Science of the Total Environment 408 (2010) 5784–5793
Contents lists available at ScienceDirect
Science of the Total Environment
j ourna l homepage: www.e lsev ie r.com/ locate /sc i totenv
Burden of disease attributed to anthropogenic air pollution in the United ArabEmirates: Estimates based on observed air quality data
Ying Li, Jacqueline MacDonald Gibson ⁎, Prahlad Jat, Gavino Puggioni, Mejs Hasan, J. Jason West,William Vizuete, Kenneth Sexton, Marc SerreDepartment of Environmental Sciences and Engineering, Gillings School of Global Public Health, University of North Carolina, Chapel Hill, NC, USA
⁎ Corresponding author. Department of EnvironmenGillings School of Global Public Health, 148A Rosenau HaChapel Hill, NC 27599-7431, USA. Tel.: +1 919 966 789
Article history:Received 9 April 2010Received in revised form 6 August 2010Accepted 9 August 2010Available online 9 September 2010
Keywords:Environmental burden of diseaseMortalityAmbient air pollutionExposure assessmentRisk assessmentRelative riskUnited Arab Emirates
This study quantifies the national burden of disease attributed to particulate matter (PM) and ozone (O3) inambient air in the United Arab Emirates (UAE), a rapidly growing nation in which economic developmentand climatic conditions pose important challenges for air quality management. Estimates of populationexposure to these air pollutants are based on observed air quality data from fixed-site monitoring stations.We divide the UAE into small grid cells and use spatial–statistical methods to estimate the ambient pollutantconcentrations in each cell based on the observed data. Premature deaths attributed to PM and O3 arecomputed for each grid cell and then aggregated across grid cells and over a year to estimate the totalnumber of excess deaths attributable to ambient air pollution. Our best estimate is that approximately 545(95% CI: 132–1224) excess deaths in the UAE in the year 2007 are attributable to PM in ambient air. Theseexcess deaths represent approximately 7% (95% CI: 2–17%) of the total deaths that year. We attributeapproximately 62 premature deaths (95% CI: 17–127) to ground-level O3 for the year 2007. Uncertainty inthe natural background level of PM, due to the frequent dust storms occurring in the region, has significantimpacts on the attributed mortality estimates. Despite the uncertainties associated with the integratedassessment framework, we conclude that anthropogenic ambient air pollution, in particular PM, causes aconsiderable public health impact in the UAE in terms of premature deaths. We discuss importantuncertainties and scientific hypotheses to be investigated in future work that might help reduce theuncertainties in the burden of disease estimates.
tal Sciences and Engineering,ll, University of North Carolina,2; fax: +1 919 966 7911.. Gibson).
Ambient air pollution is an increasing problem in the United ArabEmirates (UAE). Fueled by abundant oil resources, the UAE – afederation of seven emirates – is one of the most rapidly growingnations in the world. Poor air quality can be observed throughdegraded visibility, ambient measurements of air pollutants, and theattention this issue receives in the media. The UAE is home to largeindustries that emit air pollutants and to two international urbancenters (Abu Dhabi and Dubai) with large vehicular sources ofemissions and problems of traffic congestion. Other factors thatcontribute to degraded air quality include the frequent and severedust storms in the Arabian Gulf region and the transport of pollutantsfrom other continents. The rapid growth in the UAE and thesurrounding region – as reflected in the increases in population,energy use, and vehicle traffic – poses important challenges for airquality management.
A nation's burden of disease attributed to ambient air pollutionprovides important information for developing air quality manage-ment strategies. This study estimates the number of premature deathsattributable to ambient air pollution in the UAE in 2007 based on thebest available observed air quality data. The disease burden estimatesprovide an important indication for decision makers in the country ofthe improvements in health that could be achieved by targetedactions to control ambient air pollution (Ostro, 2004).
Current scientific evidence, derived largely from studies inwesternindustrial countries but increasingly including the developing world,demonstrates that exposure to ambient air pollution causes a widespectrum of adverse health outcomes, from acute respiratorysymptoms to premature death. The air pollutant considered to havethe greatest impact on health is particulate matter (PM), which isfrequently measured as PM10 and PM2.5 (particles with diameter lessthan 10 and 2.5 μm, respectively). Both of these indicators have beenconsistently associated with premature mortality and cardiopulmo-nary diseases. A second key pollutant is ground-level ozone (O3),which has been linked to adverse health effects similar to thoseinduced by PM. Although associations also have been reported forother common air pollutants (such as sulfur dioxide, carbonmonoxide, nitrogen dioxide, and air toxics), given the strength of
5785Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
the existing epidemiological evidence we focus on PM and O3 as thepollutants most likely to be the leading contributors to the burden ofdisease due to ambient air pollution in the UAE.
We use an integrated procedure that combines exposure assess-ment and health impact assessment to quantify the burden of diseaseattributed to ambient air pollution. Our methods are based on thoseused previously by the World Health Organization (WHO) and othersto assess the global burden of disease (Ostro, 2004; Cohen et al., 2004;Anenberg et al., 2009). They also are similar to the U.S. EnvironmentalProtection Agency's health impact assessment approach to quantify-ing the benefits of proposed actions to improve air quality (USEPA,1999), as well as methods recently recommended by the U.S. NationalAcademy of Sciences (National Research Council, 2008).
The WHO recently released a country profile of the environmentalburden of disease for the UAE (WHO, 2009). The WHO's profileattributed 200 deaths per year in the UAE to PM10 but did not reportestimates of the burden of disease due to other air pollutants. Ourstudy improves on WHO's estimate in several important ways. First,while the WHO's assessment relied on a single estimate of annualaverage pollutant concentration across the entire UAE, we create gridcells with fine resolution (approximately 0.89 km2) across the UAEand use statistical methods to estimate the ambient pollutantconcentrations in each grid cell based on observed air quality data.We also use daily average concentrations to reflect the temporalvariation of pollution levels, which is important for understandingshort-term health effects. Therefore, our method captures both thespatial and temporal variations of pollutant concentrations, inaddition to population and baseline health statistics, to the extentthat available data support. Second, while the WHO's estimateincludes only excess mortality caused by PM10, we also estimate themortality caused by PM2.5 and ground-level O3. Third, we conduct asystematic analysis to understand how the major uncertainties in ourmethods might affect the burden of disease estimates. In particular,we assess the effects of different assumptions about the naturalbackground levels of PM — an important uncertainty, given that theArabian Gulf region experiences frequent dust storms. We investigatehow the uncertainty associated with the natural background level ofPM may change estimates of the burden of disease attributable toanthropogenic pollutants. We also estimate uncertainties in ambientair quality data and in the predicted relationships between pollutantlevels and premature mortality.
