What do current mortality figures tell us about the danger of Sars-CoV-2? Actually, how can excess mortality be calculated? Our podcast with Dr. Felix zur Nieden, expert for mortality figures at the Federal Statistical Office.
New methods and timely results in our new podcast from 23. March 2021.
Podcast from 30 October 2020:
Audio podcast only in German. Transscript in English.
Mortality figures and excess mortality during the corona pandemic
How deadly is the Sars-CoV-2 corona virus really? And how dangerous is the current pandemic when compared to other infectious diseases? These are questions which have been worrying people since the outbreak of the corona virus – in Germany and all over the world. Answers are provided by current data on deaths in a given country. But how can excess mortality be calculated? What is the information value of provisional data in this context? And how can we distinguish between the impact of the corona virus and that of a flu epidemic? These are topics we are going to discuss with Dr. Felix zur Nieden, an expert for demography and mortality figures at the Federal Statistical Office.
Mr. zur Nieden, when the coronavirus crisis broke out, it put the focus of public interest on your work. How did you feel about that?
We realised pretty soon that our figures would gain in significance. The first country in Europe where the pandemic broke out is Italy, and a relatively large number of COVID-19-related deaths was counted there. Very soon there was the question of what impact this would have on total death figures. Is there an effect of a similar magnitude - meaning that the excess of total death figures is similar to the number of COVID-19 related deaths we have counted? Are there lots more, or lots less? Even back in early March, people were showing an interest in such data. They requested more timely figures for Germany. At the time, we had published figures up to November 2019. We felt it was exciting and challenging to deal with such a great interest in the data. Of course, we wanted to provide the best possible information at the earliest possible time to the general public. And we had to deal with that great number of requests for background information and data in the best and most patient way possible. We had requests from the fields of politics and science, from the media, and also from interested citizens. This often led to new aspects of the topic we had to deal with.
You have mentioned it: there has been immense public, political and scientific interest in current data that quantify the effects of the pandemic on the total number of deaths. How did you and your colleagues react to that?
In death statistics, our experts have traditionally taken a meticulous bookkeeping approach, which means they have provided very precise data. This is not only about absolute death figures but also about really precise information on the date of birth of the deceased person, the date of death and all other variables covered by the statistics. We examined the system to see how we could provide the data as rapidly as possible. And how perhaps we could take an unconventional approach to answer the questions that are urgent now. We then considered to take the data directly when they come in, that is, before they are checked for plausibility. These data are largely all right, but corrections are needed in some cases. For example, when a variable is missing or when there is an implausible combination such as the data of death being earlier than the data of birth. Or when there are transposed digits in the place of residence code - all these aspects. Normally, these things are painstakingly corrected by the statistical offices of the Länder in cooperation with the registrar's offices - this is what we mean by checking for plausibility. If, however, we want to get a direct overview of the total number of deaths and answer the most urgent question, that is “What impact is corona having?” - then it is absolutely possible to use those raw data that have not been checked for plausibility yet. Since then, we have released every week our ad-hoc evaluation of mortality figures, including any late deliveries that arrived in the course of the week.
Let’s talk again about the results of the ad-hoc evaluation: what is the impact of the coronavirus pandemic on death figures?
We have seen effects as from the last week of March 2020 and for the entire month of April, when we had markedly higher death figures. Looking at April alone, the death figures were by 10 percent above the average of the previous few years. That difference from the average of the previous few years matches exactly the COVID-19 deaths reported by the RKI. At first glance, we might conclude that other explanations of that increase cannot play a major role. If we take a closer look at the data, two things stand out. First, the phenomenon of excess mortality was very much limited to specific regions. It occurred above all in Bayern and Baden-Württemberg. Other Länder were very little affected or not affected at all. If we take an age-specific perspective, we see that it is above all the over 80 year olds who were affected. There the figures were in part 20 percent higher than previous years’ average. It should be noted, however, that today there are more over 80 year olds in Germany than there were in the past few years. If we put deaths in relation to the population, we still see death rates that are 10 percent higher for that age group - despite the measures taken to contain the spread of infections as much as possible.
