ORIGINAL PAPER

The effects of the 2009 dust storm on emergency admissionsto a hospital in Brisbane, Australia

Adrian G. Barnett & John F. Fraser & Lynette Munck

Received: 27 January 2011 /Revised: 29 June 2011 /Accepted: 30 June 2011 /Published online: 26 July 2011# ISB 2011

Abstract In September 2009 an enormous dust stormswept across eastern Australia. Dust is potentially hazard-ous to health as it interferes with breathing, and previousdust storms have been linked to increased risks of asthmaand even death. We examined whether the 2009 Australiandust storm changed the volume or characteristics ofemergency admissions to hospital. We used an observation-al study design, using time series analyses to examinechanges in the number of admissions, and case-onlyanalyses to examine changes in the characteristics ofadmissions. The admission data were from the PrinceCharles Hospital, Brisbane, between 1 January 2009 and31 October 2009. There was a 39% increase in emergencyadmissions associated with the storm (95% confidenceinterval: 5, 81%), which lasted for just 1 day. The healtheffects of the storm could not be detected using particulatematter levels. We found no significant change in thecharacteristics of admissions during the storm; specifically,there was no increase in respiratory admissions. The duststorm had a short-lived impact on emergency hospitaladmissions. This may be because the public took effectiveavoidance measures, or because the dust was simply nottoxic, being composed mainly of soil. Emergency depart-ments should be prepared for a short-term increase inadmissions during dust storms.

Keywords Dust storm . Air pollution . Emergencyadmissions . Particulate matter

Introduction

In September 2009 an enormous dust storm swept acrosseastern Australia. The storm originated in the Lake EyreBasin of central Australia, and soil from this arid regionwas blown north and east over some of Australia’s biggestcities. On 24 September the storm measured 3,450 km long(from north to south) (Schmaltz 2009). A second dust stormhit the east coast just a few days later. The storm reducedvisibility severely, which caused major disruptions to roadand air travel. The media reported people experiencingbreathing difficulties and increased ambulance call-outs.People with asthma and other respiratory conditions wereadvised to stay indoors. It was thought that this would givesome degree of protection, but the dust was sucked into airconditioners and seeped through gaps in floors and walls,leaving a layer of dust on the inside and outside of manyhomes.

Dust storms are potentially hazardous to health on anumber of levels. Large dust particles can deposit in thelung airways, causing tracheo-bronchial irritation, andreactive airway disease exacerbations (Taylor 2002).Smaller particles progress further down the tracheobronchialtree, inducing inflammation and particulate pneumonitis. Duststorms have been associated with an increased risk of asthmain children (Gyan et al. 2005; Yoo et al. 2008), and increasesin mortality and hospital admissions due to cardiovascularand respiratory diseases (Hashizume et al. 2010; Perez et al.2008; Chen et al. 2004). Dust storms can act as a vector formicroorganisms, and have been linked to outbreaks ofcoccidiomycosis (Griffin 2007), and may have the

A. G. Barnett (*)School of Public Health & Institute of Health and BiomedicalInnovation, Queensland University of Technology,60 Musk Avenue,Kelvin Grove, Queensland, Australiae-mail: a.barnett@qut.edu.au

J. F. Fraser : L. MunckCritical Care Research Group, The Prince Charles Hospital,Chermside, Queensland, Australia

Int J Biometeorol (2012) 56:719–726DOI 10.1007/s00484-011-0473-y

potential to cause pneumonia, meningitis and bacteremia(Polymenakou et al. 2007).

The September 2009 dust storm was an unusual event,although a similar storm occurred in Australia in 2002(Chan et al. 2005). Dust storms may become more commonin Australia as the driest continent on Earth becomes evendrier due to worsening global warming. There is a clearneed to estimate the health effects of dust storms inAustralia, which may become more frequent and severe.We focussed on the acute effects of the storm by examiningemergency admissions to the Prince Charles Hospital, alarge tertiary referral hospital in the northern suburbs ofBrisbane.

We had two questions. Did the number of emergencyadmissions to hospital increase after the dust storm, and ifso for how long? Did the characteristics of emergencyadmissions change during the dust storm? In particular wewere interested in whether the proportion of respiratoryadmissions increased.

