One can over-estimate seroprevalence (+ thus under-estimate IFR) by measuring seroprevalence in a sample that does not represent the general population, and then extrapolating that sample to the general population. Ioannidis does this.

Unlikely policies caused more excess deaths; non-COVID-19 deaths dropped. https://sciencedirect.com/science/article/pii/S0091743520303625… https://academic.oup.com/aje/advance-article/doi/10.1093/aje/kwab062/6169297… https://medrxiv.org/content/10.1101/2020.08.28.20183699v3… https://bloomberg.com/opinion/articles/2020-09-17/child-mortality-covid-19-lockdowns-may-have-saved-kids-lives… https://twitter.com/jburnmurdoch/status/1354158357754601472… https://twitter.com/tylerblack32/status/1367239480130740224… https://twitter.com/GidMK/status/1371045429232631810… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Most infected people increase antibody levels. In the general population that antibody increase persists for ≥6 months in most people, besides with some assays like Abbott. https://twitter.com/AtomsksSanakan/status/1301777937008652294… https://twitter.com/AtomsksSanakan/status/1362918654141202432… https://twitter.com/AtomsksSanakan/status/1356806587273379840… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Seroprevalence studies (serosurveys) measure antibody levels to estimate the number of infected people. Dividing COVID-19 deaths by that number of infected people gives a seroprevalence-based IFR.

And "non-participating invitees" are less likely to be infected, so Ioannidis was wrong. We don't the response rate for his Santa Clara study, since he has no targeted sample. https://twitter.com/AtomsksSanakan/status/1363989598498676742… https://medrxiv.org/content/10.1101/2020.08.24.20181206v1… https://twitter.com/AtomsksSanakan/status/1341296083767599104… https://medrxiv.org/content/10.1101/2020.11.02.20221309v1.full.pdf

Scientists know methods that get representative samples that are more likely to match the general population; they applied them to diseases before COVID-19. Ioannidis discards those methods, + relies on non-representative sampling luckily matching.

So his "[n]o consensus" claim is misleading. There's an evidence-based consensus (outside of Ioannidis) that those samples could *luckily* match, but are not designed to + are thus less likely to. Covered in another thread: https://twitter.com/AtomsksSanakan/status/1341288191249297408… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

The same point applies to seroprevalence studies. Non-representative sampling might *luckily* get results that match the overall population. But representative sampling is *designed* to be more likely to match the population. https://academic.oup.com/cid/advance-article/doi/10.1093/cid/ciaa1868/6041690?login=true

Suppose you want to know what proportion of people in a city like dogs. You could survey people in 1 building. By luck the percentage you get might match the percentage you would get for the city overall. But you didn't design the survey to make that more likely.

And now in his discussion section, Ioannidis turns to the core point. I'll spend a few tweets on this because this is *the* central pillar of his position, and is how he's been misleading millions of people for over a year. https://twitter.com/AtomsksSanakan/status/1375943659779198976… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

#3 is worst because it extrapolates from inaccurate samples, under-estimating IFR. Yet that's what Ioannidis chooses to do + uses Bobrovitz for. #1 makes sense; that's what "Meyerowitz-Katz" (@GidMK) did. But if you must have data for policy or planning, #2 can work.

There are at least three approaches to dealing with areas lacking representative samples: 1) exclude those areas + wait for data 2) use regions with representative samples to extrapolate over 3) include non-representative samples from those areas https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

I'll leave to others (maybe @GidMK?) to discuss the meta-analysis details. But I can say Ioannidis under-estimates seroprevalence-based IFR in southeast Asian countries such as Japan + South Korea. https://twitter.com/AtomsksSanakan/status/1364464684548644869… https://twitter.com/AtomsksSanakan/status/1364466754337071106… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

His adjustment makes no sense since it's already implicit in test adjustments for sensitivity. And IgA assessment isn't required, given IgG. https://thelancet.com/action/showPdf?pii=S0140-6736%2821%2900238-5… (table 2) https://ncbi.nlm.nih.gov/pmc/articles/PMC7882210/… https://bmj.com/content/370/bmj.m3364/rapid-responses… https://twitter.com/AtomsksSanakan/status/891040491214688257… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Some context: Infection fatality rate, or IFR, is the proportion of people infected with the virus SARS-CoV-2 who die of the disease COVID-19. There are many IFR estimates, including some from Ioannidis. https://twitter.com/AtomsksSanakan/status/1343836703996440577… https://institutefordiseasemodeling.github.io/nCoV-public/analyses/first_adjusted_mortality_estimates_and_risk_assessment/2019-nCoV-preliminary_age_and_time_adjusted_mortality_rates_and_pandemic_risk_assessment.html

- the New York sample under-estimated IFR (see 18/J) - low response rate biases seroprevalence up, under-estimating IFR https://twitter.com/AtomsksSanakan/status/1366078699964149763… - the IFR in Italy was likely over-estimated, due to lower sensitivity of the Abbott assay https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Why the following studies were non-representative: - Luxembourg: non-probabilistic selection step https://twitter.com/AtomsksSanakan/status/1341298484708839425… page 6: https://medrxiv.org/content/10.1101/2020.05.11.20092916v1.full.pdf… - New York: sampled shoppers https://ncbi.nlm.nih.gov/pmc/articles/PMC7454696/… https://twitter.com/AtomsksSanakan/status/1341303286272413696… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Taking a break for a bit. The thread so far covers *less than a page* of the distortions + misleading statements in Ioannidis' paper. I hope people understand why many experts in this field no longer invest time in addressing his nonsensical under-estimating of IFR.

