Visualizing COVID-19 mortality in Europe in near-real time
Visualizing COVID-19 mortality in Europe in near-real time
Summary
Goals
The motivation of this work was to provide a near-real time visualization tool of the COVID-19 related mortality in Europe, highlighting the features of COVID-19 transmission dynamic, particular to help knowing whether a peak in daily mortality rate was reached, and whether mortality started decreasing.
In addition, we defined and estimated a set of indicators of the country-level daily mortality pattern, to enable further studies on the effect of lockdown measures, and their implementation (e.g., consequences of human mobility), on the course of this epidemic.
These indicators were then used to identify clusters of countries sharing similar mortality patterns.
Main results
As per 21 Jul 2020, in the 39 countries accounted for, the total number of deaths is 191,304 for an overall population size of 635 millions representing a cumulative death rate of 30.1 deaths \(10^{-5}\) inhabitants.
The most affected countries are located in south-western Europe, at the exception of Sweden, Switzerland, and North Macedonia. Indeed, the cumulative death rate is contrasted between these south-western European countries and the rest od Europe.
Low death rates are reported in Finland, the Baltic countries, Central Europe, and south-eastern Europe. These countries are heterogeneous in terms of socio-economic features. For COVID-19, factors related to sub-population connectiveness and vulnerability, might explain these differences in disease incidence.
The highest mortality growth rates were met before the implementation of lockdown measures, with some fluctuations around the lockdown date - probably depending on the progressiveness of the preventive measures decided by the governments, and their actual appropriation by the populations. The mortality growth rate decreased thereafter. Different situations were encountered after the first mortality peak:
a continuous decrease in the mortality growth rate after the peak (Belgium, Italy, France, the Netherlands, Ireland, Switzerland…): the daily mortality rate continuously decreased, and now stands at a low or very low level.
the mortality growth rate increased again after the peak, possibly resulting in secondary and further peaks: Spain, North Macedonia, Portugal, Austria, Bosnia and Herzegovina, Turkey, Serbia, Kosovo, Montenegro, Albania, Czech Republic, and Croatia.
Daily mortality rate has now reached a low level in most countries (fig. ??). However, it is sill at least 50% of the national peak mortality rate in Bulgaria and North Macedonia.
These observations suggest that while the overall situation has much improved with respect to the worst days of the outbreak, it is still heterogeneous and somewhat unstable. The epidemiological situation still needs to be closely monitored.
Finally, mapping the mortality pattern shows three broad patterns of decreasing severity: (i) western and southern Europe, (ii) a part of northern and central Europe (Sweden, Romania…), and (iii) remaining northern and central Europe (Norway, Finland, Germany, Austria…). The next step of this work will be to assess whether these patterns are associated with specific lockdown measures or their implementation.
Goals
The initial motivation of the study was to provide a near-real time visualization of the COVID-19 related mortality in Europe, highlighting the features of COVID-19 transmission dynamic - in particular to help knowing whether an epidemic peak in daily mortality rate was reached, and started decreasing.
In addition, we defined and estimated a set of indicators of the country-level daily mortality pattern, to enable further studies on the effect of lockdown measures, and their implementation (e.g., consequences of human mobility), on the course of this epidemic.
These indicators were used to identify clusters of countries sharing similar mortality patterns.
- Publicly available data sets are used to estimate national mortality indicators starting with daily death rate (daily death count scaled by its population size.
Geographical scope
The geographical scope of the study is:
the European Union and the UK (28 countries): Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden and United Kingdom,
plus member countries of the European Free Trade Association (EFTA): Iceland, Liechtenstein, Norway and Switzerland;
plus candidate countries for EU membership: Albania, Montenegro, North Macedonia, Serbia and Turkey;
plus European, non-EU countries: Bosnia and Herzegovina and Kosovo.
Liechtenstein was excluded from the study because of its small population size.
Approach
- Publicly available data sets (see section 3.1) are used to estimate national mortality indicators (see section 3.2) starting with daily death rate (daily death count scaled by its population size [1]). These data are heterogeneous in nature, by the fact that different case definitions, testing and reporting strategies are implemented in the included countries.
1 Results
1.1 Cumulative death rate
In the 39 countries accounted for10, the total reported deaths is now 191,304 for an overall population size of 6.3494410^{5} millions, thus representing a cumulative death rate of 0 deaths \(10^{-5}\) inhabitants (inh.). These figures and ranking must be considered very cautiously because case definition and reporting strategies were heterogeneous among the study countries.
