Explaining the shape of a typical COVID 19 epidemic curve

Here is the shape of the COVID19 epidemic curve in Sweden.

This curve is the typical curve observed in many countries. It is characterized by a rapid increase in the number of deaths, followed by a regular decrease in that number. Or Didier Raoult, one of the world best expert in infectious diseases, said that we don’t really understand the reasons for such a mysterious pattern. Drawing on the work of John Sanford, I would like to propose a hypothesis, namely that the gradual decline in the number of deaths is due to the fact that the virus loses its virulence over time as a result of the accumulation of mutations that eventually damage the ability of the virus to replicate and infect new hosts. Do you have alternative explanations?


Of note, the United States did not follow this pattern. Do you know the reason why @Giltil? This will give you a way to test your hypothesis.


No, I don’t know the reason. But I am wandering whether the same pattern would not be observed at the state level, instead of the one of the whole country.

A key point is that this isn’t what you’d expect if your hypothesis was correct, not be a long shot. It is easy to go find the data and see for yourself what it is showing.


It seems that the curve in the U.S. will take a camel back shape rather than a dromedary back shape. Still a mysterious pattern!

1 Like


I think population genetics can help here. Maybe denser populations allow the deleterious mutations to fix in the population of Covid. The US is very spread out and behavior was not consistent from state to state.

A paper that was authored and posted here by a contributor showed a lot of variation in the Covid sequences.

Yes, population can help here.

No, this has nothing to do with this mechanism.

In this case, we are facing the second bump because we are relaxing social distancing to quickly before we are prepared. It has nothing to do with the genetic deterioration of the virus, which remains quite stable at this point in time.


Not sure the second bump in the US is due to relaxation of social distancing. Here in France, the government complains about such relaxation but despite this, the number of hospitalizations and deaths keep decreasing. Same thing in Sweden, where no lockdown happened, no mask imposed. How do you explain these facts?


Not just the relaxation of social distancing, but doing this before we are ready. France is far more prepared than are we in the US.


Then the only other choice is mutations that have increased the ability for the virus to spread, and possibly increased virulence. Take your pick.


@Giltil @swamidass
You guys are discussing different graphs. One is the death rate, and the other is the no: of cases detected.

Totally different things.

No: of cases detected has very little to do with @Giltil claim.

I suspect the 2nd bump in the US is due to the virus reaching parts of the country (Texas, Florida) that weren’t hit during the 1st bump.

So let me see if I understand correctly. The hypothesis is that the COVID 19 epidemic curve looks the same everywhere because the following happened: in China, the virus spread rapidly until mutations decreased its fitness. Then the virus – with most of those mutations already present – spread to Italy and other places, where again it spread rapidly until mutations decreased its fitness, whence it spread to the Northeastern US, where again it spread rapidly because its fitness was somehow now unaffected by all of those deleterious mutations, until once again mutations reduced its ability to transmit, after which it started spreading rapidly in the rest of the US despite the deleterious mutations because… reasons?



Then the claim makes even less sense. If Sanford were right, deleterious mutations should be decreasing the virus’s fitness, which should be reflected in the number of cases. The death rate should only matter insofar as it is a proxy for the case rate.


Death rate is a proxy for the actual case rate… many scientists are basing the infection fatality rates based on the number of deaths.

However, the number of cases detected is many factors different from the actual number of cases, and dependent on other factors such as testing strategy.
The discussion was happening based on no: of cases detected.

On the whole, I don’t think things have gone the way you say they have, since the epidemic spread very quickly, almost simultaneously and not successively, in a very large number of countries around the world. In fact, the window of attack for the epidemic in many countries was very small, about a month, between the end of February and the end of March.
Anyway, how do you explain the typical shape of the epidemic curve in so many places around the world?

We explain it with epidemiological modeling, such as SEIR models, which treat the fitness of the virus as a fixed quantity. There is not reason, in this case, to infer fitness changes.


China peaked in early February; Italy peaked around the third week of March; NY peaked two weeks later; Florida peaked in July. Yes, many other countries also saw peaks in March, but so what? If deleterious mutations cause the shape and the shape is the same from early February through July, then. . . how could deleterious mutations be causing the shape? If two months worth of deleterious mutations were enough to stop spread in China, why was the outbreak accelerating in NY when the virus had twice as many deleterious mutations? And why did it accelerate last month in Florida when the virus had eight times as many deleterious mutations?

It’s infecting the same host everywhere, and that host reacts similarly when a lot of people start getting sick.


A word of caution about this sort of time series data - it very easy to think we have spotted trends, but it may be difficult to determine the cause when there are multiple factors in play. I advise against looking at a curve and speculating about unknown causes in the past. It’s far better to start with known factors and see how the curve changes moving forward.

Mostly unrelated, I time series data set I worked on years ago: