Yellow Fever Reemergence in Tolima, Colombia 2024–2025: An Eco-epidemiological Study

Abstract

Background

Yellow fever virus is transmitted by mosquitoes among humans and non-human primates (NHPs) in South America. The 2024–2025 yellow fever (YF) outbreak was notable for its spread into new areas, including the department of Tolima in Colombia's Andean region. We investigated the eco-epidemiology of human and NHP YF cases to understand the patterns and factors associated with the Tolima outbreak.

Method

We collected spatiotemporal, sociodemographic, and mortality data on human YF cases in Tolima, as well as the locations of deceased NHPs with YF. We then conducted exploratory, descriptive, and spatial statistical analyses to identify risk factors and hotspots for YF. We then used inferential spatiotemporal modeling to compare ecological and sociodemographic factors.

Results

From September 2024 to October 2025, 116 human and 53 non-human primate YF cases were detected in Tolima. Among the human cases, 77% were male (89/116), and the median age was 47 (interquartile range 34–63). Of these cases, 45 (39%) died, and older age was associated with increased odds of death (adjusted odds ratio, 1.43 per 10 years; 95% confidence interval, 1.18–1.78). There was significant clustering among human and NHP cases. Higher rainfall and poverty/deprivation were associated with increased YF incidence at the neighborhood level, and rainfall during the outbreak was above average due to La Niña.

Conclusions

These findings demonstrate significant mortality from YF, which reemerged in Tolima after nearly a century. Rural populations with greater ecological risk and poverty/deprivation could benefit from targeted vaccination strategies for YF.

Yellow fever reemerged in Tolima, Colombia, after nearly a century, causing mortality among humans and non-human primates. This outbreak occurred primarily in rural areas with low vaccination coverage, highlighting the need for preparedness in areas with similar risk factors.

During 2024–2025, a yellow fever (YF) outbreak in South America expanded into new regions, causing global concern. Yellow fever is a viral hemorrhagic fever caused by a mosquito-borne flavivirus, yellow fever virus (YFV). In South America, the Amazon region is often considered the primary endemic area for YFV circulation. However, in October 2024, YF cases were reported in Tolima, Colombia, a department in the Andean region, which had no known YF cases in the 21st century and low YF vaccination coverage [1–7]. Concerning features of the 2024–2025 Tolima outbreak included substantial morbidity and mortality, as well as potential peri-urban cases [4, 7].

The majority of both historical and recent YF cases in South America have been reported in Brazil, Bolivia, Colombia, Peru, and Ecuador [8, 9]. A significant YF outbreak occurred in Brazil in 2016–2018, notable for the reemergence of YF in the Atlantic forest region and proximity to major cities. In Colombia, there have been no major YF outbreaks since 2005, aside from sporadic cases [10]. Thus, the YF outbreak in Tolima is highly concerning and part of a recent trend of YF outbreaks occurring among different ecoregions during the past decade.

Yellow fever virus circulates via 2 transmission cycles in the Americas—the sylvatic and urban cycles [8]. The sylvatic cycle involves non-human primate (NHP)-to-mosquito-to-human transmission near forested areas via primarily Haemagogus and Sabethes mosquitoes. Some new-world NHPs, especially howler monkeys (Alouatta spp.), are susceptible to severe disease and mortality from the sylvatic cycle of YFV [11]. Yellow fever outbreaks in South America have almost exclusively been attributed to the sylvatic cycle since the 1940s, in the context of urban mosquito control and YF vaccination efforts [7].

Given that the severe form of YF has an estimated case fatality ratio of 39% [12], and there are no licensed therapeutics for YF, prevention is key. Understanding the eco-epidemiology of the YF outbreak in Tolima could yield novel insights into risk factors and spatiotemporal patterns of YF, enabling targeted preventive measures such as vaccination and vector control. While multiple hypotheses have been generated surrounding the drivers of the Tolima outbreak [4, 5], spatial statistical and modeling analyses could help understand these factors. Therefore, the goals of our study were to (1) characterize the epidemiology and disease burden of YF in Tolima, (2) analyze spatiotemporal dynamics of YF spread, and (3) identify ecological and sociodemographic risk factors associated with YF incidence.

METHODS

Setting

The administrative structure of Colombia comprises 32 departments, divided into municipalities, which are further divided into the lowest level, called veredas (neighborhoods). Tolima is a department in Colombia's central-west region and has 47 municipalities. These municipalities include many rural areas and dispersed settlements, as well as urban areas such as the capital city, Ibagué.

