COVID-19 numbers, step by step

20 July 2020 by Jennifer

What do we know about covid numbers? And how do we know? Key points, plus explanations in common-sense terms. (Most examples from England or UK.)

Doctors and science researchers are learning more every day about covid and the virus which causes it.1 Out of my own wish to understand, I’ve been reading science papers and news stories, and following scientists discussing among themselves on Twitter.

Often, it seems to take a while for that info to get out into everyday knowledge. (Plus, in the UK, the government keeps saying different things.) I thought I’d try to bridge the information gap, and do some plain-English explanations.

Today’s theme is numbers and measurements – e.g. deaths, cases, testing. I’ll talk partly about the actual numbers, and partly about how we try to find them out, and the things we don’t know yet. And then I’ve got this to refer back to if I write other things which draw on those numbers.

It got long, so I don’t necessarily suggest you read it all:

  • Just want a quick overview? See the key points section.
  • More explanation? Don’t like statistics but do want to understand? Jump from a key point to the particular topic you’re interested in, or browse through.
  • Don’t trust it until you’ve checked the sources? (wise!) Footnotes give more details of how I got to these conclusions.

(Content note for the rest of it: plain talk about death. Feel free to come back another day if you’re not in the mood.)

Covid-19 Numbers Step By Step

Key points covered

  • The measurements of deaths and known infections don’t tell us what infections are happening right now today. All the measurements have a bit of a time-lag.
  • So far this year – mostly during April and May – about 67 to 69 thousand people in the UK have died that we wouldn’t normally expect at that time of year. This is called the “Excess deaths“.
  • About 45 thousand of those were of people who’d tested positive for covid, who are counted in the official numbers from the Department of Health and Social Care (DHSC).(There’s also a separate count by the Office of National Statistics, based on death certificates; I’ll get into the differences.)
  • Excess deathsdoesn’t measure exactly the same thing as “official number of deaths from covid”. But it can be more useful for comparing, because it doesn’t depend on side questions, such as how many people got tested, or what exactly you count as “a death from covid”.
  • You can’t automatically assume that the unexpected deaths were all covid. But for other reasons, we can deduce that these ones probably mostly were. So back in the spring, there were probably about 20 thousand deaths in the UK which didn’t get written down as covid, but actually were. The official covid deaths from the DHSC at the moment won’t include this other 20 thousand.
  • The official figure for the total number of cases (about 294 thousand so far, at 19 July) should really be thought of ascases that we know about“. There are (and were) probably still a lot of infections that we don’t know about.
  • The COVID-19 Symptom Study is a good source of estimates for how many UK people in the 20 to 69 age range currently have covid symptoms. That estimate is currently being updated every day. (It doesn’t include care homes, or people outside those age ranges, or people who have the virus without symptoms.)
  • There have been attempts to work out how many people have had the virus so far. We don’t yet have a good way to find that out; I’ll talk about some guesses.
  • Related to that question, it’s important not to assume that having the virus once gives you immunity forever. We don’t know yet whether that’s the case; it’s probably more likely that immunity wears off over time.
  • If two areas are doing equally well at protecting people from the virus, one might still have more deaths than the other, simply because more people live there. So sometimes, instead of comparing the total for the whole country, it can make more sense to compare same-size chunks. Typically, you’d look at how many people died unexpectedly out of each million people in the area. This would be called “excess deaths per million people“.
  • Based on all the numbers discussed here, it looks as though in the year so far, up to 19 July, the UK has had something in the region of 800 to 1,000 excess deaths out of every million people in the country. So far, this is similar to Spain, a little bit worse than Italy, about twice as bad as France, and much worse than Germany, Greece or Denmark.
  • Of every 200 people who gets infected with covid, at the moment it looks as though typically 1 or 2 would die. The exact number isn’t known yet. However, that doesn’t mean that any random person has the same risk; for example, older people are more likely to die of it than younger people.
  • We don’t know yet how many previously-healthy people will remain long-term disabled as a result of a covid illness, such as a stroke, damaged lungs, or post-viral fatigue.
  • Because the illness is very new, death risks and disability risks are both likely to improve over the months, as doctors & researchers learn more about the best treatments.
  • The “R” number is way of summing up roughly how successfully the virus is managing to spread around. It’s how many other people would typically be infected by 1 infected person. For a particular virus, it changes over time, depending on what’s going on with the humans.
  • Another way to talk about the epidemic growing or shrinking is the “growth rate“. For example, if 1,000 people caught the virus yesterday, and 1020 people catch it newly today, that would be a “growth rate” of 2% (two per cent). Or if 980 people catch it newly today, that would be a “growth rate” of -2% (minus two per cent).
  • At the moment, the UK‘s epidemic is probably about holding steady, with deaths around 100 a day, mostly in England. Scotland is doing better than England; they’re down to about 1 death a week at the moment.
  • “Pillars 1 & 2” of the English government’s testing plan both relate to “who’s got the virus now” type testing, where you want the answer back as quick as possible. Pillar 1 is via labs in the National Health Service and Public Health England. Pillar 2 is the privatised services, e.g. drive-in testing.Some useful measurements to track would be:
    • How quickly can people get a test if they want one?
    • How soon do they get the results?
    • If 100 people get tested, how many results come back positive? (The idea is that if it’s more than a handful, there are probably a lot more infected people you haven’t found yet, and you should be aiming to do more tests.)

In the rest of this article, I shan’t bring in much different stuff – I’ll just spell out the reasoning in a more step-by-step way, and give links to find out more.

Simple example of a tricky question

The first thing I want to explain is: with statistics, some tricky questions never go away however expert you are. I would even say that the art of understanding stats is largely the understanding of all the many many ways they can get wiggly :-)

Here’s an example: if you’re counting apples in your kitchen, and you’ve got four apples, but one is half manky already… would you count the half-manky one, so you’d say you had 4? or not, so you’d say you had 3?

If you were counting up to check that your receipt from the shop was correct, you probably would count the half-gone one. You did buy it, even if you don’t get to eat all of it. So then the answer is 4.

But if you were counting up to see how many days you could have an apple for a snack, you probably wouldn’t want to count that one. Or you might count it as “a half”, because you could cut the squidgy bit off and eat the other half. So then the answer is either 3, or 3½.

So even with that relatively simple example, you’re already dealing with a question of: “what are we counting exactly, and why does it make sense to do it that way“.

This kind of question comes up in stats all the time – including with the covid numbers.

Simple example of another tricky question

Another type of uncertainty is where there was a definite answer, but it didn’t get noted at the time, and now you can’t find it out.

For example, you might go swimming at a pool one day, and swim up and down. And later you think, I wonder how many lengths I swam?

There was an answer. You can probably even guess roughly what it was, from memory or from the amount of time you were at the pool. “Maybe about 15 or 20, definitely not as many as 50.”

But if you didn’t count the exact number at the time, you don’t know it now. You can’t go back in time and watch yourself swim.

This is a bit like the situation where someone died of a stroke back in February and now we’re wondering whether the stroke could’ve been caused by covid. If we could go back in time, we could test their body for the virus. But now we can’t.

The more you get into stats geekery, the more you bump into things like those two examples! where there’s either uncertainty, or a “what makes sense here” situation, or both. Even the experts don’t always agree.

For now, though, let’s get into the real questions about covid.

What counts as a death from covid

As soon as you ask the question of how many people are dying of the illness, you have to decide what counts as “dying from covid”.

In some cases, it’s really obvious. The person was fine until they got ill; after a week or so of feeling ill, they had trouble breathing; they went to hospital; a test result said the virus was in their body; after a couple more weeks, they died. Definitely counts.

In some cases, it’s not so clear-cut.

When the test doesn’t confirm

For example: Someone died, and it looked as though they had the illness, but the test result came back negative.

  • Was the test wrong? That can happen.
  • Was the illness really something else? That can happen too.

What if they would’ve died anyway

Suppose someone has a stroke, and dies. That gets written down as: they died of a stroke.

Most strokes are caused by a blood clot getting stuck and cutting off the blood supply to a bit of the brain. (a few are caused a different way.)

We know that covid can cause blood-clotting in the wrong places (or as doctors call it, “thrombosis”),2 and some people with covid infections have had strokes.

What if this person’s blood clot was caused by the coronavirus affecting their blood? Then if the person hadn’t caught the coronavirus, their blood wouldn’t have formed that clot, and they wouldn’t have had the stroke.

How do you tell that apart from a stroke they would’ve had anyway?

Even when you know they had the covid virus at the time… strokes do happen without covid too. You might wonder.

Clues from around the place

Another question you might have to think about is: are you going to allow clues from what’s going on around the person who died?

For example, let’s imagine that you’re a doctor looking after some people in a care home. And one day, one of the people you look after there gets ill with pneumonia – inflamed lungs – and it gets worse, and a bit later, they die.

If that’s all you know, you might put on the certificate “died of pneumonia”.3

But what if you also know that two other people died of pneumonia in the same care home that week? and those two people were both tested for the covid virus, and it came back that they did have it?

Knowing that covid often causes pneumonia, you’re probably going to put that those two did die of covid – or to be specific, something more like “pneumonia caused by covid”.

Then you’re thinking about that first person. Should you count their death as covid-related as well? now you know that covid was in their care home, and that person had the same symptoms as the other two? even though that one person for some reason didn’t get tested?

Definitions, choices and wiggle room

So, when you count up the covid deaths, whenever you get to one of those choices which could reasonably go in either direction, someone has to decide whether it does count or it doesn’t. (same as, in the apples example, you’d have to decide whether to count that half-an-apple or not.)

This “human judgement call” factor is part of the reason why it’s hard to compare between different countries. The people who define the counting-up rules in one country aren’t necessarily making the same judgement calls as the people in another country.

If your priority is “we must definitely keep track of what’s going on, and not be lulled into a false sense of security”, then you make a big effort to offer testing to anyone who’s been near to an infected person. And if you suspect that a death is probably due to covid, you do count it, even if you’re not 100% sure.

For example, in Belgium, if someone dies in a care home and doctors think it was covid, that gets counted as a covid death, even if the person who died wasn’t tested.

“It’s based on the assessment of the medical doctor, usually taking into account whether the coronavirus is present in the same care home,” says Prof Van Gucht.

“For example: if you have one or two confirmed cases, then the week after you have 10 deaths in the same home based on similar symptoms.”

In other words, Belgium is an example where people are “erring on the safe side” in their counting. The numbers might “look worse” than some other countries – but the medics and government there aren’t going to be caught out with “uh oh, actually there were lots more infections we hadn’t realised were there”.

On the other hand, if your area relies on tourism… or you just want your government to look better… or for whatever reason, you decide you want to make the covid deaths look as low as possible, without actually going so far as to write down fake numbers… then you’d want to err on the side of counting the “uncertain” deaths as something else. You might even intentionally not test people, so as not to find out for sure.

“This person died of pneumonia.” Yeah they did, but why did they get the pneumonia? If there’s suddenly thousands more people dying of pneumonia than last year… that’s a bit suspicious!

