Identifying the high-risk areas and associated meteorologicalfactors of dengue transmission in Guangdong Province,China from 2005 to 2011

J. FAN1†, H. LIN2,3†, C. WANG1,4, L. BAI1, S. YANG1, C. CHU5, W. YANG6*AND Q. LIU1,5,6,7*1State Key Laboratory for Infectious Diseases Prevention and Control, National Institute for CommunicableDisease Control and Prevention, China CDC, Beijing, China2Guangdong Provincial Institute of Public Health, Guangzhou, China3Center for Disease Control and Prevention of Guangdong Province, Guangzhou, China4Department of Preventive Medicine, College of Basic Medical Sciences, Shandong University of TraditionalChinese Medicine, Jinan, China5Centre for Environment and Population Health, Griffith University, Brisbane, Australia6Key Laboratory of Surveillance and Early-Warning on Infectious Disease, Chinese Center for Disease Controland Prevention, Beijing, China7Collaborative Innovation Center for Diagnosis and Treatment of Infectious Diseases, Beijing, China

Received 17 December 2012; Final revision 26 May 2013; Accepted 31 May 2013

SUMMARY

We examined the spatial distribution pattern and meteorological drivers of dengue fever (DF) inGuangdong Province, China. Annual incidence of DF was calculated for each county between2005 and 2011 and the geographical distribution pattern of DF was examined using Moran’s Istatistic and excess risk maps. A time-stratified case-crossover study was used to investigate theshort-term relationship between DF and meteorological factors and the Southern OscillationIndex (SOI). High-epidemic DF areas were restricted to the Pearl River Delta region and theHan River Delta region, Moran’s I of DF distribution was significant from 2005 to 2006 andfrom 2009 to 2011. Daily vapour pressure, mean and minimum temperatures were associatedwith increased DF risk. Maximum temperature and SOI were negatively associated with DFtransmission. The risk of DF was non-randomly distributed in the counties in GuangdongProvince. Meteorological factors could be important predictors of DF transmission.

Key words: Analysis of data, dengue fever, infectious disease, public health, tropical diseases.

INTRODUCTION

Dengue fever (DF) has been known from antiquityand is still a major global public health concern.The disease, which can be caused by four distinctdengue viruses, is transmitted by Aedes mosquitoes[1] and is epidemic in more than 100 countries. DF

is now the fastest spreading mosquito-borne diseasein the world, and in the past 50 years, case incidencehas multiplied 30-fold, with about 30–54·7% of theworld’s population (2·05–3·74 billion) now living inareas where dengue viruses can be transmitted [2].

In 1978, a sudden outbreak of DF occurred inFoshan City, Guangdong Province after an absenceof 30 years. Since then, Hainan, Guangxi, Fujian,Yunnan and Zhejiang have all reported DF outbreaks[3]. Guangdong Province has had the highest DF inci-dence in mainland China during the past decades,accounting for more than 65% of all cases in the

* Author for correspondence: ProfessorW.YangorProfessorQ.Liu,155 Changbai Road, Changping District, Beijing, China, 102206.(Email: yangwz@chinacdc.cn) [W. Yang](Email: liuqiyong@icdc.cn) [Q. Liu]† These authors contributed equally to this work.

Epidemiol. Infect., Page 1 of 10. © Cambridge University Press 2013doi:10.1017/S0950268813001519

country [4]. However, the incidences in the countieswithin Guangdong is not homogenous [4]. A betterunderstanding of the spatial pattern of DF wouldhelp to identify high-risk areas and might assist inthe development of DF control and prevention strat-egies in the province.

Meteorological factors can influence the lifespan,breeding, survival, and population dynamics of thevector mosquitoes both directly and/or indirectly.Studies have shown that meteorological factors,including rainfall, temperature, and relative humidity,are important environmental factors that could influ-ence the transmission dynamics of DF. El Niño/Southern Oscillation (ENSO) is a significant indexof quasi-periodic climate pattern, which affects mostcountries in the Pacific and Indian Oceans. TheSouthern Oscillation Index (SOI), the atmosphericcomponent of El Niño, is defined as the normalizeddifference in atmospheric pressure between Darwin(Australia) and Tahiti (French Polynesia). InGuangdong Province, negative SOI values are associ-ated with El Niño conditions (dry and warm), andpositive SOI values with La Niña conditions (wetand cool). Other studies have linked the SOI to therisk of DF transmission [5]. The use of geographicalinformation systems (GIS) with spatial statistics,including spatial smoothing and autocorrelationanalysis, has been increasingly applied to infectiousdiseases, especially zoonoses and vector-borne dis-eases over recent years [6–9]. In this study, weexplored the spatial pattern of DF transmission inGuangdong Province from 2005 to 2011 with theaim of identifying the high-risk areas of DF trans-mission, to provide a basis for dengue control andprevention measures. We further examined the short-term effect of meteorological factors and SOI on DFepidemics to assess their ability to predict dengueoutbreaks.

