4,785 Matching Annotations
  1. Jun 2020
    1. David Spiegelhalter said this morning, “Peer review has just disappeared from scientific analysis.” This is complete and utter nonsense. Our editors across 19 Lancet journals do nothing else but peer review. We intensively review all COVID-19 research papers. You know this David.
    1. I'm getting asked more about the 'k' parameter that describes variation in the reproduction number, R (i.e. describes superspreading). But what does this parameter actually mean? A short statistical thread... 1/
    2. As 'k' is a bit hard to interpret directly, we can also use it to calculate what proportion of infections generate a given amount of transmission, e.g. does transmission follow something like a '20/80 rule'? Here's a conversion table for R=3... 8/8
    3. lternatively, we can estimate k from the distribution of outbreak sizes after infections are introduced to a new location, using a simple transmission model: https://wellcomeopenresearch.org/articles/5-67 7/
    4. So how do we calculate k? One way is to estimate directly from transmission chains reconstructed from contact tracing data, e.g. https://researchsquare.com/article/rs-29548/v1… 6/
    5. If k is smaller, then there is more variability - some cases generate a lot of new infections, while most generate very few. Here's the negative binomial distribution when R=3, k=0.2 (plausible for COVID, SARS). Note x-scale is cropped at 20, but can obviously go higher 5/
    6. If k is very large, every case generates transmission randomly at constant rate with mean=R (i.e. equivalent to a Poisson process as k->infinity). Here's what the distribution of transmission looks like when R=3, k=1000 (dashed line shows R). 4/
    7. We can do this by fitting a curve to the distribution of secondary infections, and see how much variation there is. A commonly used tool is the negative binomial distribution, which has mean=R and variation captured by a dispersion parameter 'k' https://nature.com/articles/nature04153… 3/Superspreading and the effect of individual variation on disease emergFrom Typhoid Mary to SARS, it has long been known that some people spread disease more than others. But for diseases transmitted via casual contact, contagiousness arises from a plethora of social...nature.com
    8. R measures average transmission per case, but in reality some cases may generate more infection than others, e.g. because of events/places they visit while infectious. So we need a way to estimate variation in R at the individual-level... 2/
    9. I'm getting asked more about the 'k' parameter that describes variation in the reproduction number, R (i.e. describes superspreading). But what does this parameter actually mean? A short statistical thread... 1/
    1. 1) ACTIVE cases...shows which countries have 1) Peaked: Germany, S Korea, Japan, Italy, Spain... 2) Plateaued: France 3) Yet to peak: US, UK, Brazil, India...active cases still rising. 4) Second wave: Iran and.... Spain (?)
    2. How are countries recovering? 1) Some have peaked. 2) Some have plateaued. 3) Some still reporting rising ACTIVE cases...yet to peak. 4) Iran reporting big second wave...after initial peak.
    3. 1) Total cases in hotspot countries (>150K cases) 2) Total deaths in hotspot countries (>3000 deaths)
    4. The Compound Daily Growth Rate = 10.43%
    5. 1) TOTAL cases doubling in 15 days. 2) ACTIVE cases doubling in 25 days.
    6. #DailyUpdate #COVID19India As Total cases rise, the growth rate of ACTIVE cases= 2.8% - so doubling in 25 days. This is the lowest recorded rate (even after removing 'large recovery of 29th May' in Maharashtra). Need to double down on contact tracing & testing to slow further.
    1. UK could have reduced R much further by implementing contact tracing during lockdown & brought the total number of cases right down, making it much safer for schools to go back. But we haven't, and still have by far the highest number of cases and deaths per day in Europe... 31/
    2. This comes onto a broader point. It is 18 weeks since @WHO declared a #PublicHealthEmergency & 7 weeks into #lockdown. And yet there is still no community #ContactTracing. This paper shows how even moderate contact tracing brings case numbers down. https://twitter.com/ScienceShared/status/1262320191872131073?s=20… 30/
    3. We can't draw conclusions by comparing UK w/ Denmark as the situations are so different (100-1000 fold difference in #COVID19 pop prevalence). @TheBMA and teaching unions are right to say "Until we have got case numbers much lower we shouldn't consider re-opening schools." 29/
    4. ... (likely closer to 20,000 true cases) a day & hundreds of deaths a day. The massive difference in the prevalence of #COVID19 between the UK & Denmark means the risk of infection to children (and their contacts) in Denmark is much lower when compared to the risk in the UK. 28/
    5. What should the UK do? The UK gov seems to be using Denmark as evidence that opening schools is safe. However the situation there is very different from the UK. They have <100 cases a day & recently reported 0 COVID deaths in a day, whilst UK has ~3500 confirmed cases... 27/
    6. @NYGovCuomo: "We are still learning so much about #COVID19 & the more we learn the worse it seems to get. Whilst the disease is so prevalent in the pop (NY/UK) we shouldn't be risking our children when we know so little."
