Thread by @davidwdowdy

The golden rule of epidemiology:
Association does not equal causation.

New variants are associated with epidemic waves, but what is the evidence they *cause* them?

An application of classical causal thinking to the question:
"Did the alpha variant cause a COVID wave in the US?"

An introduction to the Helix data (tabsoft.co/3h6fTUd):

Blue bars show % of samples that are the alpha variant, from Jan 2021 to present. Numbers in each state are # of specimens.

Caveat: Data are not representative. But have been used to argue that alpha caused a surge.

Alpha reached 50% of specimens ~Mar 20. Four weeks later, the US saw 30% rise in cases.

Current dogma: This surge was caused by alpha variant.

But this is only an association. Let's look at evidence for causality, taking the 5 states w most data (not cherry-picked).

A few of the classical Bradford Hill criteria for causality (bit.ly/3qKaCFd), again w caveat that these are not perfect:

- strength of assn
- consistency
- temporality
- dose-response
- plausibility

Plausibility is undeniable. But plausible doesn't mean causal.

Let's look first at Florida. Data here look very similar to the US as a whole.

Worth noting that - like the US - alpha never achieved more than ~70% prevalence. Also that (albeit small sample size) alpha was already highly prevalent back in Jan. But there was no surge in Feb.

Next, let's look at Michigan. Surge here was a 6-fold increase in cases.

Strong association? Check.

Alpha also became highly and rapidly predominant in MI. So looking good for dose-response as well.

But the discrepancy in size of the surge speaks to other factors at play.

Next, consider Pennsylvania. A large surge was seen in April (larger than in FL) - but alpha didn't reach 50% in PA until the same day as that surge reached its peak.

Not great for temporality - would have expected at least a slightly later peak if alpha wave was later.

Things get interesting in California. Here, alpha took over in a similar trajectory as in PA - but there was zero April surge.

Not great for consistency or strength of assn.

(It's also convenient that all 4 states all had similar rates of vaccination.)

But Georgia is the most interesting case. Very rapid and complete alpha takeover in early March - but again, zero surge in April.

Vaccination in GA? Lowest of the 5.

This looks bad for dose-response, consistency, strength of association (and temporality).

In other words, there is a clear association between the alpha variant and surge in cases on the national level.

But applying causal thinking to data at the state level does not provide a convincing story that the variant *caused* the surge.

Specifically:
- Strength of association & consistency: Despite alpha takeover in CA/GA, no surge seen.
- Temporality: Later alpha takeover didn't translate to later surge (e.g., in PA).
- Dose-response: No clear correlation b/w depth of alpha takeover & size of surge (e.g., GA).

To summarize, evidence that the alpha variant caused a surge in cases in the US is not convincing at the state level.

We should carefully apply similar thinking to the delta variant. Just because a variant<->epidemic wave association is plausible, does not make it causal.