cmas special session oct 13, 2010 air pollution exposure estimation: 1.what’s been done?...
TRANSCRIPT
CMAS special sessionOct 13, 2010
Air pollution exposure estimation:
1. what’s been done?
2. what’s wrong with that?
3. what can be done?
4. how and what to evaluate?
what’s been done?
1. use existing monitoring network(s)
2. estimate “exposure”• average monitors• nearest monitor• inverse-distance weighting• spatial statistics (eg, kriging)
3. plug estimate into health model
what’s wrong?
1. basically, exposure measurement error• the dogma: if non-differential, bias of health effect
estimate to the null
2. what else?
• features of air monitoring network data• intensity - spatial, temporal, components• coverage - spatial, geographic covariates
3. unknown impacts on health effect estimates
improvements
1. to monitoring• to reflect sources• study subject residences/transit/indoors/personal
2. enhance value of monitoring • land-use regression
early example
Hoek et al. Lancet 2002; 360: 1203-9
1. local scale - distance to roads (BS & NO2)
2. urban scale - BS/NO2 regressed on population density (land use regression)
3. regional scale - inverse-distance weighting using population-oriented monitors
use of deterministic air quality models in health studies
1. attempt to address many deficiencies in using monitoring data
• concentrations at unmonitored locations and times• unmeasured pollutants & sources
2. two (at least) ways to use them1. have monitoring data and estimates of exposure and
supplement with AQM predictions
2. start with AQM and supplement with data (“data assimilation”)
evaluating value added
1. performance in predicting “exposure”• validation
2. impact on health effect estimates• variance, coverage, bias, RMSE
3. role of simulation
problems with using AQMs
1. for pollutants that are relatively homogeneous spatially, we have monitoring data
2. most models smooth and therefore reduce variability
• addition of covariates with less variability than that of study subjects
3. for pollutants that are not spatially homogeneous, spatial scale often too large
= MESA Air region
Chestnut Hill, MAFall River, MA
Jacksonville, FL
New Brunswick, NJ
Greensboro, NC
Des Moines, IADavenport, IA
Phoenix, AZLa Jolla, CA
Evanston, IL
Epidemiology:WHI-OS and MESA Air cardiovascular cohorts
NPACT 2-week sampling protocols in MESA communities (in addition to CSN data)
type of monitoring
measurements(sampler)
(1) fixed PM2.5
(2) home
outdoorPM2.5
(3) home
indoorPM2.5
(4) personal
PM2.5
(5) traffic
gradientsnapshots
(6) PM2.5/PM10
snapshots
mass, BC,elements/metals(Teflon filter)
EC, OC,
DTT(quartz filter)
EC, OC only
NOx, NO2, O3, SO2
(Ogawa sampler)
no O3
no O3
NOx & NO2
(1) STN (collocated), roadside, household (3-7/city [26 total])(2) rotating every 2 weeks > 50 homes/city in 2 seasons(3) rotating ~ 50/city in 2 seasons(4) rotating ~15/city in 2 seasons(5) ~100 sites/city (roadside, near road, population) in 3 seasons(6) ~ 40 sites/city (Chicago, Minneapolis, Winston-Salem) in 2 seasons
Silicon, nickel and EC concentrations at New York STN and MESA Air fixed and home outdoor sites
As expected, the spatial distribution of concentrations(quintiles) is different for each species
Si Ni EC
monitoring data structure is complex
1 2 3 4 5 6 7 … 24 25 26 27 28 29 30 31 32 33 … 45 46 47 48 49 501 X X X X X X X … X X X X X X X X X X … X X X X X X
… … … … … … … … … … … … … … … … … … … … … … … … … …5 X X X X X X X … X X X X X X X X X X … X X X X X X1 X X X X X X X … X X X X X X X X X X … X X X X X X
… … … … … … … … … … … … … … … … … … … … … … … … … …5 X X X X X X X … X X X X X X X X X X … X X X X X X1 X X2 X X3 X X4 X X
… … …49 X X50 X X1 X X X2 X X X3 X X X
… … … …99 X X X
100 X X X
Spatial locations
Time
CSN/STN (number varies by
location)
Fixed (MESA) (3-7 sites)
Home outdoor (50 sites)
Snapshots (100 traffic gradient
&40 speciation)
Approach to estimating household concentrations:
• estimate time trend from temporally rich CSN/STN sites (and less rich fixed sites) to estimate time trend (which can vary over space,
esp. for some PM2.5 components) at home outdoor sites
• then, remove time trends from spatially rich (but
temporally sparse) home outdoor measurements to allow spatial information to be used
• use spatio-temporal modeling methods, incorporating geographic information
Long-term EC concentrations at individual monitoring sites adjusted for temporal trend in LA
• solid black line = temporal trend• dotted black line = mean of trend• black dot = observed value at monitoring site• red circle = temporal adjustment for observed value• red line = adjusted mean at monitoring site