As part of proactive steps to get ahead of the Wuhan Coronavirus, many state agencies have started conducting random sampling surveillance testing. In addition to giving valuable early warning information about the spread of the virus, this also dramatically reduces the uncertainty surrounding the true extent of the pandemic and its mortality rate. Quality data can stifle a panic and show what can and should be done to combat the virus.
Considered one of the most reliable forms of data collection, surveillance testing means picking out a small number of demographically representative random people to use as an estimate for the whole population.
Lesson in Statistics
During the presidential election cycle of 2016, polling companies made fools of themselves by confidently predicting Hillary Clinton would win. However, despite their blunder, the national exit polls were only off by 3%. Only in ten states were the exit polls wrong by more than 5%. Considering that a typical survey involves asking only around 1000 people, they were still amazingly accurate.
When aggregating many measurements, the random noise tends to cancel out, and only a shared underlying signal will remain. Consider coin flips. A single toss gives you only a 50/50 chance of predicting the outcome. If you flip a coin 1000 times, you can, with 95% confidence, say that you will get between 376 and 624 heads.
The pollsters get good results by randomly sampling the population’s opinions and making sure that they have a demographically representative group. The same method can be used with medical testing to find out how widespread COVID-19 is.
Application to the Coronavirus
Being able to estimate the number of infected people to within 10% of the number of real cases would be highly instructive in how to act. If the confirmed cases turn out to be close to the real number of infected people as determined by surveillance testing, we then know that the virus has a relatively high mortality rate of more than 3% but with moderate contagiousness. If, on the other hand, it turns out that the real number is 50 times higher than the confirmed cases, the virus is exceptionally viral but harmless to most people.
If the latter scenario is accurate, it means that the virus in practice is unstoppable but not much more dangerous than the regular flu. Random surveillance testing should, therefore, start as soon as possible to establish the mortality rate and virality of the Coronavirus.
Some Case Studies
In the Netherlands, the authorities randomly tested 1097 hospital staff in Brabant between March 6 and 9. They found that 3.9% of them were infected with the Coronavirus – a figure 50 times higher than the confirmed infection rate throughout the Dutch population as a whole. It could either mean that hospitals are infection hotspots, or that there are far more people with the virus than detected – or both. More random sampling in the general population is needed.
The Diamond Princess cruise ship is still the best study to date because all people on board were tested. In addition to teaching us that the death rate among the elderly is around 1%, we also learned that nearly 60% of the infected people were asymptomatic. Today, governments around the world are mostly testing only people showing symptoms of flu, and the Diamond Princess tells us that these tests underestimate the real numbers by at least a factor of 2.4.
Due to the high rate of asymptomatic cases, Dr. Marty Makary at Johns Hopkins University indicated that more than 500,000 Americans could already be infected: “There are probably 25 to 50 people who have the virus for every one person who is confirmed.”
Good and Bad News
If surveillance testing reveals Dr. Makary’s assessment to be accurate, it indicates the virus is already about to become endemic in the population and cannot be stopped. The good news, though, is that this means the virus is far less dangerous than feared. Still, those at risk should self-isolate as much as possible until the pandemic has receded, especially those older than 60.
Read more from Onar Åm.