Watts et al. Int J Health Geogr International Journal of (2020) 19:44 https://doi.org/10.1186/s12942-020-00241-1 Health Geographics RESEARCH Open Access Influence of socio‑economic, demographic and climate factors on the regional distribution of dengue in the United States and Mexico Matthew J. Watts1*, Panagiota Kotsila1,4, P. Graham Mortyn1,5, Victor Sarto i Monteys1,3 and Cesira Urzi Brancati2 Abstract Background: This study examines the impact of climate, socio-economic and demographic factors on the incidence of dengue in regions of the United States and Mexico. We select factors shown to predict dengue at a local level and test whether the association can be generalized to the regional or state level. In addition, we assess how different indi- cators perform compared to per capita gross domestic product (GDP), an indicator that is commonly used to predict the future distribution of dengue. Methods: A unique spatial-temporal dataset was created by collating information from a variety of data sources to perform empirical analyses at the regional level. Relevant regions for the analysis were selected based on their recep- tivity and vulnerability to dengue. A conceptual framework was elaborated to guide variable selection. The relation- ship between the incidence of dengue and the climate, socio-economic and demographic factors was modelled via a Generalized Additive Model (GAM), which also accounted for the spatial and temporal auto-correlation. Results: The socio-economic indicator (representing household income, education of the labour force, life expec- tancy at birth, and housing overcrowding), as well as more extensive access to broadband are associated with a drop in the incidence of dengue; by contrast, population growth and inter-regional migration are associated with higher incidence, after taking climate into account. An ageing population is also a predictor of higher incidence, but the relationship is concave and flattens at high rates. The rate of active physicians is associated with higher incidence, most likely because of more accurate reporting. If focusing on Mexico only, results remain broadly similar, however, workforce education was a better predictor of a drop in the incidence of dengue than household income. Conclusions: Two lessons can be drawn from this study: first, while higher GDP is generally associated with a drop in the incidence of dengue, a more granular analysis reveals that the crucial factors are a rise in education (with fewer jobs in the primary sector) and better access to information or technological infrastructure. Secondly, factors that were shown to have an impact of dengue at the local level are also good predictors at the regional level. These indi- ces may help us better understand factors responsible for the global distribution of dengue and also, given a warming climate, may help us to better predict vulnerable populations on a larger scale. Keywords: Dengue, Climate-change, Global-warming, Socio-economic, Mosquito-borne, Vector-borne-diseases, GDP, GAM Introduction The dengue virus (DENV) is one of the most important *Correspondence: matthew.watts@uab.cat mosquito-borne viral diseases in the world today. Two 1 Institute of Environmental Science and Technology (ICTA), Autonomous main arthropod vectors are responsible for transmission University of Barcelona (UAB), Bellaterra, Spain Full list of author information is available at the end of the article of dengue viruses: Aedes aegypti (commonly known as © The Author(s) 2020. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat​iveco​ mmons​.org/licen​ses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creat​iveco​mmons​.org/publi​cdoma​in/ zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Watts et al. Int J Health Geogr (2020) 19:44 Page 2 of 15 yellow fever mosquito) and Aedes albopictus (commonly product (GDP), which is a standard measure of the mar- known as tiger mosquito). A. aegypti mainly feeds on ket value of all the final goods and services produced humans and is highly adapted to human habitations and over a specific time period in a given location. Some urban areas; A. albopictus feeds on animals and humans studies have looked at the interaction between climate, and is more prevalent in rural and peri-urban environ- socio-economic factors and demographics at a local level ments. While A. albopictus is also responsible for den- [16–20], focusing on factors specific to local areas, which gue transmission among humans, it is a less likely vector means that their findings cannot be easily extrapolated to than A. aegypti since it is adapted to a wider range of the macro level. To get better estimates of where dengue environments and has less restrictive feeding habits [1]. may spread, there is a need to understand how climate Both Aedes mosquitoes are highly adapted to breeding factors, socio-economic factors and demographic factors in aquatic habitats like ponds and lakes, but also micro interact over a greater geographic scale to reveal com- habitats, such as tree-holes, rock crevices and even leaf mon global patterns. axils [2]. The latter behaviour in recent times has ben- The original contribution of this article is that it efited both species by allowing them to exploit a range selects factors shown to predict dengue at a local level of man-made aquatic breeding habitats, where water and tests whether the association can be generalized can accumulate, like urban gardens, vases in cemeteries, to the regional or state level. In addition, we propose a discarded bottles and plant pots; therefore, both species more comprehensive set of socio-economic predictors can survive in drier climates than expected, by exploiting of dengue transmission, to disentangle the role of GDP artificial water sources. from other measures. Although a useful and parsimoni- Dengue is a disease caused by any one of four closely ous indicator, GDP is a very broad measure and it is not related viruses: DENV 1, DENV 2, DENV 3, or DENV 4. necessarily reflective of population health and well-being, Currently, all four dengue serotypes are in circulation in distribution of wealth, discrimination and spending on the Americas and can co-circulate within a region; the public welfare [21]. More importantly, GDP alone may actual distribution of each serotype is difficult to estab- not be able to capture cross-regional differences. The lish for a number of reasons, such as inadequate surveil- predominance of using GDP as an indicator has been lance, under reporting, high numbers of asymptomatic largely questioned [22–25]; for some time now research- carriers, and so on, as laid out by [3]. DENV causes an ers in human health geography, critical public health, and acute flu-like illness that affects people of all age groups. social epidemiology have requested more careful consid- Those who recover from a dengue infection can expect eration of the contextual social and economic conditions lifelong immunity against that serotype and some par- that shape diseases at the local level [26, 27]. tial, but temporary, cross-immunity to the other sero- To this end, this study investigates regional differences types, although secondary infections by other serotypes in the incidence of dengue by evaluating the impact of increase the risk of developing severe dengue, which may socio-economic and demographic factors such as house- cause lethal complications, and sometimes death [4]. hold income, regional rates of education, housing over- There is currently no specific antiviral therapy for den- crowding, life expectancy, medical resources, migration gue fever; once the disease is contracted, there is no way flows, age structure of the population (the proportion of to combat it other than relying on the host’s immune people under 14 and over 65), and population density. response. Several vaccines are currently in development; The study focuses on the occurrence and distribution however, given the current cost-effectiveness, efficacy, of dengue in Mexico and southern