New York City and Rockland County, so the modeling region included locations near both areas. In Winston-Salem, only one AQS monitoring location for NO2 and NOx met inclusion criteria. To have a complete time series for the 14-year modeling period, an AQS monitor in Charlotte, North Carolina, was included for estimating time trends. In Chicago, the modeling region was further restricted to locations west of –87.5°W longitude because some covariates were unavailable east of that meridian. In Los Angeles, only locations south and/or west of the San Gabriel Mountains were included.

Bottom Line: The mean structure was derived from a large set of geographic covariates that was reduced using partial least-squares regression.This novel spatiotemporal modeling approach provides accurate fine-scale predictions in multiple regions for four pollutants.These successes highlight modeling advances that can be adopted more widely in modern cohort studies.

View Article: PubMed Central - PubMed

Affiliation: Department of Biostatistics, and Department of Environmental and Occupational Health Sciences, University of Washington, Seattle, Washington, USA.

ABSTRACT

Background: Cohort studies of the relationship between air pollution exposure and chronic health effects require predictions of exposure over long periods of time.

Objectives: We developed a unified modeling approach for predicting fine particulate matter, nitrogen dioxide, oxides of nitrogen, and black carbon (as measured by light absorption coefficient) in six U.S. metropolitan regions from 1999 through early 2012 as part of the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air).

Methods: We obtained monitoring data from regulatory networks and supplemented those data with study-specific measurements collected from MESA Air community locations and participants' homes. In each region, we applied a spatiotemporal model that included a long-term spatial mean, time trends with spatially varying coefficients, and a spatiotemporal residual. The mean structure was derived from a large set of geographic covariates that was reduced using partial least-squares regression. We estimated time trends from observed time series and used spatial smoothing methods to borrow strength between observations.

Results: Prediction accuracy was high for most models, with cross-validation R2 (R2CV) > 0.80 at regulatory and fixed sites for most regions and pollutants. At home sites, overall R2CV ranged from 0.45 to 0.92, and temporally adjusted R2CV ranged from 0.23 to 0.92.

Conclusions: This novel spatiotemporal modeling approach provides accurate fine-scale predictions in multiple regions for four pollutants. We have generated participant-specific predictions for MESA Air to investigate health effects of long-term air pollution exposures. These successes highlight modeling advances that can be adopted more widely in modern cohort studies.

No MeSH data available.


Related in: MedlinePlus