Stochastic Compartmental Modelling of SARS-CoV-2 with Approximate Bayesian Computation

Vedant Chandra

For most provinces, there is an excellent agreement between the SIR-ABC model and the total number of confirmed cases. The fit is less perfect for the individual infected-recovered curves. This is to be expected, since the real-world obviously does not truly follow an SIR model. There are various externalities like spatial effects and government/healthcare responses. Our simple SIR model also lacks vital statistics like births and deaths. For a fatal illness like SARS-CoV-2, it would be valuable to add these parameters to the model. However, for the purpose of this proof-of-concept study, we estimate that adding these parameters will negligibly affect the goodness-of-fit of the total confirmed cases (Chen & Li 2020) . We extrapolate the model for each region by allowing it to run until no active infections remain (Fig. 3) . We find a consistent extrapolated infection profile for all the provinces under study. This indicates a similar level of government response after the first infections were reported, despite differing population sizes in each region. We quantify the 'steepness' of the infection curve by dividing the maximum number of active infected patients by the total length of the extrapolated infection curve, i.e. the duration of the epidemic. We compare the steepness of different Chinese provinces in Fig. 4 .