Read on Twitter
Vaccine Impact As of February 6, deaths in Over80s have reduced 29.38% faster than Under80s, and it's diverging fast
The tipping point was January 17, when 50% of Over 80s had been vaccinated with a first dose
There will now be increasing evidence of impact on deaths
I'm absolutely certain that deaths in Over80s, (within 28 days of a positive test), are diverging and the only plausible explanation is the vaccines
I can't say whether it's saving lives or only prolonging them, but it is a *compelling* signal
You heard it here first folks!
I can't say whether it's saving lives or only prolonging them, but it is a *compelling* signal
You heard it here first folks!
For context, this is the same chart from two days ago...
It's diverged another ~10% in that time, so it's moving much faster than I expected and that's why I say it's *compelling* evidence
It should become really obvious over the next 7-10 days https://twitter.com/PsyberAttack/status/1359275170406555657
It's diverged another ~10% in that time, so it's moving much faster than I expected and that's why I say it's *compelling* evidence
It should become really obvious over the next 7-10 days https://twitter.com/PsyberAttack/status/1359275170406555657
The trend continues nicely...
Let me briefly explain what this is and why I believe it's compelling early evidence of vaccine impact
Forgive the manually drawn trendlines in the next tweets. I'm just more autistic than artistic, so DataViz is not my thing
Let me briefly explain what this is and why I believe it's compelling early evidence of vaccine impact
Forgive the manually drawn trendlines in the next tweets. I'm just more autistic than artistic, so DataViz is not my thing
The charted "ratio" comes from time-shifting deaths to match their peaks, (Under80s +3 days), and then calculating relative daily percentages of peak for Under/Over80s throughout the 2nd wave
A stable trendline that's only diverged +5 across 4 months now diverges -10 in 5 weeks!
A stable trendline that's only diverged +5 across 4 months now diverges -10 in 5 weeks!
The tipping point appears to be January 17, with an even larger impact of -15 divergence over 4 weeks
That risks overly fitting data to a hypothesis, but the trend across the whole 2nd wave points to the UK's one-dose vaccine rollout reducing deaths, or at least prolonging life
That risks overly fitting data to a hypothesis, but the trend across the whole 2nd wave points to the UK's one-dose vaccine rollout reducing deaths, or at least prolonging life
I checked back to last March, and we only saw this divergence before we had mass testing, and it's implausible that similar ratios of elderly are dying untested now
I'll release the code for the model in the next days, I just need to refactor variable names and add some comments
I'll release the code for the model in the next days, I just need to refactor variable names and add some comments
The ratio of deaths in Under/Over80s has converged for the 2nd day. A convergance of 19 causes a breach of the lower boundline, resting at -8.3
There's still a strong divergent trend from Jan 1, with today's divergence levels unseen since mass testing became readily available
There's still a strong divergent trend from Jan 1, with today's divergence levels unseen since mass testing became readily available
Divergence increased 9 today, breaching the lower bound to -17.5
A pattern is forming, and today I could reliably forecast the expected divergence from the waveform
Of course, divergence without context is meaninless, so I've illustrated the major crises periods for comparison
A pattern is forming, and today I could reliably forecast the expected divergence from the waveform
Of course, divergence without context is meaninless, so I've illustrated the major crises periods for comparison
The badly drawn arrows are pointing to a divergence of ~-20, which is what around what I was expecting, but more importantly they obscure what appears to be a meaningful signal in the wave form
I'll show you what I mean in the next tweet
I'll show you what I mean in the next tweet
The last 2 weeks of the waveform have a clearly discernible pattern, something any sound engineer would recognise
The data represents upto Tueday, Feb 9, and the noise in the waveform is settling into a clearly defined pattern of weekly reporting days that illustrate a trend
The data represents upto Tueday, Feb 9, and the noise in the waveform is settling into a clearly defined pattern of weekly reporting days that illustrate a trend
As we now have a demonstrable signal, that appears to be predictable, let's macro out so you can see the "Divergence" I've been illustrating in real-world data
It's there. It might be predictable. And over the next 10 days it'll become pronounced, even on this format of chart
It's there. It might be predictable. And over the next 10 days it'll become pronounced, even on this format of chart
To illustrate those charts are illustrating the same data, today's "divergence" is -17.47
Under80s are timeshifted forward 3 days to align, and latest data shows they've remained static, whilst Over80s tend to suggests cumulative decline over next 3 days and increased divergence
Under80s are timeshifted forward 3 days to align, and latest data shows they've remained static, whilst Over80s tend to suggests cumulative decline over next 3 days and increased divergence
A small but expected convergence of 2.45 today, moves divergence to -14.82 and ensures it remains in breach of the lower bound
The trend remains downward, but also quite modest. The next week of data will be critical because it's now 4 weeks since 50% of Over80s were vaccinated
The trend remains downward, but also quite modest. The next week of data will be critical because it's now 4 weeks since 50% of Over80s were vaccinated
The waveform appears to be following the pattern of the 2 prior weeks, possibly reporting cycles, which suggests there'll be further divergence over the next days
Divergence may also be affected by 70+ being vaccinated, but there's no clear way points to define effect window yet
Divergence may also be affected by 70+ being vaccinated, but there's no clear way points to define effect window yet
Interestingly there's almost zero divergence between cases in Under/Over 80s, they're falling in exact proportion
The data is lagged by 5 days, so it's not covering the "impact period" of 4 weeks after first dose yet, (but then we can compare results to one week after 2nd dose).
