The role of climate variability and change in the transmission dynamics and geographic distribution of dengue

Introduction

Dengue is the most common vector-borne viral disease worldwide, and is ranked among the most important infectious diseases by the World Health Organization (WHO). ¹ Dengue poses a major challenge to international public health (World Health Assembly resolution, 2002). The disease burden and geographic range of dengue have expanded over the past 50 years, from approximately 15,000 cases reported annually to WHO from fewer than 10 countries during the 1960s to close to 1 million cases annually across more than 60 countries in 2000–2005. ² Recent estimates suggest that transmission occurs in up to 124 countries with at least 35 million symptomatic cases per year and around 20,000 deaths, and that approximately 3.5 billion people, or 55% of the world's population, live in countries at risk for dengue. ³ In Vietnam, dengue was first reported in 1959 and since then the number of reported cases has increased steadily, with major outbreaks in 1987 and 1998 throughout all provinces. ^4–7 Figure 1 shows the incidence of dengue and the case fatality rate in southern Vietnam since 1996. In endemic areas of southeast Asia, dengue is mainly a childhood disease and children experience an annual exposure risk of ∼10%. ^8,9 An upwards shift in the median age of dengue cases has recently been reported from Thailand; ¹⁰ however, it is not yet clear whether this is replicated in other endemic settings.

OPEN IN VIEWER

Figure 1

Dengue incidence and mortality in southern Vietnam, 1996–2009. Bars show the incidence of dengue (dengue fever and dengue hemorrhagic fever) per 100,000 population in the southern 20 provinces of Vietnam between 1996 and 2009. The line shows the mortality rate among reported dengue cases over the same period. Data represent cases reported to the national dengue surveillance control program and includes hospitalized cases only. These data are reproduced with the permission of the Pasteur Institute, Ho Chi Minh City

Humans are the primary host of dengue virus (DENV) and transmission of DENV takes place through the bite of the principal mosquito vector Aedes aegypti, and to a lesser degree, Aedes albopictus. Human infection by DENV can lead to a broad spectrum of outcomes. The majority of DENV infections are inapparent (50–90%) ^11,12 ; clinical manifestations range from dengue fever (DF), a self-limiting syndrome characterized by fever, malaise, arthralgia and myalgia, to severe disease that has been classically classified into dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) and is characterized by vascular permeability, coagulopathy and, in the case of DSS, hypovolemic shock. ^13,14 Mortality rates vary from less than 1% to 10% in different settings, and depend critically on access to health-care facilities and the degree of clinical experience in hematological monitoring and management of vascular leakage. ¹⁵

The geographical distribution of dengue largely reflects the distribution of the principal vector A. aegypti. The expansion in the range and magnitude of the dengue disease burden over the past half-century has paralleled rapid population growth and urbanization, particularly in endemic areas of southeast Asia, and unprecedented increases in global human mobility and trade (reviewed in ref. ¹⁶ ). All of these factors are conducive to an intensification of transmission in early endemic settings and the export of DENVs and their vectors from these early foci in southeast Asia and parts of Central America into new areas, as occurred in the decades following the 1950s. In particular, many countries in the Americas experienced a re-emergence of dengue during the 1970s and 1980s after prior success in controlling A. aegypti through concerted campaigns to eliminate yellow fever, which shares the same primary vector. ¹⁷

The rapid and continued expansion of dengue represents an increasing challenge to national and regional health authorities, both in endemic settings and in currently unaffected or newly affected areas. The ability to accurately estimate the current disease burden and to predict future trends is key to the planning of effective prevention and control interventions.

Climate is an important determinant of the spatial and temporal distribution of dengue and other vector-borne diseases. ^18–22 It seems logical therefore that long-term changes in global climate, for which there is considerable evidence, may have significant effects on the distribution of vector-borne diseases such as dengue. A substantial scientific literature has examined this question over the last two decades; however, there is still inconclusive evidence on what influence, if any, sustained global climate change will have on dengue. The aim of this paper is to review the current knowledge of the relationship between climate variability or climate change and dengue transmission and spread, and to discuss the tools that are being used to explore this relationship. In particular, we aim to highlight the distinction between models that explore climate variability with respect to dengue transmission, and those that aim to model the effects of climate change, as the required input parameters as well as intended outputs of these models are quite distinct. We consider the variable endpoints for models of dengue and climate, which tend to describe and predict either the geographic range or the intensity of dengue transmission, and the differences in how climate and environmental parameters are included to these ends. We further discuss the ways in which non-climatic determinants of dengue transmission may mediate or constrain the predicted effects of climate variability or change on dengue transmission, and the importance where possible of including these sociodemographic–economic factors into models, or at least giving due consideration to their potential influence on model outputs.

Effects of climate on dengue transmission

Climatic conditions play a key role in the biology and ecology of mosquito vectors and the viruses they transmit, and consequently also exert a strong influence on the risk of dengue transmission. ^18,23–27 Higher temperatures increase the rate of larval development and therefore emergence of adult vectors, increase the vector biting rate and reduce the time required for virus replication within the vector, known as the extrinsic incubation period, ^28,29 meaning vectors are infectious earlier and bite more frequently. Higher temperatures may reduce vector survival time, which may offset the positive effect on vector abundance to some degree. Numerous studies have linked temperature to A. aegypti abundance and dengue incidence rates. ^30–33 These studies have used various statistical approaches considering different temperature parameters (e.g. mean, maximum and minimum temperatures).

Aedes vectors breed predominantly in clean water-holding containers in close proximity to human dwellings. ^29,34,35 Variability in rainfall is likely to affect the availability of these vector breeding sites, and therefore influence vector abundance; ³⁶ however, the magnitude and direction of this effect is uncertain and may differ between settings depending on the nature of local vector breeding sites, especially whether they are predominantly outdoor rain-filled objects or indoor containers that are filled manually. ³⁷ Humidity, or vapor pressure, is governed by a combination of rainfall and temperature and influences the lifespan of the mosquito and therefore the potential for transmission of the virus. ²⁹ Annual average vapor pressure has been suggested by some authors to be the most important climatic predictor of global dengue occurrence. ³⁸

Temperature, rainfall and humidity are therefore important determinants of the geographic limits within which dengue transmission can be expected to be sustained, primarily through their effects on the survival and proliferation of the Aedes vector. Furthermore, within areas where minimum thresholds of these climate parameters are sufficient to sustain DENV transmission, seasonal fluctuations in these parameters will be important determinants of the duration and potentially the intensity of transmission.

Non-climate determinants of dengue transmission and disease severity

Climatic conditions favorable for Aedes vectors are a necessary but insufficient criterion for DENV transmission to occur and to spread. The availability of susceptible human hosts is determined by local population density and levels of pre-existing immunity, and the presence of the DENV itself is the necessary third component of the transmission cycle.

Infection with any one of the four DENV serotypes is believed to elicit lifelong immunity against that serotype, but confers only partial or transient immunity against the other three serotypes. ^13,39 Secondary or subsequent infection with a heterologous virus is a well-established risk factor for severe dengue disease, ^11,40–42 a phenomenon that is thought to be mediated by a process of antibody-dependent enhancement (ADE) of DENV infection by cross-reactive non-neutralizing antibodies, resulting in a higher viral burden and an increase in the resultant pathogenic processes. ^43,44 Transient cross-protective immunity and ADE may also have implications for dengue transmission at a population level, as they may respectively reduce or increase the effective population of susceptible human hosts. ^45–47 Despite the importance of cross-protective immunity for understanding the epidemiological dynamics of dengue, little is known about the duration of cross-protection against a heterologous serotype. It is generally assumed that a first clinical dengue episode elicits cross-protective immunity that lasts 2–9 months, ^1,48 although a recent report estimates cross-protection to last as little as 1–2 weeks. ⁴⁹

There is evidence that the infecting viral serotype, genotype and sequence of infection are also important determinants of disease severity, independent of infection parity. ^50–55 This may also influence dengue transmission dynamics if increased inherent virulence results in higher viral titers and increased infectiousness to mosquitoes. Although there is limited direct evidence for this as yet, particular genotypes of DENV serotype 2 (DENV-2) are known to have displaced previously dominant strains in Vietnam, Thailand and Cambodia, suggestive of a difference in viral fitness. ⁵⁶ Age, ^11,15,57,58 host genetic background ⁵⁹ and the time interval between sequential infections ⁵⁷ are other factors associated with the risk of developing severe dengue disease.

Finally, factors that influence interactions between the virus, vector and host components of the dengue episystem will be critical in determining the overall effect on the establishment, persistence and intensity of dengue transmission in a given setting. This includes factors that determine the accessibility of vectors to human hosts, such as built environments that create vector breeding sites in close proximity to humans, increases in population density associated with urbanization or the use of window screens and air-conditioning that reduce vector access into human dwellings. As the Aedes vector has a restricted flight range, movement of dengue-infected humans is thought to play a major role in dengue transmission. ^60,61 Rabaa et al. ⁶² have demonstrated substantial DENV viral exchange between the urban center of Ho Chi Minh City and other provinces of southern Vietnam, suggesting that human movement between urban and rural areas may play a central role in the rapid diffusion of DENV across southern Vietnam. Movement and transport of humans or vectors at a national, regional or international scale creates new opportunities for vectors to become established in permissive environments, and for virus transmission to be established where competent vectors exist, and inadequate or interrupted vector control activities determined by economic and political priorities will directly affect individuals' risk of exposure to infection in endemic settings. These social, demographic and economic drivers are thought to be responsible for much of dengue's expansion and intensified transmission over recent decades. ^2,16,63