Methods

We studied the effects of the dust storm in Brisbane,Australia, the capital of the state of Queensland (27°28′S,153°01′E). On 30 June 2009, Brisbane had a population of1,052,458. Brisbane has a sub-tropical climate with averageJanuary (summer) daily maximum temperatures of 30.3°C,and average July (winter) maximum of 21.9°C.

Pollution and health data

Particulate matter levels are sensitive to the levels ofdust in the air, and are monitored hourly at four fixedstations in Brisbane. These stations measure particulatematter less than 2.5 μm and 10 μm in aerodynamicdiameter (PM2.5 and PM10, respectively), and alsomeasure daily levels of air temperature and humidity.Daily estimates of particles and weather variables weremade by averaging the levels from the four sites. Dailyvisibility at midday was obtained from the station in thecity centre. The visibility data relates to visibility loss dueto fine particles (particularly particles from combustionprocesses), and not to visibility loss due to other factorssuch as rain.

The health data were the emergency admissions to thePrince Charles Hospital from 1 January 2009 to31 October 2009. The admissions in the period prior tothe start of the dust storm (23 September) were used as acontrol period to compare with the admissions during thedust storm. We examined admissions up to 31 October aswe believed that any effects of the dust storm wouldhave ended by this date (Table 1).

Time series of the number of admissions

We plotted the time series of the daily numbers of:emergency admissions, particulate matter levels, visibilityand weather variables. We also examined simple summarystatistics of the emergency admissions before and after thedust storm.

We investigated the effects of the dust storm in twoways: (1) using the daily levels of particulate matter, (2)using the timing of the dust storm. We examined both PM10

and PM2.5 measures of particulate matter, and modelled adelayed effect between exposure and admission of up to21 days. We log-transformed the particulate matter levelsbecause of the strong skew in the distribution caused by thedust storm.

We used the following Poisson regression model of thedaily number of hospital admissions,

Yt � Poisson mtð Þ; t ¼ 1; ::::n;

log mtð Þ ¼ a0 þ a1:6DOWt þ a7holidayt þ s t; 5ð Þþ ns temperature t�21;...;tf g; 4; 4

� �þ f dust storm t�21;...;tf g� �

;

where Yt is the number of admissions on day t, DOW is theday of the week, and holiday is a binary variable for holidays(Barnett and Dobson 2010). To control for any trends orseasonal patterns in admissions we used a penalised splines(,5) with five degrees of freedom per year (Bell et al. 2008).To control for temperature, we modelled a lagged effect overthe previous 21 days and a non-linear effect using a naturalspline (ns) (Gasparrini et al. 2010). We used four degrees offreedom for both the lagged and non-linear effect. f () is afunction that models the effect of the dust storm. We triedthree different functions as detailed below.

We estimated the effects of the dust storm using thefollowing third-order polynomial distributed lag model,

f dust storm t�21;:::;tf g� � ¼

X21d¼0

a0þa1 pollution t�d þ a2 pollution2t�d

þa3 pollution3t�d; t ¼ 1; :::; n;

Table 1 Basic characteristics of Prince Charles Hospital emergencyadmissions before (1 January to 22 September 2009) and after(23 September to 31 October 2009) the dust storm. IQR Inter-quartilerange

Before After

Number of days, N 265 39

Admissions, N 9,098 1,353

Admissions per day, median (IQR) 34 (29, 40) 35 (31.5, 39)

Length of stay (days), median (IQR) 3.5 (1.5, 6.5) 3.5 (1.5, 6.5)

Age (years), median (IQR) 62 (43, 78) 63 (45, 78)

Sex (male), N (%) 4,595 (50.5) 695 (51.4)

720 Int J Biometeorol (2012) 56:719–726

where pollutiont is the daily level of log-transformed PM10

or PM2.5. We fitted this model using the “dlnm” library inthe R software (Gasparrini et al. 2010).

For the two other dust storm models, we examined theeffects by time, rather than using the measured levels ofparticulate matter. We assumed a constant change inadmissions between a start and end date using,

f dust storm t�21;:::;tf g� � ¼ z; start � t � end;

0; otherwise:

nð1Þ

To estimate the start and end dates we fitted separatemodels for a start date from 0 to 7 days after the 1st day ofthe storm, and an end date from 1 to 21 days after the start.We selected the optimal start and end dates as those with thelowest deviance information criterion (DIC) (Spiegelhalteret al. 2002). This model was fitted using the JAGS software(Plummer 2010).

