Abstract

Introduction

Dengue is a mosquito-borne infectious disease caused by a virus with four serotypes (DEN 1-4) of the Flaviviridae family, which is transmitted to man by the bite of mosquitoes of the genus Aedes, mainly Aedes aegypti in urban areas (Kraemer et al., 2015). Clinical manifestations of the disease vary widely from asymptomatic to high fevers, headache, muscle and joint pain, and in some severe cases plasma leakage, hemorrhages and death (Polwiang, 2016). It stands as one of the main emergent tropical diseases, affecting 390 million people per year and a tenfold at risk in 128 countries, with an estimated annual global cost of US$8.9 billion (Bhatt et al., 2013; Brady et al., 2012; Shepard et al., 2016). The major disease burden is registered in South East Asia, South Asia and Latin America; during 2016, over 2.3 million cases were reported only in the Americas, with 244.8 cases for every 100,000 inhabitants (PAHO, 2018).

Where dengue is epidemic, the occurrence of an outbreak depends on virus arrival, the presence of a susceptible human population, a competent vector population and adequate environmental conditions for virus development and transmission. Therefore, dengue dynamics is affected by multiple mechanisms, in which temperature is an important determinant of mosquito traits relevant to transmission, namely the biting rate, egg and immature development, adult survival and fecundity, and development time of the virus in the mosquito (reviewed in Mordecai et al., 2017). While both seasonal and inter-annual climatic variability influence the geographical distribution of A. aegypti, other factors also determine habitat suitability. Importantly, the successful exploitation of artificial containers as larval habitats, which translates in a high domestic condition of the vector, allows A. aegypti to persist in regions that may otherwise be unsuitable based solely on climate (Jansen & Beebe, 2010). Along with local climate, El Niño Southern Oscillation has also been reported to play a role in dengue dynamics at the seasonal and inter-annual scales (e.g., Vincenti-Gonzalez et al., 2018).

In Argentina, A. aegypti is distributed along temperate and subtropical latitudes (Vezzani & Carbajo, 2008). Adult activity is concentrated in the warm season throughout its distribution and is absent during winter in temperate zones (Carbajo & Vezzani, 2015). As neighboring countries to the northeast (Brazil and Paraguay) and northwest (Bolivia) are endemic for dengue, Argentina represents the southern limit of dengue transmission in South America with epidemic outbreaks concentrated in the warm season (November to May) (Carbajo, Cardo & Vezzani, 2012). In this transmission fringe, dengue re-emerged in 1998 causing epidemic outbreaks of different magnitudes in tight association with the incidence in endemic neighboring countries (Fig. 1). Before 2016, the biggest outbreak had occurred in 2009, with nearly 27,000 cases of DEN 1, followed by 2013, in which 2,922 cases were reported and co-circulation of DEN 1, 2 and 4 was verified (MSN, 2015). In 2016, the biggest dengue epidemic in the country so far was experienced, concomitant with first to date autochthonous transmission of Zika and chikungunya (26 and 322 confirmed cases, respectively) (MSN, 2016). Also, a recent outbreak of yellow fever has been reported in Brazil with over 1,000 confirmed cases (Ministério da Saúde do Brasil, 2018). These arboviruses are all transmitted in urban settings by the same mosquito vector, A. aegypti (Kraemer et al., 2015). There is currently no antiviral therapy against dengue, and although the vaccine CYD-TDV “Dengvaxia^®” has been approved in the country (ANMAT, 2017), the development of haemorragic dengue in a first infection after vaccination has been recently reported (WHO, 2017). Therefore, preventing contact between mosquitoes and people is still considered the main tool in the struggle against dengue.

Open in a separate window

Figure 1

Total number of dengue cases registered in Argentina per epidemiological year (defined from 1 July Year X-1 to 30 June Year X) for the period between 1999–2016, and classification of each year in epidemic intensity 0 to 3.

