Estimating the risk for university students and staff

More universities have communicated to offer-holders and/or implemented special action to handle the Admissions situation in the past two weeks (cf. my previous post). However, there are still hesitations among some institutions on whether to open campus next autumn or not, and if so which measures should be applied.

Are these hesitations supported by data ? Let’s do a bit of fact-checking.

According to NHS England’s reports, there have been approximately 10-20 Covid-19 deaths in the 20-24 age group so far. As 50% of this age group are students, let’s assume that about 5-10 students might have been among them (probably less, taking into account socio-economic factors).

Around 500 students die every year in England and Wales (ballpark number), so we’re talking about a maximum of 1-2% of annual student deaths. By comparison, suicides and drugs poisoning/misuse each represent roughly 100 student deaths. That’s about 200 in total, i.e. 40% of student deaths.

Let’s look at it from another angle, using a source that has more portability into the future, for example the University of Bonn study I referred to in a previous post. It was conducted in real conditions in a hotspot (Gangelt), so it cannot be accused of minimizing mortality. This study yielded a mortality rate (probability of dying if infected by the disease) of 0.37%, which is pretty much in line with most common estimates. Please note that the average age in Gangelt (43.7) is higher than in the UK (39), so we would expect mortality rate in England and Wales to be a bit lower, but let’s keep the higher value, this will only make the demonstration more valid.

Let’s apply this overall mortality rate to the current age distribution of Covid-19 deaths from the latest data published by NHS England. Through age-based regression, the estimated Covid-19 mortality rate for the 20-24 age group is approximately 0.0041%. In comparison, the estimated flu mortality rate in Europe for the same age group (obtained using the same regression method as for Covid-19) would be about 0.0027%.

Therefore, for the 20-24 group, the Covid-19 mortality rate is only roughly 1.5 times the normal flu mortality rate, even with unfavourable hypotheses.

This makes complete sense. Of course, Covid-19 is significantly more dangerous than flu overall. But this lethality difference is much more marked for older age groups, whereas for younger age groups the gap is quite small.

Let’s now use the same method to compare estimated mortality rates for staff, using the HE staff age pyramid published by HESA. The estimated Covid-19 mortality rate for this age structure would be about 0.1040%, versus 0.0529% for flu.

Therefore, for a typical HE institution, the Covid-19 mortality rate for staff is roughly twice the normal flu mortality rate, there again with conservative hypotheses.

Incidentally, this factor of 2 is exactly the H1N1 2009 average mortality rate compared to normal flu’s. I cannot recall that any university closed for swine flu, (although 80% of the H1N1 deaths occurred in the under-65 age group, which is precisely the opposite of Covid-19).

Ultimately, it will be for British universities to assess whether the associated risk level is acceptable or not, and whether social distancing should be maintained, bearing in mind the many mitigation possibilities that will be available, such as testing, tracing, protecting students and staff with conditions, self-isolation, etc.

Friends and colleagues to whom I have mentioned these numbers are a bit surprised. Given what they hear daily about the Covid-19 crisis from the media, they expected much more spectacular gaps with flu. Even an unmitigated ratio of twice the flu mortality rate is unimpressive and well below any major cause of death for the adult population. Within the viral domain, SRARS 2003 was 96 times the flu mortality rate, MERS 2012 was 356 times, Ebola 500 times, and plague (so much for those who ventured to use that term to describe the current pandemic) 600 times, i.e. 60% mortality rate.

I understand their surprise: how can members of the public NOT be confused by Covid-19 data ? Being exposed to a daily mixed bag of correct and incorrect information, unchecked or mis-commented figures and erroneous conclusions, sometimes within the same source of information, is confusing. As this confusion is set in the context of the media’s natural tendency towards inflationist sensationalism, this accumulation erodes the public’s sense of scale and creates a general misrepresentation of the situation which, powered by fear as it is, becomes extremely difficult to challenge.

The problem is not just a few tabloids selling a good story. It’s deeper than that. The latest analysis published by the Financial Times and subsequently relayed by much of the high-end press illustrates this peculiar situation.

The article sets out to estimate Covid-19 excess mortality. This is a major step in the right direction: as I explained some weeks ago, excess mortality is the only measure that gives a sense of scale to the current crisis.

Unfortunately, the headline (‘UK coronavirus deaths more than double official figure’) and opening statement (‘The coronavirus pandemic has already caused as many as 41,000 deaths in the UK’) are both wrong. The error lies in three methodological problems that lead to an overestimate:

(1) There isn’t necessarily ONE cause of death, therefore this raises the question of what we define as a coronavirus-attributed excess death. That is the difference between dying ‘from’ and dying ‘with’ coronavirus. Even the NHS is not so clear about it and has changed its counting method several times. Having worked on data appertaining to causes of death in the past, I confirm this is more complicated than it looks, as explained by Carl Heneghan, professor of evidence-based medicine at Oxford University: ‘You get the death certificate, but you’d also need to have the medical notes to hand, and coroners’ reports… That is actually a large job. A very big research study’.

(2) More importantly, the cause of current non-attributed excess deaths, referred in the article as ‘mystery’ deaths, is still unclear. Some might be due to people not being able to (or being unwilling to) access care because of the Covid-19 crisis. More generally, they could be assigned to the health system’s burden, resulting in the difficulty to deliver on both Covid-19 and other usual conditions. This is not unknown in epidemics to have also higher mortality from other causes. However, the FT analysis takes a one-sided vision and assumes these are all related to Covid-19. At best, they can be counted as indirect deaths, not direct. By indirect, we mean possibly dying ‘because’ of Covid-19, not ‘from’ or ‘with’. It’s a different count and cannot be included in a ‘UK coronavirus deaths’ headline.