For our second timingmodel, we modelled a delayed effectof up to 21 days after the start of the dust storm using,

f dust storm t�21;:::;tf g� � ¼

X21d¼0

gd;

gd � N 0; s2d

� �; d ¼ 0; :::; 21;

s2d ¼ exp hdð Þ@2;

@2 � Uniform 0; 105� �

;

h � Uniform �30; 0ð Þ:

ð2Þ

This model assumes the effects of the dust stormapproach zero as the time since the start of the stormincreases (Welty et al. 2009). The model assumes anexponential decay in the variance σ2d, with the rate ofdecay controlled by the parameter η. This model was alsofitted using the JAGS software.

0.02

0.04

0.06

0.08

0.10

0.12

NO

X (

ppm

)

0.2

0.4

0.6

0.8

CO

(pp

m)

10

20

50

100

200

500

1000

PM

10 (

µg/m

3 )

2

5

10

20

50

100

PM

2.5

(µg/

m3 )

15

20

25

Tem

pera

ture

(°

C)

Jan Mar May Jul Sep Nov

20

40

60

80

Hum

idity

(%

)

Jan Mar May Jul Sep Nov

20

30

40

50

60

Em

erge

ncy

adm

issi

ons

Fig. 1 Time series of pollutants,weather variables and emergencyadmissions over time. The dottedvertical line shows the start ofthe dust storm (23September 2009). The y-axes forthe particulate matter (PM)results are on a log scale

Int J Biometeorol (2012) 56:719–726 721

All results are presented in terms of the percent changein admissions with 95% confidence intervals, or credibleintervals (CIs) for the models fitted using JAGS. Theresiduals from the final models were checked for indepen-dence using the cumulative periodogram test (Barnett andDobson 2010).

Modelling the characteristics of admissions

To examine whether the characteristics of emergency admis-sions changed during the dust storm, we used logisticregression with a dependent variable of admitted during thedust storm (yes/no), and independent variables of age, sex,length of stay (days), acute admission (yes/no), respiratoryadmission (yes/no) and cardiovascular admission (yes/no).This is a case-only analysis, and should be able to detectindividual factors that changed during the dust storm

(Schwartz 2005). We tested the importance of the variablesusing multiple regression, and found the best model using theAkaike information criterion (AIC) (Akaike 1974), using the“bestglm” library in the R software (McLeod and Xu 2010).

We defined the dust storm in two ways: (1) for 1 weekafter the start of the storm (23–29 September), (2) for thetime interval identified as the peak for emergency admis-sions using Model (1). For both definitions we used acontrol period of the week before the storm. We used twodefinitions in order to compare different assumptions aboutthe length of the storm.

As an alternative to logistic regression we also used aclassification tree (Breiman et al. 1984). We used adependent variable of dust storm (yes/no), with independentvariables of: sex, age, length of stay, acute admission,respiratory admission, cardiovascular admission, admissionstatus, admission unit, admission type, admission source,

0.01

0.02

0.03

0.04

0.05

NO

X (

ppm

)

0.20

0.25

0.30

0.35

0.40

0.45

CO

(pp

m)

20

50

100

200

500

1000

PM

10 (

µg/m

3 )

10

20

50

100

PM

2.5

(µg/

m3 )

18

20

22

24

26

Tem

pera

ture

(°

C)

20

30

40

50

60

70

Hum

idity

(%

)

25

30

35

40

45

50

Em

erge

ncy

adm

issi

ons

8 Sep

13 S

ep

18 S

ep

23 S

ep

28 S

ep3

Oct

8 Oct

51020

50100200

500

Vis

ual d

ista

nce

(km

)

8 Sep

13 S

ep

18 S

ep

23 S

ep

28 S

ep

3 Oct

8 Oct

Fig. 2 Time series of pollutants,weather variables and emergencyadmissions over time in the2 weeks before and after start ofthe dust storm. The dottedvertical line shows the start ofthe dust storm (23 September2009). The y-axes for the PMresults are on a log scale

722 Int J Biometeorol (2012) 56:719–726

diagnosis related group (Australian revision version 5;http://www.health.gov.au/internet/main/publishing.nsf/content/health-casemix-ardrg1.htm) and major diagnosticcategory. Many of these independent variables have a largenumber of categories that are not suited to logisticregression because the numbers in each category are sparse.A classification tree overcomes this problem by examiningall possible combinations of categories. This gives theclassification tree the ability to find an important changeduring the dust storm in either a narrow or broad range ofadmission characteristics. Although the classification treetests many combinations, it is protected against spuriousstatistical findings by using cross-validation (Breiman et al.1984). This model was fitted using the “rpart” library in theR software (Therneau and Atkinson 2009).

