COVID-19: A Multipeak SIR Based Model for Learning Waves and Optimizing Testing

Abstract

One of the greatest challenges of the COVID-19 pandemic has been the way evolving regulation, information and sentiment has driven waves of the disease. Traditional epidemiology models, such as the SIR model, are not equipped to handle these behavioral based changes. We propose a novel multipeak SIR model, which can detect and model the waves of the disease. We bring together the SIR model’s compartmental structure with a change-point detection martingale process to identify new waves. We create a dynamic process where new waves can be flagged and learned in real time. We use this approach to extend the traditional SEIRD model into a multipeak SEIRD model and test it on forecasting COVID-19 cases from the John Hopkins University dataset for states in the United States. We found that compared to the traditional SEIRD model, the multipeak SEIRD model improves MAPE by 10%-15% for the United States, and by 25%-40% in the specific regions that were hit by the multiple waves. We then pair this model with an optimization model for testing, which is critical in managing the epidemic and which significantly outperforms alternative testing strategies (more than 57% in detection rate). We show how to prioritize symptomatic, asymptomatic and contact tracing populations, most interestingly when balancing testing early to reach contact tracers and saving tests for later when the epidemic is worse.