Influence of socioeconomic, geographic, climatic, and infrastructure factors with dengue fever in Colombia from 2015 to 2020: A Spatial Generalized Additive Mixed Model
School of Medicine, Universidad El Bosque, Bogotá, Colombia
School of Medicine, Universidad de Los Andes, Bogotá, Colombia
Department of Global Public Health, Karolinska Institutet, Stockholm, Sweden
Otolaryngology and Allergy Research Groups, UNIMEQ-ORL, Bogotá, Colombia
Department of epidemiology, Instituto Mexicano del Seguro Social, San Luis de Potosí, México
Department of Otolaryngology, Fundación Santa Fe de Bogotá, Bogotá, Colombia
*Corresponding author Email: sm.morenoluniandes@gmail.com (SML).Abstract
Background
Dengue virus infection remains among the most prevalent infectious diseases globally, with an estimate of 100–400 million cases occurring each year. Prior ecological studies have documented spatial and temporal correlations between dengue occurrence and either socioeconomic or climatic conditions independently. The aim of this study was to analyze the association between multidimensional typologies, including socioeconomic, geographic, climatic, and infrastructure factors and the rate of dengue cases in Colombia.
Methods
An analytical, observational, ecological study with repeated measures that included six years of aggregated quantitative data study from 2015 to 2020, based on national registry data from several sources. A spatial Generalized Additive Mixed Model (GAMM) was used to identify the probable factors associated with the frequency of dengue in Colombia. Socioeconomic, infrastructure, geographic, and climatic factors included Multidimensional Poverty Index (MPI), altitude, population density, road density, education and public services coverage, altitude, temperature, precipitation, relative humidity, and presence of El Niño-Southern Oscillation (ENSO) in the period of analysis.
Results
During the study period, 448,774 cases of dengue fever were reported in Colombia, with overall monthly median of 6280 cases (IQR: 2842-9278). The highest number of cases was reported in 2019 with 118,956 cases. Higher levels of drinkable water service coverage, education access coverage, and relative humidity were negatively associated with dengue frequency, showing lower case frequency. While higher GDP per capita and greater participation in the subsidized health system were associated with higher incidence of dengue, although the latter showed substantial variability in extreme values. Precipitation and maximum and minimum temperatures showed unimodal associations, with risk concentrated at intermediate values, while altitude and population density revealed complex multimodal patterns. Municipal performance, a composite index that assesses the efficiency and management capacity of local governments, showed a negative association with dengue occurrence, indicating lower risk in better-performing municipalities.
Conclusions
A complex interaction was found between socioeconomic, geographic, climatic, and infrastructure factors. These findings may highlight the need for multidimensional approaches in dengue prevention and control policies, that consider the specific conditions of each community. Research is required to better understand the factors that affect the dynamics of dengue. Factors identified at this local level may be useful in different geographical regions, especially in the context of climate change related vulnerabilities.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
The author(s) received no specific funding for this work.
Introduction
The World Health Organization (WHO) estimates 100-400 million dengue virus infections occurring each year, with a mortality rate of 2.5% [1]. Currently, the medical treatment for dengue is based on symptomatic management, but timely detection and access to medical care can potentially decrease mortality rates to less than 1% [1,2]. By the first half of 2020, more than 1.6 million of dengue cases had been reported in Latin America, with higher frequency rates in Brazil (1,040,481 cases); Paraguay (218,798 cases); Bolivia (82,460 cases); and Argentina (79,775 cases) [2]. In 2023, Colombia reported more than 150,000 cases, and an increase in the incidence of dengue fever, with 387,2 cases per 100,000 inhabitants [3]. Likewise, recent studies have highlighted the increase in the incidence of this disease worldwide, despite calls for vector management, prevention, and control [4]. By the end of 2016, a total of 291,964 cases of dengue associated with outbreaks had been reported, of which 72.4% of dengue patients were reported in the Western Pacific region, followed by the Americas region (19.4%), the Southeast Asia region (4.8%), the Eastern Mediterranean region (1.5%), the European region (1.5%), and the African region (0,3%) [5].
