Foundation time series models for forecasting and policy evaluation in infectious disease epidemics

Epidemic forecasting and policy evaluation rely on mathematical models to predict infectious disease trends and assess the impact of public health policies. Traditional models typically require extensive epidemiological data and may struggle in data-limited settings. Transformer-based, foundation AI models have demonstrated strong predictive capabilities in various time series applications. We investigated whether they can be the basis of a new epidemic modeling framework. We evaluated five foundation models - TabPFN-TS, TimeGPT, TimesFM, Lag-Llama, and Chronos - across diverse pathogens, diseases and locations, including influenza-like illness, RSV, chickenpox, dengue, COVID-19 and neonatal bronchiolitis. Models were tested for long-term forecasting (multi-season predictions), short-term forecasting (four-week-ahead predictions), and epidemic peak timing estimation. We also assessed their ability to generate counterfactual scenarios in policy evaluation, using COVID-19 restriction measures in Italy, RSV immunization in France, and synthetic epidemic data as validation. Foundation models demonstrated strong predictive accuracy, possibly outperforming traditional statistical and mechanistic models in data-limited contexts. They generated multi-season forecasts and short-term forecasts with good accuracy and stable uncertainty. They gave reliable peak timing estimates months before the actual peak. In policy evaluation, TabPFN-TS accurately estimated intervention effects, matching estimates from an independent epidemiological study. Our findings suggest that foundation models can complement existing approaches in epidemic modeling. Their ability to generate accurate forecasts and counterfactual analyses with minimal data input highlights their potential for real-time public health decision-making, particularly in emergent and resource-constrained settings. Further research should explore domain-specific adaptations to optimize performance for infectious disease modeling.