Will the Great Lockdown’s epitaph be ‘The Greatest Mistake in History’?
The average seasonal flu has a fatality rate of 0.1%. On 5 March, based on the early data from Wuhan in China which had the first cluster of infections and deaths, the World Health Organisation (WHO) published a fatality rate of 3.4% for the novel coronavirus. Not surprisingly, this alarmed public health authorities everywhere but it also badly misled epidemiologists in their modelling of the likely infectiousness and lethality of the virus under different response scenarios.
By now we have better and more reliable data. A report from the US Centers for Disease Control and Prevention (CDC) on 20 May estimated the most likely case fatality rate (CFR or the percentage of known infected cases who die from the disease) of 0.4%. Its estimate of asymptomatic cases is 35%, giving an overall infection fatality rate (IFR) of 0.26%. The president of the Yamaneko Research Institute writes that the fatality rate in Tokyo is also 0.26%.
The Imperial College London (ICL) model of 16 March predicted up to 510,000 UK and 2.2mn US deaths in ‘an unmitigated epidemic’ (p. 7). The model’s assumption was that without intervention, 80% of the people would be infected and the IFR was 1%. On this basis 500,000 is the figure the model came up with in its ‘reasonable worst-case scenario’ (which is an oxymoron). In a subsequent interview Ferguson revised this to ‘the most likely scenario’. ‘But as information has been gathered in recent weeks, from particularly Italy but other countries, it has become increasingly clear that actually this is not the reasonable worst case – it is the most likely scenario’, he said. The ICL model has been proven just as spectacularly wrong in its best case scenario of a maximum of 20,000 deaths over two years with school and university closures, case isolation and social distancing, going down to 15,000 with home quarantine thrown into the mix as well (p. 14).
The seductive numerical precision of the ICL model provoked a herd-like panic across the world with its grim forecasts of tens of millions dead. Yet, it was undermined by data collected in the following weeks that progressively reduced its policy usefulness. ICL ignored best practice to release the codes behind their model late and reluctantly. Chris von Csefalvay, an epidemiological specialist in the virology of bat-borne illnesses, including coronviruses, examined the code and concluded its flaws fall ‘somewhere between negligence and… grave scientific misconduct’. Drawing on data from the early outbreaks in Wuhan and Italy, the model made critically flawed assumptions on infectiousness and lethality that led it to massively overstate the mortality risks. Professor John Ioannidis of Stanford University’s School of Medicine dismissed the coronavirus data as ‘utterly unreliable’ and dubbed the ICL model ‘speculation and science fiction’.
Michael Levitt from the same School won the 2013 Chemistry Nobel Prize for ‘the development of multiscale models for complex chemical systems’. Although not an epidemiologist, presumably he knows something about complex modelling. He argues the growth of coronavirus infections is never exponential ‘forever’. Although the growth rate is very rapid at first, it also decreases at an exponential rate. He therefore describes the coronavirus infection curve as ‘self-flattening’. The mortality plateaus around one month of natural deaths, not one year as Ferguson had projected.
Levitt’s argument mirrors that of Sunetra Gupta, Professor of Theoretical Epidemiology at the University of Oxford. Her team produced a competing, far more sceptical model to that of ICL back in March that has stood the test of time a lot better. In an interview with UnHerd TV on 21 May, she believes many people who have been exposed to the virus likely have other kinds of pre-existing immunities to related coronaviruses such as the common cold. According to her, the strikingly similar patterns of the epidemic across countries is better explained by this hidden immunity than by lockdowns or government interventions:
In almost every context we’ve seen the epidemic grow, turn around and die away — almost like clockwork. Different countries have had different lockdown policies, and yet what we’ve observed is almost a uniform pattern of behaviour which is highly consistent with [our] model. To me that suggests that … the build-up of immunity … [is] a more parsimonious explanation than one which requires in every country for lockdown … to have had the same effect.
Conversely, a prolonged lockdown increases ‘the vulnerability of the entire population to new pathogens’.
That said, I’d be curious to know what Professors Levitt and Gupta make of the India case, as discussed in my previous article on Monday, where cases and deaths have not ‘self-flattened’ but are still rising more than two months later.
