Comparing the Performance of Three Computational Methods for Estimating the Effective Reproduction Number

The effective reproduction number $$ is one of the most important epidemiological parameters, providing suggestions for monitoring the development trend of diseases and also for adjusting the prevention and control policies. However, a few studies have focused on the performance of some common computational methods for R. The purpose of this article is to compare the performance of three computational methods for R: the time-dependent (TD) method, the new time-varying (NT) method, and the sequential Bayesian (SB) method. Four evaluation methods-accuracy, correlation coefficient, similarity based on trend, and dynamic time warping distance-were used to compare the effectiveness of three computational methods for R under different time lags and time windows. The results showed that the NT method was a better choice for real-time monitoring and analysis of the epidemic in the middle and late stages of the infectious disease. The TD method could reflect the change of the number of cases stably and accurately, and was more suitable for monitoring the change of R during the whole process of the epidemic outbreak. When the data were relatively stable, the SB method could also provide a reliable estimate for R, while the error would increase when the fluctuation in the number of cases increased. The results would provide suggestions for selecting appropriate R estimation methods and making policy adjustments more timely and effectively according to the change of R.