Original Article

Wildfire-specific Fine Particulate Matter and Risk of
Hospital Admissions in Urban and Rural Counties
Jia Coco Liu,a Ander Wilson,b Loretta J. Mickley,c Francesca Dominici,b Keita Ebisu,a Yun Wang,b
Melissa P. Sulprizio,c Roger D. Peng,d Xu Yue,c Ji-Young Son,a G. Brooke Anderson,e and Michelle L. Bella

Background: The health impacts of wildfire smoke, including fine
particles (PM2.5), are not well understood and may differ from those
of PM2.5 from other sources due to differences in concentrations and
chemical composition.
Methods: First, for the entire Western United States (561 counties)
for 2004–2009, we estimated daily PM2.5 concentrations directly
attributable to wildfires (wildfires-specific PM2.5), using a global
chemical transport model. Second, we defined smoke wave as ≥2
consecutive days with daily wildfire-specific PM2.5 > 20 μg/m3, with
sensitivity analysis considering 23, 28, and 37 μg/m3. Third, we estimated the risk of cardiovascular and respiratory hospital admissions
associated with smoke waves for Medicare enrollees. We used a generalized linear mixed model to estimate the relative risk of hospital
admissions on smoke wave days compared with matched comparison
days without wildfire smoke.
Results: We estimated that about 46 million people of all ages were
exposed to at least one smoke wave during 2004 to 2009 in the Western
United States. Of these, 5 million are Medicare enrollees (≥65 years).
We found a 7.2% (95% confidence interval: 0.25%, 15%) increase
in risk of respiratory admissions during smoke wave days with high
wildfire-specific PM2.5 (>37 μg/m3) compared with matched non
smoke wave days. We did not observe an association between smoke
Submitted 4 November 2015; accepted 9 September 2016.
From the aSchool of Forestry and Environmental Studies, Yale University,
New Haven, CT; bDepartment of Biostatistics, T.H. Chan School of Public Health, Harvard University, Cambridge, MA; cSchool of Engineering
and Applied Sciences, Harvard University, Cambridge, MA; dDepartment
of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD; and eDepartment of Environmental & Radiological Health
Sciences, College of Veterinary Medicine & Biomedical Sciences, Colorado State University, Fort Collins, CO.
Financial supported by NIH/NIEHS R21 ES022585-01, NIH R01 ES019560,
NIH R01 ES024332, EPA RD 83479801, NIH R21 ES021427, NIEHS
R21 ES020152, Yale Institute of Biospheric Studies Doctoral Dissertation
Improvement Grant.
This publication was developed under Assistance Agreement No. 83587101,
awarded by the U.S. Environmental Protection Agency to Yale University.
It has not been formally reviewed by EPA. The views expressed in this
document are solely those of the authors and do not necessarily reflect
those of the Agency. EPA does not endorse any products or commercial
services mentioned in this publication.
The authors report no conflicts of interest.
 Supplemental digital content is available through direct URL citations
in the HTML and PDF versions of this article (www.epidem.com).
Correspondence: Jia Coco Liu, Room 8B, 205 Prospect St, New Haven, CT,
06511. E-mail: coco.liu@yale.edu.
Copyright © 2016 Wolters Kluwer Health, Inc. All rights reserved.
ISSN: 1044-3983/16/2801-0077
DOI: 10.1097/EDE.0000000000000556

Epidemiology  •  Volume 28, Number 1, January 2017 

wave days with wildfire-specific PM2.5 ≤ 37 μg/m3 and respiratory
or cardiovascular admissions. Respiratory effects of wildfire-specific
PM2.5 may be stronger than that of PM2.5 from other sources.
Conclusion: Short-term exposure to wildfire-specific PM2.5was
associated with risk of respiratory diseases in the elderly population
in the Western United States during severe smoke days. See video
abstract at, http://links.lww.com/EDE/B137.
(Epidemiology 2017;28: 77–85)

W

ildfires are a growing concern, as climate change is
anticipated to increase their frequency, intensity, and
spreading speed.1 Wildfires are known to cause substantial
ecologic and economic burden, and the economic costs may
be underestimated because they do not account for the potentially severe impact of air pollution from wildfires on human
health.2 Understanding the public health impact of wildfire
smoke can inform intervention-focused policies to protect
population health and promote more accurate estimates of the
consequences of wildfires.3
The Western United States historically suffers from
wildfires4 due to large areas of forests, vegetation, and relatively arid weather. The burning of biomass can dramatically
increase levels of toxic air pollutants, such as fine particles
(PM2.5).5 Numerous studies have demonstrated links between
all-source particulate matter (PM) measured as total mass and
health outcomes, especially for respiratory and cardiovascular
diseases.3 Many studies have indicated that PM2.5 raises more
human health concerns than coarse PM because the smaller
particles penetrate the respiratory system more deeply.6
The health effects of wildfire-emitted fine particles are
not well understood. Wildfire smoke can increase ambient PM
levels several times higher than that on days with no wildfire
sources.3 The size of fire-generated PM tends to be small, such
as fine particles (PM2.5).7 The composition of wildfire-generated PM2.5 may be different from PM2.5 from other sources,
which in turn can affect toxicity.8,9 Wildfires are episodic,
making it especially challenging to link wildfire-specific air
pollution with health.
We previously performed a literature review of the
small number of studies on health impact of wildfire smoke
www.epidem.com     77

