Temporal evolution of cases and alerts was visualized using epidemic curves stratified by region and data source. Factors associated with Rift Valley fever (RVF)–related mortality were identified using multivariable logistic regression, estimating odds ratios (ORs) with the dependent variable defined as clinical outcome (death versus survival) among confirmed cases, derived from routine DHIS2 surveillance data. Selection of the most relevant factors was performed using a stepwise procedure based on the Akaike Information Criterion (AIC), and the significance of associations was assessed using p-values (<0.05), ORs, and their 95% confidence intervals. Standard reference categories were: age 15–34 years, female sex, “no reported occupation, “ and districts with the lowest cumulative cases. A forest plot displays aORs with 95% confidence intervals (95% CI) for factors associated with death among laboratory-confirmed human Rift Valley fever cases in Senegal. Odds ratios (OR = e^β) greater than 1 indicate increased odds of death, while ORs less than 1 indicate a protective effect.

Was employed to better understand the distribution and dynamics of Rift Valley Fever (RVF) in Senegal. The study aimed to map outbreak locations, identify high-risk areas, and examine temporal trends in case occurrences. GIS-based spatial analysis was conducted using district-level surveillance data. Analyses further described the distribution and evolution of human and animal cases, deaths, and community alerts from the 3S platform at the district level. Choropleth maps, density analyses, and autocorrelation indicators were used to identify high-concentration areas and risk clusters, while weekly aggregation of early warning signals was compared with confirmed cases to evaluate the performance of the surveillance system. Confirmed human and animal cases were geocoded and plotted as points on the map. These data were overlaid with hydrological layers representing water bodies across Senegal, sourced from the national water dataset (SEN_water_areas_dcw), to investigate the relationship between RVF occurrence and proximity to water sources. Administrative boundaries at the departmental level were included to contextualize the spatial distribution within recognized territorial units. The datasets were preprocessed by removing duplicates, correcting inconsistencies, handling missing values, geocoding case locations, aggregating temporal data into appropriate intervals, and normalizing environmental variables to ensure comparability. Exploratory data analyses were conducted to identify temporal trends, such as seasonal peaks and inter-annual variations, and to visualize the spatial distribution of cases using GIS mapping tools. Spatio-temporal patterns of RVF were then represented through heatmaps and animated maps to track the progression of outbreaks across Senegal.

AI-generated alerts and laboratory-confirmed cases were geocoded and mapped over time to evaluate clustering, diffusion pathways, and the temporal alignment between early warning signals and confirmed outbreaks. Descriptive spatio-temporal analyses quantified the geographic accuracy of early alerts and their correspondence with subsequent transmission. The spatial distribution of cases was further analyzed in relation to livestock density, pastoral mobility corridors, and areas of frequent human–animal interaction. Overlay analyses integrated epidemiological data with environmental and zootechnical layers, including ruminant density and proximity to wetlands, to identify zones of spatial overlap associated with elevated RVF risk.

The 3S–AI4DECLIC-SN platform used rule-based anomaly detection and machine-learning–assisted signal prioritization to identify deviations from historical baselines in livestock health, community alerts, and environmental indicators. Models were trained on pre-2025 surveillance data, with anomalies flagged when observed values exceeded defined thresholds. Retrospective validation compared AI alert timing with official outbreak confirmation to assess lead time, sensitivity, and geographic coverage. Performance was descriptively evaluated in terms of lead time (days/weeks gained), sensitivity to early animal signals, and geographic coverage. Comparative detection delays between routine and AI-assisted surveillance were summarized descriptively (median delay, range) rather than causal inference, given the observational design.

To contextualize quantitative findings, a qualitative institutional analysis was conducted based on document review, field reports, and stakeholder feedback, which examined alert validation, cross-sector coordination, escalation pathways, and feedback to communities. Data were analyzed using thematic analysis, with codes related to governance, data sharing, operational bottlenecks, and decision-making processes. The analysis explored integration of technological tools within health governance, identifying operational bottlenecks, best practices, and factors influencing system responsiveness. Data were thematically coded and analyzed; generative AI tools were not used in manuscript production.