Or of course, you can equally well have a setup where nobody’s actively trying to wiggle the figures, but they just have a testing system that doesn’t work very well, so it misses a lot of the people.

Here’s an article with a chart showing how different countries have been defining and counting “deaths from covid”.

Here’s an article discussing similar “what are we counting exactly” questions in the US.

England’s definitions of a covid death

As I understand it, there are two sets of official covid death numbers for England.

  • One set of numbers is based on counting who’s died after testing positive for the virus. It covers the whole UK, and there’s a separate subtotal for England. It’s published by the Department of Health and Social Care (DHSC), based on input from Public Health England (PHE) and similar organisations. The latest total is at At 19 July 2020, their count stands at 45,300 “positive-tested deaths” for the UK as a whole, of which 40,706 were in England, 1,547 in Wales, 2,491 in Scotland, 556 in Northern Ireland.
  • The other set is based on what doctors said on death certificates. That one is organised by the Office of National Statistics (ONS), which keeps track of all the births and deaths in England and Wales. When you register a birth or a death in England, the info gets sent to them.Their count isn’t published as often. The latest I’ve seen is from 3 July, published 14 July: 50,548 deaths registered, in England and Wales put together, which had covid on the death certificate.In practice, so far, most of those had covid identified as the main cause of death.

    In the majority of cases (46,736 deaths, 92.8%) where COVID-19 was mentioned on the death certificate, it was found to be the underlying cause of death.

    But it can also include certificates where covid got a mention as a contributing factor. (See longer quote from them below, explaining how death certificates work.)

You can see that the “death certificates” count is coming out higher than the “had a positive test before dying” count.

Let’s look at how each organisation defines the numbers they’re counting.

DHSC definition

The definition in use (at 19 July) by the Department of Health and Social Care is summarised by this line, on the web page next to the total:

Deaths of people who have had a positive test result

The full definition (as of 19 July) says more:

Total number of deaths of people who have had a positive test result for COVID-19 reported on or up to the latest reporting date.

The data do not include deaths of people who had COVID-19 but had not been tested or people who had been tested negative and subsequently caught the virus and died.

Deaths of people who have tested positively for COVID-19 could in some cases be due to a different cause.

(The full definition also includes an explanation of where exactly they get the data from in each UK nation, and other practical details.)

So if you have someone who, going purely by common sense, might have died because of covid, but for whatever reason they didn’t get tested, they wouldn’t be included in this official count. (This probably applied to a lot of people in March and April. More on that in a bit.)

It also means that if someone’s cause of death doesn’t look like covid, but they’d tested positive around the time of their death, they would be included in the count.

When the test has been recent, that’s maybe not as daft as it sounds, given how many weird ways covid turns out to affect people:

One colleague told Belchetz about a patient who came in with a head laceration. “Everyone assumed it was nothing to worry about,” he said. Head wounds are bread and butter stuff in the ER. But after some detailed questioning, the patient revealed how he got the cut: He had passed out and fallen.

He didn’t have a cough or a fever. But he wasn’t getting enough oxygen. He got swabbed. The test came back. He had COVID-19.

I’ve seen people (online) saying things like “my uncle died of a heart attack, and they’ve written down it was covid!” or “my grandma died from a bad fall and they’ve counted that as a covid death! ridiculous!”

And, OK, if you could go and ask God, or replay a moving microscope scan of exactly what the person’s immune system was busy with when they died, then you might discover in some cases that the conclusion was wrong. But if they had the virus at the time, it isn’t all that ridiculous, because we know now that heart attacks and bad falls are both things that can be (partly) caused by covid. In a situation where you can’t be 100% sure, it’s not unreasonable to guess that there could’ve been a connection.

However, this “anything that happens after the test” definition will produce more mistakes as the test result recedes into the past. If someone had a minor run-in with covid six months ago, felt pretty much back to normal after a few weeks, and then dies of something apparently unrelated… does it really make sense to count that death as part of tracking the epidemic? Probably usually not.

So the latest news about how they’re counting it is that there’s probably going to be a cut-off point in time for making the connection: for example, if the person’s positive test result was a month ago, or three months ago (I don’t know what timing they’re going to choose), don’t count that death as covid-related.

The Secretary of State has today, 17 July, asked PHE to urgently review their estimation of daily death statistics. Currently the daily deaths measure counts all people who have tested positive for coronavirus and since died, with no cut-off between time of testing and date of death. There have been claims that the lack of cut-off may distort the current daily deaths number. We are therefore pausing the publication of the daily figure while this is resolved.

(from on 18 July, although by 19 July, the wording had already changed a bit.)

Bear in mind, a time cut-off can’t totally solve the problem either! A person could test positive while feeling fine, and next day, sheer bad luck for them, a builders’ crane tips over and crashes into their house and they die. You don’t really want to count that as a covid death, but it would still be included under the time cut-off rule.

Or, other way round: a person could die of a non-covid pneumonia, a year or two into the future after they’d had covid – and covid actually could still have contributed to their death, if it had left lasting damage in their lungs.

ONS definition, and how death certificates work

Now, what about the definition in use by the Office of National Statistics?

Here’s their explanation of how COVID-19 might end up on a death certificate, and the categories they use when they’re counting (bold type added by me):

The doctor certifying a death can list all causes in the chain of events that led to the death and pre-existing conditions that may have contributed to the death. Using this information, we determine an underlying cause of death. More information on this process can be found in our user guide. In the majority of cases (46,736 deaths, 92.8%) where COVID-19 was mentioned on the death certificate, it was found to be the underlying cause of death.

Our definition of COVID-19 includes some cases where the certifying doctor suspected the death involved COVID-19 but was not certain, for example, because no test was done. Of the 46,736 deaths with an underlying cause of COVID-19, 3,763 (8.1%) were classified as “suspected” COVID-19. Including mentions, “suspected” COVID-19 was recorded on 8.4% (4,251 deaths) of all deaths involving COVID-19.

In this bulletin, we use the term “due to COVID-19” when referring only to deaths with an underlying cause of death as COVID-19 and we use the term “involving COVID-19” when referring to deaths that had COVID-19 mentioned anywhere on the death certificate, whether as an underlying cause or not.

Here’s them putting into context what they do with those numbers, and how their numbers compare to the DHSC ones:

Because of the coronavirus (COVID-19) pandemic, our regular weekly deaths release now provides a separate breakdown of the numbers of deaths involving COVID-19: that is, where COVID-19 or suspected COVID-19 was mentioned anywhere on the death certificate, including in combination with other health conditions. If a death certificate mentions COVID-19 it will not always be the main cause of death but may be a contributory factor. …

These figures are different from the daily surveillance figures on COVID-19 deaths published by the Department of Health and Social Care (DHSC) on the GOV.UK website, for the UK as a whole and constituent countries. Figures in this report are derived from the formal process of death registration and may include cases where the doctor completing the death certificate diagnosed possible cases of COVID-19, for example, where this was based on relevant symptoms but no test for the virus was conducted.

In contrast to the GOV.UK figures, we include only deaths registered in England and Wales, which is the legal remit of the Office for National Statistics (ONS).

Here’s an example pic of what the doctor has to fill in, also taken from the ONS site:

Sample "Medical Certificate Of Cause Of Death", that a doctor would fill in after someone's died.

The doctor gets four different lines for what was going on in terms of the person’s health or death:

I(a) Disease or condition directly leading to death

(b) Other disease or condition, if any, leading to I(a)

(c) Other disease or condition, if any, leading to I(a)

II Other significant conditions
CONTRIBUTING TO THE DEATH but not related to the disease or condition causing it

(those capitals are on the original thing, not added by me)

and there’s also a footnote for I(a) which says

This does not mean the mode of dying, such as heart failure, asphyxia, asthenia, etc: it means the disease, injury, or complication which caused death.

In the advice for doctors about filling them in, I found an example of what you might put for covid:

I(a) Disease or condition directly leading to death

Interstitial pneumonitis

(b) Other disease or condition, if any, leading to I(a)


So in that example, the covid caused the pneumonitis, which caused the death.

In that example, COVID-19 would later be written down as the “underlying cause of death“, and the ONS would report that death as “due to” COVID-19.

If the person also had diabetes (which might have contributed to their vulnerability to covid), that would go into the second part, “Other significant conditions contributing to the death but not related to the disease or condition causing it”.

So the key difference from the DHSC definition is that this one involves some common sense from human beings: firstly the doctor who writes the death certificate, and secondly, someone else deciding what to count as the underlying cause, if it wasn’t totally obvious.

My guess is that in most cases, this is going to give a more sensible result. I could be mistaken, but it seems fairly unlikely to me that a doctor would write down “covid” as a contributing factor if someone died because of a crane falling onto their house.

It does still allow for a lot of covid deaths to have been missed in the early days of the epidemic, before the illness was as well understood (and while most people in England couldn’t get tested).

Learning to recognise it

The thing is, when covid first became famous, most people (including doctors) were thinking of it primarily as a “respiratory illness with fever“. It wouldn’t surprise me if in the early weeks, a stroke, heart attack, or “quiet death overnight” might not have “looked like” covid, even if actually it was.

This article gives the flavour of those early days, of doctors coming to realise how many different ways it could affect people:

… initially we had this very clear case diagnosis,” [Dr] Belchetz said. “It was travel and a cough and shortness of breath and fever.” … “But what we’ve been finding is almost anything can be a presentation of COVID-19,” Belchetz said. “We’ve seen patients whose only presenting symptom is headache or their only presenting symptom is abdominal pain and we swab them and they’re positive.”

In March, it was still newsworthy that covid “might” cause loss of smell! Yet we know now that that’s actually one of the most common symptoms.

And emerging through June & July, the COVID Symptom Study picked up that some people have a skin rash as their only symptom.

I think by now, most doctors are aware of most of the main ways it can show up (not saying there won’t be a few more weird new symptoms to be added to the list), and testing is more widely available where covid is suspected. So I don’t expect that a large number of covid deaths are still being missed in England now. But in the early part of the epidemic, it’s very possible that there were death certificates written which, if the same death happened now, would include “caused or partly caused by covid“.

To sum up: both the DHSC count and the ONS count are approximations, not spot-on measures of “how many deaths in England were caused at least partly by covid infections”. Even the one I think is probably closer (the one drawing on doctors’ common sense) isn’t perfect.

“Excess deaths”

A number that’s much simpler to define is what’s called the “excess deaths“.

Instead of trying to work out why people died, you count up how many people died overall. And then you compare that number with other recent years.

It’s not trying to measure exactly the same thing as “deaths from covid”. But it can still be useful in getting a sense of how your area’s doing.


For example, let’s imagine you work in the registry office in Fictiontown, Fictionalshire. From past experience, you know that every April for the last few years, there’s been roughly 100 deaths registered in your town. Some years it could be a bit more, like 120… some years could be a bit less, like 85.