- Kenya used non-representative sampling on blood donors https://science.sciencemag.org/content/371/6524/79… https://twitter.com/AtomsksSanakan/status/1341288191249297408… - Due to co-linearity, the nationwide study ICCRT cites supplants Rio Grande do Sul https://nature.com/articles/s41591-020-0992-3… https://twitter.com/GidMK/status/1283232054646173696… https://imperial.ac.uk/media/imperial-college/medicine/mrc-gida/2020-10-29-COVID19-Report-34-supplement.pdf

ICCRT: https://imperial.ac.uk/mrc-global-infectious-disease-analysis/covid-19/report-34-ifr/… https://twitter.com/AtomsksSanakan/status/1343877612914012167… - Ioannidis' IFRs for LA County + Scotland are impossible: https://twitter.com/AtomsksSanakan/status/1369430446271037449… https://twitter.com/AtomsksSanakan/status/1369641571247923203… - Gangelt over-estimated the seroprevalence: https://twitter.com/AtomsksSanakan/status/1329620151571001344… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

It's 0.31% IFR is unreliable anyway since, for example, the studies for Santa Clara, New York (both), + Chelsea used non-representative sampling. Miami-Dade was wrong. https://ncbi.nlm.nih.gov/pmc/articles/PMC7499676/… https://twitter.com/AtomsksSanakan/status/1363989598498676742… https://twitter.com/AtomsksSanakan/status/1341306679812644865

The "IFR = 0.31%" study Ioannidis mentioned is below. @LeaMerone + @GidMK excluded it because "did not allow for an estimate of confidence bounds" https://sciencedirect.com/science/article/pii/S1201971220321809… "to estimate an overall IFR for the United States of 0.863 percent" https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3590771

The "low IFR" Ioannidis references is one he inferred from a Los Angeles County study. That IFR is impossible since it requires more people are infected than actually exist. https://jamanetwork.com/journals/jama/fullarticle/2766367… https://twitter.com/AtomsksSanakan/status/1369641571247923203… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

Ioannidis co-authored 2009 PRISMA guidelines that stated one should competently assess studies for risk of bias. @LeaMerone + @GidMK did that. Ioannidis didn't, letting in non-representative samples https://bmj.com/content/339/bmj.b2700… https://twitter.com/AtomsksSanakan/status/1315313977539334144…

Seroprevalence-based IFR was ~0.76% in @LeaMerone + @GidMK's paper, when they focused on seroprevalence studies with a low risk of bias. Ioannidis conveniently leaves that out. https://sciencedirect.com/science/article/pii/S1201971220321809… https://twitter.com/AtomsksSanakan/status/1286771217274482689… https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

John Ioannidis published an article defending his low estimate of COVID-19's fatality rate. It contains so many distortions that I'll try something I've never done on Twitter for a paper: Go thru distortions page-by-page. This will take awhile. https://onlinelibrary.wiley.com/doi/10.1111/eci.13554

The problem with trying to work out a global IFR – i.e. the total number of people dead/infected for COVID-19 across the world – is that both the death AND infection data is scant in most places in the world

And as for non-PhD authors? I wrote four papers as a PhD student in which no author had an advanced degree. Theor. Pop. Biol., Phil. Trans. Royal Society, PNAS, Genetics. Cited 86, 116, 178, and 214 times respectively. Maybe they're all crap, but not b/c of my degree status.

As to the paper itself? There are obviously more issues – covered here in depth by @AtomsksSanakan – but oddly enough there are also places where Prof Ioannidis and I agree about our paper

But for anyone reading this who is mentoring PhD students, particularly people at Stanford, I would suggest strongly that you check in and assure them that you do indeed find their opinions and perspectives useful

Moreover, I personally find the entire focus of the piece strange. I do not think it is reasonably possible to accurately estimate the GLOBAL IFR (infection fatality rate/ratio) of COVID-19

The condescension and hypocrisy here is mind-boggling.

I will be writing to the European Journal of Clinical Investigation. Given that the immediate past Editor In Chief was one professor John Ioannidis, I’m not sure it will do much good, but at least I will have my say

I could point out that our paper was reviewed by several very senior epis before we submitted it (including one of the most senior epis in Australia), but that they did not feel they contributed enough to add their names – perhaps this would’ve saved me a tongue-lashing

But the point is that we should not have to have Big Fancy Professors on our paper for it to be considered on its own merits. I’m sure we could have twisted our colleagues’ arms, but we did not think that a professor would stoop to our PhDs as a means of attack

This issue is not a new one by a long shot. @hertzpodcast covered the issues that PhD students face several times in great detail – I recommend you listen

Imagine reading this as a PhD student at Stanford. This is a senior faculty member telling these students that no matter what work they do, their opinions will always come second to professors Not what I would hope the scientific discourse to be

I may have the wherewithal to defend myself, and I’ll be writing to the journal, but the implication that PhD students have no place in scientific discourse, that their papers are worthless scientifically will, I think, have far greater ramifications

