Artillery Row

Does peak infection sync with lockdown enforcement?

The lockdown logic’s basic arithmetic doesn’t add up

Our brains are not wired for the exponential growth at the start of an epidemic, nor for the fact that epidemics do simply level off and then decline as a natural law of the epidemic curve, well before everybody is infected, even though history shows that it just is so. But now that we are all armchair epidemiologists, sniffy about any chart without a log-y scale (that we had to google a few weeks ago), let’s do a little detective-work ourselves using nothing more than primary-level maths. And to start, let’s latch onto a couple of numbers from the current covid-19 outbreak that do seem relatively certain.

First, we are told that the average time from becoming infected to showing symptoms is about 5 days: the “incubation period.” Secondly, we are told that the average time from symptoms to death is 18 days.

Armed with just those two numbers, 5 and 18, and no maths skills more complex than addition and subtraction, let’s look at the experience of China, Italy and the US to check that we really can see a “hockey-stick” in the infection curves 5 days (incubation period) after lockdown measures were introduced.  Just, you know, to verify that the lockdown theory actually works, given that a quarter of the planet is now under lockdown with the world economy in a coma and an unprecedented suspension of civil liberties.  It seems worth some back-of-the-envelope double-checking.

Let’s take China. Here’s their official CDC data:

 

Despite the weird blip on 1 February, this classic bell-shaped epidemic curve (the blue lines) shows a clear peak in symptom-onset around 25 January, implying an infection-peak (subtracting 5) around 20 January.  But hang on: lockdown proper in Hubei (the lion’s share of cases) wasn’t until 23 January, so this implies that the curve was already turning down before they declared the lockdown.

Okay, let’s say that you’re a bit sniffy about Chinese data; let’s look at the Italian experience:

Again, a classic epidemic curve, with a peak of symptom-onset 10–13 March, implying (subtracting 5 days) an infection-peak 5–8 March: too early for the turn in the graph to have been caused by the lockdown announced on 9 March.

Finally, in the US, CDC data suggests that cases peaked around 26-30 March, implying an infection-peak (subtracting 5 days) around 21-25 March – before most stay-at-home orders had been issued:

There are obvious counter-examples to lockdown. Sweden, famously, is taking a far more relaxed approach, with sensible hygiene and light-touch social-distancing precautions but without putting the economy or civil society into comas.  Looking at their death-per-capita relative to nearby locked-down countries, as well as at the current declining trajectory of admissions to Sweden’s intensive-care units (a macabre if reliable forward-indicator of deaths), there is no obvious harm from not having locked down:

Daily covid-19 ICU admissions in Sweden

Without disputing well-established methods of slowing the rate of transmission, it just doesn’t seem that lockdowns have been what has caused the reversal of the tide in any of the countries above. And given the extraordinary harm they cause – social, economic, mental and physical health – we should set the bar very high as a burden of proof. The impact should be unignorable, impossible to miss: not invisible.

But what about the need to “flatten the curve”? If you’ll forgive me for getting a bit mathmo: so long as we do not go over NHS capacity then the “area under the curve” (representing total cases and deaths), which is what really matters, will not be changed by flattening the curve, i.e. prolonging the epidemic. It will simply mean further economic and social harm, and the potential for a stretched epidemic to overlap (nightmarishly) with a bad flu season. And as forces in government are now modelling, we will be creating a second, additive, curve of excess mortality caused by the lockdown (largely delayed procedures and presentations at hospital – widely reported on the ground – plus mental health).

A final point before conclusions: don’t hang the policy-makers or the modellers out to dry.

None of the data above was available to the Chinese, Italian, American – or  British – governments when they declared lockdowns. In the fog of war, with no reliable data, they were jumping at shadows – rightly, they should be cautious.  And the performance of the most sophisticated models (Imperial College’s being the most high-profile) is very largely dependent on the accuracy of the disease’s parameters (especially infectivity, prevalence and mortality), and these have been totally unclear from the start.

So: what could be the policy conclusions of all the above? Well, if all the below are true:

  • looking at the experience of other countries, it is not clear that a lockdown caused the epidemic to reverse, and if Sweden suggests that sensible hygiene and modest distancing measures are likely just as effective as draconian lockdown; and
  • covid-19 deaths per day are clearly levelling off in the UK; and
  • there is clear evidence of a matching slowdown in reliable forward indicators (covid-19 related calls to 111 / 999 are declining sharply, particularly in the most vulnerable age-groups, and the number of covid-19 inpatients is actually declining) implying that a peak in infections is distant in the rear-view mirror; and
  • there is mounting evidence (and, apparently, government analysis) of the death-toll that a continued lockdown could cause; and
  • if it is now clear, as it seems to be, that there is ample NHS capacity both in critical-care and, particularly with the Nightingale hospitals, overall hospital capacity …

… then there is no case to extend the lockdown a day longer.

Enjoying The Critic online? It's even better in print

Try three issues of Britain’s newest magazine for £5

Subscribe
Critic magazine cover