Following my previous post, a colleague (from another institution) asked me whether I thought British universities would be able to resume offline teaching next October.
My first reaction was to think: ‘yes, of course they will’. Then I wondered: is that so obvious ? And if it is, which data would support this view ? So let’s give it a go.
There are actually two practical questions to answer:
- Should British universities prepare to resume offline teaching next October ?
- If so, will it be business as usual ?
To begin with the beginning: will it be safe to welcome students back on campus next October ? There are several indications that make this prospect quite likely.
Such a prediction might sound somewhat out of sync for British readers as the island is now in its highest coronavirus burden period, with still high daily death tolls. To explain this, I need to go one step further than my previous post, which was deliberately lean on statistics (that couldn’t last forever, could it ?), and explore three concepts: burden, timescale and scale.
In epidemiology and health statistics, there are several ways to define burden. However, in the case of the Covid-19 pandemic, it is the number of deaths that is the least unreliable indicator of the evolution of the situation in each country.
As mentioned in my last post, the number of identified cases is the key indicator for any country to monitor the situation and lead the fight, but not so for making predictions of final outcomes. At any stage, the number of identified cases depends on how much testing is done and the capacity of the country’s health system to report and process these cases. Therefore, in terms of flow over time, it is a more complex and multi-dimensional measure than the evolution of the number of deaths.
‘Least unreliable’ does imply some imperfections, though, and there are quite a few. For example, some countries only report hospital coronavirus deaths. Two recent adjustments published by France and the UK suggest that the total number of coronavirus deaths could be about 30% higher (France’s public health organization now publishes two daily numbers : hospital deaths and total deaths). My educated guess is that this gap could increase in the next few weeks as more deaths are reported outside hospitals, particularly care homes (edit 10/04: this increase can now be seen in France’s latest figures, where total number of deaths is now 1.5 the number of in-hospital deaths, therefore the coefficient there is already 50% — edit 24/04: the coefficient seems to stabilise at approximately 60%, but could rise again with delayed death recording, so educated guess was correct, unfortunately). It will definitely be much bigger for countries where a large proportion of the population has no access to health care. Some countries also experience longer time lags to centralise and report deaths. However, in this post I am only looking at neighbouring European countries, so the assumption that this bias is relatively homogeneous across countries is not unreasonable. Also, what matters most to our question is not necessarily the absolute correctness of reported numbers but their trend, assuming that these biases are relatively consistent over time in each country.
There is also the question of how much of these coronavirus reported deaths exceed the normal mortality at a given time of year. That is the concept of excess mortality, which is used to measure all causes of death. I will discuss this question further in the post, but for our purpose the number of coronavirus deaths as reported daily by public health organizations is still the indicator to follow.
In all countries affected by coronavirus, the epidemic picks up more or less quickly, then reaches a peak and gradually comes down. In the middle of that curve is the High Burden period where 80% of coronavirus deaths happen. For any country, it is the period after the first 10% of deaths have been recorded and before the last 10% of deaths. When this most important period occurs and how long it may last will be crucial to answer our initial question, because once this period is over governments may start thinking of easing lockdown (which will also depend on the number of cases by then; that’s public health management, not forecasting).
I have modelled the evolution of the number of reported coronavirus deaths in European countries. Although the evolution in some countries is not entirely predictable yet, it is nevertheless clear, looking at countries where the epidemic is more advanced than the UK (China, Iran, Italy, Spain), that this High Burden period lasts between 3 and 5 weeks (edit 24/04: that interval is correct for most countries, but data published since writing this post shows that the later social distancing measures have been taken, the slower the comedown after the peak, so late-deciding countries like the UK could go through a 6-week High Burden period). It so happens that the UK is currently in its High burden period. Therefore, by the end of April (at the very latest), the number of daily deaths in the UK will have substantially decreased (edit 24/04: this is also turning out to be correct, although the comedown is not as pronounced as if the UK had followed the curve we see in countries that took decisions earlier). It does not mean that there will be no coronavirus deaths at all after that, but these will gradually dwindle to nought. This should also be the case for most European countries, which means that by October the European peak should be well behind us.
