Rethinking Case Fatality Ratios for COVID-19 from a data-driven viewpoint

TL;DR

This paper proposes a data-driven method to accurately estimate the true COVID-19 case fatality ratio by accounting for time delays between reporting and death, revealing a consistent CFR over time across countries.

Contribution

It introduces a simple algorithm to find the optimal time lag for CFR calculation, demonstrating that corrected CFR remains constant over time and improves early mortality estimates.

Findings

Corrected CFR is constant over months and across countries.

Optimal time lag varies per country but can be identified with the method.

Traditional CFR is artificially time-dependent and less reliable.

Abstract

The case fatality ratio (CFR) for COVID-19 is difficult to estimate. One difficulty is due to ignoring or overestimating time delay between reporting and death. We claim that all of these cause large errors and artificial time dependence of the CFR. We find that for each country, there is a unique value of the time lag between reported cases and deaths versus time, that yields the optimal correlation between them is a specific sense. We find that the resulting corrected CFR (deaths shifted back by this time lag, divided by cases) is actually constant over many months, for many countries, but also for the entire world. This optimal time lag and constant CFR for each country can be found through a simple data driven algorithm. The traditional CFR (ignoring time lag) is spuriously time-dependent and its evolution is hard to quantify. Our corrected CFR is constant over time, therefore an important index of the pandemic in each country, and can be inferred from data earlier on, facilitating improved early estimates of COVID-19 mortality.