Black, Asian and Minority Ethnic groups in England are at increased risk of death from COVID-19: indirect standardisation of NHS mortality data

Background: International and UK data suggest that Black, Asian and Minority Ethnic (BAME) groups are at increased risk of infection and death from COVID-19. We aimed to explore the risk of death in minority ethnic groups in England using data reported by NHS England. Methods: We used NHS data on patients with a positive COVID-19 test who died in hospitals in England published on 28th April, with deaths by ethnicity available from 1st March 2020 up to 5pm on 21 April 2020. We undertook indirect standardisation of these data (using the whole population of England as the reference) to produce ethnic specific standardised mortality ratios (SMRs) adjusted for age and geographical region. Results: The largest total number of deaths in minority ethnic groups were Indian (492 deaths) and Black Caribbean (460 deaths) groups. Adjusting for region we found a lower risk of death for White Irish (SMR 0.52; 95%CIs 0.45-0.60) and White British ethnic groups (0.88; 95%CIs 0.86-0.0.89), but increased risk of death for Black African (3.24; 95%CIs 2.90-3.62), Black Caribbean (2.21; 95%CIs 2.02-2.41), Pakistani (3.29; 95%CIs 2.96-3.64), Bangladeshi (2.41; 95%CIs 1.98-2.91) and Indian (1.70; 95%CIs 1.56-1.85) minority ethnic groups. Conclusion: Our analysis adds to the evidence that BAME people are at increased risk of death from COVID-19 even after adjusting for geographical region, but was limited by the lack of data on deaths outside of NHS settings and ethnicity denominator data being based on the 2011 census. Despite these limitations, we believe there is an urgent need to take action to reduce the risk of death for BAME groups and better understand why some ethnic groups experience greater risk. Actions that are likely to reduce these inequities include ensuring adequate income protection, reducing occupational risks, reducing barriers in accessing healthcare and providing culturally and linguistically appropriate public health communications.


Background
There is increasing international evidence that Black, Asian and Minority Ethnic (BAME) people are at higher risk of death from COVID-19 1 . As of 26th April 2020, there were 20,732 reported COVID-19 deaths in hospital in the UK 2 , but to date, there have been no officially reported analyses of the risk of death by ethnicity 3 . An inquiry has been announced that will examine the impact of COVID-19 on BAME people 4 .
Ethnicity data are currently available in the intensive care national audit and research centre (ICNARC) reports on patients with confirmed COVID-19 that have been admitted to intensive care for at least 24 hours. On 24th April 2020 these data showed that BAME people were at higher risk of developing severe COVID-19 disease 5 . A total of 5,993 patients with confirmed COVID-19 had reported data on ethnicity and 34.2% (2,055/5,993) of these patients were from BAME groups. Analyses matched by area (ward) of residence showed differences are significant for all BAME groups but there is substantial variation by minority ethnic groups. There were 1.63 times more Black patients in critical care than expected based on the matched population (10.6% vs 6.5%). For Asian patients the differential is reduced but still significant with 1.25 times more Asian patients than expected (15.3% vs 12.2%).
Ethnicity is not recorded in death certificates in England which is an important limitation on our ability to study the differential impact of COVID-19 on mortality in different BAME groups. However, daily NHS hospital death data are provided by geographical region, age and ethnicity 6 . Adjusting for region is potentially important because in England COVID-19 has affected different parts of the country to a different extent. For example, London and the West Midlands, the two regions with the highest levels of BAME residents have had most COVID-19 cases. Using these data, we aimed to examine the risk of death from COVID-19 by BAME group and through a sensitivity analysis test whether differences between BAME groups could be explained by regional differences in the ethnic make-up of the population.

