The importance of space and different routes of transmission in the dynamics of endemic cholera has been neglected mainly because of the absence of high resolution spatio-temporal data. More generally, the stochastic and spatial dynamics of infectious diseases with both an environmental reservoir and human-to-human transmission has not yet been addressed from a metapopulation perspective. An extensive data set for cholera infections from 1983 to 2004 in Matlab (Bangladesh) was geographically mapped to the level of baris (groups of households), providing an opportunity to address both the role of space and transmission routes in detail.

We present a stochastic SIRS (Susceptible-Infected-Recovered-Susceptible) model for cholera dynamics at the level of baris. In the model, a bari is considered infected (I) if at least one case is present. The transitions from the infected to the recovered state (R) and back to the susceptible state (S) are unobserved events in this framework. We modify this basic model in a series of formulations of increasing complexity to examine the respective roles and seasonalities of primary and secondary transmission, as well as the spatial scale of transmission.. We further incorporate forcing by climate variability (ENSO) and estimates of socio-economic status relevant to education and sanitation levels. .

An MCMC algorithm is used to evaluate the likelihood of these models and estimate the values of epidemiological parameters. The Akaike Information Criterion (AIC) was used for model selection. 
Results/Conclusions

Seasonality differ significantly for the two routes of transmission, indicating not just their different roles during epidemics but also between the two seasonal peaks typical of cholera in this region. A spatially restricted transmission improved the fit significantly; so did consideration of ENSO. We discuss our findings on the duration of immunity in light of conflicting results from previous non-spatial models. The metapopulation model presented here is applicable to other infectious diseases for which high-resolution spatio-temporal data is available only for infections, with other states remaining unobserved.