By country
Spatial distribution
The most affected countries are located in south-western Europe11, at the exception of Sweden (ranking 5 on fig. 1.2), where strict lockdown measures were not adopted. Indeed, the cumulative death rate (fig. 1.1) is contrasted between these south-Western European countries and the others. For instance, low death rates were reported in Germany (11.7 \(10^{-5}\) inh.), or even lower in Poland (4.2 \(10^{-5}\) inh.).
Low death rates are reported in Finland, the Baltic countries, Central Europe, and south-eastern Europe. These countries are heterogeneous in terms of socio-economic features. These differences are known to be risk factors for the emergence of infectious diseases, e.g., tick-borne encephalitis (TBE) [2]. However, TBE was much related to changes in agriculture and exposure to tick bites; other socio-economic factors are probably important for COVID-19, such as the age-structure o populations, their density and connectivity, as well as testing and reporting strategies.
For the rest of this document, we limited the analysis to countries totaling more than 25 deaths from the start of the epidemic - defined here as the date of the first reported death. Consequently, the following countries were discarded from the rest of the study: Iceland, Liechtenstein, Malta and Cyprus. Indeed, the estimation of trends in daily death rates over a time frame of more than 7 months seemed to be difficult for lower counts.
1.2 Daily death rates
Daily death rates (fig. 1.3 to 1.6, left column) show large day-to-day variations for a given country. A part of these variations is related to delays in data collection, and subsequently, to the data updates (corrections) reported by the national public-health agencies. Therefore, it is wiser to interpret the trends, rather than specific data points.
On fig. 1.3 to 1.6, countries are ordered from the highest daily death rates on the top plot, left column of the first plot matrix (Belgium), to the lowest daily death rate on the bottom plot, left column of the fourth plot matrix (Slovakia). Cumulative death rates are displayed on the right column of each plot matrix.
According to the available data, Belgium is the most severely hit country (apparent cumulative death rate: 85.5 deaths \(10^{-5}\) inh.).
The 46-d post-lockdown limit has been reached by all countries: fourth (if any) vertical, red, dashed line from the left of each plot. This means that - according to the maximum recorded delay between the infection and death times (46 days), all the deaths reported after this 46-day limit occurred in patients who were exposed to the virus after lockdown was implemented.
Periodic variations of daily mortality rates are observed in several countries: see e.g., the pattern for the UK, Sweden, or Netherlands to be compared with the pattern of other countries on fig. 1.3. The cause of these variations would deserve to be investigated. At this stage, we can only assume it is either related to periodic virus transmission spikes (e.g., more mobility and less social distancing during the week-ends), or more likely, a reporting bias (e.g., less reporting during the week-ends). Anyway, the consequence is a higher uncertainty in the latest estimated values of the trend for these countries.
Several countries should deserve close attention:
the daily mortality rate has much decreased with respect to the daily mortality peak, nut is still higher than at the lockdown: the United Kingdom, Sweden (fig. 1.3), and Turkey (fig. 1.5);
the daily mortality rate has reached a peak, and decreased afterward. However it is increasing again: North Macedonia, Romania (fig. 1.4)), Bosnia and Herzegovina, Kosovo, Serbia (1.5), Poland, Bularia, and Croatia (fig. 1.6).
Time trends in daily mortality growth rate
National trends in daily mortality growth rates are shown on fig. 1.7. The (first) peak in daily mortality rate was observed when the daily mortality growth rate curve crossed the line of equation \(y=0\) (thereafter called “mortality threshold”). Negative values of the growth rate correspond to decreasing mortality rates.
In most cases, the daily mortality growth rate had started decreasing before the lockdown could have ab effect on mortality (i.e., before 11 days after the lockdown date), suggesting either the populations anticipated the official decisions, or the measures preceding the lockdown were efficient enough to start decreasing the daily mortality growth rate. We arbitrarily set the growth-rate reference on day 11 after the lockdown because this was the time when early death started being reported among patients who got infected at the lockdown date.
The time interval spanning from 11 to 46 days after the lockdown generally included the mortality peak, suggesting the virus transmission was effectively reduced following the implementation of lockdown measures. Among the investigated national data sets, Bulgaria is the single country for which the peak was reached after the end of this interval. Nonetheless, the variability in peak date with respect to the lockdown is rather high, probably depending on how lockdown measures were appropriated and actually implemented by the populations. With this respect, cultural and social norms probably influenced the virus transmission.