    7. With such little consensus, this now turns into risk assessment and ultimately a political decision, in which the individual risk to children & teachers (& parents) is weighed up against the value of children being in school (and ofc many other things but these are main 2). 25/
    8. And this is what it basically comes down to. There is no consensus view amongst experts. There is little strong evidence for either side, and much of it is contrasting. Can the UK afford to risk it? 24/
    9. ... normally healthy children being in ICU/dying and can have long-term cardiac complications. The disease shows ~4 weeks after initial infection & so likely to see cases increase. We are finding more out about this every day, but more studies (time) are needed.
    10. There is also increasing concern with now hundreds of seemingly #COVID19 associated Pims-TS syndrome cases confirmed in several countries. Pims-TS (whole system inflammatory toxic shock syndrome similar to #Kawasaki symptoms) has resulted in...
    11. And a retrospective cohort study based on 1286 close contacts of 319 #COVID19 cases in Shenzen showed that children (and younger people generally) were as likely as adults to be infected (but also much less likely to have mild or severe symptoms).
    12. Additionally, work by @ONS, who have been doing some excellent research on the impact of #COVID19 in the UK, showed there is currently no evidence that age affects the likelihood of being infected with #COVID19 in the UK.
    13. The man leading the coronavirus response in Germany, @c_drosten, heeds caution, showing evidence there is no sig difference in viral loads between children & adults. This highlights how much we have to learn & that all steps should be taken w/ caution.
    14. It seems clear that, unlike with influenza, children aren't key drivers of transmission & most will suffer no symptoms. But not being 'the key driver' doesn't mean they can't still drive transmission, & there is increasing concern over a related inflammatory syndrome... 18/
    15. These @nytimes & @guardian articles summarise Denmarks approach.
    16. And in Denmark, re-opening of schools (w/ effective social distancing measures), led to an increase in R from 0.7 to 0.9 shortly after (although now returned to 0.7). This seems to be the main 'case study' the UK is basing it being safe for schools to re-open. 16/
    17. Still, Prof @GrahamMedley, UK chief modeller, says children not the 'key drivers' of #coronavirus transmission... And @WHO says in countries where schools have remained open, outbreaks in schools have been rare & are normally associated with contacts with adults. 15/
    18. However, in regards to contact tracing studies suggesting children are less infectious, there needs to be more rigorous data to draw conclusions from this data alone, as need to take into account the potential for #superspreaders. 14/
    19. There is also a fair bit of contact tracing data suggesting that children are less infectious, such as this in a French chalet cluster, where an infected child did not transmit the disease despite close interactions within schools.
    20. There are also many studies showing lower incidence in children, such as a population-based study in Iceland in which 9199 were tested, of the 564 children <10y, 38 (6.7%) tested positive, vs 1183/8635 (13.7%) adolescents & adults, suggesting lower incidence in children.
    21. And this @ScienceMagazine paper shows risk of infection upon exposure in children may be about 1/3 (sorry for the simplification) that of adults, although also suggests proactive school closures can reduce peak incidence by 40-60% & delay the epidemic.
    22. ...on the spread of the virus. Should be noted that this models school closures and effect on peak rather than the re-opening of schools.
    23. So let's dig a little into the evidence for & against re-opening schools. A @ucl review published in @LancetChildAdol concluded that the evidence to support the closure of schools to combat Covid-19 is “very weak” & school closures are likely to have a minimal impact... 9/
    24. It should first be made clear that there is no scientific consensus on this, w/ many contrasting views & little strong evidence on the issue. This uncertainty means there is inherently significant risk involved & increases the need for evidence to be discussed #transparently. 8/
    25. But what is the evidence, and when will UK schools really be able to open safely? This @guardianscience article summarises the contrasting views between experts, with some experts saying it's low risk & a difficult but necessary decision.
    26. However, teaching unions & parents have raised legitimate concerns over whether it will really be safe, and @TheBMA announced that "Until we have got case numbers much lower we should not consider re-opening schools. The NEU is right to urge caution."
    27. Whilst @michaelgove said he can 'guarantee teachers and children will be safe.' More reasonably, @WHO chief scientist @doctorsoumya has said children are "less capable of spreading the virus, and are low risk."
    28. The increasing educational disadvantage means it is urgent to find a SAFE way to resume education. Dismissing concerns from teaching unions as 'scaremongering,' the Education Sec @GavinWilliamson announced that R being below 1 is the ‘green light’ for schools to re-open... 4/
    29. School closures have presented a huge challenge to children, parents & policymakers. A study by @TheIFS showed that better-off children now spend 30% more time on education than poorer children & school closures will increase educational inequalities.