We used routinely collected data that were de-identifiedand so did not seek individual patient consent. The studywas approved by The Prince Charles Hospital HumanResearch Ethics Committee.

Results

Figures 1 and 2 show the massive increases in particulatematter (PM) due to the dust storm. On 23 September 2009,the PM10 particle levels were 894 μg/m3 and PM2.5 levelswere 138 μg/m3. These levels far exceeded the AustralianNational Environment Protection Measures for AmbientAir Quality standards of 50 μg/m3 for PM10 and 25 μg/m3

for PM2.5. The levels remained unusually high for around10 days. The particle levels peaked again on 26 Septemberbecause of the secondary storm. The visible distancedropped to 5 km on the day of the storm (the QueenslandEnvironmental Protection Air Policy target visibility is20 km). There was a large drop in humidity during the duststorm due to the increased wind speed.

Increases in the percentage of emergency admissions dueto the dust storm

Figure 3 shows the estimated effects of particulate matterfor PM2.5 and PM10. There were no increases in emergencyhospital admissions associated with increases in particulatematter.

Figure 4 shows the fit for Model (1) using the DIC and arange of different start and end dates for the increase inemergency admissions due to the storm. The smallest DIC(best fit) was for an increase in emergency admissions onjust one day: 24 September. The estimated increase inadmissions using Model (1) on 24 September was 39%more admissions than average (95% CI: 5, 81%).

Figure 5 shows the estimated increased in admissionsusing Models (1) and (2). Both models show that the effect

of the storm was short lived, with no increase in admissionsfrom 2 days after the storm and no impact from the secondstorm that hit 3–4 days after the first. Model (2) estimated a

Start date

End

dat

e

Sep 24

Sep 29

Oct 04

Oct 09

Oct 14

Oct 19

Sep 23 Sep 24 Sep 25 Sep 26 Sep 27 Sep 28 Sep 29 Sep 30

1925

1926

1927

1928

1929

1930

1931

1932

Fig. 4 Deviance information criterion (DIC) for different start andend dates using Model (1) to estimate the change in emergencyadmissions. The smaller the DIC, the better the model fit. The smallestDIC was 1,926 for a start date of 24 September and end date of25 September

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5PM2.5

Per

cent

cha

nge

−2

−1

0

1

2

3

0 5 10 15 20Lag (days)

PM10

Per

cent

cha

nge

Fig. 3 Effects of a log 10 μg/m3 increase in PM on emergencyadmissions over days 0 to 21 since exposure (lag days). Estimatesusing a third-order polynomial lag for particulate matter. Mean effect(solid line) and 95% confidence interval (dashed lines). The y-axesshow the rate ratios; a horizontal line is shown at a rate ratio of 1corresponding to no change in the rate of admissions

Int J Biometeorol (2012) 56:719–726 723

17% increase in emergency admissions over a lag of 0–1 days (95% CI: −17, 68%).

Differences in emergency admission characteristicsduring the dust storm

Table 2 shows the results from comparing the admissionsduring the dust storm to those admissions a week prior tothe storm. There were no significant changes in thecharacteristics of admissions for either definition of a duststorm. The best multiple variable model (according to theAIC) was one with no independent variables for a duststorm defined as the week after the storm started, and amodel with length of stay only for a dust storm defined asthe day after the storm started (24 September). For theshorter definition of the dust storm, there was someevidence of reduced lengths of stay (P-value=0.14). The

classification tree did not find any variables that differedduring the storm, indicating no change in patient character-istics during the dust storm.

There is a danger that this small study is underpoweredto detect a difference in admission characteristics. Toquantify this we used a power calculation to find thedetectable difference in odds ratio given the sample size.With 246 admissions during the dust storm and 240 controladmissions, the smallest detectable odds ratio is 1.8 for abinary independent variable with a 50% prevalence (e.g.sex), or an odds ratio of 2.3 for a binary variable with a10% prevalence. These calculations assumed a 90% powerand 5% two-sided significance level.