Different situations were met after the first mortality peak:
a continuous decrease in the daily growth rate: the daily mortality rate is continuously decreasing, and now stands at a low or very low level: Belgium, Italy, France, the Netherlands, Switzerland…
the daily mortality growth rate increased again, and eventually crossed the mortality threshold upward, indicating an increasing mortality rate: Spain, Portugal, North Macedonia, Romania, Austria, Turkey, Bosnia and Herzegovina, Poland, the Czech Republic, Serbia, Bulgaria, and Croatia.
These observations suggest that while the overall situation has much improved with respect to the worst days of March and April, it is still heterogeneous and somewhat unstable.
Mortality patterns
We used a panel of mortality indicators to characterize the national mortality patterns, both for the overall features and growing step of the daily mortality pattern (tab. 1.1), and for the decay step (tab. 1.2):
for the overall features and growing-mortality step:
the estimated maximum daily death rate, i.e., the daily mortality rate at the first daily mortality peak (
maxdr
), and the time (d) when it was reached, with respect to the first reported death (d2peak
);the daily mortality growth rate 11 days after the lockdown (
gr11
) before lockdown was implemented;the relative change (%) in fitted mortality growth rate on day 46 after the lockdown, with respect to
gr11
:rgr46
);
for the daily mortality decay step, after the first daily mortality peak: the estimated times (d) from the mortality peak to a 10%, 20%, …, 90%, and 95% reduction of the daily mortality rate, with regard to the peak
T10
,T20
,…T95
.
d2peak | maxdr100k | gr11 | rgr46 | |
---|---|---|---|---|
Albania | 24 | 0.04 | 0.11 | -52.30 |
Austria | 99 | 0.23 | 0.09 | -136.39 |
Belgium | 101 | 2.62 | 0.12 | -45.56 |
Bosnia and Herzegovina | 37 | 0.05 | 0.21 | 29.33 |
Bulgaria | 76 | 0.04 | 0.06 | 26.58 |
Croatia | 118 | 0.06 | 0.14 | -36.12 |
Czech Republic | 100 | 0.09 | 0.23 | 4.18 |
Denmark | 98 | 0.27 | 0.22 | -10.26 |
Estonia | 105 | 0.14 | 0.25 | -15.83 |
Finland | 117 | 0.17 | 0.18 | -24.09 |
France | 102 | 1.37 | 0.15 | -37.62 |
Germany | 107 | 0.29 | 0.14 | -35.14 |
Greece | 96 | 0.04 | 0.02 | -106.83 |
Hungary | 47 | 0.14 | 0.15 | -17.09 |
Ireland | 115 | 0.96 | 0.10 | -69.62 |
Italy | 89 | 1.31 | 0.09 | -43.62 |
Kosovo | 39 | 0.04 | 0.12 | -38.12 |
Latvia | 58 | 0.02 | 0.12 | -6.30 |
Lithuania | 111 | 0.05 | 0.11 | -11.25 |
Luxembourg | 103 | 0.48 | 0.12 | -34.50 |
Montenegro | 32 | 0.03 | 0.04 | -157.73 |
Netherlands | 98 | 0.89 | 0.14 | -29.33 |
North Macedonia | 109 | 0.11 | 0.16 | -19.27 |
Norway | 105 | 0.14 | 0.15 | -46.98 |
Poland | 52 | 0.06 | 0.17 | -9.38 |
Portugal | 45 | 0.31 | 0.13 | -61.21 |
Romania | 105 | 0.06 | 0.08 | -24.43 |
Serbia | 35 | 0.08 | 0.28 | -12.68 |
Slovakia | 47 | 0.02 | 0.24 | -32.92 |
Slovenia | 38 | 0.15 | 0.15 | -38.02 |
Spain | 92 | 1.71 | 0.14 | -40.68 |
Sweden | 112 | 0.96 | 0.20 | -9.84 |
Switzerland | 100 | 0.54 | 0.12 | -54.92 |
Turkey | 42 | 0.15 | -0.05 | 70.94 |
United Kingdom | 105 | 1.46 | 0.11 | -31.76 |
