By Richard Marshall

COVID-19 Actuaries Response Group – Learn. Share. Educate. Influence.

Choice of benchmark

Using weekly death registration data from 1 January 2015 to 31 December 2019, the number of daily deaths for the UK was calculated. Two benchmarks for weekly mortality were generated:

• Deaths in 2019 only
• Average number of deaths in the same weeks across the years 2015 to 2019.

The deaths in 2020 were compared to each benchmark (adjusting for incomplete weeks) on a weekly and a cumulative basis. The results are shown in Figures 1 and 2.

Figure 1: Weekly excess deaths in 2020

Figure 2: Cumulative excess deaths in 2020

The difference in the excess mortality between the two benchmarks (to 7 August) is approximately 9,100 deaths (excess is 65,250 relative to 2019 and 56,150 relative to 2015-19). The difference is not evenly distributed through the year; almost all of it (8,600 deaths) relates to the winter period (January to March).

Neither benchmark has been adjusted for population ageing or mortality improvements. These should broadly offset each other and their effect on the estimate of excess mortality would be immaterial.

Had mortality in 2019 been in line with the 2015-18 average (without change to the deaths in 2020), the excess deaths would have been 53,850.  That is 11,400 fewer than the comparison to 2019 (a reduction of 17%) and 2,300 fewer than the comparison to 2015-19 (a reduction of 4%).

Potential excess COVID-19 deaths amongst ‘influenza non-deaths’ from 2019

There is the possibility that some of the excess deaths (including COVID-19 deaths) amongst the frail elderly during the period since March 2020 are individuals who might have been expected to die during 2019, if winter mortality had not been relatively light at that time.

To investigate this, it is useful to re-group months so that they are organised into three periods, a central ‘Winter’ period which runs from December to March, along with periods either side of this: ‘Pre-winter’ from August to November and ‘Post-winter’ from April to July.

In this section, the year from 1 August to 31 July is referred to by the two calendar years which intersect it. So 2015/16 refers to the year from 1 August 2015 to 31 July 2016.

Using the same data as in the previous section, the ‘Excess Winter Deaths’ for 2015/16 to 2019/20, have been calculated as follows:

Excess Winter Deaths = Winter deaths – ½ (Pre-winter deaths + Post-winter deaths)

For 2019/20 only pre-winter deaths can be used (the margin of error is small as the difference between pre-winter deaths and post-winter deaths is typically small).

Excess winter deaths have been calculated net of approximately 5,500 deaths due to COVID-19 registered up until the end of March 2020.

Table 1: Excess Winter Deaths in the UK, 2015/16 to 2019/20

Relative to the average of the previous three winters (35,333 excess deaths), both the 2018/19 and 2019/20 excess winter deaths appear light.

It is worth noting that following the light winter in 2018/19, pre-winter mortality for 2019/20 was higher than average, reducing the number of survivors exposed to COVID-19 from March 2020. The average number of deaths in a pre- or post-winter period in 2015/16 to 2018/19, was 190,200. There were 195,200 pre-winter deaths in 2019/20, an excess of 5,000.

A crude (and probably prudent[1]) estimate of the number of additional lives exposed to risk of death in 2020 could be the sum of the shortfalls of excess winter deaths in 2018/19 and the first quarter of the winter 2019/20 season (relative to the prior 3-year average), less the excess pre-winter deaths for the 2019/20 season (the 5,000 deaths referred to in the previous paragraph). This comes to:

In this calculation, the winter deaths from January to March 2020 have been excluded, as they already contribute to the 2020 cumulative excess deaths.

So, there are perhaps 7,550 additional (potentially higher-risk) individuals at risk of death in 2020 as a result of surviving the prior winters. The light winter mortality in 2019/20 (prior to COVID-19) may have increased the population at risk of COVID-19 further still, but as noted above this would have no net effect on the excess deaths in 2020.

As noted, this is likely to be a high estimate and not all of these will necessarily have died during 2020.

Even if it were the case that the whole of this ‘excess survivor’ population had died during 2020, this would explain no more than 15% of the total excess deaths (based on the 5-year average benchmark).

We therefore conclude that the excess mortality in 2020 cannot be explained by excess deaths amongst those who might have been expected to die during the winter of 2018/19 or December 2019 (had those periods seen excess deaths in line with the prior three winters).

Comparison to Germany and Sweden

Table 2 shows the total number of deaths in Germany for January to July (inclusive) in each year from 2015 to 2020, and also for the full calendar years from 2015 to 2019[2].

Table 2: German death statistics

Based on these figures, the excess deaths versus the same two benchmarks (2019, and the 2015-2019 average) would be 1,000 and minus 450 respectively. The choice of benchmark makes relatively little difference. It is noteworthy that deaths for 2019 were higher than the average over 2015 to 2018 by approximately 8,700. Had all of these excess deaths been delayed from 2019 to 2020, then the excess mortality in 2020 to date (versus 2019) would have been approximately 18,400.

The SCB (Statistics Sweden) has published daily death registration data for 2015 to 2020[3]. Table 3 shows total deaths from January to July and over the full calendar year from 2015 to 2019.

Table 3: Swedish death statistics

Based on these figures, the excess deaths versus the same two benchmarks (2019, and the 2015-2019 average) would be 6,650 and 4,150 respectively.

Swedish full-year deaths in 2019 were approximately 3,450 fewer than the average of the previous four years; this is similar to the UK’s light mortality in 2019. Had these additional deaths occurred in 2019 and the equivalent number been removed from 2020, there would be roughly minus 250 excess deaths in Sweden for 2020 (up until 31 July).

Table 4 summarises the comparison between Germany, Sweden and the UK.

Table 4: Comparison of excess deaths in Germany, Sweden and the UK

So, even after adjustment in a consistent manner to the UK, excess deaths (as a proportion of the average annual deaths over 2015-18) in Germany and Sweden have been significantly lighter than in the UK.

Conclusion

We have shown that the excess deaths in 2020 can be measured relative to different benchmarks and that this can result in a difference of around 9,100 deaths.

Had 2019 experienced heavier mortality (in line with the recent average) then the excess deaths in 2020 might have been further reduced to 53,850.

In addition, ‘excess survivors’ from the previous low-mortality winters might have contributed some additional deaths in 2020. However, an upper bound on these might reasonably be set at 7,550 deaths. This leaves at least 46,250 excess deaths even against the 2015-18 average.

Whilst the calculations set out here have been approximate, they show it is improbable that more than 15% of the UK’s excess deaths can be explained by the low-mortality winter. Given the prudent assumptions required to get close to 15%, it seems likely that the true proportion is much lower.

Comparing the UK excess deaths with Germany and Sweden, adjusted in a similar manner, shows that the excess deaths in the UK are materially higher and the difference in the total excess mortality between the UK and these countries is, at most, only partly explained by differences in winter mortality.

25 November 2020 (based on information available to end October)

[1] This assessment is based on the particularly severe mortality seen in 2018 being included within a relatively short period of averaging.

[2] https://www.destatis.de/EN/Themes/Society-Environment/Population/Deaths-Life-Expectancy/_node.html

[3] https://www.scb.se/contentassets/edc2b33f85ad415d8e7909002253ed84/2020-10-05-preliminar_statistik_over_doda_inkl_eng.xlsx