The remainder of the paper is organized as follows. Section 2reviews the current epidemiologic literature on the health effects ofPM and O3 and describes the methods used to quantify the burden ofdisease in the UAE. Section 3 presents our estimates of the number ofpremature deaths attributable to ambient air pollution, includingmaps to visualize spatial variation in health risks due to ambient airpollution. Section 3 also analyzes the uncertainty in our estimates andthe causes of that uncertainty. Section 4 summarizes the majorconclusions of this study. We identify important uncertainties andscientific hypotheses to be investigated in future work.
2. Methods and epidemiological basis
2.1. Health effects of particulate matter and O3
2.1.1. Particulate matterTime-series studies of the short-term effects of air pollutants
examine the relationship between daily changes in air pollution (24-h average concentrations in most cases) and the daily occurrence ofmortality or morbidity in an area. The key advantage of the time-series method is that it potentially reduces the confounding effects ofmany factors, which otherwise might be difficult for researchers tocontrol. Several important confounding factors – such as smokinghabits, health-care status, activity patterns, characteristics of theworking and living environment, and socioeconomic status – do not
vary considerably over time. Time-series studies have consistentlyreported significant associations between daily mortality and dailyexposures to both PM10 and PM2.5 and thus provide compellingevidence that PM increases mortality rates (Ostro, 2004). Recentmulti-city studies and meta-analyses in the United States havereported that all-cause mortality increases by 0.2–0.8% per 10 μg/m3
increase in the daily average PM10 concentration (Dominici et al.,2005; Levy et al., 2000; Samet et al., 2000; Schwartz et al., 2002; Zekaet al., 2005). Ameta-analysis of 33 European studies suggested ameanincrease in the risk of premature mortality of 0.6% per 10 μg/m3 PM10
(Anderson et al., 2004). Studies in Brisbane and Sydney, Australia,reported increases in mortality of 1.6% and 0.05%, respectively, per10 μg/m3 PM10 (Ostro, 2004). Consistent associations also have beenobserved in cities in developing countries, but the effects tend to beslightly greater than those reported in the United States and Europe.For example, the following all-cause mortality effect estimates havebeen reported for total populations and a 10-μg/m3 change in PM10(with 95% confidence intervals): 1.7% (1.1–2.3) in Bangkok, Thailand;1.83% (0.9–2.7) in Mexico City, Mexico; and 1.1% (0.9–1.4) inSantiago, Chile (Ostro, 2004).
Cohort studies follow a group of initially healthy people for along period (e.g., 10–20 years) to observe whether they developdiseases or die. These studies are valuable in investigating possiblelong-term chronic effects of exposure to air pollution. In the UnitedStates, two large-scale cohort studies – the Harvard Six Cities Studyand the American Cancer Society (ACS) Study – have beenconducted over the past two decades. Both studies observedincreased mortality associated with an increase in time-averagedPM2.5 levels but not with increases in time-averaged PM10,suggesting that long-term adverse health effects are influenced bythe fine portion of PM (Dockery et al., 1993; Krewski et al., 2000;Laden et al., 2006; Pope et al., 1995, 2002).
Compared to the acute effects observed by time-series studies,chronic effects reported by cohort studies are generally 5–10 timeslarger (Künzli et al., 2001). In the United States, the ACS study hasbeen commonly used as a basis for assessing health impacts ofambient PM, because it included the largest population sample size. Inan extended analysis of the ACS study, Pope et al. (2002) reportedresults using 16 years (1982–1998) of follow-up data for approxi-mately 500,000 adults (ageN30) throughout the United States. Theyfound that each 10-μg/m3 elevation in PM2.5 was associated withapproximately 4%, 6%, and 8% increases in all-cause, cardiopulmonary,and lung cancer mortality, respectively. A new, extended follow-upanalysis of the ACS study, conducted to clarify outstanding scientificissues arising from earlier analyses, reported results that areconsistent with those from other studies, further supporting thehypothesis that long-term exposure to ambient PM2.5 increasesmortality in the general population (Krewski et al., 2009).
The Arabian Gulf region frequently experiences episodes of highPM concentrations that are dominated by windblown desert dust. Inthese episodes, coarse-mode particles (those with diameters ofbetween 2.5 and 10 μm, denoted as PM10–2.5) are likely to comprisea greater proportion of the total PM10 than is typical in most otherregions. For instance, a study in the Coachella Valley, an arid, desertregion located in southern California in the United States, reported aPM2.5:PM10 ratio of 0.35 (Ostro et al., 2000), compared to ratios of0.50–0.65 regularly observed in urban areas (Ostro, 2004). Fineparticles (PM2.5) have generally been considered to be more toxic andto pose a greater health risk for people than coarse particles (PM10).However, there is evidence that the health effects of PM2.5 may varyby source, with inorganic crustal and soil-based components of PM2.5
posing little if any health risk (Thurston et al., 2005; Schlesinger,2007). Given this, in areas affected by desert dust, a unit increase inPM concentration (e.g., 1 μg/m3 or 10 μg/m3) may or may not causehealth effects similar to those observed in less dusty regions. A fewtime-series studies of PM acute health effects have been conducted in
5786 Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
arid desert areas in the United States that have windblown dustepisodes similar to those in the Gulf region. The following increases indaily mortality associatedwith a 10-μg/m3 increase in PM10 (with 95%confidence intervals) were reported:
• in Coachella Valley, California, 0.09% (0.01–0.17%) for all-causemortality (Ostro et al., 1999) and 1.2% (0.4–2%) for cardiovascularmortality (Ostro et al., 2000);
• in Salt Lake City, Utah, 0.8% (0.3–1.3%) for all-cause mortality (Popeet al., 1999); and
• in Spokane, Washington, no significant health risk increase(Slaughter et al., 2005).