Despite its recent history of epidemics, several studies with different geographic scopes have been performed to study dengue dynamics in Argentina. The spatio-temporal pattern of dengue cases during the 2009 epidemic has been analyzed within a city (Seijo et al., 2009; Estallo et al., 2014) and at the country scale (Carbajo, Cardo & Vezzani, 2012). Regarding the 2016 outbreak, studies have been undertaken only at the city scale (Rotela et al., 2017; Carbajo et al., 2018). As no country-wide comprehensive analysis has been performed so far, and given the recent upsurge of the disease, the objective of this work was to identify the environmental factors that trigger epidemic events in Argentina by analyzing the spatio-temporal pattern of dengue cases since the re-emergence of the disease.

Materials & Methods

Mainland Argentine extends from 22° to 55°S, encompassing subtropical and temperate latitudes (Fig. 2). Along with its neighboring countries, it is located between the two great oceans of the Southern Hemisphere, the Atlantic and Pacific Oceans. This configuration reduces daily and annual thermal amplitudes in comparison to similar latitudes of the Northern Hemisphere. The Andes Mountains, with a height range of 2,600—6,000 m.a.s.l., also greatly influences the regional climate by preventing the passage of moisture from the Pacific Ocean (Barros et al., 2015).

Open in a separate window

Figure 2

Study area and meteorological stations. Squares indicate stations adjacent to districts with dengue cases in at least one year since 2009.

Mean temperature has increased by about 0.5 °C across most of Argentina during the past century. The strongest positive changes since 1960 occurred in the mean summer minimum temperature, even though in this season the mean maximum temperatures mostly decreased, except in Patagonia. In the second half of the past century, there was a general warming in Patagonia where both the maximum and the minimum temperature had a positive trend that was consistent with a more frequent northern flow component in the low levels of the atmosphere. Regarding precipitation, annual positive trends in northern Argentina can be partly attributed to variations in the frequency and intensity of the El Niño-Southern Oscillation (ENSO) phases. The rest of subtropical Argentina was influenced by the southern shift of the western edge of the South Atlantic high that enhanced moist advection from the Atlantic Ocean (Barros et al., 2015).

Information on dengue cases was obtained from dengue notifications reported in the National Health Surveillance System. During laboratory surveillance (SIVILA), diagnosis was confirmed through the detection of viral genome by polymerase chain reaction or detection of neutralizing antibodies (IgG) by plaque reduction neutralization test (PRNT). After the onset of epidemics in a given city, cases were further confirmed by detection of virus-specific immunoglobulin M (IgM) antibodies or NS1 antigen detection. Unknown data regarding department and province of residence was reconstructed using different sources: (1) district and locality in the domicile section of the SIVILA database; (2) direct consultation with the provincial Health Surveillance Area referents; or (3) data of province and district of sample collection consigned in SIVILA. Data of province or country of contagion was reconstructed from epidemiological comments of each case report. Cases registered in the SIVILA database were classified as autochthonous or imported according to the site of acquisition of the infection. All cases whose possible site of infection coincided with the jurisdiction of the patient’s habitual residence, without a history of travel to an area with dengue virus circulation, were defined as autochthonous. All cases with residence in an area without dengue virus circulation and with a history of travel to an area with circulation of the dengue virus in the last 15 days prior to the onset of symptoms were defined as imported. The database was available per district and epidemiological week.

For all analyses hereafter epidemiological years were defined as from 1 July Year 1 through 30 June Year 2 (e.g., reference to Year 2004 includes the period from 07/01/2003 to 06/30/2004). In accordance, climatic seasons were defined as follows: winter 1 Jul–30 Sep, spring 1 Oct–31 Dec, summer 1 Jan–31 Mar, autumn 1 Apr–30 Jun.

Results

Spatial approach

Out of the 503 districts in the country, 272 were negative for dengue cases in both major epidemics (2009 and 2016), whereas 73 reported one or more cases in both occasions. A strong increase in the number of positive districts was verified in 2016 compared to 2009, with 133 new districts reporting cases, and only 25 with the opposite behavior (Fig. 3). The relation autochthonous/imported cases was, however, very similar in both years, around 26 autochthonous cases per imported case (47,894/1,850 for 2016 and 26,923/~1,000 for 2009). Although not reflected in the number of cases and districts affected, considering temperature conditions expressed in terms of days of possible transmission (DPT) the situation in 2009 was worse than in 2016 (isolines of 30 and 120 DPT spread further south and west, Fig. 3).

Open in a separate window

Figure 3

Dengue cases per district and days of possible transmission (DPT) isolines for 2009 (A) and 2016 (B).

