For the purpose of showing closer to real numbers on fatalities world wide than using reported numbers, coronasim.eu uses an underestimations formula to calculate these numbers.
The current coronasim.eu model is based on using data from research reports from Italy in addition to seroprevalence research in India. Using those two data points a linear function is used where the x-axis is the assumed healthcare system quality. This gives an approximate estimate on an underestimation percentage for all countries that can be assigned a value for the healthcare system quality.
The reasoning behind both the Italian and Indian values are well documented in the referenced research on https://coronasim.eu/page2.php. However, the Italian research was done quite early in the pandemic, and includes the chaotic situation in parts of Lombardia. It is reasonable to believe that correct classification suffered during those early winter and spring days and that correct classification on cause of deaths has most likely improved since then. Therefore, the current underestimation percentages in most countries should be lower than in earlier months of the pandemic. Still, any calculations need to take into account the underestimations that have happened before.
Deaths | Coronavirus in the UK shows fatalities in the United Kingdom due to covid-19. The “Deaths within 28 days of positive test” 1) is the figure that is reported by ECDC. However, the “Deaths with COVID-19 on the death certificate” 2) is considerably higher.
The numbers on UK fatalities up to Friday 6 Nov were:
If a person gets SARS-CoV-2 and dies, it is reasonable to assume that in most cases the virus was the factor that triggered the death. Therefore, we assume that of the 14680 unreported deaths, ⅔ can be classified as covid-19 deaths, i.e. 14680 * ⅔ = 9787. This results in an underestimation percentage for the UK on Nov 5 to become: (49193+9787) / 49193 - 1 = 0.199 = 19.9%.
(This also means that we do not add any underestimation from death certificates that might possibly erroneously not mention covid-19 as a cause even if it could have been. This might be a source of error.)
As this is the best estimate we are able to derive for the UK at this point, we set the underestimation percentage for a country with the healthcare system quality as UK's to 18%, “rounded down” since we are assuming it to be declining gradually.
Assessing the Impact of the Covid-19 Pandemic on US Mortality: A County-Level Analysis estimates the underestimation to be 35.8% for the United States from February until September. This would then include the early, chaotic period in e.g. New York. In lack of better data, we set the assumed underestimation until now in the US to be approximately 27%, considerably lower due to assumed improvements over time from September and on.
We are then also assuming that countries with a similar healthcare system quality as the US' will have the same percentage.
Belgium is still counting suspected, unconfirmed fatalities in their official death toll figure. Additionally, we observe that Belgium adjusted their figures retrospectively on June 26, meaning that possible underestimations from the early days actually are intended to be included in the now published data. (“Update 27 June 2020, Belgium: Retrospective counts adjusted based on additional data received to apportion excluded cases and deaths to the date they were initially reported. Retrospective adjustments made from 2 March to 26 June as national authorities revised their historical data.” - https://www.who.int/docs/default-source/coronaviruse/data-change-log.pdf?sfvrsn=c15ed704_341 )
Until now, underestimation for Belgium was fixed at 15%. Due to new knowledge on this matter, now we do not adjust the death toll, but keep it as reported by Belgium authorities.
As of November 22, 2020, the calculations use the new underestimation function. This gives expected results: Countries with seemingly good healthcare systems get a reduced number of estimated fatalities, while countries with low quality healthcare systems still would struggle with underestimation comparable to previous figures.
An improvement on the underestimation calculation would be to incorporate excess deaths statistics into the equation. However, the way to do so reliably for all countries of the world is not obvious, as excess deaths from other causes than covid-19 are now heavily affected by various covid-19 restrictions and measures.
Calculating underestimations does of course assume that the source data is correct. Coronasim.eu calculations rely on correct reporting from each countries’ authorities and it cannot “correct” numbers that are reported wrongly (mostly too low) on purpose, which is believed to happen in many countries for political or other reasons. Therefore, the total number of fatalities and number of cases estimated by coronasim.eu will also be wrong (mostly too low) for these countries and for the world total.