SlideShare a Scribd company logo
1 of 47
Download to read offline
Environmental exposure in environmental
epidemiological studies:
modeling approaches and challenges
Veronica J. Berrocal
Department of Biostatistics
University of Michigan
SAMSI Workshop on Remote Sensing, Uncertainty Quantification, and
Theory of Data Systems
Veronica J. Berrocal Air pollution exposure
Environmental epidemiological studies
• Environmental epidemiological studies aim to establish an
association (potentially, a causal one) between an health outcome
and an environmental exposure.
• Typically, health outcome refers to:
• mortality (all-cause non accidental deaths, or due to
cardiovascular diseases, respiratory diseases, etc.)
• hospitalizations/emergency visits
• pregnancy outcomes and birth defects
• ...
• Environmental exposure refers to:
• exposure to temperature, heat/heatwave, cold/cold waves
• air pollution/wildfires
• pesticides
• precipitation
• pollen
• ...
Veronica J. Berrocal Air pollution exposure
Environmental epidemiological studies:
this talk
• Focus only on air pollution and temperature as environmental
exposures.
• The statistical methods used are very similar:
• same type of health models are used to link health and
environmental exposure
• methods for air pollution exposure assessment are slighly
ahead of those for temperature, but the gap is closing
Veronica J. Berrocal Air pollution exposure
Health data
• Typically health data is obtained from:
• administrative database: mortality from the National Center
for Health Statistics; death and birth records from local and
State Departments of Health
−→ easier to obtain but aggregated over some areal unit
• hospital and emergency visit records: individual level data
−→ hard to obtain
• Department of Human Health and Services: Medicare and
Medicaid databases
−→ expensive to obtain; require a lot of data cleaning and
formatting
• cohort studies: individual level data, smaller sample size
−→ need a collaborator to obtain
• Earlier studies used mostly administrative databases aggregated over
counties or larger metropolitan areas.
• Now, common to use geocoded health data, e.g. individuals are
assigned specific geographical coordinates
Veronica J. Berrocal Air pollution exposure
Health data
• Typically health data is obtained from:
• administrative database: mortality from the National Center
for Health Statistics; death and birth records from local and
State Departments of Health
−→ easier to obtain but aggregated over some areal unit
• hospital and emergency visit records: individual level data
−→ hard to obtain
• Department of Human Health and Services: Medicare and
Medicaid databases
−→ expensive to obtain; require a lot of data cleaning and
formatting
• cohort studies: individual level data, smaller sample size
−→ need a collaborator to obtain
• Earlier studies used mostly administrative databases aggregated over
counties or larger metropolitan areas.
• Now, common to use geocoded health data, e.g. individuals are
assigned specific geographical coordinates =⇒ potential for
geocoding error
Veronica J. Berrocal Air pollution exposure
Environmental exposure data
• Typically environmental exposure data is obtained from:
• outdoor monitors (air pollution, temperature, humidity) -
sources: NOAA/NCEI, EPA
• dispersion model outputs (mostly for traffic related pollutants)
- sources: EPA, collaborators
• air quality model outputs - sources: EPA, collaborators
• satellite data (MODIS AOD, Aura OMI, etc.) - sources NASA
• personal monitors
• private monitors (e.g. Harvard Six City study, etc.) - sources:
NASA
• Temporal and spatial resolution varies across data sources.
• Informative sampling.
• Availability and access vary across data sources.
• Uncertainties and measurement errors vary across data sources.
Veronica J. Berrocal Air pollution exposure
Health analysis
• The statistical model used in the health model depends on the
spatial resolution of the health data.
• Most commonly used study design is a time series design
• Health outcome: number
of deaths on day t
aggregated over a unit
• Poisson regression
(GAM, overdispersion)
• logExpected mortalityt =
S(t;λ1)+
S(env expt;λ2)+
S(lagged env exp.;λ3)+
g(confound.)
• S(·;λ): smooth function
with λ d.f. Figure: From Bhaskaran et al. (2003)
Veronica J. Berrocal Air pollution exposure
Health analysis
• In time series study, the environmental exposure is
representative of the entire areal unit
−→ usually, area-wide average from outdoor monitors
Veronica J. Berrocal Air pollution exposure
Health analysis
• In time series study, the environmental exposure is
representative of the entire areal unit
−→ usually, area-wide average from outdoor monitors
• No accounting for intra-urban variation in exposure
• Usually fit at individual locations separately and then
combined together through a hierarchical approach
• Effect of environmental exposure at each areal unit linked to
areal unit characteristics
−→ use multiple data sources (Census, National Land Cover
Dataset, etc.)
Veronica J. Berrocal Air pollution exposure
Health analysis
• In time series study, the environmental exposure is
representative of the entire areal unit
−→ usually, area-wide average from outdoor monitors
• No accounting for intra-urban variation in exposure
• Usually fit at individual locations separately and then
combined together through a hierarchical approach
• Effect of environmental exposure at each areal unit linked to
areal unit characteristics
−→ use multiple data sources (Census, National Land Cover
Dataset, etc.)
• If individual lavel health data is available, other study design
are possible
• case-crossover design
• Cox proportional hazard models
• linear regression models
Veronica J. Berrocal Air pollution exposure
Environmental exposure
• Earlier studies on the health effect of air pollution or
temperature used area-wide averages from monitors as
exposure
=⇒ this approach ignores intra-urban variation in exposure.
• Exploiting intra-urban contrasts has several advantages:
• increased power
• rule out unmeasured confounders
• differentiate between different environmental exposures (e.g.
pollutant, temperature vs dew point, etc.)
• Environmental exposure assessment methods are used more
frequely for air pollution exposure, but are starting to be more
routinely used also for temperature
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to locations/individual subjects:
1 Nearest monitor(s)/Average over a buffer: derive exposure
using measurement from the nearest monitor or the monitors
within a buffer.
• Pro: simple
• Cons: too simplistic; does not capture small-scale spatial
variation
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to locations/individual subjects:
3 Land use regression (LUR): predict environmental exposure
using a linear regression model with predictors GIS covariates,
e.g. type of roads within a buffer, air conditioning (Briggs et al.
1997, 2000; Brauer et al. 2003; Ross et al. 2006).
• Pro: simple
• Cons: large number of GIS covariates (variable selection
technique needed: backward/forward selection; LASSO; PLS),
limited transferability to different locations, residual
independence between sites.
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to locations/individual subjects:
3 Land use regression (LUR): predict environmental exposure
using a linear regression model with predictors GIS covariates,
e.g. type of roads within a buffer, air conditioning (Briggs et al.
1997, 2000; Brauer et al. 2003; Ross et al. 2006).
• Pro: simple
• Cons: large number of GIS covariates (variable selection
technique needed: backward/forward selection; LASSO; PLS),
limited transferability to different locations, residual
independence between sites.
• In recent studies, satellite data has also been used (Kloog et
al. 2011).
• Multiple-stage models needed to account for extensive missing
satellite data.
• Uncertainty in each model stage is not properly accounted for.
Veronica J. Berrocal Air pollution exposure
Using satellite data (Kloog et al. 2017)
• (1) Daily average temperature from monitoring stations from M´et´o
France, (2) daily temperature data from MODIS, Terra instrument,
at 1km resolution
• Two step-model:
1 Calibration of satellite data
2 Prediction
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to locations/individual subjects:
4 Universal kriging (UK) models/Hybrid models: predict
exposure using a spatial statistical model with predictors GIS
covariates (Mercer et al. 2011; Bergen et al., 2013; Kloog et
al. 2012 - includes satellite data).
• Pro: accounts for spatial dependence between sites
• Cons: requires knowledge of spatial statistics; cannot be
implemented directly in ArcGIS, although possible through a
two-stage approach.
Still requires use of variable selection techniques to reduce
number of GIS covariates.
