Thread by @HcqInvestigator, 1/n Most of my optimism re HCQ efficacy rests on observational ecological [...]

1/n Most of my optimism re HCQ efficacy rests on observational ecological data (eg drastic changes in mortality when countries introduce HCQ)

I'm 50/50 this data's legit. So I think we should recommend early HCQ use

best-case: game-changer
worst-case: placebo+mild side effects

This ecological data is very noisy and confounded. But if the noise and confounders are mostly independent, it's more parsimonious to conclude there's signal in the noise than to think the noise just happens to look like signal (cf Occam's razor: https://en.wikipedia.org/wiki/Occam%27s_razor)

Noisy data can still lead to strong conclusions, even w/o statistical tests

(What % of important decisions in your everyday life do you base off clean data that pass statistical tests? lol)

Potentially fatal confounder: this data is cherry-picked by HCQ supporters

Enough ecological datapoints showing LACK of efficacy would lead me to back down on HCQ

I'd still think doctors should face no stigma/barriers to prescribing HCQ. Not sure I'd stilI recommend it broadly

Wouldn't be surprised if my ecological datapoints turned out to be heavily cherry-picked, or 50%+ noise. Twitter, pls help debunk me

Note: even if 60% is noise, the other 40% could still imply strong conclusions, so long as the cherry-picking isn't too severe, which I'm 50/50 on

Let's dive in

Time series data first (what happens when HCQ is introduced / removed?)

Then CFR performance by HCQ usage

Time series data highly confounded by other stuff going on at the time of HCQ policy changes, eg changes in testing, reporting, public health measures

https://twitter.com/gummibear737/status/1283840213123575809

@gummibear737 has interesting time series graphs for Algeria, Morocco, Brazil, ~12 days after broad HCQ use -- about when we should expect to see results

median time to ICU is ~12 days: https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-guidance-management-patients.html#:~:text=Among%20patients%20who%20developed%20severe,admission%20from%20the%20onset%20of

https://twitter.com/gummibear737/status/1283840213123575809

Obvious effects on mortality in all three cases.

Replications, including sources re start dates:
* Brazil https://docs.google.com/spreadsheets/d/1YoBCgheRwNzgh-ZLpFmGt-xvpo8_p4Z-kdmjO-Q1eIM/edit?usp=sharin
* Algeria https://docs.google.com/spreadsheets/d/1W9_r9kVnuJ2yPzxvpkDuXLFzi25YRU-NvNtRBx8q1jA/edit?usp=sharing
* Morocco https://docs.google.com/spreadsheets/d/1LXmtshvb0kjSr20NzPFna3fh_CKu9QrZnwBiwC9Rlbw/edit?usp=sharing

Morocco corona

graphs HCQ usage dates https://www.palmerfoundation.com.au/moroccos-ministry-of-health-has-adopted-chloroquine-and-hydroxychloroquine-to-treat-covi...

https://docs.google.com/spreadsheets/d/1YoBCgheRwNzgh-ZLpFmGt-xvpo8_p4Z-kdmjO-Q1eIM/edit?usp=sharin

Zooming in on Pará, Brazil, a region with lots of access to HCQ. Excerpt and figure from Harvey Risch, showing large effects in mortality shortly after Brazil recommended widespread usage (~May 20)

https://academic.oup.com/aje/advance-article/doi/10.1093/aje/kwaa152/5873640

Note: the Brazil data isn't fully trustworthy

https://www.reddit.com/r/slatestarcodex/comments/hzoxh1/learned_epistemic_helplessness_covid19_and_hcq/fzlt2fd/.

Would like some help from in-the-know Brazilians to interpret this mortality data

Switzerland

Massive mortality spikes ~12 days after HCQ got banned, that disappeared ~12 days after HCQ got reinstated: http://www.francesoir.fr/societe-sante/covid-19-hydroxychloroquine-works-irrefutable-proof

Replication: https://docs.google.com/spreadsheets/d/1doX3KQu60splewIGiBoWkztEox32-oP_pjXTEN7JdGY/edit#gid=1167197222

Underlying data might be noisy. Two major death spikes according to JHU (6/8: 11, 6/15: 15), but only one according to ourworldindata (6/12: 12)

With two death spikes, Switzerland's mortality data seems very anomalous in the timeframe when HCQ's ban should have effects

But with only one death spike, it's more comparable to noise from other countries. (Still gives some signal, just weaker)

Would love to explore base rates of anomalous mortality spikes. Would also love to hear from Swiss doctors what actually happened

Uganda used HCQ since early April: https://www.independent.co.ug/uganda-using-hydroxychloroquine-to-treat-covid-19-patients/?fbclid=IwAR0NvjaWMmSJBiIHKAF3O0PVaQTFRAJQlUSVpm3cxyzf-0qY1-_X7cbvCMA