Discussion

Our results show that the dust storm had a substantial butshort-lived effect on emergency admissions to hospital.Admissions increased by 39%, but this increase lasted only1 day. We detected no change in the characteristics ofemergency admissions during the dust storm (using eitherlogistic regression or a classification tree); in particular,there was no increase in respiratory admissions. However,the study was only powered to detect relatively largechanges (odds ratios above 1.8).

If a 39% increase in emergency admissions wererepeated in every hospital in the path of a dust storm, thenthis would be a significant public health problem. Also, ourresults may underestimate the problem because we exam-ined only one aspect of the storm’s possible health effects.General practitioners or paediatric hospitals may haveborne the brunt of the health problems [a borderlineincrease in asthma hospital admissions was found on theday of a dust storm in a Brisbane children’s hospital(Rutherford et al. 1999)].

A possible reason for the limited effect could be thesuccessful avoidance strategies taken by the public (pre-dominantly staying at home), particularly those vulnerableto air pollution such as asthmatics and the elderly. Queens-

Date(s) covering the health effects of the storm

23–29 September 24 September

Variable (change) Mean (95% CI) P-value Mean (95% CI) P-value

Sex (male vs female) 1.19 (0.83, 1.72) 0.35 1.23 (0.66, 2.29) 0.51

Age (10 year increase) 1.03 (0.94, 1.13) 0.47 1.00 (0.86, 1.17) 0.96

LoS (1 day increase) 0.99 (0.96, 1.01) 0.29 0.95 (0.88, 1.02) 0.14

A&E (yes vs no) 0.76 (0.33, 1.76) 0.52 0.65 (0.14, 2.93) 0.57

Respiratory (yes vs no) 1.15 (0.67, 1.99) 0.61 0.64 (0.22, 1.86) 0.41

Cardiovascular (yes vs no) 0.95 (0.62, 1.46) 0.83 0.95 (0.47, 1.91) 0.88

Table 2 Case-only analysiscomparing emergency admis-sions during the dust storm(using two definitions) toadmissions 1 week before thestorm. Results from singlevariable logistic regressionmodels. LoS Length of stay,A&E admission via accident andemergency

−20

0

20

40

60

80

Days since storm

Per

cent

cha

nge

0 1 2 3 4 5 0 1 2 3 4 5Days since storm

Fig. 5 Percent change in emergency admissions due to the dust stormby the number of days since the storm using Model (1) (left) andModel (2) (right). Mean effect as a solid line and 95% credibleintervals using a dashed line. A dotted horizontal line at zerorepresents no change in admissions. Although we modelled anincrease up to 21 days the estimated effects were zero from 2 daysafter the storm for both models

724 Int J Biometeorol (2012) 56:719–726

land Health recommends the following actions during adust storm: staying indoors with windows and doors closed;using air conditioning on recirculate mode to reduceexposure to the outdoor air; and avoiding vigorous exercise(Queensland Health 2009). However, these avoidancemeasures are not universally available. In 2008, airconditioning was not available in 35% of homes inQueensland, and 20% of homes were made of timber,which is often not well sealed (Australian Bureau ofStatistics 2008).

We were not able to detect the effects of the dust stormusing particulate matter levels (Fig. 3), but we did seeeffects using the timing of the storm (Fig. 5). Particulatematter levels are based on the size of the particlesregardless of their chemical composition. However, thechemical composition of particles has been shown to alterthe risks associated with high exposures, with higher risksfor particles made from metals such as aluminium andnickel, or elemental carbon (Bell et al. 2009; Franklin et al.2008). A recent study in New York found no increased riskin hospital admissions with soil particulate matter, but didfind increases with traffic and steel particulate matter (Lallet al. 2011). The September 2009 dust storm in Australiawas mainly soil from the Lake Eyre region, and socontained mainly mineral oxides (including nitrogen andiron). Particles from traffic are mostly carbonaceousmaterials including hydrocarbons and carbon monoxide(Robert et al. 2007). These chemicals are far more toxic tohumans than soil dust, and, unlike the intermittent problemof dust storms, exposure to traffic pollution is a daily publichealth problem. An example of the negative health effectsof air pollution in Brisbane is an inter-quartile rangeincrease in carbon monoxide being associated with a 3.2%increase in cardiac hospital admissions in the elderly (95%CI: 1.5, 4.9%) (Barnett et al. 2006). This increase is farsmaller than the 39% increase shown here due to duststorms, but dust storms happen rarely, whereas increases incarbon monoxide have the potential to happen at any time(although levels are higher during winter, Fig. 1). Over 80%of the carbon monoxide in south-east Queensland comesfrom traffic, so a spike in carbon monoxide could be causedby a busy traffic day.