MATERIALS AND METHODS

Setting

Guangdong Province is located in southeasternChina (20° 12′ N, 25° 31′ N, 109° 45′ E, 109° 45′ E)as showed in Figure 1. The province comprises 21prefecture-level cities and 122 counties with a totalland area of 179800 km2 and a population of about104303132 according to the national census in 2010.Guangdong has a subtropical climate influenced bythe Asian monsoon, with an annual average

temperature of 19–24 °C. The summer is wet withtemperatures between 28 and 29 °C in July, whilethe winter is usually mild, dry and sunny with tem-peratures between 16 and 19 °C in January. Theaverage annual rainfall in the study area was between1300 and 2100 mm and the humidity was between69% and 83%.

Materials

Records of DF cases in Guangdong Province be-tween 2005 and 2011 were obtained from the Nation-al Notifiable Disease Surveillance System. Bloodsamples from all DF patients’ were collected in hospi-tals and sent for diagnostic confirmation to the labora-tory of Guangdong Provincial Center for DiseaseControl and Prevention. Demographic data wereobtained from the Guangdong Statistical Yearbook.

To conduct a GIS-based analysis on the spatial dis-tribution of DF cases, the county-level polygon mapat 1:250000 scale was obtained from Data SharingInfrastructure of Earth System Science (www.geodata.cn), on which a county-level layer containinginformation regarding latitudes and longitudes ofcentral points of each county was created. All DFcases were geocoded and matched to the county-levellayers of polygon and point by administrative codesusing ArcGIS 10.0 software (ESRI, USA).

Daily meteorological data, including mean tem-perature, maximum temperature, minimum tempera-ture, relative humidity, rainfall, atmospheric pressureand SOI from 2005 to 2011, were retrieved from theChina Meteorological Data Sharing Service System(http://cdc.cma.gov.cn/index.jsp) and the website ofthe Australian Bureau of Meteorology (http://www.bom.gov.au/).

GIS mapping and spatial analysis

To manage variations of incidence in small popu-lations and areas, the annual average incidences ofDF per 100000 in each county over the 7-year periodwere calculated, and spatial rate smoothing wasimplemented. Based on the overall annual averageincidence (2·5/100000), all counties were groupedinto four categories [9]: non-epidemic areas, lowepidemic areas with annual average incidence between0 and 2/100000 persons, medium epidemic areas withan incidence between 2 and 4/100000 persons, andhigh epidemic areas with an incidence >4/100000 per-sons. The counties were coloir-coded according to the

2 J. Fan and others

epidemic level. To assess the risk of DF in eachcounty, an excess risk map was produced. The excessrisk was defined as the ratio of the observed incidenceat each county divided by the average incidence. Theaverage incidence of the whole province was calcu-lated as the number of cases in the province dividedby the total population at risk [10].

Global spatial autocorrelation analysis was per-formed in GeoDa095i software. Moran’s I, spatialautocorrelation statistic was calculated and visualizedin the form of a Moran scatter plot. First, acontiguity-based spatial weight was constructed foreach county by creating a rook contiguity weightsfile. Second, Moran’s scatter plot was produced witha spatial lag of incidence on the vertical axis and astandardized incidence on the horizontal axis. Thenumber of permutation tests was set to 999 and thepseudo-significance level was set at 0·01.