The study followed the guidelines of the National Ethics Committee for Health Research (NECHR) (SEN 23/121) to ensure adherence to ethical principles in the management of health data. Personal information was anonymized before analysis, with no direct contact with patients or animals. Confidentiality and data minimization principles were strictly applied, in alignment with the World Health Organization (WHO) data governance and ethical guidelines for public health surveillance. Specifically, data handling complied with the WHO’s principles on ethical and responsible data use (28), including the collection and processing of only data strictly necessary for public health purposes, and the protection of individual privacy through anonymization and restricted access. These practices are consistent with the International Health Regulations (2005), which require that the collection, use, and sharing of public health data respect the principles of confidentiality, necessity, proportionality, and protection of personal data, while ensuring that information is limited to what is essential for public health purposes (29). The study relied on secondary surveillance data collected as part of routine public health and veterinary activities. All human data were anonymized before analysis. National authorities granted access to DHIS2 data and is subject to data protection regulations; therefore, raw datasets cannot be publicly shared.

Surveillance activities focused on known RVF hotspots in Podor, Matam, and Saint-Louis. Reported livestock abortions were used as the primary event indicator, acknowledging that this approach likely underestimates overall animal morbidity, particularly in remote pastoral settings with limited veterinary access and formal reporting. This limitation is consistent with established RVF surveillance challenges, as abortions represent a specific but incomplete proxy for infection. In this study, abortions were intentionally selected as a sentinel indicator rather than a comprehensive measure of disease occurrence. Their strong association with RVF outbreaks and their tendency to prompt rapid community notification support their use for early warning surveillance. The study objective was to assess the timeliness and comparative performance of an AI-supported community surveillance system relative to conventional reporting mechanisms, rather than to estimate absolute disease burden. To partially address under-reporting, the 3S-AI4DECLIC-SN platform integrated multiple data sources, including community alerts, veterinary observations, and human health indicators, in accordance with a One Health approach. In addition, a qualitative institutional analysis was conducted to characterize structural factors influencing reporting and validation processes, providing contextual information for interpretation of surveillance performance.

The integrated approach enhanced ecological validity through triangulation across data types and sectors and provides a replicable framework for other settings with routine surveillance, community reporting, and basic AI-supported anomaly detection. However, logistic regression analyses were constrained by small sample sizes in some occupational and geographic categories, potential collinearity between age, occupation, and residence, and residual confounding from unmeasured factors such as clinical severity or healthcare access. Spatio-temporal analyses were limited by district-level aggregation, which may obscure localized transmission patterns, reporting delays affecting temporal resolution, spatial autocorrelation between neighboring districts, and incomplete capture of dynamic environmental or livestock changes. These factors should be considered when interpreting the accuracy, predictive value, and operational utility of early alerts generated by the AI-assisted surveillance system.

In Senegal, 6, 070 human samples were collected for epidemiological surveillance, of which 363 tested positive for RVF, including 29 fatalities. Cases were reported in eight regions, with an intense concentration in the northern Saint-Louis (78%, 294 cases), a result consistent with recent outbreaks concentrated in the Senegal River valley (wetlands with Aedes and Culex mosquitoes). Matam, Louga, and Fatick recorded fewer cases (16–24) but showed relatively high fatality rates (12.5%–22.2%), likely reflecting delayed detection or limited access to care. In contrast, regions such as Kaolack, Dakar, Kédougou, Tambacounda, and Thiès reported only sporadic cases and no deaths, suggesting lower transmission or faster case identification. Analysis of attack rates confirms an unequal geographic distribution. Richard-Toll department is the primary hotspot, with the highest attack rate (66.51), followed by Dagana and Saint-Louis, indicating intense circulation in the northern zone (Figure 1). Podor, Ranérou, Thilogne, and Dioffior show moderate but active transmission with intermediate attack rates, warranting continued vigilance. Most other localities reported between 1 and 10 cases, with attack rates below 5 (e.g., Linguère, Nioro du Rip, Thiès), suggesting low population density, reduced mobility, or rapid containment of initial infections.

Districts like Louga and Ranérou reported 6–15 cases, indicating secondary outbreaks, while isolated cases elsewhere may stem from livestock movement or mosquito-friendly environments. Even areas with no cases could still transmit due to animal movement and irrigation. The high death toll underscores health risks for herders and vets. The spread may reach central regions, and the south, with fewer cases, needs monitoring to prevent wider outbreaks.

