Table 1.
Reported CFRcrude, CFRactual, and corrected IFR values for China, the United Kingdom and the United States.
The table summarizes CFRcrude for each country region at the time of the report, calculated CFRactual and IFR values through early May 2020. Details are available in the cited references [2, 4, 6–14, 16, 19, 21–24, 32, 33, 40, 41]. For the USA and UK the CFRcrude on April 15, 2020 is listed. Studies are listed by their first author or by the location of the modeling group that reported them.
Fig 1.
CFRcrude(t) and CFRclosedcase(t) versus time for Germany.
The bottom curve (red) shows CFRcrude(t) plotted versus day after outbreak. The top curve (blue) shows the same for CFRclosedcase(t). CFRcrude(t) increases over this period from a value of 0.12% to a value of 4.36%. It is seen that CFRclosedcase(t) converges to the projected true of CFRcrude earlier than the CFRcrude(t) curve itself.
Table 2.
Comparison of the time corrected CFRcrude values calculated using the closed case convergence method versus the standard time to fatality time correction method [8–12, 14, 15, 26].
Fig 2.
Simulated and reported CFRcrude(t) versus time curves for Germany.
The reported CFRcrude(t) curve is plotted in black. Even though the reported CFRcrude(t) curve rises by more than 10-fold, it is well matched throughout the duration by the simulated CFRcrude(t) curve (blue) using the CFRcrudetimecorrected value of 5.0% determined from the entire time course. Therefore, even early in the outbreak, when CFRcrude(t) was 10-fold lower than on May 7, 2020 (the last day used) the time correction method would have accurately predicted the true CFRcrude(t) value.
Table 3.
Age specific fractions of cases, age specific corrected CFR, and contributions of each age group to the overall corrected CFR for each country.
Fig 3.
Linear regression analysis of CFR*crudetimecorrected(70+), CFR*crudetimecorrected(70+)A, and CFR*crudetimecorrected(0−69) versus percent of cases 70 years old and above (p(70+)).
A shows a plot of CFR*crudetimecorrected(70+) (blue) and CFRcrudetimecorrected(0−69) (green) for each country versus the percent of cases 70 years old and above (p(70+)). It is seen that for all countries the CFRcrudetimecorrected(70) term explains the large majority of CFRcrudetimecorrected (81% +/- 8%). The majority of the variance in CFRcrudetimecorrected(70) is explained by cases 70 years old and above (R2 = 0.82). B shows a plot of CFRcrudetimecorrected(70+A) (blue) for each country. The value of cCFR70+A for each country was calculated by adjusting the fraction of cases in the 70 and over group who are 80 years old and above to be 40% (p(80+)/p(70+) = 0.40), which is the mean of the countries examined (Table 3). The higher fraction of the variance explained by age for CFRcrudetimecorrected(70+A) (R2 = 0.89) indicates that the percentage of the population 80 years and older are an important factor in determining the average population value of CFRcrude.
Fig 4.
Reported percentage of New York City adults infected with COVID-19 versus percentage calculated from our and other reported IFR values prior to May 7, 2020.
As shown in the inset, the predicted maximum and minimum percent of the population in New York City infected with COVID-19 is within the range determined from random adult serological testing [42, 43]. For comparison, we plotted the percentage infected using the IFR values in Table 1.
Table 4.
Comparison of age specific IFR coefficients from the present study with serological testing studies internationally [44] and a comprehensive analysis of results from NYC [25] To facilitate comparison, we calculated a 0–64 group mean value for Yang et al. [25] and a 0–69 mean value for Seoane [44] by averaging their reported age sub group IFR values and weighting by percentage of each subgroup of the total infected population.