1,000+ cases, zero fatalities through mid-July, when the WHO asked it to stop: https://www.independent.co.ug/uganda-halts-use-of-hdyroxychloroquine-for-covid-19-treatment/?fbclid=IwAR0-RhFgl9byIpWv3BU7K3gBhqIGyF4bjWtqljZpECAX5hJcVQnR1_e32X4

Since then: 13 deaths

h/t @EduEngineer

Turkey started HCQ around Mar 21? https://www.klimik.org.tr/2020/03/20/hidroksiklorokinin-covid-19-hastalarinda-ve-hastaligin-profilaksisinde-kullanimi-konusunda-klimik-dernegi-gorusu/?fbclid=IwAR0IT9TAqZRkDnvEZ3HCXlHkLQbIXqaFP4TfVyUx-1gQcB9LaZ4tWpFtFVk

Pneumonia rates among cases dropped 60.3% -> 19.5% https://www.middleeasteye.net/news/coronavirus-turkey-hydroxychloroquine-malaria-treatment-progress?fbclid=IwAR2Et9gM5p6kzATt_n9WCqqhNGfPP5otiQXawrjsePQZPV40QTQtuRRIAY0

Not an artifact of increased testing -- % positive tests increased over this time frame. (Maybe artifact of earlier testing?)

h/t @EduEngineer

National Consumption of Antimalarial Drugs and COVID-19 Deaths Dynamics: An Ecological Study https://www.medrxiv.org/content/10.1101/2020.04.18.20063875v1.full.pdf

Compares explosiveness of early mortalities by CQ/HCQ usage

Many confounders, most notably wealth. Would love to see comparison with next 16 non-HCQ countries

Most compelling time series counterevidence I've seen: Colombia, Chile, Panama, and Belarus all stopped using HCQ after fraudulent Lancent study in late May: https://health.economictimes.indiatimes.com/news/industry/hydroxychloroquine-a-drug-dividing-the-world/76154571 https://www.palmerfoundation.com.au/infectologists-propose-to-resume-use-of-hydroxychloroquine-in-panama/ https://eng.belta.by/society/view/belarus-suspends-use-of-chloroquine-for-covid-19-outside-of-hospitals-130777-2020/

Belarus suspends use of chloroquine for COVID-19 outside of hospitals

The World Health Organization suspended trials on using chloroquine and hydroxychloroquine to prevent and treat COVID-19 as part of the Solidarity international clinical trial.

https://health.economictimes.indiatimes.com/news/industry/hydroxychloroquine-a-drug-dividing-the-world/76154571

If HCQ substantially reduces mortality, CFRs should rise. They only rise in Chile

Data not necessarily trustworthy. Also unclear how much HCQ usage actually changed after ban

Above points also apply for pro-HCQ data. Absent further info, still a hit against HCQ

Now for CFR comparisons by HCQ usage.

Many confounders, like: death underreporting, age pyramids, comorbidities, testing rates, climate, genetics, immune systems (different pathogen environments)

Raw CFR comparisons aren't too meaningful. A 5x difference in CFRs means most variation is due to non-HCQ factors, which makes it hard to ascribe anything to HCQ

Better to compare CFRs among neighboring countries, which controls for a lot of confounders

Central Asia

Some evidence for Uzbekistan, Tajikistan, Kazakhstan, Russia, and Georgia using HCQ early, Armenia using HCQ late, Krygyzstan not using HCQ. CFRs consistent with early HCQ helping

Source: https://docs.google.com/document/d/1IkiESIizLxwtvpDcGJHd4ZqUV50XVFkero9HVN080eQ/edit?usp=sharing

h/t @EduEngineer

Indian subcontinent

Evidence of HCQ use in Afghanistan, Pakistan, India, Burma, Bangladesh, Nepal, Sri Lanka, and Bhutan. All have low CFRs (except Afghanistan)

Source: https://docs.google.com/document/d/1UALtvXA0ja1D7TYCbBYVjwPXf_zIKJNDojyHD_aNGEU/edit

(low CFRs might stem from non-HCQ confounder)

h/t @EduEngineer

Southern Europe

Some evidence for HCQ use in Turkey, Greece, Bulgaria, and Serbia. Albania banned HCQ. No data for Montenegro, Kosovo, or Northern Macedonia. Known HCQ countries seem to do a bit better?