• UK has been shown to yield better predictions than LUR
(Beelen et al. 2009; Mercer et al. 2011)
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to location/individual subjects:
5 Dispersion modeling/Regional climate model: use as exposure
the output of a dispersion model/regional climate model
(Nafstad et al. 2004; Penard-Morand et al. 2006).
• Pro: Based on physical principles; non-linearity; better than
LUR for specific-source related component.
• Cons: output often uncalibrated; smooth exposure surfaces;
different spatial resolution.
Veronica J. Berrocal Air pollution exposure
Environmental exposures: modeling
approaches
• Several different strategies are used to assign environmental
exposure to locations/individual subjects:
6 Statistical downscaling/data fusion models: predict exposure
combining the output of air quality/dispersion/regional climate
models with monitoring data (Fuentes and Raftery 2005;
McMillan et al.2010; Berrocal et al. 2010a,b, 2012, 2014;
Reich et al. 2014; Rushworth et al. 2014; Lee et al. 2016).
• Pro: accounts for physical and chemical processes; no need to
include GIS covariates (Lindstrom et al. 2014); can account
for complicated spatio-temporal dependence structure.
• Cons: numerical model output not easily/publically available;
requires knowledge of spatial, spatio-temporal and Bayesian
statistics; software not always available.
Veronica J. Berrocal Air pollution exposure
Downscaler/data fusion models (Berrocal et al., 2010,
2011, 2012, 2014):
100 95 90 85 80 75 70
30354045
Longitude
Latitude
20
40
60
80
100
120
140
Observed ozone concentration
August 1, 2001
100 95 90 85 80 75 70
30354045
Longitude
Latitude
20
40
60
80
100
120
140
CMAQ estimate of ozone concentration
August 1, 2001
• Main idea: leverage two sources of information:
• the true sparse, point-referenced monitors measurements:
Y (s), s ∈ S
• the spatially dense, grid-based, potentially uncalibrated, model
output X(B) with B 12-km grid cell
Veronica J. Berrocal Air pollution exposure
Data fusion model (Berrocal et al. 2012)
• Our most recent data fusion model is:
Observed concentration at s = β0(s)+β1 · ˜X(s)+ε(s)
Y (s)
with
• β0(s) modeled as a spatial process (e.g. Gaussian process with
a given covariance function).
• ˜X(s) smoothed version of model output at each point s,
obtained using random, directional, spatially-varying and
spatially correlated weights.
• ε(s) ∼ N(0,τ2
).
Veronica J. Berrocal Air pollution exposure
Data fusion model (Berrocal et al. 2012)
• Our most recent data fusion model is:
Observed concentration at s = β0(s)+β1 · ˜X(s)+ε(s)
Y (s)
with
• β0(s) modeled as a spatial process (e.g. Gaussian process with
a given covariance function).
• ˜X(s) smoothed version of model output at each point s,
obtained using random, directional, spatially-varying and
spatially correlated weights.
• ε(s) ∼ N(0,τ2
).
• Other extensions of this data fusion model have been proposed.
• Used by EPA to derive fused surfaces of daily average PM2.5 and
daily 8-hour maximum ozone concentration at census tract
centroids for US
• Website https://www.epa.gov/air-research/
downscaler-model-predicting-daily-air-pollution)
provides predictions and their standard deviations
Veronica J. Berrocal Air pollution exposure
More monitoring data? Caution!
(Berrocal and Holland, 20XX)
• The downscaler model uses monitoring data to calibrate the
air quality model output.
• We used data from the Federally Regulated Monitors for
PM2.5: the FRM monitors.
• Measurements of PM2.5 are nowadays available also from
other monitors. Are they useful?
• Here we consider “new”, automated monitors that measure
PM2.5 semi-continuously, the SC-FEM monitors, and we
evaluate whether adding these set of monitors in a downscaler
model leads to better predictions.
Veronica J. Berrocal Air pollution exposure
Is more monitoring data helpful?
−120 −110 −100 −90 −80 −70
253035404550
Longitude
Latitude
qq
q
qq
qq
q
q
q
q
q
q
q
q
qq
q
q
qq qq
qq
q
q
q
q
qq
qq
q
q
qqq
q
q
qq
q
q
qqq
qq
q
q
q
q
q
q
qqq
q
q
q
q
qq
q
q
q
qq
q
q
q
q
q
q
q
q
q
qq
q
q qqq
qqq
q q
q
q qq
q
q
q qq
q
q
q
q
q
q q
q
q
qq
q
q
q
q
qq
q q
qqqq
q
qq
q
q
q
q
q
q
q
qqq
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
qq
qq
q
q
q
q
q
q
qq
q
q
q
q qq
qq
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qqq
q
q
q
q
q
qqqqqqqq
q
q
q
qqq
q
qq
q
q
q
q
q
qq
qqq
q
qq
q
qq
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q qq
q
q
q
qqqqqq
q
qq
q
q
qq
q
qqq
q
q
qq
q
q
q q
q
qqq
qq
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
qq
q
qq
q
qq
q
q
q
qq
q
q
qq
q
qq
q
q
q
qq
q
qqqq
qqq
q
q
qqqqqq
q
q
qq
q q
q
q
q
q q
qqqqqqqq
q
q
q
q
q
qq
q
q
q
q
qq
qq
q
q
q
qq
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
q
qq
q
q
q
qq
q
q
q
qq
qqqq
qqq
q
q
qq
q
qqq
q
qq
q q
q
q
q
q
q
q q
q
q
q
q
q
qq
q
q
q
q
q
q q
q
q
q
q qq
q
qq
q qq
q
q
q
q
q
qq
qq
q
q
q
qqqqqq
qqq
qq
q
q
q
q
qqqq
qq
qqqqq
qq
q
q
q
q
q
q
qqq
q
q
qq
q q
q
q
qq
q
q
q qq
q
qqqqq
q
q
q
q
q
q
q
qq
q
q
q
qq
q
qq
q
q
qq
qqq
q
q
q
q
qq
q
q
q
qqq
q
q
q qq
q
qq
q
q
q
q
q
qq
q
q
q
q
qq
q
qq
q
qq
q
q
q
q
q
q
q
qq
q
q
q
q
qq
qq
q
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
q
q
q
qq
q
q
q
qqqq
q
q
qq
q
q
q
q
q
q
qqq
q
q
q
qq
q
qqq
q
qqq
q
qq
q
q
q
q
qqq
q
q
q
qqq
qq
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
qq
q
q
qq
q
q
q
q
q
q
q
q
q
q q
q
q q
q
q
q
q
q
q
q
q
q
q
q q
q
q
q
q
qq
q
q
qqq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
qq
q
qq
q
q
q qq
q
q q
q
qqq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
qqq
q
q
qq
q
q
q
q
q
q q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
q
q
q
q
q
qq
qq
qqqqqqqq
q
qqq
q
qq
q
q
q
q
q
q
q
q
q
q
q
q qq
qq
q
qq
q qq
qq
q
q
q
q
q
q
qq
q
qq
q
q
q
q
qqq
qq
q
qqqq
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
qq
q
qq
qq
q
q
qqqqqq
q
q
q
qq
q
qq
q
qq
q
q
qq
q
q
qq
q
qq
qq
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
qq
q
q
q
q
q
q
q q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
q
q
q
q
q q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q q
q
q
q
q
q
q
q
q
q
q q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q q
q
q
q
qq
q
q
q
q
q
q
q
FRM
SC/FEM
• We apply a downscaler model to daily
average PM2.5 concentration in the US
for year 2011.
• Each day we fit the model to the data,
using 80% of the observations
and holding 20% for validation of the
predictions.
• We compare a model that uses both
FRM and SC/FEM monitoring data
vs one that uses only FRM data.
Veronica J. Berrocal Air pollution exposure
Results
• MSE= Average of (observed−predicted)2
.
• MAE= Average of |observed−predicted|.
MSE MAE Width 95% CI Coverage 95% CI
Monitors (µg/m3
) (µg/m3
) (µg/m3
) (%)
Only
FRM 4.09 2.28 16.82 95.59
FRM and
SC/FEM 4.03 2.09 17.45 96.29
• Moderate improvement overall.
• Larger improvements when stratified by season, with large gains in
Summer and Fall.
Veronica J. Berrocal Air pollution exposure
Caution with monitoring data!
• Reduction in Mean Absolute Error by state when we compare the
model that uses FRM + SC/FEM data vs a model that uses only
FRM data.
• Positive means better predictions using FRM + SC/FEM data.
Percent reduction in MAE:
SC/FEM + FRM data vs FRM only
Fall
PercentreductioninMeanAbsoluteError(%)
Alabama
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
DistrictOfColumbia
Florida
Georgia
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
NewHampshire
NewJersey
NewMexico
NewYork
NorthCarolina
NorthDakota
Ohio
Oklahoma
Oregon
Pennsylvania
RhodeIsland
SouthCarolina
SouthDakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
WestVirginia
Wisconsin
Wyoming
−40−20020
Veronica J. Berrocal Air pollution exposure
Challenges: measurement error
• Predicted exposure is not the true exposure.
• Using predicted exposure in a health model introduces two
sources of measurement errors:
• Berkson error: from smoothing the exposure surface =⇒
inflated SE of health effect estimates.
• Classical measurement error: from estimation of the
parameters in the exposure model =⇒ bias and SE inflation.
• Not easy to separate the two.
Veronica J. Berrocal Air pollution exposure
How to correct for measurement error
• Various papers have addressed this issue (Gryparis et al. 2009;
Szpiro et al. 2011; Szpiro and Paciorek, 2013; Alexeef et al.
2015; etc).
• Proposed approaches include:
1 Bayesian joint model of exposure and disease: this uses the
correct imputation scheme (see missing data literature - Rubin
1987, Little 1992).
2 Two-stage Bayesian approach: exposure model first, then
health model.
Posterior distribution for exposure is the prior in the health
model.
=⇒ Does not cut the feedback between exposure and health
outcome.
3 Bootstrap approximation: resample health and exposure datat.
• Option 1 is the ideal, but most computationally intensive.
Veronica J. Berrocal Air pollution exposure
Accounting for uncertainty in personal
exposure