In the last decades, the increase in dengue incidence and the unavailability of treatments for advanced stages have increased its global burden and represent a significant economic impact on affected countries. This scenario is particularly important in regions of the world that are most vulnerable to climate change, such as Asia, Africa, and Latin America, which share socioeconomic characteristics and climatic factors [6,7]. Regarding the average annual costs of dengue in Latin America, this amount is estimated to be $3 billion in both direct and indirect costs and losses in Disability-Adjusted Life Years (DALYs) [8,9]. On the other hand, the reporting and surveillance of dengue cases globally is limited in regions such as sub-Saharan Africa or the Eastern Mediterranean [1,10]; regions where an increase in the dissemination of Aedes mosquitoes is expected in the next 50 years [11,12].
Moreover, the increase in the global incidence of dengue has been attributed, in part, to infrastructural factors related to social inequality such as the lack of access to public services and the decrease in the implementation of preventive health programs [13,14], as well as high levels of poverty [15,16], governmental deficiencies, and low institutional effectiveness [6,17]. Overall, this scenario is also exacerbated by the climatic variability mainly attributed to climate change [11,18,19]. Current literature discusses the complex interactions among socioeconomic, geographical, climatic factors that may influence both the biology, reproductive dynamics, and survival of the vector, as well as the infrastructure-related and intermediate social determinants of the populations that are most vulnerable to this disease [20,21]. It is crucial to identify how the combined effect of multiple socioeconomic, climatic, and governmental factors affect the dynamics of dengue, as the interaction among them may exacerbate the frequency and widespread patterns of the disease in many territories worldwide [22,23]. Therefore, this study aims to assess the possible impact of socioeconomic, geographic, climatic, and infrastructure factors on the frequency of dengue, all within a framework of vulnerabilities present in the Colombian territory.
Materials and methods Ethics Statement
This study was approved by the Ethics Committee of the Hospital Universitario Fundación Santa FE (CCEI-15374–2023).
Study design
Colombia has a population of 52.6 million inhabitants according to the 2020 national estimate from the National Administrative Department of Statistics (DANE). The country is divided into 32 departments, which are first-level territorial entities, and a total of 1,099 municipalities, which are second-level administrative divisions within departments. Each municipality has local governance structures such as a mayor and municipal council. Dengue surveillance and public health control measures are implemented at both the national (Ministry of Health) and local (municipal health secretariats) levels. Local governments collaborate with both the National Health Institute (INS) and regional health institutions to monitor diseases like dengue [24,25]. In terms of altitude, most of Colombian territory ranges between 1,000 and 2,000 meters above sea level, and its average temperatures range from 11 to 17°C. [26]. In terms of human development, Colombia ranks in the 88th position worldwide and had a Human Development Index (HDI) of 0.756 in 2020, categorizing it as having a middle to high level of human development. However, the country has significant inequalities, exhibiting a Gini index of 0.553 of 2023 (the Gini index determines a nation’s level of income inequality by measuring the income distribution or wealth distribution across its population) [27].
Analytical, ecological, observational study with repeated measures based on data collected from several national sources, including: the Bank of the Republic (central bank of Colombia), the National Ministry of health, the National Administrative Department of Statistics (DANE), the National Planning Department (DNP), and the National Health Institute (INS) [24,26,28]. The study population included the entire population with dengue registered cases at the municipal level during the specified period. Monthly data from January 2015 to December 2020 were analyzed, covering a total of 1,110 municipalities in Colombia.
Data collection of Dengue data
Data on confirmed cases of dengue fever in the 32 departments and municipalities of Colombia were obtained from the National Public Health Surveillance System (SIVIGILA) [29], based on the ICD-10 registry definition of dengue case (A90: Classic dengue and/or A91 Dengue hemorrhagic fever) established by healthcare institution as either confirmed with laboratory testing or a probable case of dengue fever with clinical warning signs such as headache, retro ocular pain, myalgia, arthralgia, rash or rash and also has a history of displacement (up to 15 days before the onset of symptoms), severe and continuous abdominal pain, persistent vomiting, diarrhea, drowsiness and/or irritability, postural hypotension, painful hepatomegaly >2cms, decreased diuresis, drop in temperature, mucosal bleeding, abrupt drop in platelets (<100.000) associated with hemoconcentration or residing in a dengue endemic area [30]. The data used in this study was extracted from this system: the authors gathered all the weekly dengue reports at a municipal/local level based [24,25]. For the analysis, the data were aggregated into monthly case counts per municipality to facilitate temporal and spatial comparisons.
