skip to primary navigation skip to content

COVID: Analysis of excess deaths (updated 15th May)

Winton Centre for Risk and Evidence Communication

COVID: Analysis of excess deaths (updated 15th May)

From the latest ONS data we can obtain the total deaths for each sex and age-group, and the number that are labelled as COVID deaths, and compare with five-year averages in the figure below.

total-age-sex.png

This reveals that

  • excess deaths are occurring in all age-groups over 45,
  • a substantial proportion of excess non-COVID deaths in the over 70's.

We can expand the graph for ages 45.

total-younger.png

Briefly -

  • There is no evidence of excess deaths in the under 15s.
  • There has been fewer deaths than expected in young men aged between 20 and 24: 58 deaths compared to the average of 89 over this period over the last 5 years, a 35% reduction from normal [see statistical note below]. If we exclude the 10 COVID deaths in this group, non-COVID deaths have been reduced by 45%. The usual 'lump' in mortality for young men, caused by accidents and other risky behaviours, appears to have been ironed out.
  • There is no evidence of excess non-COVID deaths in the under-45s, suggesting they have not had increased risk due to health service disruption.

Just as for COVID deaths, we can examine how the rate of non-COVID excess deaths varies with age and sex.

excess-non-covid-rates.png
  • There is a clear exponential increase with age, but little minimal effect of sex. This is in stark contrast to the doubling of COVID death rates in males.
  • If the non-COVID excess deaths were actually COVID-related deaths that were not diagnosed as such, we would expect a similar sex-ratio in non-COVID excess deaths.
  • The lack of a strong sex-effect suggests that the non-COVID excess deaths are not primarily due to under-diagnosis.

Statistical note

  • Assuming a Poisson distribution with mean 89, the chance of observing such an extreme change in either direction, by chance alone, is 0.0006.
  • I am fully aware that I did not pre-specify my hypothesis and therefore this P-value is not strictly valid. However I would claim that this is an obvious group to look at, and therefore it's not entirely wrong.

Data sources