This is a really careful, interesting, and thoughtful analysis of non-pharmaceutical interventions for COVID-19 and what works (i.e. masks, distancing etc) to control spread

A quick review thread 1/n https://www.medrxiv.org/content/10.1101/2020.05.28.20116129v3

2/n The paper does something relatively simple - marks when various policies were enacted in various places, and tracks deaths/cases of COVID-19 from that point

2.5/n This is the list of places

Mostly European, with a few other locations around the world

3/n The researchers then created two statistical models. The first inferred the reproductive number (Reff) at any point, using the cases/deaths data

The second associated Reff with the measures put in place to get an idea of what works

4/n (FYI - my Bayesian stats are a bit rusty, so please correct me if I'm misunderstanding, but that seems like the simplest explanation of what was done here)

5/n Quick epi reminder - the Reff of a disease is the number of people that each infected person will go on to infect, on average, at any moment in time

This is important, because it's what the results were presented in

6/n So what did the paper find?

Here's the main results. The further to the right an intervention is, the better it was at reducing Reff (and thus reducing the spread of COVID-19

7/n To further explain, what this shows is the expected marginal benefit of each intervention

So, for example, say you've already cancelled all gatherings over 100. Cancelling gatherings over 10 people will have an additional 36% reduction in the number of new cases

8/n Two very interesting points:

1. Mask mandates for public spaces had negligible impact on the spread of COVID-19

2. Stay at home orders similarly didn't have much additional benefit

9/n On the other hand, as the authors note, closing schools and universities appears to have had a disproportionate impact on the spread of COVID-19

10/n The authors then estimate the interventions required to bring Reff down to 1:

- close schools
- close universities
- close some businesses
- limit gatherings to 10 people

11/n Some strengths of this trial are that they had multiple people rating the interventions and timing, the sample size was reasonably large, and they used a relatively objective outcome (Reff) to track efficacy

12/n Limitations are a bit more numerous, but two in particular I wanted to highlight:

- Hard to dissociate some interventions
- Some interventions not included

13/n In particular, they found that while mask regulations appeared to have minimal benefit, it was hard to dissociate mask mandates from other interventions, so this might not be indicative of a lack of efficacy

14/n Indeed, many places only implemented these mandates after other regulations, so it could be that the MARGINAL (i.e. additional) benefit of masks on top of other social distancing regulations was small, but that by themselves the benefits could be larger

15/n Now, as ever it is hard to infer causal conclusions from studies like this (correlation=/=causation)

BUT

This is a careful, well-thought-out attempt to define the benefits associated with each intervention

16/n For example, I'd say a reasonable conclusion is that the marginal benefit of stay-at-home orders on top of other interventions is probably pretty small

Conversely, the benefit associated with closing universities is probably pretty big

17/n I should also note that I am not an expert in Bayesian statistical methods, so I might've missed something important in terms of limitations of the models used

18/n Another worthwhile point is that I think that this paper is pretty good, but as with everything I could be wrong

Point out any errors I've missed!