Estimating case fatality rates of COVID-19

In their model, David Baud and colleagues¹ exclude individuals who die within 14 days of testing positive for severe acute respiratory syndrome coronavirus 2. If an individual contracts symptoms on March 1, tests positive on March 10, and dies on March 11, they would not be included in the denominator for case fatality rate (CFR) on March 11. In addition, patients might test positive up to 13 days after recovery.² As testing is expanded, an asymptomatic patient infected 1 week before testing positive on March 3 should, but will not, be included in calculations for March 16.

For the stated purpose of the authors, it might be useful to also include everyone who was symptomatic for 14 days before the calculated date. Consider a CFR calculation for March 15. If an individual was symptomatic on March 1 and tested positive on March 10, they would not be included in the denominator, even though the patient probably contracted the disease before the 14-day lag time.

Moreover, patients with mild symptoms might not undergo testing and so might not be included in the overall denominator. A further consideration is the delay between testing and receipt of results. Consequently, individuals might not test positive until after the suggested 14-day incubation period. With disease spread, indications for COVID-19 testing will expand, thereby increasing the denominator size. These factors might act as time-varying confounding variables in the authors' calculation of CFR.

One other published lag-time calculation has included half the additional cumulative deaths in the numerator and time from illness to death as the lag-time (13 days).³ For example, if calculating the CFR for March 15, the denominator would be cumulative cases until March 2, and the numerator would be cumulative deaths until March 2, in addition to half of the deaths recorded from March 2–15. This method assumes a normal distribution of time from illness to death.

Although underestimation of CFRs risks the population not taking the threat seriously, overestimation might lead to unnecessary additional panic and concern. During a rapidly evolving pandemic, accurate measures of disease characterisation are important. Future estimates will probably involve patient-specific data for refined calculation. However, the provided CFR estimate of 15·2% for countries outside China might be a premature statistic owing to the limitations of their methods.