Projections of ICU need by Country
Modeling current and future ICU demand.
An Exception was encountered at 'In [5]'.
Warning: This dashboard contains the results of a predictive model that was not built by an epidemiologist.
Note: Click a country name to open a search results page for that country’s COVID-19 news.
Execution using papermill encountered an exception here and stopped:
World map (interactive)
Hover mouse over map for detailed information.
- Details of estimation and prediction calculations are in Appendix and in Tables, as well as Plots of model predictions.
- Recent cases and Recent deaths refer to cases or deaths in the last 5 days.
Tip: Select columns to show on map to from the dropdown menus. The map is zoomable and draggable.
Tables
Projected need for ICU beds
Countries sorted by current estimated need, split into Growing and Recovering countries by current transmission rate. Only for countries with ICU need higher than 0.1 beds per 100k. More details in Appendix.
Projected Affected Population percentages
Top 20 countries with most estimated recent cases. Sorted by number of estimated recent cases during the last 5 days. More details in Appendix.
Methodology
- I'm not an epidemiologist. This is an attempt to understand what's happening, and what the future looks like if current trends remain unchanged.
- Everything is approximated and depends heavily on underlying assumptions.
- Projection is done using a simple SIR model (see examples) combined with the approach in Total Outstanding Cases:
- Growth rate is calculated over the 5 past days by averaging the daily growth rates.
- Confidence bounds are calculated by from the weighted standard deviation of the growth rate over the last 5 days. Model predictions are calculated for growth rates within 1 STD of the weighted mean. The maximum and minimum values for each day are used as confidence bands.
- Transmission rate, and its STD are calculated from growth rate and its STD using active cases estimation mentioned above.
- For projections (into future) very noisy projections (with broad confidence bounds) are not shown in the tables.
- Where the rate estimated from Total Outstanding Cases is too high (on down-slopes) recovery probability if 1/20 is used (equivalent 20 days to recover).
- Total cases are estimated from the reported deaths for each country:
- Each country has a different testing policy and capacity and cases are under-reported in some countries. Using an estimated IFR (fatality rate) we can estimate the number of cases some time ago by using the total deaths until today.
- IFRs for each country is estimated using the age adjusted IFRs from International IFRS study and UN demographic data for 2020. These IFRs can be found in
df['age_adjusted_ifr']
column. - The average fatality lag is assumed to be 8 days on average for a case to go from being confirmed positive (after incubation + testing lag) to death. This is the same figure used by "Estimating The Infected Population From Deaths".
- Testing bias adjustment: the actual lagged case fatality rate is then divided by the age adjusted IFR to estimate the testing bias in a country. To account for testing bias changes (e.g. increased testing capacity) this is done on a rolling window basis of two months (with at least 300 deaths). The estimated testing bias then multiplies the reported case numbers for each date to estimate the true case numbers (=case numbers that would be consistent with the deaths and the age adjusted IFR).
- ICU need is calculated and age-adjusted as follows:
- UK ICU ratio was reported as 4.4% of active reported cases.
- Using UKs ICU ratio, UK's testing bias, and IFRs corrected for age demographics we can estimate each country's ICU ratio (the number of cases requiring ICU hospitalisation).
- Active cases for ICU estimation are taken from the SIR model.
- Pre COVID-19 ICU capacities are from Wikipedia (OECD countries mostly) and CCB capacities in Asia. The current capacities are likely much higher as some countries already doubled or even quadrupled their ICU capacities.