Socioeconomic, geographic, climatic, and infrastructure data
Sociodemographic and economic variables included population density (inhabitants/km²) and educational coverage (%), obtained from the National Administrative Department of Statistics (DANE) [31]; the multidimensional poverty index (MPI) and government-subsidized health coverage (%), obtained from the National Planning Department (NPD) through its Terridata platform [32]; and GDP per capita (millions of dollars), obtained from the Central Bank [33]. On the other hand, infrastructure and access to public services indicators, such as aqueduct coverage (%), education coverage (%), healthcare coverage subsided by the government (%), and access to public services (%), were also obtained from the Terridata platform [32]. Regarding to governance variables, included the Municipal Performance Measure (%) defined as a composite indicator developed by the National Planning Department of Colombia that evaluates the efficiency and management capacity of local governments, covering administrative, fiscal, and planning dimensions (%) were obtained from Terridata platform of NDP [32] while Multidimensional poverty index (MPI) incidence (%) were obtained from the National Administrative Department of Statistics (DANE) [31].
Geographic and cartographic information were obtained from the DANE Geoportal public website [34]. Climatic and environmental data were provided by World Bank Group (WBG) in the Climate Change Knowledge Portal (CCKP), what is a hub for climate-related information for the WBG [35] or purposes of provides an online platform from which access and analyze comprehensive data related to climate change and development, [35] from this platform were obtained data from monthly maximum temperatures, monthly minimum temperatures, monthly relative humidity and total monthly rainfall. On the other hand, the National Oceanic and Atmospheric Administration (NOAA) provided information of the ENSO phenomenon by the period of analysis [36].
Most variables were collected on a monthly or annual basis depending on their reporting frequency and were available at the municipal level, except for GDP per capita and climatic variables, which were reported at the departmental level.
Statistical analysis
Statistical analysis was conducted using R 4.2.0 and Stata 17MP software. About the descriptive analysis, absolute and relative frequencies were calculated for the qualitative variables. Measures of central tendency (average and median) were estimated for the quantitative variables. Standard deviation and interquartile range were assessed for the dispersion measures. These analyses were performed stratified by the years of analysis included in the study.
For purpose of the analysis, all municipalities with at least one case reported of dengue in the sample were included. With the aim of comparing number of cases of dengue reported during the study analysis period at the territorial level, we visualized the spatiotemporal patterns of dengue in Colombia, using choropleth maps based on quantiles groups, using ‘spmap’ command in Stata for this purpose. The geographic base layer used for the maps (Colombian municipal boundaries) was sourced from the publicly available shapefile provided by the DANE geoportal. The shapefile used is available from: https://geoportal.dane.gov.co/
To analyze the relationship among socioeconomic, geographic, climatic, and infrastructure data factors with number of cases reported in the country between 2015-2020 data was analyzed using a Generalized Additive Mixed Model (GAMM), the estimations were conducted using the ‘mgcv’ package. A GAMM model was estimated to study the expected number of cases 𝑌𝑖𝑗𝑡 for each territory 𝑖 in each time 𝑗, assuming that 𝑌𝑖𝑗𝑡 ∼ 𝑁𝑒𝑔𝐵𝑖𝑛(𝜇𝑖𝑗𝑡, 𝜃), with 𝜃 the dispersion parameter of the negative binomial distribution, based on 𝑘 variables as follows:
Where 𝐸𝑖𝑗𝑡 is the municipal population, 𝑧𝑖𝑗𝑡 is the effect of categorical variables (ENSO, municipality category), 𝑓𝑝(.) are smooth functions for the variables included in the analysis in the time 𝑗 trend defined via penalized cubic regressions splines; 𝑓𝑠𝑝(𝐿𝑜𝑛,𝐿𝑎𝑡) is the two-dimensional spatial term
and
are time-specific and department-specific randoms effects to capture the effects of potential unobserved confounders in the model. Considering the high level of dispersion in the number of dengue cases, a negative binomial model was estimated based on the fast restricted maximum likelihood method (fREML). In order to adjust the models to the underlying population at risk, we included the annual municipal population as an offset term in the models, to estimate rates based on the number of events that occur in a defined population during a specified period, divided by the population at risk. In a negative binomial model linked to the logarithm, this compensation limits its coefficient to 1, ensuring that the model estimates rates rather than absolute counts.
The generalized cross-validation or GCV criterion was used to select an appropriate smoothing parameter value for the proposal models. Given the possible spatial autocorrelation structure, a spatial Generalized Additive Mixed Model (spatial GAMM) was fitted to evaluate the association between dengue cases and sociodemographic, climatic and performance government factors accounting for spatial and temporal structure. The model included a smooth function of the MVI, a two-dimensional spline over geographic coordinates (longitude and latitude), and a random effect for year.