But imagine, for a second, that I had not been in the news a bit and grown a social media platform. Imagine I was one of 1,000s of faceless PhD students watching a tenured professor at Stanford publicly defame one of their comrades It’s quite chilling

I appreciate the many wonderful people who have come to my defense against these attacks, but in all honesty it’s not me that I’m worried about. For better or worse, I have a large platform, and I’m not in any huge danger from a professor being publicly mean to me

It is also worth noting that while I am still doing my PhD, I have been working in public health for more than half a decade, because often the more sought-after qualification is an MPH not PhD

I make no secret of my junior status (it’s there in my twitter bio and every paper I publish), but to say that my research is flawed because of it is a remarkable piece of gatekeeping and I think really quite harmful

John's defenders have done this in the past, but I'm stunned that he'd stoop to the same. Science doesn't work like that, to say the least. Gideon's degree status is irrelevant and in the entirety of my career I've never seen this issue raised in a scientific paper before.

The article itself is here, and honestly it’s a bit of an odd piece. If I were to commission a review on the small number of SR/MAs on the COVID-19 IFR, I’d probably want it to be written by someone who hadn’t authored one of the 6

For my followers who don’t publish academic research, it’s worth noting that these attacks not only were written by the author, but approved by at least one editor and (usually) 2-3 peers as well

Now, to the personal attacks I must admit, I was quite shocked to read this published in a scientific paper I’m not going to go over them, but please do have a read in the paper itself (appendix 1)

But overall, I think that Prof Ioannidis' review really shows the issues with having people who have staked their reputation on an issue author perspective pieces on the issue. We all tend to think that our own research is the best

I would argue that one of the biggest STRENGTHS of our meta-analysis was the time we spent EXCLUDING biased research, because as has now become fairly obvious these studies often overestimate seroprevalence in a population

There are also parts of this paper that are bizarre. It is, for example, not a strength of meta-research to include MORE studies. Indeed, the phrase “garbage in garbage out” is commonly used to describe analyses that do not attempt to exclude poorly-done studies

The paper which he co-authored is, I suppose, a matter for discussion – perhaps @LeaMerone and I were presumptuous in reading “selection bias is likely...” as an explicit warning against extrapolating to the entire population of LA County

For example, this tabulated estimate includes studies that we reference elsewhere in the review, with 5 of these estimates ~included in our meta-analysis~ It would actually be BAD scientific practice to include these figures twice!

I’m not sure how it is possible to say that something is “overtly biased” when it is transparent and open, but nevertheless there are quite obvious explanations for all of these things (that we give in the paper)

That being said, I disagree with many of these statements. For example, this passage argues that we excluded studies in “overtly biased ways” with these three pieces of research

Now, one thing to note is that these are judgement calls rather than actual scientific critiques. We laid out our methodology quite transparently – saying that this is “implausible” is an opinion, not a fact

In his latest paper about COVID infection fatality rates, John Ioannidis does not address the critiques from @GidMK, but instead engages in the most egregious gatekeeping that I have ever seen in a scientific paper.

Recently, Professor John Ioannidis, most famous for his meta-science and more recently COVID-19 work, published this article in the European Journal of Clinical Investigation It included, among other things, a lengthy personal attack on me Some thoughts

...Participants will be different to the earlier trials. 24% of participants are 65 or older - a dramatic difference, for a critical question. If efficacy is less >65 - & that's a totally open question - it could lower overall efficacy. Complicated...4/8 https://s3.amazonaws.com/ctr-med-7111/D8110C00001/52bec400-80f6-4c1b-8791-0483923d0867/c8070a4e-6a9d-46f9-8c32-cece903592b9/D8110C00001_CSP-v2.pdf

...not designed to test the hypothesis that 8-week or 12-week interval increases efficacy. But as with other large vaccine trials, there'll be data relevant to onset of immunity to chew on. It's 2:1 randomization (twice as many in vax as placebo), final goal 150 events...

..With results for AstraZeneca's large trial of the Oxford vaccine on the horizon , some things I'm keeping in mind. It's a single, adequately powered, standardized, double-blind, fully placebo-controlled trial: the first for this vaccine. It's 2 doses, 4 weeks apart, so...

... Protection against moderate to severe Covid-19 was reduced in SA overall for the J&J vaccine (57%), but data on the variant specifically not released yet...

...& such a small group, you wouldn't expect anyone to get severely ill. Here's the efficacy calculation. No appreciable difference, based on 39 people with symptomatic Covid-19, an efficacy rate of 10.4% with extreme uncertainty: CI range -79 to 55..

...The Oxford press release doesn't include any numbers - just says that it didn't provide protection against mild to moderate Covid-19 caused by B.1.351 ("SA" variant) was "minimal" & the study was too small for them to assess severe disease: with an average age of 31,.

..The SA trial was too small to provide conclusive answers. The press release doesn't say how many people are in the analysis, but the most recent they released analyzed 1,476 of the approx 2,000 participants..