For this study, we assume that overall PM2.5 in the UAE should betreated as having the typical risk associated with PM2.5 in highlydeveloped and industrialized areas, although we subtract naturallyoccurring PM from observed ambient air data. While this may beconsidered an upper bound assumption, it indicates the importance ofquantifying the nature of PM in the UAE through chemical speciationto reduce uncertainty in future risk estimates.
2.1.2. O3
Scientific evidence increasingly supports that ground-level O3,even at low levels, can damage people's health, causing (among othereffects) increased rates of hospital admissions, exacerbation ofrespiratory illnesses and premature mortality. Meta-analyses of city-specific studies have consistently reported associations between O3
and daily mortality (Bell et al., 2005; Ito et al., 2005; Levy et al., 2005).A recent, large time-series study examined the link between ambientO3 and short-termmortality for 95 large U.S. urban communities from1987 to 2000 and reported that a 10-ppb increase in the daily averageO3 concentration was associated with a 0.52% increase in daily non-accidental mortality (Bell et al., 2004). A meta-analysis of studies inEurope found a 0.3% increase in all-cause mortality per 10 μg/m3
increase in 8-h O3 (Anderson et al., 2004). While cohort studies hadnot clearly identified a relationship between O3 and mortality in thepast, a recent follow-up of the ACS cohort reported that long-term O3
exposure was significantly associated with an increase in deaths fromrespiratory causes, with a relative risk of 1.040 per 10 ppb increase inO3 concentration (Jerrett et al., 2009). This evidence collectivelysupports an association between ground-level O3 and mortality, withrespect to both short-term and long-term exposure, over a wide rangeof concentrations.
2.1.3. Summary of the evidence on health effects of air qualityIn summary, the negative impacts of both PM and O3 on public
health have been well documented. Epidemiologic evidence isparticularly strong in indicating that high PM levels cause excessmortality. The key uncertainty issues include the magnitude andvariability of relative risk estimates, possible thresholds or disconti-nuities in the concentration–response functions, the extent to whichfindings in one location can be generalized to other locations, and thelack of a clear understanding of underlying biologicalmechanisms. ForPM, whether the mass of PM or particular chemical componentswithin the PM is responsible for health effects is unclear. Uncertaintyis also involved in years of life lost associated with prematuremortality attributed to PM or O3. In addition, epidemiologic studies donot clearly indicate whether O3 itself causes health effects directly, orwhether O3 is an indicator of other constituents produced byatmospheric photochemistry that may influence health (Bell et al.,2006).
Although local epidemiological studies may better reflect theinfluence of pollutant characteristics and baseline health status on theassociations between air pollution and health, studies conducted atvarious locations across the United States and in Europe that involve awide range of underlying conditions and studies conducted in citiesoutside the developed world report generally consistent effect
estimates. Given this, the WHO has suggested that it is reasonableto extrapolate existing relative risk estimates to areas where studieshave not been undertaken (Ostro, 2004), which is the case for theUAE. Based on the WHO's suggestion, we select representative multi-city studies or meta-analyses conducted in the United States as thebasis for the health impact assessment for ambient air pollution in theUAE (see Section 2.2).
2.2. Methods for quantifying the disease burden
2.2.1. OverviewOur quantitative assessment of the health impacts of ambient air
pollution in the UAE is based on the following components:
• air pollutant concentration assessment based on measurements atmonitoring stations;
• determination of the size of population groups exposed to specificair pollutants, the health effects of interest, and the baselineincidences of those health effects; and
• concentration–response functions abstracted from epidemiologicliterature.
These factors are combined in an integrated procedure to estimatethe burden of disease attributable to ambient air pollution.
The health impact function to quantify the excess mortalityattributed to ambient air pollution is
Δy = y0 × 1−e−β⋅Δx� �
= I0 × P × 1−e−β⋅Δx� �
ð1Þ
where
Δy attributable mortality (deaths per year);y0 baseline incidence (current number of cause-specific deaths
per year, which is equal to the baseline incidence rate (I0)times the number of people exposed to the pollutant (P));
β concentration–response coefficient from epidemiologicalstudies (increase in cause-specific mortality per unitincrease in PM or O3); and
Δx change in concentration of the pollutant of interest (PM10,PM2.5, or O3) (μg/m3 for PM and ppb for O3), taken here asthe total concentration minus the estimated backgroundconcentration from natural sources.
Eq. (1) can be derived from the equation used in epidemiologicliterature to calculate the attributable fraction (AF), defined as thefraction of the disease burden attributable to a risk factor. TheWHO hasproposed calculating the AF due to exposure to ambient air pollutionusing the following equation (Cohen et al., 2004; Ostro, 2004):
AF =RR−1RR
ð2Þ
where RR is the relative risk due to exposure to the pollutant of
concern. In epidemiologic studies on the health effects of ambient airpollution, RR is commonly estimated from the following equation(Aunan, 1996):
RR = eβΔx: ð3Þ
Hence, the health impact function in Eq. (1) can be written as:
Δy = y0 × AF = y0 ×RR−1RR
= y0 ×eβΔx−1eβΔx
= y0 × 1−e−βΔx� �
:ð4Þ
The following sections describe the health endpoints associated withPM or O3 considered in this study and how we characterized the keyvariables in Eq. (4) (β, Δx, and y0) to estimate the burden of diseasedue to ambient air pollution in the UAE.
5787Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
2.2.2. Health endpoints and concentration–response coefficientsTable 1 summarizes the health outcomes considered in this study,
their concentration–response coefficients (β), and sources in theliterature used to estimate β.
The health endpoints chosen for PM10 are based on the WHO'srecommendations for national-level burden of disease assessments(Ostro, 2004): short-term health effects including all-cause mortalityin the general population and respiratory mortality in childrenyounger than 5 years old, using daily average concentrations as theexposure indicator. The health effects of PM2.5 are based on Pope et al.(2002). These effects are all-cause, cardiopulmonary, and lung cancermortality for adults aged over 30, all due to long-term exposure toPM2.5, using annual average concentration as the exposure indicator.As explained under “Results and discussion,” in aggregating the totalmortality attributed to PM, only all-causemortality attributed to long-term exposure to PM2.5 in adults over age 30 and premature mortalityfrom respiratory disease attributed to short-term exposure to PM10 inchildren younger than 5 were summed, in order to avoid double-counting, whereas the remaining health endpoints were assessed toprovide additional information for decision makers.
For O3, we estimated all-cause mortality due to short-termexposure (using the daily average concentration as the exposureindicator) and respiratorymortality due to long-term exposure (usingthe annual average of the daily maximum concentration as theexposure indicator), based on the best available literature (Bell et al.,2004; Jerrett et al., 2009).
The extent to which concentration–response relationships for PMand O3 are independent of each other is subject to differentinterpretations. Therefore, in this study, deaths attributed to thetwo pollutants are reported separately to avoid possible double-counting. Likewise, we do not add short-term and long-term healtheffects, as it is generally expected that short-term effects will be asubset of long-term effects.
In assessing uncertainty in the estimated burden of disease, weassume, based on the general epidemiologic literature, that theconcentration–response coefficients (β) are normally distributedwithmean values as shown in Table 1 and standard deviations computedfrom the 95% confidence intervals reported in Table 1.
2.2.3. Exposure assessmentIn order to assess population exposure to ambient air pollution in
the UAE, we rely on measurements of PM and O3 from fixed-site air
Table 1Concentration–response coefficients selected to estimate the disease burden attributable to
Pollutant Exposure indicator (unit) Exposure type Cause of death
PM10 Daily average (μg/m3) Short-term All causes
Respiratory failureb
PM2.5 Annual average (μg/m3) Long-term All causes
Cardiopulmonary failu
Lung cancerd
O3 Daily (24-h) average (ppb) Short-term All causes other than a
Cardiovascular or respfailuref
Annual average of daily 1-hmaximum concentration (ppb)
Long-term Respiratory failureb
a Units are % increase in incidence per 1 μg/m3 PM or per 1 ppb O3.b Respiratory mortality refers to International Classification of Diseases, Ninth Revision (c Cardiopulmonary mortality refers to ICD-9 codes 401–440 and 460–519.d Lung cancer mortality refers to ICD-9 code 162.e Total non-accidental excludes deaths caused by injuries and other external causes.f Cardiovascular and respiratory mortality refers to ICD-9 codes 390–448, 480–486, 490–
quality monitoring stations in the UAE. At the time of this study,monitored PM10 data were available from ten stations, and O3 datawere available from seven stations, all located in the Abu Dhabiemirate, for the years 2007 and 2008. In total, Abu Dhabi emiratecomprises more than 80% of the geographic area of the UAE, so thatAbu Dhabi's monitoring network can serve as a reasonable proxy forUAE-wide air monitoring until data can be obtained from otheremirates. Since baseline mortality data were only available for theyear 2007, we use 2007 air quality data to estimate the annualmortality attributed to ambient air pollution. We use statisticalmethods to interpolate ambient concentrations at locations and attimes for which monitored concentrations are unavailable. Toestimate the pollution attributable to human activities, we subtractan estimated natural background concentration from the estimatedambient concentration.
The first step in our estimation of the ambient concentration is theextrapolation of pollutant levels at different spatial locations in theUAE based on observed air quality data from the 10 monitoringstations. To develop this information, we used the universal krigingprocedure (see, e.g., Cressie, 1993). Given a finite set of n observationsY=(y(s1, t),…,y(sn, t))′ from a space/time random field, krigingprovides an estimation of the variable of interest (i.e., PM10 or O3
concentration) at a location and time, say y(s0, t), that have not beenobserved. We use a mean trend that varies linearly with respect tospace and time. Hence, the universal kriging estimates are essentiallydriven by observed values when the estimation point is near amonitoring station, while estimates are driven by the linear trendwhen the estimation point is far from any monitoring stations. Inorder to visualize the spatial variability in pollutant exposureestimates, we divide the UAE into a grid of cells (measuring 0.3times 0.25°, or approximately 0.86 km2), use kriging to estimatepollutant concentrations in each cell, and draw maps using theconcentrations estimated in each grid cell at each point in time. Fig. 1shows two maps reporting estimates for daily PM10 concentration ontwo different days. Fig. 2 shows analogous plots for O3 concentration.
At the time this study was conducted, no data existed to enableestimates of the fraction of PM10 that is PM2.5. Therefore, weestimated PM2.5 concentrations based on an empirical PM2.5/PM10ratio of 0.35 suggested by the WHO for arid, desert regions (Ostro,2004). A recent study on the characterization of PM for three sites inKuwait reported that PM2.5 comprised 47% of PM10 at two sites and41% at the other (Brown et al., 2008). A recent, preliminary,
ambient air pollution in the UAE.
Reference (study type) β (95%CI)a Age group
Ostro (2004)(meta-analysis and expert judgment)
0.08(0.06–0.1)
All ages
Ostro (2004)(meta-analysis)
0.166(0.034–0.3)
Ageb5
Pope et al. (2002)(multi-city cohort study)
0.6(0.2–1.1)
AgeN30
rec 0.9(0.3–1.6)
AgeN30
1.4(0.4–2.3)
AgeN30
ccidentse Bell et al. (2004)(multi-city time-series study)
0.052(0.027–0.077)
All ages
iratory Bell et al. (2004)(multi-city time-series study)
0.064(0.031–0.098)
All ages
Jerrett et al. (2009)(multi-city cohort study)
0.4(0.1–0.67)
AgeN30
ICD-9) codes 470–478, 480–488 and 490–493.
497 and 507.
Fig. 1. Daily average PM10 concentration (μg/m3) for the UAE on May 22, 2007 (top)and April 5, 2008 (bottom). Fig. 2.Daily average O3 concentration (ppb)map for the UAE on June 21, 2007 (top) and
May 6, 2008 (bottom).
5788 Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
unpublished analysis of the PM2.5/PM10 ratio at three air monitoringstations in Abu Dhabi reported average ratios of 0.38, 0.40, and 0.38,with ranges of 0.20–0.70, 0.18–0.74 and 0.20–0.61, respectively(Boehler, 2010). This evidence suggests that the WHO's 0.35 scalingfactor is a reasonable initial assumption for the UAE. However, theburden of disease estimate presented here should be updated as new,confirmed data on PM2.5 concentrations in the UAE become available.
In estimating the burden of disease, we assume that ambientconcentrations of pollutants in each grid cell are lognormallydistributed, with means and standard deviations estimated by thekriging procedure.