The best model for the occurrence of two or more dengue cases per district included autumn minimum temperature and human population (log scale) as fixed factors, and the province as a random grouping variable (Table 2). Mild temperatures during the previous autumn and higher population numbers were associated with the occurrence of dengue cases per district. Prediction accuracy of the observed values was 80% better than chance (K = 0.8) at a cut off-point of 0.4. Explanatory power of the fixed factors (marginal R²) was 0.74, whereas including the random factor (conditional R²) raised explanation to 0.86.

The model published in Carbajo, Cardo & Vezzani (2012) for the 2009 epidemics was re-run using the occurrence of two or more dengue cases per district in 2016 as a response variable, and the same explanatory variables that predicted the previous outbreak. These were the DPT, distance to water courses, population number (log scale) and the percentage of population change in the period 2001–2010 (last two national censuses). In Argentina, national censuses are typically performed on a decadal basis; therefore, updating of the demographic variables was unfeasible. The model was not very accurate in predicting the new cases (K = 0.47, correct classification 0.7, sensibility 0.4, specificity 0.9). An alternative explanation for the low predictive model is the difference in the origin of both epidemics. The one in 2009 came down from Bolivia, whereas the latter 2016 outbreak was caused by an unusually high incidence in north-eastern neighboring countries (Brazil and Paraguay). Also, unexpectedly two northeastern provinces did not communicate any cases during 2009 (Fig. 3A).

Conversely, the new model was re-run with the 2009 epidemics data. In this case, the predictive accuracy was high (K = 0.71 at a cut off point of 0.35), however the fraction explained by the fixed factors was significantly reduced (marginal R² = 0.45), whereas the variance explained by the Province as a random grouping increased (conditional R² = 0.84).

Discussion

Modeling disease outbreaks in time and space is a powerful tool to qualitatively predict future potential epidemics (Myers et al., 2000). Given the epidemic nature of dengue in Argentina, virus pressure from endemic neighboring countries along with climatic conditions are crucial to explain disease dynamics. Remarkably, in all three methodological approaches used to analyze the spatial and temporal pattern of dengue cases, different proxies for temperature of the previous autumn were among the best predictors. The number of days necessary for transmission presented higher explanatory power than the raw variable in two approaches. Although this proxy is mainly used for transmission risk calculations, it may well resume conditions for the mosquito. As temperatures seldom reach extreme inhibition values during autumn, the temporal detail and the inclusion of temperature amplitude in its calculation may favor DNT instead of mean temperature.

We built a temporal model based on the period 2009–2016, which lacks predictive power because it relies on the ongoing year virus pressure from endemic neighboring countries. However, it does provide a hint regarding the conditions of the previous autumn. If the autumn of a given year is propitious for transmission, caution should be taken for the following summer and the situation in bordering endemic areas should be followed up closely. We propose that mild autumns represent an advantage for mosquito vector populations. In temperate regions with no mosquito reproduction during winter, this advantage could manifest as a larger egg bank deposited during autumn from which the adult population will re-emerge in spring.

Including information of the years 1999–2008, we reached a model only relying on in advance climatic conditions. In this way, with easy-available temperature data we can anticipate if the following year will present an epidemic of mid-high intensity. Models predicting epidemic intensity rather than number of cases are more robust to particular and/or stochastic conditions of a given year. The presented model properly predicted the 2017 situation; however as only two major outbreaks (2009 and 2016) have occurred so far in Argentina to train the models, more time is needed to test its efficiency.

Spatio-temporal models of dengue cases distribution between 2009 and 2016 presented significant association with the same type of variables, prioritizing the days necessary for transmission during autumn months and the incidence in neighboring endemic countries. These models also lack predictive power and are particularly sensitive to the characteristics of the on course epidemics. This could also be appreciated when comparing 2009 and 2016 spatial models, in light of the poor correspondence in terms of numbers and geographical distribution of positive districts that presented the Carbajo, Cardo & Vezzani (2012) model for the 2009 epidemics to predict the situation in 2016. Both main dengue outbreaks in Argentina started in locations close to endemic borders and further expanded to other localities, however their origin was different. In 2009, the epidemic was triggered by high incidence in Bolivia, spreading south from the northwest, whereas in 2016 it started in the northeast, due to high incidence in south Brazil and Paraguay. For this reason, the proper identification of imported cases is especially relevant during transmission onset. We also found no relation between El Niño events and dengue patterns. It has been described that river flow of water courses located within the northeast region of the country are strongly affected by ENSO events, whereas those located in the northwest region are not (Pasquini & Depetris, 2007). This may partly explain the lack of association of dengue outbreaks and El Niño in Argentina, as described elsewhere, however we were unable to test this hypothesis due to the short history of northeast and northwest derived epidemics in the country.