So you release provisional figures, and you do so earlier than usual to satisfy the public need for timeliness. What is the information value of those data, for example, regarding completeness?
We did release provisional data in the past, too, although those data had been checked for plausibility and they were published with a time lag of about two and a half months. Even then we had to expect late data deliveries, though to a very small extent. I think we really can get a lot of information from data that have not been checked for plausibility. The main restriction really is the lack of completeness and, above all, that it differs between regions. Now we are releasing data that have a time lag of four weeks from the date of death. These data can be almost complete in one Land, while in another there are still more than 15 percent of the data missing. This means that it is not possible to definitely assess the regional development of the most recent cases.
You take only the past four years to determine the base mortality as a reference value for measuring excess mortality. Why don’t you use a longer period?
Our goal is to assess the absolute death figures by comparing them with an average. But death figures depend not only on mortality and the current number of deaths but also on the size and age structure of the population. To put it in a simplified way: more older people means more deaths. In the medium to long term, the age structure will change so much that comparing current deaths with absolute figures is becoming meaningless. This will not happen for week-to-week comparisons, but if we look at a period of several years, we have considerable effects. Even in this four-year period, we have to consider age-structure effects. For example, the age group of the over 80 year olds rose by 15 percent from 2016 to 2019, that is, from 4.9 to 5.7 million. Of course we have to take this into account. However, we cannot just compare things with the previous year because the effects of flu epidemics and heat waves slightly differ from year to year. Taking a four-year period is a trade-off between the two aspects. On the one hand, we take account of the variance between years that occurs in the data even without corona. On the other hand, we avoid letting the age-structure effects I mentioned earlier get too strong.
By comparing the current data with the average across previous years, you have chosen a different method than, for example, that applied by the European mortality monitoring EuroMOMO - a project that has covered excess mortality in many European countries for over ten years. What is the reason for your decision?
Our main emphasis is on producing the death figures and making them available as rapidly as possible. We also try to give a first assessment of the figures that is easy to understand. We think that comparing them with the average across previous years is a very transparent and easily comprehensible approach. EuroMOMO is a research network that originally was set up to address the world of experts. It uses a generalised, linear Poisson regression model to estimate expected developments of death figures. This is certainly justified when it comes to details, but it is a very complex, scientific approach. Another aspect to remember is that EuroMOMO analyses many countries at the same time and therefore has to find some kind of standardisation. This involves other restrictions than when you look at just one country. When thinking about the methodology and the conclusions, we can say that this complex approach does not lead to entirely different conclusions than those obtained by just examining averages. We decided to use an easily comprehensible average calculation because we address the general public, rather than a world of experts.
You have mentioned it already. In 2017 and 2018, flu epidemics led to an increase in death rates, whereas this summer the high temperatures were the reason for rising figures in August. How can the different effects on death figures be distinguished? And, considering the multitude of factors, how is it possible to make any assessment of the consequences of the coronavirus pandemic?
That’s a very complex question. Let’s start with the example of the flu epidemic. The RKI publishes a seasonal influenza report. In that report, the flu-related excess mortality is estimated. The same is done with total death figures - as the difference between what would have been expected and what really happened during the flu season and the total number of deaths recorded. If we take a look at the very serious flu season of 2017/2018, which had a particularly strong impact in early 2018, that calculation shows an excess mortality of about 25,000 cases. But only 1,700 deaths of people with laboratory confirmed influenza were counted. The assumption behind the estimate is that not all influenza-related deaths were identified as such. In such cases, the death certificate stated a different main cause of death. The only possible explanation of the large number of additional deaths occurring at the same time is the flu epidemic. This is also suggested by additional knowledge from other sources. For example, medical practices had laboratory findings that confirmed an influenza particularly often in that period. This tells us, irrespective of the death figures, that the flu was spreading rapidly at the time. That’s basically the method of elimination. What might explain the large increase in deaths during that period if not the flu epidemic? It is quite similar with the coronavirus impact we saw in April. The flu epidemic was considered to be over at the time. It is entirely unusual at that time of the year to have death figures rising as strongly as observed in 2020 and not going down. Also, the difference from the average of the previous few years matches very well the COVID-19 deaths reported to the RKI. All that is not conclusive evidence, but it does strongly suggest a connection. And this is what we said in our press releases. Things are quite similar with heat waves. We know that very old people suffer most from high temperatures and, therefore, have higher probabilities of death. When we see particularly high death figures during the hottest weeks of the year, we assume that there is a connection. Again, we have that combination between knowledge from experience and the method of elimination. What else could explain the one-time increase we experienced in August 2020 if not the heat wave that occurred at exactly the same time?