2.2.4. Natural background levels of pollutantsFor each pollutant, a natural background level that reflects the
non-anthropogenic concentration is needed in order to capture thedisease burden attributable to man-made air pollution. Here,“background level” does not mean a threshold concentration belowwhich no health effects occur. Current scientific evidence does notclearly indicate the existence of such a threshold for either PM or O3.Instead, some studies report that the association between PM or O3
and mortality persists at concentrations close to zero (Schwartz et al.,2002; Jerrett et al., 2009). This study assumes no threshold forpollutant health effects. However, a natural background level issubtracted from the ambient pollutant concentration to estimate thedisease burden due to anthropogenic pollution rather than pollutionfrom natural sources, which is not easily controlled by humans andthus of less interest.
TheWHO recommended using natural background levels of 10 μg/m3 and 5 μg/m3 for PM10 and PM2.5, respectively, based onobservations in typical urban areas in the United States (Ostro,
2004). However, these values might be too low for the UAE, wherefrequent dust storms occur. Lacking further evidence from the MiddleEast region, we assumed the natural background concentration ofeach pollutant is uniformly distributed. For PM, the levels recom-mended by the WHO (Ostro, 2004) are employed as the lower bound,and the upper bound is based on the possible high natural backgroundlevels that may exist in an arid dessert region as indicated in previousresearch studies in desert regions of the western United States (Ostroet al., 2000). Thus, we assume that for PM10, the parameters of theuniform distribution are 10 μg/m3 and 90 μg/m3; for PM2.5, theparameters are 5 μg/m3 and 35 μg/m3. For O3, a zero background isassumed to be the lower bound, and 25 ppb is applied as the upperbound, based on other studies of the disease burden attributable toground-level O3 (Anenberg et al., 2009).
2.2.5. Population sizeTo estimate the spatial variation in population, we mapped the
UAE population on to the same grid created for estimating pollutantconcentrations, using data from the LandScanTM Global PopulationDatabase 2007 (available at http://www.ornl.gov/landscan/). Allpeople living within each grid cell are assumed to be exposed to thesame pollutant concentration.
2.2.6. Baseline incidence ratesThe annual report of the UAE Ministry of Health provided
mortality data for 2007 (Ministry of Health, 2008). The reportprovided cause-specific mortality data in specific age groups for thehealth outcomes shown in Table 1; we estimated mortality rates bydividing mortalities by the age-specific population. Given that age-
5789Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
specific population data were available only from the Abu Dhabiemirate (Health Authority-Abu Dhabi, 2008) and the entire UAE(Ministry of Health, 2008), the UAE-wide age distribution is applied tothe population in each of the six smaller emirates to derive age-specific populations. Annual baseline incidences are assumed to beconstant across an emirate over the study year. Table 2 summarizes allthe baseline mortality rates by emirate for each of the healthendpoints considered in this study.
2.2.7. Burden of disease calculationUsing the concentration–response coefficients, estimates of natu-
ral background pollutant levels, population data, and mappedpollutant concentrations, we estimated the number of excess deathsdue to PM and O3 across the UAE.We apply Eq. (4) using estimates foreach variable specific to each grid cell and then aggregate theattributed deaths across all the grid cells to estimate the totalmortality burden due to each pollutant. We used two different gridresolutions: a coarse grid that divides the entire UAE into 1409 cells(approximately 55 km2), and a fine grid consisting of 90,176 cells(30×30 arc second resolution, approximately 0.8649 km2). Thecoarse grid serves as the basis for our estimate of the total burdenof disease across the UAE. We also use it to analyze the uncertainty inour estimate. We use the fine grid to map the spatial variation inmortality attributed to ambient air pollution.
In applying Eq. (4), we represent uncertainty in the concentrationof the pollutant (designated as Δx) and in the concentration–responsecoefficient (ß). The representation of uncertainty in the pollutantconcentration includes uncertainty in the measured levels of thepollutant in the ambient environment and in the assumptions aboutthe natural background levels of the pollutant. We used Monte Carlosimulation (coded in Analytica software from Lumina DecisionSystems, Inc.) to quantify uncertainty. We used the median Latinhypercube sampling method in Analytica and a sample size of 1000.
To map the spatial variation in the burden of disease using the finegrid resolution, we used deterministic estimates for each variable inEq. (4). We assume that ambient pollutant concentrations are fixed atthe mean value in each grid cell; that the concentration–responsecoefficients are fixed at their mean values; that the background levelsof PM10 and PM2.5 are 10 and 5 μg/m3, respectively, as recommendedby the WHO; and that the background level of O3 is zero.
3. Results and discussion
3.1. Total burden of disease due to ambient air pollution
Figs. 3 and 4 summarize the results of our estimates of the numberof deaths due to each combination of health endpoint and pollutantconsidered. The figures show estimates for the entire UAE and also forAbu Dhabi emirate alone, since the observational air quality data arebased on Abu Dhabi's air monitoring network. Although Figs. 3 and 4
Table 2Estimated baseline ratesa of selected causes of death in the UAE in 2007.(Source: Ministry of Health Annual Report, Ministry of Health, UAE; 2007).
Emirate All-cause mortality All-cause mortalityin adults over age 30
a Units are cases per year per person in the population or for a particular age group as spb Cardiopulmonary, respiratory and lung cancer mortalities refer to the ICD-9 codes liste
show estimated deaths for several health outcomes, these numbersare not additive, because some of the health end points are subsets ofothers.
As Fig. 3 shows, our best estimate is that anthropogenic PM mayhave accounted for 545 (542 adults and 3 children) (95% CI: 132–1224) premature deaths in the UAE — about 7% of the total deaths(7414) in 2007. The number of deaths attributed to anthropogenic PMin Abu Dhabi is estimated to be 209 (208 adults and 1 children) (95%CI: 40–449). The WHO suggests that when estimating the burden ofdisease due to PM in ambient air, it is reasonable to aggregatecardiopulmonary and lung cancer mortality related to long-termexposure with respiratory mortality in infants and children related toshort-term exposure (Ostro, 2004). However, we believe thatcardiopulmonary and lung cancer mortality may not capture all thepremature deaths caused by PM exposure. Given this, we combine all-cause mortality in adults related to PM2.5 and respiratory mortality inchildren related to PM10 in aggregating the total PM-relatedpremature deaths, in order to avoid double-counting. This estimate,shown on the far right-hand side of Fig. 3, is likely to be conservativebecause the population in the 5–30 age range is not included.