To count up those numbers, no-one has to make a medical diagnosis: each one is simply one death registered at the Fictiontown registry office on an April calendar date.5

Now that covid’s reached the UK, what’s happening in Fictiontown?

Like every other year, you can count up how many people’s deaths were registered in April 2020.

If the total came out about 100, you can be pretty sure covid hasn’t had a big effect in your town – because that’s about the number of deaths you’d expect in any average year.

If the Fictiontown deaths in April were 170, you’re gonna be like: uh oh! Something is going on.

You can do the sum. 170 deaths overall, take away 100 you were expecting in an average April: you know you’ve got about 70 deaths you weren’t expecting.

That’s what they call the “excess”: deaths that wouldn’t normally have happened at that time.

Deaths for other reasons

It’s important to realise, we don’t just jump to the conclusion that all those extra deaths are directly due to covid! There are other reasons that “excess deaths” can happen in an epidemic.

For example, someone might have felt ill, and they might have put off going to hospital, thinking “a lot of people in the hospital right now have covid, and I don’t want to catch it”. Then they feel worse, and it turns out they had something that ought to have been treated quickly, and they die.

Or let’s say that because of the combination of the epidemic itself and the emergency measures thrown together to handle it, someone was under a lot more stress than usual (maybe financial worries, or a friend dying), and died of a stress-related condition.

Those are deaths which are indirectly due to the epidemic and how it’s being handled here, but aren’t someone dying of covid. It’s kind of a “knock-on effect“.

(To complicate things even further, it’s likely that some people stayed alive in this version of the world, who would’ve died if the epidemic hadn’t happened – e.g. people who would’ve died in a car crash if they’d gone to work that day, but instead had worked from home. And we don’t know those numbers for sure either. We can never be 100% sure of all the “maybes”.)

When the epidemic got out of hand in the UK in April, lots of people with cancer had their treatment deferred. Some of those people will die in the next few years, who could’ve otherwise lived longer if their treatment had been done at the right time. So we haven’t yet seen all of the “knock-on effect” deaths which have already been caused.

(This type of “death by delay” also happens to an extent without covid, due to underfunding of the NHS. In some countries, people with cancer get treated much quicker than here.)

So the “excess deaths” doesn’t tell you directly “This is how many people died of covid”. What it tells you is more like: overall, how is this area doing in coping with the situation.

“Excess deaths” isn’t very wiggly

The great advantage of the “excess deaths” measurement is how straightforward it is. Stats people might call this kind of measurement “robust”.

There are already UK laws that say every birth and every death must be “registered”. And because that’s been true for so long, even the tricky questions have generally already been sorted out by years of tradition that you can just stick to.6

So unless you’re going to do some fairly bold-faced cheating (like intentionally making up fake numbers), you can’t really fiddle it.

It’s also a measure which is likely to be done the same in different countries – whereas different countries have come up with different rules about what counts as “a covid death”, and some have done much much more testing than others to find infections.

(OK, it’s not impossible for countries to fail at tracking total deaths accurately – especially if there’s a war going on. But in that case, they probably wouldn’t be able to track the other measurements either.)

Some real numbers for England

Remember I said how the Office of National Statistics (ONS) keeps track of births and deaths? If you have a spreadsheet program, you can look for yourself at the numbers of deaths in the different towns and areas of England & Wales, month by month.

To get a sense of the typical death numbers for April in England, I looked back at the five spreadsheets from 2015 to 2019.

Of those recent years in England, the lowest total of April deaths was a bit over 36 thousand, in 2017. The highest was nearly 44 thousand, when there was a particularly bad Spring flu in 2016.

So if covid hadn’t come along, we’d have been expecting a number of April deaths in between those two numbers: that is, around 40 thousand, give or take a few thousand.

As the “typical April in England” number to compare with, I’ll use the average of those five previous Aprils in a row, from 2015 to 2019.7

(There’s a tradition that when you’re going to do a comparison like this, you look back over 5 years, not just one earlier year – because it goes up and down a bit every year. If you just took 2019 as a supposedly “typical” year to compare with, you don’t want to find out later that 2019 was actually an odd one out itself. That’s why I’m working out the average of 5. The details are in that previous footnote.)

That 5-year average is about 41,400 deaths.

At the moment,8 the England deaths for April this year add up to about 83,500.

To get the “excess deaths” for the month, we look at this year’s total, 83,500, then take away the amount we’d normally expect, 41,400. The result is: 42,100.

That is, in England in April, we had about 42 thousand deaths we weren’t expecting, alongside the similar number which we were expecting. It works out about twice as many deaths as would’ve typically happened in recent years.

What were those other deaths?

As I said above, we can’t just assume that all those extra, unexpected deaths were people dying of covid. I was wondering myself about this, and looking out for info. Aside from the “officially covid” number, were they people who genuinely did die of something else? like a stress-related thing, or connected with not going to the hospital early enough when they felt ill? Or were they people who did have covid, but weren’t diagnosed?

Nick Stripe‘s job is health analysis at the Office of National Statistics. On 5 June, he did a thread analysing what we know about “excess deaths” in the UK.9

Note that whereas my example was specifically England, he’s talking about the UK as a whole. And whereas my example was specifically April, he’s including a few extra weeks – from 7 March to 1 May. But still, the puzzle to solve will be similar.

Short version: they probably mostly were covid.

(Skip ahead if that’s all you want to know about that.)

For anyone who’s interested, a bit more detail, based on points from Nick’s thread:

  • About three-quarters of those “excess deaths” match up with the known covid-related deaths in the same timespan. In other words, we already have a reason for three-quarters of the unexpected ones. That leaves a quarter to think about.
  • Of the remaining quarter, i.e. the ones which weren’t officially covid: about two-thirds were certified as connected with dementia, old age or frailty. The dementia-related deaths in particular went up very quickly compared to the typical pattern:

    Dementia increases are so sharp it’s implausible that they are unrelated to COVID

    How would those connect?

    • Some people may have had a “quiet” form of covid which nobody noticed:

      Some evidence has been observed for atypical hypoxia in frail COVID patients – well preserved lungs but severely compromised pulmonary gas exchange without signs of respiratory distress

      This refers to people who don’t feel breathless, and whose lungs would look pretty good if you scanned, but in reality they aren’t getting enough oxygen. (This can happen in younger people too, with covid or in high-altitude parachute jumping. You feel OK, and unless your oxygen level gets measured, no-one might realise how close you were to suddenly losing consciousness.)10

    • In the same tweet, he also points out,

      People with dementia are more likely to have communication problems describing symptoms

      Without the details of invisible symptoms (such as losing your sense of smell), maybe no-one realised it was covid.

    • So overall, it’s likely some of the “dementia” deaths could have been covid deaths that weren’t picked up on at the time.But not necessarily all:

      … we cannot discount the impact of changes to normal routines for vulnerable care home residents following lockdown. These could have had adverse consequences too

      (Maybe someone always looked forward to the music session once a week, and because of covid that didn’t happen. Their mood dipped, they got less exercise, their overall health went down. Or maybe there was one particular carer who was particularly skilled at helping an elderly person with their food – and maybe while that carer was off sick with covid themself, the older person lost some weight, making them more vulnerable to some other health problem.)

  • There were more deaths than usual of people with high blood pressure. Some could’ve been because of stress – and some could’ve been unrecognised covid deaths, because high blood pressure seems to be a risk factor for getting covid badly. From just looking at this one bit of information, we can’t tell what the connection was.
  • In areas where the officially-covid deaths were high, the other deaths were high as well. The two things went together.It seems to me this could be because more covid deaths in an area meant more disruption in the area (e.g. makes it more likely that someone didn’t get treatment for something else). But it would also make sense that the “mystery deaths” went up alongside covid if they mostly were more covid.
  • A point in favour of that interpretation is what then happened in May:

    Note – excess deaths during May are so far all accounted for by COVID being mentioned on death certificates

    This may reflect improving knowledge of its complex effects, increased testing, and the fact that some earlier deaths will have been brought forward by COVID

    In other words, once we get into May, the “mystery” starts to disappear, and we see that the “unexpected for the time of year” deaths exactly match the “we know this many were covid” number.

Nick sums up:

The balance of evidence so far points to undiagnosed COVID in the elderly being the most likely explanation for a majority of excess deaths that did not mention CV on certs

This fits: demography, locations, esp where testing was sparse, causes of death & timings of peaks

Here’s a much much much more detailed analysis of these stats, from the ONS web site.

To me, another bit of evidence pointing that way is how doctors were learning through the spring about the different ways that covid can “present”. If no-one realises that covid was a factor, then (unless there’s routine testing for the virus), it won’t end up on the death certificate.

So – although we can’t be 100% sure – the overall picture suggests that the “puzzle” in March and April came mostly from fewer of the covid infections being recognised.

This means that, even if individual cases don’t all get diagnosed accurately, for now we can make a pretty good guess of “covid deaths” by looking at “excess deaths“, and it won’t be too far off the true covid numbers.

UK excess deaths

The Financial Times (FT) has been tracking “excess deaths” across the world. By the time you look at the FT’s page, it might have updated again – but at the time of writing this, its latest figure for the UK was for the year up till 26 June, and they reckon the excess UK deaths up till that point stood at 65,700.

Because of the reasoning which Nick Stripe explains (discussed above), we can guess those were probably mostly covid.

(Jamie Jenkins, an independent statistician, has also been tracking UK death statistics, and he estimated 69,005 excess deaths up till 23 June. Not far different from the FT’s estimates, but a good example of how the experts don’t always agree. He discusses in his thread why it might be different. For now, I’ll go with the FT’s numbers, the lower set.)

We can compare that with the number the Department of Health and Social Care counted for covid deaths in the UK up till 26 June. Their total – the one based on positive tests – was 43,414.

So that’s 65,700 estimated excess deaths, and the DHSC has 43,414 of them officially written down as covid. The remainder is: 22,286 – about 22 thousand.

In other words, during the spring of 2020,11 there were about 22 thousand deaths which wouldn’t have happened in a typical spring but weren’t initially put down to covid. And we can guess most of them probably were covid, but didn’t get into the DHSC’s numbers.

I don’t think it would be unreasonable to guess that pretty much all those 22 thousand were actually due to covid. But if we want to be sure not to overestimate the covid death numbers, we could also speculate that 10% or so were really other things, and knock off a couple of thousand. Let’s say 20 thousand.

(Or, if you disagree with the reasoning and/or the estimates, then you might want to adjust further. I don’t think it makes sense to assume there would be no covid deaths missed, given the learning curve that the doctors were on at the time, and the shortage of tests.)

The extra 20 thousand or so

To recap, that 20 thousand is an estimate of the covid deaths from the spring which were “missed” by the DHSC’s stats. I’ve calculated that based on three pieces of information from other people: (a) the “excess” deaths estimated by the Financial Times up till 26 June, (b) the DHSC’s official covid deaths number for the same time-frame, and (c) Nick Stripe’s analysis of why it makes sense to conclude that most or all of the other “unexpected” ones were also covid.12

(My working assumption is that almost all of these will be people who never got tested at all, either because nobody thought of it at the time, or because there weren’t enough tests available. It’s not impossible there could’ve been a few false negatives too – where the person did have the virus, did get tested, and the test didn’t pick it up.)