    30. On 20 March 2020, UK schools closed their gates to all but the children of essential workers and those deemed most vulnerable. This remains the case; although the gvmnt has now said that more children will return to school at the start of June. 2/
    31. When should #Schools #Reopen in the UK? This thread summarises the key points & evidence on the issue & finishes with some things the government must do before allowing schools to re-open.
    1. and please consider joining the initiative, by joining the reddit community forum, or volunteering in other ways http://scibeh.org 7/7
    2. behav. sciences will still be central in all of this (if anything, their role looks set to broaden), but the balance of content in the @SciBeh feed will likely change. That also means adapting our sources Please help by tagging @SciBeh in any relevant material you find! 6/7
    3. in a context that already saw rising levels of instability pre-pandemic, and a flood of mis- and dis-information and polarisation fuelling means of information exchange 5/7
    4. and divergent views on what our future should look like, all to be negotiated with fragile democratic legitimacy (pandemic response options featured in no election manifesto, and political decisions will be made after limited discussion at warp speed) 4/7
    5. All of this change will take place in a societal and political context of increasingly divergent (or perceived to be divergent) individual interests (high risk vs. low risk, young vs. old, rich vs. poor, crisis winners vs. losers, now immune vs. 'haven't had it yet') 3/7
    6. what will only continue to grow over the next months is the demand for science guiding the restructure of key aspects of society (how we work, travel, education) short, mid and long term, and how we manage economic fallout and economic transformation 2/7
    7. A thought and a plea from @SciBeh: as the pandemic unfolds, we will see shifting in what aspects of the beh. sciences are most relevant to crisis response - the "first wave" emphasised risk communication, behaviour change, and mental health - 1/7
    1. Problem in using lag indicators: 21 days on average from infection to death- and countries seem to track 'progress' based on deaths. So by the time the data is worrying & increase in deaths is exponential, it's already too late.
    2. Then ease measures while testing widely & w/ good data systems that alert public whether it is red/amber/green in their area. Need clusters of cases identified rapidly & broken up before tips over into sustained community transmission. If it tips, hard to avoid another lockdown.
    3. My suggestion: bring down daily new cases to a low level, get test/trace/isolate in place and core infrastructure build up, get regular testing going for essential workers/teachers/students, monitor borders for imported cases, & move to mandatory masks in shops/public transport.
    4. Looking at the estimates for daily new cases in England (8K/day), the openings of shops/schools on Monday, watching carefully what's happening in East Asia & combining this with what we know so far about this virus --> feels like mistakes are being repeated from early March.
    1. #Italy remains one of the worst outbreaks & one of the best & most consistent responses to lockdown/NPI measures. 0.6% positive rate; STILL testing at rate of greater than 1/1000 each day. The US is NOT currently on this path. (some regions are). 33K fatalities.
    1. We should be able to explain good faith third parties how science works and why we do what we do.In Germany we just had an open science flare up. A famous virologists (Prof. Christian Drosten) published a preprint and colleagues gave feedback on it, mostly how to improve the statistical analysis and as far as I can judge this only made the conclusion stronger. Our Daily Mail (Bild Zeitung) spun that into a series of stories about Drosten doing shady science and one former public health official and professor was willing to help them by calling for a retraction, while the key finding stood firm and all that was needed were some revisions.There was close to a popular uprising against the Bild Zeitung. Science kept Germany safe and we would not let the Bild Zeitung drag us to the USA or UK. You can see the burning buildings and looted Target Store under the hashtags. #TeamScience and #TeamDrostenIt was perfectly possible to explain to good faith third parties that preprints were preliminary, that peer review and disagreements belong to science, that feedback is normal (one of the reviewers is now an author) and that no work of science is perfect, but that it was good enough to come to the carefully formulated conclusion, which was only a small part of the puzzle. I am sure for nearly everyone this was a bizarre world they did not know, normally peer review in closed. Surely they did not understand how it works in the short time this flare up happened, but they trusted science and the scientists from many fields who told them all was fine.Even if this could be abused by bad faith actors, I think it was good to publish this study as a preprint, to have people see the peer review in the open. That is good science, especially in these times were we cannot afford to wait too long, and we should do so.
    1. Argh it is so frustrating that these regulations don't get put to parliament at least for debate. It's madness. These are huge changes and they affect every single person in England, and there are loads of potentially unclear and difficult bits. Why so reluctant to engage debate?
    2. Wait, believe it or not this could be correct! https://twitter.com/icecolbeveridge/status/1267103055293681664?s=20… But the people in the shop need to have gathered there in order to shop "with each other" so unless you all met up there deliberately, then you're safe
    3. Ooh, cheeky little change in the pre-amble to add a proportionality test, was necessity, perhaps a response to the recent @HumanRightsCtte report... chuffed about that actually https://committees.parliament.uk/writtenevidence/5454/default/…
    4. Also, the definition of "elite athlete" gets a lot of attention in these amendments. Long story short, I don't qualify. @holland_tom maybe you do?