As our best estimate of the total burden of disease due to ground-level O3, we select all-cause mortality caused by short-term O3
exposure, given that only respiratorymortality in adults over age 30 isincluded in the estimated effects of long-term exposure. Therefore, in2007, anthropogenic O3 is estimated to cause about 62 (95% CI: 17–127) premature deaths (Fig. 4), which accounts for 0.8% of the totaldeaths in the UAE in that year. In Abu Dhabi, the number of deathsattributed to anthropogenic O3 is estimated to be 23 (95% CI: 6–51).Premature deaths caused by ground-level O3 thus appear to be muchfewer in number than those caused by PM.
3.2. Uncertainty and sensitivity analysis
The wide confidence intervals around the estimated burden ofdisease due to ambient air pollution (Figs. 3 and 4) show theconsiderable uncertainty in these estimates. Uncertainties exist inevery component of the burden of disease assessment. As notedabove, we estimated how uncertainty in three of the key inputs – theconcentration–response coefficients, the natural background levels ofpollutants, and the measured ambient concentrations – propagatesthrough our risk model and affects the estimated burden of disease.We assessed the contribution of each of these uncertain inputvariables to the uncertainty in the estimated number of prematuredeaths in adults over age 30 due to PM2.5 exposure — the health endpoint with the highest burden of disease. We also conductedsensitivity analyses for all the health end points to examine theeffects of adjusting these three categories of input variables upwardsor downwards.
Fig. 5 shows the mean absolute value of the rank-order correlationbetween the three uncertain input variables and the estimated total
Fig. 3. Estimated annual excess deaths due to PM air pollution in the UAE and in the emirate of Abu Dhabi. Note that the numbers indicated are not additive, since some healthoutcomes are subsets of others. The estimate on the far right indicates the total mortality attributed to PM.
5790 Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
long-term all-cause mortality due to PM2.5 in adults over age 30. Therank-order correlation between a model input and its output showsthe degree to which uncertainty in the input contributes touncertainty in the output. As shown, the uncertainty in thebackground PM2.5 concentration is the greatest contributor to theuncertainty in the estimated number of deaths. The uncertainty in theconcentration–response coefficient is the second-most importantcontributor to the uncertainty. Thus, reducing uncertainty in theseinput variables would substantially reduce the uncertainty (asindicated in the wide confidence bands shown in Fig. 3) in theestimated number of deaths attributable to PM in ambient air.
To estimate the sensitivity of our predictions to changes in the keyinput variables, we adjusted each key input one at a time by ±25%,while holding all the remaining variables at their baseline values.Table 3 shows the results. As shown, our estimates are sensitive to allthree of the uncertain input variables, but changing the estimatedambient pollutant concentrations has the greatest effect on theburden of disease estimates for each pollutant and health end point.
Fig. 4. Excess deaths attributed to O3 pollution in ambient air in the UAE and in the emirate oend points are subsets of others shown.
Estimates of the burden of disease change by 34–42% in response to a25% increase or decrease in estimated ambient pollutant levels. On theother hand, while most of the uncertainty in the estimated burden ofdisease is due to uncertainty in the estimated natural backgroundlevels of pollutants, the estimated burden of disease is somewhat lesssensitive to estimates of the natural background level than it is toestimates of the ambient concentration of pollutants, with a 25%change in estimated background levels changing estimated attribut-able diseases by 12–21%. Changing the concentration–responsecoefficient has an approximately proportionate effect, in general, onthe estimated burden of disease.
3.3. Spatial variability in attributed mortality estimates
Figs. 6 and 7 show how the attributable fraction (Eq. (2)) and all-cause mortality rates due to long-term exposure to PM2.5 vary acrossthe UAE. We focus on this health end point since it accounts for thebulk of the disease burden attributable to PM. These maps are based
f Abu Dhabi for the year 2007. Note that the numbers are not additive, since some health
Fig. 5. Contribution of the indicated variables to the total uncertainty (i.e., width of the confidence intervals) in the estimated burden of disease due to PM2.5. As shown, uncertainty inthe background PM2.5 concentration contributes the most to uncertainty in the estimated burden of disease.
5791Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
on the high-resolution grid and deterministic values for all para-meters. The attributable fraction is highest in the Abu Dhabi City area(as seen in Fig. 6), most likely because the highest pollution levels arefound in this area. The estimates of all-cause deaths attributed toambient PM2.5 exposure also are highest in the Abu Dhabi City area(Fig. 7), not only because pollutant levels are high but also because all-cause mortality rates are relatively high. On the other hand, Dubaiemirate has the lowest estimated PM2.5-related mortality rates(Fig. 7). This is in part because the baseline all-cause mortality rateis lower in Dubai than in the other emirates. However, it also may bean artifact of the lack of air quality data obtained for this analysis inthe Dubai City area. Since we were unable to obtain measurements
Table 3Sensitivity of mortality estimate to changes in model input variables.
Pollutant Exposure term Mortality health outcome Total mort
Original es
Effect of changing assumed ambient concentration of pollutantPM10 Short-term All-cause 331
Respiratory 2.14PM2.5 Long-term All-cause 542
Cardiopulmonary 206Lung cancer 19
O3 Short-term All-cause 62Cardiovascular and respiratory 17
Long-term Respiratory 59
Effect of changing assumed natural background level of pollutantPM10 Short-term All-cause mortality 331
Cardiopulmonary mortality 206Lung cancer mortality 19
O3 Short-term All-cause mortality 62Cardiovascular and respiratory mortality 17
Long-term Respiratory mortality 59
specific to Dubai, we had to rely on measurements at air qualitymonitors in Abu Dhabi emirate to estimate air quality in Dubai. Theseestimates likely do not capture the expected peak concentrations inDubai. If urban concentrations in Dubai are similar to those in AbuDhabi, many more premature deaths would be expected in Dubai.