Methods
We used NHS data on patients with a positive COVID-19 test who died in hospitals in England, in separate tables by age group, region, and ethnicity. We used data published on 26th April, that included deaths by ethnicity from 1 March 2020 up to 5pm 21 April 2020 6 . Where the age group and region tables showed a different total number of deaths to the ethnicity table, we applied a scaling factor to align the totals to the ethnicity table. We assumed that decedents with unknown ethnicity had the same ethnicity structure as other decedents.
We used indirect standardisation to calculate standardised mortality ratios (SMRs) by ethnic group (where the reference group is the whole population). We first calculated age-specific mortality rates using the COVID-19 deaths data and population estimates from the UK Census 2011. All ages were used, and age ranges included were 0-19; 20-39;40-59; 60-79 and 80+. We then calculated an expected number of deaths for each ethnic group by applying these mortality rates to population estimates by both ethnic group and age, also from the UK Census 2011. We calculated the SMR as the observed deaths divided by the expected deaths. We assumed that deaths occurred over the same time period for all ethnic groups, and used the population point estimate as the denominator for simplicity. We then conducted a sensitivity analysis to account for regional differences in the ethnicity of the population. The number of COVID-19 deaths by age and region was not available, and we assumed that the proportion of deaths in each age group was the same across regions. We calculated age-and region-specific mortality rates using this assumption, and calculated an expected number of deaths by applying these rates to population estimates by ethnic group, age, and region. The data included seven regions: London, South East, South West, Midlands, East of England, North West, and North East & Yorkshire and the Humber. We then estimated SMRs adjusted for region by dividing the observed by the expected deaths. We calculated 95% confidence intervals using the exact Poisson method. All analyses were conducted using R version 3.6.2. Data and code required for replication are provided as Underlying and Extended data 7 .

Results
A total of 16,272 deaths were observed over the study period. Ethnicity was missing for 9.4% (1,537/16,272) of NHS England hospital deaths. The largest total number of deaths in minority ethnic groups were Indian (492 deaths) and Black Caribbean (460 deaths) people. In comparison to the whole population, SMRs in the unadjusted analysis were reduced for White British and White Irish groups, but were increased for all BAME groups.

Discussion
Our analyses showed that several BAME groups have a higher risk of death from COVID-19 and that regional differences in ethnicity explains some but not all of the differences between

