Advancing real-time infectious disease forecasting using large language models.

Area of Science:

• Epidemiology
• Artificial Intelligence
• Computational Biology

Background:

• Forecasting short-term disease spread is complex due to multi-modal variables, public policy, and human behavior.
• Existing models struggle to integrate diverse, real-time data effectively.

Purpose of the Study:

• Introduce PandemicLLM, a framework using multi-modal large language models (LLMs) for real-time disease spread forecasting.
• Reformulate disease spread prediction as a text-reasoning problem to incorporate non-numerical data.

Main Methods:

• Developed an AI-human cooperative prompt design and time-series representation learning to encode multi-modal data for LLMs.
• Utilized textual public health policies, genomic surveillance, and epidemiological time-series data.
• Applied and tested the model across all 50 US states for 19 months using COVID-19 data.

Main Results:

• PandemicLLM successfully incorporated heterogeneous pandemic-related data formats.
• Demonstrated performance benefits over existing disease spread forecasting models.
• Enabled real-time forecasting by treating disease spread as a text-reasoning problem.

Conclusions:

• PandemicLLM offers a novel framework for integrating diverse data types in disease outbreak forecasting.
• The approach shows promise for enhancing the accuracy and timeliness of public health predictions.
• Opens new avenues for AI-driven analysis of complex, real-world epidemiological challenges.