A one health visual analytics approach for syndromic surveillance

Effective syndromic surveillance requires integrating diverse data streams while maintaining transparency and interpretability to support timely and informed public health action. Traditional surveillance approaches often rely on single-domain data, such as clinical reports or environmental measurements, or employ black-box models that limit stakeholder trust and validation. To address these challenges, we present a One Health visual analytics framework that integrates heterogeneous surveillance signals with complementary predictive models and interactive model analysis, enabling systematic comparison of data contributions, model configurations, and predictive uncertainty across infectious disease surveillance tasks. The framework integrates mixed-effects models, Transformer-based deep learning architectures with spatiotemporal embeddings, and interactive dashboards to balance predictive performance with explainability in a white-box environment. We demonstrate the framework's utility through three mixed-effects case studies and the complementary Transformer forecasting experiment: (i) prediction of influenza and RSV cases in Arizona, (ii) modeling of COVID-19 cases and deaths in Arkansas, and (iii) prediction of COVID-19 hospitalizations in Oklahoma using wastewater, animal health, and environmental data. Across the case studies, integrating multi-source signals improved predictive accuracy for most configurations, although the incremental value of individual streams proved species- and configuration-dependent, while hierarchical random effects (e.g., county, month, age group) captured contextual variation and enhanced robustness. For longer-term forecasting, Transformer models with spatial and spatiotemporal embeddings outperformed baseline statistical models, highlighting the value of contextual information. By transforming complex modeling outputs into interpretable visual tools, the framework supports proactive decision-making, enhances public health preparedness, and strengthens One Health collaboration across human, environmental, and animal health domains. This work introduces a generalizable One Health approach for transparent, data-driven syndromic surveillance with direct implications for health security and community resilience. The novelty of this work is the integration of heterogeneous One Health signals with complementary modeling strategies and interactive model provenance so that prediction and interpretation can be examined together across health domains.