The statistical detective work required to lift the lockdowns
Anyone prone to cynicism about “damned lies and statistics” should be prompted to think again by the pandemic. Admittedly, distorted or fictional statistics have been press-ganged into their familiar roles of spin and propaganda.
But the real thing — statistical information, carefully gathered — can save lives.
The UK Office for National Statistics has announced a new survey of 25,000 people, designed to test a demographically representative cross-section of the UK population for infection and antibody response. Given that the UK already tests many tens of thousands of people a day for infection, that news might provoke a shrug. But it is an example of the data detective work that we desperately need if we are to find our way through the crisis.
Consider the question we’re all desperate to have answered: when is the right time to lift the lockdowns? Without delay? In a week or two? Months from now? The answer depends on how much weight we put on livelihoods versus lives, and on how quickly we can prepare ourselves to carry out mass testing and contact tracing.
But it also depends on just how deadly the virus really is, something we do not yet know. Clearly, Covid-19 is dangerous. It has already killed more people than the Indian Ocean tsunami of 2004, and nobody is asking whether tsunamis are dangerous. But the degree of risk remains unknown.
This is the most important question in the world right now: what is the “infection fatality rate”, the number of infections — measured or unmeasured — which end in death? It is a hard thing to judge.
We do not know how many people have died of Covid-19: official figures overcount in some ways, by attributing deaths to this virus that would have happened at much the same time without it; they undercount in others, when people are killed by the virus at home or in a care home without a formal diagnosis.
In the UK, for example, Financial Times research suggests that there have been more than 50,000 excess deaths in this epidemic, and many are plausibly attributable to Covid-19.
More problematically, we are still guessing at how many infections have occurred below the radar. Everyone agrees that the official global case-count of more than 3.5m misses many mild or even asymptomatic cases. But much hinges on just how many of those undetected cases there are.
Let’s explore two possible scenarios, then. If the true infection fatality rate in the UK is 1 per cent, an estimate consistent with much of the alarming early modelling, then that suggests up to 5m people have been infected, 60m people have not, and an incautious relaxation of lockdown could cause a second wave of deaths even worse than the first.
In contrast, if the true infection fatality rate is around 0.1 per cent, as asserted by the veteran Swedish epidemiologist Johan Giesecke, then that implies that the clinical cases are the tip of the iceberg. It would also mean that up to 50m people have been infected in the UK, enough to confer herd immunity on the entire country, and the lockdown should be lifted now.
Knowing the truth would be of enormous value — which is why systematic serological surveys are now so vital. Serological tests look for the antibodies that suggest a person has already been infected. These antibody tests should give more clarity but the early results remain a statistical patchwork for now.
A serological study conducted in Santa Clara, California, suggests an infection fatality rate as low as 0.12 to 0.2 per cent. That would be very good news, if true.
But serological data from New York City, in contrast, suggests an infection fatality rate above 0.5 per cent and perhaps even close to 1 per cent. If so, the frightening forecasts of hundreds of thousands of deaths if the virus was not suppressed in the UK were not far wrong. A study from Germany points somewhere in the middle.
Which is correct?
None of this work has been peer reviewed and, as much as I would love to believe the Santa Clara results, they seem fragile. One problem is that participants were recruited on Facebook. The study might be packed with people who signed up because they were convinced they’d been infected; that would overstate the prevalence of the virus and understate the true death rate.
The New York data, regrettably, look more plausible. But they are hardly conclusive. New York City’s death rate may be unusually high due to nursing homes being centres for viral outbreaks — a fate other places might avoid.
Tempting as it is to adjudicate, we need more and better surveys, from all over the world — such as the one now in progress from the ONS. The results will help us make informed decisions as we lift the lockdowns.
Because the threat we face is both serious and novel, there is no hope of producing a well-calibrated response without this kind of information. We need solid statistics to blow away the fog of this epidemiological war.
Written for and first published in the Financial Times on 8 May 2020.
In a few days my NEW book The Next Fifty Things That Made the Modern Economy is out in the UK. It is available to pre-order – pre-orders help other people find the book and are a huge help.