T10 | T20 | T30 | T40 | T50 | T60 | T70 | T80 | T90 | T95 | |
---|---|---|---|---|---|---|---|---|---|---|
Albania | 4 | 6 | 8 | 11 | 13 | 17 | 22 | 28 | 39 | NA |
Austria | 6 | 10 | 14 | 17 | 20 | 22 | 25 | 28 | 36 | 83 |
Belgium | 5 | 8 | 10 | 13 | 15 | 18 | 23 | 31 | 42 | 55 |
Bosnia and Herzegovina | 3 | 46 | 47 | 50 | 53 | NA | NA | NA | NA | NA |
Bulgaria | NA | NA | NA | NA | NA | NA | NA | NA | NA | NA |
Croatia | 5 | 8 | 10 | 12 | 15 | 18 | 22 | 28 | 43 | NA |
Czech Republic | 5 | 7 | 10 | 12 | 14 | 30 | 36 | 43 | 53 | NA |
Denmark | 5 | 8 | 12 | 18 | 26 | 30 | 35 | 41 | 61 | 72 |
Estonia | 4 | 6 | 9 | 12 | 16 | 23 | 30 | 39 | 50 | 55 |
Finland | 4 | 6 | 9 | 12 | 16 | 19 | 23 | 28 | 37 | 45 |
France | 5 | 7 | 9 | 12 | 15 | 18 | 23 | 31 | 42 | 57 |
Germany | 6 | 10 | 13 | 16 | 20 | 24 | 30 | 37 | 51 | 69 |
Greece | 5 | 8 | 11 | 17 | 22 | 27 | 40 | 55 | 76 | 92 |
Hungary | 8 | 12 | 16 | 20 | 25 | 31 | 40 | 49 | 67 | 80 |
Ireland | 4 | 6 | 8 | 10 | 13 | 15 | 19 | 26 | 39 | 54 |
Italy | 6 | 10 | 16 | 23 | 28 | 33 | 41 | 51 | 64 | 83 |
Kosovo | 5 | 8 | 10 | 12 | 14 | 17 | 20 | NA | NA | NA |
Latvia | 6 | 11 | 18 | 26 | 36 | 44 | 51 | 59 | 74 | NA |
Lithuania | 5 | 9 | 15 | 34 | 41 | 46 | 53 | 61 | 76 | NA |
Luxembourg | 3 | 6 | 9 | 13 | 17 | 22 | 28 | 34 | 41 | 47 |
Montenegro | 7 | 10 | 13 | 16 | 19 | 22 | 26 | NA | NA | NA |
Netherlands | 8 | 15 | 19 | 22 | 26 | 30 | 35 | 40 | 53 | 63 |
North Macedonia | 11 | 16 | 21 | NA | NA | NA | NA | NA | NA | NA |
Norway | 5 | 8 | 10 | 13 | 15 | 18 | 23 | 29 | 37 | 79 |
Poland | 7 | 11 | 17 | 22 | 27 | 73 | NA | NA | NA | NA |
Portugal | 8 | 11 | 13 | 15 | 17 | 21 | 50 | 54 | 95 | NA |
Romania | 32 | 34 | 36 | 40 | 45 | NA | NA | NA | NA | NA |
Serbia | 21 | 24 | 26 | 28 | 31 | 34 | 38 | NA | NA | NA |
Slovakia | 3 | 6 | 7 | 9 | 11 | 13 | 16 | 19 | 24 | 28 |
Slovenia | 4 | 7 | 10 | 13 | 16 | 20 | 25 | 29 | 39 | 51 |
Spain | 5 | 9 | 13 | 17 | 21 | 26 | 31 | 39 | 55 | 56 |
Sweden | 9 | 14 | 21 | 28 | 38 | 48 | 61 | NA | NA | NA |
Switzerland | 7 | 10 | 13 | 16 | 19 | 23 | 27 | 34 | 43 | 51 |
Turkey | 3 | 6 | 8 | 11 | 15 | 21 | 27 | NA | NA | NA |
United Kingdom | 8 | 12 | 16 | 20 | 25 | 32 | 43 | 57 | 83 | 97 |
- Daily mortality rate has decreased to a low level in most countries (fig. 1.8). However, it is still at least 50% of the national peak mortality rate in Bulgaria and North Macedonia.
We selected variables from the two sets of indicators, (i) based on their
completeness regarding the study countries, and (ii) the separate
dimensions they describe. We retained T10
, T30
and T50
from the
second set. Bivariate scatter plots and linear correlation between all
item pairs are shown on fig. 1.9.
We used a principal component analysis (PCA) on this dataset, followed by a hierarchical ascending classification analysis on the PCA country scores to identify clusters of countries sharing similar mortality patterns (fig. 1.10).
The two first eigen values of the PCA correspond to 57% of the total variance (fig. 1.10A: bar plot of PCA’s eigen values). We limit the interpretation of results to these two axes.