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017

Liu et al. 

on community populations. We found that the results on the
effects of wildfires on hospital admissions were inconsistent,
especially for cardiovascular diseases, in the Western United
States.3 To date, most of the literature focused on a single fire
episode and small population.10–12 It is unknown whether the
health impacts of wildfire-emitted PM2.5 differ from those of
PM2.5 from other sources. As a result, research that investigates health impact from wildfires on a large geographical
area and over a long time is needed.
The understanding of the health impact of wildfirerelated air pollution is hindered by the challenge of estimating
exposure to air pollution that can be specifically attributable to
wildfires. Ambient monitors measure PM2.5 concentration but
cannot distinguish the proportion directly attributable to fires
versus other sources. The majority of current wildfire-health
studies used air monitoring data, which are limited in spatial
(no monitors available in rural areas) and temporal resolution
(generally measure every 3–6 days) and cannot isolate wildfire-specific pollution.3
Our study aimed to address many of these challenges
described above. Using a chemical transport model, we could
fill in the spatial and temporal gaps of monitoring data and
make source attributions of the modeled PM2.5. We estimated
daily 2004–2009 PM2.5 concentrations specifically from wildfires for 561Western United States counties and linked them
to daily numbers of Medicare admissions for respiratory and
cardiovascular diseases. We applied statistical methods that
have not been previously used in wildfire-health studies and
estimated health impacts of wildfire-specific PM2.5, incorporating rural populations into statistical analysis.

METHODS
Study Domain
The study domain is the Western United States (latitude:
31–49, longitude: −101 to −125; eFigure 1; http://links.lww.
com/EDE/B109), where wild fires occur frequently.13 The
study region consists of 561 counties in 16 states.

Wildfire Modeling
We employed wildfire simulations from the GEOSChem chemical transport model (v9-01-03) to generate daily
wildfire-specific PM2.5levels for 6 years (2004–2009). GEOSChem is a global 3D atmospheric chemistry model driven by
meteorology.14 It has been used to understand the pollution
impact of present-day fires15,16 and to predict future wildfirespecific aerosols.1,17 The modeling integrates meteorological
data from Goddard Earth Observing System (GEOS-5) of
the NASA Modeling and Assimilation Office and observed
wildfire area burned based on the Global Fire Emissions Database (GFED3). GFED3 combines satellite observations of fire
counts, area burned, and fuel load to produce gridded, daily
maps of wildfire emissions.18,19
The GEOS-Chem simulation model outputs used
in this study are daily (24-hour average), gridded surface
78     www.epidem.com 

PM2.5concentrations for fire seasons (May 1–October 31)
2004–2009. The grid size is 0.5 × 0.67 degrees (approximately 50 × 75 km) latitude-by-longitude. We generated estimates under two simulations: (1) the “all-source PM2.5”: total
PM2.5 levels from all sources including wildfires; and (2)
“no-fire PM2.5”: PM2.5 from all sources except the contribution from wildfires, by performing model simulations without wildfire emissions. Non fire sources for PM2.5in the West
include fossil fuel combustion from transportation, industry,
and power plants.20,21 The difference between outputs from
these two simulations provides an estimate of the wildfirespecific PM2.5 for each day and grid cell. We defined exposure based on daily wildfire-specific PM2.5 estimates, which
may differ from the actual locations of wildfires as smoke
can travel large distances.22 This model provided exposure
estimates for all study subjects in the spatial domain, including those far from monitors. The results of GEOS-Chem
simulations on particulate matter have been validated against
observations.16,23 We use ground-based or aircraft measurements, not satellite data, to validate the GEOS-Chem surface PM2.5, including wildfire PM2.5(eAppendix Methods 2;
http://links.lww.com/EDE/B109).
The modeled estimates of PM2.5 from wildfires were
spatially misaligned with health and weather data, with
GEOS-Chem exposure data in a gridded form, health
data at the county level, and weather data at the point level
(i.e., monitor location). We converted daily grid-level wildfire-specific PM2.5 and all-source PM2.5 into daily county-level
values using area-weighted averaging24 (eAppendix Methods
3; http://links.lww.com/EDE/B109). We assumed that all persons residing in a given county have the same exposure to
wildfire-specific PM2.5 on a given day.