    5. A few more places that can open
    6. Some more places which must shut (although they prob weren't open anyway). What is a "landmark"?
    7. 2 months after the regulations appeared we finally have a definition of a gathering "two or more people are present together in the same place in order to engage in any form of social interaction with each other, or to undertake any other activity with each other" @SeethingMead
    8. Third big change - now illegal for there to be a gathering of 2 or more people in private places, which includes your own home, unless it falls within one of the (it seems exhaustive) list of reasonable excuses - can be people from same household (obvs) - for work etc
    9. Second big change: Regulation 7 has been completely replaced - Gatherings of 6 people or less allowed outside in any formation (i.e. from any number of households) - Gatherings over 6 people prohibited without "reasonable excuse", there is an *exhuastive* list of excuses AND..
    10. To be clear - from tomorrow - he police can no longer get involved with why you are outside of the place you are living. - No more power to direct people back home - no more power to fine for leaving/being outside of home without reasonable excuse
    11. The lockdown regulations have changed very significantly: - No more prohibition on leaving the place you are living or being outside of it without a "reasonable excuse" - Regulation 6 replaced by prohibition on staying over somewhere without a reasonable excuse
    12. The Amendment Regulations are here - made today (!) and laid before Parliament tomorrow morning at 11:30am http://legislation.gov.uk/uksi/2020/558/pdfs/uksi_20200558_en.pdf
  2. May 2020
    1. This despite signing up 90,000 new subscribers since March. The coverage has been much praised. Bradley says it’s to ensure longterm viability.
    2. .@TheAtlantic to cut staff by 68 positions, or 17 percent, in response to current economy, per chairman David Bradley statement
    3. .@TheAtlantic to cut staff by 68 positions, or 17 percent, in response to current economy, per chairman David Bradley statement
    1. South Korea really bringing the hammer down on quarantine violations. 27-year-old man who twice left mandatory self quarantine, with *two days* to go until the end of his 14-day isolation, has been sentenced to four months in prison. First covid-19 sentencing, per @YonhapNews
    1. Then ease measures while testing widely & w/ good data systems that alert public whether it is red/amber/green in their area. Need clusters of cases identified rapidly & broken up before tips over into sustained community transmission. If it tips, hard to avoid another lockdown.
    2. My suggestion: bring down daily new cases to a low level, get test/trace/isolate in place and core infrastructure build up, get regular testing going for essential workers/teachers/students, monitor borders for imported cases, & move to mandatory masks in shops/public transport.
    3. Looking at the estimates for daily new cases in England (8K/day), the openings of shops/schools on Monday, watching carefully what's happening in East Asia & combining this with what we know so far about this virus --> feels like mistakes are being repeated from early March.
    1. This is because we can be confident that if our analysis isn't the right analysis, it's at least extremely close to the right analysis in all the ways that impact the conclusion. Interrogating findings from different angles is the best defense against bugs in the analysis. 6/6
    2. Statisticians like to do various checks of the results. Doing checks allows us to confirm that our analysis has certain properties which we think are properties of a good analysis. I think this makes it less necessary that our analysis contain zero bugs. 5/6
    3. As statistical analyses become more complex, we might want to ensure that our statistical analyses satisfy certain properties instead of worrying about small mistakes in implementation. I believe statisticians already intuitively do this. 4/6
    4. The way I understand unit tests in coding is that instead of repeatedly reading over the code to ensure logical correctness, we use empirical tests of the code to constrain its possible behaviors. I think this would be a good idea to import into statistical analyses. 3/6
    5. I firmly believe that 100% bug free code is impossible to achieve for any sufficiently complex code and statistics isn't any different. The right goal isn't bug free code, it's minimizing the impact of bugs on the scientific conclusions. 2/6
    6. I want to talk about bugs in statistical analyses. I think many data analysts worry unnecessarily about this. I do think it's important to put a good faith effort into avoiding bugs, but I know data analysts that live in terror of hearing there's a bug in published work. 1/6
    1. Correction. One of many things that I have learned since I posted this thread in March: "R0" should read "Rt" in this opening tweet.