Previous studies have similarly found small adversehealth effects from dust storms. A study in Taipei showedthat hospital admissions for ischaemic heart disease were16–21% higher on sandstorm days, but there were noincreases in admissions for cerebrovascular disease, asthmaor pneumonia (Bell et al. 2008). Other studies in Taipeifound a non-statistically significant increase in hospitaladmissions and deaths due to dust storms (Chen et al. 2004,2005). A study in British Colombia found no effects of duststorms on hospital admissions (Bennett et al. 2008). Astudy in Cyprus found all-cause and cardiovascular admis-

sions were 4.8% (95% CI: 0.7, 9.0%) and 10.4% (95% CI:−4.7, 27.9%) higher on dust storm days, respectively(Middleton et al. 2008). Studies in Korea and Washingtonfound no increase in daily mortality due to dust storms(Kwon et al. 2002; Schwartz et al. 1999).

Our study had a narrow focus as we examinedadmissions from only one hospital. It would be useful tosee whether mortality increased during the dust storm(accidental and non-accidental). However, at the time ofwriting the daily data on mortality were not yet availablefor 2009.

Australia is likely to face more dust storms due to globalwarming. Australian hospitals should prepare for a short-term spike in admissions when the next dust storm hits.Vulnerable members of the public should limit theirexposure by staying indoors, turning on air conditioning(if it is available), and closing doors and windows(Queensland 2009).

Acknowledgements Computational resources and services used inthis work were provided by the High Performance Computer andResearch Support Unit, Queensland University of Technology,Brisbane, Australia. Thanks to Don Neale from the QueenslandDepartment of Environment and Resource Management for help withthe visibility data.

References

Akaike H (1974) A new look at the statistical model identification.IEEE Trans Autom Control 19:716–723

Australian Bureau of Statistics (2008) 4602.0.55.001 Environmentalissues: energy use and conservation, Mar 2008

Barnett AG, Dobson AJ (2010) Analysing seasonal health data.Statistics for Biology and Health. Springer, Berlin

Barnett AG, Williams GM, Schwartz J et al (2006) The effects of airpollution on hospitalizations for cardiovascular disease in elderlypeople in Australian and New Zealand cities. Environ HealthPerspect 114:1018–1023

Bell ML, Levy JK, Lin Z (2008) The effect of sandstorms and airpollution on cause-specific hospital admissions in Taipei, Taiwan.Occup Environ Med 65:104–111

Bell ML, Ebisu K, Peng RD, Samet JM, Dominici F (2009) Hospitaladmissions and chemical composition of fine particle airpollution. Am J Respir Crit Care Med 179:1115–1120

Bennett CM, McKendry IG, Kelly S, Denike K, Koch T (2008)Impact of the 1998 Gobi dust event on hospital admissions in theLower Fraser Valley, British Columbia. Sci Total Environ366:918–925

Breiman L, Friedman J, Stone CJ, Olshen R (1984) Classification andregression trees. Wadsworth International Group, Belmont, CA

Chan Y, McTainsh G, Leys J et al. (2005) Influence of the 23 October2002 dust storm on the air quality of four Australian cities. WaterAir Soil Pollut 164:329–348

Chen Y-S, Sheen P-C, Chen E-R, Liu Y-K, Wu T-N, Yang C-Y (2004)Effects of Asian dust storm events on daily mortality in Taipei,Taiwan. Environ Res 95:151–155

Chen L, Verrall K, Tong S (2005) Effects of Asian dust storm eventson daily hospital admissions for cardiovascular disease in Taipei,Taiwan. J Toxicol Environ Health A 68:1457–1464

Int J Biometeorol (2012) 56:719–726 725

Franklin M, Koutrakis P, Schwartz P (2008) The role of particlecomposition on the association between PM2.5 and mortality.Epidemiology 19:680–689