Also important to consider are the day-to-day variations inmortality caused by short-term exposure to PM. Fig. 8 shows thetemporal variation in mortality attributed to short-term exposure toPM10 over the year 2007, averaged across the entire UAE. As shown,substantial temporal variability exists, with a general trend of highermortality associated with PM occurring during the warm summermonths.
ality in the UAE
timate Estimate if variable decreases 25% Estimate if variable increases 25%
Fig. 6. Attributable fraction of all-cause mortality in adults over age 30 due to long-termexposure to PM2.5 for the year 2007. Blue lines show state and international highwayswith four lanes or more, and black lines show emirate boundaries. Abu Dhabi (thelargest emirate) comprises most of the geographic area of the UAE, with the smaller,northern emirates shown in the upper right side of the map.
Fig. 8. Daily mortality caused by short-term exposure to ambient PM10 in the UAE(January 1, 2007–December 31, 2008).
5792 Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
4. Conclusions
Our best estimate is that the total number of premature deaths inthe UAE caused by exposure to anthropogenic PM is approximately545 (95% CI 132–1224). These deaths account for approximately 7%(95% CI 2–17%) of the total deaths occurring in the UAE in 2007. Thenumber of deaths attributed to ground-level O3 exposure isapproximately 62 (95% CI 17–127), or about 1% (95% CI 0.2–2%) ofthe total deaths in 2007. This study reports themortality related to PMand to O3 separately, since current epidemiologic evidence does notclearly indicate whether the health effects of the two pollutants areindependent of each other.
Despite the uncertainties in our estimates, some general conclusionscan be drawn from the overall mortality estimates: (1) anthropogenicambient air pollution was a causative factor in a substantial number ofpremature deaths in the UAE in 2007, and (2) PM causes a significantlygreater number of premature deaths compared to ground-level O3.Further, the mean estimates for the burden of disease due to PM in theUAE are greater than the global total percentage of premature deathscaused by PM. Cohen et al. (2004) estimated that outdoor air pollution
Fig. 7. Attributed all-cause mortality rate (mortality per 100,000 population per year)caused by long-term exposure to PM2.5 for the year 2007.
accounts for approximately 1.4% of total mortality in urban areasworldwide (as compared to our estimate of 7% for the UAE). Therefore,ambient PM should be set as a top priority in developing nationalprograms to protect public health from environmental risks in the UAE.
These burden of disease estimates are highly sensitive to assump-tions about ambient pollutant concentrations. Further, a lack ofinformation about the fraction of pollution that can be attributed tonatural sources (including dust storms) is the major source ofuncertainty in the estimates presented in this paper. Additionalmonitoring sites, in particular in the Dubai area, where peakconcentrations are expected, will be very helpful in improving burdenof disease estimates in the future. For PM, given that frequent duststorms occur in the study area, natural background levels of PM couldbe considerably higher than those found in other locations. Further,background concentrations of PM may vary quite substantially on adaily basis, due to natural variations in windblown dust loadings.Sensitivity analyses suggest that increases in the assumed naturalbackground levels result in significant decreases in the attributedmortality estimates for both pollutants. However, separating thenatural and anthropogenic contributions to PM in ambient air may bedifficult, especially as human activities such as construction mayincrease the dust load to the environment. Future research shouldconsider using an ambient air quality model to simulate atmosphericconcentrations in the absence of anthropogenic emissions and defineanthropogenic contributions to air pollution accordingly (e.g. Anen-berg et al., 2010).
Another significant source of uncertainty derives from publishedestimates of concentration–response coefficients. This study relies onconcentration–response coefficients from representative epidemio-logical studies in the United States, because local studies are notavailable. The bias caused by extrapolation of findings from onelocation to another might be reduced in future research as localepidemiological studies become available and are incorporated intohealth impact assessments.
Finally, although this study only focuses on premature mortality,which is the most serious health outcome and thus of the greatestconcern, previous studies also have found that exposure to ambientair pollution causes a wide range of nonfatal outcomes, includingincreases in hospital admissions due cardiovascular problems andincreased rates of asthma and other respiratory symptoms (Ostro andChestnut, 1998). These nonfatal illnesses can cause significanteconomic losses due to the associated health-care costs and loss ofproductivity. Future research should include morbidity estimates asthe related baseline health statistics become available so that a morecomprehensive assessment for the disease burden of ambient airpollution in the UAE can be presented.
5793Y. Li et al. / Science of the Total Environment 408 (2010) 5784–5793
Acknowledgements
The work reported here was possible only due to the support ofnumerous individuals. We especially thank H. E. Majid Al Mansouri,Secretary General of the Environment Agency-Abu Dhabi (EAD), forsponsoring this research. Sincere thanks are also due to many staffmembers at EAD who provided critical data and other information:Dr. Jaber Al Jaberi, Dr. Ahmed Bashir, Dr. Mahmoud Abdulrahim, Dr.David Blackmore, Eng. Hazem Qawasmeh, Ms. Ayesha Al Suweidi, Ms.Ruqaya Mohamed, Ms. Ayesha Abushahab, and many others. We alsoowe our sincere thanks to Mr. Chris Davidson for managing the datasets required for this analysis and to Dr. Zeinab Farah for her tirelessefforts to pursue the data without which this work would have beenimpossible.
References
Anderson HR, Atkinson RW, Peacock JL, Marston L, Konstantinou K. Meta-analysis oftime-series studies and panel studies of particulate matter (PM) and O3 (O3):report of a WHO task group. Copenhagen: World Health Organization; 2004.
Anenberg SC,West JJ, Fiore AM, Jaffe DA, Prather MJ, Bergmann D, et al. Intercontinentalimpacts of ozone pollution on human mortality. Environ Sci Technol 2009;43(17):6482–7.
Anenberg SC, Horowitz LW, Tong DQ, West JJ. An estimate of the global burden ofanthropogenic ozoneandfineparticulatematteronprematurehumanmortality usingatmosphericmodeling. EnvironHealth Perspect 2010. doi:10.1289/ehp.0901220.
Aunan K. Exposure-response functions for health effects of air pollutants based onepidemiological findings. Risk Anal 1996;16:693–702.
Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F. O3 and short-term mortality in95 US urban communities, 1987–2000. J Am Med Assoc 2004;292:2372–8.
Bell ML, Dominici F, Samet JM. A meta-analysis of time-series studies of O3 andmortality with comparison to the National Morbidity, Mortality, and Air PollutionStudy. Epidemiology 2005;16:436–45.
BellML, PengRD,Domenici F. Theexposure–responsecurve forO3and riskofmortality andthe adequacy of current O3 regulations. EnvironHealth Perspect 2006;114:532–6.