The correlation circle (1.10B) provides two keys for data interpretation:
The first (horizontal) axis is negatively correlated (left side) with high values of
rgr46
(relative mortality growth rate on day 46 after the lockdown with respect to mortality growth rate on day 11 after the lockdown), as well as high values ofT50
(time to 50% decay of daily mortality after the peak), andgr11
(mortality growth rate on day 11 after lockdown). It is also negatively correlated with high values ofd2peak
andmaxdr
(left side).The second axis is positively correlated with high values of
maxdr
, the mortality rate at the peak, andd2peak
, the time lag between the first reported death and the mortality peak.The two groups of variables at the basis of axes interpetation (
T50
,gr11
, andrgr46
for axis 1, vs.maxdr
andd2peak
for axi2) are approximately othogonal, i.e., they are weakly correlated.The projection of country scores on the plane made of axes 1 and 2 allows discriminating countries with unfavorable patterns - located in the bottom part of the plane, from the others. The right part of the plane encompasses countries with high daily mortality rate at the peak (
maxdr
), and late mortality peak (d2peak
). The left part of the plane gather countries with persisting daily mortality rate after the peak: high values ofT50
, `gr11
, andrgr46
.On the basis of the variance shut down according to the partition level, we select a partition of the countries into 7 classes: fig. 1.10C, dendrogram of the hierarchical ascending classification.
The scatter plot of countries according to their PCA scores reveals contrasted patterns (fig. 1.12). The first axis discriminates countries with slow mortality decay on the left (e.g., Bosnia and Herzegovina), versus others on the right (e.g., Ireland). Countries are ordered along the second (vertical) axis, according to their daily death rate at the peak (highest rate at the top: e.g., Belgium, lowest rate at the bottom, e.g., Montenegro). Greece seems to be in the most favorable situation, with low death rate at the peak, and fast mortality decay after the peak. However, it is bordered by countries with a persisting virus transmission, like Bulgaria, North Macedonia, or even Turkey. Therefore, even in the case of Greece, the epidemiological situation should be closely monitored.
The median value of the items used to define the categories are shown in table 1.3. We can order the 7 country categories according to their decreasing severity regarding COVID-19 incidence (axis 1: decreasing from top to bottom) and persistence (axis 2, decreasing from right to left).
: 1
[1] "Belgium, France, Ireland, Italy, Netherlands, Spain and United Kingdom"
------------------------------------------------------------
: 2
[1] "Lithuania, Romania and Sweden"
------------------------------------------------------------
: 3
[1] "Croatia, Czech Republic, Denmark, Estonia, Finland, Germany, Luxembourg, Norway and Switzerland"
------------------------------------------------------------
: 4
[1] "Bosnia and Herzegovina, Hungary, Latvia, Poland, Serbia and Slovakia"
------------------------------------------------------------
: 5
[1] "Turkey"
------------------------------------------------------------
: 6
[1] "Albania, Kosovo, Portugal and Slovenia"
------------------------------------------------------------
: 7
[1] "Austria, Greece and Montenegro"
These conclusions should be modulated by the recent evolution of the daily mortality rates, and daily mortality growth rates described above, showing that even the countries in the most favorable situations still needs monitoring (Austria, Greece, Turkey…).
The next step will be to assess whether these patterns are associated with specific lockdown measures or their implementation.
Category | d2peak | maxdr | gr11 | rgr46 | T50 |
---|---|---|---|---|---|
7 | 96 | 0 | 0.04 | -136 | 20 |
4 | 47 | 0 | 0.19 | -11 | 29 |
2 | 111 | 0 | 0.11 | -11 | 41 |
6 | 38 | 0 | 0.13 | -45 | 15 |
5 | 42 | 0 | -0.05 | 71 | 15 |
3 | 105 | 0 | 0.15 | -34 | 16 |
1 | 101 | 0 | 0.12 | -41 | 21 |
2 Discussion and conclusion
Because COVID-19 mortality is highly dependent on the age of infected patients [3], death rates standardized on the age structure of the target population are needed for safer country comparisons. Unstandardized rates are used here.
National epidemics, starting here by definition on the day of the first reported death, are not synchronized. Therefore, country comparisons for overall death rates must take this into account.
We introduced indicators or mortality persistence after the peak. Results of the PCA revealed that this dimension was different from the overall size of the epidemic, as well as features of the growing step.
European countries are heterogeneous in areas and population densities. In the largest countries (e.g., France, Italy…), there are strong sub-national differences, including in disease transmission, both in time and incidence rate.
Case definition (what is a COVID-19-related death?) is not the same according to the country. In addition, death notifications were initially based on mortality occurring in the hospitals, ignoring those recorded in the retirement homes (at least in Belgium, France, and Spain). Progresses were made to correct these issues, but there is still room for improvements. Moreover, diagnostic tests were not always done in deceased people without a prior COVID-19 diagnostic (both in hospitals and in retirement homes), thus introducing a downward bias in the estimates.