Amendments from Version 1
We have updated the manuscript in relation to the helpful comments from the reviewers, particularly noting some additional limitations of our analysis and data. We have tried to improve the image quality of Figure 1 and we have added p-values to Table 1 as requested in the review by Oliver Razum and Odile Sauzet.
Any further responses from the reviewers can be found at the end of the article REVISED ethnic groups. After accounting for geographical region SMRs reduced, but there remained large differences in SMRs between ethnic groups -White British and White Other have lower SMRs, but Bangladeshi, Pakistani, Indian, Black African and Black Caribbean ethnic groups all have substantially increased SMRs.
There are limitations to these data relating to the reporting of COVID-19 deaths. The NHS data we have used are currently only available in broad age groups, and are not broken down by both region and age, which meant we had to assume there were no differences in age structure of deaths across regions within these age bands. Data are also not disaggregated by sex and social deprivation and therefore we were unable to explore the effect these would have on our adjusted SMR estimates. There is increasing evidence that men are more likely to die from COVID-19 and therefore our lack of disaggregation by sex could account for some of the remaining differences in SMRs we see between ethnic groups, particularly as those occupations found to be at higher risk include greater numbers of men working in them. Publication of COVID-19 death data by age, region, gender, social deprivation, and ethnicity would improve these adjustments further. The data we used only include people who died in hospital. ONS data from 28 April suggest that 77.4% (14,796 deaths) of all deaths in England and Wales occurred in hospital with 3,096 deaths occurring in care homes 8 and the SMRs may be biased if deaths that occur in hospitals differ from those occurring in the community by ethnic group, which might arise due to, for example, differences in use of residential care homes 9 . Deaths in residential care homes are likely to include a larger number of White British people 10 which could lead to an under-estimation of the SMR in this group within our estimates. Our analysis was based on the 2011 census data and therefore will not reflect recent changes in the age, ethnic and region across England in the last nine years. Our use of census data from 2011 is likely to result in overestimation of mortality ratios in minority ethnic groups that have grown the fastest during this time period. We found raised  SMRs in several BAME groups including Asian Other, Black Other, Mixed Other and White Other and further analysis should be undertaken to examine whether there are particular groups at risk within these broad groups to ensure we can better understand their increased mortality risk.
Our analysis is consistent with Intensive Care National Audit and Research Centre (ICNARC) data which suggests that Black ethnic groups are substantially over-represented amongst critical care patients, and that BAME groups in critical care are generally more likely to require ventilation and therefore more likely to die. Further analyses of the ICNARC data are required to assess the extent to which these associations are due to differences in age, comorbidity and socioeconomic status. A recent analysis of COVID-19 deaths in health and social care workers was undertaken using data from mainstream and social media reports. This analysis suggested that BAME deaths in nursing and support staff accounted for 64% of deaths, and 95% in medical staff, whilst these groups accounted for 20% and 44% percent of all staff 11 . This analysis of deaths in health and social workers, however, did not adjust for regional differences in the proportion of NHS staff coming from BAME groups.
Adjusting for geographic region reduced the high SMRs for BAME groups as shown in Figure 1. Several other factors, some of which will be associated with geographic region, may further explain this increased risk. There is increasing evidence from ONS and Public Health England on the role of occupation and socioeconomic deprivation in relation to risk of COVID-19 and the increased risk of infection and poor outcomes found in BAME communities [12][13][14][15] . Occupation is also likely to play an important role in terms of increased risk of infection as BAME people are more likely to have occupations that involve greater social mixing and less ability to work from home. For example, Black groups are overrepresented in caring and leisure industries; Pakistani and Bangladeshi groups are overrepresented in sales and consumer service occupations; and Black groups in public administration, education and health 16 . BAME groups are more likely to have a low income, be in zero hours contracts and non salaried jobs than white ethnic groups. This may make it harder to comply with social distancing restrictions that prevent people from working and those who are self-employed or working in the gig economy will have their earnings stop unless they sign up to a government scheme. There may be barriers to this and some migrants, for example, may not want to make themselves known to the authorities. Ethnicity is socially constructed and correlates poorly with biology. Biological differences are therefore highly unlikely to underpin these inequalities 17 .
Living in overcrowded housing likely increases transmission risk, and BAME households were more likely to be overcrowded than White British households in recent analysis by ONS 18 . This is true even when restricting analyses to those living in poverty, where BAME groups living in poverty are more likely to be in overcrowded conditions than white groups living in poverty 19 . Increased levels of pre-existing medical conditions such as diabetes, hypertension and heart disease are known to increase the risk of severe COVID-19 disease 1 and these are also increased in some ethnic groups. Finally, differences in risk factors such as obesity, may also be relevant. Research to disentangle these potential pathways appears highly limited, with only one study having been conducted, to our knowledge 20 . This was based on laboratory-confirmed diagnoses using the UK Biobank study and suggested that socioeconomic differences might make an important contribution, but differences in pre-existing health and risk factors appeared less important. However, this study was based on a non-representative sample and relied on routine testing which is likely subject to substantial ascertainment bias.
Ethnicity is not recorded in death certificates in England, which is a major limitation in our ability to study the differential impact of COVID-19 on mortality in different ethnic groups. However, this has been achieved in Scotland and the COVID-19 pandemic highlights the potential utility of introducing it in England 21 . Further analysis of deaths for BAME people will require urgent linkage to other records that contain ethnicity information such as NHS hospital episode statistics and primary care electronic health records. A key unanswered question is to understand why mortality risks differ between ethnic groups. This may arise from an increased risk of developing infection, worse prognosis or care once infection has occurred or a combination of the above 22 . While it is important to conduct and report such analyses rapidly, this must not delay immediate action to begin to mitigate these extreme inequities.
We believe there are several important and urgent public health actions to be taken to address the high mortality rates in BAME groups described in our analyses. First, some BAME groups face barriers in accessing high quality healthcare. The NHS must remove these barriers working with minority ethnic people to understand the issues. For example, some people in BAME communities will also be international migrants and Public Health England recently reported the increased risk of death from COVID-19 in this group 15 . This analysis showed that the largest relative increase in death from COVID-19 was for people born in Central and Western Africa, the Caribbean, South East Asia, the Middle East and South and Eastern Africa. Some groups of international migrants in the UK avoid the use of the NHS because of the current NHS charging regime for migrants or through fear of their data being shared with the Home Office for immigration enforcement purposes 23 . Limited healthcare entitlement results in untreated conditions, poorly managed chronic conditions and deterrence from healthcare for migrants is well documented, rendering a context of distrust and fear 24 . Whilst migrants diagnosed with COVID-19 are exempt from healthcare charges, not all migrants will be aware of these exemptions and the exemption first requires a diagnosis. Some migrants may fear the charge being imposed through a lack of diagnosis due to limited testing opportunities. We therefore call for the removal of all NHS charges during this public health emergency to ensure that no migrant or individual from a BAME group delays seeking healthcare and risks death through fear of being charged for their NHS care. Second, we must ensure that linguistically and culturally appropriate public health communication and engagement is being provided and appropriately targeted at those populations at greatest risk. This needs to be developed with affected communities and tailored to specific challenges including addressing culturally specific disinformation and, for example, addressing the difficulties of preventing transmission in overcrowded households or of shielding vulnerable people in multigenerational households. Third, we must take urgent action to reduce the risk of SARS-CoV-2 infection in BAME populations. For example, BAME groups are more likely to work in care settings such as nursing homes, where adequate PPE to prevent infection is vital. BAME groups are also more likely than others to be in key worker occupational groups who have high levels of exposure to the general public and therefore high risk of infection. The effectiveness of personal protective equipment in preventing infection outside health and social care settings remains uncertain, however, there are a range of other measures that are likely to reduce infection risk. These include: ensuring that workplaces are not overcrowded so that staff can maintain social distancing at work; providing distancing measures and physical barriers to reduce exposure to droplets from the members of the general public (e.g. perspex screens at supermarket counters); ready availability of handwashing materials at the workplace and access to testing and workplace contact tracing. Fourth, there is a risk that some ethnic minority groups might not only experience greater risks from COVID-19 itself, but also greater adverse consequences of the extensive social distancing measures in place 25 . There is a need for adequate income protection to ensure low paid, non-salaried and zero-hours contract workers can afford to follow isolation and "stay at home" recommendations.
The unacceptable differences in COVID mortality between white and BAME groups demand immediate action. They are an extreme example of the long-standing inequities affecting BAME groups in our society. As we emerge from the COVID-19 pandemic we must ensure that these unfair and avoidable disparities are addressed. Governments in the UK, and elsewhere, must consider how to best protect minority ethnic groups from experiencing further disadvantage and indirect health harms during the recovery process. The public health response to COVID-19 must be equitable and urgent if it is to address the unacceptable ethnic disparities our analyses show.