However, there are two conditions for this to happen according to prediction.
The first condition is that no large-scale second wave of infection (or worse: a later seasonal resurgence of the virus) will get in the way of a return to normality. That is difficult to predict at this stage and will depend on a combination of various factors: immunological, political (how governments handle the situation) and behavioural. If everyone starts hugging frantically on their first day out, that might be asking for trouble.
The second condition is that no scale effect will result in the epidemic getting suddenly out of hand.
Fighting epidemics means controlling hotspots. Scale effects can happen when a sufficiently big hotspot (a region or town, for example — Wuhan, Lombardia, New York, London, Birmingham, etc.) gets out of control for too long. Looking at countries which are near the end of the epidemic (China) or near the end of their High Burden period (Italy, Spain, Iran), it seems that so far measures have prevented this from happening.
This is, so to speak, the underlying mechanism of the epidemic, and that’s where the sheer number of daily deaths is not enough to appreciate the situation. A benchmark is necessary to give a sense of scale.
To understand this, let us compare the number of deaths due to the coronavirus so far to the number of deaths we usually get from flu in Europe in a given year. There are several methods to calculate flu-caused ‘excess mortality’, e.g. the number of additional deaths caused by flu compared to the ‘normal mortality’ over a period of time. In this post, I will refer to a 2019 study based on the 2017-18 season (https://www.sciencedirect.com/science/article/pii/S1198743X19300588), which was close to average by European standards. According to this study, the flu death rate per 100,000 inhabitants was 25.4. For coronavirus reported deaths at hospital, by the end of the epidemic the corresponding rate in Europe will be somewhere between 20 and 30 (according to my model — too early to be more precise — edit 10/04: Sweden is the last European country where the evolution of the current wave is unpredictable, due to the uncertainty around government policies and inconsistencies in daily reports), with some countries above their usual flu rate and some below.
Even that comparison is not strictly relevant because a proportion of reported coronavirus deaths is part of the ‘normal mortality’ that occurs at any point in time, whereas past flu numbers are ‘excess mortality’, e.g. the number of additional deaths caused by flu. In other words, the final excess mortality of Covid-19 in Europe will be lower than what we would think looking at the current burden on health systems (edit 24/04: at this stage, information released on excess mortality is not consistent statistically, let’s wait until reporting contradictions clear up).
This could be quite different outside Europe, for example in India or some African countries, which we cannot model because the number of deaths is not reported reliably at this stage.
The point here is not to try and predict whether coronavirus will eventually kill more or less people than flu, but to give the current situation a sense of scale by comparing both epidemics (one being regarded as exceptional, the other annual) from a quantitative standpoint.
I am aware that these numbers might surprise some readers, least of all because we are hearing how overwhelmed most European health systems already are, including the NHS. However, there is no contradiction between both views, here’s why. When there is a flu peak, health systems are already under pressure. In the case of Covid-19, what makes the pressure on hospitals much higher is that, due to pathological complications of the virus’s acute symptoms (and the resulting mortality risk), the number of coronavirus-infected patients ending up in A&E and ICU is higher than for flu. In European countries, the proportion of coronavirus deaths occurring in hospitals might be somewhere between 50% and 80% (edit 24/04: this range of percentages holds, according to latest stats), whereas for flu that percentage is much lower. Also, the coronavirus epidemic seems more concentrated over time in each country. Flu annual outbreaks usually occur over 5-6 months with varying intensity, whereas coronavirus burden lasts about 3 months in each country (but not at the same time in all countries). These two factors explain the intensity of the coronavirus burden on health systems.
Therefore, both past quantitative facts and current news reflect the reality of their respective context, but in one case (daily coronavirus news) we are looking at burden, particularly peak burden, whereas in the other (annual flu statistics), we are looking at scale over an epidemic cycle.