    2. The symptoms to meet SARS-CoV-2 testing are: patients who have fever >37.5°C/respiratory symptoms & had a close contact history with an infected patient, visited or contacted with those who visited to high risk countries, or may have pneumonia requiring hospitalization. 31/
    3. Second, RT-PCR-based testing generates a lot of false negatives and also false positives. To effectively identify infected patients with limited resource, it is important to focus the testing on patients with high likelihood of COVID-19. https://twitter.com/Dr_yandel/status/1232463168422567937… 30/
    4. First, different countries are using different PCR protocols and primers to detect SAR-CoV-2. What I heard from my microbiologist friend in Japan is that NIID made great efforts to establish a reliable protocol, which is now being used in Japan. https://who.int/emergencies/diseases/novel-coronavirus-2019/technical-guidance/laboratory-guidance… 29/
    5. So, back to testing. I know that there has been some doubts on COVID-19 case counting in Japan due to low numbers of testing. I have limited information, but I can point out a few things. 28/
    6. What I am hoping to do here is to see if we can learn some hints for a long term solution for COVID-19 from Japan, which is apparently handling the situation relatively well so far. 27/
    7. I'm back. Please keep in mind that the strategy working for Japan may not be directly applicable to cities like NYC, which is facing hospital overloads. For us (New Yorkers), it is important now to do whatever possible to do to suppress the current burst of infections. 26/
    8. OK. Here comes the limit of max tweet numbers in a thread. I'll take some break, and will continue the thread this later. 25/
    9. So how the team was able to successfully detect infected cases? Population of Sapporo, capital of Hokkaido, is 2M. The size of Hokkaido is similar to Maine. I'd like to start with testing. 24/
    10. Thanks to early detection and isolations of patients, a burst of infection in Hokkaido was successfully prevented, and now R0 is maintained below 1. On March 18, the mayor of Hokkaido declared State of Emergency to be terminated next day as has initially planned. 23/
    11. Meanwhile, cluster Response Team of Ministry of Health, Labour and Welfare (MHLW), National Institute of Infectious Diseases (NIID) and local public health center worked together to isolate patients and their close contacts through PCR testing. 22/
    12. The mayor requested Hokkaido citizens to refrain from going outside during weekends. Public schools were closed from 2/28. Large-scale events were asked to be canceled. https://nikkei.com/article/DGXMZO56188700Y0A220C2L41000/… 21/
    13. On 2/28, State of Emergency was declared in Hokkaido, after detecting 15 and 12 positive cases on 2/27 & 28, respectively. Total# was 66 at that point. 20/
    14. There was Sapporo Snow Festival held from 1/31 to 2/11, where 2M people visited, likely causing clustered infections at this period. A lot of closed, crowded temporal spaces without good ventilation where many tourists gathered, talked and ate foods were generated. 19/
    15. Now, as an example, I'd like to explain a case in Hokkaido, where clusters of COVID-19 were detected from late January. 18/
    16. It is worth noting that following this advice, the government even encourage people to go outside. An opposite of what is happening in NYC. https://twitter.com/Kantei_Saigai/status/1240541777712443393… 17/
    17. 1) closed space with poor ventilation 2) crowded with many people 3) conversations and vocalization in close proximity (within arm's reach of one another) 16/
    18. So, the team set the priority to suppress super-spreading events, and urges people to avoid places where the following three conditions were met simultaneously: 15/
    19. Supporting data from Nishiura are available in this preprint. “The odds 44 that a primary case transmitted COVID-19 in a closed environment was 18.7 times greater compared to an open-air environment (95% confidence interval [CI]: 6.0, 57.9). “ 14/
    20. "2) Meanwhile, cases have been reported where a single infected patient has transmitted the infection to multiple persons in areas such as athletic gyms, houseboats, buffet style meals, mahjong rooms, guest houses at ski resorts, and air-tight provisional tents." 13/
    21. "1) So far, approximately 80% of the confirmed infection cases in Japan have not transmitted the infection to others." 12/
    22. Cluster Response Team of Ministry of Health, lead by Hiroshi Nishiura, Prof at Hokkaido Univ. and an expert of mathematical modeling of infectious disease, defined characteristics of the transmission rout as: 11/
    23. The underlying principle is laid out in this handout by MHLW. “The important thing is to minimize the spread of infection in the country by preventing one cluster of patients from creating another cluster.” 10/
    24. 1) early detection of and early response to clusters 2)early patient diagnosis and enhancement of intensive care and the securing of a medical service system for the severely ill 3) behavior modification of citizens 9/
    25. The Government of Japan, Cluster Response Team of Ministry of Health, Labour and Welfare (MHLW) has stated The three pillars of Japan's strategy:
    26. Remember that a lot of tourists from China visited Japan during Chinese New Year. There was also a case of Diamond Princess. However, they did not trigger catastrophe. So how did Japanese do? 7/
    27. If the R0≅1 can be maintained with this minimum restrain, and if the R01 can be naturally reduced when humid summer arrives, there is a reasonable chance that the society can maintain good activities without risking hospital overload. 6/
