Good piece on the value of digital contact tracing in future pandemics by @marcelsalathe – combined with better linkage to venues of transmission (e.g. superspreading events), potential for a lot of impact here. 1/nature.com/articles/d4158… During COVID, countries were competing with an exponential process, which meant any individual targeted intervention (like testing, isolation and contact tracing) had to be able to scale easily. Some places understood this more than others... 2/

https://twitter.com/adamjkucharski/status/1256140521065664515?s=20

It's remakable some people are still claiming COVID had a 'susceptible-infected-recovered-susceptible' dynamic early on, i.e. claiming most in UK got it in 1st wave and 2nd wave was driven by reinfections. Let's look at the heroic assumptions that this claim requires... 1/ 1. Assumes first waves declined not because of reduction in contacts, but because of lots of infections and resulting strong immunising responses - and yet these widespread strong immune responses somehow weren't detectable on any antibody test. 2/

I recently gave a talk at @JuniperConsort1 outlining some of the work we've been doing in @Epiverse_TRACE with @DataDotOrg and a range of collaborators to try and improve software tools for epidemic response - and how others can contribute to these collective efforts... 1/ As a motivation, I asked the question 'What could the final size of an epidemic be?' - as a first pass, there's a relatively simple method we could use based on an SIR model, but even implementing this can be complicated... 2/

Why it makes no sense to use total overall COVID deaths as the comparison metric when evaluating the impact of COVID measures, and why we need to focus on transmission dynamics instead. A thread… 1/ Suppose we have two countries, A and B. Country A adopts a lighter touch strategy X early on that gets the reproduction number down to 1 (i.e. epidemic remains flat). Country B leaves it later, then adopts a more stringent strategy Y to bring epidemic down (i.e. R below 1)… 2/

In the past year, @LSHTM_CEPR has (co-)hosted events on a range of epidemic topics, from public trust and global treaties to analytics software and response strategies.

In case you missed them, here are few to catch up on… Vernon Lee on Experience, evidence and some intuition in responding to COVID-19 in Singapore: lshtm.ac.uk/newsevents/eve…

There's something a eerily familiar about todays' 'new' IEA report on lockdowns, right down to the text, tables, and half-baked methods. And, of course, the massive estimated effect of masks that somehow hasn't made it into the headlines... 1/

https://twitter.com/adamjkucharski/status/1489529756592787460

Lots has been written already about this issues with this analysis (e.g. above thread and factcheck.org/2022/03/sciche…), from a lack of accounting for epidemic dynamics to performing a 'meta-analysis' on datasets that aren't independent... 2/

A year ago I talked to @guardianscience about lockdowns and COVID control measures: theguardian.com/science/audio/…

A few summary thoughts:
- In discussions, need to be clear what we mean by 'lockdown'
- Need to consider information available at time, rather than just in hindsight

1/ - Lockdowns (i.e. stay at home orders) were introduced as a last-ditch measure in many countries – it wasn't clear where transmission was or what would suppress it
- Closures delayed epidemics, at enormous cost. Some places used that delay much better than others.

2/

In hindsight, it can be easy to forget that all the parameters we now have for COVID - from incubation period and clinical progression to severity and transmissibility - had to be estimated by someone. A thread on some of the crucial early work in this area... 1/ 17 Jan 2020. Estimation of unreported cases in Wuhan based on exported international cases, by Imai et al: imperial.ac.uk/mrc-global-inf… 2/

To share some of the discussions and ideas happening during Epiverse software development with others than might find them useful, we have a new blog: epiverse-trace.github.io/blog.html

Here's a thread with the posts so far... 1. Ensuring & Showcasing the Statistical Correctness of your R Package by @grusonh epiverse-trace.github.io/posts/statisti…

Need to write up a scientific paper but got writer's block and staring at a blank document? A few tips I've found helpful over the years... 1/ Start by copying your key results figures/tables into the document. They don't have to be totally polished in terms of colour scheme etc. Get them down on paper, and arrange them in an order that makes sense in the context of your overarching research question. 2/