• Personal exposure is not the same as ambient exposure
• Using ambient exposure as a proxy for personal exposure
1 Biased estimates of the effect of air pollution (Zeger et al.
2000)
2 Wrong assessment of its uncertainty (Zeger et al. 2000;
Gryparis et al. 2009)
• Prototype analysis that shows how to account for exposure
uncertainty in a study on the effect of personal or ambient
maternal exposure to PM2.5 on birthweight.
Veronica J. Berrocal Air pollution exposure
Personal exposure
Veronica J. Berrocal Air pollution exposure
Ambient vs personal exposure
• Given a day d, let Cj (d) be outdoor concentration of PM2.5 at
the location where subject j lives .
• Ambient exposure for individual j on day d is the outdoor
concentration Cj (d).
Veronica J. Berrocal Air pollution exposure
Ambient vs personal exposure
• Given a day d, let Cj (d) be outdoor concentration of PM2.5 at
the location where subject j lives .
• Ambient exposure for individual j on day d is the outdoor
concentration Cj (d).
• Personal exposure for individual j on day d is the sum of
contributions from the various “microenvironments” the
individual visits during the day:
Personal exposure =
m
∑
k=1
wjk (d)·Cik,amb.(d)+wjk (d)·Cjk,non-amb.(d)
• wjk(d): time individual i spent in microenvironment k on day d
• Cjk,amb.(d): PM2.5 concentration in microenvironment k on
day d due to outdoor sources
• Cjk,non-amb.(d): PM2.5 concentration in microenvironment k
on day d due to indoor sources
Veronica J. Berrocal Air pollution exposure
Exposure simulators
• How to derive estimates of personal exposure if we don’t have
data on individuals’ movement?
• Exposure simulators are stochastic models that estimate the
distribution of average personal exposure to a contaminant
• Initially developed for regulatory purposes: pNEM (Law et al.
1997), SHEDS-PM (Burke et al. 2001), APEX, pCNEM
(Zidek et al. 2003, 2007)
Veronica J. Berrocal Air pollution exposure
SHEDS-PM
• Given a day d
• SHEDS-PM simulates individuals with certain demographic
characteristics (i.e. age, sex, smoking status, etc.) living in a
spatial unit (usually, census tracts) according to proportions
obtained from the Census.
Veronica J. Berrocal Air pollution exposure
SHEDS-PM
• Given a day d
• SHEDS-PM simulates individuals with certain demographic
characteristics (i.e. age, sex, smoking status, etc.) living in a
spatial unit (usually, census tracts) according to proportions
obtained from the Census.
• To each simulated individual, SHEDS-PM assigns an activity
diary. The activity diaries come from the CHAD database.
Veronica J. Berrocal Air pollution exposure
SHEDS-PM
• Given a day d
• SHEDS-PM simulates individuals with certain demographic
characteristics (i.e. age, sex, smoking status, etc.) living in a
spatial unit (usually, census tracts) according to proportions
obtained from the Census.
• To each simulated individual, SHEDS-PM assigns an activity
diary. The activity diaries come from the CHAD database.
• SHEDS-PM simulates values for the input parameters, uses
PM2.5 concentration for the given day and derives PM2.5
concentration within each “microenvironment”.
Veronica J. Berrocal Air pollution exposure
SHEDS-PM
• Given a day d
• SHEDS-PM simulates individuals with certain demographic
characteristics (i.e. age, sex, smoking status, etc.) living in a
spatial unit (usually, census tracts) according to proportions
obtained from the Census.
• To each simulated individual, SHEDS-PM assigns an activity
diary. The activity diaries come from the CHAD database.
• SHEDS-PM simulates values for the input parameters, uses
PM2.5 concentration for the given day and derives PM2.5
concentration within each “microenvironment”.
• Finally, SHEDS-PM derives a personal exposure for the
simulated individual on the given day d.
Veronica J. Berrocal Air pollution exposure
Example: metrics of ambient exposure
without uncertainty
Day
DailyambientexposuretoPM2.5
0102030
2001−01−20 2001−04−21 2001−07−21 2001−10−27
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
Veronica J. Berrocal Air pollution exposure
Example: metrics of ambient exposure
without uncertainty
Day
DailyambientexposuretoPM2.5
0102030
2001−01−20 2001−04−21 2001−07−21 2001−10−27
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
1 Average exposure:
13.36µg/m3
Veronica J. Berrocal Air pollution exposure
Example: metrics of ambient exposure
without uncertainty
Day
DailyambientexposuretoPM2.5
0102030
2001−01−20 2001−04−21 2001−07−21 2001−10−27
Threshold: 15µg/m3
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
1 Average exposure:
13.36µg/m3
2 Percentage of days over
threshold: 35.92%
Veronica J. Berrocal Air pollution exposure
Example: metrics of ambient exposure
without uncertainty
Day
DailyambientexposuretoPM2.5
0102030
2001−01−20 2001−04−21 2001−07−21 2001−10−27
Threshold: 15µg/m3
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
1 Average exposure:
13.36µg/m3
2 Percentage of days over
threshold: 35.92%
3 Area above a threshold:
1.98(µg/m3)2
Veronica J. Berrocal Air pollution exposure
Example: metrics for personal exposure
Day
PersonalexposuretoPM2.5
05101520253035
2001−01−20 2001−04−21 2001−07−21 2001−10−27
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
• Set of 30 simulated individuals
Veronica J. Berrocal Air pollution exposure
Example: metrics for personal exposure
Day
PersonalexposuretoPM2.5
05101520253035
2001−01−20 2001−04−21 2001−07−21 2001−10−27
• Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
• Set of 30 simulated individuals
• For each simulated individual, we
compute the 3 metrics of exposure.
• The 30 values are considered
equally likely and represent
the distribution of metrics of
personal exposure for this particular
subject.
Veronica J. Berrocal Air pollution exposure
Example: metrics for personal exposure
0 10 20 30 40 50 60
0.000.050.100.15
Average personal exposure to PM 2.5
Density
Average personal exposure • Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
• Set of 30 simulated individuals
• For each simulated individual, we
compute the 3 metrics of exposure.
• The 30 values are considered
equally likely and represent
the distribution of metrics of
personal exposure for this particular
subject.
Veronica J. Berrocal Air pollution exposure
Example: metrics for personal exposure
0 20 40 60 80 100
0.000.010.020.03
Percentage days personal exposure over 15µg/m3
Density
Percentage of days over threshold • Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
• Set of 30 simulated individuals
• For each simulated individual, we
compute the 3 metrics of exposure.
• The 30 values are considered
equally likely and represent
the distribution of metrics of
personal exposure for this particular
subject.
Veronica J. Berrocal Air pollution exposure
Example: metrics for personal exposure
0 5 10 15
0.000.050.100.150.200.250.30
Normalized area personal exposure over 15µg/m3
Density
Area above threshold • Spatial unit: census tract in
North Carolina
• Period of exposure (Tij ):
January 20 to October 27, 2001
• Set of 30 simulated individuals
• For each simulated individual, we
compute the 3 metrics of exposure.
• The 30 values are considered
equally likely and represent
the distribution of metrics of
personal exposure for this particular
subject.
Veronica J. Berrocal Air pollution exposure
Birthweight and air pollution
• Health outcome: birthweight (gr) of children born between 2001
and 2002 in 14 counties in North Carolina (N=49,689).
• Exposure: predicted ambient maternal exposure to PM2.5 using a
data fusion model, and personal exposure using SHEDS-PM.
• Considered different exposure metric and time window of exposure.
• Window of exposure: entire pregnancy.
• Estimates of coefficients with 95% credible intervals.
Exposure metric Personal exposure Ambient exposure
Average exposure 11.04g (-193.74g; 160.65g) 20.27g (-179.23g; 209.42g)
Percentage days
above threshold 0.27g (-0.28g; 0.82g) 0.28g (-0.25g; 0.84g)
Area
above threshold 19.57g (-89.17g; 135.18g) 32.05g (-28.04g; 96.18g)
Veronica J. Berrocal Air pollution exposure
Discussion
• Environmental epidemiologists use multiple data sources
• Varying spatial and temporal resolution
• Varying level of precision and quality (what data to use? and
where?)
• Some are provided with measures of uncertainty (e.g. Census
estimates), most are not
• Some data formats are not easy to use
• Many unresolved issues (and new potential data sources)
• What metric of exposure? Avg vs apparent temperature? Etc.
• Spatially resolved estimates of exposure, but how to account
for uncertainty?
• Data on individuals’ movement to derive personal exposure
• How to handle multiple environmental exposures?
• Larger spatial datasets, long time series: computational burden.
• Privacy issues with health data.
• Typical to do sensitivity analyses, changing definitions of exposure
metrics.
Veronica J. Berrocal Air pollution exposure