The RKI publishes daily figures of people who died from, or with COVID-19 disease. What is the connection between these data and your results, and how do you cooperate?
The RKI first of all focuses on the reporting date so that they can get an overview of the death figures really quickly. For example, if we have 40 more deaths than yesterday, this does not mean that those people really died yesterday - it just means that the deaths were reported yesterday. They died some time before, maybe a week ago or several weeks ago, in some cases the death may have occurred even longer ago. For the total death figures, however, we always take the date of death. If we want to establish a meaningful relationship between these COVID-19 related deaths and the total number of deaths, we have to take the date of death also for the COVID-19 related deaths. Otherwise we wouldn’t be comparing like with like. At the beginning of the pandemic, these figures were not directly available. Now they can be downloaded from the RKI website in a breakdown by week of death. We were not the only ones to consider this perspective as meaningful. The RKI was very cooperative and provided the data very rapidly - just in the way we needed them for our comparisons. We often exchange views with our RKI colleagues. Of course, they also use our data and ask for background information and data. The cooperation is very, very good.
You have already mentioned the difference between the date of death and the reporting date. I would like to ask you another question in this context. Why do you need four weeks to release informative figures?
That is a trade-off we have to make between completeness and timeliness. Many people want to have yesterday’s death figures available at the push of a button. But the legal reporting system does not work like that. For example, people have three working days to report the death of a family member to the registrar's office. And the registrar's offices cannot transmit the case to the statistical office until it has finally been registered. In a few cases, this can take several weeks - for example, if the family members cannot provide the death certificate at once. The death certificate, however, is needed for final registration of the death. The registrar's offices are legally required to transmit death figures to the statistical offices of the Länder at least once a month. So, it may happen that there is one month between one delivery of death data to the next. To sum up, we can say that the way of data reporting prevents more rapid data processing. The consequence is that when we push that button today there is just one percent of yesterday’s deaths in the system. But incomplete data do not help anyone. After four weeks, we have 97 to 98 percent of the data. Then we can give first information about whether the number of deaths in that week was significantly higher than usual or if it will be at average level.
Mr zur Nieden, to conclude our interview, let’s take a look ahead. What new findings do you hope to obtain in the next few months, for example, as regards reliable data?
I think the figures obtained from the raw data are pretty reliable. That was confirmed when we replaced the remaining raw data of 2019 by final data. It had hardly any influence, for example, on the average from 2016 to 2019. This means that many effects will be confirmed by the final data. But when we have the final data, it will be possible to perform more in-depth analyses. For example, when we have the detailed population figures, it will be possible to establish a proper relationship between deaths and population figures. Then I will be interested to see what effect the corona virus and the development in 2020 has on further indicators that can then be calculated - for example, life expectancy. I find that interesting not only for Germany, but also for many other countries where the coronavirus effects on death figures were in part even more massive than in Germany - despite the fact that even stronger restrictions were introduced. What else I find important is that it is not possible to derive from the development of death figures what would have happened without the restrictions. We should always bear this in mind when we want to use death figures to calculate how dangerous the pandemic is. In any case, it won’t be possible to assess the full impact of the pandemic until it is over. This means that next year we will have to take a very close look at how the figures have developed before we can draw a final conclusion.