So when you see the DHSC “dashboard” numbers going up (which of course they have done a bit more since then), it makes sense to add on about another 20 thousand to those – or whatever you estimate is the number who died of covid without being tested.

For example, on 19 July, the DHSC’s total was 45,300. So we can estimate that it’s really more like 65,300. It’s only approximate anyway, so we may as well say 65 thousand.

The ONS’s stats will include some of the deaths which the DHSC count “missed”: where the person didn’t get tested, but the doctor worked out anyway that it likely was covid they had. The ONS stats will also have missed some. So if you happen to see one of their totals (the ones based on death certificates), bear in mind that’s not the full total either.

Would some of the people have died by now anyway?

Everyone dies eventually!

Life expectancy” is basically “how long an average person like you could expect to live, with reasonable luck”.

Typically it’ll be based on your age, sex and health conditions, and what’s happened in the past to people in the same categories.

The ONS has a simple life expectancy calculator which anyone can try. By “simple”, I mean that it doesn’t take into account your health or habits at all – it’s based purely on what age you are now, and male or female.

For example, it says a man in the UK who’s made it to the age of 70 would have an average life expectancy of 86. That’s only the average, so that same category includes all the 70-year-old men about to die now and the ones who’ll live to be 90 or 100.

Life insurance companies are always dealing with this kind of stuff. There’s a kind of life assurance where you pay in a bit every month while you’re alive, and there’s a payout to someone else when you die. Every time they give someone a quote, they’re like “OK, let’s see… woman of 35, she smokes, otherwise in good health so far, she’ll probably live X many years”.

Then if they think they’ve got years and years to make money off you, they give you a low quote per month, and if they think you’re going to die pretty soon, they give you a high quote.

In those business calculations, they use wayyyy more details than the little online calculator linked above – if you’re signing up for life assurance, any health condition you’ve ever had will probably have to go on their form.

Even with every single detail, it doesn’t mean their forecast will be correct about any particular person. They only have to be right “on average” to make money.

So, anyway, people and companies in that sort of business have lots of incentive to know loads about when everyone’s likely to die. And this is the sort of data which can also be used to estimate how many years of life have been lost to covid.

I’m not saying it’s quick to work that stuff out. It’s quite complicated. You have to categorise the people who died, and compare their life expectancy with the average for people with comparable health conditions. And there’s hundreds or thousands of categories, combining all the possible ages and health conditions! But people do know how to do it.

Jamie Jenkins has done some sums around that. As of 7 July, he’d estimated that 300 to 350 people who died of covid earlier in the year would’ve died by now anyway, even if covid hadn’t come along.

Time-lag in measuring

Here we get to one of the trickiest things about tracking the epidemic: the delays before you find stuff out.

When someone gets symptoms from covid, typically the symptoms would start about 5 days into the infection, a day or two after the person became infectious to others. Symptoms can start as quickly as 2 days in; it’s usually within a fortnight.13

(Not everyone who gets the virus ever notices any symptoms; some people just catch it and get over it, without realising. The virus might still do things to their body that they weren’t aware of.14 I’m not sure where the research stands on whether people can be infectious if they never have symptoms; you can definitely be infectious before you have symptoms.)

When people die of covid, that would be later still: typically around 15 to 22 days after symptoms start, though it could be shorter or longer.15

This adds up to mean that if someone dies of covid today, they probably caught it around three weeks ago, maybe more, maybe less.

So when we look at the numbers for “deaths which happened this week“, that’s not telling us much about the infections happening today. A death today is telling us something about the infections happening back a few weeks ago, when that person originally caught the virus.

Likewise, if someone gets a positive test result today, that might be from an infection a few days ago, or last week, or the week before – depending on how fast they got tested and how fast the result comes back. (You might get tested even before you’d had symptoms, e.g. because someone close to you had tested positive, or as part of some research that chooses people randomly to be tested. On the other hand, some people get as far as being quite ill and going into hospital before they get tested.)

(By the way, a positive test result doesn’t necessarily mean you’re still infectious to other people. Research is still happening about how long people typically stay infectious to others, but a study in Taiwan suggested probably mostly not after Day 5 of symptoms.16 You might even be over the illness yourself and still test positive: the “have you got it now” type of test looks for bits of virus, and there might be some inactive remnants of virus in your body while your immune system is “tidying up” at the end of an infection.)

Ideally, we’d have quick testing of anyone who’d been near an infected person. If people can get their test results back the same day – maybe even before they show any symptoms – then that gives a better picture of what’s going on right now, and a better chance of reaching people before they infect anyone else.

As it is, though, most of the measurements we have are a bit like “looking in the rear-view mirror”.

How many infections in an area

Like I said, “excess deaths” is a pretty solid kind of number. People are either alive or dead, and outside of war zones, it would be rare for more than a handful of deaths to go unnoticed.

The question of “how many infections” is much harder to be certain of, because it’s so easy not to know about some of the infected people.

In a country or area that’s “on top of things”, they will know almost exactly how many people currently have the virus. As soon as someone reports symptoms or tests positive, a contact tracer person has a chat with them, then reaches out to all the people they’ve been close to recently. Soon, most of those other people have been tested, and if they test positive, that’s one more person on the “case numbers” count. And they’ll all be supported to stay in isolation till they’re not infectious any more.

However, if an area (like England at the moment) doesn’t have a reliable system for testing and tracing, its “case numbers” might miss out lots of people:

  • the ones who only just caught the virus, and don’t have any symptoms yet.
  • the ones who have symptoms, but wrongly think their symptoms are something else.
  • the ones who think they have covid, but haven’t yet been able to get a test.
  • anyone who gets over the virus without even having noticeable symptoms.

That could be a lot of people!

Infections in the UK

Back in March, a team led by Prof Tim Spector of King’s College London launched the COVID-19 Symptom Study. They made an app which is available on iPhone or Android, which invites people to input what symptoms they’re having each day (if any). People also input the test result if they go for a test. About 3 million people in the UK are taking part. Then the researchers have been analysing what people said about their test results, symptoms etc.

If you want an estimate of current covid cases in the UK, or up-to-date info on what symptoms are common, this project is the best source I’ve seen so far. Advantages:

  • they’re getting their info very quickly, because it’s based on people putting things directly into the app, the very day they notice a symptom.
  • a lot of people are taking part.

The main limitations are:

  • they only report on the age range 20 to 69, due to insufficient data coming in from younger and older people.
  • they don’t know about infections with no symptoms.

Here’s the page which shows their estimates of UK cases, changing every day.

Today, 19 July, they estimate there are 28,368 people in the UK with symptomatic COVID, in the age range 20-69, not including people in care homes. (And note that they said “symptomatic”, so this number also doesn’t include people who have the virus without symptoms.)

That number also doesn’t include the people who are experiencing the long-term aftermath of the virus: “long covid“, as it’s sometimes called.17

Separate from that, the Office of National Statistics runs the COVID-19 Infection Survey. They invited some thousands of households to take part, by doing the “swab tests” at home and sending them in. Because it takes time to send in the tests and have them processed, the estimates from this have more of a time lag than the Symptom Study. On the other hand, unlike the Symptom Study, this one will be detecting infections in people who don’t have any symptoms.

Meanwhile, separately again, the UK government’s official tracker page for covid deaths and infections describes its infection count (running total 294,792 on 19 July) as “lab-confirmed UK cases”, or “Total number of people who have had a positive test result”. If you didn’t have a test, you won’t be counted in that number.

So when you see the total for “case numbers” on the “dashboard”, the key thing to keep in mind is: That isn’t a measure of how many cases there are. It’s a measure of how many cases they know about.

That brings me to…

What do we want these numbers for?

One of the reasons we want these numbers is so we can see what’s happening near to us: how risky is it now? are things getting better or worse?

Ideally we’d have all the numbers in quite fine detail, and linked to areas and groups. When you’re deciding what to do about things like reopening shops or cafés, what makes sense in one area might not be what makes sense in another.

(In some countries, you can see that data right down to individual streets. Has someone in the area had the virus recently? Maybe you want to walk down a different street instead.)


Another thing people often want to do with these stats is to make comparisons.

Sometimes of course this is just “yay, my country is doing well” type stuff. But it also has some much more practical uses.

One is: when two different areas are facing a similar challenge, and we see that one area has done much better than another, it shows us that there might be something to learn from the successful place. What did people there do, that worked well, that we could perhaps copy or adapt?

Another is: by tracking how the numbers change, we’re also learning about the virus itself. How infectious is it? What are the environments where it’s most likely to spread? What helps to protect you? If someone gets it, how likely is it they’ll have an easy time, versus how likely they’ll get badly ill or die?

Fair comparison

When comparing how different areas are coping, it makes sense to bear in mind how places can differ from each other. For example:

  • people in towns tend to live closer together than people in rural areas, may have worse air quality, and may be more likely to use public transport.
  • many office workers can work from home, whereas families or communities who rely on earnings from retail, delivery, manufacturing, construction, tourism etc might be under financial pressure to take more risks.
  • the climate could play a role, e.g. if people spend more time outside (where there’s a breeze helping to disperse any virus in the air), or if the virus becomes unviable more quickly at different temperatures.18
  • in some countries there are more older people.
  • in some areas or traditions, multiple generations of a family would typically live together, whereas in others, grandparents are likely to have separate houses.

But one of the most obvious differences is of course the number of people in each place.

Taking into account the size of population

If you’re going to say, for example, “the US had more deaths than Hong Kong”, you have to take into account that the US is an enormous place with an enormous number of people in it, whereas Hong Kong is relatively small.

With all these numbers you might be tracking, if you want to look at how one place is doing compared to another, you’ll want to take into account how many people live there overall. (You might see this referred to as “adjusting for population”, or a “per capita” number.)

So for example, instead of just saying “1 person died”, you might say “Out of 100 people, 1 died”.

By this method, instead of comparing all the deaths in a big country with all the deaths in a small one, you can compare a same-size chunk of the people.

To compare covid death rates between countries, they’re often worked out in terms of “excess deaths per million”. That takes one of the most reliable, non-fiddleable measures, and makes it fairer between countries of different sizes.

(It still might not be a completely fair reflection of how a particular government’s done at coping with the epidemic, because of other factors I namechecked above, like some countries having more older people.)19

So let’s look at that next.

“Excess deaths” per million people in the population

For this example, I’ll go back to the UK as a whole, and the epidemic as a whole.

As of mid-2019, the UK population was estimated at 66,796,807. In recent years, typically it’ll grow a bit rather than shrink, so let’s say now it’s probably about 67 million.20

So that’s: 65,000 extra deaths, in a population of about 67 million people. That works out as about 970 deaths per million people who live here.

