Thank you for holding the
#Lockdown!
So important, as ~1 in 11 UK people (6.4 million) might have #COVID19.
That would mean there are infected people in every park, in every shop, in every street. It might include you!
Thread w/calculations if interested:

So important, as ~1 in 11 UK people (6.4 million) might have #COVID19.
That would mean there are infected people in every park, in every shop, in every street. It might include you!
Thread w/calculations if interested:
Assume: 9,239 deaths (hospital deaths + 50% for lag + non hosp)
Infection (not case) Fatality Rate : 1%
Adj. days infection to death ~18
So ~914k infected up to 18 days ago
~6 days to double so ~3 doublings since
So infections to date: ~7.3m, ~13% recovered so ~6.4m now
Infection (not case) Fatality Rate : 1%
Adj. days infection to death ~18
So ~914k infected up to 18 days ago
~6 days to double so ~3 doublings since
So infections to date: ~7.3m, ~13% recovered so ~6.4m now
Model is v.sensitive to inputs, so low confidence + wide ranges apply.
More deaths --> more infections
Higher IFR% --> fewer infs (don't need as many for same deaths)
Longer days to death --> more infs (more time to double)
Higher doubling rate --> more infs
More deaths --> more infections
Higher IFR% --> fewer infs (don't need as many for same deaths)
Longer days to death --> more infs (more time to double)
Higher doubling rate --> more infs
This is a simple, top-down model.
For a pro version that also uses a "death, days to death, transmission speed" approach, read this from Imperial (1.8m).
V.low given deaths/growth since + non-hospital + reporting lag additions.
http://imperial.ac.uk/news/196556/coronavirus-measures-have-already-averted-120000/
h/t @StefanFSchubert
For a pro version that also uses a "death, days to death, transmission speed" approach, read this from Imperial (1.8m).
V.low given deaths/growth since + non-hospital + reporting lag additions.
http://imperial.ac.uk/news/196556/coronavirus-measures-have-already-averted-120000/
h/t @StefanFSchubert
Another interesting approach via symptom tracking.
https://covid.joinzoe.com/post/how-we-are-predicting-covid-in-the-uk
Estimates 1.9m, but need to add 0-20 and over 69 years infected, all asymptomatic infected and also apply growth (double?) in last 6 days.
@Join_ZOE @timspector @KingsCollegeLon
https://covid.joinzoe.com/post/how-we-are-predicting-covid-in-the-uk
Estimates 1.9m, but need to add 0-20 and over 69 years infected, all asymptomatic infected and also apply growth (double?) in last 6 days.
@Join_ZOE @timspector @KingsCollegeLon
It's heartening to see more press references to % infected, not just the "confirmed cases" iceberg tip.
https://unherd.com/2020/04/how-likely-are-you-to-die-of-coronavirus/ @TomChivers
https://news.sky.com/story/coronavirus-the-data-suggests-the-uk-is-on-course-for-many-thousands-of-deaths-11966517
https://telegraph.co.uk/news/2020/03/29/least-16m-people-could-infected-coronavirus-uk-new-estimates/
https://metro.co.uk/2020/03/30/coronavirus-slowing-uk-thanks-social-distancing-measures-12476442/
https://ftalphaville.ft.com/2020/04/04/1586015208000/Imperial-s-Neil-Ferguson---We-don-t-have-a-clear-exit-strategy-/ @jemimajoanna
https://unherd.com/2020/04/how-likely-are-you-to-die-of-coronavirus/ @TomChivers
https://news.sky.com/story/coronavirus-the-data-suggests-the-uk-is-on-course-for-many-thousands-of-deaths-11966517
https://telegraph.co.uk/news/2020/03/29/least-16m-people-could-infected-coronavirus-uk-new-estimates/
https://metro.co.uk/2020/03/30/coronavirus-slowing-uk-thanks-social-distancing-measures-12476442/
https://ftalphaville.ft.com/2020/04/04/1586015208000/Imperial-s-Neil-Ferguson---We-don-t-have-a-clear-exit-strategy-/ @jemimajoanna
Illustration implies total UK death toll of ~73k...
But only if IFR is 1% + we see zero new infections.
If our ICU capacity is overrun, the IFR% will spike hard.
New infections depends on hard #Lockdown + test/trace.
Similar to: http://covid19.healthdata.org/united-kingdom @IHME_UW
But only if IFR is 1% + we see zero new infections.
If our ICU capacity is overrun, the IFR% will spike hard.
New infections depends on hard #Lockdown + test/trace.
Similar to: http://covid19.healthdata.org/united-kingdom @IHME_UW
A wave is crashing over our ICUs.
We can't reduce the size of that wave because it's caused by infections from 2-5 weeks ago.
We can improve our readiness by scaling ICUs + protecting NHS staff.
We can stop the wave getting bigger + longer by hard #Lockdown + test/trace/quar.
We can't reduce the size of that wave because it's caused by infections from 2-5 weeks ago.
We can improve our readiness by scaling ICUs + protecting NHS staff.
We can stop the wave getting bigger + longer by hard #Lockdown + test/trace/quar.
It's tempting to think... "There's 9,239 deaths and #Covid19 kills about 1 in 100, so there must be ~923k infections".
The problem is you're nearly right. But those ~923k infections were ~3 weeks ago. That's how long it takes to kill.
They've been doubling every ~5-6 days since
The problem is you're nearly right. But those ~923k infections were ~3 weeks ago. That's how long it takes to kill.
They've been doubling every ~5-6 days since
This isn't the best way of estimating % infected!
We're waiting for people to die, assuming fatality rate to judge how many infections drove those deaths, then using doubling rates to judge infections today.
A better way? Random pop. sample testing! https://twitter.com/JamieWoodhouse/status/1247187400876580864?s=20
We're waiting for people to die, assuming fatality rate to judge how many infections drove those deaths, then using doubling rates to judge infections today.
A better way? Random pop. sample testing! https://twitter.com/JamieWoodhouse/status/1247187400876580864?s=20
Why is estimating # infected more important than just tested "cases"?
Infected people are doing the infecting. The "cases" are in hospital/dead/immune
If you know # infected and # deaths you can assess how deadly this thing really is (IFR%, not just "case" fatality)



BJ's speech should have started:
"There are infected people in every train, in every shop, in every park, in every tube... you might even be one of them... please take what I am about to say very seriously..."
This poll shows a dangerous level of complacency across the UK. I suspect this is worse in the real world.
Most people think less than a million are infected! That's what happens when govt + press only talk about the "confirmed cases" iceberg tip. https://twitter.com/JamieWoodhouse/status/1245291470107672576?s=20
Most people think less than a million are infected! That's what happens when govt + press only talk about the "confirmed cases" iceberg tip. https://twitter.com/JamieWoodhouse/status/1245291470107672576?s=20

If it is, #Covid19 will kill much more than 1%. When you run out of ICU beds / ventilators, many more die as in Italy/Spain.
If we'd started projecting total infected from day one (ideally via. regular pop. sample tests), we might have acted sooner.
We might also have persuaded people to comply better with our weak #Lockdown.
Instead... "a national scandal" @richardhorton1 https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30727-3/fulltext
We might also have persuaded people to comply better with our weak #Lockdown.
Instead... "a national scandal" @richardhorton1 https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30727-3/fulltext
A powerful visualisation of deadly UK #Covid19 complacency:
"Right things... at the right time..."?!?
H/t @Imperial_JIDEA
"Right things... at the right time..."?!?
H/t @Imperial_JIDEA
Amateur epidemiologists like me are annoying, but as long as we have pro epidemiologists saying things like:
- 1000x deaths or
- "2-5x cases" or
- 734k or
- 50%+ or
- "We have no way of knowing" (we do - pop sample tests!)
It feels useful to have a top down sanity check?
- 1000x deaths or
- "2-5x cases" or
- 734k or
- 50%+ or
- "We have no way of knowing" (we do - pop sample tests!)
It feels useful to have a top down sanity check?
Thanks to data journalists grappling with this critical estimate - how many are really infected.
@TomChivers @EdConwaySky @alexwickham @Ashley_J_Kirk @PaulNuki @AlbertoNardelli @nicholascecil @jburnmurdoch @jemimajoanna
Wish you were asking questions at the daily pressers!
@TomChivers @EdConwaySky @alexwickham @Ashley_J_Kirk @PaulNuki @AlbertoNardelli @nicholascecil @jburnmurdoch @jemimajoanna
Wish you were asking questions at the daily pressers!
+ Final thanks as ever to @cheianov who helped improve the maths.
Particularly modelling an adjusted "infected to death" duration given exponential skew over time.
On average, deaths to date are caused more by recent deaths than the 23.5 days from infection to death implies.
Particularly modelling an adjusted "infected to death" duration given exponential skew over time.
On average, deaths to date are caused more by recent deaths than the 23.5 days from infection to death implies.