Tolima has varied topography, including 2 major branches of the Andes Mountains separated by the Magdalena River Valley [13]. This altitudinal variation contributes to diverse ecosystems in Tolima, including lowland riparian forests, mid-altitude Andean forests, and high-Andean forest and páramos [13]. Agriculture is the most important economic sector and has been associated with deforestation as well as increased seasonal human mobility [2]. Additionally, there has been an increase in refugees and migration into the region from Venezuela, as well as armed conflict in neighboring departments of Colombia [2].

Historical YF outbreaks in Tolima occurred from 1830 to 1940 (Supplementary Fig. 1) [14–16]. Given the absence of further cases in the region, Tolima was considered a low-risk zone for YF and vaccination was optional. In 2005, routine YF vaccination in Tolima was implemented only for infants. An estimated 54% of adults had been vaccinated prior to the outbreak [2].

Human and Non-human Primate Cases

For the 2024–2025 YF outbreak, we analyzed confirmed human YF cases in Tolima. We included cases from the earliest confirmed case (based on symptom onset), which occurred on 8 September 2024 to 31 October 2025. Cases were confirmed at the Laboratorio Nacional de Referencia del Instituto Nacional de Salud de Colombia via reverse transcription polymerase chain reaction (RT-PCR) [17]. Demographic data, date of symptom onset, mortality status, and location of residence were collected on all cases. Additionally, data were collected by the Corporación Autonoma Regional del Tolima on deceased NHPs in Tolima that were confirmed to have YFV infection postmortem via RT-PCR and/or histopathology. The testing of NHPs began in February 2025, and the locations of these NHPs were collected, including municipalities and veredas, and georeferenced GPS coordinates, when available.

Statistical Analysis

Descriptive statistics of confirmed human YF cases were calculated in R version 4.5.2. We conducted univariable and multivariable logistic regression to compare the associations between age and sex with mortality status. To analyze areal-level spatial relationships, we aggregated cases to the municipality and vereda levels and assessed spatial autocorrelation using Moran's I (Supplementary Figs. 2 and 3). For point-pattern analysis, the geographic coordinates of residences of confirmed human cases and the locations of infected deceased NHPs were obtained, and the K and L functions were calculated in R (Supplementary Figs. 4–7).

Ecological and Sociodemographic Factors

Using a conceptual framework, we identified key covariates potentially associated with YF incidence in Tolima (Supplementary Fig. 8). These covariates were selected based on whether they had previously been associated with YF in other regions, as well as on discussions with public health officials in Tolima [18–20]. The sources and spatiotemporal resolution for these covariates are shown in Supplementary Table 1. We included ecological covariates (precipitation, temperature, elevation, vegetation, and land cover) and sociodemographic covariates (population density, poverty/deprivation, and built-up area) (Supplementary Tables 1–3). We also assessed the species distribution of NHPs in Tolima using the IUCN Redlist (Supplementary Fig. 9) [21]. We conducted an initial exploratory statistical analysis by comparing each covariate between veredas that reported YF cases and those that did not (Supplementary Appendix, section 5). Because granular vaccination coverage data prior to the outbreak were unavailable, vaccination was not included in the analysis.

Spatiotemporal Modeling Approach

We used a Bayesian inferential modeling approach to identify which covariates were associated with YF incidence in Tolima from 2024 to 2025. First, a correlation matrix was used to exclude covariates that had a correlation coefficient of >|0.7| (Supplementary Appendix, section 2). We then developed 4 types of Bayesian hierarchical spatiotemporal models using integrated nested Laplace approximation with the R-INLA package. Model I includes fixed effects only. Model II includes fixed effects and spatially structured and unstructured random effects using a Besag-York-Mollié (BYM2) model. Model III adds an unstructured space-time interaction term to capture unexplained temporal variation at the vereda level. Model IV is a full spatiotemporal model that additionally includes a temporally structured random effect, modeled as a first-order random walk. We compared the best fitting of these models to a baseline model with only random effects. Model fit was assessed using the deviance information criterion (DIC) and the Watanabe-Akaike information criterion (WAIC). All covariates were standardized, and we assessed lag values of monthly temperature, precipitation, and normalized difference vegetation index (NDVI) ranging from 1 to 3 months to identify the optimal lag to include in the models. To account for overdispersion, monthly YF cases were modeled as a negative binomial process:

$${\mathrm{Y}}_{\mathrm{i},\mathrm{t}}\sim \mathrm{NegBinom}({\mu }_{\mathrm{i},\mathrm{t}}\phi )$$ (1)

where φ is the overdispersion parameter and Y_i,t is the observed number of YF cases in vereda i during month t, while μ_i,t is the expected mean number of cases in i at t, modeled in the spatiotemporal model as the log-incidence rate:

$$\log ({\mu }_{\mathrm{i},\mathrm{t}})=\alpha +\log (P_{\mathrm{i},\mathrm{t}})+{\mathrm{X}}_{\mathrm{i},\mathrm{t}}\beta +u_{\mathrm{i}}+v_{\mathrm{i}}+{\omega }_{\mathrm{t}}+{\epsilon }_{\mathrm{i},\mathrm{t}}$$ (2)

where α is the intercept, log(P_i,t) is an offset to adjust case counts by the logarithm of the population, and X is a matrix of covariates with regression coefficients β. The spatial component follows the Besag-York-Mollié (BYM2) model, where u_i is the spatially structured random effect (conditional autoregressive), and v_i is the unstructured (i.i.d.) vereda-level random effect. The temporal trend, ω_t, is modeled as a first-order random walk, and ε _i,t is an unstructured space-time interaction.

Patient Consent Statement

This study was deemed exempt from institutional review board review (IRB) by the University of California, Los Angeles IRB, and informed consent was waived, as it involved only secondary analyses of data previously collected for public health purposes.

RESULTS

Human Yellow Fever Cases

From 2024 to 2025, 116 laboratory-confirmed cases of YF were identified in Tolima (Table 1). Among these cases, 45 (39%) died, all of whom were unvaccinated for YF. Of the 9 YF cases that were previously vaccinated, per case investigations, these were not attributed to vaccine-associated viscerotropic disease. Overall, these cases in Tolima represented 85% (116/136) of all YF cases reported in Colombia from 1 January 2024 to 31 October 2025 [9].

Table 1.

Yellow Fever Cases in Tolima, Colombia, September 2024–October 2025

	Confirmed Death (N = 45)	Confirmed Alive (N = 71)	Overall (N = 116)
Age
Median [IQR]	62 [40, 71]	43 [28, 57]	47 [34, 63]
Age group
Age 0–18	0 (0%)	10 (14.1%)	10 (8.6%)
Age 19–39	11 (24.4%)	21 (29.6%)	32 (27.6%)
Age 40–59	8 (17.8%)	24 (33.8%)	32 (27.6%)
Age ≥60	26 (57.8%)	16 (22.5%)	42 (36.2%)
Sex
F	8 (17.8%)	19 (26.8%)	27 (23.3%)
M	37 (82.2%)	52 (73.2%)	89 (76.7%)
Yellow fever vaccination status
Unvaccinated	45 (100%)	62 (87.3%)	107 (92.2%)
Vaccinated	0 (0%)	9 (12.7%)	9 (7.8%)

The majority of YF cases were male (77%, 89/116), and the overall median age of cases was 47 (interquartile range 34–63). A higher proportion of deaths occurred among those ≥60 years old (58%, 26/45). In univariable and multivariable logistic regression analyses, adjusting for sex, age was associated with increased odds of death (adjusted odds ratio, 1.43 per 10 years; 95% confidence interval, 1.18–1.78) (Supplementary Table 5).

The initial YF cases were detected in the southeastern municipalities, with the highest incidence in Cunday, followed by Villarrica, Prado, Purificacion, and Dolores. Subsequently, cases were detected in the central part of the department with single cases in Ibagué, El Espinal, and Valle de San Juan, as well as a single case in the northern municipality of Palocabildo. Multiple YF cases were then detected in the southern municipalities of Ataco, Chaparral, and Rioblanco. There were 74 veredas with YF cases. The incidence of YF cases by municipality and vereda is shown in Figure 1.

Figure 1.

Incidence Rate of Human Yellow Fever Cases in Tolima, Colombia 2024–2025. The incidence rate of human yellow fever cases from 8 September 2024 to 31 October 2025 in Tolima is shown aggregated at the (A) municipality level and (B) vereda level.

Non-human Primate Yellow Fever Cases

Of 53 confirmed NHP cases with vereda locations, 49 individuals had georeferenced coordinates available (Supplementary Fig. 4). The NHP cases occurred in the following municipalities (in order of highest incidence): Chaparral, Ataco, Planadas, Cunday, Purificacion, Villarrica, Rioblanco, San Antonio, and Prado. The genera of NHPs infected included Alouatta and Aotus.

Spatiotemporal Analysis

Human YF cases occurred between 8 September 2024 and 8 October 2025. Non-human primate cases were detected from 10 February 2025 to 1 October 2025. Figure 2 depicts the epidemic curve with the timing of human and NHP cases.

Figure 2.

