The government’s mantra to justify tough lockdowns has been the need to save the NHS from being overwhelmed with Covid cases. The actual number of COVID hospital cases is tracked on the government’s dashboard, and the last update was from 26th November and states
Patients admitted Daily: 1369 Last 7 Days: 10415
I had assumed that these figures were members of the public who had contracted COVID in their everyday lives and then become so sick that they had to be hospitalised, like Boris Johnson. So I was surprised to see this tweet:
6. The data pic.twitter.com/Drs3Sv8OU7
— Statistics Guy Jon ? (@Jon_statistics) November 30, 2020
I decided to look into this. Digging down on the dashboard under “Health Care” we find there are actually 2 categories of hospital patients. 1. Patients admitted to hospital and 2. Patients already in hospital. This is confusing because at first sight 2. seems to be the total number of COVID patients within Hospitals. However these figures are all based on a very large spreadsheet reporting data from each NHS trust. In that spreadsheet it is explicitly clear that 2. is “Total number of inpatients diagnosed with COVID (Last 24h)”. So that means these patients were admitted to hospital for other reasons (stroke, cancer, injury etc.) and who have then contracted COVID in hospital !
I took the Total England NHS data from this spreadsheet and plotted it:
I get exactly the same result as @StatisticsGuy. This means that by far the majority of patients contracted “COVID” in hospital after being admitted for something else! The number of community infections leading to hospital admission is much less! Hospitals seems to be a risky place for patients getting infected by COVID.
Here is the ratio of the two groups of cases with time
Note how the ratio varies inversely to the infection rate.
The conclusion is that when infection rates in the community are low as during the summer, then cases are mostly community infections. However when the community rates are high then hospital infections dominate. That means that infections are probably being spread to inpatients within hospitals by staff, cleaners & visitors because they get infected, perhaps asymptomatically, within the community. Hospitals are often too warm with little or no ventilation and many patients are held together on a ward. It is easy to imagine how easily infections can spread in such an environment.
What about death statistics? Nearly all “COVID” deaths occur within hospitals. These are defined by PHE as deaths which occur within 28 days (4 weeks) of a positive test. So the question is : How many of these “hospital” cases would have died anyway from the underlying condition that put them in hospital in the first place? These are then so-called deaths with COVID. If we assume that half of them were deaths with COVID then that still has a dramatic effect on the overall death rate. It reduces it by ~40% during the first peak and by 33% during the second peak.
COVID Deaths in England by date of death. Last week still uncertain. Corrected is scaled as described.
This then implies that the number of deaths caused directly by COVID should be nearer 33,000 rather than the official figure of 51718 as of 1st December.
It also reduces the estimate for the Infection Fatality Rate (IFR) by 33%
If you are going to claim that COVID deaths are less than the official government numbers, then it would be relevant to simultaneously explain the excess death count which is _higher_ than official government numbers: https://www.bbc.com/news/health-54976362
The excess deaths are from the ONS and indeed show more deaths in the first peak than recorded but that is not true in the second peak. I think there are probably three separate effects. 1) Deaths due to COVID alone 2) Deaths due to a lack of access to healthcare and 3) Deaths from other causes where COVID was the final blow. The example of 3 were deaths in care homes in the first wave. I think 3 is not included in the NHS hospital trust figures. Under 2. we have
– Stopping certain cancer treatment eg. immunotherapy
– Late diagnosis for cancer
– reduced treatment for strokes and heart disease etc.
What I am looking at is just the number of cases of COVID in the public which led directly to hospitalisation.
This item from BBC news is relevant:
https://www.bbc.co.uk/news/health-55151832
HI Clive, thanks for the post. The more you look, the more the data seem a can of worms.
1) https://www.euromomo.eu/graphs-and-maps/ for England shows the z scores for excess deaths. In April there was a very high score. The high z score and the Covid deaths recorded correlate well. So it would seem reasonable to assume the 1st wave of Covid deaths was a reasonable estimate. This correlation does not exist for the ‘2nd wave’. Whilst excess deaths are slightly above the normal band, this does not match the reported magnitude of ‘Covid deaths’. This, I think, supports your contention.
2) Your posts on HIT (and those of others such as Nic Lewis and Gabriella Gomes) show that the HIT drops below (1-1/Ro) when there is heterogeneity in mixing, susceptibility, and also when susceptibility is seasonally modulated. All 3 factors must surely apply to Covid.
3) If we look at Euromomo for Sweden, the z score has been within the normal range since the summer (i.e. no ‘2nd wave’ present’). Sweden has allowed its citizens far more freedom than here. Importantly, schools were not closed.
I would suggest that the absence of the 2nd wave, suggests that Sweden has (on average) reached HIT for the current amount of intermixing. Clearly, there will be pockets in the country where this has not happened. In addition, as and when mobility of citizens returns to pre-covid levels, then the HIT will shift upwards. More infections would then occur until a new equilibrium is reached.