The effect of sociodemographic, climatic and performance government variables were adjusted using the following confounding variables identified in the directed acyclic diagram (DAG, see supplementary material Fig. S1): altitude, population density, aqueduct coverage, education coverage, subsidized survey coverage, public services coverage, municipal performance measurement, precipitation, maximum and minimum temperature, relative humidity, Multidimensional Poverty Index (MPI) and GDP per capita. These variables were used as covariates, based on previous literature, what reported and the effect of the climate variables on dengue fever, the impact of socioeconomic conditions on dissemination of vector-borne diseases and evidence associated with the role of performance institutional in the control of Aedes.
Several models were built to adjust the sociodemographic, climatic and performance government variables according to the multilevel component of the information, the performance of these models was evaluated using Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC) and ANOVA test. The smallest AIC/BIC indicates the model with the best fit. Moreover, a diagnostic analysis of the models was performed based on simulated residual diagnostics from the ‘DHARMa’ R package [37], evaluating residual uniformity (Kolmogorov-Smirnov test), overdispersion, and outliers. Additionally, QQ plots and residuals versus predicted plots were inspected to evaluate potential systematic deviations. The spatial GAMM estimations were conducted using the ‘mgcv’ R package [38], applying penalized cubic regression splines to allow for flexible smoothing of the relationship between selected variables and dengue occurrence over time. Statistical analysis was conducted using R 4.1.1 and Stata 17 MP software.
Results
Study population
A total of 1,100 municipalities were included in the study. Table 1 describes the characteristics of these territories. The median altitude above sea level was 1,010 meters (range: 1.00-3,850 meters). The median values of health and education coverage indicators were above 50%, contrasting with the median water supply coverage, which was below this value. The trend of dengue cases is illustrated in Figure 1. During the study period, a total of 421798 dengue cases were reported in Colombia. At the national level, the annual number of cases showed marked interannual variation, with a median of 69688 cases per year (IQR: 43820–95288) and a peak of 118956 cases in 2019. Across municipalities, the annual median number of reported cases was 6280 (IQR: 2842–9278), reflecting the high spatial heterogeneity of dengue transmission. The highest numbers of cases were reported in 2019 (118956) and 2016 (91378) (red color in Fig 1), whereas the lowest occurred in 2017 (24448). (blue color in Fig 1). Between 2015 and 2020, dengue incidence showed marked interannual fluctuations, peaking in 2015 and 2019. Socioeconomic and municipal performance indicators improved steadily, while the multidimensional poverty index decreased over time. These results are summarized in Table 1.
Figure 2 shows the spatial and temporal patterns in the incidence of dengue for all municipalities in Colombia during 2015–2020. Most cases occurred in the north and southwest coast region, with the highest incidence in the Caribbean region. The frequency in southwest part of the coast region and eastern region increased in 2016 and 2020 compared with that in the previous years.
Spatial GAMM Analysis
Table 2 presents the estimates of the Negative Binomial Spatial GAMM for the number of dengue cases reported. Final model was selected after comparing alternative specifications, including models with fixed effects for department and year, models without the spatial component, and models with different random effect structures. Model selection was based on Akaike’s Information Criterion (AIC) and Bayesian Information Criterion (BIC), with the chosen specification yielding the lowest AIC and BIC values. The selected model explained 54% of the deviance in dengue cases. The high values of the effective degrees of freedom (edf) for the smooth functions indicate strong nonlinear associations between dengue incidence and sociodemographic, climatic, and governance-related variables; most of these findings were statistically significant.
Figures 3, 4 and 5 show the relationships estimated by the final model. Altitude showed a multimodal pattern, with higher risk at low and medium altitudes (<750 m.a.s.l), a reduction around 1000–2800 m.a.s.l, and an uptick toward >3000 m.a.s.l (Figure 3a). Population density displayed a bimodal relationship, with peaks around 2000 and 10,000 P/km², followed by declines at higher levels (Fig. 3b). Multidimensional Poverty Index (MPI) showed irregular oscillations without a consistent gradient, although incidence appeared higher in territories with 35–65% of people in multidimensional poverty (Fig. 3c). Finally, GPD per capita (Fig. 3d) show directly proportional trends, where dengue fever is concentrated in areas with high GDP levels, particularly concentrated in territories above 9000 USD.