...Oxford issued a press release with some data, & announced they are working on trying to develop a version that can target new variants. South Africa has halted its rollout of the Oxford/AstraZeneca vaccine. https://ox.ac.uk/news/2021-02-07-chadox1-ncov-19-provides-minimal-protection-against-mild-moderate-covid-19-infection…, https://wsj.com/articles/astrazeneca-vaccine-doesnt-protect-against-mild-and-moderate-cases-from-south-africa-strain-11612700385…

..Additional context on the issue of the SA trial: the median age was reportedly 31 - so it's not a group at high risk of severe outcomes. (Note: 2,000 enrolled &c doesn't mean that they are all in the analysis that will be reported.)

Confirmation on the report of reduced efficacy against mild & moderate "SA" variant: AstraZeneca stressing efficacy against severe disease & subgroup data on dosing intervals https://reuters.com/article/uk-health-coronavirus-astrazeneca-respon-idUKKBN2A60S2… HT @sailorrooscout

Context for the question of impact of the "SA" variant for the Oxford/AstraZeneca data: numbers in SA are small. It was a phase 1/2 trial, with results for 1,476 people to the latest data reported (45 of whom had a single shot). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3777268

Unofficial unnamed AstraZeneca insider says they are doing the interim analysis for the US trial of the Oxford vaccine. AstraZeneca spokesperson says 4-6 weeks till data release. https://reuters.com/article/uk-health-coronavirus-astrazeneca-usa/astrazeneca-expects-us-trial-results-in-next-4-6-weeks-research-chief-says-idUSKBN2A522V… One is wrong? Or they'll release only when have FDA minimum follow-up?

could you elaborate? Given that, in general, it is false that 'Absence of evidence that x is, is no evidence of absence of x, how does the specific 'research object' you might have in mind come into play?

This is not true for the reality of the research object which this fallacy, in my eyes, is all about. Absence of evidence that x is, is no evidence of absence of x.

Yeah, I think people often mean “absence of evidence is not *necessarily* evidence of absence” and leave of the “necessarily”. Obviously absence after a thorough search is evidence of absence!

great list, but I think one of the main problems with "absence of evidence fallacy" is its phrasing: "absence of evid. is not the same as evidence of absence" is a true statement, "absence of evidence is not evidence of absence" is literally false @richarddmorey

What about infections? Yesterday 69K cases 2 wks ago 49K cases So nationally, test positivity 5.5% to 7.1% Wait, there's more In 45 states, test positivity has increased past 2 weeks Yup, in 45 states, increases in testing not keeping up with increasing infections

Best example is NY, which tests more than 100K people daily, test + of about 1.3% So now, let's look at where we are as a nation vs 2 weeks ago All data from @COVID19Tracking using 7d moving avg Yesterday, did 960K tests 2 weeks ago, 898K tests So testing is going up

Easiest is to wait 2-4 weeks If underlying infections stable but you test more, hospitalizations and deaths will remain stable But what if you don't want to wait? Look at test positivity As testing increases, becomes harder to find the next case So test positivity drops

President keeps saying we have more cases because we are testing more This is not true But wait, how do we know? Doesn't more testing lead to identifying more cases? Actually, it does So we look at other data to know if its just about testing or underlying infections Thread

All cases occurred 4 to 16 days after vaccination with AstraZeneca COVID-19 vaccine. This pattern "flagged" in the surveillance data. This is what we hope good surveillance does- detect potential concerning events & investigate further.

6 of the cases were women who had a particular form of cerebral venous thrombosis, namely cerebral venous sinus thrombosis. (https://hopkinsmedicine.org/health/conditions-and-diseases/cerebral-venous-sinus-thrombosis…) The 7th case was medically very similar (cerebral hemorrhage + platelet deficiency + thrombosis)

Update: English report now available here: https://pei.de/SharedDocs/Downloads/EN/newsroom-en/hp-news/faq-temporary-suspension-astrazeneca.pdf?__blob=publicationFile&v=5

Since #AstraZeneca was the primary vaccine strategy in Germany and much of Europe, every day we wait is a frustrating one. I do appreciate the transparency of this report. Wish these numbers had been released yesterday though!

Now, we eagerly await more info & the #EMA recommendation. Of course, serious events in close temporal relation to vaccine should be taken seriously & weighed carefully against the strong, demonstrated benefits of the vaccine.

As of Monday (March 15th), 2021, 7 cases of specific form of severe cerebral venous thrombosis associated with platelet deficiency (thrombocytopenia) were identified. The individuals were 20-50 years old and 3 of them subsequently died.

On Monday, the 2 additional cases of cerebral venous thrombosis were reported following #AstraZeneca vaccination. These additional 2 cases put the number of observed cases "well above the expected number" & ultimately led to the decision to suspend, they write.

If the frequency of an event (like the cerebral venous thrombosis, in this case) is near the expected frequency, this speaks for a *random* occurrence in temporal connection with the vaccination.

They describe the process of observed vs. expected analysis in pharmacovigilance (drug safety) and go on to say that on Friday, the threshold for the "red flag" had not yet been reached.

Q7 asks "What changed since Friday?" The report explains that on Friday, the number of incident cerebral venous thrombosis events that occurred within the vaccinated population was still within a range that would be expected in the general (unvaccinated) population.

Thus, the @PEI_Germany advises that people who have received the AstraZeneca COVID-19 vaccine and feel *increasingly* unwell more than four days after vaccination - with severe and persistent headaches or pinpoint bleeding on the skin seek professional medical advice.