Epidemic Curve of Human and Primate Yellow Fever Cases in Tolima, 2024–2025. The histogram depicts yellow fever cases by week of symptom onset for humans and reported date of death for non-human primates. The line graph shows the overall mean monthly precipitation in Tolima, given the significant association between YF incidence in Tolima and precipitation.

Significant spatial autocorrelation among human YF incidence rates was found via Global Moran's I at the municipality level (I = 0.24, P < .001), and weaker spatial autocorrelation was found at the vereda level (I = 0.03, P < .001). Neighboring municipalities and veredas with high YF incidence rates (hotspots) were identified using Local Moran's I (Supplementary Figs. 2 and 3). These hotspots occurred in 2 areas in southern and southeastern Tolima.

The spatial patterns of human versus NHP YF cases are shown in Figure 3. Human YF cases were found to significantly cluster based on the L function, with clustering strongest at 23.9 km (Supplementary Fig. 6). Non-human primate YF cases were found to form statistically significant clusters with human cases, with clustering strongest at 25 km (Supplementary Fig. 7).

Figure 3.

Kernel Density of Human Yellow Fever Cases in Tolima, Colombia 2024–2025. The georeferenced locations of human and non-human primate yellow fever cases in Tolima are shown. The L function was used to calculate the distance at which maximal human case clustering occurred (23.9 km). Thus, 23.9 km was selected for the bandwidth of the kernel density estimate to depict the intensity of clustering.

Exploratory Analysis of Ecological and Sociodemographic Factors

Figure 4 shows the distribution of ecological and sociodemographic covariates in Tolima. Monthly mean precipitation (Supplementary Figs. 10–12) and temperature (Supplementary Figs. 13–15) were higher in veredas with YF cases compared to those without cases (P-value <.001). Yellow fever cases occurred at a mean elevation of 1066 m (range 318–2190 m) (Supplementary Table 3), and mean elevation was lower in veredas with cases (P = .007, Supplementary Fig. 16). The highest proportion of YF cases occurred in a mosaic of agriculture and pasture habitat (71%, 82/116), followed by forest (23%, 27/116), infrastructure (3%, 4/116), other non-vegetated area (1%, 1/116), and other non-forest formation (2%, 2/116) (Supplementary Table 4, Supplementary Fig. 17). Four YF cases occurred in cities/dense towns (including Ibagué) based on urban built-up surface classification. There were no statistically significant differences in overall monthly mean NDVI, population density, poverty/deprivation index, or built-up surface between veredas with YF cases and those without cases (Supplementary Figs. 18–21).

Figure 4.

Ecological and Sociodemographic Covariates in Tolima, Colombia 2024–2025. The multipanel figure depicts the vereda-level mean (A) temperature 2024–2025, (B) Normalized Difference Vegetation Index (NDVI) 2024–2025, (C) precipitation 2024–2025, (D) Global Gridded Relative Deprivation Index 2020, and (E) elevation.

Spatiotemporal Modeling Results

Comparing our Bayesian hierarchical spatiotemporal modeling results, the spatiotemporal model with standardized covariates and a 3-month lag for precipitation, temperature, and NDVI had optimal performance, as indicated by the lowest WAIC and DIC (Supplementary Table 5). For this model, the incidence rate ratio for precipitation was 2.32 (95% credible interval 1.62–3.47), temperature 1.25 (95% credible interval 0.79–1.98), poverty/deprivation 1.74 (95% credible interval 1.34–2.25), and NDVI 1.37 (95% credible interval 0.99–1.89). All models identified an increased incidence rate of YF associated with higher mean monthly precipitation and higher mean poverty/deprivation at the vereda level (Supplementary Table 6). The full model had better fit and spatial precision than the baseline model, and structured spatial effects contributed most to the variance (Supplementary Table 7).

DISCUSSION

The 2024–2025 YF outbreak in Tolima was the largest regional YF outbreak during this period in South America. While YF cases had occurred in Tolima from 1830 to 1940, there were no reported cases in this department for nearly a century. Overall, the epidemiological features of this outbreak are reminiscent of other YF outbreaks in South America, but with key differences. As with other YF outbreaks in South America, the highest proportion of YF cases occurred among middle-aged males, a demographic that includes agricultural workers most likely to be exposed to the sylvatic cycle. However, there was also a high proportion of patients aged 60 years or older with YF (36%), who had a very high case fatality ratio (CFR) (58%), and increased age was found to be significantly associated with mortality. The elderly population was particularly vulnerable during this outbreak because Tolima was considered a low-risk department for YF and had low vaccination coverage.

