Relaxing UK COVID-19 control measures over the Christmas period will inevitably create more transmission risk. There are four main things that will influence just how risky it will be... 1/
We can think of as epidemic as a series of outbreaks within households, linked by transmission between households. This is particularly relevant over Christmas, given school holidays and some workplace closures. 2/
We can also think of R in terms of within and between household spread. If the average outbreak size in a household is H, and each infected person in household transmits to C other households on average, we can calculate the 'household' reproduction number as H x C. 3/
This brings us to the four main factors that will influence Christmas transmission risk. First, it depends on likelihood an outbreak will emerge in a household bubble. This depends on what people do in the 1-2 weeks before they meet others over Christmas... 4/
If 2% of people in a community are currently infected, and 10 random people get together, there's an 18% chance at least one will be infected. If 0.2% are currently infected (perhaps because they can take more precautions) this chance drops to 2%. 5/
Second, risk depends on how many people are in bubble. Per-contact transmission risk can be high in households (eg. cdc.gov/mmwr/volumes/6… & medrxiv.org/content/10.110…), so larger bubbles mean infection not only more likely to join, but also cause larger outbreak if it does. 6/
Third, it depends on how who is in a bubble. The riskiest bubbles will be ones that mix people from groups with a high chance of recent infection with people from groups with a high chance of severe disease if infected. 7/
Finally, it depends on whether a bubble outbreak leads to more outbreaks (i.e. how big the C is in H x C above). If people stick to exclusive bubbles, and are cautious in post-Christmas interactions, it will help reduce chance of one household cluster turning into many more. 8/8
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Some people are interpreting the below study as evidence that people who test positive without symptoms won't spread infection, but it's not quite that simple. A short thread on epidemic growth and timing of infections... 1/
If we assume most transmission comes from those who develop symptoms, there are 2 points where these people can test positive without having symptoms - early in their infection (before symptoms appear) & later, once symptoms resolved (curve below from: cmmid.github.io/topics/covid19…) 2/
So if people test positive without symptoms, are they more likely to be early in their infection or later? Well, it depends on the wider epidemic... 3/
Why do COVID-19 modelling groups typically produce ‘scenarios’ rather than long-term forecasts when exploring possible epidemic dynamics? A short thread... 1/
Coverage of modelling is often framed as if epidemics were weather - you make a prediction and then it happens or it doesn’t. But COVID-19 isn’t a storm. Behaviour and policy can change its path... 2/
This means that long-term COVID forecasts don’t really make sense, because it’s equivalent of treating future policy & behaviour like something to be predicted from afar (more in this piece by @reichlab & @cmyeaton: washingtonpost.com/outlook/2020/0…). 3/
Data take time to appear on gov website, so deaths for 1st Nov now average over 320, not “just over 200” as claimed in this article. Either CEBM team aren't aware of delays in death reporting, or they are & for some reason chose to quote too low values. cebm.net/covid-19/the-i… 1/
Worth noting models above were preliminary scenarios, not forecasts. (Personally, I thought there were more than enough data/trends to be concerned about last month, regardless of results of one specific long-term modelling scenario from early Oct.)
Why do data delays matter? Because it's difference between simplistic narrative of 'all model values were too high' and realisation that despite model variation, median of quoted worst-case estimates (376 deaths) concerningly close to Nov 1st average (which will rise further). 3/
Because SARS-CoV-2 testing often happens after symptoms appear, it's been difficult to estimate detection probability early in infection. So great to collaborate with team at @TheCrick & @ucl to tackle this question, with @HellewellJoel & @timwrussell – cmmid.github.io/topics/covid19… 1/
We analysed data from London front-line healthcare workers who'd been regularly tested and reported whether they had symptoms at point of test (medrxiv.org/content/10.110…)... 2/
To estimate when people were likely infected and hence detection probability over time, we combined the HCW data with a model of unobserved infection times. 3/
As epidemic trajectories in Europe rise back towards their spring peaks, a thread on normalisation during a pandemic... 1/
In late Feb & early March, we modelled scenarios for what effect widespread social distancing measures might have in UK. Like others, we'd already modelled contact tracing (thelancet.com/journals/langl…) but characteristics of infection suggested additional measures would be needed. 2/
At the time we were doing this modelling, fewer than 3000 COVID deaths had been reported globally (below from 1st March). I remember doing media during this period and could tell many saw the potential impact of COVID-19 as rather abstract, given observed numbers to date... 3/
If the reproduction number drops below 1, an epidemic won't disappear immediately. But how many additional infections will there be before it declines to very low levels? Fortunately there's a neat piece of maths that can help... 1/
Let's start by considering a situation where one person is infectious. On average they'll infect R others, who in turn will infect R others, and so on. We'd therefore get the following equation for average outbreak growth: 2/
If R is above 1, the outbreak will continue to grow and grow until something changes (control measures, accumulated immunity, seasonal effects etc.) But if R is below 1, above transmission process will decline & converge to a fixed value. (Proof here: en.wikipedia.org/wiki/Geometric…) 3/