When other people have worked out the same sum, I’ve also seen it come out slightly differently. Most likely that’s because they used either a different estimate of the extra deaths, or a different estimate of the population, or both.

Here’s a Financial Times article comparing excess deaths per million across Europe at the end of May. They estimated the UK’s number as: 891 deaths per million people.21

If we take instead the DHSC’s official number of covid-related deaths, that gives only 676 deaths per million up till 19 July.22 But that would only be accurate if no covid deaths were missed in the early stages, which seems to me unlikely.

For comparison, Denmark has had 611 known covid deaths in total (not per million) so far, and Greece has had 194.23

How many people had it already?

There have been rumours that the virus has been very widespread already – that lots of people had it without symptoms, so that perhaps a quarter of people or half of people across England (or the whole world) are immune by now.

It’s important to understand that immunity is still something we don’t know much about. Even if someone has had the virus once, it doesn’t necessarily mean they can’t get it again.

(At the moment, it’s looking fairly likely that having the virus once gives you immunity for some weeks or months, and eventually you’d be able to catch it again. But that’s only a guess of how it could play out, based on comparing it with similar viruses. We really really don’t know yet.)

But even if catching the virus doesn’t equate to being immune afterwards, it would still be useful to know how many people have already caught it.

So then the question is: how can we tell who’s had it & who hasn’t?

How can we tell who’s had it?

For this question, it wouldn’t be any good to look for the virus itself in your body – because once you’ve got better, the virus would be gone.

What you can do is look at the state of someone’s immune system. Your immune system responds to the virus in ways that can be detected afterwards.

For example, you might have antibodies which are specific to a particular virus. An antibody is a tiny bit of your immune system24 which recognises a virus it’s “seen” before.

And researchers have already developed tests for antibodies that “match” SARS-CoV-2, the virus that causes covid. Yay!

It’s not quite as simple as “whoever has the antibodies, that’s the people who’ve had the virus”. For one thing, the antibodies sort of “fade away” after a few months.

(Your body might still remember how to make the antibodies again, if they were needed. So when the levels go down, it doesn’t necessarily mean you’ve lost immunity that soon. It just means that “have I got covid antibodies right now” isn’t a guaranteed method for finding out “have I had covid already”.)

And researchers have suggested that some people don’t make the antibodies at all, e.g. if other parts of their immune system handled the virus quickly. More on that in a bit.

But even though the antibodies aren’t a perfect marker for who’s had the illness… do we know how many people have them?

The answer is… we do know something about that.

Antibodies in Spain

So far, the biggest lot of testing of who had the covid antibodies was in Spain, over two weeks in April-to-May of 2020.

Spain is one of the European countries initially hardest hit by the virus, similar to Italy and England.

Between 27 April and 11 May, researchers took blood samples from sixty thousand people across Spain, being sure to include rural areas as well as towns.25

From this, they could work out that in Spain overall, at the time, probably about 5% of people had antibodies which are specific to SARS-CoV-2, the virus for covid. That is, for every 1 person who did have the antibodies, another 19 didn’t.

In parts of the country where lots of people had died, it was higher: of every 20 people, about 3 showed the covid-specific antibodies, 17 didn’t.

Remember, that doesn’t mean you can just go “Only 5% of people in Spain had the virus!” When the Spain antibody research first came out, a lot of people thought it did mean that, so you might have seen newspapers reporting it along those lines. But no. It was “Probably only about 5% of people in Spain had the antibodies“, which is not the same thing.

Testing for T cells

So if some people didn’t have the antibodies… how do we know they ever had the virus?

T cells” are another bit of your immune system, which also help to eradicate viruses.

Compared to finding if someone’s got particular antibodies, it’s more of a faff to see if they’ve got particular T cells. But it can be done.

Some researchers in France looked at families where one person definitely had covid back in March, and another person was exposed. The first person was called the “index patient”; the other person in their family was called the “contact”.

In May, they tested the “index patients” and the “contacts”. For all the “index patients”, the researchers could see covid-specific antibodies and covid-specific T cells – not very surprising.

What was interesting was what they saw with the eight “contacts”. They didn’t test positive for the covid-specific antibodies, but 6 out of 8 did have the covid-specific T cells.26

(The researchers’ commentary puts this down to them never having had the antibodies, but I’m not sure they’ve proved that, because their description says the blood samples were from May. Did anyone test the “contacts” earlier than that? Maybe at some time in between, they would have tested positive for the antibodies. But whichever – the main point for what we’re talking about here is, it was true a couple of months later.)

In other words, someone could encounter the virus, and their immune system responds and fends it off – yet, if their body did what happened with those people in the research, they wouldn’t be picked up by the tests used in Spain, which only looked for antibodies.

What it shows is: if you want to know whose immune system has responded to the virus, you can’t only test for antibodies and expect to find everyone.

It’s worth remembering that everyone’s antibodies will probably “wear off” after some weeks. If they’d tested the same people a few months later, a lot of people who had had the antibodies wouldn’t any more – whereas most or all would still have the T cells. (People who survived the first SARS in 2003 still have T cells specific to that, 17 years on.)27

So the big question is… how many people in Spain, who tested negative for the antibodies, actually had encountered the virus? and tested negative because by the time of testing, their antibody response had quietened down again? (Or, if the French researchers’ idea is correct, they’d never made the antibodies in the first place? And there could also be a few people who actually still had some antibodies when the testing happened, but had tested negative due to the tests not being perfect.)

More research required :-)

It would be great to do a big research project where thousands of people got tested for the T cells as well as the antibodies, or make that an easily-available test. However, the limitation on that is the faff of the testing: the T cell testing process is more bothersome and expensive than the antibodies one.28 And most labs who could do the processing for this test are already extremely busy with other stuff, including tests that are more urgent. It wouldn’t be easy to roll it out so everyone in the country could get their T cells tested.

However, there could be work-arounds. When we know more about

  • whether everyone who gets covid even makes the antibodies at all
  • how soon they typically appear, and
  • how soon they typically disappear

(which are all things that researchers are very very interested in at the moment!)29

… then we could go back to the Spanish research and do more sums and estimates.

Here’s a very good blog post which goes into more detail about all this area: antibodies, T cells and so on.


Antibodies in the UK

Some people from the Office of National Statistics have been researching how many people in the UK have the covid-specific antibodies. They didn’t test as many people as the Spanish research, which means the numbers might not be quite as accurate. But what they found out suggests that in the UK in May, about 7%, or 1 person in every 15, had the antibodies that match up with SARS-CoV-2, the covid virus.30

That doesn’t mean it was the same proportion everywhere. For example, there have definitely been more cases than average in London – which would mean lower than average somewhere else.

The same cautions apply: that number doesn’t necessarily reflect everyone who’s had the virus, and, having the antibodies now doesn’t mean you can’t get the illness again.

Infection Fatality Rate

Having talked about all that, we now come to one of the things there’s been the most argument over.

Suppose 100 people get the virus. How many of them would typically die? That’s known as the “Infection Fatality Rate“, or “IFR“.

(It’s not the same as the “Case Fatality Rate”, or “CFR”, which is based on only the people who had symptoms. IFR includes people who got over it without realising they’d had it.)

We don’t know!

A lot of the uncertainty goes back to numbers already discussed above.

  • How many people have died of covid?
  • How many people have been infected so far, including the ones who didn’t even realise?

To find out what proportion of people would generally die, you’d have to know both of those things. And we don’t definitively know either of them!

An early guess, based on the numbers from China and Italy, was that of every 100 people who caught the virus, on average 1 would die.

Some people said like: “naah, it’s much less risky than that – yes a lot of people have died, but you don’t realise how many people were infected – when you consider how many people had it without knowing, the number of people who died looks small.”

However, from the research that’s been done so far, it does look as though that initial guess of 1 in 100 was about right (for the current stage of knowledge about how to treat people). Could be 1 in 200.

Let’s test it against the numbers already discussed.

What if we do the sums while assuming it’s approximately right that in the UK, 1 person in 15 has had the virus already? There’s about 67 million people in the UK, so 1 in 15 would be about 4.7 million people who’ve had it so far (many without realising). And we’ve already estimated that 65 thousand people is probably about right for how many people died so far.

That works out as: about every 72 UK people who had the infection, 1 of them died. (This can also be put as 1.4%, one point four percent.)31

Or what if we do the sums again, but this time assuming that some people weren’t found by the UK antibody research? Suppose, for every one person who did still have the antibodies, there was another person who’d had the virus who wasn’t picked up by that type of test.

In that case, about 9.4 million people in the UK would’ve already had some level of covid infection, and the IFR would be more like 0.7%: 1 person dying in every 144 infected.

Remember, these possibilities are based on estimated numbers. They might not be spot-on.

For comparison, here’s an explanation from epidemiologist blogger Gideon Meyerowitz-Katz. He and a colleague reviewed a lot of other people’s research about this same thing.

They concluded that it looked as though the rate was about 0.64% (nought point sixty-four per cent), which is 1 person dying in about every 156 people infected. (But they do add: “…it is difficult to know if this represents the true point estimate. It is likely that different places will experience different IFRs.”)

So I’m not going to say that 1 in 72 is the definitive amount, or that 1 in 144 or 1 in 156 is the definitive amount. I’m going to say that yeah, based on what we know so far, 1 in 100 probably isn’t far off.

Age groups and the Infection Fatality Rate

When they say “different places will experience different IFRs”, part of what they’re talking about is how the age groups can be different sizes in different countries – e.g. some countries have more young people than others. (This is sometimes referred to as “age distribution”.) We know already that covid tends to hit older people harder. With an illness showing that pattern, aside from whatever you do with treatments, the countries with a lot of older people are always likely to have more people dying of it – simply because those countries have more of the people most vulnerable to the illness.

In other words: there won’t even be one definitive Infection Fatality Rate that applies everywhere across the world. What research could eventually find out is a set of IFRs which are typical for different age groups. And from there, you can work out an expected IFR for a particular country, based on its pattern of age groups.

Risks reduce as doctors learn more

Of course, how risky it is to get the disease will change as doctors learn more about the disease, and about what drugs or other treatments are helpful. So, even if 1 in 100 is a reasonable estimate for now, while we don’t know much about treatments, that doesn’t necessarily mean it’ll be the same risk in a few years’ time, or even a few months’ time.

For example, in April, doctors were discussing which patients should be lying on their front, & for how long, which can help with breathing difficulties.

In May, a trial started of whether a drug called Interleukin 7 will help people who have the illness badly. Many other drug trials are in progress.

In June, there was a report from a big study on the safety of giving sick people the antibodies from recovered people (via blood plasma donations).

People who’d already died in March didn’t have the benefits of this developing knowledge.

Long-term disabilities

At the moment, we know even less about long-term disabilities as a result of covid illness. We know it can happen; we don’t know how likely it is.

We know that complications of covid “can include delirium, brain inflammation, stroke and nerve damage”.32 Usually, some people would recover completely from a stroke, and others not.

