SHIVIR - An Agent-Based Model to assess the transmission of COVID-19 in India

Abstract

Background: COVID-19 has tormented the global health and economy like no other event in the recent past. Researchers and policymakers have been working strenuously to end the pandemic completely. Methodology/ Principal Findings: Infectious disease dynamics could be well-explained at an individual level with established contact networks and disease models that represent the behaviour of the infection. Hence, an Agent-Based Model, SHIVIR (Susceptible, Infected, Admitted, ICU, Ventilator, Recovered, Immune) that can assess the transmission dynamics of COVID-19 and the effects of Non-Pharmaceutical Interventions (NPI) was developed. Two models were developed using to test the synthetic populations of Rangareddy, a district in Telangana state, and the state itself respectively. NPI such as lockdowns, masks, and social distancing along with the effect of post-recovery immunity were tested across scenarios. The actual and forecast curves were plotted till the unlock phase began in India. The Mean Absolute Percentage Error of scenario MD100I180 was 6.41 percent while those of 3 other scenarios were around 10 percent each. Since the model anticipated lifting of lockdowns that would increase the contact rate proportionately, the forecasts exceeded the actual estimates. Some possible reasons for the difference are discussed. Conclusions: Models like SHIVIR that employ a bottom-up Agent-Based Modelling are more suitable to investigate various aspects of infectious diseases owing to their ability to hold details of each individual in the population. Also, the scalability and reproducibility of the model allow modifications to variables, disease model, agent attributes, etc. to provide localized estimates across different places.