Deaths by Covid-19 in Sweden

An analysis of geographical patterns of Covid-19 in Sweden

It all started at the end of December 2019, when a new strange disease was discovered in Wuhan in China (WHO Timeline - COVID-19, n.d). No one knew that this disease would be spread across the whole world only 3 months later, that this disease would close the whole world down. The year 2020 did not start as I expected… The spread of Covid-19 has put great pressure on governments all over the world and how different countries handle it has been interestingly to follow. Lock-downs, social distancing, work from home are today more common words than ever. Although our world has historically been through previous pandemics, e.g. the Spanish Flu or Yellow Fever, Covid-19 is the most widespread pandemic in modern times (LePan, 2020). There is no obvious way to handle a pandemic and learning from historical pandemics is difficult as the world has changed and is at a different level technologically and hence also looks different globally. Today, there is still no correct answer to how to handle the pandemic, Covid-19 correctly. It will take months or even years until it can be seen. Sweden, the small elongated country up north, have excelled in how they have handled Covid-19 this far. The little country with its 10 million population had on the 22 of June 2020 5000 deaths from Covid-19. If Covid-19 deaths per million are observed, seen in figure 1, it can be seen that Sweden together with Italy, Belgium, France, and the UK stand out from the rest of the world. During the spread of Covid-19 in Sweden the Caring homes have been severely affected, for the target group 70+, almost 50% of those who has died in Covid_19 died in elderly housing /caring homes (Thomsen, 2020). Therefore, the analysis will be made from different perspectives. Both to explore if geographical patterns of Covid-19 can be seen from the total death numbers as well as the death numbers from the caring homes. An interesting pattern would be if the deaths of the Covid-19 is spatially clustered. In Sweden it can be seen that the spread of the virus differs from county to county and it would be interesting to investigate closer to this. Can some kind of geographical pattern of the spread of Covid-19 be seen in Sweden?

Figure 1: Total deaths by Covid-19 per million, all over the world

Methodology

To find an answer to the questions a statistical analysis has been conducted. For this research two focus-areas has been chosen, Sweden as a whole and specifically on Stockholm, the most affected city in Sweden, see figure 2. Data of deaths by Covid-19 was chosen for this analysis. This choice was made because data on measurement of the infection spread did not seem like accurate data, since the government recently started to increase the number of Covid-19 test. Figure 2: Spatial distrubution of percent of total deaths in Sweden

This analysis aims to investigate if geographical patterns of the spread of Covid-19 can be seen in Sweden as well as in the Stockholm region. A spatial autocorrelation are commonly used to analyze spatial patterns and clusters, specially characterizing correlation between nearby locations in space, and will therefore be used to answer if spatial clusters of Covid-19 can be seen in Sweden. Because of its multidimensional, multidirectional feature and complex shapes, the spatial autocorrelation is a more complicated than a one-dimensional autocorrelation. Moran’s I is a measurement within spatial autocorrelation and explains if a spatial correlation can be seen as well as if it is positive or negative. To explain the process of this analysis the spatial autocorrelation can be divided in 5 steps (Gimond, 2019). First the neighboring polygons must be defined, in our case are the municipalities of Sweden or Stockholm the polygons. In this analysis we define the neighboring municipalities, this is shown in figure 3. Second step is to assign weights to the neighbors, in this case each neighboring polygon will be assigned equal weight. Then the Moran’s I value is calculated. This is done by creating a regression model with the weighted polygon values and the death data. When plotting this the Moran’s I value is represented by the slope. Then the Moran’s I value has to be tested to observe if spatial autocorrelation can be seen. This is done by randomly permuting the Covid-19 death values across all the municipalities and then fit a regression model to each permuted set of values. The original Moran’s I value is then compared with the distribution from the simulation. Last step is to calculate a p-value, from the simulation a one sided p-value is calculated. This is done by taking the values from one side of the Moran’s I value and divide it with all simulated values. The p-value helps the analysis and explains the probability of the event happening by chance. For the spatial Autocorrelation, both country- and county wise, the investigated areas has been divided in municipalities. This means that, the spatial Autocorrelation indicates if the number of deaths of Covid-19 within the municipalities are clustered. When performing a spatial autocorrelation, a number of adjustments must be made. NA values constitute problems and must be excluded. This may change the map to some extent but not significantly in ours example since only a few was excluded. Data have been collected from the Swedish Statistical Central Bureau (SCB), World Health Organization (WHO) and the Swedish National Health Authority (WHO Coronavirus Disease (COVID-19) Dashboard, no date; Coronaviruset covid-19, no date; Bekräftade fall i Sverige — Folkhälsomyndigheten, no date). Figure 3: Map of Sweden, neighboring municipalities

Result and Discussion

Analysis of total deaths by Covid-19 in Sweden and Stockholm

In this section the results will be presented. Results that aim to answer if spatially clusters of the deaths of Covid-19 be seen in Sweden? First, the country as a whole has been observed. From the spatial autocorrelation of deaths of Covid-19 in Sweden’s municipalities, a positive Moran’s I value is obtained, 0,14. This value implies that the death of Covid-19 are somewhat spatially clustered through the municipalities. In figure X the y-axis are the weights of the neighbors and the x axis are the number of deaths. The dots represent the municipalities of Sweden and the slope of the blue line represent the value of Moran’s I, 0.14. Figure 4: Graph over the weights and total deaths, Moran’s I value is represented by the slope