Dengue transmission dynamics and disease modeling

Tools for modeling dengue transmission dynamics

Traditionally, Fourier analysis has been used to analyze relationships between oscillating time series. This technique decomposes time series into their different periodic components that can then be compared between time series. Since Fourier analysis cannot take into account temporal changes in the periodic behavior of time series (i.e. their lack of stationarity), this method, and others such as generalized linear models, may be inadequate for investigating the determinants of dengue transmission dynamics. Wavelet analysis is suitable for investigating time series data from non-stationary systems and for inferring associations between such systems. ⁷⁶ This approach reveals how the different scales (i.e. the periodic components) of the time series change over time. Wavelet analysis is able to measure associations (coherency) between two time series at any frequency (period) band and at every time-window period.

Wavelet coherency analyses have been used to compare time series of disease incidence across localities and countries for the characterization of the evolution of epidemic periodicity and the identification of synchrony. Wavelet analyses have been used in analyzing various human infectious disease dynamics such as measles, influenza, leishmaniasis and dengue. ^{65,71,77–80}

Although annual periodic patterns are a common phenomenon in dengue endemic areas, the identification of a periodic multiannual (e.g. 2–3 years) cycle differs between countries and in analyses used. Cazelles et al. ⁶⁵ used wavelet approaches to demonstrate a highly significant but discontinuous association between ENSO, precipitation and dengue epidemics in Thailand. Johansson et al. investigated the relationship between ENSO, local weather and dengue incidence in Puerto Rico, Thailand and Mexico since the mid-1980s with an annual scale and multiyear cycles. Temperature, rainfall and dengue incidence were strongly associated in all three countries for the annual cycle. The associations with ENSO varied between countries in the multiannual cycle. ENSO was associated with dengue incidence in Puerto Rico, but characterized with non-stationarity while there was no association in Mexico. In Thailand, ENSO was associated with climate and dengue incidence. ⁷¹ Thai et al. provided insights into dengue transmission dynamics in Binh Thuan province, southern Vietnam, between 1996 and 2009. A continuous annual mode of oscillation with a non-stationary 2–3-year multiannual cycle was observed (Figure 2a) with strong irregular associations between dengue incidence and ENSO indices and climate variables (Figure 2b). The dengue incidence time series for Phan Thiet City and eight neighboring districts within Binh Thuan province are shown in Figure 2c. Figure 2d shows the coherency between these two time series, which demonstrates high coherence in the annual periodic band and in the 2–3-year periodic band between 1996 and 2001. In addition, the spatio-temporal patterns seem to be different between the annual and multiannual cycle. ⁷³

OPEN IN VIEWER

Figure 2

Wavelet analysis and wavelet coherency analysis with dengue incidence time series and with an El Niño Southern Oscillation (ENSO) index in Binh Thuan province, Southern Vietnam. (a) Wavelet analysis of the dengue incidence time series with monthly data. Color scheme shows increasing intensity, from blue to red; dotted lines show statistically significant area; the black curve delimits the cone of influence. (b) Wavelet coherence analysis of the dengue incidence time series with the ENSO index NIÑO 4. Dark blue and dark red indicates low coherence and high coherence, respectively. The dotted lines show α = 5% and 10% significance levels. The cone of influence (black curve) indicates the region not influenced by edge effects. (c) Time series of dengue incidence in Phan Thiet City (solid red line) and the eight neighboring districts combined (dotted blue line). (d) Wavelet coherence analysis of time series in panel (c) showing a continuous annual mode of oscillation with a non-stationary 2–3 year multiannual cycle. Dark blue indicates low coherence and dark red indicates high coherence. The dotted-dashed lines show α = 5% significance levels computed based on 1000 bootstrapped series. The cone of influence (black curve) indicates the region not influenced by edge effects. This figure has been published previously⁷³ and is reproduced with the permission of the authors

Climate change and dengue

Impact of changes in dengue transmission range and intensity on clinical outcomes