    28. Although most sports and music events have been cancelled in Japan, a large majority of people go outside and work, and restaurants are open. Schools are currently spring break period, but they are trying to open in April. 5/
    29. The future of next 2 years predicted by Fergusson is depressing, as it demands strong restriction of school/university and business to accomplish R<1. In Japan, so far R0≅1 has been accomplished with volunteer-based cancellations of large-scale events, and school closures. 4/
    30. Many of you are now familiar with a simulation analysis of Fergusson et al, indicating that mitigation (R0<1) is not sufficient to prevent hospital overload, while suppression (R0<1) would take years to gain herd immunity until we get vaccines. 3/
    31. Disclaimer: I am a cell biologist studying mitosis, and not an expert of infectious disease. I've gathered most information is from internet. It would be great if experts @PaulBieniasz, @danmucida, @VirusesImmunity & others . 2/
    32. It is perplexing to see that the article ignores tactics of Japan, where R0 of COVID-19 is clearly lower. The purpose of this thread is to search for reasons why Japan has been successful. 1/
    1. RT @thehowie: #SouthKorea follow-up 0.19% positive rate with 13 new cases (7 imported; 5 from #itaewon CONTACTS (not from the clubs); 1 new…
    2. The stigma associated with this outbreak will reverberate for some time. It would be a problem for any country to find infection with #SARSCoV2 associated with stigma: it will make testing, contact tracing, and isolation more difficult. Yet we know it is already present. 3/3
    3. 19 new cases found today: 10 were imported; 1 was a visitor to the #Itaewon clubs; 5 were contacts of #Itaewon cases. 3 were new community spread cases. < 0.1% positive rate (a new low), with 13.9K tests done. cCFR 2.4%. 2/3
    4. #SouthKorea 88+ patrons got #Covid19 from (presumably) a single contact in #Itaewon on May 1/2. 74 additional cases have been found from contacts. While testing goes on, they have this under control. Homophobia will take longer to extinguish.
    1. This is meant to highlight some of the best aspects I see, and is not an endorsement of these countries’ overall strategies. There is still a long way to go, but we should pay close attention to how even places with limited resources respond (= hard work, people power). 7/7
    2. In South Africa, the government has deployed 28,000 health workers to test and screen communities across the country, drawing on the health infrastructure that had been built in response to the HIV/AIDS pandemic. 6/7
    3. South Korea learned a lot of valuable lessons during their 2015 Middle East Respiratory Syndrome outbreak, particularly strengthening surveillance systems, diagnostic capacity, and hospital infection control. 5/7
    4. India is leveraging an existing gigantic integrated disease surveillance network to trace and quarantine infected people. Interestingly, these networks are strongest in the poorest areas. 4/7
    5. The Ugandan health authorities are testing around 1,000 truck drivers a day. But many of those who test positive have come from Tanzania and Kenya, countries that are not monitoring as aggressively. 3/7
    6. Sierra Leone repurposed disease-tracking protocols established for Ebola. The government set up emergency operations centers in every district and recruited 14,000 community health workers, including 1,500 contact tracers. 2/7
    7. In comparing outbreaks across countries, I think there is a lot of luck involved in the timing (it’s when, not if). But places that have faced outbreaks in the past have a collective memory and can respond quickly. Highlighting a few examples. 1/7
    1. 24. I'm sorry that Dr. Heald felt that I was being being insulting by calling his study "odd". I should have taken high road that he did when questioned about his odd assertion that bonfire smoke is a vector for COVID-19.
    2. 23. I do legitimately feel for those risking their own safety to work on the frontline. Though you know what would make their jobs safer and less hectic? Not publishing dangerous, unjustified policy advice such as the notion that it is safest to reopen the hardest-hit areas.
    3. 22. And finally, speaking of causal inference and directionality, when I read this paper I can't help but wonder whether science is driving policy aims, or vice versa. /fin
    4. 21. The authors end by drawing conclusions about about economic and social factors that they did not even pretend to address in the paper. This sort of thing feels OK to me in an OpEd, but not in a scientific manuscript.
    5. 20. The problematic (if implicit) causal inference in this paper leads to a bizarre and I think dangerous conclusion, should anyone take it seriously. This is an extraordinary claim, countering common sense, with immediate relevance. One needs far stronger evidence in my view.
    6. 19. In the discussion the authors suggest their regression reveals susceptible depletion. But the historic number of confirmed cases is likely influenced by the same common causes that influence declines in R, e.g. control measures. This is a huge causal inference failure.
    7. 18. And then we get to the conclusion: the authors think COVID19 is about 1/5th as deadly as most others think, and about 5-10 times as prevalent. I.e., we're much closer to herd immunity than we thought and the cost of getting there is much lower.
    8. 17. 5) Testing effort varies over time and across locations. How does this play into the estimation procedure? 6) To extrapolate in this way implicitly assumes all regions will follow the same trajectory given enough time. How reasonable is that?
    9. 16. 3) Why extrapolate to everyone being infected instead of the final epidemic size or even herd immunity threshold? 4) The regression shown accounts for only a modest fraction of the variation. How does the remaining variation impact the predictions?
    10. 15. I have SO MANY questions. Just a few of them: 1) What is the causal basis for the relation between R and reported cases? Is this susceptible depletion? Something behavioral? Or is the claim it doesn't matter? 2) Given (1), why can you extrapolate and why linearly?
    11. 14. And this turns out to be about 400,000. Yet there are 60M+ in the UK, which gives a scaling of at least 150 cases per reported cases if everyone gets infected.
    12. 13. And as always, I welcome corrections from any authors that haven't blocked me already. The idea seems to be to extrapolate to figure out how many cases per capita would be reported by the time you reach R=0.
    13. 12. OK, let' try to get back to the paper. I'm really struggling to understand what is going on here. This doesn't look like any infectious disease epidemiological method I've ever seen, and there's no citation given. But I can try to reconstruct the thought process.
    14. 11. Oddly enough, I do know a little bit about peer review. In addition to writing a thousand of them or so in my career, I've written a little bit about peer review and what it does and does not guarantee.
    15. 10. And I guess now he's falling back on the old "It's been peer reviewed" defense. Well, Adrian, consider this a post-publication peer review.
    16. 9. Here was the original post to which I was responding, and which I suppose no longer appears in the first post of my thread.
    17. 8. Aside: I haven't even had a chance to explain what is wrong with the study, and I've already been blocked by the author. Friends, this is not an ordinary scientific response to criticism. Especially criticism that hasn't even arrived yet.
    18. 7. Now it starts to get really weird.
    19. 6. Of the predictors in the regression, only cases/1000 people is predictive, albeit with an r^2 of 0.2. The authors then posit a linear relationship between R and case density: R_ADIR = 1.06 - 0.16 x Current Total Cases/1,000 population. Here's that data.
    20. 5. What determines R? In an effort to estimate this, the authors use a regression approach across local regions. There's no underlying mechanistic model of how R changes with time, interventions, etc., nor any temporal analysis. Notice case density is from an April 8th snapshot.
    21. 4. But let's get to the science. What did the authors do? They start by estimating a local R value that they call the Average Daily Infection Rate, and estimating its derivative. One could dig into this more deeply, but let's keep going instead.
    22. 3. Things get odd right from the very start. The first line of the paper's abstract is not your usual way of beginning a scientific report.
    23. 2. The principal claim is that "unreported community infection may be >200 times higher than reported cases", meaning that "29% of the population may already have had the disease." (Most estimates from the US, EU, UK are closer to 10x than 200x) The tabloids are there:
    24. 1. An odd research study out of U. Manchester today uses an indirect and, frankly, bizarre method to estimate the incidence in the UK as being vastly higher than that inferred using more direct approaches.
    1. Antibody testing suggests ~15% of NYers (~20% in NYC) have had Covid https://nytimes.com/2020/04/23/nyregion/coronavirus-antibodies-test-ny.html… vs • "I did some back of the envelope extrapolations and found that 83% of NYers have had Covid. Here are my charts from Google Sheets." I know which one I’m going with...
    2. Right on cue, this drops into my inbox
    3. My other tip: follow lots of experts. For me, that means @CT_Bergstrom, @AdamJKucharski, @nataliexdean, @EricTopol, @cmyeaton, @globalhlthtwit, @ActuaryByDay and others. They don’t always agree! And that’s *good*. This stuff is complicated.
    4. We already know from comprehensive research in other countries that the share of people who've had Covid in even hard hit countries, is around 5%. Claims that differ significantly from that require extraordinary evidence
    5. My tip for anyone, fellow journalist or otherwise: weigh any surprising new claims against the balance of evidence already out there on the issue in question.
    6. Yesterday's Manchester paper is a particularly acute case, as the claims in that study concern a critical issue that people will use to justify policy — how many people in the UK may have already had Covid.
    7. It's absolutely vital that as journalists we do all the necessary checks before reporting on highly sensitive issues like this.
    8. Few weeks ago media reported studies saying air pollution levels had big impact on Covid death rates. Problem 1: studies hadn't been peer reviewed. Problem 2: air pollution & pop dens are correlated. Over at SMC, experts pointed out flaws: https://sciencemediacentre.org/expert-reaction-to-preprint-on-air-pollution-in-england-and-covid-19-severity/
    9. And here's a detailed, point-by-point take-down of the same paper by @CT_Bergstrom, including an explanation that even peer review isn't enough to ensure a study's findings are watertight:
    10. Thread: Critical assessment of scientific papers by the media has never been more important than during the pandemic That new Manchester study saying 25% of UK has HD Covid *was* peer reviewed, but has already been comprehensively debunked by many leading epidemiologists.