To assess if the observed Moran’s I value, in this specific case 0,14, is significant, it is compared to the random simulated distribution. Figure X show the simulation where the red line presents our Moran’s I value, 0,14. Figure 5: Random simulation of deaths of Covid-19 in Sweden, the Moran’s I value is represented by the red line

A one-sided p-value have been calculated to analyze the probability of this being a coincidence, this is done by dividing the values on one side of our Moran’s I value and divide it by all values. The p-value for observing clustering of deaths from Covid-19 in Sweden municipalities is 0.0066. This tells us that it is a low probability, 0.66%, that the seen autocorrelation has arisen by chance. Therefore, it can be seen that the deaths by Covid-19 are somewhat spatially clustered.

Interestingly if only the municipalities of Stockholm are observed, a different result is obtained. Spatial autocorrelation gives a small negative Moran’s I value, of -0.082. The negative nature of the coefficient implies a spread distribution like a chessboard, instead of a clustered for a positive value. Although, the value is also relatively close to zero which implies that no spatially autocorrelation can clearly be seen. Figure 6: Random simulation of deaths of Covid-19 in Stockholm, the Moran’s I value is represented by the red line

As seen the Moran’s I is in the middle of the simulated distribution, the simulation can be seen in figure 6. This implies therefore that the seen spatial autocorrelation is probably emerged by chance and that no cluster can be seen in the Stockholm area. The calculated p-value gives a value of 0.375 and with a significance level at 38% no spatial autocorrelation can be determined.

Our results this far show that for the whole country, spatially clusters of the deaths by Covid-19 can be seen but if only Stockholm is observed, no pattern can be seen. Since Stockholm has the most cases of Covid-19, it may be the Covid-19 cluster of Sweden. To look further into this an additionally analysis has been made, a spatial autocorrelation of Sweden’s all municipalities except the ones in the Stockholm region. This gave a Moran’s I value of 0.026 and a p-value of 0.195. The spatial autocorrelation for this example implies that the cluster pattern that was seen for the Sweden as a whole is not as strong if Stockholm municipalities are not included. This implies that the Stockholm region is probably the cause for a spatial autocorrelation in Sweden, this does however need further analysis to establish.

Analysis of deaths by Covid-19 at Caring homes in Sweden and Stockholm

The problem with Covid-19 spreading in elderly /caring homes has been a serious in Sweden. Can similar geographical patterns be seen for those who died in Covid-19 on caring homes?

If a spatial autocorrelation is done on all the municipalities in Sweden and the total deaths in a caring home, it gives similar results as the spatial autocorrelation of total deaths. The Moran’s I value was 0.17 and a low p-value at 0.008. The deaths by Covid-19 in caring homes can therefore be seen to be somewhat spatially clustered. A relatively expected result since the majority of deaths are over 70 years and that half of these occurred at the caring home. When only Stockholm was observed similar to the earlier Stockholm’s results was found, seen in figure 6.

Conclusion

To answer if some kind of geographical pattern of the spread of Covid-19 be seen in Sweden a spatial autocorrelation has been performed. Focusing on if the virus form spatially clusters within the municipalities in Sweden and Stockholm. From the analysis following patterns could be seen.

• The deaths from Covid-19 at a municipal level indicate spatially clusters. Both when observing the total death numbers as well as the deaths at caring homes.

• No clear geographical pattern can be observed in the Stockholm region.

When performing a spatial autocorrelation of the municipalities of Sweden where Stockholm region is excluded, gives a weaker spatial autocorrelation than when all municipalities are included. This implies that Stockholm may be the reason for the spatial clusters of Covid-19 deaths in Sweden, however this needs further analysis to be established.

Short reflection of the gained knowledge

In the previous project, Covid-19 was investigated with the whole world in focus and how they handled the pandemic differently. The goal was to try to find Social / Economic patterns within the country that led the governmental decisions. An issue here was among others that it is a large national for error margins, since influencing factors are difficult to define. It is usually several factors that together make a contributing effect. By focusing on one country and its regions makes it more possible to investigate closer to geographical patterns and the socioeconomic influences. This means that countries’ differences are excluded, like correctness off reports etc. which was a bit overwhelming during the last project. It made me realize the importance of narrow the study area to be able to more easily see patterns and reduce influencing factors, even though these are still present. From the beginning of this subject visual data was something I wanted to look more into, I started with this in assessment 1.

References

Bekräftade fall i Sverige — Folkhälsomyndigheten (no date). Available at: link (Accessed: 27 June 2020).

Coronaviruset covid-19 (no date) Sverige i siffror. Available at: link (Accessed: 27 June 2020).

Gimond, M. (2019) A basic introduction to Moran’s I analysis in R. Available at: link (Accessed: 27 June 2020).

LePan, N. (2020) Visualizing the History of Pandemics, Visual Capitalist. Available at: link (Accessed: 23 June 2020).

WHO Coronavirus Disease (COVID-19) Dashboard (no date). Available at: link (Accessed: 27 June 2020).

WHO Timeline - COVID-19 (no date). Available at: link (Accessed: 23 June 2020).

Thank you for reading!

Emma Välme