Hospital Admissions Data
The hospital admission data are based on billing records
2004–2009 from the Medicare Cohort Air Pollution Study
(MCAPS).25 Ethical review was not required for this study.
We included county-level data for all Medicare beneficiaries (US residents ≥65 years) enrolled in fee-for-service plan
(70.0% of all Medicare beneficiaries) in 561 counties including rural and sparsely populated counties (eFigure 1; http://
links.lww.com/EDE/B109). The Medicare data contain daily
counts of cause-specific hospital admissions by county along
with detailed information on date of admission, age category,
sex, race, and daily total numbers of Medicare enrollees, representing the population at risk, in each combination of age
category, sex, and race. The hospital admissions counts can
include repeated admissions.
We selected emergency hospital admissions for cardiovascular (CVD) and respiratory diseases as health outcomes.
A visit coded as an emergency admission might not be admitted from an emergency room/department directly but the
admission was emergency (admission type is emergency not
elective). Previous studies connected these disease categories
© 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017  Health Impact of PM2.5 from Wildfires in Western United States Counties

with total mass PM2.5.25–27 The ICD-9 codes of diagnoses are
in eAppendix Methods 1; http://links.lww.com/EDE/B109.

Air Monitoring Data and Weather Data
Daily total PM2.5 measurements from the monitoring
data, reflecting real-world PM2.5 from all sources, were used
to calibrate the total GEOS-Chem PM2.5results (“all-source”
PM2.5). The air monitoring data were acquired from EPA
AirData (http://aqsdr1.epa.gov/aqsweb/aqstmp/airdata/download_files.html#Daily). When a county had measurements
from multiple monitoring sites on a given day, we averaged
all monitor measurements to estimate the county’s total PM2.5
level on that day.
Weather information was used in statistical analysis
because temperature may confound health impact of air pollution.28 Daily county-level weather data, including temperature
and dew point temperature, were obtained from the National
Centers for Environmental Information of National Oceanic
and Atmospheric Administration.

Calibration
As in other chemical transport models, the GEOSChemPM2.5 estimates were biased low during extreme events,
reflecting the challenge in capturing smoke plumes on fine
spatial scales.23 To address this bias, we calibrated the daily,
county-level 2004–2009 modeled total PM2.5 estimates ([“allsource” PM2.5] in all 561 counties) with the county-level
total PM2.5 data from air monitors, by matching the quantile
functions of the two datasets. This approach scales the distribution of modeled PM2.5 data to more closely resemble the
distribution of the monitored data.29 This method maintains
the ordering of PM2.5 in the original (modeled) data (e.g., any
day above the 98th percentile of PM2.5 in the original modeled
data is above the 98th percentile in the calibrated data). This
calibration process results in empirical cumulative distribution functions for the simulated total PM2.5 that matches that
of the observed PM2.5. Hence, the overall proportion of PM2.5
that comes from wildfire smoke is identical in the original and
calibrated data. We calibrated the daily total modeled PM2.5
using county-average monitoring data, calculated the proportions of total modeled PM2.5 contributed by modeled wildfirespecific PM2.5on each day, and then multiplied the calibrated
total modeled PM2.5 with the proportions to obtain the calibrated wildfire-specific PM2.5. Results from the calibration
process are shown in eTable1 and eFigure2 (http://links.lww.
com/EDE/B109).

Definition of a Smoke Wave
Traditionally, the short-term effects of PM2.5 have been
investigated by estimating the association between day-to-day
variations in pollutant levels with the day-to-day variation in
health outcome rates. For example, some researchers applied
time-series analysis to associate daily ambient air pollution
exposures with daily hospital admission rates in large multicity studies.25–27 However, the frequency distribution of
© 2016 Wolters Kluwer Health, Inc. All rights reserved. 

wildfire-specific PM2.5 data differs from that of traditional
ambient levels of total PM2.5. Absent a wildfire smoke event, the
wildfire-specific PM2.5 level is near zero. Among all the days
with an estimated wildfire-specific PM2.5 levels, only 28.1%
have values >1 μg/m3 but levels can reach >200 μg/m3during
the wildfire days. To estimate health effects associated with rare
but extreme episodes of wildfire-specific PM2.5, we introduced
a new modeling approach that to our knowledge has not previously been used in the wildfire-health literature.
Specifically, we first introduce the concept of “smoke
wave.” The concept of smoke wave allows us to capture periods with high concentration, sporadic, and short-lived characteristics of wildfire PM2.5. We define a smoke wave as at
least two consecutive days with daily calibrated wildfire-specific PM2.5 > 20 μg/m3 (near the 98th percentile of all county
days across all 561 counties). This definition is based on daily
wildfire-specific PM2.5 levels above a designated threshold
and the daily levels in all days in a smoke wave must exceed
the threshold. We conducted sensitivity analyses that varied
the definition of smoke wave with respect to duration and
intensity; for example, we also defined smoke wave as at least
one day with daily calibrated wildfire-specific PM2.5 > 20 μg/
m3 (“single-day smoke waves”). Among all smoke-wave days,
we investigated whether health impact differs on smoke wave
days with different intensity and considered intensity thresholds of 23, 28, and 37 μg/m3corresponding to the 98.5th, 99th,
and 99.5th quantile of all county days across all 561 counties,
respectively. We investigated whether timing within smoke
waves (during the first 2 days, 3rd to 7th day, and 8th or later
day of a smoke wave) affects health risks, that is, whether the
health risks on an earlier day in a smoke wave differed from
those for a later day in a smoke wave. We also conducted sensitivity analysis on counties with fee-for-service enrollment
≥75% among Medicare beneficiaries.