Monitoring confirmed cases from Week 36 to Week 44 reveals three epidemic phases (Figure 2). The initial phase (W36-38) has few cases but high lethality, reaching 67% in W37, likely due to late detection. The growth phase (W39-41) shows a sharp rise, peaking at 83 cases in W41, with the fatality rate dropping to 1%, indicating better care. The decline phase (W42-44) shows declining case counts and low fatality rates (<10%), reflecting the success of health measures. The difference between the number of live cases and deaths highlights the importance of early detection, integrated surveillance, and swift interventions.

Demographic analysis indicates that the majority of cases involve young adults aged 20–49 (such as students, apprentices, and family helpers), representing 62% of total cases, often linked to livestock farming or slaughtering activities. The distribution of confirmed cases by occupation and age (Figure 3) reveals epidemiological patterns primarily influenced by occupational exposure, risk behaviors, and socioeconomic factors. Farmers emerge as the most affected group, consistent with known RVF transmission modes, which include direct contact with infected animals, ruminant abortions, and carcass handling. These individuals tend to be the most economically active and most exposed to environments with vectors or animal contact. The significant proportion of middle-aged adults (15–34 years), followed by other age groups, reflects their active role in intensive pastoral work, reinforcing the link between occupational exposure and infection risk. Housewives are the second most affected group, with their high presence among adults aged 35–59 indicating indirect transmission at home, primarily through handling meat or animal products, caring for domestic animals, or frequenting areas where the virus circulates. The group of pupils and students shows a high proportion of cases among 15–34-year-olds and, to a lesser degree, among 5–14-year-olds.

Farmers, the unemployed, and other middle occupational categories, such as traders, drivers, laborers, and fishermen, experience broader but steady exposure, especially among working adults. Their risk stems from a mix of environmental exposures—wetlands, irrigated fields, mosquito vectors—and occasional contact with animals, signifying a moderate yet notable risk. Fewer cases are observed among certain occupations, such as administrative staff, teachers, healthcare workers, police officers, sailors, and gold miners, suggesting lower exposure to or contact with hotspots. Nonetheless, the sporadic cases in these groups highlight that RVF, primarily occupational, remains a vector-borne disease capable of affecting populations with less direct contact.

The four pie charts (Figure 4) offer a clear and helpful overview of the distribution of confirmed cases, their clinical progression, and related sociodemographic characteristics.

Gender analysis shows men constitute 67.5% of confirmed cases and 58.6% of deaths, suggesting higher exposure and risk, potentially due to behavioral, occupational, or biological factors. Furthermore, 76.9% of cases occur in rural areas compared to 23.1% in urban areas, likely reflecting socioeconomic status, population density, or resource access. Despite these differences, the overall clinical outlook remains positive, with 92.4% of patients still alive and only 7.6% deceased. Most individuals experience mild symptoms, though some have faced hemorrhagic forms.

Analysis of factors associated with mortality among confirmed RVF cases, based on odds ratios (ORs), demonstrates a clear risk gradient driven primarily by individual characteristics (age and clinical symptoms), with additional contributions from occupational exposures (livestock farming, agriculture, fishing) and regional context (hyperendemic areas), as summarized in the multivariable forest plot (Figure 5). This forest plot shows adjusted odds ratios (aORs) and 95% confidence intervals for factors associated with death among laboratory-confirmed human Rift Valley fever cases. Points represent aOR estimates, and horizontal lines indicate 95% confidence intervals. The vertical dashed line at an odds ratio of 1 denotes no association. Values greater than 1 indicate increased odds of death, whereas values less than 1 indicate reduced odds. Odds ratios are displayed on a logarithmic scale to facilitate comparison across a wide range of effect sizes.