We can make some guesses based on what other viruses do: for example, some of the people who lived through SARS in 2003 still had lung damage years after the immediate illness.33

We already know that many people who recover from covid have damaged lungs at the time – even some of the ones who hadn’t noticed any symptoms. But obviously we can’t yet look at, for example, “how many people’s covid lung damage got completely better after a year”, because a year ago, nobody had had it yet!

The COVID Symptom Study is tracking the question of how long the initial illness would typically last:

Data from our COVID Symptom Study suggests that while most people recover from COVID-19 within two weeks, one in ten people may still have symptoms after three weeks, and some may suffer for months.

Ed Yong at The Atlantic interviewed some of the “long-haulers”, and described a crowd-sourced report from some of them (including previously-healthy people):

As many people reported “brain fogs” and concentration challenges as coughs or fevers. Some have experienced hallucinations, delirium, short-term memory loss, or strange vibrating sensations when they touch surfaces. Others are likely having problems with their sympathetic nervous system, which controls unconscious processes like heartbeats and breathing: They’ll be out of breath even when their oxygen level is normal, or experience what feel like heart attacks even though EKG readings and chest X-rays are clear.

(If some people did have a long-term fatigue syndrome after covid, it wouldn’t be surprising, because for a lot of people who have chronic fatigue, it seems to have been triggered by a viral infection of one type or another.)

At the moment, there’s a lot of “wait and see”.

The “R” number

The R number is to do with how likely the virus is to spread itself around. The “R” stands for Reproduction, meaning the virus infecting new people with copies of itself. It’s never an exact number – it’s more of a shorthand for how things are going.

(You might also see Rt or Reff meaning the same thing.)

Let’s say for example that Adrian infects Bell, and then Bell infects Caz, and then Caz infects Don. That would be: the R number is 1. Each 1 infected person is infecting 1 other.

What if it spread out to more people? Let’s say Adrian infects Bell, Becks and Bob. Then Bell infects Caz, Cal and Carol, while Becks infects Colin, Corin and Coral, and Bob infects Chuck, Chaz and Charlie. Then the R number is 3 – each 1 infected person is infecting 3 others.

(This is what it was like at the start, back in February or March. The R number in the UK then was probably not far off 3, so that as the days went along, the numbers of infected people went like 1 → 3 → 9 → 27 → 81 → 243, or a bit more than that. You can see why the virus managed to turn into an epidemic.)34

The R number being 3 wouldn’t have to mean that every single person infects exactly 3 others. It might not be as neat as that. For example,

  • Maybe Becks happened to be at home in the few days she was most infectious. Maybe Bob did nip to the shop, but wasn’t there long, and wore a homemade mask to catch most of his germs. And because of these circumstances, neither of them ended up infecting anyone.
  • Maybe Bell works somewhere busy, and infects nine other people before she starts to feel a bit rough and realises she’s got the virus.

But even though it wasn’t exactly 3 per person, it was still 9 people in that example who got infected in the “next round” of infections. It went from 3 in one round, to 3×3 in the next, so R would still be 3. For counting up R, it doesn’t matter exactly who passed it on – it’s about how many people are infected overall, in the “next round”.

(In fact, that last example is more like what we’ve seen with covid, when researchers have tracked back who gave it to whom. Typically, it’ll turn out that some people had spread it to several others, whereas some for whatever reason hadn’t spread it at all.)

It’s worth noting that the R number is never all that precise, & especially not if not many people are being tested, or the overall numbers are low.35

The epidemic growing, shrinking or staying the same size

When R is 1, we’re talking about an epidemic that isn’t growing – and it isn’t shrinking either. As one lot of people get over it (or die, or stay disabled), another lot have caught it, so that the overall number of infected people is holding steady.

For example, if you have 1,000 people already infected, and an R number of 1, in a few days’ time you’ll have a new lot of 1,000 people infected. And a bit later, you’ll have another lot of 1,000 people infected. It doesn’t actually end.

When the R number is bigger than 1, we’re talking about an epidemic that’s growing. The bigger the number, the more it’s growing.

When the R number is smaller than 1, we’re talking about an epidemic that’s shrinking. The smaller the number, the more it’s shrinking.

Pointy maths symbols for bigger and smaller

When R is bigger than 1, you might see this written down as “R>1“. The pointy thing is from maths. It’s called “greater than”, so that statement is saying “R is greater than 1”.

When R is smaller than 1, that one could be written down as “R<1“. That way round, it’s called “less than”, and you’d read it out as “R is less than 1”.

It’s easy to remember which way round is which, because the little pointy side is always towards what you’re saying is the smaller number.

Growth rate

R isn’t exactly the same as how fast the epidemic is spreading. How fast this all happens would also depend on the time lag between “rounds”. If Adrian catches it on Monday, starts to become infectious on Thursday, sees Bell on Saturday and gives it to her, that’s a time lag of 5 days, which would be typical for covid.

For this reason, it can also be useful to talk about the “growth rate” of infections, which includes time. Typically it would be described as percent of increase or decrease per day.

For example, let’s say on a Monday, 1,000 new people test positive. On Tuesday, it’s “only” 970 new people. You could describe that in terms of: minus three percent, -3%.

(It might sound odd that you still call it “growth” even if it’s shrinking. You’re describing the “shrinking” as “negative growth”, which is why it has the minus sign.)

R and growth can change

R partly depends on the nature of a particular disease – for example, how it travels from one person to another, or whether you need a big or small amount of virus to start the illness. For example, one of the reasons that measles is so infectious is that you only need a tiny bit of measles virus to kick off the disease.36

Luckily for us, the spread of a virus also depends on what people do.

That’s why we’ve tried out staying at home, meeting outdoors, keeping 2 metres apart, washing our hands, and/or covering our mouths & noses. Those are all to try to make it harder for the covid virus to jump to the next person. And it worked: fewer people got infected.

R can be different in different areas

You don’t have to only look at the R (or the growth) for a whole country. It’s often useful to look at a particular area, or a particular type of place – if you can do enough testing and tracing to find out. And in different places or situations, you might find different results.

For example,

  • Sometimes the virus will spread more easily in a town (e.g. because of more people living closer together, or more people using public transport), and less in a rural area.
  • There was a suggestion around May/June that the virus had been spreading more within care homes, compared with the wider community.
  • This page at shows differing estimates of R for the different “regions“. For example, the figures for 10 July show that R in the Midlands was then estimated at 0.7 to 0.9, whereas in the South East, it was estimated at 0.8 to 1.0.

(When it’s given as a range, like the “0.8 to 1”, it’s acknowledging that we don’t have enough info to be 100% sure, and/or it’s different in different places within that region.)

Part of what’s tricky about doing the estimates is the time-lag – the fact that the people who are getting infected today won’t show up till maybe next week, or the week after.

The UK’s R number going down and up

From the testing that’s happened so far, and from the numbers of people going to hospital, we have some idea of how R has been changing in the UK.

  • Before the so-called “lockdown” (which was really more of a “slowdown”, in my opinion), it looks as though R in the UK was around 2.4 to 4.37
  • Towards the end of April, while lots of us were staying at home, R in the UK had got down below 1. At the time, it was said to be between 0.6 and 0.9.38
  • A study that’s just been previewed has looked back over infections in May: for England only, and not counting hospitals & care homes. According to that, the R number in the community in England during May was down to 0.57.39 (Due to spread in care homes and hospitals, the overall UK number at that time was between 0.7 and 1.0.)40
  • Since the middle of May, the government has been offering an official estimate of R each week, based on things like hospital admissions, testing people for the virus, and surveys.41 The latest update as I write, from 17 July, was 0.7 to 0.9 for the UK as a whole, 0.8 to 1.0 for England.

How can England’s be higher than the UK’s? It’s because Scotland currently has a lower rate of transmission than England. They’re down to about 1 death a week at the moment. The most recent estimate of R in Scotland (published in their own report from 15 July) is 0.5 to 0.9.

I suspect the England number is actually closer to 1 than 0.8, because the COVID Symptoms Tracker project has already said that based on their data up to 11 July, it looks as though the UK epidemic has stayed roughly the same size since the beginning of July:

The latest data suggests that the number of daily new cases has now stopped dropping, with a definite leveling off of cases since the beginning of July. The latest figures were based on the data from almost 3 million users and 14,429 swab tests done between 28 June to 11 July.

Don’t only look at R by itself

In practical terms, it’s important to keep in mind how many people are infected right now, as well as how fast the infection is spreading.

If someone told you R was 2, that could be referring to one person infecting two people – or it could be referring to ten thousand people infecting twenty thousand people, which is a much bigger problem!

If the R number is a little bit smaller than 1, the epidemic is shrinking – but you still might not be doing great if the overall numbers of infections are big. For instance, if R is 0.8 and you have ten thousand people already infected, the “next round” will still mean another eight thousand people catching the virus.

If it keeps on shrinking like that, eventually it will shrink away to nothing, but a lot of people would have died in the meantime.

This is rather like what we’ve got in the UK at the moment: the Symptom Study team are estimating 28 thousand people with symptoms as of 19 July, with a couple of thousand new infections each day.


I’m not going to get into all the details of testing here. I’ll just address the question of: What statistics are useful to keep track of, to see how well the testing processes are working?

The government started off by making a big deal of how many tests were being done overall.

There was a palaver over them fiddling the numbers – partly by including tests that had only been posted out to someone’s house and not actually done, and partly by, if one person had spit/snot samples taken both from their nose and from their throat, counting that as two separate tests!

But aside from whether it’s being counted up correctly, it’s maybe not the best measure to focus on anyway. You want to know how many people are being tested – not just how many tests are done.

Leaving aside the “cheat” of counting two parts of one test separately, it’s actually not that unusual for the same person to be tested twice. For example, someone who’s ill might initially test negative, then get re-tested a few days later, to help doctors work out what’s going on with them. That’s genuinely two tests, but only one person.

Two other key measurements to track are:

  • How quickly people can get a test when they want one.
  • How quickly people then receive their results.

If someone does test positive, you want to be warning the people they’ve spent time with – preferably before that “next round” of people starts to be infectious. So the speed of turnaround is crucial.

Are you testing enough?

Why does it matter how many people got tested, as long as everyone can get a test when they need one?

It’s partly because that then forms part of another measure: the “positivity rate“.

If you test a thousand people for bits of virus in their body, and only three or four results come back positive, you can be fairly sure that you’re not missing a lot of cases by failing to test enough.

If you test a thousand people, and two hundred results come back positive, it’s very very likely that there are some other people out in the community who didn’t get tested, but do have the virus. If you seriously want to get on top of the situation, that means you need to ramp up the testing and find those other people.

In other words, if a big proportion of people turn out to test positive, it’s probably a hint that a lot of other people have the infection as well, who never got tested. It means you aren’t doing enough testing yet to know where the virus is.

So the “positivity rate” is useful for working out whether you’re doing enough testing, or not.

Note, the “positivity rate” isn’t the same as “what percentage of people in this area have the virus”. This is “what percentage of people who got tested have the virus”.