In the UK alone, flu has killed over 17,000 people every year in average over the past 5 years (Source: Public Health England). Here’s a little thought experiment: as I write, there are 6,159 coronavirus reported hospital deaths; according to my model, it is unlikely that the final toll will exceed 3 times that number (edits 09/04 and 24/04: extra hospital capacity + slower comedown after the peak in countries where lockdown has been decided later ⇒ higher proportion of Covid-19 deaths in hospitals; which could mean a 4.5-5 factor in the end for the UK); allow 50% extra for non hospital deaths (edit 24/04: this is still the current official figure, there are indications this could rise as a result of the situation in care homes, but this has not materialised yet in data), then remove 30% of the total for normal mortality (not precise estimates, just ballpark percentages — edit 24/04: the data that is piling up on the question of excess mortality is highly contradictory, to say the least, but appears more favourable than initially thought; some more recent estimates have been published, they are very interesting in terms of calculations but there are at least three potential flaws: see my full analysis here — nevertheless, using these more recent sources with a correction factor still places our conservative ballpark at about -30 to -40%); compare the result to 17,000 (edit 24/04: the estimated maximum goes from 6,159×3×1.5×0.7 to 6,159×5×1.5×0.6, higher but still on the same scale as a bad flu year). That’s as high as it (edit 10/04: e.g. the current wave, which does not exclude the possibility of additional deaths through resurgence or by-effect) can possibly get, most probably lower (edit 02/05: that possibility still holds, but difficult to prove at this stage).
You may then wonder whether official epidemiological forecasting models overestimated the number of coronavirus deaths, or whether the difference with reality is the effect of lockdown measures. That will be for scientists and public health specialists to answer when the time comes. I think it’s both: official models were scenario-based and probably overestimated the outcome (worst-case scenarios predicted 250,000 potential deaths in the UK, e.g. 15 times the average annual flu death toll – here again, the comparison will give readers some appreciation of scale), AND lockdown measures are having an effect (through a straightforward probabilistic mechanism reducing propagation), particularly in hotspots.
Second practical question: will it be business as usual, even if universities teach offline again next October ?
That’s less likely.
As already analysed by many university professionals from Admissions or Planning departments, the student supply chain is long and complex, which means we may have already lost many students with the current crisis, particularly international students.
The second reason is that, even supposing European countries are free from coronavirus in October, that may not be the case for other regions. It is possible that countries and continents like India or Africa, where local conditions make it more difficult to control both propagation and contamination, may have to remain in lockdown longer. More generally, the coronavirus may end up polarising the world in a dividing line between wealth and poverty (again).
There is also the unknown of how the virus will behave over the winter in the Southern hemisphere.
Thirdly, many implications of the current crisis are yet to unfold: to get students in, we need enough borders to re-open, which requires countries to accept each other’s testing standards, we need airplanes back in service, etc. Plus the financial and psychological aftermath. We need families to recover from financial hardship. We also need people to switch back to a mindset where they can start planning again.
Finally, there is the issue of trust and image. On that record, British universities may not be in a great place if you add up the effects of the virus crisis, this year’s strikes, Brexit and a notoriously underfunded health system (coronavirus inevitably spotlights how health systems are coping across the globe).
To make things worse, some British universities are now cancelling summer exams. Although made in the name of inclusion, this is the least inclusive ‘one size fits all’ decision they could possibly make. ‘No detriment’ is undeniably the right way forward, but ‘no detriment’ does not mean ‘nothing at all’. The students I’ve spoken to may not openly articulate their views, but they are not happy.
First year students are particularly affected. Some of them came a long way to join a British university. With the strikes and the virus, they have barely received more than one term of tuition. That’s not much for their money, and even less so for the debts they or their families are contracting through student finance schemes.
For all these reasons, it may not be enough for British universities to say ‘we’re back in business’ for business to come back.
Top universities with historically high QS World rankings will remain attractive. Those who are most at risk of losing out by leveraging entry tariffs and filling their places with lower-qualification home students are institutions from the 5th-20th band in British league tables, because once you drop out of the top 20, rankings become much more volatile. But as with every challenge comes an opportunity, universities that are able to rebuild a positive student experience after this crisis despite the financial loss will have an advantage. In this new context, widening participation and raising attainment will be hot topics.
One possible evolution is for universities to move some activities online, offer more online degrees, etc. This is a highly complex matter that every university, indeed every faculty or department, will decide according to their own strategy. But it is no mystery that much will happen there in the next few months and years.