Anecdotes

"3% of Dutch blood donors have antibodies to Covid-19" was the headline. I picked it up from a pleasant vlogger by the name of Dr John Campbell. He is another vlogger, along with Dr Chris Martenson (Peak Prosperity), who has been following and interpreting reports about Covid-19 from a very early stage. He is overall very concerned about the pandemic but also does quite neutral interpretations, i.e., he qualifies comments well and raises not just depressing but also optimistic news that Covid-19 might not be as bad.

This particular result can cause many conclusions depending on the angle you're thinking about at any one time. The Netherlands who apparently have been fairly transparent and complete with their numbers has the eighth most deaths per million, not far behind the UK. Herd immunity, if it exists for this, is at the 80% mark for this kind of thing so it indicates that this crisis is a long way from finishing.

But it was interpreted for another key metric. The heart of a lot of the controversy is to know how many people have had Covid-19, whether confirmed or not, because if you only county confirmed cases, then the current fatality rate in that country, ignoring the lag between diagnosis and death, is an unlikely 11.4% (3601 death divided by 31,589 cases after testing 0.9% of the population). But if 3% of the population have already had it, actually deaths should be divided by 514,046 cases instead, and the case fatality rate drops to 0.7%, which would make it really bad, but not as bad as feared from the official numbers.

Should we breathe a sigh of relief? Well, perhaps it's my innate catastrophism but I'd say no. In any snapshot of any complex situation or phenomenon, you can see it one way from one angle that makes it all look black and white, when reality is prone to being grey. Where is the grey here?

1. Blood donors might be a cross section of the population. On the positive side, it would tend to be the healthier portion, too. (You are questioned before donations and rejected if you've had symptoms of an infectious disease or been to an affected country.) So it could be thought that the prevalence in the Netherlands could be higher.
2. If it was known that blood donors were to be tested for antibodies, it may have biased the group though for those who thought they may have been exposed.
3. There is some interesting fine print: "Preliminary results show that the presence of antibodies differs per age group. 3.6 percent of young blood donors between 18 and 20 years old (688 individuals) have Covid-19 antibodies. That percentage decreases as donors get older. No antibodies were found among donors between the ages of 71 and 80, though the number of donors in that age group is also much lower - only 10 individuals." This could mean quite a few things, from the older group not travelling as much and not being exposed, to the existed of a repository of vulnerable but safe unimmune groups, perhaps through the measures put in place.

4. Maybe more significant though is the accuracy of tests. Any test has two metrics, sensitivity (the ability to signal the existence of the antibody) and specificity (only signalling the exist of one antibody and not others, or other substances). It sounds good to have a test that is, say, 99% accurate, but if it comes up with false positives 1% of the time and false negatives 1% of the time, then it can lead to wrong conclusions at different levels of prevalence:
Note how at a low actual prevalence, 0.1%, the test might give a 1.1% prevalence. If this were the parameters of the 3% test, then it is possible that the actual prevalence could be more like 2% and then the case fatality rate becomes 1%.

For amusement, here is the case of a 98% reliable test, where a 3% test prevalence actually corresponds to a 1% actual prevalence. (And the corresponding case fatality rate would be 2%...):

It is unlikely that the sensitivity and specificity are the same percentage, and ideally for the purposes of a non-diagnostic test, specificity should be as high as possible at the cost of some sensitivity. (And a diagnostic test, for the purposes of isolation, not treatment, should be high in sensitivity at the cost of specificity.)

And you will find even 98% is not a realistic number: The government’s “game changer” antibody tests are inaccurate. What next?

So what does the Netherlands news mean? Well, it is just a data point. Neither a black or white, but a shade of grey that you need to accept till other anecdotes and cases bring more definition in. People who rely on a single case, article or anecdote and interpret it for their purpose, perspective or politics.

Another similar case, also picked up by the vloggers was: Pregnant women without symptoms are testing positive for coronavirus, study says. This article described some research described the situation for women who came to deliver their babies at one medical centre in New York, 4 of whom were symptomatic for Covid-19 which was confirmed by test. But 23 of the other 210 pregnant women were also tested as positive by the PCR test for Covid-19. The hopeful conclusion reached by some from this was that the vast majority of the infected were asymptomatic (or at worst, pre-symptomatic).

The latter point is a cry for optimism. If over 80% of the infected are asymptomatic or have very mild symptoms that would not cause them to be voluntarily tested, then the actual case counts in any country that tests only symptomatic cases would be underestimating by 80% the actual rate that the population has been infected.

There is well-documented asymptomatic cases and asymptomatic spread of this virus, which is not in doubt. The numbers seem quite powerful again in this case, but it is once again prone to the wrong extrapolations. But the virus is more dangerous, and presumably symptom-causing, as you get older, and the more male you are. Pregnant women tend to be younger, and some communities in New York might have even younger average ages of pregnancy too. They also tend not to be male. In this case it is not the cross section you'd use to base conclusions on the whole population. The other conclusion that should have been made from this article was that that, if they were representation of the current infection rate for the population of New York City at the time of the research, late March/early April, over 10% of the city population was positive with the virus and shedding virus particles. Disturbing.

In the so-called Plan B camp (the people who advocate letting the virus do its thing in the population), there is often the desire to find the true infection number, which they surmise is significantly higher than the confirmed case number. This is a way to minimise the case fatality rate and to demonstrate the scale of the overreaction from lockdowns. Anyone who is biased the other way will likely indicate that the confirmed death number isn't accurate either, as deaths at home, care homes and the pre-testing covid-related deaths are also not being taken into account.

People used to raise the Diamond Princess cruise ship as the perfect case study because everyone on board was tested. There were 712 cases and 14 deaths (almost 2% CFR), but again the meaning of this is qualified because the clientele of cruise ships are older and all the fatalities were over 60. (And even at this age, over half of cases were asymptomatic.) And even though this case seems an age ago, there are still many passengers still in hospital. It is possible that the CFR may grow pass 2.5%. There is also a possibility that though the quarantining at Yokohama might have reduced the impact. There was no further exposure to the previously exposed and many of the early symptomatic cases may have been discovered early.

Then South Korea was said to be a good case study as they were the reigning testing champions. Early on the Plan B people were optimistically crowing that there was a 0.6% CFR for a time forgetting the lag in deaths. Now, based on official numbers only from PCR testing, their PCR has slowly grown to almost 2.2%. But again the explosive outbreak that happened may have meant many people got it without knowing it.

Taiwan is also another case study who has managed to be scrupulous with screening and testing to the point that they were barely affected and are harder to judge because of their lower numbers, just six deaths out of 398 cases (1.5% CFR). But an undue number of those are young travellers and tourists. It has never really broken out in the population to get a better cross-section.

Each situation that you could singularly quote has its own parameters, history, idiosyncrasies and quirks. Using any single one is not enough to mean anything for the whole. They need to be all taken in to make any definitive calls.