Intended for healthcare professionals

Rapid response to:

Feature Global Health

Covid-19: Why Germany’s case fatality rate seems so low

BMJ 2020; 369 doi: https://doi.org/10.1136/bmj.m1395 (Published 07 April 2020) Cite this as: BMJ 2020;369:m1395

Read our latest coverage of the coronavirus outbreak

Rapid Response:

Re: Covid-19: Why Germany’s case fatality rate seems so low

Dear Editor

A rate comprises two numbers: a "numerator" - number of incidents; divided by a background number, the denominator. For them to be meaningful, you have to understand what both numbers mean.

When it comes to Covid-19 fatality rates, both numerator and denominator are far more slippery than you might imagine.

Take the deaths. The UK government has been reporting deaths that occur in hospital, in people who have tested positive for Covid-19. Died somewhere else, not in hospital? Not counted. Died in hospital but not tested, or a false negative? Not counted. Died in hospital of something else, but had had a positive Covid-19 test (false positive or incidental finding) - not really a Covid-19 death, but counted as one, anyway.

We've also started to receive data on death certification. But anybody who works in epidemiology knows how unreliable death certificates are - especially if (as has been reported) there is pressure on doctors, from some coroners not to include Covid-19 without a test. When health protections teams deal with an outbreak of cases in a care home, for example, they want a few tests done, to confirm that the virus is circulating in the home. But once that's been confirmed, there really isn't any point in doing any further testing - if other residents have symptoms suggestive of Covid-19, you should assume that's what they have. With a sensitivity rate of only 70-80%, who would believe a negative test result? The negative predictive value isn't good enough to be confident. But if the coroner insists that if they didn't test positive you shouldn't put Covid-19 on the medical certificate of the cause of death (MCCD), it's easier just to put "bronchopneumonia" or similar... So those cases won't be counted either.

All of which means that the number of Covid-19 deaths is far from clear. It doesn't mean that the data are useless - but you have to understand their limitations.

So if getting the numerator - the number of Covid-19 deaths - clear isn't as straightforward as you'd think, how about the denominator?

The denominator used is, often, the number of cases. But how do you know how many cases you've got? In the UK, cases are people who are admitted to hospital, who test positive for Covid-19. We've already established that the test isn't perfect. The RT-PCR tests are highly specific for SAR-CoV-2 RNA (although they don't distinguish between viable, potentially infectous virus, and "dead", non-viable fragments of RNA). But many patients don't have virus particles present on the swabs done - and getting a good nasopharyngeal swab is not straightforward. So the sensitivity rate is estimated at 70-80% at best.[1] And remember - in studies in China and elsewhere, a considerable number of people had very minor illness or were asymptomatic. One of the first UK cases demonstrates this - he was unwell, but not so unwell as to curtail his normal activities, including a few days skiing. Which is partly how he came to infect so many other people.[2]

So if you only count people who are sick enough to be admitted to hospital, and who get offered the test, you will miss a lot of people. They will have had the infection, perhaps passed it on to others, but they won't count as cases. They'll also be a lot sicker - ill enough to be admitted - and therefore most likely to die. Which explains why early estimates of the case fatality rate were so high.

So, when you compare the UK's and Germany's "case fatality rate" - are you comparing like with like?

Just imagine, for a moment, that the same percentage of the population in Germany and England is infected; and the same proportion die; and that both systems identify deaths in a similar way, so their numerators will be similar. But if Germany tests twice as many people the denominator will be twice as large, and the rate will be half the UK rate. But what matters is that the same proportion of the population got infected and the same proportion of those infected died.

What you really want to know is not the case fatality rate, where caseness is such a slippery concept, but the infection fatality rate - which can only ever be estimated.

If you want to compare populations you need to compare true numbers. You could, for example, compare the number of deaths with the number of people in the population - the Covid-19 fatality rate (not the case-fatality rate). You would also want to take differences in demographics into account. Death rates increase significantly at older ages, for example; so if you were to compare, say, Italy (with a relatively elderly population), and Syria (with a very young population), you'd expect a much higher death rate in Italy, simply because of the demographics. It wouldn't indicate that the disease, per se, were more serious in Italy; or that medical treatment in Syria is better. (These things could be true, of course; but it could just be the demographics affecting the fatality rates.) You have to use methods such as age standardisation before such comparisons are meaningful.

What matters is that you know what's going on in the population, and you can take the appropriate steps to control it. For now, the main question is whether the so-called "social distancing" measures and shielding of the vulnerable are working. If they are, the effective reproduction number (Re) will drop. Once it is below one - so that each case on average infects fewer than one other person - the incidence will fall; and then the case numbers will fall... Death rates are a clumsy proxy for incidence, because deaths generally occur weeks after infection, so there will be a considerable lag.

Peter English.

References:

1 Zhao J, Yuan Q, Wang H, Liu W, Liao X, Su Y, et al. Antibody responses to SARS-CoV-2 in patients of novel coronavirus disease 2019. Clin Infect Dis 2020 PMID: 32221519, DOI: 10.1093/cid/ciaa344 (https://academic.oup.com/cid/article/doi/10.1093/cid/ciaa344/5812996).
2 Ball T, Ellery B, Wace C. Coronavirus ‘super-spreader’ named as Brighton scout leader Steve Walsh. The Times, 2020; Updated 12 Feb 2020; Accessed: 2020 (17 Apr): (https://www.thetimes.co.uk/article/coronavirus-super-spreader-named-as-b...).

Competing interests: No competing interests

17 April 2020
Peter M B English
Public Health Physician
N/A - this is a personal response
Surrey, UK