regions of the United safety and estimated impact of vaccination, the WHO’s States (US) where dengue has been reported, as some US present recommendation is to introduce it only in geo- regions share very similar environmental conditions but graphic settings (national or sub national) where the dis- have distinct socio-economic conditions [12]. This study ease is particularly problematic [5]. takes advantage of time series data between 2011 and 2019 and it is, therefore, able to exploit cross sectional Motivation for study variation between states, and variation over time for each Climate change, specifically rising temperatures, is likely state. playing a crucial role in dengue transmission, poten- tially driving its expansion across the globe, as predicted Conceptual framework by several studies [6–10]. Socio-economic conditions Dengue transmission is determined by interactions in a given location can be vital for a disease to persist between host, vector and pathogen, and modulated by once local transmission has occurred [11–15]; however, ecological, climatic and geographic factors, including research in this domain, generally, does not account socio-economic factors. Regions were selected for the for socio-economic factors other than gross domestic empirical analysis if conditions were met in terms of Watts et al. Int J Health Geogr (2020) 19:44 Page 3 of 15 their receptivity and vulnerability, based on principles Vulnerability laid out in the WHO’s framework for malaria elimina- As direct measures of vulnerability we include spatial tion [28]. effects (neighbourhood structures) in our models in order Receptivity is defined as the ability of an ecosystem to explicitly account for spill over effects with infected to allow transmission of a virus (dengue in this study). neighbouring regions (for a more detailed description see An ecosystem can be considered receptive if competent the methods section). Indirect measures of vulnerabil- vectors, a suitable climate and a susceptible popula- ity can be derived from traditional patterns of travel and tion are present; in other words, regions are selected if population flow in the area; indeed, well connected areas, autochthonous virus transmission may occur because in terms of trade and transport with considerable human human populations and vector populations overlap/ movement, can benefit both mosquito species and den- interact. Vulnerability occurs when either (1) a region gue, by facilitating their movement and spread [30–32]. was receptive and had regularly reported cases over the study period (endemic) or (2) bordered an endemic region and occasionally reported cases (likely due to Modulating factors spread or importation from neighbouring regions). We Modulating factors can either speed up or slow down defined modulating factors as variables that influence transmission. The transmission cycle of dengue is com- the transmission dynamics of dengue such as host pop- plex, since there are several key interactions at play ulation size, host density, climate factors and medical between the virus, host and vector. Density of both the interventions. vector and host are fundamental factors in disease trans- mission, as contact between infected vectors and suscep- tible hosts is the source of new infections [33]. Mosquito Receptivity breeding habitat can be increased by precipitation and Since dengue is a vector borne disease, understanding flooding [34], temperature heavily influences mosquito the key ecological requirements of its vectors is crucial to hatching rate, development time [35, 36] and optimal assessing the receptivity of a region. As explained below, temperature can shorted the extrinsic incubation period some of the main factors determining the receptivity of a (EIP) [37]. While there are no datasets covering mosquito region to dengue (due to the presence of its vectors) are: population abundance in all of our study regions, we its physical environment (land use), the overlap with the selected meteorological variables that predict mosquito human population, and its climate. abundance and therefore are related to dengue transmis- Both types of Aedes mosquitoes that transmit the sion. Furthermore, there are several socio-economic risk dengue virus are ectothermic organisms and are highly factors of dengue including home water storage (rather sensitive to colder temperatures and extreme high tem- than receiving piped water), poor sanitation, and poor peratures. A. albopictus adults can survive in tempera- public services (e.g. litter not collected) [12, 38–42]. Such tures from 15 to 35 °C and A. aegypti from 10 to 35 °C factors can be responsible for creating breeding habitat [29], while their growth and development are severely for mosquitoes and bringing them into closer contact inhibited in ambient temperatures below 13 °C or above with humans, therefore increasing the risk of dengue. By 35 °C. A. albopictus eggs though, can go through diapause contrast, use of mosquito nets, insect screens, and air- (suspended development) when exposed to extreme cold conditioning, can limit the chance of being bitten. Simi- (down to − 10 °C). This adaptation allows them to inhabit larly, knowledge and education of mosquito ecology can environments with a wider annual temperature range, also help people make personal interventions and reduce with more distinct seasonal changes than in tropical cli- risk of being bitten [43]. Because there are no direct mates, where climate is more homogeneous. A. aegypti measures of home water storage or the use of mosquito can endure a wider range of temperatures, but its survival nets, we use a range of socio-economic indicators as at temperatures below 14–15 °C is limited to short peri- proxies, capturing a latent variable that would represent ods, since its mobility is severely restricted and its ability vector risk. The rationale is that people living in locations to imbibe blood impeded. A. aegypti is also highly sen- with better socio-economic conditions can avoid contact sitive to fluctuations in temperatures. As for most mos- with mosquitoes and restrict virus transmission, either quito species, availability of freshwater habitat, humidity from the bottom-up (e.g. personal interventions) or the and precipitation are highly indicative of their distribu- top-down (e.g. regional government pest control). How- tion in the environment. ever, it is important to note that factors associated with To account for this, we selected a range of humidity higher economic status can also bring humans into closer and temperature variables for analysis which would cap- contact with mosquitoes, for example home owners with ture mosquitoes’ living requirements. gardens, potted plants and ponds, or having good access Watts et al. Int J Health Geogr (2020) 19:44 Page 4 of 15 to recreational space where mosquitoes can breed [44]. and frequency of each sample, we selected global obser- In terms of post-infection factors that influence dengue vations from 1970 onward to capture the entire range of transmission, access to health care, risk perception and climatic conditions that each species can survive in, and access to information on dengue infection symptoms to limit potential sample bias caused through the selec- had positive effects on people’s decision to seek medical tion of localised seasonal collections. We also removed help when presented with dengue infection symptoms any duplicate observations i.e. replicate coordinates. [14, 43, 45]. To reflect this in the conceptual framework Climate data were extracted using R’s DISMO pack- we selected variables that would proxy access to health age in all point locations where mosquitoes occurred. care and variables which would represent access to infor- Climate data for the species distribution prediction mation and personal knowledge. Finally, younger peo- modelling were sourced from the MERRAclim, a data- ple are more likely to be infected by dengue [46], so we base complied by members of the Department of