Age was the strongest predictor of mortality. Compared with the reference category, individuals aged ≥60 years exhibited the highest risk of death, followed by those aged 35–59 years and 15–34 years. Although the 5–14-year age group also demonstrated an elevated OR, the wide confidence interval indicates limited precision. Overall, a clear age-related gradient in mortality risk was observed. The presence of hemorrhagic manifestations was strongly associated with an increased risk of death, underscoring the contribution of severe clinical presentation to fatal outcomes. Male sex was associated with a modestly increased risk of death compared with females, though the magnitude of the association was limited. Several occupational categories, including housekeepers, students/pupils, masons, tailors, shepherds, livestock breeders, traders, and farmers, were associated with ORs greater than 1. These jobs typically involve regular contact with infected animals or environments that promote vector transmission, or may reflect disparities in access to healthcare, delays in treatment, or fewer medical checkups. However, confidence intervals frequently overlapped the null value, indicating moderate associations with considerable uncertainty. Districts like Podor, Richard-Toll, Dagana, and Saint-Louis (major hotspots) exhibit high ORs, consistent with the large number of cases and deaths in the Senegal River valley, probably due to intense viral transmission, high animal density, numerous livestock activities, ecological factors like irrigation and mosquito vectors, and limited healthcare access. Conversely, districts in the central or southern regions, such as Kaolack and Dioffior, show low ORs, reflecting reduced morbidity and mortality. Urban residence was not significantly associated with mortality.

Our analysis identifies advanced age as the primary determinant of RVF-related mortality, consistent with prior outbreaks in East and West Africa, where older individuals exhibited higher severity and fatality, likely due to immunosenescence and comorbidities (30). Hemorrhagic manifestations were strongly associated with death, aligning with clinical reports indicating case-fatality ratios of 30–50% in severe RVF, underscoring the need for early recognition and intensive management. Men showed a modestly increased risk, potentially reflecting occupational exposures such as livestock handling or work in vector-prone environments, although the effect size was limited. Occupational associations, particularly among livestock breeders, shepherds, and farmers, similarly showed moderate associations, suggesting that it serves more as a proxy for exposure than as a direct determinant of mortality (31). Geographic location did not strongly influence fatal outcomes after adjustment, indicating that mortality is driven primarily by individual-level and clinical factors, rather than residence. Wide confidence intervals for some estimates reflect small subgroup sizes, highlighting the need for pooled analyses, hierarchical modeling, or age-stratified approaches in future studies to improve precision and explore effect modification.

Epidemiological data across 20 affected departments found 243 positive cases, mainly in 128 sheep, 105 goats, and 10 cattle. Diagnoses were confirmed through 1, 642 blood samples and 68 aborted fetuses confirmed from 1, 811 reported abortions in ruminants, indicating a substantial effort to verify infections biologically. Nonetheless, the ratio of samples to reported abortions hints at possible under-detection, particularly in remote areas with limited veterinary access. The weekly trend in reported abortions over about two months (Figure 6) shows high variability, with notable fluctuations from week to week:

Analysis combining reported abortion data with animal confirmed cases shows a strong link between abortion rates and epidemiological trends. Early signs of abortions (2 to 3 cases) appeared when confirmed cases were still low (9 cases in week 38), indicating low-level virus spread within multiple herds. From September 15, a gradual rise in abortions (reaching 35 to 45 cases) slightly before the increase in confirmed cases between weeks 39 and 40 (28 to 20 RVF cases). The connection becomes clearer at the end of September: 57 abortions occurred just before the rise in confirmed cases in week 41 (18 cases), before the sharp increase in week 42 (85 cases). Peaks in abortions in October coincide with weeks of maximum viral detection (62 cases in W43, 50 in W44). These patterns indicate an epidemic where high abortion numbers serve as early signals and sensitive warning signs of rising virulence, reflecting faster transmission likely driven by seasonal vectors or livestock movements. The resurgence of abortions at the end of October (53 and 49 cases), despite declining confirmed cases, may indicate residual virus circulation or diagnostic delays due to field capacity.

The trend of confirmed RVF cases in animals shows a sharp rise between weeks 38 and 44 (Figure 7), illustrating the rapid growth typical of RVF outbreaks. After a modest start with 9 cases in week 38, the number increased to 28 in week 39. Then, the trend stabilized, with fluctuations between 20 and 18 cases in weeks 40 and 41, indicating that virus circulation remains limited in herds. The dynamics shifted notably in week 42, with a peak of 85 cases, likely due to a vector surge or secondary amplification in animal populations. The following weeks showed a gradual but steady decline, with 62 cases in week 43 and 50 in week 44, indicating that transmission remained active despite the decrease after the peak.

