Associations Between Meteorological Factors and Influenza A/B Incidence in Subtropical China: A Six-Year Surveillance Study with Deep Learning Modelling for Influenza Early Warning
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Influenza burden in subtropical regions like southeastern China is shaped by meteorological factors-driven complex transmission patterns that differ from temperate zones, challenging traditional surveillance and necessitating advanced predictive tools. This study aimed to characterize non-linear associations and lagged effects between meteorological variables and influenza A/B incidence in subtropical China and develop a deep learning model for predictive surveillance. From 2018-2023, we analyzed 3155 laboratory-confirmed infections among 20488 influenza-like illness reports from seven sentinel hospitals in Putian. Distributed lag non-linear models (DLNM) examined meteorological impacts and type-specific influenza incidence, while a Bayesian-optimized Long Short-Term Memory (LSTM) neural network, incorporating COVID-19 pandemic indicators, was constructed, trained on 2018-2022 data, and validated with 2023 data. DLNM analysis identified distinct meteorological drivers for influenza subtypes. Influenza A incidence significantly increased at 26– 31°C (relative risk [RR] = 97.73, 95% CI: 15.24–626.71 at 30° C) and 82– 92% humidity (RR = 16.27, 95% CI: 6.39–41.42 at 87%), relative to reference values (24.1°C, 75.8%). Conversely, influenza B risk surged at ≤20°C and >20 mm rainfall. Cumulative lagged effects extended up to 15 days, indicating type-specific vulnerabilities. The LSTM model demonstrated strong predictive accuracy, outperforming conventional models (Mean Absolute Error [MAE]: 1.71 vs. 2.03 for Influenza A, 0.38 vs. 1.02 for Influenza B). External validation in neighbouring Sanming confirmed the LSTM network’s robustness and generalizability across subtropical regions. Our study unravels complex meteorological influences on subtropical influenza and introduces a DLNM-LSTM framework for climate-adaptive digital solutions. Its validated precision supports real-world applicability and targeted public health interventions, offering actionable insights for climate-sensitive disease surveillance.
Importance
This study addresses critical gaps in understanding how meteorological factors affect infectious disease transmission in subtropical regions, where traditional seasonal patterns are less pronounced. By identifying specific meteorological dynamics that trigger increased influenza incidence, with distinct patterns for influenza A versus B, our research enables more targeted public health responses. Our innovative integration of distributed lag non-linear models (DLNM) with Bayesian-optimized Long Short-Term Memory (LSTM) neural network represents a significant advancement in digital health surveillance technology. This approach not only outperforms traditional forecasting methods but also demonstrates robust cross-regional applicability, offering a scalable solution for real-time monitoring. As climate change continues to alter disease transmission patterns globally, our framework provides health systems with a powerful digital tool to anticipate outbreaks and optimize resource allocation. This work exemplifies how machine learning can transform climate-sensitive infectious disease surveillance, ultimately enhancing pandemic preparedness in an increasingly unpredictable world.