More Related Content

What's hot

The Development of a Fire Vulnerability Index for the Mediterranean Region200...
The Development of a Fire Vulnerability Index for the Mediterranean Region200...The Development of a Fire Vulnerability Index for the Mediterranean Region200...
The Development of a Fire Vulnerability Index for the Mediterranean Region200...grssieee
 
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...Assessment of the Environmental Impact of Landfill Sites in the East Riding o...
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...Mark Kwabena Gadogbe
 
Remote sensing application in monitoring and management of soil, water and ai...
Remote sensing application in monitoring and management of soil, water and ai...Remote sensing application in monitoring and management of soil, water and ai...
Remote sensing application in monitoring and management of soil, water and ai...Jayvir Solanki
 
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...Manat Srivanit
 
32-2_ZBKlaic_et_al_2
32-2_ZBKlaic_et_al_232-2_ZBKlaic_et_al_2
32-2_ZBKlaic_et_al_2Sarah Ollier
 

What's hot (7)

Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
 
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
Program on Mathematical and Statistical Methods for Climate and the Earth Sys...
 
The Development of a Fire Vulnerability Index for the Mediterranean Region200...
The Development of a Fire Vulnerability Index for the Mediterranean Region200...The Development of a Fire Vulnerability Index for the Mediterranean Region200...
The Development of a Fire Vulnerability Index for the Mediterranean Region200...
 
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...Assessment of the Environmental Impact of Landfill Sites in the East Riding o...
Assessment of the Environmental Impact of Landfill Sites in the East Riding o...
 
Remote sensing application in monitoring and management of soil, water and ai...
Remote sensing application in monitoring and management of soil, water and ai...Remote sensing application in monitoring and management of soil, water and ai...
Remote sensing application in monitoring and management of soil, water and ai...
 
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...
A Classification Urban Precinct Ventilation Zones using Key Indicators of Spa...
 
32-2_ZBKlaic_et_al_2
32-2_ZBKlaic_et_al_232-2_ZBKlaic_et_al_2
32-2_ZBKlaic_et_al_2
 

Similar to CLIM Program: Remote Sensing Workshop, Environmental Exposure in Environmental Epidemiological Studies: Modeling Approaches and Challenges - Veronica Berrocal, Feb 13, 2018

25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx
25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx
25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptxPriyankaSharma89719
 
Ecological study
Ecological studyEcological study
Ecological studyNik Ronaidi
 
Environmental Epidemiology in Small areas
Environmental Epidemiology in Small areasEnvironmental Epidemiology in Small areas
Environmental Epidemiology in Small areasNik Ronaidi
 
Climate change and health epidemiologic methods - Dr Dung Phung
Climate change and health epidemiologic methods  - Dr Dung PhungClimate change and health epidemiologic methods  - Dr Dung Phung
Climate change and health epidemiologic methods - Dr Dung Phungintasave-caribsavegroup
 
Exposure to traffic related air pollution and the onset of childhood asthma ...
Exposure to traffic related air pollution and the onset of childhood asthma  ...Exposure to traffic related air pollution and the onset of childhood asthma  ...
Exposure to traffic related air pollution and the onset of childhood asthma ...Institute for Transport Studies (ITS)
 
ANALYTICAL STUDIES.pptx
ANALYTICAL STUDIES.pptxANALYTICAL STUDIES.pptx
ANALYTICAL STUDIES.pptxpayalrathod14
 
Unit 6 environmental_impact_assessment
Unit 6 environmental_impact_assessmentUnit 6 environmental_impact_assessment
Unit 6 environmental_impact_assessmentHarish kumar Lekkala
 
Lecture on Environmental Impact Assessment.pdf
Lecture on Environmental Impact Assessment.pdfLecture on Environmental Impact Assessment.pdf
Lecture on Environmental Impact Assessment.pdfapratim7
 
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...ahmedameen85
 
Assessment of Exposure to Environmental Health
Assessment of Exposure to Environmental HealthAssessment of Exposure to Environmental Health
Assessment of Exposure to Environmental HealthThomas Ayalew
 
Module-1-prediction-of-health-impacts.ppt
Module-1-prediction-of-health-impacts.pptModule-1-prediction-of-health-impacts.ppt
Module-1-prediction-of-health-impacts.pptAbshiroBeyene1
 
Impact prediction, evaluation and mitigation in eia
Impact prediction, evaluation and mitigation in eiaImpact prediction, evaluation and mitigation in eia
Impact prediction, evaluation and mitigation in eiaMizanur R. Shohel
 
Forbes co2 and temperature presentation for earth day at cua april 22 2015 ...
Forbes   co2 and temperature presentation for earth day at cua april 22 2015 ...Forbes   co2 and temperature presentation for earth day at cua april 22 2015 ...
Forbes co2 and temperature presentation for earth day at cua april 22 2015 ...Kevin Forbes
 
HEALTH AND CLIMATE CHANGE.pptx
HEALTH AND CLIMATE CHANGE.pptxHEALTH AND CLIMATE CHANGE.pptx
HEALTH AND CLIMATE CHANGE.pptxROBIN VAVACHAN
 
Eia prediction, evaluation & mitigration
Eia prediction, evaluation & mitigrationEia prediction, evaluation & mitigration
Eia prediction, evaluation & mitigrationMizanur R. Shohel
 
Raskob Iscram 2009
Raskob Iscram 2009Raskob Iscram 2009
Raskob Iscram 2009guestee5a52
 
3 Research methods and materials (1).pptx
3 Research methods and materials (1).pptx3 Research methods and materials (1).pptx
3 Research methods and materials (1).pptxestelaabera
 

Similar to CLIM Program: Remote Sensing Workshop, Environmental Exposure in Environmental Epidemiological Studies: Modeling Approaches and Challenges - Veronica Berrocal, Feb 13, 2018 (20)

25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx
25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx
25.-Introduction-to-Air-Pollution-Epidemiology_23Sep2020.pptx
 
AP Epidemiology.pptx
AP Epidemiology.pptxAP Epidemiology.pptx
AP Epidemiology.pptx
 
Ecological study
Ecological studyEcological study
Ecological study
 
Environmental Epidemiology in Small areas
Environmental Epidemiology in Small areasEnvironmental Epidemiology in Small areas
Environmental Epidemiology in Small areas
 
Climate change and health epidemiologic methods - Dr Dung Phung
Climate change and health epidemiologic methods  - Dr Dung PhungClimate change and health epidemiologic methods  - Dr Dung Phung
Climate change and health epidemiologic methods - Dr Dung Phung
 
Exposure to traffic related air pollution and the onset of childhood asthma ...
Exposure to traffic related air pollution and the onset of childhood asthma  ...Exposure to traffic related air pollution and the onset of childhood asthma  ...
Exposure to traffic related air pollution and the onset of childhood asthma ...
 
ANALYTICAL STUDIES.pptx
ANALYTICAL STUDIES.pptxANALYTICAL STUDIES.pptx
ANALYTICAL STUDIES.pptx
 
Unit 6 environmental_impact_assessment
Unit 6 environmental_impact_assessmentUnit 6 environmental_impact_assessment
Unit 6 environmental_impact_assessment
 
Lecture on Environmental Impact Assessment.pdf
Lecture on Environmental Impact Assessment.pdfLecture on Environmental Impact Assessment.pdf
Lecture on Environmental Impact Assessment.pdf
 
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...
Identification and Evaluation of Cercospora Leaf Spot of Sugar Beet by using ...
 
Assessment of Exposure to Environmental Health
Assessment of Exposure to Environmental HealthAssessment of Exposure to Environmental Health
Assessment of Exposure to Environmental Health
 
Module-1-prediction-of-health-impacts.ppt
Module-1-prediction-of-health-impacts.pptModule-1-prediction-of-health-impacts.ppt
Module-1-prediction-of-health-impacts.ppt
 
Impact prediction, evaluation and mitigation in eia
Impact prediction, evaluation and mitigation in eiaImpact prediction, evaluation and mitigation in eia
Impact prediction, evaluation and mitigation in eia
 
Forbes co2 and temperature presentation for earth day at cua april 22 2015 ...
Forbes   co2 and temperature presentation for earth day at cua april 22 2015 ...Forbes   co2 and temperature presentation for earth day at cua april 22 2015 ...
Forbes co2 and temperature presentation for earth day at cua april 22 2015 ...
 
HEALTH AND CLIMATE CHANGE.pptx
HEALTH AND CLIMATE CHANGE.pptxHEALTH AND CLIMATE CHANGE.pptx
HEALTH AND CLIMATE CHANGE.pptx
 
CLIM: Transition Workshop - The Relationship Between Extreme Events and Human...
CLIM: Transition Workshop - The Relationship Between Extreme Events and Human...CLIM: Transition Workshop - The Relationship Between Extreme Events and Human...
CLIM: Transition Workshop - The Relationship Between Extreme Events and Human...
 
Eia prediction, evaluation & mitigration
Eia prediction, evaluation & mitigrationEia prediction, evaluation & mitigration
Eia prediction, evaluation & mitigration
 
Raskob Iscram 2009
Raskob Iscram 2009Raskob Iscram 2009
Raskob Iscram 2009
 
Sbs
SbsSbs
Sbs
 
3 Research methods and materials (1).pptx
3 Research methods and materials (1).pptx3 Research methods and materials (1).pptx
3 Research methods and materials (1).pptx
 

More from The Statistical and Applied Mathematical Sciences Institute

More from The Statistical and Applied Mathematical Sciences Institute (20)

Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
Causal Inference Opening Workshop - Latent Variable Models, Causal Inference,...
 
