Correlation and causality of climate with global human seasonal influenza

Abstract

The recurring epidemics of human influenza has been causing a wide impact on the global health. Understanding the drivers for global epidemiology of seasonal human influenza would help to develop better control measures. Although evidence has been presented to relate the seasonally varying pattern of the influenza prevalence to the climatic variation, yet our understanding of their relationships remains limited. Since the climate itself is a complex system of multiple factors varying over the time and geographical space, accurate identification of the climatic drivers of influenza and understanding the underlying mechanism are challenging and require the holistic analysis that considers multivariate effects and non-stationarity in various dimensions. This thesis aims to study the relationship (correlation and causality) between various climatic factors and the occurrence of influenza in both temporal and geographical dimensions, with some innovative extensions of existing analysis methods.

Seasonality of human influenza disease was qualitatively and quantitatively analyzed using the influenza data sets collected from global sampling sites, which detect strong seasonality in high-latitude regions. This indicates a possible correlation between the influenza prevalence and climatic factors, which is further complicated with the spatiotemporal non-stationarity detected in the periodicity of the influenza epidemics. Moreover, the different pattern between influenza A and B reveals that their drivers and the underlying mechanisms may be different.

A novel statistical method named the geographically weighted temporally correlated logistic regression (GWTCLR) model is proposed to study the correlation between the influenza occurrence and climatic factors when spatiotemporal non-stationarity is also considered. The construction of the estimate and its asymptotic properties are presented. Simulation is done to verify the robustness of the model. The model is applied to the influenza dataset and reveals the geographical and temporal variation of the correlation between temperature and influenza occurrence.

Since correlation could not straightly imply causality, the empirical dynamic model is used to identify the climatic drivers of influenza and generalize the mechanism of such causality. Temperature and absolute humidity are found to be important factors that cause the influenza A and B. Given the strong link between the temperature and humidity, the combined causal effect of these two factors was particularly studied. For influenza A, the combined causal effect is not stationary. A threshold is detected where this causal effect changes its direction between positive and negative. This provides an explanation of the difference between the annual pattern and the semi-annual pattern of influenza in different regions.

In summary, this thesis reveals and characterizes the correlation and causality between the influenza epidemics and several climatic factors including temperature and absolute humidity. Such relationships are non-stationary in time and geographic space, which importantly lead to the discovery of the dynamic causal links capable to explain the diverse patterns of influenza epidemics in different regions and seasons including the enigmatic semi-annual epidemics in subtropical regions.