What is the "Fatality Rate"? IFR, CFR, and Mortality

There are a few clarifications needed to answer this question well. The "fatality rate" is not specific enough.

We typically speak of infection fatality rates (IFRs), case fatality rates (CFRs) and mortality rates.

• The IFR is the percentage of deaths among people infected with the disease;
• The CFR is the percentage of deaths among people who sought medical help and tested positive; and,
• The mortality rate is the percentage of the population which will die.

The IFR may be difficult to estimate without broad testing of the population.

The CFR is usually easier to measure since people feel symptoms, go to the doctor, and get tested. (They "opt in" to testing.) In the recent pandemic, however, even people with symptoms sometimes avoided going to the doctor, so the CFR may be biased higher -- because only people with severe symptoms go to the doctor. Since the CFR is conditional on being symptomatic (perhaps severely), we expect CFR > IFR.

Unless an entire population gets infected, we expect mortality rates to be lower than IFRs.

Estimation of CFRs (and Maybe IFRs)

You looked at Worldometer's measure of deaths/resolved cases. Since most countries are not doing universal (nor near-universal) testing, this is an estimate of the CFR: people have to have symptoms to go get tested and thus be in the denominator.

However, there is also a time element at play. Early on in an epidemic/pandemic, some people are more susceptible to infection and death. Thus we will see those people get sick and die at rates greater than for the rest of the population. That makes this number likely to be an overestimate. We also would expect it to decrease as time goes on.

Some people look at a measure of deaths/total cases. This is surely an underestimate because some of those cases have not yet been resolved and it is likely some will result in death.

Cross-Country Comparisons

If we look across countries, the picture seems to become more muddled: countries such as the US are reporting very high CFRs while places like South Korea report very low CFRs. The truth is that some of the countries (like South Korea) have tested widely. This makes their reported "CFR" much more like an IFR. Furthermore, you will see variation in their deaths/resolved cases metrics depending on their population demographics and if the virus is just starting to work through their vulnerable populations or not. We saw very high rates (near the 21% you mention) for deaths/resolved cases in the US and Italy early on because the infections in those countries hit older populations very hard early on. If you look at those numbers now, you will see they are closer to 5% to 7%.

Current Estimates of SARS-CoV-2 IFR and CFR

Finally, what are the IFR and CFR for SARS-CoV-2? So far, it seems like the IFR is around 0.3%, though some work is now suggesting a number perhaps as high as 0.6%. The CFR has generally been 10x to 20x higher which agrees with over 85% of people having no or very mild symptoms (i.e. too mild to merit going to a doctor). That would put the CFR at 5% to 9% (using a 15x multiplier).

I think the only fair answer is that it is complicated. What you want to say (e.g., fatality rate if infected vs. fatality rate if confirmed positive vs. fatality rate by cohort vs. expected population fatalities) makes a fairly big difference.

We can think of the percent of deaths vs closed cases as a reasonable upper bound on the case fatality rate (the fatality in confirmed cases). That is, it is reasonable to assume that the fatality rate in the open cases is not higher than the fatality rate in closed cases (this may not always be true — for example, if those recover usually do so quickly while fatalities usually take a long time — but it is likely reasonable for SARS-CoV-2 where the opposite appears to be true).

We can think of the percent of deaths vs all cases as the lower bound on the case fatality rate, since nothing can happen (from current cases) to lower that rate.

The true value likely falls somewhere in between the two, though where in between depends primarily on the speed of progression for fatalities and for recoveries. If every single case either died or recovered exactly 14 days after confirmation, the fatality rate vs closed cases would be a good estimate of the true case fatality rate (because it accurately measures the case fatality rate for cases that have hit 14 days). If every case that dies does so immediately after confirmation, but recovery takes an indeterminate amount of additional time, the fatality rate in all cases is accurate (because we can assume that all currently living cases will recover).

The reality is somewhere in between because we know that most cases that live beyond a certain point (say, 7 days, for example) are substantially more likely to survive. Then, the question becomes how to weight your cases to account for this difference in progression (to fatality or recovery) appropriately.

However, this neglects the prevalence of undiagnosed cases. The true infection fatality rate (the fatality rate in all infections, not just those that are diagnosed) requires an estimation of the infections that are undetected (e.g., asymptomatic carriers and those that skip testing) and the deaths that are not attributed to the disease (e.g., deaths before testing or those that die at home without ever going in).

In the case of SARS-CoV-2, the infection fatality rate is almost certainly lower than the case fatality rate. So, it can be argued (though I have not seen it publicly) that the fatality rate vs. all cases may accidentally be closer to the correct value if the undetected infections closely balance the fatalities that are still to come.

I don't think this answered your question directly (though, I am not sure what the actual question is), but I hope that it at least explains the complexity (particularly to the nearly stats-illiterate general public) in estimating and explaining the fatality rate.