Figure 4 shows the climatic component in the analysis: rainfall was associated with the highest incidence around 500–700 mm. Maximum and minimum temperatures presented unimodal associations, with higher incidence at intermediate ranges (25–28 °C and 21–24 °C, respectively), and lower incidence at extreme values (Figs. 4a, 4b, 4c) showing an increase in dengue frequency towards the middle of the distribution and gradually decreasing at that point. Relative humidity showed a negative, nearly linear association, with decreasing dengue occurrence at higher humidity levels (Fig. 4d).
Figure 5 shows the government performance in the analysis, water service coverage exhibited a negative, linear pattern, with lower incidence at higher coverage (Fig. 5a). Educational coverage showed a negative relationship (Fig. 5b), whereas subsidized health coverage showed a positive association, albeit with wide uncertainty at extreme values (Fig. 5c). Access to public services displayed a bimodal pattern, with higher incidence at intermediate levels (40–70%) and lower incidence at higher levels (Fig. 5e). Finally, municipal performance showed a marked negative relationship, with better-performing municipalities consistently exhibiting lower reported dengue cases (Fig. 5f).
On the other hand, differences between municipality category (Small versus Intermediate municipality: p value<0.01) and ENSO dynamic (p value<0.01) were found in the frequency of dengue for the period of analysis (Table 2). These patterns suggest that the level of environmental, socioeconomic, and governmental vulnerability jointly influence the risk of dengue in Colombia, highlighting that climatic factors alone do not fully explain incidence trends.
The DHARMa diagnostics indicated adequate dispersion (p = 0.108), but significant deviations from uniformity and presence of outliers (p<0.05). Nonetheless, residuals vs. predicted plots showed no major systematic biases, supporting overall model adequacy (Fig. 6).
Discussion
This study aimed to determine the relationship between the frequency of dengue fever and socioeconomic, geographic, climatic, and infrastructure data factors in Colombia between January 2015 and December 2020. The interannual variation in dengue incidence, with peaks in 2015 and 2019, aligns with previously reported national epidemic cycles in Colombia. Given the complex interaction between the socioeconomic, geographic, and infrastructure factors and the dynamics of dengue, spatial GAMM models provide more flexible modeling to analyze the spatiotemporal dynamics of dengue, allowing the inclusion of non-linear relationships of the different factors included in the model. [39]. The results showed that the incidence of dengue is strongly associated with climatic factors, with a nonlinear association between rainfall and temperature, and with complex socioeconomic and infrastructure conditions. The spatial GAMM approach allowed us to capture temporal and spatial structure relationships and reduce residual spatial autocorrelation compared to conventional models.
Access to public services, such as water, was associated with a decrease in dengue case frequency, which may highlight the need for improved access to basic services in affected communities as suggested by prior studies [40,41]. Moreover, the study highlighted the links between multidimensional poverty index and dengue incidence, with higher prevalence in areas with higher poverty levels in line with prior studies describing this association [42,43]. The study also identified education and information access as crucial factors in dengue prevention and control, along with an increase in the GDP. Similarly, a prior study described that education and access to information are crucial for dengue prevention and control, and gains in GDP can reinforce these measures by improving living conditions, health infrastructure, and vector control capacity [44]. On the other hand, the results show a significant influence of climatic components (temperature, humidity, and rainfall) on the frequency of dengue in Colombia. Precipitation showed a unimodal association, with a higher incidence around 500-700 mm, which may be related with the optimal conditions for Aedes reproduction due to water storage in containers and the accumulation of standing water, while very heavy rainfall can wash away breeding sites, reducing mosquito populations [45,46]. Maximum and minimum temperatures also showed concave associations, with the highest risk in the range of 26-28 °C and 22-24 °C, respectively, consistent with the optimal temperature ranges for mosquito survival and viral replication described in the literature [47,48]. These non-linear relationships may be explained by the biology of the vector, as higher precipitation rates combined with higher temperatures result in increased humidity, which is associated with an increase in the feeding activity, survival, and egg development of Aedes aegypti [49].