If you're interested in the maths of disease emergence (and we probably all should be to some extent), here's a thread with a few more papers... 1/

https://twitter.com/adamjkucharski/status/1621164561066762240

What characterizes a successful invader? ncbi.nlm.nih.gov/pmc/articles/P…

2/

How many infections have actually been happening in China? And how big was the epidemic peak? New analysis with @PungRachael et al, using data on recent arrivals from mainland China into Singapore to reconstruct case dynamics… 1/

medrxiv.org/content/10.110… Based on recent arrivals from China who were detected as cases within community, we estimated the epidemic peaked in mid-December with around 3% of people getting infected per day (among those would then travel)… 2/

If an epidemic is growing, most infected people will have been infected very recently. Which also means they're less likely to test positive...

Below shows estimate of who might be detected with a test 1 day pre-departure, with another day for travel... 1/ The obvious question that folllows is 'how many infections are missed'? But this is actually not so straightforward, because people infected a while ago are also more likely to be missed – but they are also less likely to be infectious. 2/

For context, Omicron waves peaked with an estimated 5-8% testing positive in UK, US and France last year (ons.gov.uk/peoplepopulati… & medrxiv.org/content/10.110…)... 1/

https://twitter.com/business/status/1608136965244813314

To get a true infection prevalence in a city anywhere near 50%, you'd need an extremely high R at the start of the epidemic, and testing that happened to catch the peak (because, all things being equal, the larger the initial R, the shorter the epidemic will be)... 2/

Targeted travel restrictions are often popular as a ‘seen to be doing something’ measure, but they will have limited effect on a domestic epidemic unless very specific criteria are met… 1/

bbc.co.uk/news/world-us-… The first issue is that targeted measures assume that out of all arrival countries, the level of infection is much larger in place(s) being targeted. Which has rarely been the case for COVID, given paucity of systematic community or arrival testing. 2/

Some useful reflections (and more to come soon from our team). One thing I’d add about incentives is that as well as funders, health agencies and governments, journal editors and reviewers could also play a role in shifting software culture in academia… 1/

https://twitter.com/johnlees6/status/1605180531980877824

If journals refused to accept modelling/analysis papers unless they came with documented code (I.e. at least a README clearly outlining analysis pipeline) and underlying data (or dummy data if too sensitive to share), I bet habits would change fast. 2/

These tools have potential to transform teaching/training, but they also have a number of implications – and not always obvious ones – for analytics work in general. A few thoughts… 1/

https://twitter.com/marcelsalathe/status/1598769656776384512

When I used to teach mathematical biology to undergraduates, it sometimes had a reputation for being an ‘easy’ area of maths, because the equations were less technical (e.g. ODEs rather than PDEs)... 2/

It's been 10 years since I got my PhD. So I few scattered reflections on things I've come to realise over the years, which I hope might be useful to others starting out... Never underestimate value of reading widely and deeply in and around your field. As well as improving quality of your own research, it helps you to contribute usefully to a range of research/policy discussions.

Quantitative analysis methods in the form of GIFs. A Sunday thread... Grid search

Below is correct. When teaching vaccine efficacy calculations, a key lesson is that if a test returns a lot of false positives, it will bias efficacy estimates downwards. Here's how to work out the impact…. 1/

https://twitter.com/whippletom/status/1588118265980764162

Typically, we define:

Vaccine efficacy = 1-(incidence in vaccinated group)/(incidence in unvaccinated group)

For brevity, let’s write this equation as

VE = 1-v/u

2/

A few science writing lessons I’ve picked up over the years… 🧵 Show don’t tell. If something is interesting or surprising, the reader should be interested or surprised by what they’ve read - they shouldn’t need to be told. en.m.wikipedia.org/wiki/Show,_don…