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

Area of Science:

• Epidemiology
• Public Health Surveillance
• Computational Biology

Background:

• Globalization increases imported epidemic risks.
• Current surveillance systems are fragmented and rely on lab confirmation.
• China's multipoint trigger early-warning framework needs enhancement.

Purpose of the Study:

• To develop an open-source, data-driven hybrid modeling system for earlier and more reliable epidemic alerts.
• To complement China's existing national early-warning system.
• To integrate diverse data sources for improved surveillance.

Main Methods:

• Integrated heterogeneous signals: official epidemiology, digital traces, mobility, meteorology, and pathogen genomics.
• Utilized semantic harmonization and a hybrid analytic stack.
• Employed seasonality-adjusted baselines, anomaly detection, SEIR models, and short-horizon learners for early-warning scores.

Main Results:

• Achieved 83.3% sensitivity and 76.9% positive predictive value.
• Provided a median lead time of 9.3 days before official confirmation.
• Demonstrated high forecasting accuracy for COVID-19 and SFTSV.

Conclusions:

• An open-source hybrid modeling system provides calibrated, timely alerts for diverse pathogens.
• The system enhances China's national early-warning system and has potential for scaling.
• Broadened inputs and cross-agency linkage improve public health response and resource allocation.