Climate Factors, Air Pollutants, and Extreme Events Shape Influenza Risk in China: A Hospital-Based Case-Crossover and Machine-Learning Study

Abstract

Although the link between seasonal influenza and environment has been established, the heterogeneous impacts of climate extreme events with their interactions remain poorly understood. Based on 630,120 influenza-positive cases from hospitals in 335 Chinese cities (2005–2019), this case-crossover study adopted distributed lag nonlinear models to quantify exposure-lag-response relationships between environmental exposures and influenza, applied machine learning to rank dominant variables, and performed a Long Short-Term Memory (LSTM) model using Beijing as proof of concept to test the forecasting utility of extreme events. Meteorological variables, air pollution, and extreme events are significantly associated with influenza risk. Machine learning models indicate that atmospheric temperature, humidity, and particulate matter may be the primary contributors. Co-exposure to extreme events showed a higher joint risk than single exposure, with a significant submultiplicative interaction on the multiplicative scale and no evidence of additive interaction. Based on the Beijing-specific LSTM proof-of-concept framework, integrating extreme events significantly enhanced daily influenza forecasting performance. Climate factors, air pollution, and extreme events potentially shape influenza risk, with heterogeneous effects across subtypes, demographics, and geography. These findings underscore the potential benefits of integrating extreme climate events into influenza forecasting and precise public health responses in the era of climate change.