Below analysis was two years ago (bbc.co.uk/news/health-51…). As well as providing an early warning about the COVID threat, it’s a good illustration of what is often an under-appreciated point: if we want to make sense of epidemic data and dynamics in real-time, we need models… 1/
At the time, only 41 cases of 2019-nCoV (aka COVID-19) had been reported in Wuhan. But two exported cases had just been detected in Thailand and one in Japan. How plausible was it that there were really just 41 cases in Wuhan? 2/
To answer this, we need to outline a model: if there are X cases in Wuhan, and travellers leave to different destinations at given rates, how likely is it we'd observe those three exported cases? With this model outlined, we can then use it to infer X given the observed data. 3/
This is what Imperial did, estimating that those 3 exported infections were consistent with 1723 cases (95% CI: 427 – 4471) in Wuhan imperial.ac.uk/mrc-global-inf… 4/
Now, of course, it's well known there were loads of infections not appearing the data early on (thelancet.com/journals/lance… & thelancet.com/journals/lanpu…). Everything is obvious in hindsight etc. But at the time this modelling went against raw data, so valuable situational awareness. 5/
I think above is a useful example of real-time analysis because 1) it's pretty intuitive why you'd need a model, and 2) shows these approaches can provide crucial early insights that wouldn't have been possible by just looking directly at the (noisy, biased, incomplete) data. 6/
Others would later apply similar methods to estimate cases elsewhere (e.g. ncbi.nlm.nih.gov/labs/pmc/artic…). And we'd use exported cases and evaluation flights to help estimate changing transmission in Wuhan as measures came in: thelancet.com/article/S1473-… 7/
Although some use modelling of future scenarios as a synonym for 'modelling', important to remember a lot of the modelling work during COVID has focused on very different questions:
Suggestions to include spontaneous behavioural responses to a growing epidemic as default in scenario models might sound simple and obvious, but it's worth considering the assumptions involved... 1/
What magnitude of behaviour change in response to epidemic would you assume as a baseline scenario? And how would this relate to transmission reduction? 2/
Would you assume disruptive-but-ineffective change (e.g. avoiding low risk interactions and continuing to have high risk ones) or well-targeted change (e.g. using more rapid tests effectively)? 3/
Given below observation, an obvious question to ask is 'did any models consider a scenario that could give insight into what a Tier 4-like change in behaviour could look like?' 1/
If we look at scenarios that assume a prolonged return to behaviour levels seen in earlier steps of the reopening in England (e.g. Step 0, S1, S2 etc. below from 11 Dec: cmmid.github.io/topics/covid19…), models predict far lower levels of hospitalisations and deaths. 2/
Could summarise above scenarios as: 'If you want epidemic curve to look like that, this is the sort of virus characteristics and behaviour change that would be required - whether behaviour change comes from a policy, or pin hopes on spontaneous individual-level change' 3/
Some thoughts on the evolutionary trajectory of SARS-CoV-2 so far, how it compares to other viruses, and what might happen next... 1/
For seasonal coronaviruses & flu, we see a pattern of 'antigenic turnover' over time - circulating viruses give rise to new variants that escape prior immunity against infection, immunity builds against these new variants, then these in turn spawn new variants... 2/
When infections evolve to escape immunity like this, we typically end up with an evolutionary tree that looks like a lopsided ladder as new variants sequentially replace their 'parent' variant lineages (below from: journals.plos.org/ploscompbiol/a…) 3/
Flu is often quoted as an example of 'living alongside' an infection (even though countries work hard to reduce disease burden via vaccination). But I think sexually transmitted infections are also an important example to consider... 1/
For STIs, preaching abstinence to prevent infection has gradually fallen out of favour (e.g. thelancet.com/journals/langl…). Instead, societies deal with ongoing circulation of these infections using other tools... 2/
Testing & informing partners gives people information on their infection status (and treatment can reduce duration of infectiousness). Contacts can be made less risky with protection (e.g. condoms). And susceptibility can be reduced through vaccination (or, for HIV, PrEP)... 3/
Unfortunately, this oft-quoted Spectator tracker of COVID 'scenarios vs outcomes' seems at best muddled and at worst actively misleading. A thread on some weird comparisons - and how to do better critiques... data.spectator.co.uk/category/sage-… 1/
First plot is comparison of scenarios for increased R values with later data. Crucially, these weren't predictions about what R would be (R estimate that week was 0.9-1.1, so pretty flat epidemic). Rather, report showed what could happen if R increased beyond current range... 2/
Not sure why they cut out the R=2 scenario from original plot (which would've made it obvious these were illustrative - assets.publishing.service.gov.uk/government/upl…). TBF SPI-M plot could have included horizontal line to illustrate what R=1 looks like, but don't need a model to draw a flat line... 3/
Any writer who claims 'UK models always overestimate COVID numbers' is either lazy or trying to mislead their readers. A few illustrations... 1/
A surprisingly under-reported example is early UK scenario models (although @anthonybmasters & @d_spiegel correctly flag it in 'Covid by Numbers'). Some point estimates from initial 'recurrent lockdown' scenarios were lower than what actually occurred:
Then there’s the July 2020 report from @acmedsci, which used Imperial modelling. It was designed to be a plausible worst case scenario, not a forecast, but ended up strikingly close to reality (with Alpha increasing risk & vaccines decreasing it): 3/