People have been linking me to this ecological analysis of ivermectin for COVID-19 from Peru Long story short - I don't see much meaning to this preprint at all, certainly not that ivermectin is effective 1/npic.twitter.com/m9MDJc97vt
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
3/n They authors do this by dividing states into 3 categories - maximal, medium, and low ivermectin use. States with maximal use saw the greatest reductions in excess mortality from their peak and thus ivermectin is beneficial for COVID-19pic.twitter.com/Selb33oHvA
4/n So, first things first - how do we know how many people took ivermectin in each area to divide them into those 3 categories? Well, there was a government program in Peru in late July/August 2020 that involved testing citizens in regional areas for COVID-19 using rapid testspic.twitter.com/nwuodD0K36
5/n People who tested positive were given care packages that included food (so that they could isolate more effectively) as well as medicines. In some cases, the medicines included ivermectinpic.twitter.com/kstDh3MZm2
6/n But there's an immediate issue here. Peru authorized and promoted the use of ivermectin in early May (as the authors acknowledge), and the primary use was through over-the-counter prescriptions
7/n Moreover, the number of packages distributed through this program appears relatively small - looking at the references, they imply numbers in the 1,000s, maybe 10,000s. We also don't know how many packages contained ivermectin at allpic.twitter.com/kVV01jg9IV
8/n On the other hand, we KNOW that ivermectin was very widely used across all states of Peru prior to July 2020. The government had already distributed 100,000s of doses at that point, while wealthier citizens had purchased doses for themselves as wellpic.twitter.com/neC8My8mpT
9/n This means that the exposure in this paper makes no sense. By all indications, the use of ivermectin in Peru was widespread in EVERY region, not just the 'maximal' one
10/n To conduct any sort of analysis here, we'd need a good estimate of HOW MANY PEOPLE WERE GIVEN IVERMECTION by area and week, and HOW MANY PEOPLE ACTUALLY TOOK IVERMECTIN We have neither of these
11/n Moving on, the outcome measure here is called "excess" deaths Here's how the authors calculated excess deaths. This is, uh, a very idiosyncratic way to make the calculationpic.twitter.com/ASdi03zQxA
12/n See, the problem is that when you calculate excess deaths you make an estimate of the TREND of mortality, and then look at how many deaths you see above this trend (adjusting seasonally) (H/T @ArielKarlinsky)pic.twitter.com/6MEcEj3ALg
13/n Using just two months of mortality data from the Peruvian summer to calculate an excess just doesn't make sense. A minor fluctuation in mortality could totally throw off your estimates
14/n On top of all of that, we've got potential confounders. The authors attempted to correct for some of these but I don't think they've succeeded unfortunately
15/n For example, age. Older people die more from COVID-19, so places with more old people might see higher death rates that have nothing to do with ivermectin
16/n To examine this influence, the authors conduct an analysis where they just look at people age>60, but this is a really crude dichotomization. The risk of COVID-19 death increases EXPONENTIALLY by age, so even above 60 there's a huge variancepic.twitter.com/dOWzzx2wPP
17/n For example, a 65yo is about 10x less likely to die than an 85yo. A population where the median age for >60yos is 65 might have a VERY different death rate than if the median is 75, depending on who gets infected
18/n The authors also produced these graphs, and argued that since they were relatively similar that government restrictions against COVID-19 didn't explain the difference in mortality that they sawpic.twitter.com/2ojQMWHrxQ
19/n As far as I can see they didn't actually TEST this possibility - they just eyeballed the graphs, and said they were similar enough But are they?
20/n For example, the graph from Puno is plotted on a different axis to the others, which means that the declines in movement (the coloured lines) are all at least ~10% different to the other placespic.twitter.com/GhGMIt4omb
21/n Also, the "excess" mortality is wildly different between regions, peaking at different times in different places. That's not that similar imo!
22/n The other fascinating thing about these graphs is that they appear to completely undercut the entire argument in the study You see, the program that they've used to delineate exposure started at the end of Julypic.twitter.com/NLqIR8srzS
23/n As far as I can tell from the study and news reports, it consisted of first identifying high-risk people for a week or two, then going house-by-house to test them and deliver care packages over the next few weeks/months
24/n This means that the EARLIEST that the program could've been delivering ivermectin is around the second week of August But look at the peaks of excess deaths in those graphspic.twitter.com/QjaXUNrslR
25/n It appears that in most cases, the peak of deaths happened in August ~or earlier~, which means that this program wasn't even started until deaths had already peaked in most places. This is a pretty huge issue for the analysis!
26/n Anyway, at a basic level there's absolutely no way to estimate from the data how many people were actually taking ivermectin in any of these places, so this analysis cannot possibly show that ivermectin is effective or ineffective 
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.