Thread by @WesPegden

A quick note on this thread by @joel_c_miller before bedtime.

Joel is summarizing a nice argument coming from random graphs to suggest that increasing transmission rates among low-risk groups cannot be good, unless accompanied by other decreases. 🧵 1/9

This argument is valid if comparing two scenarios with constant transmissions. It is not valid if we expect (as I think we do!) that transmission patterns will eventually increase.

In particular, it is worth noting, that... 2/9

even in the simplest single-population models with time-varying transmission rates, epidemic sizes (and thus mortality) can be decreased by increases in transmission.[🤯]

Time dynamics make coupled systems complex, and intuitive reasoning about the effects of changes is tricky.

In particular, Joel suggests that increasing low-risk transmission cannot help if we do not correspondingly manage to decrease transmission to or within high-risk groups. But in our paper with @ChikinaLab, counterexamples to this principle can be found.

For example, 4/9

the only change between transmission patterns in our Figures 2B & 4B are that in the latter, transmission levels are pointwise higher for <40 year olds than in the former. Mortality drops by more than 70%.

Why isn't this precluded by Joel's argument? 5/9
journals.plos.org/plosone/articl…

Joel suggests that to improve things, these strategies would have to enact corresponding decreases among high-risk groups. But we don't do that: these transmission levels are the same in these two figures in our paper.

On the other hand, because transmission is... 6/9

... not constant (in particular, we assume that eventually, it will increase), changing *when* low-risk people become infected can significantly reduce transmission to older people.

This is why, in this tweet, it is not true that... 7/9

"the number of H individuals per L infection are unchanged." (Also, because of the 🤯 above, the last line can also be false.)

The analysis we present in our paper is relatively simple and based on mainstream formulations of simple epidemic models.

Some references for 🤯:
8/9

We gave some simple counterexamples to monotonocity principles (for homogeneous models!) in our manuscript here:



Nonmonotonicity was previously addressed by Bacaer and Gomes:

link.springer.com/article/10.100…

These both show the importance of time dynamics.

I did not notice before posting that 21 tweets in, Joel acknowledges this caveat.

I hope it is clear this is the core disagreement from those who ignore age-targeting:

10/9

Are we sure we can delay any transmission increases until a new game-changing development?

I do not think most decision-makers realize this is the assumption required to believe their policies are not dangerous.

And we have already seen this assumption fail. 11/9

Let me point out that Joel thinks I am being unfair here and that actually he clearly only intended to talk about the case where transmission rates are constant.

Let's all take part in the contest I propose in the tweet after this one: 12/9