Time-stratified case-crossover analysis

A time-stratified case-crossover model was utilizedto examine the short-term association of daily DFwith daily meteorological factors (including tempera-ture, relative humidity, precipitation, and atmosphericpressure) and SOI. This design compares the exposure

in the case period when events occurred with theexposures in nearby control periods to determine thedifferences in exposure that might explain the differ-ences in the daily number of DF cases [11, 12]. Inthe present study, the case and control periods werematched by day of the week in order to control forany weekly patterns of DF cases or weather variables.Control periods were compared with cases usingthe time-stratified method with 28 days strata [13].Such a time-stratified case-crossover study designensured that each case day had three matchingcontrol days to avoid an overlap bias in the riskestimation [14]. For this study, the first stratum was1–28 January 2005, and the second was 29 Januaryto 25 February 2005; for a DF case on 28January 2005, the control days were 7, 14 and 21January 2005. The case-crossover method inherentlycontrols for any long-term trends and seasonalpatterns of disease and explanatory variables, andindividual characteristics such as age, sex, and edu-cational level [15]. Studies have demonstrated thatthe time-stratified case-crossover analysis providesunbiased estimates in the presence of strong seasonalconfounding [16, 17]. Here, public holidays (as abinary variable) were included in all models as apotential confounding factor. We did univariate and

Guangdong

Fig. 1 [colour online]. Location of Guangdong Province in China.

Meteorological factors and dengue transmission 3

multivariate models, only two weather variables wereincluded in one model. We also performed analysisto include all the weather variables in one model. Weestimated the linear effect of the meteorological factorsaccording to different lag structures including the cur-rent day (lag 0) up to 3 days prior (lag 3). Results ofthe analysis were expressed as excess risk (ER) ofDF based on a one-unit increase in the weather vari-ables. ER was calculated using the formula: ER=(OR−1) × 100%, where OR (odds ratio) was obtainedthrough the conditional logistic regression model. Allstatistical tests were two-sided. The ‘season’ packageof R version 2.14.1 (R Foundation, Austria) wasused to fit the time-stratified case-crossover model.

RESULTS

Spatial distribution of DF in Guangdong Province

There were 1709 DF cases reported in GuangdongProvince during 2005–2011. The time series of dailyDF cases and key meteorological variables aredepicted in Supplementary Figure S1 (availableonline). There were seasonal patterns in these vari-ables. Annual average incidences at the county levelranged from 0 to 8·07/100000. High-epidemic DFareas were mainly restricted to the Pearl River Delta

region and Han River Delta region, the DF incidencesof the counties in these two areas were 2–4/100000persons or more than 4/100000 persons (Fig. 2).

Figure 3 shows the risk map of DF, where the ER isdefined as a ratio of the observed number divided bythe expected number of DF cases. Counties in lightgrey had lower incidences than expected, with riskratios <1. In contrast, counties in dark grey to blackshading had incidences higher than expected or riskratios >1.

Spatial autocorrelation of DF in Guangdong Province

A Moran scatter plot was created and a significanceassessment through a permutation test was im-plemented by global spatial autocorrelation analysisfor annualized average incidence of DF (Fig. 4). Thenumber listed on the top of the graph (0·2433) isMoran’s I statistic (Fig. 4a), and a histogram wasgenerated by performing the significance assessmentof Moran’s I statistic (Fig. 4b). The brown part isthe reference distribution and the yellow bar is thestatistic (Fig. 4b). In addition the number of permu-tations (n=999) and the preudo-significance level(0·002) were also listed in the upper left corner, aswell as the value of the statistic (0·2433), its expected

Incidence (1/100 000)

0

0–2

2–4

>4

0 37·5 75 150 225 300km N

Fig. 2. Annual average incidence of dengue fever in the counties in Guangdong Province, China, 2005–2011.

4 J. Fan and others

mean [E(I)=0·0083], and the mean and standard devi-ation of the empirical distribution were −0·0057 and0·0494 (as shown at the bottom of Fig. 4b). The stat-istic was significant for Moran’s I at a level of 0·01.Spatial autocorrelation analysis for annualized inci-dence of DF in Guangdong Province from 2005 to2011 showed that Moran’s I was significant from

2005 to 2006 and from 2009 to 2011, but not signifi-cant in 2007 and 2008 (Table 1).

The association between DF and meteorological factors

Figure 5 illustrates the results of the time-stratifiedcase-crossover analyses. In the figure, w/Tmin on the

Excess risk

<0·5(101)

0·5–1·0(2)

1·0–2·0(2)

2·0–4·0(4)

>4(10)

0 37·5 75 150 225 300km

N

Fig. 3. Risk map of dengue fever at the county level in Guangdong Province, China, 2005–2011.

