Deaths per capita from COVID-19 are more informative than confirmed cases

5 min read COVID-19

The rate of testing for COVID-19 is increasing in many countries. As such, using the number of confirmed cases over time is not a reliable method for tracking the spread of the disease. Nevertheless, many articles—including this one in the Financial Times—still talk about the number of confirmed cases. That article also focuses on number of deaths though, which is much better and leads us to what I think should be used.

We should be using deaths per capita

First of all, we should be using per capita statistics. For example, 100,000 cases in China vs 100,000 cases in Ireland would be very different things, because of what they imply for concentration of cases. Per capita (assuming a billion people in China and 5 million in Ireland), this would mean 1% of people in Ireland have the disease compared to 0.01% of people in China: 1 in 100 vs 1 in 10,000. Of course, just using per capita by country, you miss a lot too. A particularly high concentration of cases in Wuhan is not the same as a more even spread of cases throughout China. You can fix this by dividing by province or county (if you have that level of granularity in your data); I do this by state for the US. I’m going to use per million people instead of per capita, because it’s a bit easier to think about.

Second, reported deaths from the virus are more reliable than confirmed cases. That’s because the more ill you are, the more likely you are to get a test. The number of people reported to have died of COVID-19 infection will be closer to the true number of people who died of COVID-19 than the number of people reported to be infected will be to the true number of infected individuals.

For timelining, I’ll use days since the tenth confirmed death from COVID-19 in a particular country, just like the Financial Times article.

Deaths per capita will lag other measures

I don’t mean to say that deaths per capita is the only useful thing. It actually lags other measures (in particular the number of infected individuals) by a few days because the disease does not kill quickly. In particular, if a countries rate of testing is steady or increasing but the absolute number of people testing positive per day is decreasing, that’s a great sign and a sign that you’ll see sooner than a drop in deaths.

Data source, preparation and limitations

Johns Hopkins University Center for Systems Science and Engineering (JHU CSSE) has put together a great COVID-19 data repository. Note that I’ve smoothed the data to make the plots more readable. This means that only sustained changes will appear on the plots (e.g. a one-day drop in deaths will be smoothed out). This is a good thing as it prevents us from reading too much into single data. Bear in mind that different countries have different measurement methods and capabilities, so they aren’t strictly comparable. However, provided that within a given country, the measurements are done in a consistent manner, one can still do reliable comparisons of the shapes of the countries’ curves.

A note on social distancing

Social distancing is mostly new (where it exists at all) and due to the long (approximately 2-week) latency period of the virus (latency is roughly the period between a person encountering the virus and showing symptoms of infection), we probably won’t see the benefit of social distancing in these statistics for at least another week. If by then there’s still no improvement, that would be cause for huge concern. In Taiwan, for example, the government is using people’s mobile phone location to ensure they stay at home. If you leave home without permission, the police come and find you within a half hour and make you return home. That is a much better lockdown than exists anywhere in the western world. I was in a crowded supermarket yesterday in Oakland, California. I’d be surprised if there wasn’t a single transmission in there that day. The local lake is more crowded than ever with people who are desperate to get outside. They try to keep their distance, but they’re not as distant as they would be if they’d stayed at home. Are we doing enough? We’ll have a better idea in a few weeks.

World data

All countries

I’m going to limit the data to countries that have reported at least 100 deaths. First up, I’ll just plot all of these countries. I’ll give a table of country codes at the end. Note that the data is not cumulative: each data point is deaths per million on that day. The lines are colored by how bad the situation is in that country right now: how many deaths there were there yesterday.

Countries with less than 10 deaths per million yesterday

Countries with less than 2 death per million yesterday

US data

I’m going to use the same limitations for the US state data: states with at least 100 deaths. For reference, I’m also going to include Spain and Italy.

States with fewer deaths per million than Italy yesterday

Appendix

iso3c country iso3c country
AUT Austria GBR United Kingdom
BEL Belgium IDN Indonesia
BRA Brazil IRL Ireland
CAN Canada IRN Iran
CHE Switzerland ITA Italy
CHN China KOR Korea, South
DEU Germany NLD Netherlands
DNK Denmark PHL Philippines
DZA Algeria PRT Portugal
ECU Ecuador ROU Romania
ESP Spain SWE Sweden
FRA France TUR Turkey
COVID-19 coronavirus
comments powered by Disqus