The report emphasizes how rare these cerebral venous thrombosis events were (7 out of 1.6 million vaccinations, according to current knowledge). They point out that these individuals continued to feel unwell and experienced increasing headaches 4-16 days after vaccination.

The main report is followed by a series of Q&A. I will highlight the @PEI_Germany answer to Q#6: "What can I do if I have received vaccination with AstraZeneca COVID-19 vaccine?" since many are likely faced with this question today.

At this point, it remains unclear whether there is a causal relationship between the #AstraZeneca vaccine and these cases of cerebral thrombosis.

Now we await the #EMA comments & decision on whether and how these observations (together with other European surveillance data) affect the risk-benefit profile of the AstraZeneca COVID-19 vaccine and the EU approval of the vaccine.

The German Federal Ministry of Health (BMG) followed this recommendation. The Pharmacovigilance Risk Assessment Committee (PRAC) at the European Medicines Agency (EMA) will also review this information this week.

New @PEI_Germany report provides much needed clarity to the #AstraZeneca "pause" in Germany. Not yet available in English. I will try to summarize. /thread https://pei.de/SharedDocs/Downloads/DE/newsroom/meldungen/faq-temporaere-aussetzung-astrazeneca.pdf?__blob=publicationFile&v=2

3. Stein's paradox If your goal is prediction, you may *not* be after unbiased predictor effects in your prediction model http://scholar.google.nl/scholar_url?url=https://www.researchgate.net/profile/Carl_Morris/publication/247647698_Stein%27s_Paradox_in_Statistics/links/53da1fe60cf2631430c7f8ed.pdf&hl=nl&sa=X&ei=m7YXYK3sPOPKsQLKtYWYBw&scisig=AAGBfm299RJEZV-VWCdNETwpmsjPalO4xg&nossl=1&oi=scholarr

2. Ecological fallacy Hard to resist those sweet population level data to make inferences about health effects on the individual level https://web.stanford.edu/class/ed260/freedman549.pdf

1. Absence of evidence fallacy Absence of evidence is not the same as evidence of absence. Wouldn't it be great if not statistically significant would just mean "no effect"?

10. The low birth weight paradox To adjust or not to adjust for birth weight in the analysis of infant mortality?

9. Lindley’s paradox (not Lindsey's...) If you are interested in the Bayesian vs frequentist statistics wars, make sure you study the paradox

Personal top 10 fallacies and paradoxes in statistics 1. Absence of evidence fallacy 2. Ecological fallacy 3. Stein’s paradox 4. Lord’s paradox 5. Simpson’s paradox 6. Berkson’s paradox 7. Prosecutors fallacy 8. Gambler’s fallacy 9. Lindsey’s paradox 10. Low birthweight paradox

Evaluating whether an intervention works when trialled in groups of individuals can pose complex challenges for clinical research. Paper 3 assesses the role and challenges of cluster randomised trials.

Paper 2 discusses newer data-driven approaches that make clinical trial research more efficient, including adaptive trial designs and master protocols—showing the importance of considering different study designs. Read http://hubs.li/H0LlRr80

Paper 1 provides a historical perspective—from the first documented randomised control trial in English (1948) to the current state of clinical trial research and challenges in LMICs.

Quantity > quality? The magnitude of #COVID19 research of questionable methodological quality reveals an urgent need to optimise clinical trial research—but how? A new @LancetGH Series discusses challenges and solutions. Read http://hubs.li/H0LlR2q0

AZ also continues to fail. The 2.6m doses due to be sent to the EU next week have been halved, says the FT. This is half of the third originally promised. In Fr AZ take-up is now so low – 37,000 doses a day after 86,000 in March – that the cut should make little difference.

Weekend stats also continue to be poor – despite Macron’s promise to vax morning, noon and night. Around 300,000 jabs in 3 days over Easter. To hit its 15 May target of 20m 1st shots, France has to jump from 214,000 to 266,000 a day – doable but more weekend jabs would help

There are still problems, however. I was premature last week to say there was NO widespread anti-AstraZeneca feeling in France. Resistance to AZ because of the clot issue meant use fell to 61% this week, with 1.7m doses now in stock.

In next 5 weeks, to hit its 15 May target of 20m, Fr. must give 9.5 m 1st doses. About 3m 2nd doses are due. That’s 12.5m doses in 36 days – 347,000 a day. The av. in the last 7 days was 285,000. With 40 big vaccinodomes in use, it's doable, if Sat/Sun jab-rate improves

Here, for the record, are the basic figures for the Fr roll-out up to last night. First dose: 10,507,294 Total doses: 14,108,557.

The UK has finally admitted that the AZ vax is linked to v v rare, fatal blood clots for younger people. Maybe, some of those who last month accused France and the EU (but mostly France) of political bias against the sainted AZ would now like to apologise. Some hope…

Weekly French vaccination thread. The French roll-out, still described as “stuttering” or “glacial” in UK media (and even some Fr media) continues to boom. Over 500,000 doses (1st/ 2nd) were given yesterday, a record. Fr should exceed its 10m 1st jabs 15 April target by 2m.