2019 Fall Series: Special Guest Lecture - 0-1 Phase Transitions in High Dimen...
2019 Fall Series: Special Guest Lecture - 0-1 Phase Transitions in High Dimen...2019 Fall Series: Special Guest Lecture - 0-1 Phase Transitions in High Dimen...
2019 Fall Series: Special Guest Lecture - 0-1 Phase Transitions in High Dimen...
 
Causal Inference Opening Workshop - Causal Discovery in Neuroimaging Data - F...
Causal Inference Opening Workshop - Causal Discovery in Neuroimaging Data - F...Causal Inference Opening Workshop - Causal Discovery in Neuroimaging Data - F...
Causal Inference Opening Workshop - Causal Discovery in Neuroimaging Data - F...
 
Causal Inference Opening Workshop - Smooth Extensions to BART for Heterogeneo...
Causal Inference Opening Workshop - Smooth Extensions to BART for Heterogeneo...Causal Inference Opening Workshop - Smooth Extensions to BART for Heterogeneo...
Causal Inference Opening Workshop - Smooth Extensions to BART for Heterogeneo...
 
Causal Inference Opening Workshop - A Bracketing Relationship between Differe...
Causal Inference Opening Workshop - A Bracketing Relationship between Differe...Causal Inference Opening Workshop - A Bracketing Relationship between Differe...
Causal Inference Opening Workshop - A Bracketing Relationship between Differe...
 
Causal Inference Opening Workshop - Testing Weak Nulls in Matched Observation...
Causal Inference Opening Workshop - Testing Weak Nulls in Matched Observation...Causal Inference Opening Workshop - Testing Weak Nulls in Matched Observation...
Causal Inference Opening Workshop - Testing Weak Nulls in Matched Observation...
 
Causal Inference Opening Workshop - Difference-in-differences: more than meet...
Causal Inference Opening Workshop - Difference-in-differences: more than meet...Causal Inference Opening Workshop - Difference-in-differences: more than meet...
Causal Inference Opening Workshop - Difference-in-differences: more than meet...
 
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
Causal Inference Opening Workshop - New Statistical Learning Methods for Esti...
 
Causal Inference Opening Workshop - Bipartite Causal Inference with Interfere...
Causal Inference Opening Workshop - Bipartite Causal Inference with Interfere...Causal Inference Opening Workshop - Bipartite Causal Inference with Interfere...
Causal Inference Opening Workshop - Bipartite Causal Inference with Interfere...
 
Causal Inference Opening Workshop - Bridging the Gap Between Causal Literatur...
Causal Inference Opening Workshop - Bridging the Gap Between Causal Literatur...Causal Inference Opening Workshop - Bridging the Gap Between Causal Literatur...
Causal Inference Opening Workshop - Bridging the Gap Between Causal Literatur...
 
Causal Inference Opening Workshop - Some Applications of Reinforcement Learni...
Causal Inference Opening Workshop - Some Applications of Reinforcement Learni...Causal Inference Opening Workshop - Some Applications of Reinforcement Learni...
Causal Inference Opening Workshop - Some Applications of Reinforcement Learni...
 
Causal Inference Opening Workshop - Bracketing Bounds for Differences-in-Diff...
Causal Inference Opening Workshop - Bracketing Bounds for Differences-in-Diff...Causal Inference Opening Workshop - Bracketing Bounds for Differences-in-Diff...
Causal Inference Opening Workshop - Bracketing Bounds for Differences-in-Diff...
 
Causal Inference Opening Workshop - Assisting the Impact of State Polcies: Br...
Causal Inference Opening Workshop - Assisting the Impact of State Polcies: Br...Causal Inference Opening Workshop - Assisting the Impact of State Polcies: Br...
Causal Inference Opening Workshop - Assisting the Impact of State Polcies: Br...
 
Causal Inference Opening Workshop - Experimenting in Equilibrium - Stefan Wag...
Causal Inference Opening Workshop - Experimenting in Equilibrium - Stefan Wag...Causal Inference Opening Workshop - Experimenting in Equilibrium - Stefan Wag...
Causal Inference Opening Workshop - Experimenting in Equilibrium - Stefan Wag...
 
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
Causal Inference Opening Workshop - Targeted Learning for Causal Inference Ba...
 
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
Causal Inference Opening Workshop - Bayesian Nonparametric Models for Treatme...
 
2019 Fall Series: Special Guest Lecture - Adversarial Risk Analysis of the Ge...
2019 Fall Series: Special Guest Lecture - Adversarial Risk Analysis of the Ge...2019 Fall Series: Special Guest Lecture - Adversarial Risk Analysis of the Ge...
2019 Fall Series: Special Guest Lecture - Adversarial Risk Analysis of the Ge...
 
2019 Fall Series: Professional Development, Writing Academic Papers…What Work...
2019 Fall Series: Professional Development, Writing Academic Papers…What Work...2019 Fall Series: Professional Development, Writing Academic Papers…What Work...
2019 Fall Series: Professional Development, Writing Academic Papers…What Work...
 
2019 GDRR: Blockchain Data Analytics - Machine Learning in/for Blockchain: Fu...
2019 GDRR: Blockchain Data Analytics - Machine Learning in/for Blockchain: Fu...2019 GDRR: Blockchain Data Analytics - Machine Learning in/for Blockchain: Fu...
2019 GDRR: Blockchain Data Analytics - Machine Learning in/for Blockchain: Fu...
 
2019 GDRR: Blockchain Data Analytics - QuTrack: Model Life Cycle Management f...
2019 GDRR: Blockchain Data Analytics - QuTrack: Model Life Cycle Management f...2019 GDRR: Blockchain Data Analytics - QuTrack: Model Life Cycle Management f...
2019 GDRR: Blockchain Data Analytics - QuTrack: Model Life Cycle Management f...
 

Recently uploaded

Paris 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityParis 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityGeoBlogs
 
Hybridoma Technology ( Production , Purification , and Application )
Hybridoma Technology  ( Production , Purification , and Application  ) Hybridoma Technology  ( Production , Purification , and Application  )
Hybridoma Technology ( Production , Purification , and Application ) Sakshi Ghasle
 
Z Score,T Score, Percential Rank and Box Plot Graph
Z Score,T Score, Percential Rank and Box Plot GraphZ Score,T Score, Percential Rank and Box Plot Graph
Z Score,T Score, Percential Rank and Box Plot GraphThiyagu K
 
Interactive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationInteractive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationnomboosow
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingTechSoup
 
Student login on Anyboli platform.helpin
Student login on Anyboli platform.helpinStudent login on Anyboli platform.helpin
Student login on Anyboli platform.helpinRaunakKeshri1
 
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991RKavithamani
 
Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactPECB
 
CARE OF CHILD IN INCUBATOR..........pptx
CARE OF CHILD IN INCUBATOR..........pptxCARE OF CHILD IN INCUBATOR..........pptx
CARE OF CHILD IN INCUBATOR..........pptxGaneshChakor2
 
Sanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfSanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfsanyamsingh5019
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeThiyagu K
 
How to Make a Pirate ship Primary Education.pptx
How to Make a Pirate ship Primary Education.pptxHow to Make a Pirate ship Primary Education.pptx
How to Make a Pirate ship Primary Education.pptxmanuelaromero2013
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfciinovamais
 
Mastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionMastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionSafetyChain Software
 
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdfBASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdfSoniaTolstoy
 

Recently uploaded (20)

Código Creativo y Arte de Software | Unidad 1
Código Creativo y Arte de Software | Unidad 1Código Creativo y Arte de Software | Unidad 1
Código Creativo y Arte de Software | Unidad 1
 
Paris 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activityParis 2024 Olympic Geographies - an activity
Paris 2024 Olympic Geographies - an activity
 
Hybridoma Technology ( Production , Purification , and Application )
Hybridoma Technology  ( Production , Purification , and Application  ) Hybridoma Technology  ( Production , Purification , and Application  )
Hybridoma Technology ( Production , Purification , and Application )
 
Z Score,T Score, Percential Rank and Box Plot Graph
Z Score,T Score, Percential Rank and Box Plot GraphZ Score,T Score, Percential Rank and Box Plot Graph
Z Score,T Score, Percential Rank and Box Plot Graph
 
Interactive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communicationInteractive Powerpoint_How to Master effective communication
Interactive Powerpoint_How to Master effective communication
 
Grant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy ConsultingGrant Readiness 101 TechSoup and Remy Consulting
Grant Readiness 101 TechSoup and Remy Consulting
 
Student login on Anyboli platform.helpin
Student login on Anyboli platform.helpinStudent login on Anyboli platform.helpin
Student login on Anyboli platform.helpin
 
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991
Industrial Policy - 1948, 1956, 1973, 1977, 1980, 1991
 
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdfTataKelola dan KamSiber Kecerdasan Buatan v022.pdf
TataKelola dan KamSiber Kecerdasan Buatan v022.pdf
 
Beyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global ImpactBeyond the EU: DORA and NIS 2 Directive's Global Impact
Beyond the EU: DORA and NIS 2 Directive's Global Impact
 
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"Mattingly "AI & Prompt Design: The Basics of Prompt Design"
Mattingly "AI & Prompt Design: The Basics of Prompt Design"
 