(a) (b)

5

0

–5

–10 –5 5 100

W_A

VE

_RAT

E

AVE_RATE

Moran’s I = 0·2433

permutation: 999p-value: 0·0020

(0.2433) E(I ) = 0·0083 Mean –0·0057 Sd 0·0494

Fig. 4. Global spatial autocorrelation analysis for annualized average incidence of dengue fever (DF) in GuangdongProvince, China, 2005–2011. (a) The Moran’s scatter plot for annualized average incidence of DF. (b) Histogram forsignificance assessment of Moran’s I.

Meteorological factors and dengue transmission 5

y-axis means that the minimum temperature was con-trolled for in the model. However, when one weathervariable was controlled in the model of the samevariable, it was the result of univariate model forthat variable. It was found that atmospheric pressure(at lag 0–3 days) was associated with increased DFrisk in both single- and two-factor analyses, and theER appeared to be larger when other meteorologicalfactors were controlled for. Daily mean and minimumtemperatures have delayed effects on DF with lags of3 and 2 days, respectively. Daily maximum tempera-ture was associated with decreased DF risk at a lagof 0–2 days, and this effect persisted when othermeteorological factors were included. The SOI wasfound to be associated with decreased DF at lags of0–1 days. Corresponding ERs were 7·68 [95% confi-dence interval (CI) 5·58–9·83] per 1 hPa increase inatmospheric pressure on the current day, 6·99 (95%CI 3·38–10·74) for a 1 °C increase in daily mean tem-perature after controlling for pressure at a lag of 3days, 0·83 (95% CI 0·50–1·17) for a 1 °C degreeincrease in daily minimum temperature after control-ling for pressure at a lag of 3 days, −0·50 (95% CI−0·94 to −0·06) for a 1 °C increase in daily maximumtemperature after controlling for mean temperatureon the current day, and −0·99 (95% CI −1·46 to−0·52) for a one-unit increase in SOI on the currentday, respectively. No significant association wasobserved for relative humidity and rainfall in theresults. When all the weather variables were includedin one model, a similar result was produced.

In the sensitivity analysis, we conducted a time-stratified case-crossover study for the high-riskareas of the Pearl River Delta and Han River Deltaregions, separately, which yielded similar results withthat of the whole province (Supplementary Figs S2and S3).

DISCUSSION

In this study, both exploratory spatial analysisand case-crossover analysis of DF and meteorologicalfactors were conducted in Guangdong Province. Ouranalyses demonstrated that spatial distribution ofDF in the counties in Guangdong Province was non-random, and the high-epidemic areas were mainlyrestricted to the Pearl River Delta and Han RiverDelta regions. From the analysis of the associationbetween DF and meteorological factors, we foundthat atmospheric pressure (at lags of 0–3 days)and daily mean and minimum temperatures with2- to 3-day lags were statistically significantly andpositively associated with DF, while daily maximumtemperature (at lags of 0–2 days) and SOI (at lagsof 0–2 days) were negatively associated with DFepidemics.

Since 2005, the high-epidemic areas of DF havebeen mainly restricted to the Pearl River Delta andHan River Delta regions, which is consistent with pre-vious studies [4, 10]. Efficient allocation of healthresources in DF control and prevention programmesrequires such accurate information on the geographi-cal pattern of DF transmission. Thus, the results ofthis study may assist with the allocation of healthresources more effectively and efficiently. Accordingto the findings of this study, we suggest that moredengue control resources should be placed in thePearl River Delta and Han River Delta regions. Weidentified areas of increased risk that are again locatedin the Pearl River Delta and Han River Delta regions(Fig. 3). Yi and colleagues [18] analysed the distri-bution pattern of Aedes mosquitoes in GuangdongProvince, and found that both the Pearl River Deltaand Han River Delta had higher Aedes mosquitodensity than other areas, which suggests that thehigh risk of DF in these two areas must be partlyattributed to the high density of Aedes mosquitoes.Additionally, Guangdong is also the major sourceof labourers in South East Asia. As a result, manyoverseas Chinese communities had their origins inGuangdong. The residents in these regions keepclose connections with South East Asian countriesthat are in the dengue-endemic regions. Importeddengue cases were believed to have been the cause ofthe outbreaks according to some studies [4]. Since2000, frequent migration and rapid urbanization haveemerged in Guangdong Province, especially in thePearl River Delta and Han River Delta regions [19],which are the core economic region in the Province.