Among other contenders, the US lockdown intervention strategy may have its origins in a high school science experiment by a 14-year old girl named Laura Glass whose father worked as a complex-systems analyst with the Sandia National Laboratories. When the strategy of quarantine-cum-forced separation made its way into the policy bureaucracy in 2006–07 at President George W. Bush’s request to look for ways to deal with the next pandemic, it was rejected with the recommendation that the measure should be eliminated from serious consideration. Instead, a pandemic should be allowed to spread, people falling sick should be treated, and a vaccine developed to prevent it from coming back.
There are two striking European examples to date of observational data that contradict the dominant epidemiological model on which the lockdowns have been based. Belarus has been ruled by a dictator since 1994. President Alexander Lukashenko has dismissed Covid-19 panic as a ‘psychosis‘, rejected lockdown and social distancing measures, and refused to close schools and cancel football matches. On 1 April, a WHO official warned it was entering a ‘concerning’ new phase and urged the imposition of new measures to control the infection but was rebuffed. By 22 May, Belarus had just 190 Covid-19 deaths.
The best-known example of a country bucking the model is Sweden. Without compulsory lockdowns and with much of activity as normal, 99.998% of Swedes under 60 have survived. Applied to Sweden, the ICL-like model projected that, without a lockdown instituted by 10 April, between 70,000-90,000 people would die by mid-May. The actual total on 22 May was 3,925 – significantly higher than its Nordic neighbours but far lower than most of Europe. The two points of comparison are likely to be among the most watched over the next year. Figure 1 is visually stunning in dramatising the discrepancy between two epidemiological models on either side and the empirical reality in the centre chart.
For all the flurry of domestic and international criticism directed at it, Sweden held its nerve and the results are there to see (Figure 2). Sweden is hardly alone in the vastly inflated projections of the epidemiological modellers. On 29 March New York-based Columbia University projected 136,000 hospital beds would be needed in the City; the peak demand stayed below 12,000. Levitt is on the mark with his caustic comment: ‘It seems that being a factor of 1000 too high is perfectly OK in epidemiology, but being a factor of 3 too low is too low’.
In two complementary articles for The Mail on Sunday on 3 May and then The Sunday Times two weeks later, Lord Sumption, retired judge of the UK Supreme Court, made two important observations:
- The original justification for the lockdown was to save the National Health System (NHS) by averting the ICU capacity from being overwhelmed by ‘flattening the curve’. The policy does not reduce the number of cases but spreads them more slowly so the health system can cope. In the event the nation’s hospital system never approached peak capacity. The infections peaked around 10 April, at which time only 60% of ICU beds were occupied;
- When the initial lockdown period ended, it was extended on an entirely new rationale: suppression or elimination of the virus. However, the rhetoric of second wave of infections seems to be an invention to justify a policy that politicians have become too invested in and are afraid of reversing. Lockdowns are continuing, that is, to protect ‘politicians’ backs. They are not wicked men, just timid ones… But it is a wicked thing that they are doing’.
It’s ironic that new pharmaceutical products must undergo rigorous testing for side-effects and collateral harm before being approved for public use, but lockdowns were mimicked by one country after another with little apparent consideration of the unintended and perverse health, economic, educational and other human consequences.
By now we also have some preliminary data from jurisdictions that have eased lockdown restrictions to varying degrees. Figure 3 is from a study by JP Morgan of the impact of countries and US states easing various lockdown measures. In both charts, what is immediately striking is that ‘R’ has mostly gone down, not up, suggesting that the virus might evolve according to its own internal logic.
‘The fact that re-opening did not change the course of the pandemic is consistent with studies showing that initiation of full lockdowns did not alter the course of the pandemic either’, the report concludes. The report’s author Marko Kolanovic said: ‘virtually everywhere infection rates have declined after re-opening even after allowing for an appropriate measurement lag … This means that the pandemic and Covid-19 likely have [their] own dynamics unrelated to often inconsistent lockdown measures that were being implemented’. While failing to alter the course of the pandemic, lockdowns have destroyed tens of millions of livelihoods.
It will be a year or two, perhaps more, before we know how many lives Covid-19 took in individual countries and worldwide, how many lives were saved by lockdowns, and how many were killed indirectly from the unintended, perverse and long-term consequences of the lockdowns.
Ramesh Thakur is emeritus professor at the Australian National University and a former United Nations Assistant Secretary-General. Of Indian origin, he is a citizen of Canada, New Zealand and Australia.
Comments
12 responses to “The rise and fall of coronavirus modelling”
I agree with many of the comments above. Let me add two more.