Spatial analysis of reported animal cases reveals a highly variable distribution, with a significant concentration in the northern regions, as noted for human cases. The main hotspots are Saint-Louis (125 cases), Louga (81 cases), and Matam (20 cases). Central and southern regions—Thiès, Fatick, Kaolack, Tambacounda, and Kolda—show lower-to-moderate rates, likely due to less mobile systems or reduced vector exposure. Dakar and Diourbel, with urban or peri-urban settings, have the lowest levels. Isolated outbreaks are also reported in central and southeastern areas, such as Koumpentoum, Tambacounda, and Vélingara, likely due to seasonal livestock movements and connections between pastoral regions.

The distribution of cases in animals and humans highlights zoonotic and vector-borne transmission, mainly in northern Senegal along the River valley, including Saint-Louis, Dagana, Podor, and Matam (Figure 8). The following map highlights the clustering of confirmed cases in relation to water areas, allowing identification of potential high-risk zones where environmental conditions may favor vector breeding and virus transmission. Scale bars and north orientation were added to ensure accurate spatial interpretation.

This pattern forms an epidemiological corridor aligned with the Sahelian ecological zones, where the presence of temporary pools, seasonal flooding, and the proliferation of Aedes and Culex mosquitoes facilitate virus transmission. Viral transmission is aided by transhumant movements, enabling secondary introductions into suitable yet initially uninfected regions. Outbreaks across regions indicate extensive viral spread driven by animal movement, immune variation, and vector ecological niches. Few cases in the deep south and coastal west may be due to environmental factors or detection limits, especially where animal densities are low or monitoring is limited. These patterns reveal complex interactions among ecology, animal production, and detection efforts in RVF transmission, underscoring the need for integrated strategies grounded in the One Health approach.

Overall, the data indicate a large-scale epizootic marked by high abortion rates and widespread outbreaks across different regions, linked to ecological conditions, herd mobility, and climatic factors that encourage vector development. The following section will explore in detail the factors related to RVF cases and human fatalities to better understand how transmission occurs and the disease’s severity.

Results show a strong correlation between exceptional rainfall, flooding, and increases in human and animal cases. The main breeding sites for vectors are the large riverbeds of the Senegal River, temporary pools, and shallow waters. Hydrological changes in the valley, particularly the expansion of irrigated areas following the commissioning of the Diama Dam, have accentuated the seasonality of mosquitoes. Over the past thirty years, local rainfall has exhibited significant interannual variability, alternating between dry phases and prolonged wet periods. The annual average rainfall of approximately 315 mm varies significantly from year to year. Rainfall ranges from 134.5 mm in 2001, an arid year, to 600.8 mm in 2010, a very wet year. Some years stand out for their extremes: 2001 and 2002, with less than 200 mm, in contrast with 2010, 2012 (458.1 mm), and 2020 (569.5 mm), which exceed the average. Temporal analysis reveals prolonged periods of deficit, especially between 1997 and 2004, as well as wetter phases, particularly between 2009 and 2013, then between 2020 and 2023. This alternation is confirmed on a decade-by-decade basis: 1995-2005, high heterogeneity with dry and wet years; 2006-2015, higher overall rainfall; 2016-2025, increased variability, with stormy years and significant deficits. These fluctuations directly influence the filling of water points (Figure 9) and dictate the seasonality of vector populations. September remains the period of maximum risk, when temporary pools are at their maximum, and herds concentrate in wetlands.

Temperatures indicate favorable conditions, with stable maximum around 45 °C. Since 2015, the rise in minimum temperatures, often above 12 °C, has been indicative of nighttime warming that favors vector survival. Warmer nights reduce mortality, prolong the mosquito life cycle, and stimulate reproduction. This increase in minimum temperatures also reduces the annual temperature range, making the environment more stable and conducive to the sustainability of vector populations. The relative humidity, averaging 42.5%, fluctuates between 38% and 46% with no clear trend. A comparison with rainy years shows that hot, rainy years (2010, 2012, 2020) offer very favorable conditions for proliferation, with abundant larval habitats and ideal temperatures. Conversely, hot but dry years (2002, 2015) limit reproduction by reducing the availability of aquatic habitats, while promoting adult survival, which maintains a high risk of transmission. Additionally, climatic anomalies such as El Niño, which cause excessive rainfall, can heighten outbreak risk by creating more larval habitats and extending mosquito activity (32).

