Predictions, role of interventions and effects of a historic national lockdown in India's response to the COVID-19 pandemic: data science call to arms

Debashree Ray; Maxwell Salvatore; Rupam Bhattacharyya; Lili Wang; Shariq Mohammed; Soumik Purkayastha; Aritra Halder; Alexander Rix; Daniel Barker; Michael Kleinsasser; Yiwang Zhou; Peter Song; Debraj Bose; Mousumi Banerjee; Veerabhadran Baladandayuthapani; Parikshit Ghosh; Bhramar Mukherjee

To implement the eSIR model, a Bayesian hierarchical framework is assumed where the proportions of infected and the removed people are modeled using a Beta-Dirichlet state-space model while a latent Dirichlet distribution is assumed for the underlying unknown prevalence of the three states. Priors for the basic reproductive number R0, disease removal rate (consequently, the transmission rate) and the underlying unobserved prevalence of the susceptible, infected and removed states at the starting time are considered. Using the current time series data on the proportions of infected and the removed people, a Markov chain Monte Carlo implementation of this Bayesian model provides not only posterior estimation on parameters and prevalence of all the three compartments in the SIR model, but also predicted proportions of the infected and the removed people at future time point. The posterior mean estimates of the unobserved prevalence at both observed as well as future time points come along with 95% credible intervals (CI). To get predicted case-counts from the predicted prevalence, we used 1.34 billion as the population of India, thus treating the country as a homogeneous system for the outbreak. 15