How to better account for the dark figure

[ccpf]

#264 🙋 Feature Request

It would be nice to add 2 parameters to the user input: 1 parameter to account to the suspected dark figure in deaths (e.g., in Italy this factor is between 4-10) and a second parameter to estimate the dark figure in the total cases.

🔦 Context

You base your age-group-specific parameters on Chinese data, i.e., CONFIRMED cases. Hence your percentages of hospitalization, ICU, and mortality are likely to be overestimated.

By being able to estimate the dark figures, we could still try to fit the hospitalization curve to the actual data for instance, while having modelled deaths and total cases numbers that are much higher than currently achievable. These higher case numbers are important when you want to estimate at which point herd immunity might be reached.

😯 Describe the feature

You can keep using the same death and hospitalization rates as you do now, but allow the user to specify the factors by which the dark figure (in case numbers and deaths) exceeds the confirmed cases.

💻 Examples

E.g., for Northern Italy it has been reported that the total deaths are between 4-10 times the official figure (see 1st link below). At the same time, the infection fatality rate (IFR) is expected to be around 0.5% (based on the cruise ship study, 2nd link below) or even only 0.37% (based on the German antibody study, 3rd link below). Based on this, we could already have about 100k deaths (vs 20k officially reported) and with the 0.5% IFR, this would put the total number of positives in Italy at ~20M (vs the 160k of currently reported cases). Such high case and death numbers are not achievable with the model in its current form if we want a reasonable fit with the most reliable number we have: the number of hospitalizations.

💁 Possible Solution

As mentioned above: allow the user to specify a factor by which the real cases exceed the confirmed cases. Same for deaths.

Related

• https://www.corriere.it/politica/20_marzo_26/the-real-death-toll-for-covid-19-is-at-least-4-times-the-official-numbers-b5af0edc-6eeb-11ea-925b-a0c3cdbe1130.shtml
• https://www.medrxiv.org/content/10.1101/2020.03.05.20031773v2
• https://www.land.nrw/sites/default/files/asset/document/zwischenergebnis_covid19_case_study_gangelt_0.pdf

[rneher]

This is an interesting suggestion. When fitting the model to case count data to get the initital parameters roughly right, we actually do account for incomplete case counts, but take the number of death at face value. We think the data from Nembro is a bit of an extreme outlier with typical undercounting of fatalities to be more like 2-fold, so fitting to death is a more accurate procedure.

would you want to down-scale the model trajectories or up-scale the case-count data?

[ccpf]

I suppose up-scaling the case count data would be good, although I am not sure whether this would affect your hospitalizations as those should not really change. We are also working on the Spain data, for instance, and there the ministry of health admitted that real numbers may be about a factor 15 higher than officially reported (mainly due to limited testing - see here: https://elpais.com/sociedad/2020-04-07/mas-del-90-de-contagios-estan-ocultos.html), so if we could incorporate these up-scaling factors (without affecting hospitalizations), we would get a more accurate picture and better fits, which is also important when looking at present and future mitigation measures. Thanks.