Spatio-temporal dynamics of three diseases caused by *Aedes*-borne arboviruses in Mexico

Dengue, Zika, and Chikungunya viruses, transmitted primarily by *Aedes aegypti* mosquitoes, pose a significant global health burden. These *Aedes*-borne diseases (ABDs) are amplified by human transmission, creating a complex dynamic. While Dengue has seen an eightfold increase in cases over the past two decades, the true global impact of Chikungunya and Zika remains unknown. Mexico experiences high endemicity for all three viruses, highlighting the critical need for understanding their spread. The transmission intensity of these viruses varies significantly across space and time, influenced by both climatic factors (temperature, rainfall) and socio-economic conditions (access to healthcare, sanitation, poverty levels). This heterogeneity complicates control efforts and necessitates a detailed investigation of the underlying drivers. This study hypothesizes the existence of geographic clusters of these ABDs in Mexico, linked to either socio-economic or climatic variables. The research aims to identify these clusters, pinpoint the dominant influencing factors (socio-economic or climatic) within each cluster, and determine the independent contribution of individual risk factors. Such an analysis, to the authors' knowledge, has not previously been conducted nationally in Mexico, representing a significant knowledge gap in understanding and controlling these diseases. The research employed a combination of spatial statistics and machine learning to achieve its objectives.

Existing literature highlights the complex interplay of climatic and socio-economic factors in ABD transmission. Temperature and rainfall exhibit both positive and negative effects on outbreaks, depending on specific conditions and mosquito life cycle stages. Rapid urbanization and unplanned development, frequently linked to socio-economic disparities, modify transmission potentials. Specific studies on dengue have identified correlations between temperature ranges (23.8–33.1 °C), rainfall, and outbreaks, sometimes with lagged effects. Zika prevalence has been linked to inadequate water infrastructure, while Chikungunya shows a weaker correlation with urbanization. Some research in Mexico points to climate as a primary driver, while other studies emphasize socio-economic determinants. This disparity in findings emphasizes the need for a comprehensive study integrating long-term data on all three ABDs with detailed socio-economic and climate information.

This study analyzed data from 2469 municipalities in Mexico over eight years (2012-2019). The data included laboratory-confirmed cases of CHIKV, DENV, and ZIKV obtained from the Mexican Ministry of Health. This data was complemented by socio-economic variables (illiteracy rates, access to healthcare, housing quality, sanitation) from the Mexican National Council for Evaluating the Social Development Policy (CONEVAL), and climate data (temperature and rainfall) obtained from the Climate Forecast System Reanalysis (CFSR) and Climate Hazards Group Infrared Rainfall with Stations (CHIRPS) datasets. Entomological data on *Ae. aegypti* and *Ae. albopictus* presence were also incorporated. SaTScan software was used for spatial cluster detection, employing a discrete Poisson probability model. The influence of socio-economic and climatic factors on cluster prevalence was evaluated using the Pearson correlation coefficient, Randomized Dependence Coefficient (RDC), and SHapley Additive exPlanations (SHAP). Six machine learning (ML) algorithms (XGBoost, decision tree, SVM with RBF kernel, KNN, random forest, and neural network) were compared to predict ABD prevalence. Model performance was assessed using accuracy, weighted accuracy, precision, recall, and F1 scores, with 10-fold cross-validation employed to prevent overfitting. Stratified analysis addressed potential data distribution bias across urban and rural areas. Ethical approval was obtained from the ethical committee of UNIVERSIDAD DE SONORA, Mexico, with informed consent waived due to the de-identified nature of the data.

Dengue was the most prevalent ABD across Mexico (60.6% of municipalities), followed by Zika (31.2%) and Chikungunya (29.3%). Only 2.1% of municipalities reported all three diseases. Twenty-one statistically significant spatial clusters were identified. Analysis revealed that for many clusters, socio-economic factors exerted a stronger influence on ABD prevalence than climatic factors, although some clusters showed the opposite pattern. XGBoost consistently outperformed other ML algorithms in terms of precision, indicating its effectiveness in predicting ABD outbreaks. The attributes of altitude and minimum rainfall showed marginal influence on the models, while average and maximum rainfall were more influential. The presence of houses without toilet facilities, water pipelines, and access to improved water sources were strongly associated with higher disease prevalence, especially in areas with high illiteracy rates. For all three ABDs, there was generally a stronger correlation with socio-economic indicators than with climate factors, although this varied between specific clusters. The study provided evidence that certain clusters were influenced by climate-related factors (temperature, rainfall), while others were more strongly affected by socio-economic factors. The detailed analysis of the various clusters provides valuable information for targeted interventions.

The findings underscore the complex interplay of socio-economic and environmental factors driving ABD transmission in Mexico. The fact that socio-economic factors were more influential than climate in some clusters suggests that targeted interventions focusing on improving sanitation, water access, and public health infrastructure could be highly effective in reducing disease burden. Conversely, the existence of clusters driven by climate factors highlights the importance of climate-informed vector control strategies. The success of XGBoost suggests that ML techniques can offer valuable tools for prediction and early warning systems. While the municipality-level analysis provides a useful overview, the study’s limitations include potential confounding factors arising from data aggregation. Individual-level and household-level data would provide a more granular understanding of the disease dynamics. Further research incorporating fine-scale spatial resolution, microclimate data, and detailed entomological information could further refine our understanding of transmission dynamics and improve the effectiveness of intervention strategies.

This study provides valuable insights into the spatio-temporal dynamics of CHIKV, DENV, and ZIKV in Mexico, revealing a complex interaction of socio-economic and climatic drivers. The superior performance of XGBoost highlights the potential of machine learning for prediction and early warning. Future research should focus on finer-scale spatial resolution, integration of individual and household-level data, and detailed entomological studies to further refine our understanding and develop targeted intervention strategies. Such studies would be instrumental in improving disease surveillance and vector control programs.

The study utilized aggregated municipality-level data, potentially masking the impact of finer-scale socio-economic and environmental variations. Individual-level data would offer a more detailed understanding of the disease dynamics. The analysis relied on laboratory-confirmed cases, potentially underrepresenting the true burden of disease due to underreporting. Additionally, the use of passive surveillance data might introduce biases. The focus on climate variables at the municipality scale might not capture the micro-climate variations within municipalities that could influence mosquito breeding and virus transmission.