Boehler T. Unpublished memo to UNC: PM2.5/PM1 ratios at Abu Dhabi sites. Addressedto Dr.Kenneth Sexton, UNC. From Trond Boehler, Norwegian Institute for AirQuality Research. Reference O-1098123; February 8, 2010; 2010.
Brown KW, Bouhamra W, Lamoureux DP, Evans JS, Koutrakis P. Characterization ofparticulate matter for three sites in Kuwait. J Air Waste Manage 2008;58:994-1003.
Cohen AJ, Anderson HR, Ostro B, Pandey KD, Krzyzanowski M, Kunzli N. Mortalityimpacts of urban air pollution. In: Ezzati M, et al, editor. Comparative quantificationof health risks: global and regional burden of disease due to selected major riskfactors. Geneva: World Health Organization; 2004.
Cressie NAC. Statistics for spatial data. New York: Wiley; 1993.DockeryDW,PopeIIICA,XuX,SpenglerJD,WareJH,FayME,etal.Anassociationbetweenair
pollution andmortality in six U.S. cities. N Engl JMed 1993;329:1753–9.Dominici F, McDermott A, Daniels M, Zeger SL, Samet JM. Revised analyses of the
National Morbidity, Mortality, and Air Pollution Study: mortality among residentsof 90 cities. J Toxicol Environ Health Part A 2005;68:1071–92.
Health Authority Abu Dhabi. Health statistics 2007. Health Authority Abu Dhabi; 2008.Ito K, De Leon SF, Lippmann M. Associations between O3 and daily mortality: analysis
and meta-analysis. Epidemiology 2005;16:446–57.Jerrett M, Burnett RT, Pope III CA, Ito K, Thurston G, Krewski D, et al. Long-term O3
exposure and mortality. N Engl J Med 2009;360:1085–95.Krewski D, Burnett RT, Goldberg MS, Hoover K, Siemiatycki J, Jarret M, et al. Reanalysis
of the Harvard six cities study and the American Cancer Society study of particulate
air pollution and mortality. A special report of the Institute's particle epidemiologyreanalysis project. Cambridge, MA: Health Effects Institute; 2000.
Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, et al. Extended follow-up andspatial analysis of the American Cancer Society study linking particulate airpollution and mortality. Res Rep Health Effect Inst 2009;140:5-114.
Künzli N, Medina S, Kaiser R, Quénel P, Horak Jr F, Studnicka M. Assessment of deathsattributable to air pollution: should we use risk estimates based on time series oron cohort studies? Am J Epidemiol 2001;153:1050–5.
Laden F, Schwartz J, Speizer FE, Dockery DW. Reduction in fine particulate air pollutionand mortality: extended follow-up of the Harvard six cities study. Am J Respir CritCare Med 2006;173:667–72.
Levy JI, Hammitt JK, Spengler JD. Estimating themortality impacts of particulate matter:what can be learned from between-study variability? Environ Health Perspect2000;108:109–17.
Levy JI, Chemerynski SM, Sarnat JA. O3 exposure and mortality: an empiric Bayesmetaregression analysis. Epidemiology 2005;16:458–68.
Ministry of Health. 2007 annual statistical report. UAE: Preventive Medicine Sector,Ministry of Health; 2008.
National Research Council. Estimating mortality risk reduction and economic benefitsfrom controlling O3 air pollution. Washington, D.C.: National Academies Press;2008.
Ostro BD. Outdoor air pollution: assessing the environmental burden of disease atnational and local levels. WHO Environmental Burden of Disease Series, No. 5;2004.
Ostro BD, Chestnut L. Assessing the health benefits of reducing particulate matter airpollution in the United States. Environ Res Section A 1998;76:94-106.
Ostro BD, Hurley S, Lipsett MJ. Air pollution and daily mortality in the Coachella Valley,California: a study of PM10 dominated by coarse particles. Environ Res 1999;81(3):231–8.
Ostro BD, Broadwin R, Lipsett MJ. Coarse and fine particles and daily mortality in theCoachella Valley, California: a follow-up study. J Expo Anal Environ Epidemiol2000;10:412–9.
Pope III CA, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, et al.Particulate air pollution as a predictor of mortality in a prospective study of U.S.adults. Am J Respir Crit Care Med 1995;151:669–74.
Pope III CA, Hill RW, Villegas GM. Particulate air pollution and daily mortality on Utah'sWasatch Front. Environ Health Perspect 1999;107:567–73.
Pope III CA, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, et al. Lung cancer,cardiopulmonary mortality, and long-term exposure to fine particulate airpollution. J Am Med Assoc 2002;287:1132–41.
Samet JM, Zegar SL, Dominici F, Curriero F, Coursac I, Dockery DW, et al. The nationalmorbidity, mortality, and air pollution study, part II: morbidity and mortality fromair pollution in the United States. Res Rep Health Effect Inst 2000;94(Pt 2):5-70.
Schlesinger RB. The health impact of common inorganic components of fine particulatematter (PM2.5) in ambient air: a critical review. Inhal Toxicol 2007;19(10):811–32.
Schwartz J, Laden F, Zanobetti A. The concentration–response relation between PM2.5
and daily deaths. Environ Health Perspect 2002;110:1025–9.Slaughter JC, Kim E, Sheppard L, Sullivan JH, Larson TV, Claiborn C. Association between
particulate matter and emergency room visits, hospital admissions andmortality inSpokane, Washington. J Expo Anal Environ Epidemiol 2005;15:153–9.
Thurston GD, Ito K, Mar T, et al. Workgroup report: workshop on source apportionmentof particulate matter health effects — intercomparison of results and implications.Environ Health Perspect 2005;113(12):1768–74.
USEPA. The benefits and costs of the CleanAir Act, 1990–2010. EPA-410-R-99-001United States Environmental Protection Agency Office of Air and RadiationPolicy; 1999.
WHO. Country profiles of environmental burden of disease: United Arab Emirates.Geneva, Switzerland: World Health Organization; 2009. http://www.who.int/quantifying_ehimpacts/national/countryprofile/unitedarabemirates.pdf.
Zeka A, Zanobetti A, Schwartz J. Short term effects of particulate matter on causespecific mortality: effects of lags and modification by city characteristics. OccupEnviron Med 2005;62:718–25.
