key: cord-269455-pkjov371
authors: Faust, Jeremy Samuel
title: Towards a better case fatality estimate for SARS-CoV-2 during the early phase of the United States outbreak
date: 2020-05-30
journal: Clin Infect Dis
DOI: 10.1093/cid/ciaa639
sha: 
doc_id: 269455
cord_uid: pkjov371

nan

Given the above considerations, data from the model as proposed by Kou et al can be harnessed for calculating a denominator statement with acceptable face validity for symptomatic disease at three time points, ranging from March 7 to March 21. We can further posit an alternative denominator that takes asymptomatic infection into account-which their model does not. While some estimates state that only 11 percent of cases are ultimately asymptomatic, other estimates are closer to 18 percent. But presymptomatic disease comprises a substantial fraction of infections at any given time and should therefore also be considered. Universal screening among one healthy population detected that the rate of asymptomatic or presymptomatic disease was as high as 88 percent. 3 Another study of older patients who were sicker at baseline found that 56 percent of patients with a positive SARS-CoV-2 swab were asymptomatic at the time of testing, and only developed symptoms later (median time from test to symptoms = 4 days). Such patients would not be picked up in the final data point in use. Taken together, a reasonably conservative A c c e p t e d M a n u s c r i p t attempt to add symptom-free cases to numbers proposed by Kou et al could include a 50 percent addition to their estimates. 4 5 Numerator statements, meanwhile, can reasonably be assumed to be sufficiently close to the running cumulative total number of counted covid-19 deaths as recorded at least two weeks after the day used to estimate the denominator. These counts are relatively reliable because covid-19 is currently a reportable cause of death in all US states and territories. While excess deaths may ultimately offer an attractive alternative for use as the numerator, expected lags in all-cause mortality reporting renders these numbers incomplete for several weeks. 6 Once those numbers are available, they may serve as a partial measure of quality for numerator statements based on counted covid-19 deaths, which are prone to some degree of error. Thus, the use of excess mortality may at some point provide another lens through which to verify the accuracy of these counts, as excess all-cause mortality figures does not rely on the subjective judgement of those filling out death certificates.

March 21 st appears to be the best available date upon which to estimate a denominator for the CFR of SARS-SoV-2 using the model provided by Kou et al. This date has the advantage both of being the peak of ILI reporting to the CDC while being directly prior to the time when the effects of many of the mitigation strategies and changes in public behavior mentioned above began to become noticeable on a systemic level. As of April 5 th , public COVID-19 trackers reported a crude CFR of 3.5 percent worldwide. Using the Kou model as a source for the denominator (cases as of March 21 st ) and all deaths through April 4 th as the numerator (including all deaths that occurred on US soil prior to March 21), the calculated CFR appears to have been approximately 22 percent of estimates on public-facing COVID19 trackers-and this only accounts for symptomatic cases ( Table 1 , column 1). Further, allowing for the addition of pre-or asymptomatic cases into the denominator reveals a CFR of just 12 percent of the figures published on COVID-19 trackers ( Table 1 , column 2).

These figures mirror estimates obtained in closed systems where universal testing was achieved, such as the Diamond Princess cruise. While the crude CFR on the Diamond Princess appears to have settled at around 1.8 percent, passengers aged 70 or older were over-represented as compared to other cohorts by a factor of approximately four. 7 8 This implies an age-adjusted CFR for the Diamond Princess of 0.45, which is remarkably similar to implied rates we calculate here using the denominator based on Kou et al with adjustment for symptom-free infection ( Table 1 , column 2). These numbers are higher, though not astronomically, than estimates given in the increasingly controversial Santa Clara County serology study. 9 If we instead use some of the higher reported numbers of pre-or asymptomatic cases found in the emerging literature, the estimated CFR we might calculate would indeed approach the 0.17 percent figure proposed by the authors of the Santa Clara study.

Together, these data imply that a more accurate CFR for SARS-CoV-2 may rest between 0.5 and 0.8 percent for symptomatic cases, and 0.2 and 0.4 percent for all cases including pre-and asymptomatic infections. However, this would also appear to imply that SARS-CoV-2 has a CFR that is between one and eight times greater than reported figures for seasonal flu. Based upon recent ground conditions during the COVID-19 outbreak compared to the peak of the worst flu seasons from recent years (as well as the 2009 H1N1 pandemic), no credible case can be

Covid-19 fatality is likely overestimated

Fact sheet MEDICARE TELEMEDICINE HEALTH CARE PROVIDER FACT SHEET

Universal Screening for SARS-CoV-2 in Women Admitted for Delivery: NEJM

Department of Medicine. Asymptomatic Transmission, the Achilles' Heel of Current Strategies to Control Covid-19: NEJM. 2020. Available at

Estimating the asymptomatic proportion of coronavirus disease 2019 (COVID-19) cases on board the Diamond Princess cruise ship

Excess Deaths Associated with COVID-19

Field Briefing: Diamond Princess COVID-19 Cases. Available at

COVID-19 Antibody Seroprevalence

A c c e p t e d M a n u s c r i p t A c c e p t e d M a n u s c r i p t