Ramesh Thakur claims that under the dictatorship of Lukashenko in lockdown- and social-distancing-denying Belarus , on May 22, there had been “just 190 Covid-19 deaths”. What scholar would take the figures produced by a dictatorial regime as accurate without even a question?
Ramesh Thakur once more quotes with approval and indeed relies upon the work of Professor John Ioannidis of Stanford who once argued that coronavirus data was “utterly unreliable”. I checked the link Thakur provided. In it Ioannidis concluded that “if we assume the case fatality rate among individuals infected by SARS-Cov-2 is 0.3% in the general population–a mid-range guess from my Diamond Princess analysis–and that 1% of the US population gets infected (about 3.3 million people), this would translate to about 10,000 deaths…” On the day Thakur’s article was published in Pearls and Irritations the Covid-19 death rate in the United States was 100,000.
Re. Ioannidis, the example you cite, pl. note that I referred to that already in my reply to Harold Zweir yesterday. The Financial Times too used his ‘utterly unreliable’ quote (‘The mystery of the true coronavirus death rate’, 30 March). If he is good enough an authority for the FT for critiquing others’ work, he’s good enough for me. That clearly doesn’t stop him from making mistakes in his own projections.
Re Belarus, I don’t know what sources you rely on, but I use the Johns Hopkins University coronavirus hub and Worldometers. I don’t presume to judge the reliability and accuracy of the data they are reporting. That’s an impossible task. Are China’s statistics reliable? Italy, Spain, the UK, India?
You might want to read the FT article I mentioned – it’s very educational. Two examples: ‘in Italy, Covid-19 is listed as the cause of death even if a patient was already ill and died from a combination of illnesses. “Only 12 per cent of death certificates have shown a direct causality from coronavirus,” said the scientific adviser to Italy’s minister of health’.
‘… the vast majority of those who have died from Covid-19 in Britain have been aged 70 or older or had serious pre-existing health conditions. What is not clear is how many of those deaths would have occurred anyway if the patients had not contracted Covid-19’. Perhaps ‘as many as half or two-thirds’.
In other words there is very little we can be certain of by way of reliable data. This makes it possible to nit-pick your way through anyone’s argument. I don’t get the sense scientists have got on top of this virus yet. Indeed last October the WHO said measures like contact tracing, quarantine of exposed individuals, entry and exit screening, border closure were: ‘Not recommended in any circumstances’ (https://apps.who.int/iris/bitstream/handle/10665/329438/9789241516839-eng.pdf). Wuhan, not science, caused a U-turn.
On the big picture, I stand by my contention there is virtually no evidence to show hard lockdowns have worked. Modelling has proven unreliable. Given how extreme that policy is, the burden of proof rests with lockdown proponents. If you wish to make such an argument, I’d read it with great interest – and relief that the great cost has been worth it.
One cannot say enough in 100 words. However.
Re Ionnadis, it is one thing for the Financial Times to quote him in late March and quite another for Ramesh Thakur to quote him in late May without mentioning that his calculation of likely United States deaths was 10,000 at a time when it had passed 100.000.
Every morning I look at the Johns Hopkins information that forms the basis of World0meter’s information. I trust that information from several countries is reliable/honest, although even those countries use different kinds of calculations. Others I just ignore, for different reasons. It is obvious that the information produced by the Belarus government is worthless. Lukashenko is a dictator who controls the news, including news about Covid-19, by tightest censorship and with threats of imprisonment etc. Ten minutes of googling is enough to discover that simple fact. That is why Ramesh Thakur’s straight-faced claim that Belarus has so and so many Covid-19 deaths astonished me. The Johns Hopkins people, of course, must rely on the figures the governments of the world supply. I trust the figures of, for example, the Nordic social-democracies, Sweden and Norway. I place no trust whatever in those produced by their near-neighbour, the dictator in Belarus.
I didn’t quote this either from Ioannidis (17 March): even some mild or common cold-type coronaviruses “have case fatality rates as high as 8% when they infect people in nursing homes”. He was ignored. Reality today: 42% of all US CV-19 deaths are in care facilities that house 0.62% of the population. 70% in Ohio, 69% Pennsylvania, 81% in Minnesota… Surprise, sometimes he’s right, sometimes wrong.
Meanwhile, we still don’t know what you believe now: the quality of 16 March ICL model? Lockdowns have been effective? Don’t be coy, professor. Say it, give your evidence, write an article for P&I. Let’s debate this vital topic.