Table 1. Spatial autocorrelation analysis of annualaverage incidences of dengue fever in GuangdongProvince, 2005–2011

Year Moran’s I E[I] Mean S.D P value

2005 0·3703 –0·0083 –0·0060 0·0518 0·00102006 0·1500 –0·0083 –0·0071 0·0371 0·00502007 –0·0043 –0·0083 –0·0081 0·0413 0·81402008 –0·0004 –0·0083 –0·0057 0·0509 0·78302009 0·2727 –0·0083 –0·0070 0·0556 0·00102010 0·4476 –0·0083 –0·0083 0·0496 0·00102011 0·2544 –0·0083 –0·0070 0·0359 0·0010

6 J. Fan and others

Gubler [20] predicted that if global trends of unpre-cedented population growth, continued globalizationand unprecedented urbanization continue as projected,there will be continued increases in the severity,frequency, geographical distribution and magnitudeof DF epidemics in the future. All these factorsmay contribute to the increased DF risk in the PearlRiver Delta and Han River Delta regions. Preventingor reducing dengue virus transmission depends criti-cally on the control of the mosquito vectors or inter-ruption of human–mosquito contacts. At present,integrated vector management (IVM) is the strategicapproach to mosquito control and dengue preventionpromoted by the WHO [21], but an efficientlong-term sustainable control programme is needed,which should combine a top-down and a bottom-upapproach, including IVM, environmental improve-ments, climate and vector surveillance-based early

warning system, planned population growth and con-trolled urbanization.

Spatial autocorrelation analysis showed that therewas non-random spatial distribution of DF inGuangdong Province during the period 2005–2011.More specifically, by calculating Moran’s I annually,our study showed that the distribution of DF from2005 to 2006 and from 2009 to 2011 was non-randomgeographically, but in 2007 and 2008 distribution ofcases appeared to be random. Fluctuation in popu-lation immunity plays a major role in denguetransmission, because the disease does not induce per-manent immunity. The non-random and clustered dis-tribution in Guangdong from 2005 to 2011 might bedue to population immunity. Two major outbreaksin Guangdong Province occurred in 2002 and 2006,with more than 1000 cases each year, and the popu-lation immunity may have been maintained at a

w/TminTmin

w/Tmax

Tmax

w/Tmean

Tmean

w/Tmean

w/SOI

w/SOI

w/Humidity

Humidity Rain SOI

Lag0Lag1

Lag2Lag3

% increase (ER% with 95% CI) in dengue fever for per unit increase meteorological factors

w/Humidity

w/Rain

w/Rain

w/Pressure

Pressure

w/Pressure

w/Tminw/Tmax

w/Tmeanw/SOI

w/Humidityw/Rain

w/Pressure

w/Tminw/Tmax

w/Tmeanw/SOI

w/Humidityw/Rain

w/Pressure

0 2 4 6 8 10 –5 0 5 10–0

.5 0.0

0.5

1.0

–0.8

–0.6

–0.4

–0.2 0.0

0.2

0.4

–1.5

–1.0

–0.5 0.0

0.5

–0.0

2

0.08

–1.5

–1.0

–0.5 0.0

0.06

0.04

0.02

0.00

–0.0

4

w/Tminw/Tmax

Fig. 5. The association between dengue fever and meteorological factors in Guangdong Province, China, 2005–2011. Onthe y-axis, w/Tmin means that minimum temperature was controlled for. Lag on the right y-axis was the effect estimationon different lag days.

Meteorological factors and dengue transmission 7

high level after the outbreaks. The distribution wouldtherefore appear to occur randomly in the followingyears. Some studies have shown that both the inci-dence and geographical distribution of DF wererelated to environmental factors and climatic factors[3, 19]. In the future it will be helpful to investigatethe changes in these factors when DF distribution isnon-random and random. This study provides base-line information on the geographical distribution ofDF in Guangdong Province.

Previous studies have reported that meteorologicalfactors are associated with dengue transmission[22, 23]. The results of this study illustrated thatvapour pressure, which is a measure of the actualwater vapour content of the atmosphere, has astrong positive correlation with the incidence of DFin Guangdong Province. This finding was in accord-ance with studies in other areas. For example, onthe Caribbean island of Barbados a positive relation-ship was found between vapour pressure and DF inci-dence at a lag of 6 weeks [24]. Hales and co-workersanalysed climatic associations of global dengueoutbreaks between 1975 and 1996, and found thatthe annual average vapour pressure was the mostimportant individual predictor of DF distribution[25]. Higher atmospheric vapour pressure may favourthe vector mosquitoes; under these conditions infectedmosquitoes would feed more frequently or feed onmultiple individuals, thereby increasing the trans-mission of the disease [24].

