Survival-Convolution Models for Predicting COVID-19 Cases and Assessing Effects of Mitigation Strategies

Yuanjia Wang, Professor of Biostatistics, Department of Psychiatry

Background Countries around the globe have implemented unprecedented mitigation measures to mitigate the coronavirus disease 2019 (COVID-19) pandemic. We aim to predict COVID-19 cases and compare effectiveness of mitigation measures across coun- tries to inform policy decision making.
Methods We propose a survival-convolution model for predicting key statistics of COVID-19 epidemics (daily new cases). We account for transmission during a pre- symptomatic incubation period and use a time-varying reproduction number (Rt) to reflect the temporal trend of transmission and change in response to an intervention. We estimate the intervention effect on reducing the infection rate and quantify uncer- tainty by permutation.
Findings Our model adequately estimated observed daily new cases and could predict the entire disease epidemic using data from the early phase. A fast rate of decline in Rt was observed in China and South Korea. In Italy, Rt decreased at a slower rate and did not change significantly before the nation-wide lockdown and two-weeks after. In the United States (US), there was a significant change in Rt before and after the declaration of national emergency.
Interpretation Adopting mitigation strategies early in the epidemic is effective in reducing the infection rate. The lockdown in Italy did not further accelerate the speed at which the infection rate decreases and the epidemic is not yet under control. If the current trend continues in the US, COVID-19 may be controlled by May 24 (CI: May 15 to Jun 9). However, relaxing mitigation measures could delay the end date of the epidemic as long as 42 days.