Gangelt — A representative study on the lethality of COVID-19
The actual lethality of COVID-19 (also Coronavirus, SARS-CoV-2) can only be estimated so far. A representative study from the German community of Gangelt now provides information on this.
Looking at the official lethality figures of the individual countries, a large deviation is noticeable. As of 13.04.2020, lethality in Germany was estimated at 2.36%, worldwide at 6.21% and in Great Britain even at 13.3% [1].
But can the same virus lead to such different lethality rates in different countries? Actually, it can be assumed that a virus is similarly lethal in the case of similar medical care. To do this, one must bear in mind how lethality is calculated: It is calculated by comparing the number of deaths of people who are detected as infected with COVID-19 with the number of people who are detected as infected with COVID-19.
The problem is that not all infected people are recognized as infected. So there are people who show symptoms but are not tested. In addition, about 50% of infected persons do not show symptoms and thus fall through the raster of those to be tested (more about why the official statistics cannot be used as a representative data basis can be read here). However, it can be assumed that almost all of the dead in Germany have been recorded. It can be assumed that these persons had previously shown strong symptoms and had received intensive medical care and were accordingly tested for COVID-19. The Robert-Koch-Institut (RKI) in Germany says: “We assume that the patients are diagnosed before they die” [2]. Nevertheless, there may of course also be an estimated number of unreported cases, but based on the previous remarks, this figure can be estimated as significantly lower. The different lethality rates also indicate that the number of undetected cases among infected persons varies from country to country. However, if the number of infected persons is higher than officially known, the calculated lethality decreases if the correct number of infected persons is expected. For this reason it is necessary to look at representative studies.
Update from 23.04.2020: Meanwhile the RKI makes a different statement: “I assume that we have more dead than officially counted”. [7] However, it is not said how many more deaths are assumed. The Ministry of Health of Hesse says that so far no excess mortality in connection with the coronavirus has been determined. Also the health administration of Berlin says that in the last weeks only a low excess mortality could be determined. This is probably primarily due to influenza and not to COVID-19 [7]. It should be noted that there may well have been excess mortality in other countries. In contrast to this, there are autopsy results according to which not all of the COVID-19 deaths listed in the statistics are also due to COVID-19 (see section “Autopsy results for COVID-19” in the article “COVID-19 from a data science perspective”). From a data-scientific point of view, it cannot be said on the basis of the data available up to now whether more or less people have died of COVID-19 than is listed in the statistics for Germany. Therefore the official value will be used in the statistics in the following.
Therefore one has to look at representative studies. The COVID 19 cases on the cruise ship “Diamond Princess” can be regarded as representative examples. All passengers and crew members were tested there, regardless of their symptoms, which allows a closer look [3]. The average age on the cruise ship was higher than that of the normal population. For this reason, the lethality was approximately 0.5% age-adjusted with an uncertainty of 0.25 percentage points in both directions. This value occurs although medical care on the cruise ship could not be guaranteed in the same way as on land with intensive care beds and respirators [4].
COVID-19 also spread on aircraft carriers. On the “Charles de Gaulle” all persons were tested [19], on the “USS Theodore Roosevelt” about 94% of the crew members [20]. On the aircraft carrier “Charles de Gaulle” 1081 crew members tested positive for COVID-19. Of these, 24 had to be treated in hospital, 2 of them in intensive care, none of them died so far [8]. On the aircraft carrier “USS Theodore Roosevelt” COVID-19 was detected in 660 crew members [9]. One died, which corresponds to a lethality of 0.15%. If both aircraft carriers are considered together, one infected crew member died from 1,741 infected crew members, which corresponds to a lethality of 0.06%. Certainly, these values cannot be transferred to the total population, as the crew members are on average younger and healthier.
Recently a representative study from Germany has also been available. The University Hospital of Bonn examined about 1000 persons in the particularly severely affected community of Gangelt, both for the presence of the virus and for the presence of antibodies. The study is still in progress, but a first interim result has already been announced after about 500 persons have been evaluated. After the announcement of the interim results, the first criticism was voiced. Criticism was voiced that there is only one test that can reliably detect COVID-19 antibodies and at the same time does not react to other, harmless corona viruses. However, no information on the test used has been provided from the interim results [5]. The study is being conducted at the University Hospital in Bonn by professors from institutes of virology, clinical chemistry and pharmacology, hygiene and public health, and also medical biometrics, computer science and epidemiology. Since it appears that experts are conducting the study, it should first be assumed that a test was used that reliably assigns the antibodies only to COVID-19 antibodies. Should it be determined in the future that the test was not suitable, the statistical error will have to be examined more closely.
Update from 24.04.2020: After Prof. Dr. Drosten (Director of the Institute of Virology at the Charité Berlin) had contact with Prof. Dr. Streeck (Director of the study in Gangelt and Director of the Institute of Virology at the University Hospital Bonn), he now also expects an “extremely solid, robust study” [10].