CARE OF CHILD IN INCUBATOR..........pptx
CARE OF CHILD IN INCUBATOR..........pptxCARE OF CHILD IN INCUBATOR..........pptx
CARE OF CHILD IN INCUBATOR..........pptx
 
Sanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdfSanyam Choudhary Chemistry practical.pdf
Sanyam Choudhary Chemistry practical.pdf
 
Measures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and ModeMeasures of Central Tendency: Mean, Median and Mode
Measures of Central Tendency: Mean, Median and Mode
 
Staff of Color (SOC) Retention Efforts DDSD
Staff of Color (SOC) Retention Efforts DDSDStaff of Color (SOC) Retention Efforts DDSD
Staff of Color (SOC) Retention Efforts DDSD
 
How to Make a Pirate ship Primary Education.pptx
How to Make a Pirate ship Primary Education.pptxHow to Make a Pirate ship Primary Education.pptx
How to Make a Pirate ship Primary Education.pptx
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
 
Mastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory InspectionMastering the Unannounced Regulatory Inspection
Mastering the Unannounced Regulatory Inspection
 
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
Mattingly "AI & Prompt Design: Structured Data, Assistants, & RAG"
 
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdfBASLIQ CURRENT LOOKBOOK  LOOKBOOK(1) (1).pdf
BASLIQ CURRENT LOOKBOOK LOOKBOOK(1) (1).pdf
 

CLIM Program: Remote Sensing Workshop, Environmental Exposure in Environmental Epidemiological Studies: Modeling Approaches and Challenges - Veronica Berrocal, Feb 13, 2018

