An Open-Source Data Driven Hybrid Modeling System for Infectious Disease Surveillance and Early Warning

Introduction: The increasing trend of globalization has led to a heightened risk of imported epidemics; however, existing surveillance systems remain fragmented and reliant on laboratory confirmation. We developed an open-source data-driven hybrid modeling system to provide earlier and more reliable alerts, designed to complement China's multipoint trigger early-warning framework. Methods: This system integrates heterogeneous signals, including official epidemiology, digital traces, mobility, meteorology, and pathogen genomics, using semantic harmonization and a hybrid analytic stack. Seasonality-adjusted baselines with anomaly detection, mobility- and climate-aware SEIR models, and short-horizon learners generated calibrated early-warning scores. Thresholds were constrained by positive predictive value. Pilot studies were conducted for coronavirus disease 2019 (COVID-19) in Yantai and severe fever with thrombocytopenia syndrome virus (SFTSV) in Shandong and Henan, with tuberculosis indicators embedded for programmatic use. Results: Across deployments, the system achieved 83.3% sensitivity and 76.9% positive predictive value, providing a median lead time of 9.3 days before official confirmation. Forecasting accuracy reached 92.1% for COVID-19 in Yantai, 90.3% for SFTSV in Shandong, and 89.8% for SFTSV in Henan. Early warnings were aligned with subsequent confirmations and supported targeted screening and resource allocation. Conclusion: An open-source data-driven hybrid modeling system can deliver calibrated and timely alerts across diverse pathogens. By broadening inputs, enabling cross-agency linkage, and offering operator-oriented dashboards, it serves as a practical complement to China's national early-warning system and has the potential for scaling out with One Health inputs.