The devolved administration identified 991 cases in January and February last year. They also tracked down 29,399 contacts through local community based workers. Note the ratio: they detected 30 contacts for every case, far higher than in the UK.

A new paper from Anhui province (pop 64 million, almost same as UK) in China shows how control was achieved without any severe or prolonged lockdown. https://sciencedirect.com/science/article/pii/S2352771421000148…

In fact we had a clear statement for proper public health control of the epidemic from WHO on Jan 29 2020, and the China Report from WHO on Feb 24 2020. All measures were not controversial and not based on rocket science or modelling.

Journalists, MPs and the public should ask daily why the public measures implemented by countries and provinces (like Anhui) which eliminated Covid from their territories, are not being implemented in the UK

The reliance on vaccines alone as a strategy for control while transmission remains high, and isolation of cases and contacts is an expensive joke, is incredibly risky. The threat of vaccine resistance remains high.

And the decision of our PM not to attend the first five meetings of the main COBR committee to tackle a pandemic we knew about on January 30, a delay of five weeks, was an act of omission beyond comprehension.

There is a lot of nonsense about Zero Covid being an extreme position, only possible in repressive states (er..S Korea, Taiwan, Thailand, Norway, Finland, NZ??) and our UK strategy reflects a more sensible centrist view. So compare the UK with successful countries...

as I just said to @islaut1 if you want to force the logical contradiction you move away entirely from all of the interesting cases of inference from absence in everyday life, including the interesting statistical cases of, for example, null findings - so I think we now agree?

"absence of evidence is literally evidence of absence" would be basically true if it wasn't referring to and understood to mean a specific fallacious argument form, which is also how it is contextually applied in most conversations.

I think that's where we disagree in that "absence of evidence" isn't a statement it's a defined term and most the general public sees at as such and as such can't be used to mean a lack of evidence as you seem to be understanding it. your initial post which said "absence of

and not just misguided (as too simplistic) but part of the problem....

So, whatever we think on details, I think this thread and other replies on "inference from absence" to @MaartenvSmeden post rather illustrate my point: if statisticians, psychologists, Bayesian folk show confusion/disagreement simplistic "fallacy" label probably misguided

sorry, I have a theoretical position on this, having worked on args. from ignorance quite a bit in last 15 years. I think Wikipedia article is poor and outdated. Suggestions: https://researchgate.net/profile/Ulrike_Hahn/publication/286232860_Rational_Argument/links/5694bd9d08aeab58a9a375a2/Rational-Argument.pdf… pps. 282-286

The broad claim may or may not be true. I think there are tradeoffs. This specific opinion piece is bad argumentation and bad writing. Case in point: "My interest in searching for intelligence in outer space is partly because I gave up on finding it on Earth."

do you mean the specific article is bad, or the wider claim/argument? because as someone who does research on collective intelligence, I'd say there is some reason to believe it is true that there can be 'too much' communication in science. see e.g. the work of Kevin Zollman

This piece is bad. The argument is that conferences promote groupthink so you should stay home and read ArXiv instead. As evidence, here is a bunch of ArXiv papers that were mean to a finding that the author likes.

quite...that's why I'm interested both in the fact that Independent SAGE are still publicising his model, and in whether it will actually manage to do anything of interest - the prediction strikes me as interesting (and as 'stronger' than ones in original paper)

Friston's not an epidemiologist or a virologist.

a prediction to keep an eye on..

most notable is the drop in trust among their own citizens, with the U.S., already in the bottom quartile for trust, experiencing an additional 5-point drop since its presidential election in November 2020 and China seeing an 18-point drop since May 2020.

The Covid-19 pandemic has accelerated the erosion of trust around the world: significant drop in trust in the two largest economies: the U.S. (40%) and Chinese (30%) governments are deeply distrusted by respondents from the 26 other markets surveyed.

as this account is focussed on COVID, maybe time to move the discussion elsewhere- happy to discuss further if you want to get in touch by email - u.hahn" "http://bbk.ac.uk

Since i don't even see what we are conversating over at this point in time anymore what is your specific objection.

probably- and I think there are many interesting questions around why he is there and whether he should be there. But to answer those properly, looking at the performance of the model seems important and interesting to me- that is all I am saying

They're still publicizing his model because it's his group.

We (the public) are not great at understanding nuance and moving targets.

I have done a lot of interviews about covid in the past year. And one thing that really stays with me is something @nataliexdean said. That the public is used to hearing from scientists at the *end* of the process. And right now, we are in the middle.

2/2 so I would prefer to treat this as an opportunity for empirical observation and learning. Evaluation should focus on trying to assess actual contribution, not a priori judgments.

I don't think we should be paying attention to dilettantes when there are experts.

US now explicitly heading in same direction as http://E.Asia/Pacific. Call it whatever you want: max suppression, an 'elimination' strategy, a public health 'measles' approach, Zero COVID, no COVID, or simply just stopping the circulation of COVID-19. Influential & bold step.

Biden-Harris Administration gets that it is COVID-19 itself hurting the economy (the virus circulating, not just the restrictions). Stopping COVID-19 is best way to get people's lives & livelihoods back.

I think diff. is that your first response seemed to indicate the evidence was the search itself (contra Richard) so turning an inference from absence of something into a kind of positive evidence ('the search'). Let's call absence of evidence "not E". 1/2

Not quite sure if I read your words as you intended them, but this assumed I see it differently. The data itself points to something, more or less, or does'nt.

