Thread by @phl43

In this post, I explain why people claim that Delta is more than twice as transmissible as the original strain, argue that it doesn't show that and that Delta's transmissibility advantage is probably much smaller. I summarize the main points below 🧵 cspicenter.org/blog/waronscie…

First, I explain what it means when people say that a variant is more transmissible than another, because I think people don't really understand it and you can't interpret the evidence correctly unless you do.

A variant A is more transmissible than a variant B if and only if, *other things being equal*, people infected by A infect more people on average than people infected by B. The "other things being equal" clause is really important here.

Indeed, when other things are *not* equal, people infected by A could infect more people on average than people infected by B even if A is not more transmissible than B or even if A is actually *less* transmissible than A.

For instance, this could happen if A is circulating in networks where the prevalence of immunity is lower than in the networks where B is circulating. In this case, since immunity makes transmission harder, A would spread faster than B even if it's not more transmissible.

For A to be more transmissible than B, it has to be the case that, if the same people doing the same things in the same context had been infected by A rather than B, they would have infected more people on average than they actually did, which isn't the case in that example.

Of course, we can only observe how many people on average people who were infected by A or B infected, not how many people they *would* have infected if they had been infected by a different variant. In other words, we can't observe transmissibility, but only transmission.

I therefore make a distinction between a *transmission* advantage and a *transmissibility* advantage. This distinction is very important, but almost never explicitly made and therefore often overlooked, which as I explain in the post led to bad interpretations of the evidence.

A variant A has a transmission advantage over a variant B if and only if, *in a particular context (time, place, etc.)*, people infected by A infect more people on average than people infected by B.

A transmissibility advantage, on the other hand, just means that a variant is more transmissible than another in the sense I already explained above and isn't context-dependent.

So a variant A can have a *transmission* advantage over a variant B in a particular context even if A doesn't have a *transmissibility* advantage over B or even if B actually has a transmissibility advantage over A, as in my example above.

In practice, estimates of Delta's *transmissibility* advantage are based on estimates of its *transmission* advantage during the initial expansion of the lineage, which can be measured by comparing how fast it's growing relative to previously established strains.

It's actually not easy to measure a variant's transmission advantage and estimates are highly uncertain, but there is no doubt that, *during the initial expansion of the lineage*, people infected by Delta infected more people on average than people infected by other variants.

But it doesn't mean that it's (only) due to intrinsic characteristics of the Delta variant! It could be that it was (partly) due to the specific contexts in which Delta initially spread.

I examine French data beyond the initial expansion of the lineage and find that it's probably the case. As you can see, while Delta's transmission advantage was initially very high, it totally collapsed in recent weeks, as the lineage became dominant.

As I explain the post, this isn't due to a change in the composition of the non-Delta variants in France during that period, for whatever reason Delta's transmission advantage relative to the same non-Delta variants has fallen by a lot.

As I explained a few months ago, the exact same thing had already happened with Alpha, so this seems to be a recurring pattern that needs to be explained, but more on that later. cspicenter.org/blog/waronscie…

If you disaggregate and look at Delta's transmission advantage at the department level, you see the same pattern, but also that it varies wildly across departments.

This suggests that Delta's initial transmission advantage had more to do with the specific contexts in which it initially expanded than intrinsic differences with the previously established strains. As I explain in the post, this isn't limited to France.

So here is the problem in a nutshell. Epidemiologists measure Delta's *transmission* advantage during the initial expansion of the lineage and assume it accurately reflects a *transmissibility* advantage. They conclude that Delta is 40% to 60% more transmissible than Alpha.

Since they had previously concluded, using the same methods, that Alpha was 40% to 60% more transmissible than the original strain, they multiply those transmissibility advantages and conclude that Delta is between 2 and 2.5 times more transmissible than the original strain.

This is why people say that Delta has a reproduction number between 6 and 9. But as we have seen, both Alpha's and Delta's transmission advantages eventually collapsed, so this is almost certainly false!

The literature on Delta's transmissibility advantage is full of caveats showing that epidemiologists understand the dangers of taking the results of observational studies at face value, but in practice they just ignore their own warnings and plug those estimates in their models.

For instance, here is what Neil Ferguson wrote about Delta's transmissibility advantage. (He uses "transmission advantage" in this passage, but as I say in the footnote, he uses this expression interchangeably with "transmissibility advantage" throughout the paper.)

He would have been well-advised to take the distinction between transmission advantage and transmissibility advantage more seriously, because it might have saved him from another embarrassment 🤷‍♂️

He is hardly the only one to make this mistake though. Here is what a group of epidemiologists who make projections for the French government, and are also repeat offenders, did. Policy is partly based on their work 🙃

In the debate about vaccines and treatment, people often correctly point out the dangers of taking at face value the results of observational studies, but when they talk about Delta's transmissibility advantage they show none of that caution. It's exactly the same problem though!

As I explain in the post, if epidemiologists and public health officials continue to use the same methods to estimate the relative transmissibility of different strains, they might eventually conclude that Omega or whatever has a basic reproduction number of 125!

As I said above, the pattern I identified of a variant that has a very high transmission advantage during its initial expansion before eventually collapsing seems recurring, so we need a theory that can explain it.

In the post, I sketch a theory that can do that. The basic idea is that population structure matters and that if the population is structured in a certain way it could explain the pattern I described above.

As I argue more at length in the post, this theory can make sense of the data, whereas the theory that Delta is more than twice as transmissible as the original strain can't. It doesn't mean that Delta is not more transmissible, it probably is, just not *that* more transmissible.

There is a lot more in the post though, and I hope it will at least convince people that Delta's transmissibility advantage is much more uncertain than commonly assumed, so you should read the whole thing. Please share it if you found it interesting!

P. S. When I say there is more in the post, and that you should therefore read the whole thing, I really meant it! For instance, here is some more data that fit my theory rather nicely, but are difficult to reconcile with the standard narrative.