The World Health Organisation suggests aiming for no higher than 5% positivity before loosening any precautions.42 Some places are aiming for 3%.

Of course, eventually you want the positivity rate to come all the way down to 0%, as the virus is eliminated from a country. But meanwhile, you want to be doing enough testing that you’re finding the infections which are there.

Tell me if I got something wrong / updates policy

By the nature of a post like this, parts of it will be outdated soon. I don’t know if I’m going to update it to any significant extent on things which were right at the time; if something changed drastically and I thought it might mislead people to have the old information there uncorrected, I probably would put a note in.

But it’s also possible I’ve made mistakes already! especially as this went through a lot of drafts while I was researching. So please tell me if you notice I got something wrong.

That’s it for now! Hope it was useful!

Skip footnotes and jump to comments

1. “Covid and the virus which causes it“: The illness is officially COVID-19, and the virus is officially SARS-CoV-2. Even though the name “covid” isn’t as exact, I’ve decided I’m calling it that “for short” now, for ease of reading.

2. “Can cause blood clotting“: At a medical centre in New York, doctors investigated the bodies of 7 people who’d died from covid. All the bodies had unusual blood clotting, even if the person had been given anti-clot drugs while they were alive.

In this series of seven COVID-19 autopsies, thrombosis was a prominent feature in multiple organs, in some cases despite full anticoagulation and regardless of timing of the disease course, suggesting that thrombosis plays a role very early in the disease process.

3. “Died of pneumonia“: More on death certificates in the section about the Office of National Statistics.

4. ““: As well as the total, there’s a graph, going back to the start of the epidemic. On touch-screen, you can touch the graph line to pick out specific days, or if you have a mouse, “hover” the mouse pointer over the line.

They’re beta-testing a new page layout as I write: in the newer design, you have to click a tab which says “Deaths” to get to that graph.

5. “On an April calendar date“: There is a bit of wiggle room around death registrations and calendar dates: what if there’s some reason that more or fewer registrations happen on a particular day, which has nothing to do with the actual deaths?

For example, if the office is closed on a Sunday, you won’t get registrations that day – and it wouldn’t mean nobody died on a Sunday.

In the Spring, you do typically get a little uneven bit in the flow of death registrations, when it’s the Easter weekend. If you were looking at the numbers for just those few days or that week, you’d want to take it into account. But it wouldn’t have much effect on the numbers for the month overall. (If Easter falls near the end of one month, so that nearly all the registrations for that weekend get bumped on into the next month, there would be some effect.) So for simplicity, I’ve skipped over that in the main explanation.

6. Death law traditions: This is a bit off-topic, which is why I’ve put it in a footnote, but for example: if someone disappears “presumed dead”, there already are laws and discussions about how long you have to wait before you can treat them as “officially dead” for legal purposes.

7. “Average of those five previous Aprils“. Here’s my working:

2019 (provisional) 41,164
2018 43,478
2017 36,422
2016 43,755
2015 42,286
Average of those 5 years 41,421

Technically this average is known as a “mean”: that is, I added the numbers for all five years, then divided by five. I’ve rounded the 41,421 to 41,400 for ease of explaining.

The provisional figure for April 2020 was 83,504 when I looked it up, which I’ve rounded to 83,500 in the discussion.

(Note, I’ve chosen to use just England for this example; obviously each figure for England and Wales together would be a bit higher.)

8. “At the moment“: The ONS people confirm each year’s exact numbers after the end of the year. But as they had all of May to chase up late entries before publishing the spreadsheet I saw, that number is probably pretty close.

9. Note on viewing Twitter threads: You can view Twitter even if you don’t have an account (although some writers will have set their tweets to be private). A thread means several tweets linked together. The link I’ve given is to the first tweet in Nick’s thread. If the rest of the thread doesn’t show up straight away, you may need to click on the date of the tweet, or (in some apps) just the main middle bit with the writing. You should then be able to read down the page for the other tweets in the thread.

10. “Aren’t getting enough oxygen“: “Why don’t some coronavirus patients sense their alarmingly low oxygen levels?” Short answer: it’s because if you’re not breathing properly, normally what “sounds the alarm” to your body is the build-up of carbon dioxide, not the lack of oxygen. So if carbon dioxide isn’t building up, your body won’t notice at first that there’s a problem.

Doctors do already know something about how this plays out, because of people going up to high altitudes where the air is “thin”. Here’s a paper discussing the similarities: COVID-19 patients with respiratory failure: what can we learn from aviation medicine?

11. “During the spring of 2020“: Probably mostly during Feb, March and April, according to Nick Stripe’s analysis. He points out that by May, the excess death numbers were matching up with the covid death numbers, probably in part because doctors were getting better at spotting the different ways covid can kill you. So the excess deaths which don’t match up with known covid deaths would’ve mostly been before then.

12. Note on using the sums from the Financial Times: I could have done my own independent calculation of excess deaths for the UK, as well. Maybe if I were starting this article over again, I’d have done that one instead of the England/April example. But as I have done the England/April one, and doing another one would be a lot of faff – and as I’d probably have heard about it if the FT’s analysis was especially bad – I’m taking on trust that the FT people and Jamie Jenkins aren’t far off.

13. Typically 5 days to symptoms: one paper looking at that is The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and Application.

There were 181 confirmed cases with identifiable exposure and symptom onset windows to estimate the incubation period of COVID-19. The median incubation period was estimated to be 5.1 days (95% CI, 4.5 to 5.8 days), and 97.5% of those who develop symptoms will do so within 11.5 days (CI, 8.2 to 15.6 days) of infection. These estimates imply that, under conservative assumptions, 101 out of every 10 000 cases (99th percentile, 482) will develop symptoms after 14 days of active monitoring or quarantine.

14. “Might still do things to their body that they weren’t aware of“: Researchers did scans of 37 people who had covid with no symptoms. Here’s the paper: Clinical and immunological assessment of asymptomatic SARS-CoV-2 infections.

They found that 11 of the 37 people had what they call a “ground-glass opacity” (a hazy bit on your scan, which looks a bit like that type of glass with a matt, non-shiny finish). Another 10 had “stripe shadows and/or diffuse consolidation”, whatever that is.

So in other words, more than half of the people without noticeable symptoms had some kind of weirdness going on in one or both lungs.

To put that in perspective, though, someone with a minor infection like a cold wouldn’t usually have a CT scan. (That’s “Computerised Tomography”, a type of scan which shows more lung detail than an X-ray would.) So we don’t know how common it is to get a minor lung abnormality which quickly clears up.

15. “15 to 22 days after symptoms start“: That’s from this research study, which summed up lots of people’s experiences: Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study (PDF).

The median time from illness onset (ie, before admission) to discharge was 22·0 days (IQR 18·0–25·0), whereas the median time to death was 18·5 days (15·0–22·0; table 2).

Where they refer to “discharge”, that’s the people who got better and went home from hospital.

“Median” is a type of average. It’s the one where you line up all the answers in a row, and pick the middle one. So e.g. if the answers to something were 1, 2, 2, 3, 10, the median would be 2.

16. Mostly not infectious after day 5: The study is “Contact Tracing Assessment of COVID-19 Transmission Dynamics in Taiwan and Risk at Different Exposure Periods Before and After Symptom Onset“.

The attack rate was higher among the 1818 contacts whose exposure to index cases started within 5 days of symptom onset (1.0% [95% CI, 0.6%-1.6%]) compared with those who were exposed later (0 cases from 852 contacts; 95% CI, 0%-0.4%). The 299 contacts with exclusive presymptomatic exposures were also at risk (attack rate, 0.7% [95% CI, 0.2%-2.4%]).

“Attack rate” is what they call it when the second lot of people catch the virus from the first lot. What they’re saying is that 852 people spent time around someone who was on Day 6 of covid symptoms or later, and none of those 852 people caught it.

However, they don’t say it’s impossible for that to happen – only that it didn’t happen in the group of people they were studying.

(Their research also confirms again that you can catch the virus from someone who’s not yet shown symptoms. “Exclusive presymptomatic exposures”, means “only spent time with the infected people before their symptoms started”, and some of the 299 people they refer to in that bit did catch it.)

17. “Long covid” stats: Originally, the daily update number from the COVID Symptom Study did include people still ill after the initial infection had gone. But they revised their counting method on 8 July, explaining:

We will be separately estimating the numbers of people with long duration symptoms and updating our website with these figures. We want to emphasise that there are lots of people who continue to have symptoms long after they are no longer infectious – this is an area of huge importance, and one that our researchers are very keen to understand better with your help.

This is a good example of stats people deciding to come up with new “what are we counting” rules. More information came in (i.e. they could see from people’s symptom reports how many people still hadn’t recovered yet after months), and they took time to think over freshly what would make most sense.

I’m guessing they thought it wouldn’t make sense to mix together the count of “people who are part of the ongoing epidemic, possibly still infectious” with the count of “people who aren’t well, but probably aren’t infectious any more”.

More on the “long-haulers” in the section about lasting disabilities.

18. “If the virus becomes unviable more quickly at different temperatures“: We already knew that the flu virus survives better in dry, cold air, which is part of why flu illnesses often peak in the winter. It remains to be seen what’ll happen with covid. Good discussion here: Seasonality of SARS-CoV-2: Will COVID-19 go away on its own in warmer weather?

19. “Other factors, like some countries having more older people“: The Euromomo web site compares a measurement called Z-scores. Disclaimer, I’m not 100% sure I understood how they’re using it, but I think the point of it is you compare excess deaths with usual deaths for the time of year, and look at the ratio of those. You don’t only compare excess deaths with the overall population. This then compensates for the fact that one country might usually have fewer deaths per million than another these days, for reasons like currently having lots of younger people. If you understand the Euromomo Z-scores properly and can explain them to me, please do :-)

20. “67 million“: It looks as though the Financial Times has been using the UK population figure from mid-2019, around 66.8 million. I’m not sure of their source, and whether they actually know it is that now, or they’re using the most recent official number on principle even though it was a while ago. Anyway, by rounding to 67 million, I’m erring on the side of, if anything, underestimating the death rates. (because a bigger population means the deaths look fewer in comparison.)

21. Financial Times estimate, and other related stuff: Although some more actual deaths have happened since then, most of the excess deaths happened in the spring. So it doesn’t make a huge difference that this article is from May.

In this article, I’m not getting into why so many people have died here compared to other countries – maybe I’ll write about that another time.

22. Official deaths per million: DHSC’s count of covid deaths is 45,300 at 19 July (from the government’s web site). 45,300 divided by 67 million gives 676 deaths per million.

23. Denmark and Greece: As reported at Worldometers on 19 July.

Side note, I’m not an enormous fan of that site, because I don’t think they do enough to flag up that these are the known cases and there could be more, or that different countries are going by different definitions. But it is a very convenient place to get a rough summing-up of what’s happening across the world.

24. “An antibody is a tiny bit of your immune system“: Technically it’s a protein (not a cell).