Additionally, the overall CFR was very high (39%), which is consistent with global estimates for severe YF, indicating that most likely severe YF cases were detected in Tolima, whereas mild or asymptomatic cases were underascertained. It is estimated that an average of 12% of YF cases are severe [22], suggesting that roughly 1000 YF cases could have occurred in Tolima. The incidence of YF was higher among veredas with greater poverty/deprivation in Tolima, particularly in rural areas. These communities likely have less access to healthcare and lower levels of education. Residents of rural areas are also more likely to be exposed to sylvatic YF vectors due to work activities and proximity to forest borders. Since Tolima was classified as a low-risk department for YF, vaccination remained optional for adults, leading to a significant susceptible population. These findings emphasize the critical importance of immunization for YF (both routine childhood vaccination and mass vaccination campaigns) in areas where YF may circulate, including the Andean region.

We identified clustering of human and NHP cases, particularly around 2 subsequent foci in the southeastern and southern regions of Tolima. This clustering was strongest between 20 and 30 km for humans and NHPs, and human cases primarily occurred in transitional habitats. We found that precipitation was associated with YF incidence in the Tolima outbreak. La Niña that occurred in late 2024 through 2025 might have been a factor contributing to the above-average rainfall during the outbreak.

Further considering the ecology of the YF outbreak in Tolima, 2 primary NHP genera were diagnosed with YF postmortem (Alouatta and Aotus). The majority of YF cases occurred among Alouatta seniculus, which are highly susceptible to YF and exhibit high mortality [11]. Alouatta spp. are widespread and are able to survive in fragmented landscapes, including Andean forests that have undergone deforestation [23]. Aotus spp. also often develop fatal infection and high viremia with YFV and can live in peri-urban forest fragments [11, 24, 25]. Therefore, both Alouatta and Aotus were involved in the Tolima outbreak, but to what degree is less clear. Other species present in Tolima include Sapajus apella, Cebus versicolor, and Lagothrix lagothricha, which are thought to develop less severe infection from YFV. However, both Cebus and Lagothrix were recently found to develop fatal YFV infection under captive conditions in Colombia [26]. Multiple epizootic YF outbreaks among NHPs were reported during 2024–2025, including the department of Huila, which borders Tolima [27]. Given that there was no national surveillance system for YF in NHPs in Colombia, deceased NHPs were tested in response to the YF outbreak in humans, explaining the lag between cases. Additionally, NHP cases were reported by date of death, making it difficult to infer whether NHP cases preceded, coincided with, or followed human transmission.

The majority of YF cases occurred in rural areas near forested habitats. Initial cases occurred near Bosque de Galilea Natural Park among individuals involved in agricultural and logging activities in the forest. Deforestation could increase habitat fragmentation for NHPs and bring humans into closer contact with YFV-infected NHPs and vectors. Such forest fragmentation has been considered to be a driver of YF spread in other regions [28].

While most YF cases occurred in rural areas, 4 individual cases were reported near urban areas, including 1 in the capital city of Ibagué. Additional epidemiological and entomological analyses are needed to determine where these peri-urban cases most likely acquired infection. To truly determine the ecological transmission cycles and origins of this outbreak, entomological and viral genomic data are needed.

There are multiple limitations to our analysis. We used the locations of residence of human YF cases as a proxy for where infection occurred. The degree of clustering between human and NHP cases could be an overestimate because NHP deaths are more likely to be recognized in more populated areas. To compare covariates with YF incidence and protect privacy, we aggregated cases to the areal level. This can lead to the Modifiable Areal Unit Problem, a bias introduced by aggregating continuous geographic variables into discrete areas. However, we minimized this by aggregating to the vereda level, the smallest administrative unit with a mean area of 12 km² per vereda (Supplementary Appendix, section 4). While all covariates were aggregated to the vereda level for our analysis, monthly precipitation and temperature were at a coarser spatial resolution than the other covariates. Local downscaling techniques that account for cloud cover would be needed to obtain these covariates at higher resolution. While we were able to compare multiple factors from our conceptual framework, we were unable to directly assess the role of YF vaccination coverage due to a lack of granular data. In the future, including vaccination coverage, as well as data on human mobility (transhumance) and entomologic surveillance, could help improve our understanding of transmission dynamics.

In summary, the reemergence of YF in Tolima revealed critical lessons regarding YF eco-epidemiology and prevention. While this region had not experienced YF for nearly a century, it had the appropriate ecological conditions for the spread of YF. These conditions and a rural, unvaccinated elderly population with high levels of poverty and deprivation fueled a substantial YF outbreak. Identifying other regions with similar ecological and sociodemographic risk factors is essential for targeted YF vaccination campaigns to prevent future outbreaks.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.