(3) Much more importantly, a complete estimate of Covid-19 excess mortality should take into account gaps to average mortality over a whole year period starting in autumn 2019, not just a few weeks, because spring mortality, autumn mortality and so forth are not independent: those who die now didn’t die before, and some of those who are dying now because of the virus might have died later, nobody dies twice !

I was initially surprised to find that the experts that relayed this information did not mention the third issue, because that kind of mechanism literally jumps at you when you’re used to analyzing such numbers (especially given the Covid-19 age bias).

As it seems unlikely they didn’t notice it, I’ve been thinking about the possible reasons behind this singular omission:

  • One of the features of the current crisis is that no scientist wants to be seen as the one who is underestimating gravity and risk. If you’re wrong, it’s OK to be wrong by going over the top, not the other way round. This influences the way scientists communicate.
  • Issue (3) is about current excess mortality being compensated by future under-mortality. If we say that Covid-19 is causing some people to die earlier, this means we are storing under-mortality for the rest of the year. In which case this future under-mortality should be deduced from the current coronavirus-attributed excess mortality. But who wants to be caught in the media explaining that sombre reality ?
  • Worse, the evidence for issue (3) will be muddied by the future excess mortality we are currently storing because of the NHS burden. To take but one example, around 1,000 cancers are normally diagnosed every day in the UK: that’s not happening at the moment, so if you count lockdown time, plus the necessary time for the NHS to absorb the backlog, that could mean about 60,000 undiagnosed cancers, with many implications on delayed treatments as well. Consequently, future under-mortality and excess mortality will be adding one another and compensating in some way, but we don’t know in what proportion. There will be months, perhaps years, and considerable statistical work, before we can unravel these many causes and quantify them.

However, this silence is somewhat worrying because it indicates that even competent experts find it difficult to go against the inflationist trend, which leaves it unmoderated [edit 13/04: and even amplified, as the same mistake is repeated one week later]. In the case of the FT article, a critical review would have rather damped the ‘more than double’ claim, but it didn’t happen. That’s now in the headlines and that’s what the public will remember, growing more fearful than they probably should as a result, even though the claim is unambiguously wrong.

The ballpark I published one month ago still holds, i.e. in the region of 25,000-30,000 excess deaths at the end of the current wave. This is not only less than the ‘already 41,000 deaths’ of the FT article, it is also within the scale of a bad flu year concentrated over time.

This key difference between temporary burden and overall scale is also confirmed by Carl Heneghan, quoted in the same FT article: ‘The 2017-18 seasonal flu outbreak may have killed 50,000 in the UK but the reason we did not get alarmed then was that they were spread over many weeks’.

There is an important caveat: this episode might be within the scale of a bad flu year overall, but not so for the elderly. The lethality of Covid-19 is far greater than flu for the 70+ age group, which accounts for 85% of all UK Covid-19 deaths, as shown in NHS England’s reports and the new total number of deaths that include care homes. This will be the key factor behind ways to lift lockdown.

However, the caveat works both ways: a bad flu scale overall with a strong bias towards the elderly, therefore not much worse than a flu scale and risk for others, and that is exactly what the mortality rates estimated for university students and staff at the beginning of this post show. Which leads us to an inescapable and sad conclusion: the whole HE sector (students, staff and institutions alike) is a casualty of the blanket lockdown.

What this perspective also shows is that despite what the avalanche of daily coverage, death tolls and government briefings may suggest, there have been few new statistical facts on the Covid-19 crisis in the past six weeks. The first new item is that the comedown after the peak in some countries is a bit slower than expected, which has led to a revision of total deaths estimates (with previously identified relationships and patterns unchanged; all other predictions, including sources of likely under-reporting, UK’s high death toll and visible comedown by end of April, have followed their course).

The second new set of information is that we are now in a better position to locate the date of the peak in each country. From there on, there are many implications to analyse in terms of how the crisis has unfolded comparatively since the beginning of social distancing measures.

Maybe next week, if I — and the four fabulous friends (editor, data consultant, student, former HE colleague) that review and discuss these posts before publication — find the time… That’s all from me right now, cheerio !

[Edit 07/05: you might be interested in reading this, which just appeared today. Interesting views from Dr Banerjee, Pr Woolhouse, Pr Spiegelhalter and Stanford University. Nice not to be alone anymore… — Please note: the last chart in the BBC article mixes normal death rate (blue and orange lines), which is a probability to die, and Covid-19 mortality rate (red dots), which is a probability to die if you catch it, so for both to be comparable you need to multiply the Covid-19 numbers by the infection rate, which is going to make the Covid-19 percentages 5 to 10 times smaller, that’s how you can compare visually to the normal death rate, it’s much lower of course — also mind the log scale, differences between young and old are far greater than they look on the chart.]

[Edit 10/05: at last…. someone in the media who says things as they really are, thank you Professor Spiegelhalter ! Watch the interview or read the transcript.]

[Edit 13/05: interesting results in this study: not an easy read, but more complete and confirms the risk analysis of this post]

[Edit 14/05: the wind is turning, even the Telegraph is joining the realism bandwagon… better late than never]