Gasparrini A, Armstrong B, Kenward MG (2010) Distributed lag non-linear models. Stat Med 29:2224–2234

Griffin DW (2007) Atmospheric movement of microorganisms inclouds of desert dust and implications for human health. ClinMicrobiol Rev 20:459–477

Gyan K, Henry W, Lacaille S, Laloo A, Lamsee-Ebanks C, McKay S,Antoine RM, Monteil MA (2005) African dust clouds areassociated with increased paediatric asthma accident and emer-gency admissions on the Caribbean island of Trinidad. Int JBiometeorol 49:371–376

Hashizume M, Ueda K, Nishiwaki Y, Michikawa T, Onozuka D (2010)Health effects of Asian dust events: a review of the literature.Nippon Eiseigaku Zasshi 65:413–421

Kwon H-J, Cho S-H, Chun Y, Lagarde F, Pershagen G (2002) Effectsof the Asian dust events on daily mortality in Seoul, Korea.Environ Res 90:1–5

Lall R, Ito K, Thurston D (2011) Distributed lag analyses of dailyhospital admissions and source-apportioned fine particle airpollution. Environ Health Perspect 119:455–460

McLeod A, Xu C (2010) bestglm: Best subset GLM. http://CRAN.R-project.org/package=bestglm

Middleton N, Yiallouros P, Kleanthous S, Kolokotroni O, Schwartz J,Dockery D, Demokritou P, Koutrakis P (2008) A 10-year time-series analysis of respiratory and cardiovascular morbidity inNicosia, Cyprus: the effect of short-term changes in air pollutionand dust storms. Environ Health 7:39

Perez L, Tobias A, Querol X, Kunzli N, Pey J, Alastuey A, Viana M,Valero N, Gonzalez-Cabre M, Sunyer J (2008) Coarse particlesfrom Saharan dust and daily mortality. Epidemiology 19:800–807

Plummer M (2010) JAGS Version 2.1.0 manual (available at http://transact.dl.sourceforge.net/project/mcmc-jags/Manuals/2.0/jags_user_manual.pdf). International Agency for Research onCancer

Polymenakou PN, Mandalakis M, Stephanou EG, Tselepides A (2007)Particle size distribution of airborne microorganisms and patho-gens during an intense African dust event in the EasternMediterranean. Environ Health Perspect 116:292–296

Queensland Health (2009) Looking after your health during a dust storm.http://access.health.qld.gov.au/hid/HealthConsumerInformation/HealthDecisionMaking/lookingAfterYourHealthDuringADustStorm_is.asp. Accessed 24 May 2011

Robert MA, VanBergen S, Kleeman MJ, Jakober CA (2007) Size andcomposition distributions of particulate matter emissions: part 1-light-duty gasoline vehicles. J Air Waste Manag Assoc 57:1414–1428

Rutherford S, Emma C, McTainsh G, Simpson R, Charles M (1999)Characteristics of rural dust events shown to impact on asthmaseverity in Brisbane, Australia. Int J Biometeorol 42:217–225

Schmaltz J (2009) Dust over eastern Australia. http://earthobservatory.nasa.gov/NaturalHazards/view.php?id=40274. Accessed 15 June2010

Schwartz J (2005) Who is sensitive to extremes of temperature? Acase-only analysis. Epidemiology 16:67–72

Schwartz J, Norris G, Larson T, Sheppard L, Claiborne C, Koenig J(1999) Episodes of high coarse particle concentrations are notassociated with increased mortality. Environ Health Perspect107:339–342

Spiegelhalter DJ, Best NG, Carlin BP, Avd L (2002) Bayesian measuresof model complexity and fit. JRSS Series B 64:583–639

Taylor DA (2002) Dust in the wind. Environ Health Perspect 110:A80–A87

Therneau TM, Atkinson B (2009) rpart: Recursive partitioning. http://CRAN.R-project.org/package=rpart

Welty L, Peng R, Zeger S, Dominici F (2009) Bayesian distributed lagmodels: estimating effects of particulate matter air pollution ondaily mortality. Biometrics 65:282–291

Yoo Y, Choung JT, Yu J, Kim doK, Koh YY (2008) Acute effects ofAsian dust events on respiratory symptoms and peak expiratoryflow in children with mild asthma. J Korean Med Sci 23:66–71

726 Int J Biometeorol (2012) 56:719–726