1/ Calling @IHME_UW to correct GIGO model for Croatia which, using wrong inputs and unclear algorithms produces ludicrous projections and compromises trust in the integritiy of the whole project
@MLevitt_NP2013
@MLevitt_NP2013
2/ The model anticipates 1500 COVID victims in Croatia by 1-10-2020, the current no. being 108. That's 14 times more. Comparison with neighb. Serbia and BiH really puzzles, projected deaths/m. being 11 and 5 times higher though the current situation is inverse, especially trends
3/ The model itself screems 'unreliable' adding 1500 'expected' victims from the last 'observed' data and the 95% interval between 0 (!) and some 17,000 (!) cases growth. The latest seroprevelence figures by the gvt. are in the 2% zone, implying IFR of some 0.125% ...
4/ ... applied to the whole population would produce some 5,000 victims. Needless to say, not everyone will be sick L-T, and especially in the next 3 monts (or 2.5, having in mind victims do not die immediately after being infected).
5/ Reverse calculation from that IFR and 17,000 victims implies - if ALL inhabitants infected - Croatia needs to have 13.6 m. citizens, and it has only 4 m.
6/ The last 'observed (smoothed)' no. of deaths was 125 on 20 June, whereas on 20 June it was constant 107 for 7 days, with the total of 4 in the previous 20 days, meaning no meaningful 'smoothing' can produce 125 out of 107.
7/ For sure, the projections for today are 140 (126-206), whereas Croatia had only 1 case in some 20 days, the total being 108. There are currently only few cases on ventilators. Yesterday's 'infected' are calculated as 250, whereas Croatia had 52 (mostly asymptomatic or mild)
8/ Conslustion: Seeding a quasi-exponential growth model with wrong data can produce ANY results, and modellers should check the results, otherwise the whole process for all the countries is compromised.