Luckily we also have a few ppl who both have stuff to contribute and don't really care about being cool or getting criticized all the time. (Yes, incl Skinner)

A bit cliché but ppl will always find it cooler to point out that a given proposal is not the only one/has shortcomings/is not the Truth itself etc. than making or improving a proposal. I keep being reminded of this every single day, esp on twitter.

I challenge you to find a loopier article on the pandemic than this one, I really do

I kid you not, friends - this is one of the stories on the "Lockdown Sceptics" website today

Pre-COVID, people in UK had around 10 social contacts per day (https://medrxiv.org/content/10.1101/2020.02.16.20023754v2…), so below shows restrictions are still doing a lot of the hard work in keeping transmission down (from: https://cmmid.github.io/topics/covid19/comix-reports.html…).

UK goes into next reopening stage with relatively low case rates, so there are reasons for optimism, as vaccination will gradually pull down transmission further (as well as protecting individuals) - but also caution, as we've seen globally how quickly COVID situation can change.

Watch this short animation to learn more about #COVID19, how it spreads and how to protect yourself against it. #coronavirus

FACT: #COVID19 is NOT airborne. The #coronavirus is mainly transmitted through droplets generated when an infected person coughs, sneezes or speaks. To protect yourself: -keep 1m distance from others -disinfect surfaces frequently -wash/rub your -avoid touching your

Journalists, MPs and the public should ask daily why the public measures implemented by countries and provinces (like Anhui) which eliminated Covid from their territories, are not being implemented in the UK

The reliance on vaccines alone as a strategy for control while transmission remains high, and isolation of cases and contacts is an expensive joke, is incredibly risky. The threat of vaccine resistance remains high.

Fixed version: here's how the Imperial College model of Neil Ferguson performed over 1 year. I used their most conservative R0 assumption, so this is actually generous to them.

yes, that's what I took it to mean. My point was how can you calculate a prediction mismatch for this scenario, given that it didn't happen - ie we can't observe what happened in the cell "government didn't do anything".

"Unmitigated" appears to mean "if the government didn't do anything", compared to lockdowns.

this is utterly bizarre : how would one conceptually even begin to determine a number by which the model *overestimated unmitigated deaths*. What is the comparison unmitigated "prediction" to what actually happened supposed to mean?

One year ago today: A medRxiv preprint of a study in California suggests a seroprevalence of 2.5% to 4.2%, 50 to 85 times more than confirmed cases would suggest, though sample bias and test specificity may affect the results. In a Boston hot spot, a separate study measures 32%.

A few thoughts on the B.1.617 variant, first seen in India in late 2020, recently seen in >100 cases in the UK, and very much in the news here. TLDR: we should watch carefully, but I don't think any of our best lines of evidence on variants are yet cause for concern.

check out this obscure math theorem. only a few know about this.

Canada’s vaccine shortage: Covid-19 cases per capita pass U.S. rate as doctors plead for more vaccines @paulanewtonCNN reports

*This thread was brought to you by mild procrastination

Final points: - If you are paid to analyze data, this is what professionalism looks like. - If the work you do matters, take it seriously. Good science requires meticulous technique, not just genius ideas. - You can't learn all of this at once. But you can learn it, no doubt.

- This is something that is often used by programmers (I am not one of those), but I think the ideas translate very nicely to analysis scripts. https://r-pkgs.org/tests.html

- Or every time your code doesn't work and you eventually figure out why, write a little test to check for the underlying problem first.

- A simple example: You have 100 patients in your study. That means your dataset probably has 100 rows. So insert a bit of code that checks that this is true before running a model with those data.

Step 7: Write tests into your script - This is the step I am trying to learn more about. - It is often true that you computer will happily give you the "wrong" results if you give it the wrong inputs. - So build tests into your code to check that those inputs are correct.

- Save those little functions is a separate scripts that you might use across many different analysis projects. - It's easier than you think!

Step 6: Aspire to never repeat yourself (NRY). - Every new keystroke is an opportunity to make an error. - When you find yourself writing the same block of code over and over, write a little program (function) that does it with a few key strokes.

Step 5: Make your code *sharable* - and maybe even share it! - You've written all that nice code. It would be a shame not to let other people look. They probably won't, but the very "threat" of it can be a powerful motivator to use best practices.

- So when the data changes (and the data *will* change) you just re-run the report. No re-cutting-and-pasting. Fewer chances for mistakes.

- Use "literate programming" tools like R Markdown to generate (and re-regnerate) reports that include your "final" plots, tables, and text. https://rmarkdown.rstudio.com

Step 4: Generate finished outputs, not just numbers. - You need to try to eliminate (or greatly reduce) the amount of cutting and pasting results, plots, tables into other documents, like Word.

- Calculate differences between things (including dates and times) and look at those. - If you transform something, plot it against the original. - Don't just plot though - think about what the data *should* look like.

Step 3: "Look" at everything. - Visualize the variables you use. Plot raw values, summaries, and distributions. - Plot how variables relate to one another. - Plots dates and times to see if they are ordered as you would expect.