Daily mean and minimum temperatures were posi-tively associated with DF incidence, while daily maxi-mum temperature was inversely associated with DFincidence. Temperature has been shown to influencemany vector-borne diseases [6, 26], affecting boththe survival time and habitats of the vector, and alsotheir replication, maturation, and infective periods.Higher temperature was found to shorten the extrinsicincubation period and viral development rate, andincrease the amount of infectious mosquitoes in theenvironment [27]. Correlation studies between cli-matic factors and the distribution of A. albopictus[28] showed that they breed in those areas where theannual mean temperature is above 11·8 °C, the meantemperature in January was above 5·8 °C, and theannual precipitation was above 500mm [28]. Whenthe atmospheric temperature is too low, dengue virusdevelopment slows down and the mosquito bitingrate declines. According to a study by Hawley, theoptimum temperature limits for A. albopictus rangefrom 20 to 27 °C, with a maximum temperature

around 30 °C [29]. In laboratory studies survivalrate started to decrease when the temperature roseabove 28 °C [30]. Guangdong Province has a humidsubtropical climate influenced mainly by the Asianmonsoon. Summers are wet with high temperatureand a high humidity index. Winters are mild, dryand sunny. The annual mean temperature rangesfrom 19 to 24 °C; the annual maximum temperatureranges from 34·6 to 39·0 °C. Therefore, as temperatureincreases to the optimum temperature for A. albopic-tus, particular dengue prevention measures should beimplemented. When the temperatures rise above theoptimum levels for the development and activity ofadult mosquitoes, their survival will be adverselyaffected and risk of the disease is expected to decrease.

We did not find any significant relationship be-tween humidity or rainfall and DF incidence at lagsof 0–3 days in the present study, which was consist-ent with the findings of previous studies [22]. Kovatshas suggested that rainfall could affect the breedingof mosquitoes but this might be less important inurban areas, as the vectors often breed in small con-tainers, such as plant pots, which may contain waterin the absence of rain [31]. The distribution of DFin Guangdong Province was mainly concentrated inthe developed urban areas, where public health facili-ties and drainage system were in good condition andrainfall had little impact on increasing breeding sitesfor mosquitoes. Optimal humidity conditions canincrease mosquito survival significantly, although thelower limit of annual relative humidity (69%) was gen-erally in winter, when the temperature was too low totransmit dengue virus. However, in the dengue epi-demic season, the relative humidity was high enoughto facilitate dengue transmission in GuangdongProvince.

It has been suggested that the geographical expan-sion of DF throughout the world may be partlyattributable to global climate change. The ENSO isa systematic pattern of global climate variability,affecting most countries in the Pacific and IndianOceans [32]. Our study found that ENSO was associ-ated with dengue transmission risk in GuangdongProvince. Furthermore, ENSO has been found to berelated to various health outcomes, including water-borne infectious diseases, vector-borne disease andnatural disaster-related deaths in other areas [33, 34].In Thailand, which does not have a strong ENSOsignal, there was no correlation [35]. Alexandre [36]examined the relationship between dengue epidemicsand ENSO across the Indonesian archipelago and

8 J. Fan and others

northern South America [36], and found that therewas a statistically significant correlation at the 95%confidence level between El Niño and dengue epi-demics in French Guiana and Indonesia and at the90% confidence level in Colombia and Surinam [36].In Indonesia, which has a strong El Niño effect caus-ing drought, dengue epidemics consistently occurredin the year after El Niño [31]. Our study suggeststhat warmer temperatures (negative SOI values inGuangdong Province) were correlated with increasedDF incidence. Hu stated the SOI was a regionallyaggregated measure of climate variability, whichassociated with dengue risk [23]. Further studies areneeded to improve the understanding of the mech-anisms underlying these associations, and to assessthe extent to which relationships between ENSO anddengue may be useful as a basis for an early warningsystem.

Our findings illustrate that spatial and temporalanalysis of DF is useful for improving our understand-ing of the characteristics of this disease, and for gener-ation of new hypotheses. In the meantime, furtherepidemiological and ecological studies are requiredto investigate the mechanisms of the observed associ-ation. Climate is not the only factor affecting DFtransmission, environmental and socioeconomicfactors, host–pathogen interactions and populationimmunological factors can all influence the dynamicsof virus transmission [37]. Although climate factorscould directly influence the breeding and survival ofthe mosquito vectors and thereby affect their abun-dance and distribution. Climate is therefore an impor-tant predictor of vector-borne disease epidemics, andregional climate changes may drive an increase intransmission in larger, more populated area.