Source: https://docs.google.com/document/d/1_776regdG2AHcIZt8Rj-B2UggzN60GLDh4GGvzNcB0E/edit?usp=sharing

h/t @EduEngineer

Middle East

Some evidence for HCQ use among Bahrain, Qatar, UAE, Oman, and Kuwait. Couldn't find anything on Saudi Arabia

Source: https://docs.google.com/document/d/16J_0_oIDjyYwIJielveW2ILZCaoY9iAIStAe5lvBK_4/edit?usp=sharing

h/t @EduEngineer

Eastern Europe

https://twitter.com/gummibear737/status/1283840220178382851

@gummibear737 has noted that Ukraine and Russia, both using HCQ, have impressively low CFRs

Ony somewhat impressive when compared with neighbors

Friends in Estonia and Latvia tell me HCQ isn't a thing there. Idk about Lithuania/Belarus

https://twitter.com/gummibear737/status/1283840220178382851

Next up: piece 37 - piece 48 from @filipe_rafaeli's https://truthabouthcq.com/hcq-works/

Interesting datapoints of CFR performance by HCQ usage. See summary image

Can't trust it completely though. There's evidence of cherry-picking

eg piece 42 singles out South Dakota, comparing it to New Hampshire

But SD doesn't stand out among its neighbors: https://docs.google.com/spreadsheets/d/1V2W7LpAZBTqOeHuDstSzHaAqBi9QXKwdy2WhidvttV0/edit?usp=sharing

if SD is using HCQ, and not doing better than its non-HCQ neighbors, this is actually a small hit against HCQ

eg Peru appears to use HCQ: https://elcomercio.pe/tecnologia/ciencias/covid-19-por-que-se-sigue-usando-ivermectina-e-hidroxicloroquina-en-el-peru-pese-a-que-no-hay-evidencia-suficiente-noticia/ and http://www.xinhuanet.com/english/2020-06/01/c_139104610.htm

But its deaths per 100k is 3rd worst in the world, and its CFR is worse than Germany's: https://coronavirus.jhu.edu/data/mortality

Not sure why @filipe_rafaeli didn't include it

I asked a Tinder date from Peru what was going on. She's a lawyer, and her best friend's a doctor

She estimates 5-10% of Peru uses HCQ, mostly poor, and that it's frowned upon by doctors. Said her parents probably wouldn't get prescribed HCQ if they tested positive

So maybe HCQ isn't *really* used in Peru

But if all we're going off of is a couple of news articles, then how do we know HCQ is *really* used in all the other "pro-HCQ" countries?

Hard to draw real conclusions without surveying doctors on the ground

piece 43, the natural experiment with New Brunswick, does seem a little interesting. Great CFR relative to rest of Canada. But Newfoundland and Labrador also does great. Can't rule out that whatever's making N&L do well is also going on in New Brunswick

https://docs.google.com/spreadsheets/d/1yGZ1-gerPExxD2qpAI3x5qXDrRbMAhJO_YhPy2zV_JM/edit?usp=sharing

Finally: @CovidAnalysis's http://hcqtrial.com

Very interesting and very suggestive, but I'm hesitant to conclude anything without a clearer understanding of their country selection criteria. Might be cherry-picked

Analyzed in greater depth here: https://twitter.com/HcqInvestigator/status/1295070821061406724

Summary:

• many individually untrustworthy ecological datapoints which contain strong signal for HCQ when considered in aggregate, unless fatally confounded by cherry-picking (which I give 50/50)
• hard to draw strong conclusions without knowing what's happening on the ground

Directions for further exploration

• find more ecological datapoints suggesting that HCQ is (in)effective
• debunk some of my datapoints
• figure out the discrepancy between JHU's and ourworldindata's data re Switzerland's death spike

• quantify how anomalous Switzerland's death spike is, compared with other countries' death spikes
• figure out how to interpret Brazilian mortality data
• replicate the anti-malarial ecological study with next 16 non-HCQ countries
• replicate http://hcqtrial.com

Early treatment with hydroxychloroquine: a country-based analysis

A country-based analysis of HCQ use

http://hcqtrial.com

I'd love to see data indicating how adoption of early HCQ has changed over time in as many countries as possible. This would help settle the questions of which countries actually have widespread HCQ use, and how HCQ use changed in response to policy changes

RT to @Sermo?

I'd also love to see a proper ecological study done by someone with the relevant expertise. I can help find funding

Maybe somebody like Melissa Dell? Or someone she knows? https://marginalrevolution.com/marginalrevolution/2020/06/what-should-i-ask-melissa-dell.html

I'd love to at least hear her weigh in

RT to @tylercowen?

That's all for now!

Tune in next time for why I think the four negative RCTs studying early-use HCQ, often cited as debunking HCQ, actually give evidence for its efficacy

[fin]

You can follow @HcqInvestigator.

Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: