So PHE has some money to spend on researching how Covid-19 affects people of all genders (this is code for 'men', for people that work in la-la land), and also BME.
See here
twitter.com/DHSCgovuk/status/1257332511912460290
www.gov.uk/government/news/review-into-factors-impacting-health-outcomes-from-covid-19
and here
www.nihr.ac.uk/documents/highlight-notice-covid-19-and-ethnicity/24657
I am slightly perplexed at this because:
- in fact ethnic minority populations are less likely to die of covid-19 because they have the advantage of being younger, which is very important
- it's clear in general that poor, urban areas, which are disproportionately BME (but still mostly white) are hit harder than richer and rural areas (which are disproportionately white), but this largely implies a correlation between poverty and worse healthcare outcomes, and doesn't particularly speak to an ethnic issue.
In particular, Indian (much more likely), Chinese (marginally) and white Other (ditto), Asian other (ditto) are more likely to be in £1k/week+ income group than white British people. In addition, 30% of white British earned less than £400/week, which is more than all individual Asian groups, and only black was worse off.
When I investigated this a week or so ago, I found clear evidence for black people being overrepresented in deaths, but not for Asians. Given that black people are the poorest group, this is not inherently surprising.
What I don't quite understand is why one would jump directly to the 'race' angle rather than more obvious poverty. It is not obviously helpful to poor white people dying of covid-19 to be told that the whole population of whites are less likely to die than the whole population of blacks.
Regarding how one would produce this evidence, I would note:
- the NHS dataset is truncated, as I mentioned previously
- very few very old people are of BME backgrounds. These people will die at home or in care homes of covid-induced 'old age'without covid-19 testing. This is seen in the thousands of excess deaths
- we do not have ethnic data on death certificates.
- we should.
- the NHS dataset is disproportionately young and therefore disproportionately BME. It is less than half of total excess deaths.
- We know the ethnicity of around 40% of deaths (90% of the 45% of deaths in the NHS). We therefore do not know the other 60%.
- Given that we do not know the other 60% we can't very easily study the ethnicity of the whole.
- We can probably examine care home deaths by local authority, and then compare the BMEness of those local authorities with the country as a whole.
- If there is little difference, it's probably reasonable to assume that the care home deaths are 97% white, as care home deaths are generally.
10. We should not use 'covid-19 deaths' as our count of care home deaths, because that would be nonsense, as clearly they are not held to the same standard of testing as NHS deaths.
11. We must count EXCESS deaths.
12. We therefore probably conclude that the tens of thousands of excess deaths are 97% white.
13. We then have the issue of counting the deaths at home. This is a bit difficult in that if we believe that certain BME groups are more likely to care for their elderly relatives and less likely to put them in homes, it follows that relatively more elderly BME people will die at home and relatively fewer in hospital, albeit relative to the already small number of very old BME people.
14. We come back to the problem that if we assume that the care home deaths are 97% white, then if the elderly population is less white than that, then it would follow that the 'at home' deaths are less white than the elderly population as a whole.
15. If we therefore chuck out what we did in #12, and say that people are dying in home and in care homes in proportion to the ethnicity and age profiles of those areas, then we will have:
* 40% known deaths in NHS hospitals, essentially all the deaths up to 60, with most 80+ excluded, which we may decide to divide by 0.9 and assume the 'unknown' 10% are of the same characteristics (which doesn't necessarily follow, if 'unknown' varies by area, and ethnicity data is less likely to be collected in very white areas), to give us 45%
* 55% which we are simply assuming died in proportion to the underlying population, because we are finding it difficult to work out the ethnicity of the home/care home deaths - which might not be accurate.
16. The problem with this is that it's likely that our hypothetical extended ethnic family granny has a significantly lower risk of covid-19 death than a hypothetical white granny in a care home in the same area.
The alternative, I would suggest, is:
- find the number of deaths by age and sex, aged under 60 (to exclude the large excess aka 'unlabelled' deaths in older groups, though there are already some between 50 and 60), and derive a risk function for each age 0 to 60 (i.e. a single number for aged 40, male, 35, female, etc.), so that you have for example 6% of deaths were males aged 60 dying, 5.8% males aged 59, etc.
- find the number of deaths under 60 by locality (the ONS publish these data for all ages, but we'd need a dataset for age up to 60).
- Find the population of each age, sex and ethnicity in the same local authority (this is published), as a % of the total population aged 60. E.g., if 5% of people aged under 60 in a LA are white males age 60, and there were 100 deaths in that LA, then we'd expect 5% 6% 100 = 0.3 deaths in that LA to be of white males aged 60. Equally we'd expect (as our null hypothesis), that if 5% of the population was Indian males aged 60, the same 0.3 deaths
- Sum this by ethnicity for all LAs and ages, and you will get a total of deaths that is identical to the number of deaths for the population across each ethnicity, and which you can compare to the NHS dataset by ethnicity, so that for example if the NHS dataset shows 200 black people dying, but your model shows only 100 black people should have died based on age + location, then there's 2x more deaths than expected.