Fit model also with IVA cases

[mwigh]

Have you tried fitted using IVA cases?
How does it look?

I tried many things in python yesterday and used e.g. those parameters:

• IVA_DELAY=11 (from intensivårdsregistret)
• MEAN_LENGTH_IVA=10 (from https://www.folkhalsomyndigheten.se/contentassets/4b4dd8c7e15d48d2be744248794d1438/skattning-av-behov-av-slutenvardsplatser-covid-lombardiet.pdf))

I tried to fit to:

• Currently IN_IVA in sthlm from https://www.svt.se/datajournalistik/corona-i-intensivvarden/
• NEW_IVA in sthlm from (from intensivårdsregistret excluding cases from abroad)

For me it really does not match. If I e.g change to IVA_DELAY=5 it looks really good (and RMSE is smaller).
I tried to fit it by convolve the I_r with different types of arrays, but none gives a good result.
Here is the code for reference:
https://github.com/mwigh/SEIR-model-Stockholm-python/blob/37f76eef65804632bbab9110dc310539982925a5/Script/Estimate_SEIR_sthlm.py#L175

You can set OPTIMIZE_FOR_IN_IVA_OR_NEW_IN_IVA to 1 or 2 if you want to optimize for either IN_IVA or NEW_IN_IVA (together with infected cases).

Is there any flaw in my idea?
Also, the code is not producing any file outputs, it just show a single plot and takes approx 20s to run

IVA_DELAY=11:

IVA_DELAY=5:

[FohmAnalys]

Thanks for sharing this!

One possible reason for a shorter delay to fit better is how data is recorded. We use the 'epidate' to make the analysis. Epidate is for most individuals the date of symptom onset and for them I guess this delay of ~11 days from SIR is a good estimate. However, for some of the cases the date of sympton onset is not recorded and then the epidate is equal to the testing date and for these the delay should be shorter.

One other thing good to know, if you would use the data from FHMs webpage instead, the date reported there is the 'reporting date', which usually is even later.