During the 2025 epidemic, Cx. tritaeniorhynchus was identified as the dominant species among mosquitoes that tested positive (Figure 10), consistent with previous observations in northern Senegal (33). In areas with permanent water sources, Cx. tritaeniorhynchus and Mansonia uniformis predominate, while Aedes vexans arabiensis dominates in temporary pool systems. Older studies had already highlighted the central role of Aedes vexans and Culex poicilipes in RVF transmission in Senegal (34, 35). The 2025 Senegalese RVF epidemic pattern follows a well-documented sequence: intense rainfall triggers the synchronized emergence of infected Aedes mosquitoes, which then multiply within livestock populations. The infection spreads through animal movement, and Culex mosquitoes maintain ongoing transmission. This rapid amplification is followed by secondary culicines, especially Culex species, whose populations grow more slowly as water remains. The transition to these secondary vectors explains the typical RVF epidemic curve, marked by a sharp increase after rainfall, followed by a slow decline (36).

Overall, high humidity, combined with high temperatures, creates ideal conditions for mosquito proliferation, accelerating larval development and prolonging adult lifespan. Studies in Ferlo and other pastoral zones also confirm that temporary breeding sites are primary emergence points for vectors within days to weeks after rain (37).

Our analysis shows an overlap of cases in areas with high concentrations of sheep and goats, particularly in Saint-Louis, Louga, and Matam. Animal mobility is a key factor in the spatial and temporal dynamics of RVF in Senegal. These movements also lead to the emergence of secondary outbreaks when infected animals, often asymptomatic, arrive in areas with high mosquito density, which favors the formation of new outbreaks. Cross-analysis of environmental, animal, zootechnical, and epidemiological data reveals spatial overlap among wetlands, high livestock density, and RVF foci in Senegal. This indicates a clear correlation among ecological conditions, pastoral practices, and the disease’s epidemiology.

The 2025 epidemic also highlights the role of social and religious events in animal mobility and viral spread. The challenges associated with Tabaski, celebrated in June 2025, reinforce this phenomenon by concentrating animal flows, crowded markets, and slaughter, which increases the risk of direct contact with infected animals or their organs. This peak in mobility, occurring a few weeks before the traditional seasonal peak of Rift Valley fever, likely facilitated the silent spread of the virus via markets, livestock fairs, and slaughter networks, increasing direct contact and interregional exchanges. In addition, cross-border trade in animals between Senegal, Mali, and Guinea follows well-known seasonal cycles. The dry season (February–June) and major religious holidays coincide with an increase in the flow of ruminants into Senegal, thereby increasing the risk of viral introduction. These trends are consistent with eco-epidemiological analyses indicating that areas with high animal density, proximity to wetlands, and strong commercial connectivity are among the most vulnerable.

The recent spread of RVF results from complex interactions among climate variability, pastoral practices, and ecological conditions that facilitate transmission In mainly agricultural areas, rainfall shortages, combined with socioeconomic factors, significantly influence nationwide mobility. Changes in rainfall patterns, water sources, and pasture quality due to climate change exert greater pressure on pastoral systems. As a result, livestock farmers are relocating to more suitable areas, particularly to the south, which raises interactions among animals, humans, and vectors. Migrations, traditionally seasonal, now last longer and cover larger areas, aiding the virus’s spread between origin and host regions. Recent research (38, 39) indicates that climate anomalies are key drivers of these internal movements, establishing new epidemiological links between previously low-risk areas (40). Climate projections (41) further support these trends, estimating that up to 3.3% of the population could migrate due to climate-related factors by 2050. These anomalies serve as triggers, accelerating migration. The convergence of areas with high climate-driven migration and regions newly affected by RVF highlights a persistent risk of the disease’s geographic expansion. These insights emphasize the importance of predictive models combining climate data, pastoral dynamics, and entomological surveillance. Analyzing community signals via the 3S platform is therefore essential for understanding the role of early warning in outbreak detection and response.