Open Peer Review
Geography and Planning, University of Liverpool, Liverpool, UK Public and media attention has increasingly centred on the apparent disproportionate impact of COVID-19 on BAME groups. This paper adds empirical evidence to these discussions, going some way in highlighting that COVID-19 is far from a leveller or equalizer. The authors use publicly available daily NHS hospital death data provided by geographical region, age and ethnicity to create standardised mortality ratios (SMRs) by ethnic group which are first adjusted by age and then also by region (having looked at the data, the age groups are quite broad which may be worth stating). The indirect method of standardisation is used which is appropriate given the possibility of small numbers in some cases. Results are presented in a table and in a figure, illustrating that Asian (bar Chinese), Black, Mixed Other, and White Other ethnic groups all have a heightened risk of mortality from COVID-19 after adjusting for age and region. The authors go on to speculate as to possible reasons for the differences between ethnic groups, ranging from differences in experiences of poverty between regions and ethnic groups, differences in occupation and the resulting risk of exposure, housing situation, and differences in pre-existing medical conditions. The authors make a number of recommendations in terms of public health interventions during this crisis, as well as noting limitations. Though well written, clearly presented, and informative, there are a couple of factors that would strengthen this paper, particularly given the unavoidable limitations of the data used.
In the recommendations for intervention section, much is made of the barriers to healthcare access faced by some BAME groups. More evidence to support this would strengthen. Further, though concerns are rightly raised as to the impact of the NHS charging regime for migrants and general discussion of migrants and differential healthcare needs/access requirements, this does not feel well supported by the data used. These data are simply 'ethnicity', rather than migrant status, country of birth or length of residence in the UK -all of which matter for this theme of discussion. More reflection of this needed.