  • 1. Environmental exposure in environmental epidemiological studies: modeling approaches and challenges Veronica J. Berrocal Department of Biostatistics University of Michigan SAMSI Workshop on Remote Sensing, Uncertainty Quantification, and Theory of Data Systems Veronica J. Berrocal Air pollution exposure
  • 2. Environmental epidemiological studies • Environmental epidemiological studies aim to establish an association (potentially, a causal one) between an health outcome and an environmental exposure. • Typically, health outcome refers to: • mortality (all-cause non accidental deaths, or due to cardiovascular diseases, respiratory diseases, etc.) • hospitalizations/emergency visits • pregnancy outcomes and birth defects • ... • Environmental exposure refers to: • exposure to temperature, heat/heatwave, cold/cold waves • air pollution/wildfires • pesticides • precipitation • pollen • ... Veronica J. Berrocal Air pollution exposure
  • 3. Environmental epidemiological studies: this talk • Focus only on air pollution and temperature as environmental exposures. • The statistical methods used are very similar: • same type of health models are used to link health and environmental exposure • methods for air pollution exposure assessment are slighly ahead of those for temperature, but the gap is closing Veronica J. Berrocal Air pollution exposure
  • 4. Health data • Typically health data is obtained from: • administrative database: mortality from the National Center for Health Statistics; death and birth records from local and State Departments of Health −→ easier to obtain but aggregated over some areal unit • hospital and emergency visit records: individual level data −→ hard to obtain • Department of Human Health and Services: Medicare and Medicaid databases −→ expensive to obtain; require a lot of data cleaning and formatting • cohort studies: individual level data, smaller sample size −→ need a collaborator to obtain • Earlier studies used mostly administrative databases aggregated over counties or larger metropolitan areas. • Now, common to use geocoded health data, e.g. individuals are assigned specific geographical coordinates Veronica J. Berrocal Air pollution exposure
  • 5. Health data • Typically health data is obtained from: • administrative database: mortality from the National Center for Health Statistics; death and birth records from local and State Departments of Health −→ easier to obtain but aggregated over some areal unit • hospital and emergency visit records: individual level data −→ hard to obtain • Department of Human Health and Services: Medicare and Medicaid databases −→ expensive to obtain; require a lot of data cleaning and formatting • cohort studies: individual level data, smaller sample size −→ need a collaborator to obtain • Earlier studies used mostly administrative databases aggregated over counties or larger metropolitan areas. • Now, common to use geocoded health data, e.g. individuals are assigned specific geographical coordinates =⇒ potential for geocoding error Veronica J. Berrocal Air pollution exposure
  • 6. Environmental exposure data • Typically environmental exposure data is obtained from: • outdoor monitors (air pollution, temperature, humidity) - sources: NOAA/NCEI, EPA • dispersion model outputs (mostly for traffic related pollutants) - sources: EPA, collaborators • air quality model outputs - sources: EPA, collaborators • satellite data (MODIS AOD, Aura OMI, etc.) - sources NASA • personal monitors • private monitors (e.g. Harvard Six City study, etc.) - sources: NASA • Temporal and spatial resolution varies across data sources. • Informative sampling. • Availability and access vary across data sources. • Uncertainties and measurement errors vary across data sources. Veronica J. Berrocal Air pollution exposure
  • 7. Health analysis • The statistical model used in the health model depends on the spatial resolution of the health data. • Most commonly used study design is a time series design • Health outcome: number of deaths on day t aggregated over a unit • Poisson regression (GAM, overdispersion) • logExpected mortalityt = S(t;λ1)+ S(env expt;λ2)+ S(lagged env exp.;λ3)+ g(confound.) • S(·;λ): smooth function with λ d.f. Figure: From Bhaskaran et al. (2003) Veronica J. Berrocal Air pollution exposure
  • 8. Health analysis • In time series study, the environmental exposure is representative of the entire areal unit −→ usually, area-wide average from outdoor monitors Veronica J. Berrocal Air pollution exposure
  • 9. Health analysis • In time series study, the environmental exposure is representative of the entire areal unit −→ usually, area-wide average from outdoor monitors • No accounting for intra-urban variation in exposure • Usually fit at individual locations separately and then combined together through a hierarchical approach • Effect of environmental exposure at each areal unit linked to areal unit characteristics −→ use multiple data sources (Census, National Land Cover Dataset, etc.) Veronica J. Berrocal Air pollution exposure
  • 10. Health analysis • In time series study, the environmental exposure is representative of the entire areal unit −→ usually, area-wide average from outdoor monitors • No accounting for intra-urban variation in exposure • Usually fit at individual locations separately and then combined together through a hierarchical approach • Effect of environmental exposure at each areal unit linked to areal unit characteristics −→ use multiple data sources (Census, National Land Cover Dataset, etc.) • If individual lavel health data is available, other study design are possible • case-crossover design • Cox proportional hazard models • linear regression models Veronica J. Berrocal Air pollution exposure
  • 11. Environmental exposure • Earlier studies on the health effect of air pollution or temperature used area-wide averages from monitors as exposure =⇒ this approach ignores intra-urban variation in exposure. • Exploiting intra-urban contrasts has several advantages: • increased power • rule out unmeasured confounders • differentiate between different environmental exposures (e.g. pollutant, temperature vs dew point, etc.) • Environmental exposure assessment methods are used more frequely for air pollution exposure, but are starting to be more routinely used also for temperature Veronica J. Berrocal Air pollution exposure
  • 12. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to locations/individual subjects: 1 Nearest monitor(s)/Average over a buffer: derive exposure using measurement from the nearest monitor or the monitors within a buffer. • Pro: simple • Cons: too simplistic; does not capture small-scale spatial variation Veronica J. Berrocal Air pollution exposure
  • 13. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to locations/individual subjects: 3 Land use regression (LUR): predict environmental exposure using a linear regression model with predictors GIS covariates, e.g. type of roads within a buffer, air conditioning (Briggs et al. 1997, 2000; Brauer et al. 2003; Ross et al. 2006). • Pro: simple • Cons: large number of GIS covariates (variable selection technique needed: backward/forward selection; LASSO; PLS), limited transferability to different locations, residual independence between sites. Veronica J. Berrocal Air pollution exposure
  • 14. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to locations/individual subjects: 3 Land use regression (LUR): predict environmental exposure using a linear regression model with predictors GIS covariates, e.g. type of roads within a buffer, air conditioning (Briggs et al. 1997, 2000; Brauer et al. 2003; Ross et al. 2006). • Pro: simple • Cons: large number of GIS covariates (variable selection technique needed: backward/forward selection; LASSO; PLS), limited transferability to different locations, residual independence between sites. • In recent studies, satellite data has also been used (Kloog et al. 2011). • Multiple-stage models needed to account for extensive missing satellite data. • Uncertainty in each model stage is not properly accounted for. Veronica J. Berrocal Air pollution exposure
  • 15. Using satellite data (Kloog et al. 2017) • (1) Daily average temperature from monitoring stations from M´et´o France, (2) daily temperature data from MODIS, Terra instrument, at 1km resolution • Two step-model: 1 Calibration of satellite data 2 Prediction Veronica J. Berrocal Air pollution exposure
  • 16. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to locations/individual subjects: 4 Universal kriging (UK) models/Hybrid models: predict exposure using a spatial statistical model with predictors GIS covariates (Mercer et al. 2011; Bergen et al., 2013; Kloog et al. 2012 - includes satellite data). • Pro: accounts for spatial dependence between sites • Cons: requires knowledge of spatial statistics; cannot be implemented directly in ArcGIS, although possible through a two-stage approach. Still requires use of variable selection techniques to reduce number of GIS covariates. • UK has been shown to yield better predictions than LUR (Beelen et al. 2009; Mercer et al. 2011) Veronica J. Berrocal Air pollution exposure
  • 17. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to location/individual subjects: 5 Dispersion modeling/Regional climate model: use as exposure the output of a dispersion model/regional climate model (Nafstad et al. 2004; Penard-Morand et al. 2006). • Pro: Based on physical principles; non-linearity; better than LUR for specific-source related component. • Cons: output often uncalibrated; smooth exposure surfaces; different spatial resolution. Veronica J. Berrocal Air pollution exposure
  • 18. Environmental exposures: modeling approaches • Several different strategies are used to assign environmental exposure to locations/individual subjects: 6 Statistical downscaling/data fusion models: predict exposure combining the output of air quality/dispersion/regional climate models with monitoring data (Fuentes and Raftery 2005; McMillan et al.2010; Berrocal et al. 2010a,b, 2012, 2014; Reich et al. 2014; Rushworth et al. 2014; Lee et al. 2016). • Pro: accounts for physical and chemical processes; no need to include GIS covariates (Lindstrom et al. 2014); can account for complicated spatio-temporal dependence structure. • Cons: numerical model output not easily/publically available; requires knowledge of spatial, spatio-temporal and Bayesian statistics; software not always available. Veronica J. Berrocal Air pollution exposure
  • 19. Downscaler/data fusion models (Berrocal et al., 2010, 2011, 2012, 2014): 100 95 90 85 80 75 70 30354045 Longitude Latitude 20 40 60 80 100 120 140 Observed ozone concentration August 1, 2001 100 95 90 85 80 75 70 30354045 Longitude Latitude 20 40 60 80 100 120 140 CMAQ estimate of ozone concentration August 1, 2001 • Main idea: leverage two sources of information: • the true sparse, point-referenced monitors measurements: Y (s), s ∈ S • the spatially dense, grid-based, potentially uncalibrated, model output X(B) with B 12-km grid cell Veronica J. Berrocal Air pollution exposure
  • 20. Data fusion model (Berrocal et al. 2012) • Our most recent data fusion model is: Observed concentration at s = β0(s)+β1 · ˜X(s)+ε(s) Y (s) with • β0(s) modeled as a spatial process (e.g. Gaussian process with a given covariance function). • ˜X(s) smoothed version of model output at each point s, obtained using random, directional, spatially-varying and spatially correlated weights. • ε(s) ∼ N(0,τ2 ). Veronica J. Berrocal Air pollution exposure
  • 21. Data fusion model (Berrocal et al. 2012) • Our most recent data fusion model is: Observed concentration at s = β0(s)+β1 · ˜X(s)+ε(s) Y (s) with • β0(s) modeled as a spatial process (e.g. Gaussian process with a given covariance function). • ˜X(s) smoothed version of model output at each point s, obtained using random, directional, spatially-varying and spatially correlated weights. • ε(s) ∼ N(0,τ2 ). • Other extensions of this data fusion model have been proposed. • Used by EPA to derive fused surfaces of daily average PM2.5 and daily 8-hour maximum ozone concentration at census tract centroids for US • Website https://www.epa.gov/air-research/ downscaler-model-predicting-daily-air-pollution) provides predictions and their standard deviations Veronica J. Berrocal Air pollution exposure
  • 22. More monitoring data? Caution! (Berrocal and Holland, 20XX) • The downscaler model uses monitoring data to calibrate the air quality model output. • We used data from the Federally Regulated Monitors for PM2.5: the FRM monitors. • Measurements of PM2.5 are nowadays available also from other monitors. Are they useful? • Here we consider “new”, automated monitors that measure PM2.5 semi-continuously, the SC-FEM monitors, and we evaluate whether adding these set of monitors in a downscaler model leads to better predictions. Veronica J. Berrocal Air pollution exposure