Agree with you Ramesh, and with other critics such as Paul Frijters. We should avoid policies that make police look stupid, like when they fly drones along the beach to spy on couples too close together. Reminded me of school dances. I doubt if we are out of the woods. Reports from specialists bedside in New York are frightening. Patients’ immune systems are going haywire. Desperate surgeons are resorting to amputations. The virus appears to be beating the scientists but we can’t spend our lives locked indoors.
Models that involve a prolonged exponential increase in cases probably don’t consider the voluntary increase in social distancing (lockdown) that occurs without the need for rules, when people get frightened into keeping apart by the death toll, or more strictly, by the reports of it. That is probably an important factor dampening the increase, and one that is difficult to model. The lurid predictions and journalistic beat-ups have apparently saved many lives, though their lingering effects will cost jobs.
Be that as it may, people working in the overwhelmed hospitals of the world, or even in the ones that have had a lucky escape, are unlikely to have the time or the inclination to read more than the first few lines of either these two articles.
Sorry Ramesh, but I stopped reading your polemic after you made early mention of the mortality of influenza in modern times. As the developed world now uses a vaccine, this comparison is invalid. I recommend that you examine the mortality rates of the 1918-19 pandemic. My prediction is that history will look favourably on the leaders who opted for lockdown.
I mention the flu fatality rate in the very first sentence. You say you stopped reading at that point. I am mightily impressed then that that was enough for you to determine it’s a polemic. Bravo! Meanwhile, I note that in an article on 30 March, the Financial Times mentioned ‘a death rate of around 0.1 per cent for seasonal flu and 0.2 per cent for pneumonia in high-income countries’. I guess the FT is crossed off your reading list as a polemical rag. It stays on mine as one of the world’s most reputable and responsible newspapers.
You describe Sweden as “significantly higher than its Nordic neighbours.” In deaths per million the figures are: Sweden 409, Denmark 91, Finland 56 and Norway 43. It is an interesting interpretation of “significant” given the specificity of other data used in the article. In fact it puts Sweden in the territory of the worst European countries, (Italy, UK Spain) and 100 DPM more than the USA . Translated to Australia it would mean more than 7500 deaths, compared with the current 102.There is still a long way to go in these figures as you acknowledge, but it does seem a very selective use of the data.
Current deaths per million: Belgium 817, Spain 580, UK 557, Italy 543, France 426, Sweden 405
The first five are lockdown countries. Sweden is not. This proves the effectiveness of lockdowns? Forgive me for being dense today, but how exactly?
‘Very selective use of data’. According to you, the above statistics put ‘Sweden in the territory of the worst European countries, (Italy, UK Spain)’. But my description of Sweden’s mortality being ‘significantly higher than its Nordic neighbours’ is selective and a distortion? I’ll let readers judge for themselves.
‘Translated to Australia it [Sweden’s rate] would mean more than 7500 deaths, compared with the current 102’. Sorry, but this is about as inane as me saying Belgium’s rate translated to Australia would mean xyz… Why are you fixated on Europe as a point of comparison and not Asia-Pacific (Taiwan, Japan, South Korea, Hong Kong, Singapore, Vietnam)?
Since the purpose of the article is question the lockdown as a measure that prevented the spread of COVID-19 or reduced the number of deaths, it is hardly surprising that Ramesh Thakur quoted those sources that supported his thesis. But when I read an article that questions the health measures introduced at speed, I want something a little more objective and a little less belittling of the data on which it was based. I also want to read about where the experts quoted by the author also got things wrong. For instance, in early March Michael Levitt predicted that there would be no more than 10 deaths from COVID-19 in Israel, whereas their current death toll is 281. There are good reasons to examine the responses to the virus and to learn from what has happened – including lockdown measures – but if Mr Thakur’s article is to be the model for critical comment then we are back to politics as usual.
Where possible, I try to look at the data myself. On the experts quoted, Ioannidis I believe also got his prediction of likely US deaths, made around March, badly wrong. That does not invalidate their criticism of others’ modelling where their early caution has fared better than the lurid predictions of the modelling. Even so, I did say the self-flattening curve hypothesis, which I do find intriguing, is seemingly challenged by the data from India. Lockdown is such an extreme policy measure the proponents had better have a watertight case because their policy advice has real-world consequences for millions of lives and livelihoods.