Results from Gangelt
Due to the investigation in Gangelt, interim results from Germany have recently become available. It was found that 15% of the population in Gangelt have already carried the virus either currently or in the past [6]. 509 persons were tested, so that arithmetically about 76 persons must have carried the virus in themselves. An exact figure was not given in the presentation of the interim results.
When considering the numbers, both the specificity and sensitivity of the test must be considered. The specificity indicates the percentage of tests that actually test negative in people who have not developed antibodies against COVID-19. According to the manufacturer, the specificity is > 99%. If one assumes that of the 509 test participants, approximately 440 people did not develop antibodies, about four people would be overstated in the statistics. There would therefore only be 72 persons. However, the sensitivity of the test must also be considered.
The sensitivity indicates the percentage of people who actually developed antibodies against COVID-19 that test positive. According to the manufacturer, sensitivity is 100% if at least 20 days have passed since the onset of symptoms, otherwise it can drop to 87.5% [18]. If we now look at the 72 people who were correctly tested for COVID-19, this would mean that between 72 and 82 people were actually infected with COVID-19. However, it is doubtful that at least the required 20 days had elapsed for all those actually infected, which is why a 100% sensitivity cannot be assumed. However, if the sensitivity decreases, the number of people actually infected increases. It is therefore plausible to continue to assume that the previously calculated 76 people who were or are infected with COVID-19 are still infected. If the value were to be higher than the 76 persons, the percentage of the population already infected would increase and thus the calculated lethality would decrease.
If one looks at the beta distribution to the infected persons, one can see that the actual value for the percentage share is in the range between 12% and 18% with an accuracy of 95%. The most likely value for the percentage of infected persons in Gangelt is therefore 14.79%. If the percentage of infected persons can be confirmed from 15% to the total 1010 study participants, the certainty increases and the range of the 95% confidence interval decreases to around 13% to 17%. The most likely value for the proportion of infected persons in Gangelt would then be 14.98%.
The interim results were criticized by the epidemiologist Krause: “One should not take all results from these households and convert them into percentages, but at best one person per household” [6]. As a reason he gives that the risk of infection within the households is many times higher than in the population in general and a complete count of all family members would therefore not be correct. But is this statement tenable? In a representative study, the group under investigation should, if possible, reflect the cross-section of society in order to be representative of society. The criticism is that the risk of infection is not comparable. But the conditions in Gangelt were the same as in the rest of Germany. Also in the rest of Germany the undetected infected persons are still in contact with the persons living in the same household and can infect them, as it was the case in Gangelt.
Surely the value of 15% infected persons cannot be transferred to the whole federal territory, because Gangelt was particularly strongly affected. However, taking into account the statistical error, the value for lethality can be transferred to Germany. Because the counting of deaths with COVID-19 is the same everywhere in Germany and must be reported to the RKI.
For the interim results of the investigation in Gangelt, a lethality of 0.37% was given. When considering the beta distribution, it can be assumed with an accuracy of 95% that the actual lethality is between 0.15% and 0.69%. The density function is apparently left-stiff/right-skewed. As a result, the most likely value for lethality is 0.32%, which is lower than the calculated lethality of 0.37%.
Here it must be taken into account that the assumed lethality of 0.37% was calculated from the 7 known deaths and the 1879 infected persons extrapolated from the interim results of the study in Gangelt. This extrapolated figure is derived from the 15% infected persons observed in the study. As shown above, the actual percentage of infected persons in Gangelt is between 12% and 18% due to the confidence interval of 95%. Depending on the actual number of infected persons, the lethality rate changes. If we now look at the lower limit of 12% of infected persons in Gangelt, the lethality rate is 0.47%, whereby the 95% confidence interval for lethality ranges from 0.19% to 0.87%. The most likely value for lethality would therefore be 0.4%.
If one considers the upper limit of the confidence interval for the proportion of infected persons in Gangelt of 18%, the result is a lethality of 0.31% with a confidence interval of 0.12% to 0.57% and a most probable value of 0.26%.
Summary
In summary, according to the interim results of the study, the proportion of infected persons in Gangelt ranges between 12% and 18%, the average lethality is between 0.31% and 0.47%, the most probable lethality is between 0.26% and 0.4% and the range of lethality is between 0.12% and 0.87%. These figures correspond very well with the findings on the “Diamond Princess”, where the lethality rate was given as 0.5% and the probable range for lethality as 0.25% to 0.75%.
By means of the determined lethality, it can now be estimated how many people in Germany were or are actually already infected with COVID-19. According to John Hopkins University, there were a total of 3,022 deaths among the 127,854 detected COVID-19 infected persons in Germany on 13.04.2020 [13]. With the lethality of 0.37% calculated in Gangelt, however, this means that 816,757 people (approx. 1.0% of the total population) in Germany were or are already infected with COVID-19. The most probable lethality of 0.32% means that 944,375 people (approx. 1.1% of the total population as of 13.04.2020) are already infected. With a lethality range of 0.12 to 0.87%, this results in 347,356 to 2,518,833 infected persons (approx. 0.4 to 3.0% of the total population).