Also, the between-country comparison is limited to countries with a mortality peak. The absence of such a peak, or the existence of secondary peaks, is a concern by itself, and might be related to increasing virus transmission. For this reason, a particular attention should be paid to the situation in Bosnia and Herzegovina, Bulgaria, Croatia, Kosovo, North Macedonia, and Serbia.
European countries showing a (single) mortality peak are categorized with respect to the shape and intensity of their daily death rate pattern. Nevertheless, several issues are persisting regarding possible biases:
the age of patients at death was not accounted for, though the age of the population may vary between the countries.
There might be between-country differences with respect to the notification of deaths in retirement homes.
The next steps will be:
to account for these issues. For instance, we might use TESSy mortality data to assess some of the biases,
to use mobility and lockdown data - possibly split into sub-national data, to explain these between-country differences.
3 Data and methods
3.1 Data
Country borders were retrieved from the Global Administrative Unit Layers (GAUL) available from the FAO Geonetwork. We used the 2014 version (last updated in 2013=.
The GAUL dataset is distributed to the United Nations and other authorized international and national institutions/agencies. FAO grants a license to use, download and print the materials contained in the GAUL dataset solely for non-commercial purposes and in accordance with the conditions specified in the data license.
Because GAUL works at global level, unsettled territories are reported. The approach of GAUL is to deal with these areas in such a way to preserve national integrity for all disputing countries. Thus, data might not be officially validated by authoritative national sources and cannot be distributed to the general public.
Population predictions for 2020 were retrieved from the WorlPop database [4]. This dataset is licenses under the Creative Common Attribution 4.0 International License (CC BY 4.0).
Mortality data are available form the European Center for Disease Prevention and Control ECDC website which provides daily updates of the situation of COVID-19-related mortality. The database is copyrighted by ECDC [2005-2019]. Its content may be reproduced, adapted and/or distributed, totally or in part, irrespective of the means and/or the formats used provided that ECDC is always acknowledged as the original source of the material.
Lockdown data have been primarily extacted from the database maintained by the Oxford COVID-19 Government Response Tracker (OxCGRT), Blavatnik Scool of Government / University of Oxford [6]. This group systematically collects information on several different common policy responses that governments have taken to respond to the pandemic on 17 indicators such as school closures and travel restrictions. It now has data from more than 160 countries.Data use policy for this dtabase is the Creative Commons Attribution CC BY standard.
For the countries absent from this database, we took the lockdown dates from the national, institutional web sites.
3.2 Mortality indicators
The apparent daily mortality rate \(\pi_i\) is the daily death count \(m_i\) for a given country, scaled by the population size \(N\) of this country, expressed in units of 100,000. The population size was considered as constant during the COVID-19 epidemic. The index \(i\) is the number of days after the start of the epidemic. The start is set to the day of the first reported death related to COVID-19. In some countries, the counts \(m_i\) only refer to deaths occurring in hospitals. Therefore, there is a negative bias in death count: the actual count is higher than \(m_i\).
The cumulative mortality rate \(c_i\) is the cumulative death count, from the start of the epidemic (day of the first reported death) to the current day \(i\), scaled by the population size, and expressed in units of 100,000 inh., and considered as constant during the COVID-19 epidemic.
The mortality growth rate \(g_i\) is the difference in death counts between two consecutive days \(i-1\) and \(i\): \(g_i = m_i - m_{i - 1}\), scaled by the population size - expressed in units of 100,000 inh. Because the apparent daily death rate show strong variations, a smoothed mean and its 95% confidence band is fitted with a generalized additive model.
A negative binomial distribution is used to model the response, its “random” parameter being used to account to possible overdispersion of the observed counts, with respect to the usual Poisson assumption [7]. Such an overdispersion might be related, for instance, to the auto-correlation of successive daily death counts.
A cubic regression spline function with a fixed number of knots (one every week from the first death) is used to fit a smoothed time trend in the daily death rate. Besides its flexibility to account for sharp changes in daily death rates, this family of smoothing methods is suitable for computing the first derivative of the time trend in daily mortality rate, i.e., the daily growth rate [8].
To define and interpret the mortality indicators, we use the following clinical features of COVID-19 infections [3]:
an incubation period of 5 days on average (the time interval between the exposition to the virus and the onset of symptoms),
a time interval ranging from 6 to 41 days ’median: 14 days) between the onset of symptoms and the death of infected patients.
3.3 Interpretation of changes in mortality indicators
A decrease of mortality indicators should be observed between 11 and 46 days after the adoption of lockdown measures, with a maximum effect starting 19 days after this adoption.