During the pilot phase in Podor, Kédougou, and Pikine, the AI-One Health platform issued 330 alerts. Of these, 144 (43.6%) were confirmed by local teams, while 186 (56.4%) were still pending validation at the time of analysis. The analysis focuses only on verified alerts, which are deemed reliable signals from the system. The temporal pattern of these alerts is non-linear (Figure 11), with a peak in June 2025 (n = 35), a gradual decline until September (n = 1), and then an increase in October (n = 20). The nonlinear dynamics in validated alerts are specific to certain zoonoses and reflect the system’s responsiveness to environmental or notification changes. The June peak coincides with the onset of the rainy season, a time favorable for vector growth, while the September decline may result from reduced transmission or logistical issues such as the availability of community actors or validation teams. The October increase could indicate resumed transmission or enhanced surveillance after a period of inactivity.

These alerts primarily involved serial abortions, unusual animal deaths, and febrile syndromes in humans in agro-pastoral areas, detected weeks before validation and official lab confirmation. Two prominent peaks were noted: one in June 2025 linked to sheep abortions in Ferlo, and another in early September 2025, associated with increased livestock abortions and unexplained fevers in humans in Podor. These patterns suggest a seasonal effect: June-July, marking the start of the rainy season, promotes RVF vector mosquito proliferation, while the October resurgence may result from relaxed surveillance or a new local outbreak driven by environmental factors.

Most alerts involved humans (~75%, 109/144), while alerts for animals and the environment were few (Figure 12). Such a discrepancy may skew risk assessment because animal outbreaks generally precede human cases, especially in the case of RVF. It suggests that although human surveillance works well, inadequate detection in animals hampers early identification of zoonoses.

The geographical distribution shows a high concentration in Saint-Louis (126 reports), likely reflecting surveillance focus or the epidemic’s origin (Figure 13). Livestock localities like Fanaye and Podor, where ruminant abortions are more common, underscore the need for integrated surveillance.

Using data from the platform enables a detailed analysis of the epidemiological surveillance system, emphasizing how digitalization has enhanced responsiveness and identifying procedural bottlenecks. The integration of time-stamped and geolocated metadata provides objective documentation of the entire process—from detection to feedback—helping clarify operational dynamics in Senegal.

• Detection time: a substantial improvement in frontline response.

Detection time, or the interval between a health event and its recording on the One Health platform, is a vital indicator for assessing surveillance system performance. Analysis shows that this duration averages 2.8 days with the digital system, compared with 4.6 days with conventional methods such as paper records, phone calls, or texts. This 39% decrease underscores the platform’s capacity to reduce information loss caused by human intermediaries, geographic barriers, and slower processes. Additionally, the 3S platform, which is co-created with communities and appropriated by local frontline actors, can detect weak signals—such as abortions in small ruminants, unusual mortalities, or abnormal behaviors—up to ten days before official detection. These early signs appeared in June and September, weeks before official notifications and the first human RVF cases. This highlights how digital technology at the community level can uncover patterns that traditional systems, which depend heavily on veterinary or health services, might miss. However, the platform’s performance varies by region. In Podor, where digital literacy is higher, detection occurs quickly, with alerts and reporting becoming routine for farmers and local agents. In isolated areas like Saraya, limited connectivity and perceptions of zoonotic diseases cause longer detection times, mainly due to structural, social, and technological issues, not the platform itself. However, access to digital tools reduces delays in sharing health information, enabling near real-time data transmission, democratizing surveillance, and bypassing institutional hurdles.

Community reports of unusual human events and animal abortions in Podor illustrate that local populations serve as vital sentinels, challenging the reliance on centralized, technocratic surveillance systems. By emphasizing population-based surveillance, AI4DECLIC SN enables citizens to identify subtle signals, fostering a more balanced relationship between expertise and local insights. Nonetheless, it still reflects systemic issues inherent to urban bias, dependence on human over animal data, and insufficient environmental consideration. The observed weaknesses in animal signals should be seen not just as epidemiological limitations but also as systemic blind spots rooted in less interoperable systems.

• Notification delay: a persistent institutional challenge.