It can be seen that the range of those actually infected is still quite high. Therefore, the final results from Gangelt are very interesting in order to reduce the range of confidence intervals and thus increase the reliability of the figures. Further large-scale representative studies are also desirable. Despite the uncertainties regarding the number of infected persons, it can be shown that there is a high number of unreported cases in Germany.
Note from the author: This article is an excerpt from my article “COVID-19 from a data science perspective”. However, only the investigation from Gangelt is considered here. The main article also considers why the official statistics cannot be regarded as a representative data basis, how many people actually died from COVID-19, how many people in Germany are probably already infected and what intensive care is provided.
Note from the editors: Towards Data Science is a Medium publication primarily based on the study of data science and machine learning. We are not health professionals or epidemiologists, and the opinions of this article should not be interpreted as professional advice. To learn more about the coronavirus pandemic, you can click here.
Author: Daniel Haake, M.Sc. in Data Science, currently working as Data Scientist
Bibliography
[1] Statista, „Letalitätsrate beim Coronavirus (COVID-19) in den am stärksten betroffenen Ländern,“ 13.04.2020. [Online]. Available: https://de.statista.com/statistik/daten/studie/1103785/umfrage/mortalitaetsrate-des-coronavirus-nach-laendern/. [Access on 13.04.2020].
[2] ntv, „Warum ist die Sterblichkeitsrate so niedrig?,“ 20.03.2020. [Online]. Available: https://www.n-tv.de/panorama/Warum-ist-die-Sterblichkeitsrate-so-niedrig-article21657232.html. [Access on 10.04.2020].
[3] A. Vera und S. L. Erdman, „Nearly half of Diamond Princess cruise ship passengers and crew who had coronavirus were asymptomatic when tested, CDC report says,“ CNN, 24.03.2020. [Online]. Available: https://edition.cnn.com/2020/03/24/us/diamond-princess-cruise-ship-asymptomatic-tests/index.html. [Access on 10.04.2020].
[4] Stern, „Seit einer Woche unter Quarantäne — Situation auf “Diamond Princess” spitzt sich zu,“ Stern, 12.02.2020. [Online]. Available: https://www.stern.de/gesundheit/coronavirus--situation-auf-kreuzfahrer--diamond-princess--spitzt-sich-zu-9134930.html. [Access on 10.04.2020].
[5] K. Zinkant, „Kritik und Zweifel an Studie aus Heinsberg,“ Süddeutsche Zeitung, 10.04.2020. [Online]. Available: https://www.sueddeutsche.de/wissen/heinsberg-studie-herdenimmunitaet-kritik-1.4873480. [Access on 10.04.2020].
[6] H. Prof. Dr. Streeck, G. Prof. Dr. Hartmann, M. Prof. Dr. Exner und M. Prof. Dr. Schmid, Universitätsklinikum Bonn, 09.04.2020. [Online]. Available: https://www.land.nrw/sites/default/files/asset/document/zwischenergebnis_covid19_case_study_gangelt_0.pdf. [Access on 10.04.2020].
[7] NTV, „Das Coronavirus und die Todesstatistik,“ 21.04.2020. [Online]. Available: https://www.n-tv.de/panorama/Das-Coronavirus-und-die-Todesstatistik-article21728569.html?fbclid=IwAR2IqJe6CVPLIiJ3MFVfesj54otbEDTMxr_zMxx9HWw9_MQC1WhumYI9C_Y. [Access on 22.04.2020].
[8] Welt, „Mehr als 1000 Infizierte an Bord — Virus verbreitete sich wohl nach Stopp in Brest,“ 19.04.2020. [Online]. Available: https://www.welt.de/politik/ausland/article207365541/Corona-auf-Flugzeugtraeger-Charles-de-Gaulle-Virus-verbreitete-sich-nach-Stopp.html. [Access on 21.04.2020].
[9] Spiegel Politik, „USS “Theodore Roosevelt” — Zwei Drittel der Seeleute auf US-Flugzeugträger zeigen keine Symptome,“ 18.04.2020. [Online]. Available: https://www.spiegel.de/politik/ausland/coronavirus-auf-uss-theodore-roosevelt-zwei-drittel-der-seeleute-auf-us-flugzeugtraeger-zeigen-keine-symptome-a-b7d51acb-9969-4262-bd87-a0e9333899a1. [Access on 21.04.2020].
[10] C. Prof. Dr. Drosten, „Maybrit Illner Twitter,“ 17.04.2020. [Online]. Available: https://twitter.com/maybritillner/status/1250929078930223105?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1250929078930223105&ref_url=https%3A%2F%2Fembeds.br24.de%2Fembed%3Fid%3D534998. [Access on 24.04.2020].