Biol- selected variables that represent the age structure of the ogy and Geology, Physics and Inorganic Chemistry, Rey population. Juan Carlos University [48]. This data-set was built using 2 m above surface air temperature (Kelvin degrees) and Materials and methods 2 m above surface specific humidity (kg of water/kg In this study, we compiled a spatial temporal data-set of air) satellite observations from NASA’s Modern Era that would reflect the conceptual framework. By predict- Retrospective Analysis for Research and Applications ing the distribution of A. albopictus and A. aegypti in Reanalysis. Mexico and the United States, we could determine which Figure 1 shows the results of the modelling and Aedes regions were receptive i.e. there was an overlap between sample locations. Tables providing summary statistics for the vector distribution and the human population at risk. the climate values at Aedes point locations can be found By combing these results with reported cases of dengue, in Additional file 1. More specific information on statisti- we could determine which regions were vulnerable. We cal methods and results from this analysis can also found then went on to collect data on modulating factors of in Additional file 1. dengue transmission in vulnerable regions. Furthermore, our vector distribution maps allowed us to extract more accurate data on the host population at risk and climatic Data extraction and methods to assess the impact factors that contribute to disease transmission. of climate, demographic and socio‑economic factors on dengue Species distribution models to estimate regional The Global Administrative Unit Layers [49] data-set susceptibility along with our Aedes distribution maps (results Fig. 1, Because the exact distribution of vectors is unknown, we bottom right) were combined using R’s Sf package to cre- estimated the likelihood that a vector would occur in a ate regional shape files that could spatially capture and region conditional on a set of covariates. More specifi- process the human population and climate data for the cally, we estimated the distribution of the Aedes mosqui- main analysis. The GAUL data set contains geographic toes using a generalized additive logistic regression, with information in the form of shape files that lay out within point location occurrence data as our dependent variable, country boundaries linked to a unique nomenclature. and annual temperature range, mean temperature of the Countries are broken down into statistical subdivisions coldest quarter, precipitation during the driest quarter as e.g., ADM0 representing data at country level (e.g. US), covariates. Predictions were then used to select suscepti- ADM1 at regional level (e.g. California). ble regions. Climate data for the main analysis i.e. measuring the Point location occurrence data for A. aegypti and A. impact of the climate variables on dengue transmission, albopictus were obtained from a global geographic data- were sourced from the Climate Prediction Center (CPC) base of known occurrences between 1960 and 2014, com- of the National Centers for Environmental Prediction piled by members of the Institute of Biodiversity, Animal (NCEP), see [50]. These data represent a global summary Health and Comparative Medicine, College of Medical, of daily weather data. The CPC extracts surface synoptic Veterinary and Life Sciences, University of Glasgow [47]. weather observations from the Global Telecommunica- Point occurrence data represent spatial geo-coordinates tions System (GTS), which collects global data from a of a location in which a given individual organism was combination of weather station and satellite observa- sampled or sighted. Many of the samples in this data- tions. Files were processed in R with the NetCDF, Raster set consists of museum records or unpublished studies and Dismo packages in order to create annual bio-cli- including national entomological surveys. Since the data- matic variables. The bio-climatic variables in this study set contained sparse information relating to the timings Watts et al. Int J Health Geogr (2020) 19:44 Page 5 of 15 Fig. 1 Aedes sample locations and SDM results 1 Top left: Aedes point locations. 2 Top right: Results of Aedes aegypti SDM 3 Bottom left: Results of Aedes albopictus SDM 4 Bottom right: Receptive regions/data extraction locations were derived from daily maximum temperatures, daily data were provided at the regional level (ADM1 level). minimum temperatures and total daily rainfall. Case data for the United States were extracted from https​ Population count data to predict the number of persons ://www.cdc.gov/arbon​et, since data are provided at the at risk in a region were sourced from the Socioeconomic county level (ADM2 level) we needed to aggregate them Data and Applications Center’s Gridded Population of to the state level (ADM1 level) in order to match them the World data set [51]. This data set estimates popula- with the main data-set. tion count for the years 2000, 2005, 2010, 2015, and 2020, consistent with national censuses and population regis- OECD socio‑economic and demographic data ters. Data were extracted from areas where vector pres- Socio-economic and demographic data were extracted ence was predicted. R’s Zoo package was used to replace from the OECD’s Regional Statistics and Indicators Data- values for missing years, by implementing a linear inter- base [52]. This database provides comparable statistics polation method that would predict trends between and indicators and is presented in yearly time series. To years. This way increases or decreases in human popula- capture factors determining the vulnerability of a region, tion were controlled for in the final model. we selected the variables “Inter-regional migration rate”, All spatial data was aggregated to the state level. “Population density growth” and “Gross domestic prod- uct (GDP)”. For factors representing the socio-economic Dengue case data position of residents in a region we selected: “House- Dengue case data for Mexico 2011–2019 were obtained hold income”, “Life expectancy at birth”, and a measure from the Mexican Deputy General of Epidemiology web- of housing overcrowding “Number of rooms per person”. page, which provided reports on all positive serious and Furthermore, we selected “Secondary education” which non-serious cases of dengue (https​://www.gob.mx). All would also help to capture areas where there is a higher Watts et al. Int J Health Geogr (2020) 19:44 Page 6 of 15 proportion of manual labourer, e.g. agricultural workers Statistical methods or people working outdoors who may be more exposed Factor analysis—data processing for regional analysis to mosquitoes. We also selected “Perceived social net- A preliminary correlation analysis (see Additional file 1: work support”, “Self-evaluation of life satisfaction”, and Figures S8, S9- diagnostics) revealed how some of the “Perception of corruption” to try to capture additional socio-economic variables are strongly correlated with features of a region, such as quality of life. Since these each other, and if included in a regression would give rise three variables yield some indication of how people per- to multi-collinearity issues. By over-inflating the standard ceive their surroundings and quality of life, we assume errors, multi-collinearity makes some variables statisti- that poorer scores will capture poor infrastructure, poor cally insignificant when they should be significant. To public services, lack of basic provisions and lack of ben- address this issue, following similar methods to [53], a eficial government intervention. To represent access to factor analysis by maximum likelihood (VARIMAX rota- healthcare we selected “Active Physicians rate”, and vari- tion) was performed on socio-economic variables. ables which would represent access to information and Factor analysis is a method for investigating whether a personal knowledge i.e., “Broadband access” (however number of variables of interest Y1 , Y2 , . . . , Yn, are