To visualize these effects, vertical lines are added to the plots at days 0 (lockdown date), 11, and 46 after the official implementation of lockdown measures. Under these assumptions, the effect of lockdown on mortality reduction might start 11 days after the lockdown date.
Finally, the trend analysis described above can be used to estimate the time when the daily death rate starts decreasing, i.e., when the daily mortality growth rate becomes negative. For this purpose, we use the fitted values of the trend. Building on that, we can derive further indicators of the shape of daily mortality rate pattern.
3.4 Software
The freely available R software [9] is used for data management and analysis, as well as additional packages:
lattice
[10],latticeExtra
[11], andRColorBrewer
[12] for plotting;sp
[13] andraster
[14] for mapping.
This document is the output of an rmarkdown
[15] source
file compiled with Pandoc
, a universal document converter (http://pandoc.org).
3.5 Disclaimer
Much more comprehensive information is available in specific COVID-19 web resources like the European Center for Disease Prevention and Control (ECDC) and the World Health Organization (WHO), as well as on many national public-health agencies web sites. The website Our World in Data provides useful information on all aspects of COVID-19 [16].
The data analysis presented here has NOT been peer-reviewed, and thus, errors may exist. Comments and contributions are more than welcome.
3.6 Discussions on the report: questions and answers
3.6.1 Questions by Esther Van Kleef
Something doesn’t seem right to me. If you look at the epidemic curves and cumulative death rates, growth rates don’t seem at their highest at the time of lock down for almost all countries (Spain, Belgium, UK, Italy, NL, France all report only few number of deaths at time of lock down, and then exponential increase).
Starting from a very low rate (e.g. \(\frac{1}{65} 10^{-6}\) for France, even an exponential growth rate will make visible changes in daily mortality only after a few weeks. The estimated \(R_0\) being “only” \(\simeq\) 3, with an incubation of 5 days, the start of the epidemic is rather slow.
Then, also what is not clear to me is what the red lines are telling us. If I understand it correct, these are the number of days (11, 19, 46 days) post the mortality peak.
No, it is the time lag between the lockdown date and the early, median, and late death dates for people who got infected on the lockdown date. Therefore, we expect to see a decreasing mortality growth rate starting at lockdown date + 11 days.
But that’s not really informative for when lockdown measures were taken right? That’s probably why they are called “lockdown effects” but wouldn’t it be more interesting to plot the lines at the actual lock down times? Similar to the graph with the epidemic curve and cumulative death rates.
That’s what I did.
That then will give you better insight in what the growth rate was before and after lockdown measures were taken.
Regardless, I think that the conclusion (that growth rate was at it’s peak before lockdown) is now based on the fact that lines are constant for most countries post (i.e. x-axis>0) lockdown (rather than increasing). So constant on a log scale, meaning exponentially increasing right?
Not exactly. Because the mortality growth rate is the first derivative of daily mortality rate with respect to the time (that’s how it is estimated, from the fitted regression spline model, using a finite difference method), if it is constant (say \(b_1\)), then it means the daily mortality rate \(y_t\) increases linearly with time \(t\), with “linear” meaning an equation of the form \(y_t = b_0 + b_1 t\).
But this ignores the observation of many countries not reporting low numbers at the start of lock down. I wonder to what extend this might have to do with the GAM fitting procedure. How are the (number of) knots of the spline representing the time trend chosen? I wonder if, when automatic, could it be the GAM is fitting a linear a from the time observations are starting to increase, and be less random, hence ignoring the first phase of the epidemic for most countries?
You are right. I had only checked the results with data of countries which experienced many deaths. Consequently, I took the growth rate on day 11 post lockdown as the reference growth.
While implementing these changes, I realized I made a mistake in computing the death rate at the peak: I had taken the counts, not the rates. I fixed this. Of course these changes affected the list of indicators used to make the typology. The typology itself was changed.
I suppose this then telling the story, exponential increase in cases for an extended period was occurring for most countries despite lockdown? With exceptions for Poland, Hungary, Bosnia and Herz., Serbia, Slovenia, and Bulgaria?
Plus of interest is the difference in magnitude of the constant growth rate
might be completely off mark here, but just wanted to check with you
You were right. Thank you for your careful reading and very helpful comments.
[Talking about table 1] Isn’t maxdr representing absolute number of deaths in this column? If so, wouldn’t you want to normalize it by population and then stick in the PCA?
You are right. Done.
[Talking about table 1] Aren’t these three columns representing relative change post mortality peak (rather than after look down as listed in the text)? As this would be contradictory to figure 7 showing a constant positive growth rate post lock down for an extended period for most countries.