The geographic distribution of dengue-positive districts during the 2016 epidemics was positively associated with minimum autumn temperature and human population, the latter variable being also associated with the 2009 outbreak (Carbajo, Cardo & Vezzani, 2012). A larger human population number in a district may reflect higher virus pressure due to traveling. Also, larger cities usually present precarious and overcrowded conditions, and are therefore more susceptible to experiencing an outbreak (Brunkard et al., 2007). This is combined with a large number of artificial containers due to uncontrolled urbanization and land uses linked to urban areas (e.g., cemeteries, rubbish and tire deposits), which represent optimum habitats for immature vector development. Within the spatial model, the inclusion of Province as a random grouping factor may reflect non biological features, such as asymmetries in case records or in the efficiency of prevention and control interventions. The total absence of cases reports during 2009 in some northeastern provinces is an extreme example of this issue. These areas experienced unreported transmission, which could have substantially affected the predictability of the 2009 model as it occurred to Aström et al. (2012).

Comparing dengue distribution models may be problematic because of differing modeling approaches and scales, the diverse quality of the data used, and the selection of variables associated with disease distribution (Messina et al., 2015). Further issues arise in the particular case of Argentina due to its geography. It constitutes one of the potential areas of transmission expansion and intensification, where many models do not agree in their predictions (Messina et al., 2015). It extends along 3,800 km including a gradient of transmission risk due to climatic variables roughly decreasing from north to south, but almost a third of the population lives at the transmission fringe in the center-east of the country. The epidemic inter-annual pattern is irregular, with frequent virus circulation in the north and large sporadic outbreaks extending to almost half of the country. This implies that the maximum geographic extension does not repeat regularly and may respond to particular conditions occurring in certain years. Global models are not aimed at identifying these particular events, what would need a spatial and temporal detail not yet implemented or simply not available globally. Although the inclusion of demographic and socio-economic factors has improved global models (e.g., Aström et al., 2012; Mordecai et al., 2017), its application in Argentina would require a district detail to account for the internal heterogeneity in, for example, the geographical distribution of income along the country.

The systematic data gathering of dengue cases in a country-wide system allows for the constant updating of the epidemiological situation, as well as the identification of high risk areas to prioritize prevention and control measures both at the communitarian and individual levels. The strength of the 2016 outbreak manifests the flaws of the mitigation strategies. This, in conjunction with a higher dengue incidence in neighboring countries and the introduction of other arboviruses transmitted by the same vector (Zika, chikungunya, and potentially yellow fever), forecast a complex scenario which claims for an integrative research approach with joint work from the areas of education, infrastructure and health. Along with the need to develop better and faster diagnostic methods, we need to achieve consciousness in the population, whose contribution is essential. The optimal use of models to inform policy decisions requires a continuous dialogue between the multidisciplinary infectious disease dynamics community and decision makers (Heesterbeek et al., 2015), essential to plan and put into practice adequate prevention and control measures for each jurisdiction.

Conclusion

In the southernmost extreme of epidemic transmission in the Americas, the pattern of dengue cases concentrated during the warm season was associated with different proxies for temperature conditions during the previous autumn and virus pressure from neighboring endemic countries. This provides a valuable opportunity to prepare in advance for a qualitatively strong epidemic during the following summer if the autumn of a given year is propitious for transmission, and enable the onset of early alerts by close follow up of the situation in endemic bordering countries.

Supplemental Information

Funding Statement

This work was supported by Becas SALUD INVESTIGA “Dr. Abraam Sonis”, categoría ESTUDIO MULTICENTRICO, Dirección de Investigación para la Salud, Ministerio de Salud de la Nación. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References