Likewise, temperatures play a crucial role in vector dissemination, as Aedes mosquitoes depend on specific temperature ranges between 21°C and 31°C for survival and reproduction [50,51]. Precipitation also plays a fundamental role in increasing dengue incidence. Rainfall ranges between 400mm and 500mm increase the risk of dissemination by creating new reservoirs that allow vector oviposition in natural environments [46]. The relationship with relative humidity was negative, showing that high values reduce the incidence of dengue, possibly due to lower survival rates of eggs and larvae in saturated environments [46]. These conditions of humidity, precipitation, and temperature make the presence of Aedes widespread in tropical and subtropical environments. In these regions, temperature and humidity levels, combined with prolonged rain cycles, create an ideal environment for the mosquito’s life cycle [52]. These findings are consistent with other Latin American studies, highlighting the importance of climatic control and surveillance in territories most vulnerable to dengue dissemination in the context of climate change [53,54]. The impact of climatic vulnerability on dengue risk is also consistent with results from other global studies [55].
Regarding sociodemographic and governance-related vulnerabilities, strong non-linear relationships were observed with government performance aspects, water service coverage, sewerage, education, and health coverage. For water supply, a polynomial trend was observed rather than a decrease in dengue frequency in territories with higher coverage. This finding contrasts with previous evidence that showed an inverse relationship, with higher dengue dissemination in areas with less water access due to increased artificial reservoirs from water storage [56]. This implies an increase in breeding sites for mosquitoes, both in natural puddles formed by rain and in household water storage in areas affected by heat waves [18,57]. However, this effect can be modified by other factors such as inequity, income, and ineffective public health policies [50]. The effects of socioeconomic determinants in various Latin American countries highlight the biophysical, political-institutional, and community risks for dengue dissemination, addressing marginalization and inequity issues in primary health care and their consequences in terms of living conditions, access to public services, and education [58]. These findings are similar in other regions of the world, where the study of social inequality, poverty, and inequity effects on dengue dissemination suggests integrated efforts beyond conventional vector control strategies, aiming to control the dengue threat through improvements in socioeconomic conditions, income, and education. [59,60].
The approach of assessing climatic vulnerabilities jointly with social and economic factors is crucial to inform academia and policymakers with valuable information that facilitates climate adaptation and coping strategies at various scales [17,61,62]. The degree of vulnerability to climate change is differential and attributable to households’ socioeconomic variations and access to basic services [61,63]. Considering these data and the global spread of the Aedes Aegypti [57,64], vector control emerges as one of the most effective methods to mitigate new dengue epidemics. However, this measure has proven insufficient [65,66]. Evidence suggests implementing integrated preventive strategies to eliminate breeding sites, educate at-risk populations, disseminate climate information, and identify territories with high socioeconomic and climatic vulnerability. [67]. In the context of accelerating climate change, the nonlinear climate-dengue relationships identified in this analysis may have important implications. The unimodal associations suggest that both increases and decreases in precipitation or temperatures beyond optimal ranges could paradoxically reduce transmission in some areas while increasing it in others. However, changes in climate variability and extreme events may have effects beyond those captured by monthly averages. Future research should explore how climate change projections interact with socioeconomic development pathways to forecast dengue risk.
By incorporating spatial smoothing and random effects for year and department, spatial GAMM provided several advantages over conventional regression, as mentioned in the literature in terms of modeling complex relationships, predictive power, geographic connectivity, and model variations in true disease pattern [73]. It allowed flexible modeling of non-linear associations without imposing restrictive assumptions, accounted for residual spatial autocorrelation, and captured heterogeneity across territorial analysis units [68]. This approach complements earlier ecological studies that relied on linear or Poisson models, providing a more nuanced understanding of the climatic and social risk for dengue transmission. These methodological advances can support more accurate local predictions and better-targeted interventions. Moreover, the results presented in this study may have important implications for dengue management in Colombia and similar endemic contexts. Monitoring and analyzing dengue transmission windows presents opportunities for a climate-based early warning system, especially during “El Niño” seasons, when frequency was higher. Secondly, the nonlinear relationships with infrastructure and governance suggest that investment in improving water or sanitation services may not be sufficient and that community-level interventions are needed to mitigate risk. On the other hand, findings related to education coverage and local administration highlight the need to strengthen institutional capacity and community participation as part of integrated vector management strategies. Finally, the relationship with GDP highlights the importance of focusing not only on the most vulnerable territories, but also on major urban areas with high population density and excessive growth, where the risk of dengue remains high despite greater economic development.