1.
Relatedly, much is made of how differences in occupation and accordingly, socioeconomic position, matter for risk of exposure and the wider impact of COVID-19 policy responses. These are relevant and important debates, but are tangential to what is shown in the data due to the limitations of what is recorded. More set up/evidence as to why these will vary between ethnic groups, and the importance of region (which is a rather crude geographical scale and more could be made of this), would help counter this limitation.

2.
Though not possible to produce SMRs by sex, more reflection on that is important given wider differences in COVID-19 mortality profiles between sexes, and the relationship with some of the discussion on occupation/socioeconomic factors etc.

3.
Finally (minor point): as an open piece of research this is very useful. Yet, when looking through the link provided for the data I could only see deaths broken down by region, broad age group, and NHS Trust. More specifics as to the location of the ethnicity data would be welcome (NB, the authors provide the R code and underlying data, which is great).

Is the work clearly and accurately presented and does it cite the current literature? Yes
Is the study design appropriate and is the work technically sound?

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Partly

Are the conclusions drawn adequately supported by the results? Partly
Competing Interests: No competing interests were disclosed.
Reviewer Expertise: Population Geography, Health Geography, Social and Spatial Inequality, Ethnicity.

I confirm that I have read this submission and believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
Author Response 15 Jun 2020 Robert Aldridge, Institute of Health Informatics, UCL, London, UK Thanks very much for this helpful review -we appreciate the time you have taken to carefully read and provide constructive feedback on the article. We have tried to address all the points you have raised and below we include your comments alongside in our responses in bold italics. We have numbered each of our responses in order that we can cross-reference them across all of the reviews to this article. ****** Public and media attention has increasingly centred on the apparent disproportionate impact of COVID-19 on BAME groups. This paper adds empirical evidence to these discussions, going some way in highlighting that COVID-19 is far from a leveller or equalizer. The authors use publicly available daily NHS hospital death data provided by geographical region, age and ethnicity to create standardised mortality ratios (SMRs) by ethnic group which are first adjusted by age and then also by region (having looked at the data, the age groups are quite broad which may be worth stating). The indirect method of standardisation is used which is appropriate given the possibility of small numbers in some cases. Results are presented in a table and in a figure, illustrating that Asian (bar Chinese), Black, Mixed Other, and White Other ethnic groups all have a heightened risk of mortality from COVID-19 after adjusting for age and region. The authors go on to speculate as to possible reasons for the differences between ethnic groups, ranging from differences in experiences of poverty between regions and ethnic groups, differences in occupation and the resulting risk of exposure, housing situation, and differences in pre-existing medical conditions. The authors make a number of recommendations in terms of public health interventions during this crisis, as well as noting limitations. Though well written, clearly presented, and informative, there are a couple of factors that would strengthen this paper, particularly given the unavoidable limitations of the data used.
In the recommendations for intervention section, much is made of the barriers to healthcare access faced by some BAME groups. More evidence to support this would strengthen. Further, though concerns are rightly raised as to the impact of the NHS charging regime for migrants and general discussion of migrants and differential healthcare needs/access requirements, this does not feel well supported by the data used. These data are simply 'ethnicity', rather than migrant status, country of birth or length of residence in the UK -all of which matter for this theme of discussion. More reflection of this needed.

Author response 9: We agree that not all people in minority ethnic groups are migrants, but there is a large overlap between these groups and we feel that the particular issues faced by migrants are important to highlight along with potential important measures to reduce their increased risk of death from COVID-19. Public Health England has produced additional evidence on this point since our publication on this increased risk for migrants and we have now added the following sentence to qualify the points we make on migration in order to address this important issue raised by the reviewer: "For example, some people in BAME communities will also be international migrants and Public Health England recently reported the increased risk of death from COVID-19 in this group. This analysis showed that the largest relative increase in death from COVID-19 was for people born in Central and Western Africa, the Caribbean, South East Asia, the Middle East and South and Eastern Africa."
Relatedly, much is made of how differences in occupation and accordingly, socioeconomic position, matter for risk of exposure and the wider impact of COVID-19 policy responses. These are relevant and important debates, but are tangential to what is shown in the data due to the limitations of what is recorded. More set up/evidence as to why these will vary between ethnic groups, and the importance of region (which is a rather crude geographical scale and more could be made of this), would help counter this limitation.

