UC Berkeley

Abstract and Findings

In our previous work, we analyze, in near-real time, evolution of Covid-19 epidemic in China for the first 22 days of reliable data (up to February 6, 2020). In this work, we used the data for the whole 87 days (up to March 13, 2020) in China and the US data available till March 31 (day 70) for systematic evaluation of the logistic model to predict epidemic growth. We sequentially estimated sets of model parameters (maximum number of cases K , growth rate r , and half-time t 0 ) and the epidemic “end time” t 95 (defined as the time when the number of cases, predicted or actual, reached 95% of the maximum). The estimates of these parameters were done for sequences of reported cases growing daily (back-casting for China and forecasting for the US). In both countries, the estimates of K grew very much in time during the exponential and nearly exponential phases making longer term forecasting not reliable. For the US, the current estimate of the maximum number of cases K is about 265,000 but it is very likely that it will grow in the future. However, running estimates of the “end time” t 95 were in a much smaller interval for China (60 – 70 days vs . the actual value of 67). For the US, the values estimated from the data sequences going back two weeks from now range from 70 to 80 days. If the behavior of the US epidemic is similar to the previous Chinese development, the number of reported cases could reach a maximum around April 10 to 14 .