Our study had limitations, first, a significant limitation is related to underreporting bias in health statal databases regarding inequities in access to basic services, which has been described in the literature and can be as high as 5% [69]. Additionally, the municipal-level disaggregation of healthcare system affiliation variables can be considered a limitation in this analysis due to difficulties in obtaining more precise data on the existing gaps between the contributory and subsidized regimes. Furthermore, we could not account for population mobility between municipalities, which affects both exposure patterns and case reporting locations. Lack of dengue serotype data limited assessment of viral strain dynamics, and absence of entomological surveillance data prevented direct vector-climate validation. On the other hand, given the ecological design of this study, inference at the individual level cannot be possible and raises the possibility of ecological fallacy. Associations observed at the municipal level may not reflect individual-level relationships, and ecological fallacy remains a possibility. The cross-sectional nature of the aggregated data prevents establishment of temporal precedence for many relationships. Dengue surveillance data may also be affected by underreporting and variability in diagnostic capacity across municipalities. The use of monthly case aggregation may mask within-month transmission dynamics, and the optimal lag structures between climatic exposures and disease outcomes may vary across municipalities. Residual confounding by housing quality and sanitation practices cannot be excluded. Moreover, we used annual population projections as the denominator for incidence estimation, while cases were aggregated monthly. More studies to better understand these dynamics and develop effective interventions to reduce dengue transmission in Colombia. Despite these limitations, the large dataset, spatially explicit modeling, and sensitivity analyses provide confidence in the robustness of the findings.
Overall, the combination of climatic factors, service coverage, municipal performance, and transparency of social development presents an integral scope that reflects the complexity of the challenges Colombia faces in the fight against dengue. Our findings suggest that disparities in access to essential services and climatic variability are intrinsically linked to the prevalence of dengue.
Conclusion
This multidimensional study on dengue in Colombia between 2015 and 2020 reveals a complex interaction between socioeconomic, geographic, climatic, and infrastructure factors. Coverage of public services such as water was found to be associated with a decrease in dengue occurrence, while multidimensional poverty showed more intricate relationships, with higher incidence in areas with higher levels of poverty. Furthermore, increased education and access to information were highlighted as crucial factors, along with an increase in GDP. These findings highlight the need for comprehensive approaches in dengue prevention and control policies, adapted to the specific conditions of each community. More research is required to better understand the dynamics of dengue and develop effective interventions that reduce its burden on the population. Furthermore, factors identified at the local level can be useful predictors at the regional level, which can help anticipate and address vulnerable populations on a broader scale, especially in the context of climate change.
Data Availability
The dengue surveillance data analyzed in this study were obtained from Colombia’s National Public Health Surveillance System (SIVIGILA) and are available through the National Health Institute (INS) upon reasonable request subject to data sharing agreements. Socioeconomic and municipal performance data are publicly available from the National Planning Department (DNP) Terridata platform (https://terridata.dnp.gov.co/). Climatic data are publicly available from the World Bank Climate Change Knowledge Portal (https://climateknowledgeportal.worldbank.org/). The geographic shapefiles are available from the Geoportal of the National Administrative Department of Statistics (DANE), Colombia’s National Statistics Office (https://geoportal.dane.gov.co/). License/terms of use: https://geoportal-dane-gov-co.translate.goog/acerca-del-geoportal/licencia-y-condiciones-de-uso/?_x_tr_sl=es&_x_tr_tl=en&_x_tr_hl=es&_x_tr_pto=wapp#gsc.tab=0. All analysis code is available from the corresponding author upon reasonable request.
Data Availability
The dengue surveillance data analyzed in this study were obtained from Colombia's National Public Health Surveillance System (SIVIGILA) and are available through the National Health Institute (INS) upon reasonable request subject to data sharing agreements. Socioeconomic and municipal performance data are publicly available from the National Planning Department (DNP) Terridata platform (https://terridata.dnp.gov.co/). Climatic data are publicly available from the World Bank Climate Change Knowledge Portal (https://climateknowledgeportal.worldbank.org/). The geographic shapefiles are available from the Geoportal of the National Administrative Department of Statistics (DANE), Colombia's National Statistics Office (https://geoportal.dane.gov.co/). License/terms of use: https://geoportal-dane-gov-co.translate.goog/acerca-del-geoportal/licencia-y-condiciones-de-uso/?_x_tr_sl=es&_x_tr_tl=en&_x_tr_hl=es&_x_tr_pto=wapp#gsc.tab=0. All analysis code is available from the corresponding author upon reasonable request.