Author response 10: After our study was published ONS and Public Health England have completed several analyses that support the comments we make in this discussion. We have edited out some of our previous comments on these issues and replaced it with the following sentences in the discussion which provide direct evidence on the associations between occupation, socioeconomic position and COVID: "There is increasing evidence from ONS and Public Health England on the role of occupation and socio-economic deprivation in relation to risk of COVID-19 and the increased risk of infection and poor outcomes found in BAME communities"
Though not possible to produce SMRs by sex, more reflection on that is important given wider differences in COVID-19 mortality profiles between sexes, and the relationship with Limitations of the data set are explained. The authors could have performed a sensitivity analysis with different assumptions regarding the distribution of the various ethnicities among those persons in whom this information was missing.
The Discussion section of the manuscript is long as compared to the Results section, mainly due to problems of interpreting the findings. I see two issues here: Firstly, "ethnicity" is socially constructed; there is no plausible biological correlate (see, for example, Saini, A. (2020). Stereotype threat. The Lancet, 395(10237), 1604-1605 1 ). This should be explained in order to avoid speculation of underlying "genetic" causes. Second, having said that, probable underlying causes are socioeconomic disadvantages and possibly differentials in access and quality of care. The available data, however, are not really informative in this respect. In consequence, the second-butlast paragraph (starting with "We believe…") should be shorter and less speculative.
Minor technical issue: Some journals require reporting of confidence intervals and p-values (the latter are not reported in Table 1).

Author response 7: We agree with the reviewers about the socially constructed nature of ethnicity and in our original submission we deliberately chose not to discuss biology or genetics because of this. In relation to this point raised by the review, we have added the following to sentence to the discussion to elaborate further in agreement with your suggestion:
"Ethnicity is socially constructed and correlates poorly with biology. Biological differences are therefore highly unlikely to underpin these inequalities."

We believe the paragraph referred to by the authors (starting with "We believe…") provides important and urgent public health actions that should be taken as a precautionary principle to reduce the increased mortality from COVID in BAME communities, and as such we feel it is important to restate them in this discussion. We also note the recent objections to other reports describing the inequalities in health for BAME communities that did not include recommendations such as ours on this precautionary note.
Minor technical issue: Some journals require reporting of confidence intervals and p-values (the latter are not reported in Table 1). First, given the number of caveats in the study, the authors should qualify their 'conclusion' part of the abstract to make clear that there are limitations in the article for those who, say, only read the abstract (of which there are more than we would like to believe).

Author response 8: As suggested, we have added p-values to our updated
Second, the authors tell us that biological sex is not available but they do not go on to elaborate upon the possible consequences of its absence. Some (short) discussion around this would be welcome in light of evidence that men are more susceptible to COVID-19.
Third, the study lacks adjustment for confounding factors such as SEP (socioeconomic position). This is simply because this information is not available with the data at hand. To offset this, the authors provide a welcome discussion of these SEP factors and how they likely vary across the different ethnic groups in England. I suspect that the sizable role that region plays in attenuating, but not eliminating, the excess risk of death in some ethnic groups is because this variable captures some of the excess that would otherwise be explained by these SEP factors. In any case, given the size of the some of the ethnic-specific SMRs, it is unlikely that further adjusting for SEP factors would fully equalize the substantial differences observed in the SMRs of different ethnic groups (although its presence would be welcome).
Finally, I recommend that the assumptions made in the methods section are revisited more strongly in the discussion section. Notably, the use of denominator data that is nearly a decade old for rapidly expanding ethnic groups (is this likely to results in some over-estimation of death in the fastest-growing groups?), more explicit information on exactly how broad the age groups are (I don't think this is here, but I may have missed it), what the age range is (I don't think I see this anywhere either -is it simply all ages?), clarifying what happened to decedents with missing ethnicity (were they excluded? Combined with Other? Did you calculate the mortality of the missing group to see if the missingness was selective?), and the assumption regarding the proportion of deaths in age groups over regions (could you perhaps have used another cause of death, such as influenza, as a base instead?).
As a final note, I wonder whether Figure 1 would be more easy to interpret if rotated (and even presented as points as opposed to bars -although this is more a personal preference). Currently, it is a little hard to read the labels and pick out specific ethnic groups.
Final reflections: this is an important new study (with, admittedly, several limitations) that showcases some alarming findings that demand further research.