Statistical Modeling
We conducted a matched analysis to compare the hospital admission rates (number of admissions/number of Medicare fee-for-service enrollees) on smoke-wave days (exposure)
and matched non smoke-wave days (no-exposure to high
wildfire-specific PM2.5). We chose to conduct matched analysis because the wildfire-specific PM2.5 exposure is episodic
and occurs infrequently (1.63% days were smoke wave days
among all county days). Each smoke-wave day was matched
with up to three non smoke-wave days in the same county.
Smoke-wave days in counties with many smoke-wave days
may be matched with fewer than three non smoke-wave days
when we were not able to find three suitable no-smoke-wave
days. Among the total 10,080 smoke-wave days in all counties
in 6 years, 9,184 were each matched with three non smokewave days, 697 with two non smoke-wave days, and 199 with
one non smoke-wave days. We considered non smoke-wave
days to be eligible match days if they are (1) within the window of 7 calendar days before or 7 days after the smoke-wave
www.epidem.com     79

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017

Liu et al. 

day but primarily in a different year (before or after the year
of the smoke-wave day) and (2) separated from any other
smoke-wave day by at least 2 days. Among all eligible days
meeting the matching criteria for a non smoke-wave day,
we selected the matched non smoke-wave days at random.
By matching based on a 15-day period primarily in a different year, we accounted for larger seasonal trends such as the
greater propensity for wildfires to occur during the hotter and
drier months. We assessed the difference in daily temperature,
daily dew point temperature, and non fire PM2.5 for exposure
(smoke wave) days and no-exposure (non smoke wave) days.
All statistical analyses were conducted in Rv2.15.0.
Matching reduces the effects of confounding such as
from seasonal trend.30 We controlled for seasonal factors by (1)
including a fixed effect of study year; (2) controlling for daily
temperature; and (3) using a matched approach to ensure the
same seasonality of smoke-wave days and matched non smokewave days. The matching approach guarantees that the smoke
wave and non smoke-wave days have the same distribution
across season (eTable 2; http://links.lww.com/EDE/B109), and
hence controls by design for confounding by seasonal trends.
We also conducted sensitivity analysis with the statistical model
not adjusting for modeled non fire PM2.5 levels.
We investigated the relative risk (RR) of hospital admissions on the same day as a smoke wave (lag 0). We fitted a
log-linear (Poisson) mixed effects regression model separately
for each disease (cardiovascular or respiratory diseases) for
smoke-wave days and matched non smoke-wave days across
all 561 counties (details in eAppendix Methods 4; http://links.
lww.com/EDE/B109). Similar statistical models have been
applied in previous epidemiologic studies.31

RESULTS
Wildfire PM2.5 Characteristics

The frequency distribution of PM2.5levels from wildfire
sources (calibrated) differs from that of PM2.5 from non fire
sources. Levels of wildfire-specific PM2.5 are highly skewed,
with about 72% of daily county-level calibrated wildfire-specific PM2.5 < 1 μg/m3. Wildfire-specific PM2.5 has lower mean
and median, but higher extremes, compared with PM2.5 from
non fire sources (Table 1). The time-series pattern of wildfirespecific PM2.5 is mostly zero with occasional high peaks for
short periods.

Smoke Wave Characteristics
Based on our definition of a smoke wave (at least two
consecutive days with wildfire-specific PM2.5 >20 μg/m3),
about 66% of Western United States counties (369 of 561)
experienced at least one smoke wave during the 6-year period.
Among the 369 counties with at least one smoke wave, on
average a county had 4.6 smoke-wave days/year (Table 2). We
found that the dates and locations of smoke wave days generally matched well with MODIS records of large wildfires
(eFigure 4; http://links.lww.com/EDE/B109).
80     www.epidem.com 

The number of smoke-wave days experienced by counties is spatially heterogeneous. Coastal California and central
Idaho had the highest frequency of smoke-wave days (>10
smoke-wave days/year; Figure 1). The average wildfire-specific
PM2.5 concentration during each smoke-wave day was lower
during the first 2 days of smoke waves and gradually increased
over time during a smoke wave (eFigure 3; http://links.lww.
com/EDE/B109). The median length of a smoke wave was 3
days (ranged 2 to 58).Temperatures during smoke wave days
did not differ largely based on the smoke wave day’s intensity
(eTable 3a; http://links.lww.com/EDE/B109) or smoke wave
length (eTable 3b; http://links.lww.com/EDE/B109).

Hospital Admission Summary Statistics
The study population for the 561 counties during the
study timeframe (2004–2009) includes on average about 5
million Medicare enrollees per day. This population had a total
of 832,244 cardiovascular admissions and 245,926 respiratory
admissions during the study time frame. Within the study time
frame, 369 counties had at least one smoke wave. For these
counties, there were 648,789 cardiovascular admissions and
191,095 respiratory admissions. Counties that experienced a
smoke wave had, on average, lower rates of hospital admissions than counties with no smoke wave (Table 3). There are
3,844,414 people exposed to at least one smoke wave, and
1,114,513 with no-exposure to smoke waves.