  • 23. Is more monitoring data helpful? −120 −110 −100 −90 −80 −70 253035404550 Longitude Latitude qq q qq qq q q q q q q q q qq q q qq qq qq q q q q qq qq q q qqq q q qq q q qqq qq q q q q q q qqq q q q q qq q q q qq q q q q q q q q q qq q q qqq qqq q q q q qq q q q qq q q q q q q q q q qq q q q q qq q q qqqq q qq q q q q q q q qqq q q q q q q q qq q q q q q qq qq qq q q q q q q qq q q q q qq qq q q q q q q qq q q q q q q q q q q q q q q q q q q qqq q q q q q qqqqqqqq q q q qqq q qq q q q q q qq qqq q qq q qq q q q q q q q q q q qq q q q q qq q q q qqqqqq q qq q q qq q qqq q q qq q q q q q qqq qq q q q q q q q q q q q qq q q q qq q qq q qq q q q qq q q qq q qq q q q qq q qqqq qqq q q qqqqqq q q qq q q q q q q q qqqqqqqq q q q q q qq q q q q qq qq q q q qq q q q q q q q qq q q q q q qq q qq q q q qq q q q qq qqqq qqq q q qq q qqq q qq q q q q q q q q q q q q q q qq q q q q q q q q q q q qq q qq q qq q q q q q qq qq q q q qqqqqq qqq qq q q q q qqqq qq qqqqq qq q q q q q q qqq q q qq q q q q qq q q q qq q qqqqq q q q q q q q qq q q q qq q qq q q qq qqq q q q q qq q q q qqq q q q qq q qq q q q q q qq q q q q qq q qq q qq q q q q q q q qq q q q q qq qq q q q q q qq q q q q qq q q q q q q q qq q q q qqqq q q qq q q q q q q qqq q q q qq q qqq q qqq q qq q q q q qqq q q q qqq qq q q q q q q q q qq q q q q q q q q qq q q qq q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q qq q q qqq q q q q q q q q q q q q q q q q q q q q q q q q qq qq q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q qq qq q q q q q q q q q q q q q q q q q q q q q q q q q q qq qq q qq q q q qq q q q q qqq q q q q q q q q q q q q q q q q q q q q q q q q qq q qqq q q qq q q q q q q q q q q q q q q q q q q q qq q q q q q qq q q q q q qq qq qqqqqqqq q qqq q qq q q q q q q q q q q q q qq qq q qq q qq qq q q q q q q qq q qq q q q q qqq qq q qqqq q q q q q q q q q q q q q qq q q q q qq q qq qq q q qqqqqq q q q qq q qq q qq q q qq q q qq q qq qq q q q qq q q q q q q q q q q q q q q q q q q q q qq q q qq q q q q q q q q q q q q q q q q q q q qq q q q q q qq q q q q q q q q q q q q q qq q q q q qq q q q q q q q q q q q q q q q q q q q q qq q q q q q qq q q q q q q q q q qq q q q q q q q q q q q q q qq q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q qq q q q q q q q FRM SC/FEM • We apply a downscaler model to daily average PM2.5 concentration in the US for year 2011. • Each day we fit the model to the data, using 80% of the observations and holding 20% for validation of the predictions. • We compare a model that uses both FRM and SC/FEM monitoring data vs one that uses only FRM data. Veronica J. Berrocal Air pollution exposure
  • 24. Results • MSE= Average of (observed−predicted)2 . • MAE= Average of |observed−predicted|. MSE MAE Width 95% CI Coverage 95% CI Monitors (µg/m3 ) (µg/m3 ) (µg/m3 ) (%) Only FRM 4.09 2.28 16.82 95.59 FRM and SC/FEM 4.03 2.09 17.45 96.29 • Moderate improvement overall. • Larger improvements when stratified by season, with large gains in Summer and Fall. Veronica J. Berrocal Air pollution exposure
  • 25. Caution with monitoring data! • Reduction in Mean Absolute Error by state when we compare the model that uses FRM + SC/FEM data vs a model that uses only FRM data. • Positive means better predictions using FRM + SC/FEM data. Percent reduction in MAE: SC/FEM + FRM data vs FRM only Fall PercentreductioninMeanAbsoluteError(%) Alabama Arizona Arkansas California Colorado Connecticut Delaware DistrictOfColumbia Florida Georgia Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada NewHampshire NewJersey NewMexico NewYork NorthCarolina NorthDakota Ohio Oklahoma Oregon Pennsylvania RhodeIsland SouthCarolina SouthDakota Tennessee Texas Utah Vermont Virginia Washington WestVirginia Wisconsin Wyoming −40−20020 Veronica J. Berrocal Air pollution exposure
  • 26. Challenges: measurement error • Predicted exposure is not the true exposure. • Using predicted exposure in a health model introduces two sources of measurement errors: • Berkson error: from smoothing the exposure surface =⇒ inflated SE of health effect estimates. • Classical measurement error: from estimation of the parameters in the exposure model =⇒ bias and SE inflation. • Not easy to separate the two. Veronica J. Berrocal Air pollution exposure
  • 27. How to correct for measurement error • Various papers have addressed this issue (Gryparis et al. 2009; Szpiro et al. 2011; Szpiro and Paciorek, 2013; Alexeef et al. 2015; etc). • Proposed approaches include: 1 Bayesian joint model of exposure and disease: this uses the correct imputation scheme (see missing data literature - Rubin 1987, Little 1992). 2 Two-stage Bayesian approach: exposure model first, then health model. Posterior distribution for exposure is the prior in the health model. =⇒ Does not cut the feedback between exposure and health outcome. 3 Bootstrap approximation: resample health and exposure datat. • Option 1 is the ideal, but most computationally intensive. Veronica J. Berrocal Air pollution exposure
  • 28. Accounting for uncertainty in personal exposure • Personal exposure is not the same as ambient exposure • Using ambient exposure as a proxy for personal exposure 1 Biased estimates of the effect of air pollution (Zeger et al. 2000) 2 Wrong assessment of its uncertainty (Zeger et al. 2000; Gryparis et al. 2009) • Prototype analysis that shows how to account for exposure uncertainty in a study on the effect of personal or ambient maternal exposure to PM2.5 on birthweight. Veronica J. Berrocal Air pollution exposure
  • 29. Personal exposure Veronica J. Berrocal Air pollution exposure
  • 30. Ambient vs personal exposure • Given a day d, let Cj (d) be outdoor concentration of PM2.5 at the location where subject j lives . • Ambient exposure for individual j on day d is the outdoor concentration Cj (d). Veronica J. Berrocal Air pollution exposure
  • 31. Ambient vs personal exposure • Given a day d, let Cj (d) be outdoor concentration of PM2.5 at the location where subject j lives . • Ambient exposure for individual j on day d is the outdoor concentration Cj (d). • Personal exposure for individual j on day d is the sum of contributions from the various “microenvironments” the individual visits during the day: Personal exposure = m ∑ k=1 wjk (d)·Cik,amb.(d)+wjk (d)·Cjk,non-amb.(d) • wjk(d): time individual i spent in microenvironment k on day d • Cjk,amb.(d): PM2.5 concentration in microenvironment k on day d due to outdoor sources • Cjk,non-amb.(d): PM2.5 concentration in microenvironment k on day d due to indoor sources Veronica J. Berrocal Air pollution exposure
  • 32. Exposure simulators • How to derive estimates of personal exposure if we don’t have data on individuals’ movement? • Exposure simulators are stochastic models that estimate the distribution of average personal exposure to a contaminant • Initially developed for regulatory purposes: pNEM (Law et al. 1997), SHEDS-PM (Burke et al. 2001), APEX, pCNEM (Zidek et al. 2003, 2007) Veronica J. Berrocal Air pollution exposure
  • 33. SHEDS-PM • Given a day d • SHEDS-PM simulates individuals with certain demographic characteristics (i.e. age, sex, smoking status, etc.) living in a spatial unit (usually, census tracts) according to proportions obtained from the Census. Veronica J. Berrocal Air pollution exposure
  • 34. SHEDS-PM • Given a day d • SHEDS-PM simulates individuals with certain demographic characteristics (i.e. age, sex, smoking status, etc.) living in a spatial unit (usually, census tracts) according to proportions obtained from the Census. • To each simulated individual, SHEDS-PM assigns an activity diary. The activity diaries come from the CHAD database. Veronica J. Berrocal Air pollution exposure
  • 35. SHEDS-PM • Given a day d • SHEDS-PM simulates individuals with certain demographic characteristics (i.e. age, sex, smoking status, etc.) living in a spatial unit (usually, census tracts) according to proportions obtained from the Census. • To each simulated individual, SHEDS-PM assigns an activity diary. The activity diaries come from the CHAD database. • SHEDS-PM simulates values for the input parameters, uses PM2.5 concentration for the given day and derives PM2.5 concentration within each “microenvironment”. Veronica J. Berrocal Air pollution exposure
  • 36. SHEDS-PM • Given a day d • SHEDS-PM simulates individuals with certain demographic characteristics (i.e. age, sex, smoking status, etc.) living in a spatial unit (usually, census tracts) according to proportions obtained from the Census. • To each simulated individual, SHEDS-PM assigns an activity diary. The activity diaries come from the CHAD database. • SHEDS-PM simulates values for the input parameters, uses PM2.5 concentration for the given day and derives PM2.5 concentration within each “microenvironment”. • Finally, SHEDS-PM derives a personal exposure for the simulated individual on the given day d. Veronica J. Berrocal Air pollution exposure
  • 37. Example: metrics of ambient exposure without uncertainty Day DailyambientexposuretoPM2.5 0102030 2001−01−20 2001−04−21 2001−07−21 2001−10−27 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 Veronica J. Berrocal Air pollution exposure
  • 38. Example: metrics of ambient exposure without uncertainty Day DailyambientexposuretoPM2.5 0102030 2001−01−20 2001−04−21 2001−07−21 2001−10−27 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 1 Average exposure: 13.36µg/m3 Veronica J. Berrocal Air pollution exposure
  • 39. Example: metrics of ambient exposure without uncertainty Day DailyambientexposuretoPM2.5 0102030 2001−01−20 2001−04−21 2001−07−21 2001−10−27 Threshold: 15µg/m3 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 1 Average exposure: 13.36µg/m3 2 Percentage of days over threshold: 35.92% Veronica J. Berrocal Air pollution exposure
  • 40. Example: metrics of ambient exposure without uncertainty Day DailyambientexposuretoPM2.5 0102030 2001−01−20 2001−04−21 2001−07−21 2001−10−27 Threshold: 15µg/m3 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 1 Average exposure: 13.36µg/m3 2 Percentage of days over threshold: 35.92% 3 Area above a threshold: 1.98(µg/m3)2 Veronica J. Berrocal Air pollution exposure
  • 41. Example: metrics for personal exposure Day PersonalexposuretoPM2.5 05101520253035 2001−01−20 2001−04−21 2001−07−21 2001−10−27 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 • Set of 30 simulated individuals Veronica J. Berrocal Air pollution exposure
  • 42. Example: metrics for personal exposure Day PersonalexposuretoPM2.5 05101520253035 2001−01−20 2001−04−21 2001−07−21 2001−10−27 • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 • Set of 30 simulated individuals • For each simulated individual, we compute the 3 metrics of exposure. • The 30 values are considered equally likely and represent the distribution of metrics of personal exposure for this particular subject. Veronica J. Berrocal Air pollution exposure
  • 43. Example: metrics for personal exposure 0 10 20 30 40 50 60 0.000.050.100.15 Average personal exposure to PM 2.5 Density Average personal exposure • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 • Set of 30 simulated individuals • For each simulated individual, we compute the 3 metrics of exposure. • The 30 values are considered equally likely and represent the distribution of metrics of personal exposure for this particular subject. Veronica J. Berrocal Air pollution exposure
  • 44. Example: metrics for personal exposure 0 20 40 60 80 100 0.000.010.020.03 Percentage days personal exposure over 15µg/m3 Density Percentage of days over threshold • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 • Set of 30 simulated individuals • For each simulated individual, we compute the 3 metrics of exposure. • The 30 values are considered equally likely and represent the distribution of metrics of personal exposure for this particular subject. Veronica J. Berrocal Air pollution exposure
  • 45. Example: metrics for personal exposure 0 5 10 15 0.000.050.100.150.200.250.30 Normalized area personal exposure over 15µg/m3 Density Area above threshold • Spatial unit: census tract in North Carolina • Period of exposure (Tij ): January 20 to October 27, 2001 • Set of 30 simulated individuals • For each simulated individual, we compute the 3 metrics of exposure. • The 30 values are considered equally likely and represent the distribution of metrics of personal exposure for this particular subject. Veronica J. Berrocal Air pollution exposure
  • 46. Birthweight and air pollution • Health outcome: birthweight (gr) of children born between 2001 and 2002 in 14 counties in North Carolina (N=49,689). • Exposure: predicted ambient maternal exposure to PM2.5 using a data fusion model, and personal exposure using SHEDS-PM. • Considered different exposure metric and time window of exposure. • Window of exposure: entire pregnancy. • Estimates of coefficients with 95% credible intervals. Exposure metric Personal exposure Ambient exposure Average exposure 11.04g (-193.74g; 160.65g) 20.27g (-179.23g; 209.42g) Percentage days above threshold 0.27g (-0.28g; 0.82g) 0.28g (-0.25g; 0.84g) Area above threshold 19.57g (-89.17g; 135.18g) 32.05g (-28.04g; 96.18g) Veronica J. Berrocal Air pollution exposure
  • 47. Discussion • Environmental epidemiologists use multiple data sources • Varying spatial and temporal resolution • Varying level of precision and quality (what data to use? and where?) • Some are provided with measures of uncertainty (e.g. Census estimates), most are not • Some data formats are not easy to use • Many unresolved issues (and new potential data sources) • What metric of exposure? Avg vs apparent temperature? Etc. • Spatially resolved estimates of exposure, but how to account for uncertainty? • Data on individuals’ movement to derive personal exposure • How to handle multiple environmental exposures? • Larger spatial datasets, long time series: computational burden. • Privacy issues with health data. • Typical to do sensitivity analyses, changing definitions of exposure metrics. Veronica J. Berrocal Air pollution exposure