Is the study design appropriate and is the work technically sound? Partly
Are sufficient details of methods and analysis provided to allow replication by others?

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility? Yes Are the conclusions drawn adequately supported by the results? Yes Thanks very much for this helpful review -we appreciate the time you have taken to carefully read and provide constructive feedback on the article. We have tried to address all the points you have raised and below we include your comments alongside in our responses in bold italics. We have numbered each of our responses in order that we can cross-reference them across all of the reviews to this article. ****** First, given the number of caveats in the study, the authors should qualify their 'conclusion' part of the abstract to make clear that there are limitations in the article for those who, say, only read the abstract (of which there are more than we would like to believe).

Author response 1: We agree it would be useful to add a summary of the limitations to the abstract and have added the following to the conclusion: "Our results were limited by the broad age groups, lack of data on deaths outside of NHS settings and denominator data on ethnicity being based on the 2011 census."
Second, the authors tell us that biological sex is not available but they do not go on to elaborate upon the possible consequences of its absence. Some (short) discussion around this would be welcome in light of evidence that men are more susceptible to COVID-19.

Author response 2: To address this point we have added the following to the discussion, which also adds in comments made by the helpful review from Frances Darlington-Pollock: "There is increasing evidence that men are more likely to die from COVID-19 and therefore our lack of disaggregation by sex could account for some of the remaining differences in SMRs we see between ethnic groups, particularly as those occupations found to be at higher risk include greater numbers of men working in them."
Third, the study lacks adjustment for confounding factors such as SEP (socioeconomic position). This is simply because this information is not available with the data at hand. To offset this, the authors provide a welcome discussion of these SEP factors and how they likely vary across the different ethnic groups in England. I suspect that the sizable role that region plays in attenuating, but not eliminating, the excess risk of death in some ethnic groups is because this variable captures some of the excess that would otherwise be explained by these SEP factors. In any case, given the size of the some of the ethnic-specific SMRs, it is unlikely that further adjusting for SEP factors would fully equalize the substantial differences observed in the SMRs of different ethnic groups (although its presence would be welcome).

Author response 3: We agree with the reviewer on this point and would like to explore it further when data enabling us to do are available.
Finally, I recommend that the assumptions made in the methods section are revisited more strongly in the discussion section. Notably, the use of denominator data that is nearly a decade old for rapidly expanding ethnic groups (is this likely to results in some overestimation of death in the fastest-growing groups?), more explicit information on exactly how broad the age groups are (I don't think this is here, but I may have missed it), what the age range is (I don't think I see this anywhere either -is it simply all ages?), clarifying what happened to decedents with missing ethnicity (were they excluded? Combined with Other? Did you calculate the mortality of the missing group to see if the missingness was selective?), and the assumption regarding the proportion of deaths in age groups over regions (could you perhaps have used another cause of death, such as influenza, as a base instead?).

Author response 4: We have clarified in the methods section that:
"All ages were used, and age ranges included were 0-19; 20-39;40-59; 60-79 and 80+." We assumed that decedents with unknown ethnicity had the same ethnicity structure as other decedents.
In the discussion we have added the following which we hope address other points raised:

"Our use of census data from 2011 is likely to result in over-estimation of mortality ratios in minority ethnic groups that have grown the fastest during this time period."
Given the differing seasonal variation in influenza, and substantially different infection fatality ratios we are not sure that using this as a base instead would be helpful.
As a final note, I wonder whether Figure 1 would be more easy to interpret if rotated (and