There are certain limitations to this study. First, theuse of passive surveillance data might have introducedmeasurement and information biases. For example,including individuals who were infected with DF butexperienced only sub-clinical symptoms and did notseek medical treatment might have led to under-reporting [38]. Second, the incidence of DF has beenrising worldwide in recent decades. This is probablyrelated to global trends such as population growth,urbanization, weakening of public health infrastruc-ture, and climate change not measured here. Third,more information on the community- and individual-level risk factors (e.g. mosquito population dynamic,health education frequency, individual behavioursand awareness) is required because these might beconfounding factors between socio-ecological charac-

teristics and DF transmission [39]. Therefore, futureattempts to study the relationship between dengueand meteorological factors should consider collect-ing and analysing more data on entomological andeco-health variables. Finally, some studies have sug-gested that the time-stratified case-crossover studydesign could suffer severe bias from residual seasonal-ity [14], in this study, we did perform a sensitivityanalysis using a generalized additive model, whichyields a generally similar estimation, suggesting thatthe seasonal residual bias in this analysis was not aserious drawback.

The findings of this study may have significantimplications for the development of strategic plansto control and prevent DF outbreaks in GuangdongProvince.

ACKNOWLEDGEMENTS

This study was supported by the National BasicResearch Programme of China (973 Programme)(Grant no.: 2012CB955504), the Special ResearchProgramme for Health (Grant no.: 201202006) andthe National Science Foundation of China (NSFC)(Grant no.: 81273139, 30972563).

DECLARATION OF INTEREST

None.

SUPPLEMENTARY MATERIAL

For supplementary material accompanying this papervisit http://dx.doi.org/10.1017/S0950268813001519.

REFERENCES

1. Gubler DJ. Dengue and dengue hemorrhagic fever.Clinical Microbiology Reviews 1998; 11: 480–496.

2. Brady OJ, et al. Refining the global spatial limits ofdengue virus transmission by evidence-based consensus.PLoS Neglected Tropical Diseases 2012; 6: e1760.

3. Lu L, et al. Time series analysis of dengue fever andweather in Guangzhou, China. BMC Public Health2009; 9: 395.

4. Liu C, et al. Spatial analysis of fengue fever inGuangdong Province, China, 2001–2006. Asia-PacificJournal of Public Health. Published online: 22 January2013. doi:10.1177/1010539512472356.

5. Descloux E, et al. Climate-based models for understand-ing and forecasting dengue epidemics. PLoS NeglectedTropical Diseases 2012; 6: e1470.

Meteorological factors and dengue transmission 9

6. Lin HL, et al. Analysis of the geographic distributionof HFRS in Liaoning Province between 2000 and2005. BMC Public Health 2007; 7: 207.

7. Lin HL, et al. Spatial and temporal distribution offalciparum malaria in China. Malaria Journal 2009;8: 130.

8. Lin HL, et al. Lung cancer mortality among women inXuan Wei, China: a comparison of spatial clusteringdetection methods. Asia-Pacific Journal of PublicHealth. Published online: 1 May 2012. doi:10.1177/1010539512444778.

9. Lu L, Lin HL, Liu QY. Risk map for dengue fever out-breaks based on meteorological factors. Advances InClimate Change Research 2010; 6: 254–258.

10. Yi BT, et al. Study on the distribution of denguefever and vector in Guangdong province combinedapplication of SRS/GIS/PCA. Chinese Journal ofDisease Control & Prevention 2003; 7: 509–514.

11. Maclure M. The case-crossover design: a methodfor studying transient effects on the risk of acuteevents. American Journal of Epidemiology 1991; 133:144–153.

12. Maclure M. Taxonomic axes of epidemiologic studydesigns: a refutationist perspective. Journal of ClinicalEpidemiology 1991; 44: 1045–1053.

13. Guo Y, et al. Gaseous air pollution and emergencyhospital visits for hypertension in Beijing, China: a time-stratified case-crossover study. Environmental Health2010; 9: 57–63.

14. Whitaker H, Hocine M, Farrington C. On case-crossover methods for environmental time series data.Environmetrics 2007; 18: 157–171.

