Why estimating population-based case fatality rates during epidemics may be misleading

Lucas Böttcher; Mingtao Xia; Tom Chou

is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2020.03. 26.20044693 doi: medRxiv preprint In general, even if the cohort initially tested was only a fraction of the total infected population, trackingM 1 (t) or M 1 p (t) of this cohort still provides an accurate estimation of the mortality rate. However, the newly infecteds that contribute to CFR and M 0 p (t) at later times may not all be tested or may be tested at different times after they were infected. A reported/tested fraction f < 1 would not directly affect the CFRs or mortality ratios if the unreported/untested individuals die and recover in the same proportion as the tested infecteds. Undertesting will overestimate true CFR or mortality rates if the untested infecteds are less likely to die than the tested infecteds. In other words, if the untested (presumably because they were mildly or asymptomatic) population predominantly recovers instead of dying, the actual CFR and mortality ratios would be significantly lower than those based on tested individuals. If untested infecteds do not die, the asymptotic mortality of all infected individuals M 0,1 p (∞) ≈ M 0,1 p (∞) (see the SI). Current estimates show that only a minority of SARS-CoV-2 infections are reported (e.g., f ≈ 14% in China before January 23, 2020) [26] .