COVID-19: The extraction of the effective reproduction number from the time series of new cases

TL;DR

This study evaluates five algorithms for estimating COVID-19's effective reproduction number from case data, using simulated and real-world data, revealing underestimation issues when the number exceeds one.

Contribution

It compares the performance of five popular algorithms on simulated and real COVID-19 data, highlighting their tendency to underestimate the reproduction number above one.

Findings

All five algorithms underestimate R when it exceeds 1.

Simulated data helped establish input-output relations for each algorithm.

Corrected R values were obtained for five countries over nearly a year.

Abstract

Addressed in this work is the performance of five popular algorithms, which aim at assessing the dissemination dynamics of the COVID-19 disease on the basis of the time series of new confirmed cases. The tests are based on simulated data, generated by means of a deterministic compartmental epidemiological model \cite{Matsinos2020a}, adapted herein to also include the possibility of the loss of immunity by the group of the recovered (or vaccinated) subjects. Assuming a simple temporal dependence of the effective reproduction number (the exact details are of no relevance as far as the conclusions of this work are concerned), time series of new cases were generated in a time domain of nearly one year for the five top-ranking countries in the cumulative number of infections by January 1, 2021. These countries are (in descending order of infections): the United States of America, India, Brazil, Russia, and the United Kingdom. The processing of each simulated time series led to the establishment of relations between the input (actual) and the reconstructed values of the effective reproduction number for each country and algorithm, separately; this work argues that all five algorithms underestimate the effective reproduction number when the latter exceeds the critical value of $1$. The five algorithms were subsequently applied to the real-life time series of new cases for the aforementioned five countries, which also span a temporal interval of nearly one year; corrected values of the effective reproduction number are obtained for these countries in 2020.