...even if that means writing more lines of code (gasp!) - Use a consistent style guide. http://adv-r.had.co.nz/Style.html - Remember, a good script is a love letter to yourself in 6 months time. It needs to be *written*. Clear. Concise. Complete.

Step 2: Once you are nice and comfy with scripting, try to write nicer scripts. - Scripts should be human-readable, not just machine readable. - Use annotations to explain yourself and the code. - Make the machine-readable bits as human-readable as possible...

A progression: Step 1: Script everything. No point and click, ever. This is by far the most important step. I like R and @rstudio. You can do it. https://rstudio.cloud/learn/primers ||

I know this is obvious, but you can't find the errors in your data and analysis if you don't actively look for them.

With all due respect to @NateSilver538, he is not an expert on the psychology of vaccine confidence. He is a poll aggregator and political pundit. He is not an infectious disease specialist, epidemiologist, vaccinologist, virologist, immunologist, or behavioral scientist.

Multiple reports with this overall ranking of neutralization, relative immune evasion Here are a few: http://cell.com/cell/retrieve/pii/S0092867421004281?_returnURL=https%3A%2F%2Flinkinghub.elsevier.com%2Fretrieve%2Fpii%2FS0092867421004281%3Fshowall%3Dtrue… http://cell.com/cell/retrieve/pii/S0092867421002981?_returnURL=https%3A%2F%2Flinkinghub.elsevier.com%2Fretrieve%2Fpii%2FS0092867421002981%3Fshowall%3Dtrue… https://science.sciencemag.org/content/371/6534/1103/tab-article-info

The variants of concern/interest fall into a spectrum of immune evasiveness, w/ B.1.351 being most; B.1.1.7, B.1.429 least. This property pertains to potential for reinfection & some reduction in vaccine efficacy My prelim estimates based on publications/preprints, subject to Δ

New data source from HHS using Census Bureau Household Pulse Survey data estimates vaccine hesitancy on a county level. While the counties are quite similar to their respective states, it gives us a more granular analysis to adjust strategies. https://aspe.hhs.gov/pdf-report/vaccine-hesitancy

For those who are wondering: There is a slight association (r = 0.34) between the percentage a county voted for Trump in 2020 and estimated hesitancy levels. As @JReinerMD mentioned, GOP state, county and local levels need to do their part to promote vaccination.

Just published @TheLancet: Effect of vaccine in >23,000 health care workers https://thelancet.com/journals/lancet/article/PIIS0140-6736(21)00790-X/fulltext… Dose response: No vaccine 977 infections; 1 dose 71 infections; 2 doses 9 infections (14|8|4 per 10,000 person-days) "can prevent both symptomatic and asymptomatic infection "

If things come down to this, doctors are going to be at the recieving end of all patient wrath. As much as I pray for Delhi patients' lives, I pray for the safety of my colleagues. #DelhiLockdown

There is a Twitter account for the Great Barrington Declaration. It is totally silent on what is presently unfolding in India. Is it really possible to watch the surge there and in Brazil and not feel challenged in your beliefs about herd immunity? for otherwise smart people?

In 91 days, we are 99.71% of the way to President Biden's goal of 200 million shots in 100 days. We can expect to formally meet this goal tomorrow. #VaccinateAmerica

@STWorg @ceptional

2/2 Over the next 10 days, we will pick out key themes for critical discussion to see what progress was achieved, what aspirations were dashed, and general discussion of what we can build on going forward

It is 13 months since the piece on "reconfiguring behavioural science" for crisis knowledge management that led to http://SciBeh.org What have we learned? What did the behavioural sciences get right? What went wrong? Join the discussion! 1/2

Useful framing of current knowledge (not including newest case). Pfizer and Moderna have NO SUCH concern.

One year ago today: A paper in Clinical Infectious Diseases identifies a 16.3% rate of in-household COVID-19 infection, with a 21% rate in adults and 4% in children. Among cases where the first household member to be infected self-quarantined at home, the rate was 0%.

Inspired by @eliaskrainski's vaccine uptake pyramid for Brazil, here's one for England:

With vaccination >80% for people 65+, the group to focus on to reduce the most deaths are those aged 50-64. Is reaching them more important than vaccinating the young transmitters? I'd say yes. We can do both, but the middle aged are crucial They're how we get deaths down.

Below are the current vaccination rates by age group. https://covid.cdc.gov/covid-data-tracker/#vaccination-demographics-trends

@Bob_Wachter has an excellent piece outlining why vaccination is key, given local officials unwillingness to give up relaxing mitigation measures. But who should be the focus of vaccination efforts?

Here's the change in distribution of COVID deaths by age. - In Dec, 63% of all deaths were in ages 75+. - In Mar, most deaths were UNDER age 75, The flattening of COVID deaths according to age is extraordinary. https://cdc.gov/nchs/nvss/vsrr/covid_weekly/index.htm…

COVID deaths in nursing homes are down massively. - In Dec 2020, these accounted for 32% of all COVID deaths. - In Mar 2021, thanks to vaccination, they accounted for just 4%. So who are the remaining 96%? https://data.cms.gov/stories/s/COVID-19-Nursing-Home-Data/bkwz-xpvg/

US COVID-19 deaths remain upwards of 700 deaths a day. Who are these people who are dying?