Association Between Wildfire PM2.5 and
Hospital Admissions
Overall, smoke waves were not associated with increased
rates of cardiovascular hospital admissions. The overall association with cardiovascular admissions on a smoke-wave day compared with a non smoke-wave day was −0.74% (95% CI: −3.1%,
1.65%; RR = 0.99). The overall association with respiratory
hospital admissions on a smoke-wave day compared with a non
smoke-wave day was 2.3% (95% CI: −2.2%, 7.0%; RR = 1.0).
Smoke-wave days with different intensity (level of
wildfire PM2.5) and the various days within the smoke waves
exhibited indication of trends of different health effects. Central estimates for respiratory admissions showed an increasing trend as smoke-wave day intensity increases (Figure 2B).
Smoke-wave days with intensity >37 μg/m3 (99.5th quantile)
were associated with a 7.2% increase in respiratory admissions
by 7.2% (95% CI: 0.25%, 15%) compared with non smokewave days. Therefore, more intense smoke-wave days are estimated to have higher health impacts on respiratory diseases for
the study population. This association is robust to no inclusion
of a variable for non fire PM2.5 levels in the model (results not
shown). Results on single-day smoke waves and counties with
fee-for-service enrollment >75% are summarized in eAppendix
Results 1 and 2; http://links.lww.com/EDE/B109.
Central estimates for CVD admissions tend to be highest
during the first 2 days of a smoke wave, and decreasing over
the later days within a smoke wave (Figure 3A). Respiratory
admissions exhibit an opposite trend, with higher estimate in
© 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017  Health Impact of PM2.5 from Wildfires in Western United States Counties

TABLE 1.  Summary Statistics for Daily GEOS-Chem PM2.5 Concentrations (Calibrated) from Wildfire Sources and Non Fire
Sources in 561 Western United States Counties (μg/m3) During the Wildfire Season (May 1–Oct 31), 2004–2009

PM2.5 from wildfires
PM2.5 from non fire
sources

Minimum

25th Percentile

Median

Mean

75th Percentile

Maximum

0
0

0.09
4.4

0.3
6.2

2.0
7.0

1.2
8.7

242
45.1

TABLE 2.  Summary Statistics for SW (Defined as at Least Two Consecutive Days with Wildfire-specific PM2.5 >20 μg/m3) for the
369 Western United States Counties that Experienced SW During 2004–2009
SW Characteristics
No. SW days/yra
No. SW events/yra
SW intensity (μg/m3)b
SW length (days)b

Average (SD)

Median

Minimum

Maximum

4.6 (4.9)
1.0 (0.8)
29.3 (6.4)
4.4 (4.7)

2.5
0.83
28.1
3

0.33
0.17
20.1
2

26.5
3.8
70.0
58

a

Statistics based on the 369 county-average values.
Statistics based on all SW-level values across all SWs in the 369 counties.
SW indicates smoke waves.

b

FIGURE 1.  Average number of smoke wave days/year for 561 Western United States counties during 2004–2009. Hashed counties have population >75,000 in the 2010 Census.

later days of the smoke wave (Figure 3B). For each types of
admission, effect estimates based on timing within a smoke
wave were imprecise.

DISCUSSION
Our systematic assessment indicates an association
between respiratory admissions and intense smoke-wave
© 2016 Wolters Kluwer Health, Inc. All rights reserved. 

days, with daily wildfire-specific PM2.5 levels >37 μg/m3.
Single-day smoke waves have a potentially more certain positive association with respiratory admission rates, possibly due
to larger sample sizes and the acute response of respiratory
diseases.
To our knowledge, this is the first study to use wildfire-specific data to analyze the health impact of wildfire-specific PM2.5
www.epidem.com     81

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017

Liu et al. 

TABLE 3.  County-level Hospital Admission per 100,000 Medicare Enrollees per Day (2004–2009)

561 counties
    Cardiovascular disease
    Respiratory
369 counties with smoke waves
    Cardiovascular disease
    Respiratory
192 counties with no smoke waves
    Cardiovascular disease
    Respiratory

Minimum

25th percentile

Median

Mean

75th percentile

Maximum

1.59
0

8.18
1.81

11.5
3.33

12.2
3.59

15.0
4.87

43.7
17.1

1.59
0

7.87
1.63

10.7
3.07

11.2
3.25

13.7
4.52

39.7
11.7

4.88
0

9.03
2.43

13.5
3.91

14.1
4.25

17.4
5.74

43.7
17.8

FIGURE 2.  Associations between hospital admissions and exposure to smoke wave days (compared with non-smoke-wave days)
for (A) cardiovascular disease and (B) respiratory disease, by different intensity (level of wildfire-specific PM2.5) definitions of a
smoke wave. SW indicates smoke wave.

over multiple years at a large geographical scale. Key contributions of this study include (1) estimation of exposure to PM2.5 specifically from wildfires; (2) ability to estimate exposure to wildfire
PM2.5 every county with and without air monitors, therefore
expanding the study populations to include persons that live far
from PM2.5 monitoring stations; and (3) application of statistical
models that estimate percent increases in hospital admission by
matching smoke-wave days to non-smoke-wave days.
Although previous literature on the association between
wildfire smoke and health is limited, several studies have
made important contributions. The majority of such studies
82     www.epidem.com 

used air monitor measurements, which cannot identify pollution specifically from wildfires with current technology, and
studied a single wildfire episode and one or a small number of
communities.3 A few studies compared air pollution exposure
(from all sources) during wildfires to the periods or locations
with no fire.11,32,33 Our study results for respiratory diseases
are consistent with those found in most of the previous literatures,34,35 in which wildfire smoke was found to be associated
with respiratory diseases. Association between wildfire smoke
and cardiovascular morbidities was found in five US studies
that each examined a single local wildfire episode,3 but our
© 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017  Health Impact of PM2.5 from Wildfires in Western United States Counties

FIGURE 3. Associations between
hospital admissions and exposure
to SW days (compared with non
SW days) for (A) cardiovascular
disease and (B) respiratory disease,
by timing of the days within a SW.
SW indicates smoke wave.

multistate, multiyear study did not provide evidence for such
association.
Previous studies have demonstrated that the chemical
composition of PM2.5, which is related to source, can result
in different effect estimates for human health.9,36,37 Thus, estimates from wildfire PM2.5 may differ from those from PM2.5
from other sources, such as transportation or industry. Earlier studies examined the association between risk of hospital admission and levels of PM2.5 from all sources (i.e., PM2.5
total mass; e.g., change of risk of hospital admission for Medicare enrollees per 10 μg/m3 increase in PM2.5 in the Western
United States25,26,38). As we compared the health risk among
smoke-wave days with that of non smoke-wave days, rather
than by a specific increment of PM2.5, direct comparisons
of results is challenging. Furthermore, these studies focused
on urban counties with high populations, whereas our study
included rural populations in the analysis as well. Still, a general comparison can give some indication of whether PM2.5
from wildfire smoke is more or less harmful than PM2.5 total
mass.
For Medicare cardiovascular admissions, one study
estimated an increased risk of 0.53% (95% posterior interval: 0.00%, 1.05%) per 10 μg/m3 PM2.5 total mass (from all
sources) for 25 urban counties in the Southwest US, and 0.74%
(−1.74%, 3.29%) for nine urban counties in the Northwest.26
Our results did not indicate an association between wildfire
PM2.5 and risk of cardiovascular admissions.
For respiratory hospital admissions, we estimated an
increase of 7.2% (0.25%, 15%) comparing smoke-wave days
with wildfire-specific PM2.5 > 37 μg/m3 to non smoke-wave
© 2016 Wolters Kluwer Health, Inc. All rights reserved. 

days with wildfire-specific PM2.5 ≤ 20 μg/m3, which corresponds to an average difference of 29.6 μg/m3in those two
groups of days. The earlier study identified associations
between PM2.5 total mass and respiratory admissions for the
Medicare population in the Southwest at lag 2 days at 0.94%
(0.22%, 1.67%) per 10 μg/m3,26 which corresponds to an
increased risk of 2.8% (0.64, 5.0%) per 29.6 μg/m3. Therefore, our estimates of respiratory admissions risks indicate
that wildfire-specific PM2.5 from intense smoke waves are
associated with more harm than PM2.5 from other sources for
the elderly in the Western United States. Further research is
needed to investigate the relative toxicity of PM2.5 from wildfire smoke with that of other sources.
Our approaches for assessing pollutant exposure and
estimating health risks address key challenges in studying
the health impact of wildfire-specific pollutant. The GEOSChem model provided a new approach to distinguish wildfirespecific PM2.5 from PM2.5 from other sources. The fire scheme
in the simulation can explain up to 60% of the observed variance of area burned in the Western United States, and is ecosystem dependent.17 This method also improves the spatial
and temporal resolution of exposure estimates for air pollution. Unlike air monitoring data that generally measure PM2.5
concentrations every 3–6 days in urban areas, GEOS-Chem
estimates concentrations for every day and covers the entire
study area. Our smoke-wave methods provide an approach
suitable for the study of highly-skewed air pollution data and
enable identification and investigation of pollution episodes
with high source-specific pollutant concentrations. Matched
analysis can reduce the confounding effect of seasonality
www.epidem.com     83

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017

Liu et al. 

and county-specific effects. These methods can be applied
to future studies investigating other pollution events and
populations.
Limitations of our study include potential spatial misalignment between the exposure estimates (gridded estimates)
and health data (county). Our study population was restricted
to Medicare fee-for-service enrollees, a sample of elderly persons. Our smoke wave approach does not fully capture the
dose–response relationship, cause-specific health outcomes,
etc., that could be investigated in future studies. The GFED
emissions applied to GEOS-Chem contribute uncertainty
to the modeled estimates of wildfires-specific PM2.5. The
GFED3 data may underestimate fire contributions to background PM2.5 because of the omission of small fires39 and the
biases in the modeled fuel consumption. GFED3 relies on satellite observations of active fire counts and area burned, and
may have difficulty discerning such phenomena, especially
on cloudy days.40 Another limitation arises as EPA monitors generally measure PM2.5 values every 3–6 days and are
located in populated areas. Given a large number of days with
monitoring measurements for calibration, we assumed that the
systematic sampling of EPA monitors generate measurements
with mean and SD representing the full-time series of realworld PM2.5. While it would be ideal to have the full continuous measure, we believe that calibration using this discrete
sample of the continuous measure is the best possible alternative in using the available data. While our exposure estimates
are advances over methods that do not isolate the air pollution from wildfires specifically, additional study could address
these limitations. We choose not to a priori identify lags in
this study as little is known about how wildfire-specific PM2.5
affects human health. Most of the wildfire-health literature to
date has investigated effects of lag 0 or short lags (<5 days).3
Future studies can explore the lagged effect of wildfire-specific air pollution.
Our findings indicate that wildfires are associated with
increased risk of admissions for respiratory diseases for the
elderly population during severe wildfire episodes. As climate
change is anticipated to increase the frequency and intensity of
wildfires,1 the health burden from wildfire-specific pollutants
may increase in the future. With improvement of atmospheric
modeling, future studies can estimate daily wildfire-specific
PM2.5 at a finer spatial resolution. Future studies can also
investigate vulnerability to wildfire smoke, health impact of
different species of wildfire-specific PM2.5, the economic consequence of the health burden from wildfire smoke, combined
effect of wildfire smoke and other air pollutants, and estimated
health burden in the future under climate change.
REFERENCES
  1.  Spracklen DV, Mickley LJ, Logan JA, et al. Impacts of climate change
from 2000 to 2050 on wildfire activity and carbonaceous aerosol
concentrations in the western United States. J Geophys Res Atmos.
2009;114. Issue D20301.

84     www.epidem.com 

 2. Kochi I, Donovan GH, Champ PA, Loomis JB. The economic cost of
adverse health effects from wildfire-smoke exposure: a review. Int J
Wildland Fire. 2010;19:803–817.
  3. Liu JC, Pereira G, Uhl SA, Bravo MA, Bell ML. A systematic review of
the physical health impacts from non-occupational exposure to wildfire
smoke. Environ Res. 2015;136:120–132.
  4.  Dennison PE, Brewer SC, Arnold JD, Moritz MA. Large wildfire trends
in the western United States, 1984-2011. Geophys Res Lett. 2014;41:
2928–2933.
  5. Christopher JP et al.; Interagency Working Group on Climate Change
and Health. In: A human health perspective on climate change: a
report outlining the research needs on the human health effects of
climate change. J Current Issues Global. 2013: 621.
  6.  Kim KH, Kabir E, Kabir S. A review on the human health impact of airborne particulate matter. Environ Int. 2015;74:136–143.
 7. Janhall S, Andreae MO, Poschl U. Biomass burning aerosol emissions
from vegetation fires: particle number and mass emission factors and size
distributions. Atmos Chem Phys. 2010;10:1427–1439.
  8.  Bell ML, Ebisu K. Environmental inequality in exposures to airborne particulate matter components in the United States. Environ Health Perspect.
2012;120:1699–1704.
 9.  Peng RD, Bell ML, Geyh AS, et al. Emergency admissions for cardiovascular and respiratory diseases and the chemical composition of fine
particle air pollution. Environ Health Perspect. 2009;117:957–963.
 10. Delfino RJ, Brummel S, Wu J, et al. The relationship of respiratory and
cardiovascular hospital admissions to the southern California wildfires of
2003. Occup Environ Med. 2009;66:189–197.
 11.  Moore D, Copes R, Fisk R, Joy R, Chan K, Brauer M. Population health
effects of air quality changes due to forest fires in British Columbia in
2003: estimates from physician-visit billing data. Can J Public Health.
2006;97:105–108.
 12.  Morgan G, Sheppeard V, Khalaj B, et al. Effects of bushfire smoke on daily mortality and hospital admissions in Sydney, Australia. Epidemiology.
2010;21:47–55.
 13. Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW. Warming and
earlier spring increase western U.S. forest wildfire activity. Science.
2006;313:940–943.
 14. Bey I, Jacob DJ, Yantosca RM, et al. Global modeling of tropospheric
chemistry with assimilated meteorology: model description and evaluation. J Geophys Res Atmos. 2001;106:23073–23095.
 15. Park RJ, Jacob DJ, Logan JA. Fire and biofuel contributions to annual
mean aerosol mass concentrations in the United States. Atmos Environ.
2007;41:7389–7400.
 16. Spracklen DV, Logan JA, Mickley LJ, et al. Wildfires drive interannual variability of organic carbon aerosol in the western US in summer.
Geophys Res Lett. 2007;34.
 17. Yue X, Mickley LJ, Logan JA, Kaplan JO. Ensemble projections of
wildfire activity and carbonaceous aerosol concentrations over the
western United States in the mid-21st century. Atmos Environ (1994).
2013;77:767–780.
 18. Mu M, Randerson JT, van der Werf GR, et al. Daily and 3-hourly
variability in global fire emissions and consequences for atmospheric model predictions of carbon monoxide. J Geophy Res Atmos.
2011;116.
 19. van der Werf GR, Randerson JT, Giglio L, et al. Global fire emissions
and the contribution of deforestation, savanna, forest, agricultural,
and peat fires (1997-2009). Atmos Chem Phys. 2010;10:11707–11735.
 20. Chow JC, Watson JG. Review of PM2.5 and PM10 apportionment for
fossil fuel combustion and other sources by the chemical mass balance
receptor model. Energy Fuels. 2002;16:222–260.
 21. Park RJ, Jacob DJ, Field BD, Yantosca RM, Chin M. Natural and transboundary pollution influences on sulfate-nitrate-ammonium aerosols
in the United States: Implications for policy. J Geophys Res Atmos.
2004;109.
 22. Sapkota A, Symons JM, Kleissl J, et al. Impact of the 2002 Canadian
forest fires on particulate matter air quality in Baltimore city. Environ Sci
Technol. 2005;39:24–32.
 23. Zhang L, Jacob DJ, Yue X, Downey NV, Wood DA, Blewitt D. Sources
contributing to background surface ozone in the US Intermountain West.
Atmos Chem Phys. 2014;14:5295–5309.

© 2016 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.

 Epidemiology  •  Volume 28, Number 1, January 2017  Health Impact of PM2.5 from Wildfires in Western United States Counties

 24.  Bravo MA, Fuentes M, Zhang Y, Burr MJ, Bell ML. Comparison of exposure estimation methods for air pollutants: ambient monitoring data and
regional air quality simulation. Environ Res. 2012;116:1–10.
 25. Dominici F, Peng RD, Bell ML, et al. Fine particulate air pollution and
hospital admission for cardiovascular and respiratory diseases. JAMA.
2006;295:1127–1134.
 26. Bell ML, Ebisu K, Peng RD, et al. Seasonal and regional short-term effects of fine particles on hospital admissions in 202 US counties, 19992005. Am J Epidemiol. 2008;168:1301–1310.
 27.  Peng RD, Chang HH, Bell ML, et al. Coarse particulate matter air pollution and hospital admissions for cardiovascular and respiratory diseases
among Medicare patients. JAMA. 2008;299:2172–2179.
 28. Samet J, Zeger S, Kelsall J, Xu J, Kalkstein L. Does weather confound
or modify the association of particulate air pollution with mortality? An
analysis of the Philadelphia data, 1973-1980. Environ Res. 1998;77:9–19.
 29. Maraun D. Bias correction, quantile mapping, and downscaling: revisiting the inflation issue. J Climate. 2013;26:2137–2143.
 30. Rubin DB. The use of matched sampling and regression adjustment to remove
bias in observational studies. Matched Sampling Causal Effects. 2006:81–98.
 31. Bobb JF, Obermeyer Z, Wang Y, Dominici F. Cause-specific risk of
hospital admission related to extreme heat in older adults. JAMA.
2014;312:2659–2667.
 32.  Duclos P, Sanderson LM, Lipsett M. The 1987 forest fire disaster in California:
assessment of emergency room visits. Arch Environ Health. 1990;45:53–58.

© 2016 Wolters Kluwer Health, Inc. All rights reserved. 

 33.  Frankenberg E, McKee D, Thomas D. Health consequences of forest fires
in Indonesia. Demography. 2005;42:109–129.
 34. Lee TS, Falter K, Meyer P, Mott J, Gwynn C. Risk factors associated with
clinic visits during the 1999 forest fires near the Hoopa Valley Indian
Reservation, California, USA. Int J Environ Health Res. 2009;19:315–327.
 35. Rappold AG, Stone SL, Cascio WE, et al. Peat bog wildfire smoke exposure in rural North Carolina is associated with cardiopulmonary emergency department visits assessed through syndromic surveillance. Environ
Health Perspect. 2011;119:1415–1420.
 36.  Bell ML, Ebisu K, Peng RD, Samet JM, Dominici F. Hospital admissions
and chemical composition of fine particle air pollution. Am J Respir Crit
Care Med. 2009;179:1115–1120.
 37.  Zanobetti A, Franklin M, Koutrakis P, Schwartz J. Fine particulate air pollution and its components in association with cause-specific emergency
admissions. Environ Health. 2009;8:58.
 38.  Bell ML, Son JY, Peng RD, Wang Y, Dominici F. Ambient PM2.5 and risk
of hospital admissions: do risks differ for men and women? Epidemiology.
2015;26:575–579.
 39. Randerson JT, Chen Y, van der Werf GR, Rogers BM, Morton DC. Global
burned area and biomass burning emissions from small fires. J Geophys
Res Biogeosci. 2012;117.
 40. Giglio L, Randerson JT, van der Werf GR. Analysis of daily, monthly,
and annual burned area using the fourth-generation global fire emissions
database (GFED4). J Geophys Res Biogeosci. 2013;118:317–328.

www.epidem.com     85

Copyright © 2016 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.