linearly knowledge is also captured by “Secondary education”). related to a smaller number of latent (i.e. ∼ not directly Finally, younger people are more likely to be infected by measured) factors F1 , F2 , . . . , Fk . The basic concept of fac- dengue [47], so we selected variables that represent the tor analysis is that multiple observed variables have simi- age structure of the population i.e. “Percentage of Old lar patterns because they are all associated with a latent Population Group (65+)” and “Percentage of Youth pop- variable. The factors are constructed in such a way that ulation group (0–14)”. Missing values were filled based on they capture the maximum amount of common vari- values for previous years or subsequent years, depending ance (correlation) of the original items; the eigenvalue is on their position in the data set. a measure of how much of the variance of the observed All data were joined using the year of observation and variables a factor explains. The factor analysis can be for- region code, using R’s Dplyr package. malized as follows: Table 1 provides summary statistics of all the collected data for the final models. Table 1 Final dataset 2011–2019 Statistic N Min Max Mean St. Dev. Primary income of private households (USD per head) 306 2541.8 1,159,750.0 68,486.2 229,793.6 Regional Gross Domestic Product per capita 306 2883.4 1,044,310.0 64,153.2 206,614.7 Share of labour force with at least secondary education 306 26.9 89.5 42.1 13.0 Share of households with broadband access 306 7.3 85.2 41.5 18.9 Self-evaluation of life satisfaction 306 6 9 7.2 0.6 Perceived social network support 306 59 96 79.4 9.6 Perception of corruption 306 36.5 90.1 63.3 11.1 Active physicians rate (physicians for 1000 population) 306 0.7 4.8 1.6 0.6 Life expectancy at birth 306 70.5 79.4 75.1 1.4 Number of rooms per person 306 0.7 2.5 1.1 0.3 Inter-regional migration rate, (% migrants over population) 306 0.5 7.0 2.0 1.3 Population density growth 306 97.8 179.6 122.4 12.9 Percentage of Old Population Group (65+) 306 3.1 20.5 7.2 2.5 Percentage of Youth Population Group (0–14) 306 16.3 34.4 27.5 2.9 DGE Mexico confirmed serious dengue cases 306 0 5,041 259.3 606.0 DGE Mexico confirmed non-serious dengue cases 306 0 9,195 627.2 1,161.6 CDC US confirmed dengue cases 306 0 53 0.3 3.5 Population in aedes infected regions 306 335,728.3 28,145,145.0 4,710,557.0 5,484,386.0 Mean (C) temperature of warmest quarter 306 16.2 32.3 25.6 3.7 Mean (C) temperature of coldest quarter 306 9.0 25.3 18.0 4.3 Precipitation of driest quarter 306 0.0 8.3 1.1 1.3 Precipitation of warmest quarter 306 0.1 24.9 9.1 4.7 Watts et al. Int J Health Geogr (2020) 19:44 Page 7 of 15 Y1 = β10 + β11 F1 + β12 F2 + · · · + β1k Fk + ǫ components are smooth functions (e.g., splines) of the Y2 = β20 + β21 F1 + β22 F2 + · · · + β2k Fk + ǫ covariates, in which functions themselves are expressed YN = βn0 + βn1 F1 + βn2 F2 + · · · + βnk Fk + ǫ as basis-function expansions. The spatial auto-correla- tion in the GAM model was approximated by a Markov Before performing the factor analysis, all variables had random field (MRF) smoother, defined by the geo- to be standardized to z-scores (x − µ)/σ to ensure that graphic areas and their neighbourhood structure. We they were on the same scale. After performing the factor used R’s Spdep package to create a queen neighbours analysis, the predicted values for the factors for any indi- list (adjacency matrix) based on regions with contigu- vidual region can be estimated. These predictions, known ous boundaries i.e. those sharing one or more boundary as factor scores, are weighted sums of the values of the point. We used a medium rank MRF, which repre- observed items. Roughly, items with a stronger correla- sented roughly one coefficient for two areas. The local tion with a factor component (i.e. those with larger load- Markov property assumes that a region is conditionally ings) will receive higher weights in the calculation of a independent of all other regions unless regions share a score for that factor. boundary. This feature allowed us to model the correla- tion between geographical neighbours and smooth over Quality of life index—data processing for regional analysis contiguous spatial areas, summarising the trend of the We created a ‘Quality of Life Index’ by combining 3 vari- response variable as a function of the predictors, for ables from the OECD regional database: ‘Self-evaluation further information see section 5.4.2 of [55]. In order of life satisfaction’, ‘Perceived social network support’ and to account for variation in the response variable over ‘Perception of corruption’. The variables were standard- time, not attributed to the other explanatory variables ised, harmonised and combined into a composite indi- in our model, we used a saturated time effect for years, cator, capturing a latent quality of life measure, because where a separate effect per time point is estimated. each element on its own is unlikely to have a direct rela- We first tried to fit our model using a Poisson distribu- tionship with dengue. tion. However, the mean of our dependent variable (den- gue cases by region and year) was lower than its variance General additive regression model to assess impact − E(Y) < Var(Y), suggesting that the data are over-dis- of independent variables on dengue case data at regional persed. We also tried to fit our models using the negative level binomial, quasi poisson and tweedie distribution, all par- One of the main issues with our data-set is that it did ticularly suited when the variance is much larger than the not meet some basic assumptions for statistical infer- mean. After several tests, we concluded that the tweedie ence, and specifically the data are not independent and distribution worked well with our data and allowed us identically distributed random variables (iid). More to model the incident rate. Analysis of model diagnostic specifically, the data-set captured repeated measure- tests didn’t reveal any major issues, in general residu- ments over the same regions, and observations were als appeared to be randomly distributed (see Additional not independent because of spill over effects from file 1: S10–S19—diagnostics). neighbouring regions, therefore we needed to imple- Tweedie distributions are defined as subfamily of ment an appropriate statistical design to control for (reproductive) exponential dispersion models (ED), with both temporal and spatial pseudo replication (lack of a special mean-variance relationship. independence). We could deal with this in two ways, (1) A random variable Y is Tweedie distributed TWp (µ, σ 2 ) either using a generalized linear mixed model (GLMM) if Y ED(µ, σ 2 ), with mean = µ = E(Y ), positive disper- approach, relaxing the assumption of independ- sion parameter σ 2 and Var(Y ) = µσ 2. ence and estimating the spatial/temporal correlation The empirical model can then be written as: between residuals, or (2) model the spatial and tem- E(Y ) = f1 (Xit ) + fn (Yeart ) + fm (Regioni ) poral dependence in the systematic part of the model [54]. We opted to use a Generalized Additive Model where the f (.) stands for smooth functions; E(Y )it is (GAM) using R’s Mgcv statistical package because of its equal to dengue incidence in region i at time t , which versatility and ability to fit complex models that would we assume to be Tweedie distributed; Xit —is a vector converge even with low numbers of observations and of socio-economic, demographic and climate variables. could capture potential complex non-linear relation- Yeart is a function of the time intercept and Regioni rep- ships. One of the advantages of GAMs is that we do not resents neighbourhood structure of region. need to determine the functional form of the relation- We run two separate sets of analyses: one comparing ship beforehand. In general, such models transform regions in the US and Mexico and another one looking at the mean response to an additive form so that additive Mexico only, to check for