Ross Clark

The limits of Covid death statistics

The limits of Covid death statistics
Photo by Christopher Furlong/Getty Images
Text settings
Comments

As is often said, choose your statistics carefully and you can use them make just about any point you want to. But rarely does the Office for National Statistics put out two releases on the same day whose statistics point in totally opposite directions. If you listened to the BBC midday news, you may have heard that overall deaths in England and Wales, while they have fallen, are still running at 5.9 per cent above the average for the time of year. This was based on an ONS release entitled Deaths Registered Weekly in England and Wales.

However, poke your nose in another release put out an hour or so earlier, entitled Deaths Involving Covid-19, and you would have come to exactly the opposite conclusion: that overall deaths in England and Wales are running at 5.9 per cent below the average of the past five years (8,686 compared with 9,233). 

The two figures don’t quite cover the same time period: the first refers to the week ending 12 June while the second document contains figures only up to 31 May. Even so, it is extraordinary that two documents published by the same body on the same day should tell such different stories. How come? 

When I contacted the ONS, they told me there is a difference between the two sets of figures. The first document refers to registrations of death, while the second document refers to deaths themselves. Given that there is a lag between deaths happening and being registered – sometimes to the tune of several weeks – this confuses the picture somewhat. Some of those deaths registered in the week ending 12 June actually occurred weeks earlier, at the height of the epidemic. By contrast, not all deaths which occurred in the week ending 31 May have yet been registered – so the ONS’s count of deaths that occurred in that week could yet rise closer to the average of the past five years.

Even so, it is quite a turning point when the number of overall deaths falls below the five year average. One influence on the figures is almost certainly that Covid-19 has been killing large numbers of people who were close to death anyway – it brought their deaths forward, so we had a large spike in April that will now be followed by a long period of below-average deaths. As Professor Neil Ferguson of Imperial College told MPs back in March, his models for the predicted death toll from Covid-19 worked on the assumption that two-third of victims would have died this year of other conditions.

The causes of death listed by the ONS for the week ending 31 May bear this out. Conditions known to increase the risk of death in Covid-19 patients were significantly below their five year average: heart disease, stroke, lung cancer and chronic lower respiratory disease. Deaths from some other conditions – colon cancer and dementia were up. It is at least sadly comforting to see deaths from heart disease and stroke falling, given worries at the height of lockdown that many people were failing to seek urgent medical attention.

One of the perverse outcomes of Covid-19 is that it might briefly flatter the figures for circulatory diseases and lung cancer. People whose deaths would have been attributed to those conditions will instead have gone down as Covid-19 deaths. Beware the government, the NHS and Public Health England boasting in a year’s time how successful its efforts to reduce heart deaths and lung cancer deaths have been – how it vindicates sugar taxes and anti-smoking campaigns etc – when all that has happened is a statistical quirk.

The ONS release doesn’t list all causes of death, though it is highly likely that lockdown will itself have impacted on the mortality figures. We might expect suicides to have risen but violent and accidental deaths to have fallen as bars have been closed and there have been fewer people on the streets. One worry, however, is that fewer cancers may have been diagnosed during lockdown – which might lead to a rise deaths in years to come.