You asked
So in summary, regarding how ethnicity COVID-19 death data is calculated:
You used 2011 census and extrapolated the ethnicity breakdown to 2020 using linear regression?
You matched exactly the same Census 'persons' to NHS numbers?
And then used the matched census/NHS number to the cause of death on certificates?
But the deaths outside of hospital aren't linked to the NHS hospital admission, how did you account for this anomaly?
Please tell me what the ethnic breakdown is above 70.
We said
Thank you for your enquiry.
We used logistic regression for the publication to estimate risk of death, but ethnicity was not extrapolated in anyway. The ethnic group remained the same as it was recorded as at Census and presumed not to have changed (as this is a characteristic that generally remains constant). As we only account for exits from the population (deaths and weighting for migration outwards) there is no need to impute or estimate an individual's ethnicity. The regression was used to assess the differences in risk of dying from COVID-19 between ethnic groups, whilst also attempting to take into account a variety of geographical, socio-demographic, housing and self-reported health measures.
We are confident in the quality of our linkage between the 2011 Census data and Patient Register (PR) for the years 2011 to 2013 to acquire NHS number and Death Registrations since Census, because of the large amount of quality assurance undertaken on this linkage. That does not mean however, that false positives (links made that aren't actually correct) do not occur, however we are confident that these are minimal and do not have a disproportionate effect on specific characteristics of individuals, e.g. ethnicity or the characteristics used to adjust the models.
The cause of death was picked up from the death registration. It is not necessary for the deaths outside hospital to be linked to a hospital record. The NHS number was obtained from the patient register link, as a result of the linkage.
We currently do not use any information on hospital admission in our work. Death Registration data covers all deaths that happen in England and Wales, no matter the location. The Patient Register (PR) relies on someone being registered at a GP practice to pick up the NHS number. In the cases where we have assigned a PR NHS number to a Census record, the linkage to the deaths data was a simple NHS number to NHS number join. Where there was not a PR NHS number, a more complex series of deterministic/probabilistic linkage was done to link the deaths directly to Census. This second circumstance is also done when a death record does not have an NHS number (sometimes this can take some time).
Unfortunately, we would be unable to provide you with COVID-19 mortality statistics for over 70s specifically. In order to fulfil this request, we would need use a high level of statistical skill and judgement in order to create bespoke analysis. Under the Freedom of Information Act 2000, Public Authorities are not obligated to create information in order to respond to requests. We therefore consider this to be information not held.
However, we can provide statistics for those who are 65 and above, which may be of interest.
The counts of COVID-19 deaths by ethnic group for over 65s can be accessed via the following link: Counts and ratios of coronavirus-related deaths by ethnic group, England and Wales.
We can also provide a more recent publication, encompassing a larger time frame and with accompanying Age Standardised Mortality Rates for those at 65 and above. These can be found via the following link: Coronavirus (COVID-19) related mortality by religion, ethnicity and disability: England and Wales, 02 March 2020 to 15 May 2020.
If you have any further queries regarding your request, please contact Health.Data@ons.gov.uk.