robustness. Watts et al. Int J Health Geogr (2020) 19:44 Page 8 of 15 Results socio-economic variables individually since, there was Figures 2 and 3 provide a descriptive overview of the less correlation between Mexican regions. study regions, a characterisation of their environments and the reported disease incidence for those years. As we US/Mex analysis can observe, the majority of dengue cases are reported in Socio‑economic and demographic indices Mexico/US tropical and sub tropical climates. It was not possible to explore the individual impact of Tables 2 and 3 provide the results of the factor analy- all of the variables in our data-set because of collinear- sis i.e. the weighting of our socio-economic indicators. ity issues. Population density was found to be positively Table 4 shows the results for the regression model com- correlated with GDP and primary income. “Percentage of paring confirmed dengue cases in the US and Mexico for Old Population Group (65+)” was negatively correlated 2011–2019. Table 5 restricts the analysis to Mexico only with “Percentage of Youth Population Group (0–14)” (see since we could exploit a better data-set in terms of case Additional file 1: S8, S9) diagnostics). For this reason, reporting, scale, and we could explore the impact of the we performed a factor analysis to reduce the number of variables, as explained in more detail in the section on Fig. 2 Koppen-Geiger climate classification in study regions (Source: koeppen-geiger.vu-wien.ac.at) Watts et al. Int J Health Geogr (2020) 19:44 Page 9 of 15 Fig. 3 Crude incidence rates of dengue per 100,0000 people Table 2 Socio-economic factor analysis results US/MEX primary income of households, and yielded one compos- Factor 1 ite indicator (see Table 2) , which we included as a regres- sor. A priori, the socio-economic indicator is expected to Primary income of private households (USD per head) 0.91 have a negative association with dengue. Share of labour force with at least secondary education 0.95 We built our statistical model in a stepwise fashion so Life expectancy at birth 0.78 we could analyse it using the lowest Akaike Information Number of rooms per person 0.97 Criterion (AIC) which would help us validate the qual- ity of statistical models for our dataset. The first column of Table 4 (GDP Model) shows the association between regional GDP and dengue cases across the regions; Table 3 Socio-economic factor analysis results Mexico the second column (SE Model) shows the association Factor 1 between regional dengue cases and the socio-economic Primary income of private households (USD per head) 0.41 indicator derived through factor analysis, plus other vari- Share of labour force with at least secondary education 0.95 ables such as active physician rate, broadband access and Life expectancy at birth 0.59 the quality of life index. Column 3 (Dem Model) includes Number of rooms per person 0.65 demographic variables, such as inter-regional migra- tion rate, population density growth and the percent- age of older population (65+). Column 4 (Clim Model) includes the climate variables mean temperature of the statistical methods. The Mexico/US factor analysis cap- coldest quarter and precipitation in the warmest quar- tured the variance in 4 highly correlated variables: higher ter. The “full model” in column 5 shows the relationship share of labour force with at least secondary education, between dengue incidence and all explanatory variables more rooms per inhabitant, life expectancy at birth, in our final model. Table 4 also summarises the relevant Watts et al. Int J Health Geogr (2020) 19:44 Page 10 of 15 Table 4 Final regression models US/MEX: EDF value is reported as the coefficient, and DF is included in parentheses (not standard errors because a chi-square test is used for the smooth terms) GDP model SE model Dem model Clim model GDP full model Full model Intercept 3.19*** (0.26) 2.44*** (0.22) 2.53*** (0.23) 2.46*** (0.23) 2.59*** (0.22) 2.22*** (0.20) Per capita GDP 0.00*** (0.00) 0.00*** (0.00) Socio-economic index 1.87*** (1.97) 1.78*** (1.92) Active physicians per 1000 1.72** (1.90) 1.87*** (1.97) 1.88*** (1.98) Households (%) with broadband access 1.94*** (1.99) 1.95*** (1.99) 1.95*** (1.99) Quality of life index 1.78* (1.92) 1.00 (1.00) 1.00 (1.00) Inter-regional migration rate 1.00*** (1.00) 1.65 (1.85) 1.61 (1.82) Pop density growth 1.82 (1.96) 1.00*** (1.00) 1.00*** (1.01) Percentage of population (65+) 1.97*** (1.99) 1.93*** (1.96) 1.79*** (1.89) Mean temp (°C) of coldest quarter 1.90*** (1.98) 1.88*** (1.97) 1.83*** (1.95) Precip of warmest quarter (in) 1.57* (1.81) 1.63* (1.84) 1.53** (1.76) Year 7.27*** (8.00) 6.75*** (8.00) 6.99*** (8.00) 7.28*** (8.00) 6.61*** (8.00) 6.61*** (8.00) Region 13.3*** (13.91) 12.76*** (13.72) 13.04*** (13.69) 12.72*** (13.81) 10.46*** (12.24) 10.45*** (12.19) AIC 2339.82 2322.32 2332.83 2285.47 2233.10 2231.57 Deviance 1288.59 1205.43 1257.24 1143.83 967.55 962.59 Deviance explained 0.51 0.55 0.53 0.59 0.66 0.66 Dispersion 3.52 3.37 3.47 3.20 2.84 2.83 R2 0.36 0.42 0.40 0.40 0.51 0.51 Num. obs. 306 306 306 306 306 306 Num. smooth terms 2 6 5 4 10 11 *** p < 0.01; ** p < 0.05; * p < 0.1 statistics (AIC, Deviance, Adjusted R squared and so on) primary income of households are associated with fewer to compare the different specifications; the full model has dengue cases. Regions with better broadband access tend the best fit (lower AIC and higher adjusted R squared), to be those with lower incidence rates of dengue, how- followed by the one in which we control only for the cli- ever in this case the relationship is flat at low levels of mate variables (as well as the year, regional effects); the broadband coverage (below 40 percent) and then turns first model, controlling for GDP alone, has the highest negative and quadratic at higher levels of access. These AIC and has a worse fit than the specification including results could suggest residents are more likely to search the socio-economic indicators. for information on dengue prevention measures conse- When controlling for demographic and climate vari- quently lowering transmission potential, or when suffer- ables, the impact of the socio-economic indicators still ing with symptoms may be more likely to seek medical remains statistically significant, as well as the impact of advice, therefore breaking the transmission cycle; these temperature. results are consistent with findings by [56, 57]. This result Please note that as we are not estimating a standard also could be an indicator of more advance and urban- regression model, the figures reported should not be read ized regions vs agricultural and less developed regions. It as coefficients, but degrees of freedom of the smooth is reported that dengue tends to affect more those work- terms. Given that we cannot interpret the coefficients to ing in labour-intensive industries, such as agriculture or infer the sign and magnitude of the relationship, we visu- fishing [58, 59]. alise it by plot. Figure 4 plots the partial effects—the rela- The variable representing active physician rates has tionship between a change in each of the covariates and a a positive impact on the incidence of dengue, in that change in the fitted values in the full model; the first plot regions with more active physicians tend to have higher shows that the socio-economic index has linear negative incidence; however, this is likely due to more accurate impact, but the relationship becomes weaker at very high reporting. Even in this case, the relationship is concave— scores; given the weight of each variable in the factor positive up to 3 percent rate and flat afterwards. analysis, the results can be interpreted as an increase in The impact of the demographic variables on the inci- the share of labour force with at least secondary educa- dence of dengue also follows the expected sign, with tion, more rooms per inhabitant, life expectancy at birth, inter-regional migration rate and population density Watts et al. Int J Health Geogr (2020) 19:44 Page 11 of 15 Table 5 Final regression models Mexico: EDF value is reported as the coefficient, and DF is included in parentheses (not standard errors because a chi-square test is used for the smooth terms) SE model Dem model Clim model GDP full model SEindex model Full model Intercept 2.57*** (0.20) 2.89*** (0.21) 2.71*** (0.23) 2.20*** (0.24) 2.46*** (0.19) 2.44*** (0.18) Per capita GDP 0.00* (0.00) Socio-economic index 1.96*** (1.99) Income of private households 1.95*** (2.00) 1.83** (1.96) Share households with broadband 1.88*** (1.98) 1.93*** (1.99) 1.93*** (1.99) 1.92*** (1.99) (1.99) Active physicians (1000 pop) 1.00*** (1.00) 1.81*** (1.95) 1.75*** (1.92) 1.81*** (1.95) Number of rooms pp 1.95*** (1.99) 1.00 (1.00) 1.00 (1.00) Labour force with secondary edu 1.97*** (2.00) 1.95*** (1.99) 1.95*** (1.99) Quality of Life index 1.00 (1.00) 1.00 (1.00) 1.00 (1.00) 1.00 (1.00) Inter-regional migration rate 1.00*** (1.00) 1.00 (1.00) 1.00 (1.00) 1.00* (1.00) Pop density growth 1.51 (1.74) 1.00* (1.00) 1.00 (1.00) 1.00* (1.00) Mean temp (°C) of coldest quarter 1.83*** (1.96) 1.89*** (1.98) 1.94*** (1.99) 1.89*** (1.98) Precip of warmest quarter (in) 1.53 (1.76) 1.71* (1.90) 1.68 (1.88) 1.71* (1.90) Year 6.72*** (8.00) 6.90*** (8.00) 7.27*** (8.00) 6.45*** (8.00) 6.53*** (8.00) 6.44*** (8.00) Region 13.18***(13.89) 13.17***(13.90) 12.07***(13.57) 12.19***(13.52) 12.32*** (13.59) 12.13*** (13.47) AIC 2251.33 2340.28 2270.07 2207.77 2212.08 2204.83 Deviance 978.13 1242.10 1071.46 871.99 886.18 859.68 Deviance explained 0.61 0.47 0.57 0.66 0.66 0.67 Dispersion 3.00 3.57 3.17 2.76 2.78 2.73 R2 0.47 0.38 0.40 0.52 0.51 0.51 Num. obs. 288 288 288 288 288 288 Num. smooth terms 8 4 4 11 10 12 *** p < 0.01; ** p < 0.05; * p < 0.1 growth being associated with a linear increase in the inci- washed away during intense rainfall [60]. Furthermore, dence of dengue; the presence of an older population is both Aedes mosquitoes can survive in drier climates associated with higher incidence of dengue up to a cer- than expected, by exploiting artificial water sources and tain level—it peaks at around 14 percent—and then a man-made habitats, as already mentioned “Introduction” reduction, as can be seen from Fig. 4. One possible expla- section. nation for this is that a higher proportion of older people means a more vulnerable population, however very high Mex analysis rates are also associated with wealthier regions, which For our second analysis focusing on differential diffu- offset the main impact of age. Figure 4 also show the sion of dengue within Mexican regions, we were able impact of the Mean temperature (°C) of coldest quarter to analyse variables individually since there there is sig- variable is almost linear. We can see that most cases occur nificantly less correlation between the socio-economic in regions which have particularly mild cold seasons. This variables. However, we could not select ”Population den- is concurrent with the literature, we would expect to see sity” because of a correlation with ”Primary income of more cases of dengue in regions with tropical climates, household”s and ”GDP”. ”Percentage of old population where there is a distinct absence of a cold season, dur- Group (65+)” was negatively correlated with ”Population ing which low temperatures would kill the mosquitoes off density growth” so was not included in the final model. or cause mosquitoes to overwinter effectively inhibiting Furthermore, ”Percentage of population share (0–14)” disease transmission, instead such conditions allow the was highly correlated with ”Access to broadband” and virus and mosquitoes to persist throughout the year. ”Workforce with secondary education” (and negatively The relationship between rainfall and dengue incidence correlated with population 65+), so we didn’t include it in the full model is slightly negative and significant; even in the study. We again built our second statistical model though this finding could appear counter intuitive, it in a stepwise fashion so we could analyse it using the low- is probably due to the fact that mosquito larvae can be est Akaike Information Criterion (AIC) which would Watts et al. Int J Health Geogr (2020) 19:44 Page 12 of 15 Fig. 4 Partial effects of explanatory variables: GAM Mex/US model Fig. 5 Partial effects of explanatory variables: GAM Mexico model Watts et al. Int J Health Geogr (2020) 19:44 Page 13 of 15 help us validate the quality of statistical models for our of more granular socio-economic indicators can explain data set. Figure 5 and Table 5 present the results of our with greater accuracy the differences in the spread of second analysis focusing only on Mexican regions. disease in places with similar physical geography and Our findings for the second analysis are similar to the ecological characteristics. In addition, public health first: the most significant variables are “Share of house- authorities should be aware of the presence of non-lin- holds with internet access”, “Active Physicians Rate (1000 earities in relationships between dengue and income. pop)” and “Mean temperature (C) of coldest quarter”. Secondly, factors that were shown to have an impact of Our socio-economic indicator was a good predictor of dengue at the local level are also good predictors at the dengue incidence, although when “GDP” was paired with regional level. Given that data for these indicators are other individual variables from the factor analysis (except available at a sub-national scale for OECD countries and primary income) it helped to create a very useful model. selected OECD non-member economies, these indices The best fit model was our final specification using our may help us better understand factors responsible for the socio-economic variables individually; however, primary global distribution of dengue and also, given a warming income of households is not a reliable predictor of den- climate, may help us to better predict vulnerable popu- gue, since, by the concave relationship, it would appear lations. Although the variables used in this study do not that gains in economic activity may increase the spread represent disease transmission mechanisms directly, of the virus (for instance because of movement of goods understanding the relative impact of socio-economic, and people), but could also be correlated with higher demographic and climate factors on disease outcomes reporting. One of the strongest predictors of dengue in can help risk assessors predict where diseases are likely our final specification is “Share of labour force with sec- to occur in the future, by identifying locations with vul- ondary education”. As previously noted, this is consistent nerabilities in public health systems and/or by identifying with other findings by [58, 59] as dengue tends to affect impoverished areas that tend to be susceptible to disease. more those working in labour-intensive industries, such Our findings are not only useful for public health, but as agriculture or fishing. also contribute to a wider scholarly debate on whether and to what extent can economic growth (measured via Discussion and conclusions GDP) contribute to better outcomes of health and well- The study investigated the impact of socio-economic, being. Finally, it is important to note that, with any analy- demographic and climate variables on the distribution sis dealing with regional data, results should be taken of dengue. Its original contribution is that it selected fac- with caution because of issues of scale and uncertainty tors shown to predict dengue at a local level and tested introduced by the aggregation procedure. Further studies whether the association could be generalized to the seeking to test the robustness of the indicators examined regional or state level. In addition, it showed the potential in this study should try to source data at a more refined development of more sophisticated socio-economic indi- scale, and test how these indicators can generalise across cators using regional and internationally available data. the different scales. The study identified which regions are most at risk, by estimating where dengue vectors are likely to occur given Supplementary information Supplementary information accompanies this paper at https​://doi. their suitability to climate conditions in terms of recep- org/10.1186/s1294​2-020-00241​-1. tivity and vulnerability. By estimating the chance of a vec- tor occurring in a region, we could then assess the impact Additional file 1. Species distribution methods and results to estimate of socio-economic, demographic and climate factors on receptivity + model diagnostics (all models). the incidence of dengue. The results confirmed a strong association between our novel indices of socio-economic Abbreviations factors and dengue cases per region. Such results are GDP: Gross Domestic Product; Mex: Mexico; US: United States of America; consistent with the findings reported by [12, 14, 39–42, SDM: Species distribution model; ADM: Administrative division. 