The delay between reporting and review is a significant weakness in the system. Usually, symptoms appear after five days, followed by three days for biological confirmation. Data show an average of 34.11 hours for validated alerts, indicating low system responsiveness to zoonoses. The high rate of unvalidated alerts (56.4%) complicates verification at district and regional levels. While the 3S platform enables near-instant reception, validation faces organizational challenges, such as workload and staff shortages, especially in resource-limited areas, which reduce efficiency. Most delays involve intermediate professionals, mainly head nurses. In the animal sector, veterinarians and assistants covering large areas need to verify on-site, often without enough staff or logistics, which slows validation and reporting. From a One Health perspective, this gap is concerning because fast coordination between human and animal health sectors is vital for an effective response. Uncertainty about unusual signs can delay decision-making, increasing the time from detection to action. Although the platform enables early detection, its success depends on swift signal validation and on consideration of environmental and organizational factors. A significant challenge is the lack of a formal alert-verification threshold system and insufficient validation and correlation methods across segments, which impair its ability to detect and forecast zoonotic outbreaks. Without it, alerts may be noisy or insignificant, reducing user trust and delaying urgent actions. Addressing these issues is essential to strengthening surveillance and response capabilities against emerging health threats. Reorganizing decision-making processes, setting clear timeframes, developing triage protocols and automated models that dynamically add cross-segment correlations, and creating intersectoral escalation mechanisms are essential to improve consistency and to leverage digital technology’s full potential in One Health surveillance.

• Investigation time: varying capacity for mobilization across territories.

The investigation time, defined as the interval between official signal validation and the start of a field investigation, varies significantly across regions. In urban areas like Dakar, with dense health networks, these delays are generally short. Such teams can access adequate logistical resources—vehicles, fuel, and mobile units—that enable quick and coordinated responses. In contrast, rural areas like Podor experience much longer delays, often from 48 to over 72 hours, due to remoteness, large distances between livestock sites, herd mobility, and limited staffing. These factors reveal structural inequalities that impede the surveillance system’s responsiveness, despite digital advancements. Additionally, case biological validation is crucial. While Dakar’s Pasteur Institute has substantial expertise and advanced diagnostics, regional laboratories often lack the necessary equipment for routine analysis of emerging zoonoses. Samples must be transported over long distances, adding four to five days for confirmation. This diagnostic delay hampers efforts to control the rapid spread of transmission between animals and humans. Overall, these insights show that improving digital detection alone is insufficient; substantial investments are needed in both logistical and biological infrastructure.

• Feedback delay: a weak link compromising community engagement.

The delay in feedback after investigations is fragile. In many districts, input to farmers, women, relays, or local actors is often late, incomplete, or missing, with some informants raising alerts without knowing their outcome. Signals are neither confirmed nor denied, and results are not shared, which is seen as neglect and reduces trust. Without official responses, the motivation to report declines, weakening the system’s sensitivity. In Ferlo, community actors use alternative channels, such as private veterinarians, religious leaders, or radio, which are seen as more responsive. Human health services have centralized channels that enable faster communication, unlike dispersed veterinary services, which lack resources. This disparity hinders data consolidation, delays analysis, and slows decision-making. Effective surveillance depends on detection and transparent, consistent feedback to sustain trust, participation, and resilience.

• Need to integrate environmental data better and automate inter-segment correlation.

The platform performance analysis highlights a key limitation: the absence of climatic and ecological variables—such as rainfall, humidity, vegetation cover, water points, livestock movement, and vector density—reduces the ability to predict epidemics accurately. Transmission of zoonotic, vector-borne diseases like RVF is highly sensitive to environmental conditions (42, 43), and incorporating these variables can help distinguish true transmission events from surveillance artifacts, particularly in rural areas where human reporting is limited (44). Integrating local climate, ecological, and vector data with case reports can improve model precision and enable early warnings even in regions currently free of disease by combining epidemiology, vector dynamics, and environmental drivers.

The pilot AI system did not include real-time environmental variables, reflecting the initial scope of the study, which focused on evaluating the timeliness and operational feasibility of a community-based, AI-supported reporting platform rather than producing a fully predictive machine learning model. While environmental integration would be expected to enhance predictive performance, the current pilot should be considered a proof-of-concept for rapid, community-driven early warning rather than a predictive model. Future iterations of the platform will incorporate real-time climatic and ecological data—including rainfall, temperature, humidity, vegetation indices, and vector distribution—to enable true predictive modeling and improve the system’s capacity to anticipate outbreaks under variable conditions such as El Niño events.