Measuring excess mortality: a second look at Germany

Excess mortality is calculated as the number of deaths above the norm. To this end, national statistical offices like to compare the count of actual deaths from all causes with an average over past years. Measuring excess deaths in this way is the generally accepted approach, as demonstrated by recent analyses in The Economist (Tracking Covid-19 excess deaths across countries), the New York Times (74,000 missing deaths: Tracking the true toll of the coronavirus outbreak) and the Financial Times (Global coronavirus death toll could be 60% higher than reported).

But past averages can hide considerable variation. The approach taken here is to contrast the coronavirus outbreak in 2020 directly with the same period in each of the two previous years, i.e., to compare deaths from all causes in February to April 2020 with the same period in 2019, and then to do likewise one year earlier. The graphs below show the one year excess mortality in the five largest European countries, where the daily count in 2019 provides the point of reference (in grey), the excess mortality in 2020 is particularly evident (in blue), and the excess of 2018 over 2019 is also indicated (in green).

Excess mortality during February, March and April in 2020 and 2018, with reference to 2019

In addition to Germany’s minimal excess mortality in 2020, the other notable feature is the high incidence in 2018 when influenza was particularly severe in that country. The overall effect in 2018 and 2020 may be quantified as follows, using 2019 as the baseline:

Excess mortality, all causes (the sum of excesses on days when the count is greater than in 2019). Note that the one-year excess mortality period in 2020 continued until 5 May in the UK, increasing the total to 54,304.

So, how did excess mortality in 2018 affect that in 2020? One likelihood, in Germany in particular, was a reduction in the number of vulnerable elderly people still alive in 2020, given the age distribution underlying the high mortality in 2018. In recent weeks, especially in the light of the catastrophic outcome in the UK, much has been written in the press about Germany’s better comparative record in the face of Covid-19 — also usefully summarised in an informative new study by Janine Aron & John Muellbauer of Oxford University. The discussion to date has pointed to (i) the early increase in testing capacity that enabled effective tracing and isolation in Germany, (ii) the more efficient health service with its greater numbers of hospital beds, ICUs and ventilators, (iii) the concerted efforts to protect the most vulnerable by keeping the virus out of care-homes, and also (iv) the mitigating possibility that the initial carriers were probably fit young skiers returning from Alpine resorts who mainly re-infected other younger people. To that could be added, it seems, the lagged effect of the flu outbreak in 2018, not only the direct impact of excess mortality, but also the indirect effect of increased preparedness to cope with an epidemic both in the health service and amongst the population, as recommended at the time by the Robert Koch-Institut.

Further reading

1. Gabriele Ciminelli & Silvia Garcia-Mandicó (COVID-19 in Italy: An analysis of death registry data) examine excess deaths for seven regions of Italy.
2. Michel Coleman & others (Reliable, real-world data on excess mortality are required to assess the impact of Covid-19) provide a critique of the flaws in Covid-19 mortality reporting, in BMJ Opinion; their preliminary analysis of excess deaths during the Covid outbreak in Italy has been featured in The Guardian (UK wrong to rule out global coronavirus comparisons, experts say).
3. Holly Krelle & others at The Health Foundation (Understanding excess mortality: what is the fairest way to compare COVID-19 deaths internationally?) estimate excess mortality across European countries to indicate the comprehensiveness of each country’s counting of deaths attributed officially to Covid-19.
4. Jorge Felix-Cardoso & others (Excess mortality during COVID-19 in five European countries and a critique of mortality data analysis) measure excess mortality based on (i) the deviation from the expected value and (ii) the remainder after seasonal time series decomposition.

Note. Mortality time series have been obtained as daily counts from DESTATIS (Germany), INSEE (France), INE (Spain) and ISTAT (Italy), and, as weekly counts from ONS (for England & Wales only) and the devolved statistical bureaux NRS (Scotland) and the NISRA (Northern Ireland), as follows:- 
DESTATIS (Germany): Daily counts 1Jan2018-26Apr2020
INSEE (France): Daily counts 1Mar2020-4May2020 and 1Mar-31May 2018 & 2019, weekly counts Jan-Feb in 2018, 2019 & 2020
INE (Spain): Daily counts 1Jan2018-10May2020 (through Instituto di Salud Carlos III)
ISTAT (Italy): Daily counts 2018 & 2019 for all 7904 local authorities in Italy, and 1Jan2020-31Mar2020 for 6866 LAs and 1Apr2020-15Apr2020 for 4433 LAs (the estimates used here are scaled up proportionately from the two samples; linear extrapolation after 15Apr2020 is highlighted on the graph)
ONS, NRS, NISRA (UK): weekly counts 1Jan2018-11May20 (linear extrapolation after 11May2020 is highlighted on the graph)

See also: Coronavirus in Europe: Doubling Time, Continuous Growth and Other Measures. Personal note: I am not an epidemiologist, but there is cross-disciplinary connection in that my usual research is concerned with the econometric estimation of expectations constructed from noisy reported data, taking account of how much information such metrics may contain and how they may be compared internationally. Contact: s.j.mcleay@sussex.ac.uk & s.j.mcleay@sydney.edu.au.