Spatial structure influences the population dynamics of endemic cholera and its response to climate forcing in a megacity

Background/Question/Methods

The population dynamics of endemic cholera in urban environments remain poorly understood, because of the considerable environmental heterogeneity in emerging megacities that lead to fine-scale spatial structure.  Previous studies, in particular those demonstrating the influence of the El Niño Southern Oscillation (ENSO) on multiannual cycles of disease incidence, have used aggregated cases over whole cities and relatively large spatial scales.

More recent work with probabilistic spatial models and remote sensing analyses showed the existence of two distinct regions in the megacity of Dhaka, Bangladesh: an older core region in the central part of the city comprised by more established ‘thanas’ (administrative subdivisions), and a newer urban periphery surrounding the core, hereafter referred to as the “inner” and “outer” region respectively.  To better understand the role of this spatial structure on cholera dynamics, we formulated a coupled transmission model linking the two regions via movement, and incorporating a self-limiting effect in epidemic growth via an exponent α in the force of infection.  We used recently developed likelihood-based inference methods to parameterize the model with space-time surveillance data from several thanas in the city.  We compared models to address, among others, how the spatial and demographic heterogeneity in the two regions, and their dynamical coupling, impact local cholera transmission and its sensitivity to ENSO.

Results/Conclusions

Both the population density and disease incidence are significantly higher in the inner relative to the outer region.  While the (pre-monsoon) spring transmission season coincides during March-June for both regions, the (post-monsoon) fall season occurs later in the periphery.  The maximum-likelihood estimate of the exponent α is significantly below 1 in the inner region (~1 in the outer region), indicating a strong self-limiting effect that can result from highly non-random mixing in the inner city area.  The coupled transmission model has strong statistical support compared to an uncoupled model, which suggests long-distance movement between the two regions.  A similar comparison between models with and without ENSO indicates a significant effect of this covariate on transmission especially in the core of the city.

            Responses to climate variability reflect important heterogeneity in the spatial structure within the city.  Our approach combining process-based models and coarse-grained high-resolution spatiotemporal case data can provide effective ways to analyze other disease systems in megacities of the developing world.