Epidemic Landscape and Forecasting of SARS-CoV-2 in India

Aravind Lathika Rajendrakumar; Anand Thakarakkattil Narayanan Nair; Charvi Nangia; Prabal Kumar Chourasia; Mehul Kumar Chourasia; Mohammad Ghouse Syed; Anu Sasidharan Nair; Arun B Nair; Muhammed Shaffi Fazaludeen Koya

We used crowd sourced time series data available from the internet to estimate country specific parameters for the epidemic. 11 As of now, this is the best available database for information on SARS-CoV-2 in India in public domain. The most reliable information in this database is regarding the daily reported incidence of COVID cases. Also, it contains aggregated information on total confirmed cases, total death and total recovery. We did not use variables with limited information. The data was scraped on to R software on 12 th April 2020, cleaned and reshaped for the current analysis. 12 Package ggplot2 was mainly used to create figures along with features from base R. 13 We defined serial interval as the time difference in diagnosis of SARS-CoV-2 in infectee and infector. 14 We assumed serial interval and generation time to be same. Growth rate was computed from this incidence curve by fitting a log-linear model 15 and basic reproduction number (R0) was obtained from previously calculated serial interval. We also calculated time dependent reproductive number (Rt) to show the change in infectivity over time. State-wise growth was measured using the slope from a linear model.To estimate an R0 for projection purposes, in addition to log linear, we also estimated R0 from other methods such as Exponential growth (EG), Maximum likelihood estimation (ML) and Sequential Bayesian Method (SB) and Estimation of time dependent reproduction numbers (TD). 16 We selected Sequential Bayesian Method (SB) to account for stochasticity in incidence curves as the priors change with time and are drawn from posterior distribution from time dependent informative priors. 16 Conceptually, susceptible individuals become infectious (i.e., move from the susceptible compartment to the infectious compartment), and then ultimately recover from the infection (i.e., move from the infectious compartment to the recovered compartment). The rates at which they move from one compartment to another depend on the proportion of the population in each of these compartments, as well as the transmission and recovery rates associated with the disease. We modified standard SIR models to accommodate an additional compartment for deaths by assuming transition probabilities, the susceptible-infectiousrecovered/Death (SIRD) model. Our model can be represented by the following ordinary differential equation