15. Bateson T, Schwartz J. Who is sensitive to the effects ofparticulate air pollution on mortality? A case-crossoveranalysis of effect modifiers. Epidemiology 2004; 15:143–149.

16. Forastiere F, et al. A case-crossover analysis ofout-of-hospital coronary deaths and air pollution inRome, Italy. American Journal of Respiratory andCritical Care Medicine 2005; 172: 1549–1555.

17. Guo Y, et al. The short-term effect of air pollutionon cardiovascular mortality in Tianjin, China: Com-parison of time series and case-crossover analyses.Science of The Total Environment 2011; 409: 300–306.

18. Yi BT, et al. Relationship of dengue fever epidemicto Aedes density changed by climate factors in Guang-dong Province. Journal of Hygiene Research 2003; 2:152–154.

19. Wu PC, et al. Higher temperature and urbanizationaffect the spatial patterns of dengue fever transmissionin subtropical Taiwan. Science of the Total Environment2009; 407: 2224–2233.

20. Gubler DJ. Dengue, urbanization and globalization:The unholy trinity of the 21st century. TropicalMedicine & International Health 2011; 39: 3–11.

21. World Health Organization. Global strategic frameworkfor integrated vector management. (http://whqlibdoc.who.int/hq/2004/WHO_CDS_CPE_PVC_2004_10.pdf.)Accessed October 2004.

22. Lu L, et al. Time series analysis of dengue fever andweather in Guangzhou, China. BMC Public Health2009; 9: 395.

23. Hu WB, et al. Dengue fever and El Nino/SouthernOscillation in Queensland, Australia: a time seriespredictive model. Occupational and EnvironmentalMedicine 2010; 67: 307–311.

24. Depradine C, Lovell E. Climatological variables andthe incidence of dengue fever in Barbados. InternationalJournal of Environmental Health Research 2004; 14:429–441.

25. Hales S, et al. Potential effect of population and climatechanges on global distribution of dengue fever: anempirical model. Lancet 2002; 360: 830–834.

26. Lin HL, et al. Time series analysis of Japanese encepha-litis and weather in Linyi City, China. InternationalJournal of Public Health 2012; 57: 289–296.

27. Watts DM, et al. Effect of temperature on the vectorefficiency of Aedes aegypti for dengue 2 virus.American Journal of Tropical Medicine and Hygiene1987; 36: 143–152.

28. Wu F, et al. Distribution of Aedes albopictus (Diptera:Culicidae) in northwestern China. Vector-borne andZoonotic Diseases 2011; 11: 1181–1186.

29. Hawley WA. The biology of Aedes albopictus. Journalof the American Mosquito Control Association 1988; 1:1–39.

30. Delatte H, et al. Influence of temperature on immaturedevelopment, survival, longevity, fecundity, and gono-trophic cycles ofAedes albopictus, vector of chikungunyaand dengue in the Indian Ocean. Journal of MedicalEntomology 2009; 46: 33–41.

31. Kovats RS. El Nino and human health. Bulletin of theWorld Health Organization 2000; 78: 1127–1135.

32. Kovats RS, et al. El Nino and health. Lancet 2003; 362:1481–1489.

33. Hu WB, et al. Dengue fever and El Nino/SouthernOscillation in Queensland, Australia: a time seriespredictive model.Occupational and EnvironmentalMedi-cine 2010; 67: 307–311.

34. Lin HL, et al. Short term effects of El Nino-SouthernOscillation on hand, foot, and mouth disease inShenzhen, China. PLoS One. Published online:27 April 2013. doi:10.1371/journal.pone.0065585.

35. Hay SI, et al. Etiology of interepidemic periods ofmosquito-borne disease. Proceedings of the NationalAcademy of Sciences USA 2000; 97: 9335–9339.

36. Alexandre S, et al. Dengue epidemics and the El Ninosouthern oscillation. Climate Research 2001; 19: 35–43.

37. WorldHealthOrganization.Dengue: Guidelines forDiag-nosis, Treatment, Prevention and Control – New Edition.Geneva, Switzerland: WHO Press, 2009, pp. 111–136.

38. Mackenzie JS, et al. Arboviruses in the Australianregion, 1990 to 1998. Communicable Diseases Intelli-gence 1998; 22: 93–100.

39. Hu WB, et al. Spatial patterns and socioecologicaldrivers of dengue fever transmission in Queensland,Australia. Environmental Health Perspectives 2012; 120:260–266.

10 J. Fan and others