4) a recent paper (which I now can’t find!) reported the antibody ratios found in those under and over the age of 65. In Sweden, this was over 1.7:1. In other European countries, the ratio was generally <1. So Sweden's population 65), have received a greater exposure to Covid than in most other Euro countries. Presumably keeping schools open, must have contributed to this and the now apparent attainment of HIT in Sweden.
5) Referring to your recent post on ‘2nd wave’, if the acceleration of cases or deaths is examined for the UK, it is clear that these were decelerating (and following a Gompertz curve) from about 16 October 2020, in the UK and also in most of Europe. (see http://levitt2.herokuapp.com/) In the UK, this was well before lockdowns etc. In the absence of other interventions, the deceleration must surely indicate that the UK too is approaching HIT?
The initial acceleration of cases seemed to coincide with the start of University terms in September. I would suggest that these cases would have been far fewer if schools and Unis had opened earlier in the year because then the young would have developed immunity over the summer. The adverse effect of transmission to the rest of society would have been minimised because of seasonality.
6) There is much excitement about the roll out of a vaccine. However, if we have already reached HIT, how would we actually know if it’s been effective? Moreover, by the time a significant proportion of the population has been vaccinated, would natural immunity have rendered the vaccine unnecessary?
Hi Kevin,
Very interesting.
Herd immunity depends critically on R. What governments are doing with social distancing and lockdowns is to artificially reduce R so it can fool us into thinking that we are reaching herd immunity. So there is a “natural” value of R for say the UK under normal economic life. This was probably >2 so around 2.4 That implies herd immunity reaches 80% of the population infected before the virus disappears. Of course R reduces fast during this process. So if we assume 3-4 million people have had Covid (including asymptomatic) natural R has already reduced below 2. If some people already have immunity due to previous coronavirus infections then R0 is already reducing faster. The only way to find out would be to return to normal but no government is going to risk that.
A role out of mass vaccination will accelerate this process. If it can be done fast enough we should return to normal as soon as possible.
Hi Clive, thank you for your reply.
HIT only occurs at (1-1/Ro) for a population with homogenous connectivity and homogenous susceptibility.
Where there is heterogeneity in susceptibility, HIT is then defined as 1-(1/Ro)^(1+Cs^2).
Where there is heterogeneity in connectivity, HIT is defined as 1-(1/R0)^(1+2Cc^2) where Cc is the coeff of variability in connectivity and Cs is the coeff of variability for susceptibility.
Tkachenko et al published a paper on this and fitted their modified SEIR model to US data.
It seems that heterogeneity in connectivity is the key driver of HIT. For this Cv has a typical value (from data best fits) of around 2.5 prior to lockdowns and around 1.4 during lockdowns.
If these values are substituted into the equation, and using Ro=2.5, the HIT thresholds reduce to 11% (no lockdown) and 25% (during lockdown).
The paper uses much higher Ro values than 2.4 (typically 3-4) and they suggest likely HIT in the range 20-30%. I am pretty sure that Ro at the start of the outbreak in the UK was >3.
It is interesting that the Cc is higher when there is no lockdown. This, as I think is reasonably suggested in the paper, is because when lockdown occurs, the effect is disproportionately greater on the younger (working age) population who are inherently more mobile and more connected, than is is on the older (and more vulnerable) population. This suggests that population wide lockdowns are far from optimal in minimising mortality (vaccines excepted).
I have taken most of this information from Nic Lewis’ article https://www.nicholaslewis.org/why-herd-immunity-to-covid-19-is-reached-much-earlier-than-thought-update/. The footnote with the further update of 31 July contains a lot of important information.
It is clear that the Imperial and similar models used by SAGE do not properly account for heterogeneity, and this has serious implications for the management of the crisis – what the UK has done is vey sub optimal.
Kevin, has Nic Lewis updated his 10 May/31 July HI post now that Sweden has had a large recent upsurge in Covid-19 cases?
Not to my knowledge. But I would suggest referencing excess deaths rather than cases.
During the ‘1st wave’ there is a strong correlation between excess deaths and Covid deaths. This no longer applies to the ‘2nd wave’. Sweden has below normal excess deaths for this time of year. The cases appear to be, as described elsewhere, a ‘casedemic’. The cases are perhaps an artefact of mass PCR testing – something never done before. Cases are registered whether the case has symptoms or not.
The paper ‘SARS-CoV-2 waves in Europe:
A 2-stratum SEIRS model solution
Levan Djaparidze (E.E., B.C.S.), Federico Lois (B.C.S.) might be worth reading
Could you confirm that all of these “diagnosed in hospital” cases caught it inside hospital?
I know medical staff in Arizona and I know that most of *their* covid patients were admitted with unknown covid status, but were positive on the hospital entry test. They came in with covid but under your approach would be counted as having come in for something like respiratory distress and then having caught covid at the hospital.
Not sure how the NHS tracks this.
I can’t confirm it but here is another study that looked into this problem. In the NW of the UK as many as a quarter of all cases of COVID were acquired in health care settings.
https://www.cebm.net/covid-19/the-ongoing-problem-of-hospital-acquired-infections-across-the-uk/