The data is lagged by 5 days, so it's not covering the "impact period" of 4 weeks after first dose yet, (but then we can compare results to one week after 2nd dose).
Deaths diverge a lot compared to cases, but so far a vaccine effect is noticeable, but it's early and modest
My *VERY SHAKEY* fag packet maths quantifies the impact from divergence at 1,106 lives saved over 23 days
So far, the lockdown has had an immeasurably larger impact...
My *VERY SHAKEY* fag packet maths quantifies the impact from divergence at 1,106 lives saved over 23 days
So far, the lockdown has had an immeasurably larger impact...
Deaths in Under/Over80s diverged again, increasing 4.79 to -18.72. The pattern is for increased divergence today and tomorrow, although today's divergence was more modest than expected and did not surpass -20
The trend remains, and I expect there'll be more divergence tomorrow
The trend remains, and I expect there'll be more divergence tomorrow
The divergence in the previous chart is the measurement of the difference between the red/blue lines, and the divergence is now clearly visible on the chart below
However, recent rate of divergence has stalled, so I will make some changes and test against Under/Over75s later
However, recent rate of divergence has stalled, so I will make some changes and test against Under/Over75s later
Notably, the trend in cases for Under/Over80s still shows barely any sign of divergence, which suggests survival rates in Over80s probably are increasing relative to Under80s
England's vaccine program doesn't appear to be as impactful as Israel's yet, but it's too early to judge
England's vaccine program doesn't appear to be as impactful as Israel's yet, but it's too early to judge
Looking at the delivered vaccine doses, and assuming 50% vaccine efficacy, this should be the trend over the next 10 days, (down to -40 by Feb 21 on the chart)
It's just hard to see how this trend isn't going to continue to diverge strongly, especially if efficacy is higher
It's just hard to see how this trend isn't going to continue to diverge strongly, especially if efficacy is higher
Something unexpected today. A small convergence of 2.03 makes a divergence of -15.28
So deaths in Over80s have now decreased 15.28% more from their peak than Under80s over the same period
It might be a blip, but the trend of their divergence from Feb 1 is also quite negligible
So deaths in Over80s have now decreased 15.28% more from their peak than Under80s over the same period
It might be a blip, but the trend of their divergence from Feb 1 is also quite negligible
Needs more data, but the expected waveform trend has already broken...and we're now well into the window where divergence *should* be developing...so that's a bit unexpected
The divergence is still comparatively weak, and the Over 75s don't compensate it https://twitter.com/PsyberAttack/status/1361392554126770177
The divergence is still comparatively weak, and the Over 75s don't compensate it https://twitter.com/PsyberAttack/status/1361392554126770177
Worth adding that cases have remained very stable within their boundlines during the vaccination period, so it is unlikely but *plausible* that the "issue" here is just not enough people are being exposed to the virus to accurately measure the impact of vaccines at the moment 
Adding this to the thread as it should give a better idea of what to expect from impact and divergence...
The claims of 89% efficacy from 1 dose, by interpolating results from 14 days after first dose, are starting to look a bit shakey in real-world data from England
The claims of 89% efficacy from 1 dose, by interpolating results from 14 days after first dose, are starting to look a bit shakey in real-world data from England
Adding this data from Israel for extra context. If you reach hard enough you can argue the impact becomes visible 5 weeks after first dose, so Jan 10 cohort, (36.53% of Over80s), should be showing by Saturday, (compensating for delay in demographics data) https://twitter.com/segal_eran/status/1356313705684869121/
And in case anyone's wondering, this is why we have to examine data from the UK and not just assume it's worked in Israel so it'll work here
The UK has taken a punt on a delayed dose strategy, so we need clear evidence from the UK that it's paying off before talking of reopening
The UK has taken a punt on a delayed dose strategy, so we need clear evidence from the UK that it's paying off before talking of reopening
The thread states "'vaccinated' means one week after a second dose", (so 28 days after 1st dose in England), and now we've got a solid baseline to test against!
Spolier: if we were even getting close to this, our Over80s signal would be a lot stronger
https://twitter.com/Nadav_Eyal/status/1362416551060774913
Spolier: if we were even getting close to this, our Over80s signal would be a lot stronger
https://twitter.com/Nadav_Eyal/status/1362416551060774913
Recent divergence in relative rates deaths of Over80s appears to have stalled
There's no convergence, despite the backfilling of deaths as more data is added to the dataset, but the lack of a continued divergent trend is a bit unexpected given the increasing vaccinated cohort
There's no convergence, despite the backfilling of deaths as more data is added to the dataset, but the lack of a continued divergent trend is a bit unexpected given the increasing vaccinated cohort
There remains no apparent signal whatsoever in cases. What *might* have been the start of a trend a few days ago has petered out...
Tomorrow's data will show a full 28 days since 59.42% of Over80s have had at least one dose of a vaccine, so the impact *should* be visible already
Tomorrow's data will show a full 28 days since 59.42% of Over80s have had at least one dose of a vaccine, so the impact *should* be visible already