As explained above, I changed the list of indicators. I think the new columns are correct.
[Talking about table 1.] Are you sure these numbers represent growth rate (or absolute number of deaths?) at time of lockdown? Netherlands locked down according to figure 3 at about March 16. At that day, NL had about 25 deaths, day after the same, day before about 20.
Please take a look at new table 1.
[Talking about table 2] Might be an idea to plot these to allow for quicker and more visual comparison?
Done.
Acknowledgments
The work presented in this document is an output of MOOD (Monitoring Outbreaks for Disease Surveillance in a Data Science Context). MOOD has received funding from the EU Horizon 2020 Research and Innovation programme under grant agreement No 874850.
This document and associated code is managed on an instance of Forgemia, a Forge website held by INRAE, NUMM department:
References
1. Porta M. A dictionary of epidemiology. Oxford University Press; 2008. doi:10.1093/acref/9780195314496.001.0001
2. Godfrey E, Randolph S. Economic downturn results in tick-borne disease upsurge. Parasites & Vectors. 2011;4: 35. doi:10.1186/1756-3305-4-35
3. Rothan HA, Byrareddy SN. The epidemiology and pathogenesis of coronavirus disease (COVID-19) outbreak. Journal of Autoimmunity. 2020; 102433. doi:10.1016/j.jaut.2020.102433
4. WorldPop project. WorldPop: open access demographic data. GeoData Institute, editor. Southampton, UK: University of Southampton, GeoData Institute; University of Southampton; 2015. Available: http://www.worldpop.org.uk/
5. Tatem AJ. WorldPop, open data for spatial demography. Scientific data. 2017;4: 170004. doi:10.1038/sdata.2017.4
6. Hale T, Websterand S, Petherick A, Phillips T, Kira B. Oxford COVID-19 Government Response Tracker. Hale, Thomas, Sam Webster, AnnaBlavatnik School of Government. 2020. Available: https://covidtracker.bsg.ox.ac.uk/
7. Hinde J, Demétrio CGB. Overdispersion: models and estimation. Computational Statistics & Data Analysis. 1998;27: 151–170. doi:10.1016/S0167-9473(98)00007-3
8. Diggle PJ, Heagerty P, Liang K-Y, Zeger SL. Analysis of longitudinal data. 2nd ed. Oxford: Oxford University Press; 2002. p. 379.
9. R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2020. Available: https://www.R-project.org/
10. Sarkar D. Lattice: Multivariate Data Visualization with R. New York: Springer; 2008. Available: http://lmdvr.r-forge.r-project.org
11. Sarkar D, Andrews F. latticeExtra: Extra Graphical Utilities Based on Lattice. 2019. Available: https://CRAN.R-project.org/package=latticeExtra
12. Neuwirth E. RColorBrewer: ColorBrewer Palettes. 2014. Available: https://CRAN.R-project.org/package=RColorBrewer
13. Bivand RS, Pebesma E, Gomez-Rubio V. Applied spatial data analysis with R, second edition. Springer, NY; 2013. Available: https://asdar-book.org/
14. Hijmans RJ. Raster: Geographic data analysis and modeling. 2020. Available: https://CRAN.R-project.org/package=raster
15. Xie Y, Allaire JJ, Grolemund G. R markdown: The definitive guide. Boca Raton, Florida: Chapman; Hall/CRC; 2018. Available: https://bookdown.org/yihui/rmarkdown
16. Roser M, Ritchie H, Ortiz-Ospina E. Coronavirus Disease (COVID-19) – Statistics and Research. Our World in Data. 2020.
List (number of deaths in braces): Belgium (9,805), United Kingdom (45,312), Spain (30,340), Italy (35,089), Sweden (5,639), France (30,177), Netherlands (6,136), Ireland (1,753), North Macedonia (432), Switzerland (1,687), Luxembourg (111), Portugal (1,691), Germany (9,090), Denmark (611), Austria (711), Bosnia and Herzegovina (256), Turkey (5,508), Serbia (482), Kosovo (139), Hungary (596), Romania (2,038), Finland (328), Montenegro (32), Slovenia (111), Estonia (69), Norway (255), Bulgaria (308), Poland (1,627), Albania (113), Czech Republic (363), Croatia (122), Lithuania (80), Iceland (10), Liechtenstein (1), Malta (9), Greece (195), Cyprus (19), Latvia (31) and Slovakia (28).↩
Belgium, France, Ireland, Italy, Luxembourg, Portugal, Spain, The Netherlands, The United Kingdom.↩