25. More details on the Spanish research: 61 thousand people had a fingerprick test “on the spot”. About 52 thousand also gave blood which was tested later in a lab as a double-check.

The researchers started by testing for two different types of antibody, known as IgG and IgM. But they ended up not using the IgM results, as the IgG test was more reliable.

In research like this, you’re trying to find out about a big group (the whole country) by looking at a smaller group (the people you tested, also known as a “sample”). So you have to be careful that the small group really does have the same illness patterns (or whatever you’re interested in) as the big group.

Your results are generally going to be more reliable if you test a lot of people – which the Spanish researchers did.

And you also try to make sure that your “sample” is similar to the overall population, more or less. For example, if you know that a quarter of the population is in the age range 40 to 59, then you don’t want to only test people in their twenties – ideally you probably want about a quarter of your sample to be 40 to 59 as well. The Spanish researchers did some extra sums so that the final result would be more similar to the overall pattern in the country.

Here’s the report: “Prevalence of SARS-CoV-2 in Spain (ENE-COVID): a nationwide, population-based seroepidemiological study“.

26. Covid-specific T cells in people who didn’t have the antibodies: The preprint with that result is “Intrafamilial Exposure to SARS-CoV-2 Induces Cellular Immune Response without Seroconversion“, from 22 June in France. (Seroconversion means when someone’s blood develops the antibodies for something.)

Exposure to SARS-CoV-2 can induce virus-specific T cell responses without seroconversion. T cell responses may be more sensitive indicators of SARS-Co-V-2 exposure than antibodies. Our results indicate that epidemiological data relying only on the detection of SARS-CoV-2 antibodies may lead to a substantial underestimation of prior exposure to the virus.

In another study from June, researchers looked at about 200 blood samples from a variety of people in Sweden, including from people ill at the time, people who’d had it earlier (Feb/March) and recovered, people who had been exposed to covid but hadn’t had symptoms, plus some previously-stored blood samples from 2019 before the epidemic started.

Here’s the preprint: “Robust T cell immunity in convalescent individuals with asymptomatic or mild COVID-19

Like the French research, one of the things they found out was that some of the people who’d been exposed to covid didn’t have covid-specific antibodies, but did have covid-specific T cells:

SARS-CoV-2-specific CD4+ and CD8+ T cell responses were present in seronegative individuals, albeit at lower frequencies compared with seropositive individuals (Figure 4F).

(Seronegative means antibodies weren’t detected, seropositive means they were detected.)

27. T cells from the first SARS in 2003: “Memory T cell responses targeting the SARS coronavirus persist up to 11 years post-infection” had already shown in 2016 that the SARS T cells lasted that long. Researchers recently confirmed they can still persist to this day: “Different pattern of pre-existing SARS-COV-2 specific T cell immunity in SARS-recovered and uninfected individuals“.

We then show that SARS-recovered patients (n=23), 17 years after the 2003 outbreak, still possess long-lasting memory T cells reactive to SARS-NP, which displayed robust cross-reactivity to SARS-CoV-2 NP.

By the way, what they’re saying about the cross-reactivity in that quote I think means: surviving the first SARS gives you a bit of protection against COVID-19. They’re not saying it would completely stop you getting covid, but it could give your immune system a head start at dealing with it.

28. T cell testing more bothersome: See an interesting discussion at Derek Lowe’s blog. He says:

it’s unfortunately a lot more labor-intensive to profile CD4+ and CD8+ T cells in people than it is to profile their antibody responses.

I think the gist of it is: you get some blood from the person, and add some covid-virus-alikey pieces to the sample of blood, and look at how the T cells respond. I saw elsewhere that you need a bigger blood sample than the few drops from a fingerprick test, and it takes longer.

29. Timing of antibodies: for example, “Antibody Responses to SARS-CoV-2 at 8 Weeks Postinfection in Asymptomatic Patients” says

Seroconversion in asymptomatic patients might take longer.

… which translates as: having covid with symptoms might mean you’re making antibodies quicker than the people who have the virus and no symptoms.

30. “1 person in every 15” had antibodies: Here’s the initial results. Here’s a bit more about how they did the research.

As of 24 May 2020, 6.78% (95% confidence interval: 5.21% to 8.64%) of individuals from whom blood samples were taken tested positive for antibodies to the coronavirus (COVID-19). This is based on blood test results from 885 individuals since the start of the study on 26 April 2020.

There’s since been an update:

Between 26 April and 8 July, 6.3% of people tested positive for antibodies against SARS-CoV-2 on a blood test, suggesting they had the infection in the past.

It doesn’t necessarily mean that the number of people with antibodies genuinely went down in that time; that could be the case, due to them “fading away” for some people, but it’s also possible that was random chance, in picking different people to sample at different times.

31. 1 in 72 died: 65,000 over 4,700,000. With fractions, you can do whatever you like to them as long as you do the same to the top as you do to the bottom. So divide both by 65,000. That converts it to 1 over 72.3-and-some-more-decimals. Rounding off, let’s say 1 over 72.

32. “Delirium, brain inflammation, stroke and nerve damage“: That’s a quote from this briefing from University College London.

Here’s the actual research paper they’re talking about: The emerging spectrum of COVID-19 neurology: clinical, radiological and laboratory findings (in “Brain” journal).

33. Lung damage from SARS: “Long-term bone and lung consequences associated with hospital-acquired severe acute respiratory syndrome: a 15-year follow-up from a prospective cohort study“. This study was based on following up with 71 people who’d survived SARS in 2003. (The “bone consequences” were to do with having high-dose steroids for treatment while ill. Only the lung damage was to do with SARS itself.)

Of the 71 people studied overall, 46 people took part in getting their lungs tested in 2006:

The outcomes in 2006 revealed that 10 out of 46 (21.74%) patients had restrictive ventilation dysfunction. Sixteen out of 46 (34.78%) patients had reduced diffusion capacity with an ~70%–80% predicted value, indicating a mild reduction.

I think what that means is that 10 of the 46 couldn’t breathe in the normal amount of air, whereas 16 of the 46 weren’t getting the normal amount of oxygen from the air. It’s not entirely clear to me from how they’ve written it up whether some of those are the same people. But at any rate, that’s 2 to 3 years later, still with the SARS after-effects. And some still had damage even at the later follow-up in 2018.

34. “Probably not far off 3“: More discussion of this number later, when we get onto how R has changed in the UK.

35. R number not all that precise: Here’s a good thread about that.

36. “R partly depends on the nature of a particular disease“: R0, pronounced R-nought or R-zero, is sometimes used to refer to the aspect that depends only on the virus itself. It means how fast a virus would spread in a situation where no-one was immune, and where no-one took any particular measures to stop it. For example, the R0 of measles is sometimes given as 15: each infected person would infect about 15 others, if they weren’t immune already.

I think there are limits to this concept, in that humans are never doing nothing. You can say “without any changes to their behaviour to try to stop the virus”, but before they change their behaviour, they were already living in specific ways, not necessarily identical from one community to another, which can affect the virus’s chances. If it can transmit via food, what are the food-sharing customs? If it can transmit by air, how many people typically sleep in the same room, and how much time do people spend outside? If it can transmit via poo, how many people share a toilet, and do people have easy access to soap and water? Even at the stage when no-one’s immune, the transmission rate must depend only partly on the virus itself, and partly also on circumstances in the community.

Measles is possibly quite a good example of this too, because in reality, different researchers have found anything from R0=12 to R0=18.

It’s still useful to have worked out that it’s in that range. It means we can be pretty sure that measles is more infectious than flu (for which R0 is typically in the range 1 to 2). And it makes sense to use the expression R0 in comparing one disease with another against the background of the same circumstances. However, if you don’t describe the surrounding circumstances, then talking about R0 as a definite nailed-down thing is misleading – similar to “how long is a piece of string, let’s pretend we’re all talking about the same piece of string”.

Researchers are still trying to work out covid’s R0, by looking at how fast the infection rate spread in different places. Here’s a paper suggesting between 4.7 and 6.6; on the other hand, I’ve also seen estimates of 2 to 3. So at the moment, it’s looking more infectious than flu, but less than measles would be if there were no such thing as a measles vaccine.

Anyway, be that as it may, the current R is what’s more relevant here.

37. UK R number 2.4 to 4: This rather good article from Jasmina Panovska-Griffiths at The Conversation links to four different studies, and sums up:

At the onset of the epidemic in the UK, different studies estimated R in the UK to be 2.4, 2.6, around 3, or between 3 and 4.

From other stuff I’ve read, I suspect it maybe wasn’t as high as 4. It did look like a very fast take-off in March, but it’s possible that people had brought it into the country in a lot of different places before anyone noticed, which could’ve made it look faster when people did notice. Iceland did loads of testing early on, and their data shows that UK people were already bringing the virus to Iceland in February:

We found that a large number of the original cases came from the UK.

The spread of the virus was much greater in the UK early on than people realised. They might have even preceded those from the Alps. We don’t know exactly, but these cases could be from as early as February. …

… as soon as the population screening started, it was dominated by UK-origin virus, so this was spreading quickly through the Icelandic population from February.

38. “Between 0.6 and 0.9“: I got this figure via an article in Wired, by Matt Reynolds:

At the April 30 press conference, the UK’s chief scientific officer Patrick Vallence said that the UK’s R0 was between 0.6 and 0.9 while the figure in London was between 0.5 and 0.7.

39. R in the community in England in May: The original announcement from Imperial College is here: “First findings published from largest home COVID-19 testing programme. The paper itself is “Community prevalence of SARS-CoV-2 virus in England during May 2020: REACT study” (PDF).

The REal-time Assessment of Community Transmission (REACT) study is a nationally representative prevalence survey of SARS-CoV-2 virus swab-positivity in the community in England.

40. Overall R in the UK in May: The background context (the higher R if you include care homes & hospitals) is from a BBC article mostly about the Imperial/REACT research, “Coronavirus: R number ‘lower than thought’ before lockdown eased in England“.

41. Basis of the estimates of R for the UK: On the same page as the number, there’s an explanation of where they get the data and who discusses it.

The growth rate and R are estimated by several independent modelling groups based in universities and Public Health England (PHE). The modelling groups discuss their individual R estimates at the Science Pandemic Influenza Modelling group (SPI-M) – a subgroup of SAGE. Attendees compare the different estimates of each and SPI-M collectively agrees a range for which the values are very likely to be within.

42. Positivity rate 5% aim: I haven’t seen the WHO’s original announcement. I’ve seen it mentioned a few places, e.g. from Johns Hopkins university:

On May 12, 2020 the World Health Organization (WHO) advised governments that before reopening, rates of positivity in testing (ie, out of all tests conducted, how many came back positive for COVID-19) of should remain at 5% or lower for at least 14 days.

Appreciation, criticism & new ideas all welcome...

Optional HTML tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

Comments are moderated. Please no name-calling; please speak for yourself from your own life, or else say where you got the info. Thanks :-) Why these commenting guidelines?