45, 61]. Two main lessons can be drawn from this study: Acknowledgements first, while higher GDP is generally associated with a This research was funded by ICTA’s Maria de Maeztu Unit of Excellence, drop in the incidence of dengue, a more granular analy- awarded by the Spanish Ministry of Economy and Competitiveness. The award is the highest institutional recognition of scientific research in Spain. Thanks sis revealed that the crucial factors are a rise in education also to Patrizia Ziveri and Pedro Manuel Gonzalez Hernandez for supporting (with fewer jobs in the primary sector) and better access the project. to information or technological infrastructure. For this Authors’ contributions reason, the use of more sophisticated measures, aside MW led the work and was responsible for the conceptualization of the project, from GDP, should be taken into account when building data curation, data processing, formulation of the methodology, statistical models that try to predict disease distribution. The use analysis, modelling and writing the original draft and interpreting the results. Watts et al. Int J Health Geogr (2020) 19:44 Page 14 of 15 PK made substantive contributions to writing and formulating some of the 10. Ebi KL, Nealon J. Dengue in a changing climate. Environ Res. concepts and interpretation of the results. All contributors revised the manu- 2016;151:115–23. https​://doi.org/10.1016/j.envre​s.2016.07.026. script and copy-edited the final submission version. All the authors were also 11. Bouzid M, Colon-Gonzalez FJ, Lung T, Lake IR, Hunter PR. Climate involved in revising the manuscript critically for important intellectual content. change and the emergence of vector-borne diseases in Europe: CUB supervised statistical analysis and made substantive contributions to for- Case study of dengue fever. BMC Public Health. 2014. https​://doi. mulating some of the statistical models. PK, PGM VSM supervised the project org/10.1186/1471-2458-14-781. and are responsible for formulating ideas for the umbrella project Impacts 12. Brunkard JM, Lopez JLR, Ramirez J, Cifuentes E, Rothenberg SJ, Hunsper- of Climate Change (CC) on Human Health (HH) at ICTA-UAB: Integrating ger EA, et al. Dengue fever seroprevalence and risk factors, Texas-Mexico socio-economic and policy studies with natural science studies to enhance border, 2004. Emerg Infect Dis. 2007;13:1477–83. https​://doi.org/10.3201/ consilience of climate policy science. All authors read and approved the final eid13​10.06158​6. manuscript. 13. Abelz A, Smith B, Fournier M, Betz T, Gaul L, Robles-Lopez JL, et al. Den- gue hemorrhagic fever - us-mexico border, 2005. Morbidity and Mortality Availability of data and materials Weekly Report. 2007;56:785–9. The R project folder, main spatial dataset and R code for the project is available 14. Ramos EF. Hemoterapia e febre dengue. Revis Brasil de Hematol e Hemot. from https​://doi.org/10.5281/zenod​o.88790​9. 2008;30:64–6. https​://doi.org/10.1590/s1516​-84842​00800​01000​16. 15. Magori K, Drake JM. The population dynamics of vector-borne diseases. Ethics approval and consent to participate Book. 2013. Not applicable. 16. Vincenti-Gonzalez MF, Grillet ME, Velasco-Salas ZI, Lizarazo EF, Amarista MA, Sierra GM, et al. Spatial analysis of dengue seroprevalence and Consent for publication modeling of transmission risk factors in a dengue hyperendemic city of All authors and co-authors have approved the publication. Venezuela. PLOS Neglect Trop Dis. 2017. https​://doi.org/10.1371/journ​ al.pntd.00053​17. Competing interests 17. Toan DTT, Hoat LN, Hu W, Wright P, Martens P. Risk factors associated with No competing interests. an outbreak of dengue fever/dengue haemorrhagic fever in Hanoi, Viet- nam. Epidemiol Infect. 2015;143:1594–8. https​://doi.org/10.1017/s0950​ Author details 26881​40026​47. 1 Institute of Environmental Science and Technology (ICTA), Autonomous 18. Tipayamongkholgul M, Lisakulruk S. Socio-geographical factors in vulner- University of Barcelona (UAB), Bellaterra, Spain. 2 Joint Research Centre (JRC- ability to dengue in Thai villages: a spatial regression analysis. Geospat Seville), European Commission, Seville, Spain. 3 Servei de Sanitat Vegetal, DARP, Health. 2011. https​://doi.org/10.4081/gh.2011.171. Generalitat de Catalunya, Av. Meridiana, 38, 08018 Barcelona, Spain. 4 Barce- 19. Teurlai M, Menkès CE, Cavarero V, Degallier N, Descloux E, Grangeon J-P, lona Laboratory for Urban Environmental Justice and Sustainability (BCNEJ), et al. Socio-economic and climate factors associated with dengue fever Institute of Environmental Science and Technology (ICTA), Autonomous spatial heterogeneity: a worked example in new Caledonia. PLoS Neglect University of Barcelona (UAB), Bellaterra, Spain. 5 Department of Geography, Trop Dis. 2015;9:e0004211. https​://doi.org/10.1371/journ​al.pntd.00042​11. Autonomous University of Barcelona (UAB), Bellaterra, Spain. 20. Akter R, Naish S, Hu W, Tong S. Socio-demographic, ecological factors and dengue infection trends in Australia. PLoS ONE. 2017;12:e0185551. Received: 2 June 2020 Accepted: 19 October 2020 21. Robert C, Kubiszewski I, Giovannini E, Lovins H, McGlade J, Pickett K, et al. Time to leave GDP behind. Nature. 2014;505. 22. Stiglitz JE, Sen A, Fitoussi J-P. Mismeasuring our lives: Why GDP doesn’t add up. New York: The New Press Book; 2010. 23. Bleys B. Beyond GDP: Classifying alternative measures for progress. Soc References Indicat Res. 2012;109:355–76. 1. Murray NEA, Quam MB, Wilder-Smith A. Epidemiology of dengue: past, 24. Van den Bergh JC. The GDP paradox. J Econ Psychol. 2009;30:117–35. present and future prospects. Clin Epidemiol. 2013;5:299. https​://www. 25. Costanza R, Kubiszewski I, Giovannini E, Lovins H, McGlade J, Pickett KE, dovep​ress.com/getfi​le.php?fileI​D=17199​. et al. Development: time to leave GDP behind. Nat News. 2014;505:283. 2. Anosike JC, Nwoke BE, Okere AN, Oku EE, Asor JE, Emmy-Egbe IO, et al. 26. Navarro V. Assessment of the world health report 2000. Lancet. Epidemiology of tree-hole breeding mosquitoes in the tropical rainforest 2000;356:1598–601. of Imo state, south-east Nigeria. Ann Agric Environ Med. 2007;14:31–8. 27. Berkman LF, Kawachi I, Glymour MM. Social epidemiology. Oxford: Oxford 3. Gomez-Dantes H, Ramsey Willoquet J. Dengue in the Americas: University Press Book; 2014. challenges for prevention and control. Cadernos De Saude Publica. 28. WHO. A framework for malaria elimination. Geneva: WHO; 2017. 2009;25:S19–31. https​://doi.org/10.1590/s0102​-311x2​00900​13000​03. 29. Brady OJ, Johansson MA, Guerra CA, Bhatt S, Golding N, Pigott DM, et al. 4. WHO. Dengue and severe dengue. 2020. https​://www.who.int/news- Modelling adult Aedes aegypti and Aedes albopictus survival at different room/fact-sheet​s/detai​l/dengu​e-and-sever​e-dengu​e. temperatures in laboratory and field settings. Parasit Vectors. 2013;6:351. 5. WHO. Immunization, vaccines and biologicals. 2017; 2018. http://www. https​://doi.org/10.1186/1756-3305-6-351. who.int/immun​izati​on/resea​rch/devel​opmen​t/dengu​e_vacci​nes/en/. 