Why the daily coronavirus death figures do not capture what you think

The daily death toll of the coronavirus epidemic has become part of our new everyday lives. Released every afternoon, it portrays the devastating effect of the disease as another day’s cull is added to all the deaths that have preceded it. Ministers routinely cite these grim figures in their opening remarks at the daily press conference held at No 10. On Friday Health Secretary Matt Hancock announced 980 deaths, bringing the cumulative total in the UK to 8,958. That daily figure of 980 was the highest yet in the epidemic. On Saturday 917 deaths were announced, taking the overall total to 9,875.

But the headline numbers—the daily tally and the cumulative total—actually underestimate the scale of the death toll. This is not because of any attempt to cover up what is happening. Rather it arises because of the hospital-based coverage of the daily figures and delays in reporting at a time when deaths are surging.

The daily figures measure deaths in hospitals among those who have tested positive for the virus. It thus excludes those who have succumbed to Covid-19 outside hospitals. Because older people are so vulnerable to the disease, care homes are particularly at risk. There is growing evidence that many have been affected.

Moreover, the daily death toll in hospitals is not what it seems. For one thing, it is based on reports for deaths over the 24 hours to 5pm the previous day. The information is thus always a day behind.

More important, and something that is widely misunderstood, the deaths that are reported are not those that have actually occurred in the 24 hours in question. The NHS now provides a breakdown of the deaths announced each day in England by when they took place. Take the 765 deaths reported on Thursday 9^th April, measuring the tally on Wednesday (specifically between 5pm on Tuesday and 5pm on Wednesday). Of these only 140 actually occurred that day (the 8^th); 284 occurred on Tuesday the 7^th; 100 on the 6^th; 57 on the 5^th. The remainder included entries in double digits stretching back to 31^st March and in single digits back to 16^thMarch.

“These reporting delays are substantial and they matter,” says Sheila Bird, a leading statistician and a former programme leader at Cambridge University’s MRC Biostatistics Unit. In particular, they mean that the headline total has been undercounting the scale of the death toll from Covid-19.

In a report published on 7^th April, the Office for National Statistics (ONS) compared the cumulative death toll in England by Friday 27^th March (announced on the 28^th) with the NHS figures that allocate the deaths to when they actually occurred. The headline number was 926 up to that Friday. By contrast, the date-of-death series recorded 1,649 by 27^th March, almost 80 per cent higher. That latter figure came from NHS data out on 5^thApril. These numbers continue to be revised. Using a more recent NHS release, of 9^th April, the date-of-death figures show even more deaths—1,728—had occurred by 27^th March.

That does not mean that the current, much higher, headline totals are undercounting the actual deaths that have occurred to anything like the same extent. Looking at the position a week later, on Friday 3^rd April, the headline cumulative death toll in England by then was 3,939, as announced the following day. The NHS date-of-death series (published on 9^th April) recorded 5,037 by this point—28 per cent higher.

Reporting delays are not confined to the NHS. All deaths from Covid-19 should be eventually recorded through death certificates mentioning the disease. These registration-based figures for England and Wales from the ONS pick up deaths outside hospitals as well as those in them. But they are published with a delay of 11 days, such as the report on 7^th April covering deaths in the week ending 27^th March; and there are in any case delays between when a death occurs and when it is registered, typically of five days but quite often longer. This makes it very hard to establish in a timely fashion just how many deaths from coronavirus have been occurring in care homes and other non-hospital settings.

Speaking at the press conference on 9^th April, the government’s chief medical adviser Chris Whitty said that both the wider registration-based figures and the hospital-based numbers had their uses. The merit of the NHS figures was the speed with which they became available, making them particularly valuable for decision-makers and allowing them to “see in relatively near time the trends over time.” He also said that they were “very comparable to international figures which tend to be collected in the same way.”

Even so, the undercount in the headline series is troublesome. It underplays the gravity of the death toll. And, as Bird points out, the reporting delay will make it harder to evaluate and to communicate when the peak occurs. No doubt there is a case for keeping things simple. But the government should be clearer about what it is actually announcing with the headline totals.

 

The second reference to the ONS report of 7th April originally said 11th April