We read with interest the research letter on estimating case-fatality risk for coronavirus disease (COVID-19) by Wilson, et al. (1). In their analyses, the authors estimated the case-fatality risk adjusted to a fixed lag time to death. They acknowledged that the calculated adjusted case-fatality risk (aCFR) might be influenced by residual uncertainties from undiagnosed mild COVID-19 cases and a shortage of medical resources. However, we believe the time-varying number of cumulative cases and deaths also should be considered in the epidemic profile.

Because of the exponential growth curve of the COVID-19 outbreak, the numbers of cumulative cases and cumulative deaths have been relatively close to each other in the early stages of the outbreak, leading to a much higher aCFR. As the outbreak progresses, the ratio of the cumulative cases and deaths declines, which reduces the aCFR. Thus, a higher aCFR does not necessarily indicate increased disease severity.

To test our hypothesis, we performed a simulation study by using a susceptible-infectious-recovered–death model and parameters set according to prior studies. We set the infectious period as 10 days (2); case-fatality risk as 3% (3); basic reproductive ratio (R₀) as 2.5 (4); recovery rate as 1/13 day (5), that is, 13 days from illness onset to recovery; and the population size as 1 million. We compared crude case-fatality risk, aCFR per Wilson et al.’s method, and aCFR per Mizumoto et al.’s method (6). Although the case-fatality risk calculated from these methods all are biased at the early stage of the epidemic, case-fatality risk calculated from Mizumoto et al.’s method was closer to the true case-fatality risk of 3% (Figure).

In conclusion, we recommend the Mizumoto et al. method (6) to calculate aCFR in real time. When comparing real-time estimation of the case-fatality risk across countries and regions, our results indicate that the current trajectory of the epidemic should be considered, particularly if the epidemic is still in its early growth phase.