If it was established in India that for every of 100 road traffic accident cases there was one road traffic accident death or 1% .
And in three months there was 37,000 road traffic accident deaths.
How many road traffic accident cases would there be?
The problem that you are working on is, "How many people have unknowingly had covid-19 and survived." I think the corresponding question should read "How many people have escaped undocumented road traffic accidents without dying."
For a while, I tried to relate the number of daily deaths to the daily cases 22 days prior - a crude attempt to get to a survival rate based on average diagnosis to death period. It would only have related to documented cases.
As far as I can see, we are stuck with predicted prevalences at different moments in time. Putting aside for the moment that these predictions are shot through with false positives, we might suggest that a decrease in prevalence represents people emerging from an infection. And conversely, an increase in prevalence represents people acquiring an infection.
This is also something that I have been looking at daily. That is, I estimate the chances of already having the infection today and the chances of acquiring it today. For example,my spreadsheet shows that on 22 September (using 7-day averages), the chances of having the infection was 1 in 49 and the risk of acquiring it was 1 in 1,014 as the prevalence (assuming positives all true) was going up.
I did not look at the chances of emerging from the infection as prevalence went down. Perhaps I should have done this - and perhaps you are one step ahead and this is your plan? Are you integrating this effect in some way? I will look through your last post again. My mind is a bit slow these days.
Ok
I am going to bed soon and going to have a busy weekend
I think the ONS data is OK
I did put a logical caveat on this a few posts ago.
My prediction of 40% or whatever is the number of people that would have failed an ONS PCR test at some point from March to August or whatever.
The calculation fortunately and perhaps surprisingly doesn’t require the number of deaths to be accurate.
Because it involves
Deaths(total)/Deaths (april 26)
If one is 50% to high so will the other so Deaths(total)/Deaths (april 26) will be the same. hopefully.
The deaths and test thing not being synchronised is going to cause people not used to doing thus kind of stuff difficulty.
Eg you don’t become a case and die on the same day.
But that kind of thing irons or averages itself out.
However there is definitely something mystifying about this in that there should be a lag phase between the covid case values and the covid death values of at least 10 days , according to the doctors.
And there isn’t; the peak cases date is very close to the peak deaths.
It is just possible that people were being designated as cases very late and the testing by specimen date peak although messy was also very close to peak deaths curve.
Testing by spacimen date graph needs a statistical clean up.
I think it must be because either the fatality rate on the people infected earlier as the most susceptible was higher and thus it shunts the death curve to the left to make it overlap with the cases curve.
But I have often wondered if people are dying of something else eg influenza X
And that is peaking undetected and measured before and to the left of deaths.
And covid is an opportunist infective agent that turns up later and hence it cases peak overlaps with deaths.
It would explain a lot of things like asymptomatic cases.
I am not intergrating; I am deriving the case fatality rate from
The number of cases on a day from the ONS data.
And the number of deaths on the corresponding day.
You can do it over a range of days, or different days and you get the same very similar result which is reassuring because a change in a case fatality rate would mess things up a bit.
So
if the case fatality ratio was 1 death per 100 cases.
And there were 37,000 deaths over 3 months
There would have to have been 37,000 x 100 cases = 3.7 million.
A hundred times more cases than deaths.
Which is why are wanted you to do the road traffic calculation.
So if the case fatality ratio was 1 death per 1000 cases.
37,000 x 1000 cases = 37 million.
So if the case fatality ratio was 1 death per 750 cases.
37,000 x 750 cases = 28 million.
That 1/750 is 0.133% and is getting close to John Ionnadis figure.
The different and unknown number of “nosocomial” cases outside the community, particularly on April 26th, isn’t as difficult to estimate as it sounds.
I was sorting of pitching it high as a population of 1 million with a cases per 1000 at 10 times the rest of the community.
Which I think is OK for that date.
But any suggestions would be welcome.
That would set the upper level at around 45% of population.
Hello again Dave. I am not sure I follow what you are doing, so I will suggest one way to have a go at determining the total number of infections that have occurred at any time. I am not proposing this with any confidence because there is so much noise and so many unknowns. Otherwise, it probably would be the right way to go about it. Anyway, as my father used to say, it's better than a kick in the pants.
The following figure should explain it all in the notes:
It all hinges around the change in case numbers to represent onset of infections and passing of infections. It's full of noise (like the author) and I have hesitatingly picked a start date where the tested population might represent the general population.
The problem is, of course, that there are wild swings when the pandemic was on but it only applied to a very small minority. Extrapolating to the wider population is way off the mark. So we only have the little noisy swings to go by. So it's all swings and roundabouts, really.
And I will be back roundabout 10pm while I grab a life.
Ok
Just a brief update I have analysed the OND data in more detail and it is quite interesting.
It does look like they have [had] a very low false positive rate and that could be good as it will may be useful in exposing the NHE pillar 2 false positive rate.
Spent several hours today on a walk in the peak district talking to an old degree level maths computer programmer running the idea past her to see if I have trodden on any landmines,
She thought it was good and wants to see the data.
She came up with the idea of; as the ONS data start of the 26th April was approximately the median rather than the mode with it being a Poisson and non Gaussian and Normal distribution.
[These epidemiologist have another non mathematical name for it ]
You could just double the post 26th April ONS case data to get the total community cases.
The bitch !
I will try that though
These epidemiologist have another non mathematical name for it
https://en.wikipedia.org/wiki/Poisson_distribution
https://en.wikipedia.org/wiki/Normal_distribution
The case fatality rate of the ONS data stays pretty constant from 26th april to 14 of june.
As the cases percent of population dropped from 0.45 to 0.05%.
if there was an approximate 0.025% false positive rate the real 26th case fatality rate would have been about 5% lower or 5% higher than it should have been.
Anyway be June 14th a fixed error of 0.025% would have had a dramatic and noticeable affect on a value of 0.05%, which it didn’t.
So their false positive rate, then, was probably less than 0.01%..
After June 14th my model starts to go a bit pear shaped not that it matters much as it is all over by then.
It was extremely vulnerable at that point to statistical noise but it did look like a trend.
The death rate appeared to be have dropped by about 20-40%.
You check and do it for yourselves please!
I have not gone much past June 14th ; I have learnt over the years just ignore crap noisy data as it can send you down rabbit holes of cognitive bias.
Then I remembered this; I was following it at the time.
She suggested an ONS cases versus PHE deaths graph from excel
That should be a straight line with the gradient = to the fixed case fatality rate
And we can have a correlation coefficient done on it.
And extrapolate it out of its data set to 37,250.
And if they do correlation is not causation tell them to ……
I thought testing was fairly relatively stable over april
So a cases versus tests graph should and did mirror the cases curve which it seems to.
I think a delta cases and a delta cases /test curve should have measured the steepness of the curve which it doesn’t seem too?
[PHE Cases/ tests ]/ ONS “prevalence”
Might be interesting as change in PHE sample selection bias ?
I will have to think about that just spent too much time in a Manchester speakeasy and a bit pissed.