30. Gubler DJ. Dengue, urbanization and globalization: the unholy trinity of 6. Butterworth MK, Morin CW, Comrie AC. An analysis of the potential the 21(st) century. Trop Med Health. 2011;39(4 Suppl):3–11. https​://doi. impact of climate change on dengue transmission in the southeastern org/10.2149/tmh.2011-S05. United States. Environ Health Perspect. 2017;125:579–85. https​://doi. 31. Lana RM, da Costa Gomes MF, de Melo Lima TF, Honorio NA, Codeco CT. org/10.1371/journ​al.pntd.00072​13. The introduction of dengue follows transportation infrastructure changes 7. Ryan SJ, Carlson CJ, Mordecai EA, Johnson LR. Global expansion and in the state of acre, brazil: A network-based analysis. PLOS Neglect Trop redistribution of aedes-borne virus transmission risk with climate change. Dis. 2017. https​://doi.org/10.1371/journ​al.pntd.00060​70. PLOS Neglect Trop Dis. 2019;13:e0007213. https​://doi.org/10.1371/journ​ 32. Lana RM, da Gomes MFC, de Lima TFM, Honório NA, Codeço CT. The al.pntd.00072​13. introduction of dengue follows transportation infrastructure changes in 8. Messina JP, Brady OJ, Golding N, Kraemer MUG, Wint GRW, Ray SE, et al. the state of acre, brazil: A network-based analysis. PLOS Neglect Trop Dis. The current and future global distribution and population at risk of 2017;11:e0006070. https​://doi.org/10.1371/journ​al.pntd.00060​70. dengue. Nat Microbiol. 2019;4:1508–15. https​://doi.org/10.1038/s4156​ 33. Begon M. Ecological epidemiology. In: The Princeton guide to ecology. 4-019-0476-8. Princeton University Press; 2009. http://www.jstor​.org/stabl​e/j.ctt7t​14n. 9. Xu Z, Bambrick H, Frentiu FD, Devine G, Yakob L, Williams G, et al. 34. Moore CG, Cline BL, Ruiz-Tibén E, Lee D, Romney-Joseph H, Rivera-Correa Projecting the future of dengue under climate change scenarios: E. Aedes aegypti in Puerto Rico: environmental determinants of larval progress, uncertainties and research needs. PLOS Neglect Trop Dis. abundance and relation to dengue virus transmission. Am J Trop Med 2020;14:e0008118. https​://doi.org/10.1371/journ​al.pntd.00081​18. Hyg. 1978;27:1225–31. Watts et al. Int J Health Geogr (2020) 19:44 Page 15 of 15 35. Mohammed A, Chadee DD. Effects of different temperature regimens on 48. C. Vega G, Pertierra LR, Olalla-Tárraga MÁ. MERRAclim, a high-resolution the development of Aedes aegypti (l.)(Diptera: Culicidae) mosquitoes. Acta global dataset of remotely sensed bioclimatic variables for ecologi- Trop. 2011;119:38–433. cal modelling. Sci Data. 2017;4:170078. https​://doi.org/10.1038/sdata​ 36. Tun-Lin W, Lenhart A, Nam VS, Rebollar-Tellez E, Morrison AC, Barbazan .2017.78https​://www.natur​e.com/artic​les/sdata​20177​8#suppl​ement​ P, et al. Reducing costs and operational constraints of dengue vector ary-infor​matio​n. control by targeting productive breeding places: A multi-country non- 49. FAO-UN. Global administrative unit layers (gaul). 2014. http://www.fao. inferiority cluster randomized trial. Trop Med Int Health. 2009;14:1143–53. org/geone​twork​/srv/en/metad​ata.show?id=12691​. https​://doi.org/10.1111/j.1365-3156.2009.02341​.x. 50. CPC/NCEP. National center for atmospheric research. 1987. http://rda. 37. Watts DM, Burke DS, Harrison BA, Whitmire RE, Nisalak A. Effect of tem- ucar.edu/datas​ets/ds512​.0/. perature on the vector efficiency of Aedes aegypti for dengue 2 virus. Am 51. SEDAC. Gridded population of the world, version 4 (gpwv4): Population J Trop Med Hygiene. 1987;36:143–52. count, revision 11. 2018. https​://doi.org/10.7927/H4JW8​BX5. 38. Thammapalo S, Chongsuwiwatwong V, McNeil D, Geater A. The climatic 52. OECD. Regional statistics and indicators database. 2018. http://stats​.oecd. factors influencing the occurrence of dengue hemorrhagic fever in org/Index​.aspx?DataS​etCod​e=REGIO​N_DEMOG​R. Thailand. Southeast Asian J Trop Med Public Health. 2005;36. 53. GFC. Spatial data analysis and modeling with r. 2018;2018. http://rspat​ial. 39. Stewart Ibarra AM, Ryan SJ, Beltrán E, Mejía R, Silva M, Muñoz Á. Dengue org/index​.html. vector dynamics (Aedes aegypti) influenced by climate and social factors 54. Aswi A, Cramb SM, Moraga P, Mengersen K. Bayesian spatial and spatio- in Ecuador: implications for targeted control. PLOS ONE. 2013;8:e78263. temporal approaches to modelling dengue fever: a systematic review. https​://doi.org/10.1371/journ​al.pone.00782​63. Epidemiol Infect. 2018;147:1–14. https​://doi.org/10.1017/S0950​26881​ 40. Thammapalo S, Chongsuwiwatwong V, Geater A, Lim A, Choomalee K. 80028​07. Socio-demographic and environmental factors associated with Aedes 55. Wood SN. Generalized additive models: An introduction with r. Boca breeding places in Phuket, Thailand. Southeast Asian J Trop Med Public Raton: CRC Press; 2017. Health. 2005;36:426–33. 56. Gluskin RT, Johansson MA, Santillana M, Brownstein JS. Evaluation of 41. Qi X, Wang Y, Li Y, Meng Y, Chen Q, Ma J, et al. The effects of socioeco- internet-based dengue query data: Google dengue trends. Plos Neglect nomic and environmental factors on the incidence of dengue fever in Trop Dis. 2014. https​://doi.org/10.1371/journ​al.pntd.00027​13. the pearl river delta, China, 2013. PLOS Neglect Trop Dis. 2015;9:e0004159. 57. Romero-Alvarez D, Parikh N, Osthus D, Martinez K, Generous N, del Valle S, https​://doi.org/10.1371/journ​al.pntd.00041​59. et al. Google health trends performance reflecting dengue incidence for 42. Clark GG. Dengue and dengue hemorrhagic fever in northern Mexico the Brazilian states. BMC Infect Dis. 2020;20:252. https​://doi.org/10.1186/ and south Texas: Do they really respect the border? Am J Trop Med Hyg. s1287​9-020-04957​-0. 2008;78:361–2. 58. Nakano K. Future risk of dengue fever to workforce and industry through 43. Khun S, Manderson L. Health seeking and access to care for children with global supply chain. Mitigat adapta strateg Glob Change. 2018;23:433– suspected dengue in Cambodia: an ethnographic study. BMC Public 49. https​://doi.org/10.1007/s1102​7-017-9741-4. Health. 2007;7:262. 59. Jakobsen F, Nguyen-Tien T, Pham- Thanh L, Bui VN, Nguyen-Viet H, Tran- 44. Unlu I, Farajollahi A, Strickman D, Fonseca DM. Crouching tiger, hidden Hai S, et al. Urban livestock-keeping and dengue in urban and peri-urban trouble: urban sources of Ades albopictus (Diptera: Culicidae) refractory to Hanoi, Vietnam. PLOS Neglect Trop Dis. 2019;13:e0007774. https​://doi. source-reduction. PloS ONE. 2013;8:e77999–9. https​://doi.org/10.1371/ org/10.1371/journ​al.pntd.00077​74. journ​al.pone.00779​99. 60. Cabrera M, Taylor G. Modelling spatio-temporal data of dengue fever 45. Elsinga J, Lizarazo EF, Vincenti MF, Schmidt M, Velasco-Salas ZI, Arias L, using generalized additive mixed models. Spat Spatio Temp Epidemiol. et al. Health seeking behaviour and treatment intentions of dengue 2019;28:1–13. https​://doi.org/10.1016/j.sste.2018.11.006. and fever: a household survey of children and adults in venezuela. 61. Brunkard JM, Cifuentes E, Rothenberg SJ. Assessing the roles of tempera- PLOS Neglect Trop Dis. 2015;9:e0004237. https​://doi.org/10.1371/journ​ ture, precipitation, and enso in dengue re-emergence on the texas- al.pntd.00042​37. mexico border region. Salud Publica De Mexico. 2008;50:227–34. 46. ACAPS. ACAPS briefing note: Mexico - dengue fever (16 september 2019). 2019. https​://relie​fweb.int/repor​t/mexic​o/acaps​-brief​i ng-note-mexic​ o-dengu​e-fever​-16-septe​mber-2019.. Publisher’s Note 47. Kraemer MUG, Sinka ME, Duda KA, Mylne A, Shearer FM, Brady OJ, et al. Springer Nature remains neutral with regard to jurisdictional claims in pub- The global compendium of Aedes aegypti and Ae. Albopictus occurrence. lished maps and institutional affiliations. Sci Data. 2015;2:150035. https​://doi.org/10.1038/sdata​.2015.35. Ready to submit your research ? Choose BMC and benefit from: • fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations • maximum visibility for your research: over 100M website views per year At BMC, research is always in progress. Learn more biomedcentral.com/submissions