Nate Silver is a well-known statistician and has tried to make clear some relatively technical issues about the coronavirus disease. (Silver Paper) He earlier wrote a long piece on why he hasn’t built a model of cases and deaths. In this new piece he talks specifically about the issues in getting an accurate number or projection of cases, meaning infected people. One of his primary points is that case number comparisons between countries, or even between states in the US, are completely misleading because many different testing strategies and guidelines have been used. The other issues relate to how accurately testing to determine cases works. That of course is largely a sampling and timing issue. He creates several scenarios based on how quickly the virus might be spreading over time and how testing would or wouldn’t accurately determine that rate. Among other problems discussed are how many people seek testing based on their coronavirus infection symptoms, how many people have similar symptoms but from a different disease, the lag time between infection and appearance of symptoms, the lag in how long it takes an infected person to pass the virus to an uninfected one, lag time between taking a sample and getting a result, false negatives and false positives. His assumed false negative rate is surprisingly high, but apparently consistent with the literature. I wonder. Could be that people who test negative at one point but later have symptoms of the disease, just had a much shorter interval between exposure and development of symptoms. The estimated range for that process is quite wide. By the time you read through that list, you understand why figuring out things like a current infection rate or a current serious disease rate is very difficult, much less projecting one. Silver then runs through some examples, making varying assumptions, and he provides a spreadsheet where you too can run whatever scenario you want. One important conclusion is that unless testing growth and infection growth are perfectly matched, you are likely overestimating or underestimating the actual number of infections. A rapid increase in testing, which happened in the US, makes the growth in infection rate look much bigger than it is. And in a country that strictly limits testing to people with symptoms, you are likely underestimating the true rate of infection. The paper is very useful if you are trying to make sense of all the infection and fatality numbers and rates flying around.
Of course, a lot of this could be fixed if we just had a large randomized test of the general population for both infection and antibodies and we retested the negatives for infection regularly on a follow-up schedule. That would give us some background information that would be quite useful. Why hasn’t that happened? I am starting to wonder if there are people who don’t want to see the results of that study for whatever reason.