Changes in the geographic distribution or intensity of dengue transmission will have implications for the clinical manifestation of disease. Two key risk factors for both symptomatic and severe dengue disease are the age at infection and the acquisition of secondary infections. Children with DENV infections are less likely than adults to experience clinical disease, ^98–100 but are also less able to compensate for the increased vascular permeability that is characteristic of dengue pathogenesis, so are at higher risk than adults of severe outcomes including severe vascular leakage and hypovolemic shock (DSS). ^58,100,101 In sporadic dengue epidemics in Cuba and Taiwan, clinical cases were more common in adults than children, ^57,102–104 and in Singapore, where successful vector control has greatly reduced the force of infection and therefore the population immunity, adult cases of DF represent the majority of clinical disease. ^105,106 In contrast, in the dengue endemic countries of southeast Asia and Central America, dengue is primarily a disease of children and young adults, and by adulthood most individuals have partial or full immunity from multiple DENV exposures. ^8,11,107,108 Repeated introductions of DENV to a non-endemic area have the potential to lead to sequential epidemics with heterologous DENV strains, with the associated increase in the risk of DHF. An increased force of infection in endemic settings where multiple serotypes circulate could be expected to reduce the time between sequential infections, leading to a shift in the age distribution of disease towards younger age groups and potentially increasing the incidence of DHF/DSS in very young children. A recent analysis of hospitalized dengue patients in Ho Chi Minh City from 1996–2009 demonstrated that the highest risk of DSS was in pediatric dengue patients aged 6–10 years, but the youngest children (≤5 years) were at significantly higher risk of mortality than older children. ¹⁰⁹ Efforts to model the epidemiology of dengue based on trends in climate, environmental and other factors should therefore consider also the implications of alterations in force of infection and the likelihood of sequential or co-circulation of heterologous serotypes on the age distribution of disease and therefore the clinical spectrum of dengue.

Further issues

Evaluating the effect of climate or other factors on the current or future dengue disease burden requires defined indicators of disease or infection. The most obvious indicator is clinical disease; however, for dengue, perhaps more so than for many other infectious diseases of global importance, this is complicated in a number of ways. Firstly, national and international surveillance efforts for dengue generally rely on passively reported notifications of hospitalized cases, and therefore do not capture the full disease burden, with the ascertainment gap subject to considerable variation between settings and in time. Hospital-based data also carries inherent issues with determining the appropriate denominator for calculating burden of disease. Furthermore, dengue encompasses a spectrum of severities that may be variably included in the case definition for reporting. At the same time, surveillance in most endemic settings is based only on a clinical diagnosis with only a small proportion of cases undergoing laboratory diagnostic testing; therefore, over-reporting is an issue of variable magnitude within and between countries and over time. Furthermore, the quantitative relationship between clinical cases and total infections will vary substantially between epidemiological settings, depending on the intensity of transmission, age distribution of cases, population immunity and identity of circulating viruses. Indicators of vector abundance (i.e. house index, container index and Breteau index of pupae and larvae) have often been used as proxy measures for dengue risk, but their correlation with human disease is often poor ^110,111 and is likely to depend on local ecological and demographic characteristics. The lack of a robust and standardized indicator for dengue disease burden complicates efforts to predict and monitor future trends in dengue range and endemicity, which will rely on comparisons with a baseline, as well as models examining spatial patterns in dengue transmission, which may be confounded by geographic differences in clinical case ascertainment. Extensive work has been done to describe the geographic range and endemicity of malaria worldwide, using Plasmodium parasite prevalence as the index of endemicity. ^112,113 The existence of numerous published and unpublished parasite prevalence surveys from many locations and over a substantial time period made spatial and temporal comparisons of malaria endemicity valid and logistically feasible. An equivalent population-based virological marker for dengue burden is less obvious. Standard serological surveys give an indication of lifetime exposure risk, but must include a large age-stratified sample in order to be informative about historic and current transmission intensity. Serological markers of acute or recent DENV infection represent a more promising candidate for population-based assessments of dengue transmission risk.

We have discussed the limitations of transmission models or risk assessments that consider the climatic or immunological influences on dengue transmission in isolation from the multifactorial environmental, social and demographic determinants that are accepted to have contributed to dengue's recent resurgence. However, one explanation for the absence of these variables from models may be the difficulty in obtaining time series for sociodemographic and environmental factors that are comparable in time and geography to the case burden and climatic time series. Greater interdisciplinary collaboration between the ecology, social science and biomedical science fields would facilitate the incorporation of longitudinal data on relevant covariates as far as those data are available, or would at least allow for model findings to be considered within a broader environmental, social and demographic context.

Conclusions

Dengue, like all vector-borne diseases, is sensitive to climatic conditions that shape virus replication, vector development and survival, and therefore help define the geographical and seasonal limits that can support DENV transmission. Climatic factors do not, however, act in isolation, and non-climatic variables including population growth, human movement and environmental changes may have had far more to do with the global resurgence in dengue witnessed over recent decades than any direct effects of climate. The influences of climatic and non-climatic determinants on current and future dengue transmission are difficult to disentangle and therefore to quantitate independently. The challenge in refining models of dengue transmission to maximize their utility in predicting the location, magnitude and timing of future dengue epidemics or multiannual peaks in endemic cycles is to use data at appropriate spatial scales so that relevant ecological, social and demographic variables that operate at a local or regional scale can be incorporated into the model.

Author contributions: KTDT and KLA contributed equally to this work.