Spatial heterogeneity of influenza immunity and infections: the effect of population density and effective neighbourhood size

Abstract

BACKGROUND: Although influenza is considered a pervasive global risk, the spatial heterogeneity of risk has not been studied, particularly in urban and rural locations in the same country. Despite numerous theoretical (Ferguson et al. Nature. 2005; Germann et al. Proc Natl Acad Sci U S A. 2006) and empirical (Yang et al. PLoS One. 2012; Acuna-Soto et al. PLoS One. 2011) studies of the initial waves of pandemics, little is known about the interaction among human population structure, movement, and influenza transmission over longer periods. MATERIALS AND METHODS: We recruited 1424 individuals from 40 randomly selected spatial locations in southern China aged 2-90 years and obtained peripheral sera antibody titres against 5 historically circulating strains of H3N2. Locations were chosen randomly from a fan-shaped geographic transect extending outward 90 km from the centre of Guangzhou, China. We used generalised additive models to investigate the effect of spatial covariates in models that included age and other individual-level covariates on titre levels as well as seropositivity as defined by a threshold {Lessler:2012fw}. RESULTS: We showed that local population density was consistently positively associated with increased titres across viruses representing multiple antigenic clusters{Smith:2004jc}, even at this small spatial scale. We found that each log increase in population density resulted in a 0.78 log increase (95% CI, 0.29-2.47) in titre, averaged across 5 strains of influenza virus. We defined the effective neighbourhood size (ENS) to incorporate population density and connectivity: for a given location, the ENS was all of those living within a given perimeter added to a proportion of those not within, discounted according to a power-law function of distance. We determined the optimal ENS model to explain our data and compared it with models of transport and findings from other disease systems. Parallel ENS models of titre thresholds (seropositivity) suggest an increase in odds of 1.32 (95% CI, 1.12-1.78) in the cumulative attack rate between urban and rural areas. Even over entire antigenic clusters, is seems that less dense rural communities may avoid large outbreaks of some influenza strains.