Read on Twitter
A thread on the measurement of excess mortality and uncounted COVID19 deaths. Key point: be careful inferring number of uncounted COVID19 deaths from a look at all-cause mortality data 1/n
Excess mortality is generally understood as the number of deaths (from any cause of death) we are seeing now (e.g., this week), minus or divided by the number of deaths we would have seen IN THE ABSENCE OF THE COVID19 PANDEMIC. 2/n
How do we measure this counterfactual, “the number of deaths we would have seen in the absence of COVID19”? In most cases, we simply calculate the average number of deaths per week in the past n years, i.e., when COVID19 had not emerged as a cause of death. 3/n
Daily counts we see in media come from hospital-based reports that focus on COVID deaths. Data on excess mortality come from vital statistics. That's administrative system that continuously registers ALL deaths. It also includes home deaths, long-term care facilities… 4/n
Vital statistics take time to come online b/c administrative reporting process (e.g., Germany only reports up to 3/15). At first, no information on cause of death, that comes months later. One exception is UK, beautifully visualized by @d_spiegel
https://twitter.com/d_spiegel/status/1252526609116135424?s=20 5/n
https://twitter.com/d_spiegel/status/1252526609116135424?s=20 5/n
The NY Times has a stunning look at excess mortality from vital statistics in various countries and cities out today with quotes from outstanding demographers (I provided some tips) 6/n https://www.nytimes.com/interactive/2020/04/21/world/coronavirus-missing-deaths.html
Now, what can we say about number of uncounted COVID deaths, based on this excess of all-cause mortality? It’s tempting to use excess mortality to calculate an upper bound for the number of uncounted COVID deaths. Take the following example 7/n
Imagine country with 100 deaths per week in past years. This year, hospital-based systems report 100 COVID deaths/week, and vital stats count 300 all-cause deaths in total. Excess mortality is 300 – 100 = 200 so there could be up to 100 uncounted COVID deaths (200 – 100). 8/n
That's too simple! First, IN PRESENCE OF COVID, everything else being equal, expect FEWER non-COVID deaths this year than in past years, mechanically. Why? Because COVID now kills people who would otherwise have died from non-COVID causes. COVID is a new “competing risk”. 9/n
In our imaginary country, out of the 100 non-COVID deaths we should have expected this year based on past trends, some % will in fact die of COVID. So the gap between COVID and non-COVID deaths grows, and the potential number of uncounted COVID deaths grows along. 10/n
Second, “everything else” is not equal. The population size/age composition might change, and the rate at which people die from non-COVID causes might also vary. This latter indirect effect is gloriously described in @andrewnoymer’s thread 11/n https://twitter.com/AndrewNoymer/status/1241620288825167874?s=20
In year of lockdowns, rate at which people die from accidents might come down due to reduced car traffic, same for mortality linked to air pollution etc…. On other hand, rate at which people die from other causes might increase due to lower healthcare use, increased stress 12/n
In the end, excess in all-cause mortality is a great indicator of sheer magnitude of health crisis we are in, much better than hospital-based counts of COVID cases & deaths. This is not normal or a bad flu season. 13/n
But to know how many COVID deaths went uncounted, we need to wait for vital statistics to publish cause of death data, we need to conduct linkage studies and we need to undertake much more refined modeling exercises. That will take… quite some time. 14/14
One data addendum: the near real-time monitoring system recently set up by @tomtom_m & colleagues in #SouthAfrica ( https://www.samrc.ac.za/reports/report-weekly-deaths-south-africa?bc=254) also gives counts of deaths by broad causes, and it's amazing. Below, the effects of a lockdown on deaths from accidents & injuries 15/14
