I was shocked that that the PHE report signally failed to mention the OpenSAFELY (OS) preprint from Goldacre's team in respect to deprivation, ethnicity or comorbidities.[1] This study, based on some 17.4 million primary care records and ONS mortality data, had information on all of these factors for most of the patients. Because it accessed the primary care records directly, data (where available) for all the variables within each individual could be analysed, allowing for a single multivariable Cox proportional hazards model, stratified by geographical area to allow for local variations in prevalence, looking at all causes of mortality over an 80 day period, and exploring and also quantifying associations between mortality and possible risk factors.
This study has some weaknesses, but it has the advantages of being: orders of magnitude larger than any other such epidemiological studies of Covid-19 mortality; based on a primary care (not hospital) denominator; repeatable as required; and transparent, having all the methodology on the OS website.
The authors specifically examined ethnicity as a factor in a sub-study, using the 74% of records in which ethnicity was recorded. Even after allowing for all the other co-variates (including age, BMI, smoking, deprivation, and co-morbidities), BAME ethnicity remained an independent factor, with a hazard ratio of up to 1.7. The PHE report says: '[The data they included] were also not able to include the effect of comorbidities or obesity.' Why, given that they were happy to include the OS data in the chapter on BMI (despite the study having missing BMI data for 22% of the records), did they explicitly exclude it from the other sections, where it would have given at least a preliminary indication of an effect and its magnitude, independently of the other accepted factors?
The Association of Local Medical Advisors (ALAMA) have a model devised by the Joint Occupational Health COVID-19 Group to help assess the absolute risk of workers' vulnerability to Covid-19 (as ‘Covid-years’), partly based on OS data.[2] The BMA reference this risk assessment tool on their website. I have devised a spreadsheet to explore the impact of the various risk factors, using a hybrid of both the OS dataset and the ALAMA tool.[3] This yields an estimate of the relative and absolute risk of Covid-19-related mortality. Using this, the independent impact of both ethnicity and deprivation can easily be seen to have a major impact on mortality, in addition to age.
NHS employers recommend undertaking a risk assessment for all employees. A Risk Reduction Framework for NHS Staff at risk of COVID-19 infection has been drawn up by Prof Khunti and others.[4] This identifies that one needs to assess the environmental & behavioural factors, as well as the intrinsic individual factors. Such a calculator as mentioned above makes the latter component easy, transparent & apparent in both relative and absolute terms. Although the risk framework was already published, the PHE report makes no mention of it. Given the lack of assessment of BAME doctors[5], here’s a way of making this much easier and quicker. If I can do it, why can’t PHE?
References
1. medrxiv.org/content/10.1101/2020.05.06.20092999v1.full.pdf
2. alama.org.uk/covid-19-medical-risk-assessment/
3. www.forrestmls.org/useful-links [third bullet point on page]
4. diabetestimes.co.uk/diabetes-professor-publishes-covid-19-risk-reduction-framework-document/
5. https://www.bbc.com/news/health-53021628
Reposted 16 June with correct references
Competing interests:
I am a member of a co-operative of doctors who write medico-legal reports for asylum-seekers; some of these are paid for by Legal Aid. This response is in a personal capacity.
Rapid Response:
Re: Ethnicity and covid-19
Dear Editor
Thanks for this excellent and timely editorial.
I was shocked that that the PHE report signally failed to mention the OpenSAFELY (OS) preprint from Goldacre's team in respect to deprivation, ethnicity or comorbidities.[1] This study, based on some 17.4 million primary care records and ONS mortality data, had information on all of these factors for most of the patients. Because it accessed the primary care records directly, data (where available) for all the variables within each individual could be analysed, allowing for a single multivariable Cox proportional hazards model, stratified by geographical area to allow for local variations in prevalence, looking at all causes of mortality over an 80 day period, and exploring and also quantifying associations between mortality and possible risk factors.
This study has some weaknesses, but it has the advantages of being: orders of magnitude larger than any other such epidemiological studies of Covid-19 mortality; based on a primary care (not hospital) denominator; repeatable as required; and transparent, having all the methodology on the OS website.
The authors specifically examined ethnicity as a factor in a sub-study, using the 74% of records in which ethnicity was recorded. Even after allowing for all the other co-variates (including age, BMI, smoking, deprivation, and co-morbidities), BAME ethnicity remained an independent factor, with a hazard ratio of up to 1.7. The PHE report says: '[The data they included] were also not able to include the effect of comorbidities or obesity.' Why, given that they were happy to include the OS data in the chapter on BMI (despite the study having missing BMI data for 22% of the records), did they explicitly exclude it from the other sections, where it would have given at least a preliminary indication of an effect and its magnitude, independently of the other accepted factors?
The Association of Local Medical Advisors (ALAMA) have a model devised by the Joint Occupational Health COVID-19 Group to help assess the absolute risk of workers' vulnerability to Covid-19 (as ‘Covid-years’), partly based on OS data.[2] The BMA reference this risk assessment tool on their website. I have devised a spreadsheet to explore the impact of the various risk factors, using a hybrid of both the OS dataset and the ALAMA tool.[3] This yields an estimate of the relative and absolute risk of Covid-19-related mortality. Using this, the independent impact of both ethnicity and deprivation can easily be seen to have a major impact on mortality, in addition to age.
NHS employers recommend undertaking a risk assessment for all employees. A Risk Reduction Framework for NHS Staff at risk of COVID-19 infection has been drawn up by Prof Khunti and others.[4] This identifies that one needs to assess the environmental & behavioural factors, as well as the intrinsic individual factors. Such a calculator as mentioned above makes the latter component easy, transparent & apparent in both relative and absolute terms. Although the risk framework was already published, the PHE report makes no mention of it. Given the lack of assessment of BAME doctors[5], here’s a way of making this much easier and quicker. If I can do it, why can’t PHE?
References
1. medrxiv.org/content/10.1101/2020.05.06.20092999v1.full.pdf
2. alama.org.uk/covid-19-medical-risk-assessment/
3. www.forrestmls.org/useful-links [third bullet point on page]
4. diabetestimes.co.uk/diabetes-professor-publishes-covid-19-risk-reduction-framework-document/
5. https://www.bbc.com/news/health-53021628
Reposted 16 June with correct references
Competing interests: I am a member of a co-operative of doctors who write medico-legal reports for asylum-seekers; some of these are paid for by Legal Aid. This response is in a personal capacity.