    1. It seems like all the news articles are based on this press release, which itself does not link to any published or unpublished research: https://scs.cmu.edu/news/nearly-half-twitter-accounts-discussing-%E2%80%98reopening-america%E2%80%99-may-be-bots… When I go to the publications page of the lab, I can't find anything from 2020 about bots. http://casos.cs.cmu.edu/publications/index.php…
    2. The thing I normally do when I see an article like this is click through to the academic research being cited but....... I cannot find the academic research being cited
    3. Ah here we go, this is the paper that describes Bothunter, the algorithm described in the Asia-Pacific paper, which again I am *assuming* is what was used in the research referred to in the NPR article (which again, has not been published) http://casos.cs.cmu.edu/publications/papers/LB_5.pdf
    4. I found a couple of posters that share the same Office of Naval Research funding award numbers as the Asia-Pacific bot paper that detail some machine learning approaches to social network analysis, possibly related: http://casos.cs.cmu.edu/events/summer_institute/2019/si_portal/posters/poster-Binxuan%201.pdf…
    5. Unfortunately this presents its bot detection methodology as a handwavey machine learning black box, based on a training data set that itself isn't auditable, and with a threshold of 60% probability-you-are-a-bot being their cutoff for comfortably declaring an account a bot
    6. NPR -- Researchers: Aurora Borealis Discovered At This Time of Year, At This Time of Day, In This Part of the Country, Localized Entirely Within Your Kitchen
    7. Maybe I should send out a press release and see what mainstream news outlets run with it: "Darius Kazemi, noted Twitter bot expert, says to CMU researchers 'nuh-uh, you're definitely wrong', based on research that he has not published yet and almost certainly exists"
    8. In conclusion, holy shit, publish or at least preprint your damn research before you do a massively alarmist press release, my fuckin god
    9. ...and probably not the activity people are thinking of when they see a headline like "Nearly Half Of Accounts Tweeting About Coronavirus Are Likely Bots". But since there is no published paper for that particular press released, I am just guessing based on their prior research!
    10. This incident was definitely a bot-based attack, but of a weird DDOS style harassment attack, rather than a "take control of the conversation" style attack. In other words, their training data set (at least for this paper) is based on a very narrow slice of bot activity...120
    11. Well they based it off of this article by the Atlantic Council:#BotSpot: The IntimidatorsTwitter bots unleashed in a social media disruption tacticmedium.com
    12. So to train their model they need known bots acocunts. Instead of attempting to attract bots, or looking at accounts that were suspended for bot activity, they picked a single "known and publicized" bot attack on the Atlantic Council (!). How do they know it was a bot attack?
    13. Oh no. This paper is.... not very good in my opinion. It's 8 pages long, about 3 pages of which is the actual research, and those sections (1, 2, and 3) don't give hardly any auditable information. What they do lay out is, well, not what I would do if I were running a bot study
    14. Ah here we go, this is the paper that describes Bothunter, the algorithm described in the Asia-Pacific paper, which again I am *assuming* is what was used in the research referred to in the NPR article (which again, has not been published) http://casos.cs.cmu.edu/publications/papers/LB_5.pdf
    15. I suppose that for now I have to assume the lab is using the same methodology outlined in this 2019 paper looking at the role of bots in activist hashtags in the Asia-Pacific region (PDF): http://casos.cs.cmu.edu/events/summer_institute/2019/si_portal/pubs/Uyheng%20-%20Characterizing%20Bot%20Networks.pdf…
    16. Can anyone point me to a published paper or any of the research mentioned in this article? There's no link here, no link in an MIT Tech Review article, nothing in several other articles I've seen. @CMU_CASOS can you help me out here? I want to review your methodology.Citar TweetNPR@NPR · 21 may.Nearly half of the accounts tweeting about the coronavirus pandemic are likely bots, Carnegie Mellon researchers say, adding that the tweets appeared aimed at sowing divisions in America. https://trib.al/DFLzKEu
    1. These are very interesting initiatives, and a very useful post - it is quite heartening to see how much self-organisation has occurred in such a short space of time.One type of consideration that seems difficult to bring into the discussion is purely practical. So, for example, regarding policies of PPE equipment, and the speed with which testing might be expanded, there is almost certainly a great deal of expertise distributed around the policy, healthcare management, practitioner, and business community that would help identify what the current situation is, and what realistic options there are to help fix it.Many of these people may not be able to contribute except anonymously---it would be incredibly helpful to have some way of allowing insiders to safely (in terms of their careers) feed relevant information to the debate.It is not obvious how we do this, but perhaps something reminiscent of a prediction market, although surely with no money changing hands, might be helpful.Similarly, it would be great to have some way of aggregating experiences and judgements from relevant individuals (e.g., some kind of barometer for PPE/testing availability which could be based on judgements by frontline staff; or priorities from the frontline which may be very different from those perceived from the upper reaches of government, or indeed the academic community).