The coronavirus (COVID-19) pandemic has brought ethnic inequalities in health to the attention of the public. These inequalities are long standing and deeply rooted. Many UK studies have shown that people from black, Asian and minority ethnic groups are at higher risk of a number of diseases. A series of reports now provides clear evidence that these same groups are also at higher risk of dying from COVID-19. One recent example is the Public Health England report on disparities in the risk and outcomes of COVID-19. This shows higher age-standardised mortality rates for people from black, Asian and minority ethnic groups compared to those of white British ethnicity. The risk of testing positive for COVID-19 is also higher for NHS staff from black and Asian backgrounds. This evidence raises fundamental questions, including what is driving these differences.

Research into ethnicity and COVID-19 is important if it can improve the health of people from black, Asian and minority ethnic groups and reduce inequalities in mortality and healthy life expectancy. Unfortunately, much research is unclear about the questions being investigated, the strengths and limitations of the data used, why data have been analysed in a certain way and how the findings should be interpreted. There has sometimes been a limited view of inequality, focusing on ethnicity without acknowledging how other factors – such as employment, occupation, socioeconomic deprivation or other dimensions of disadvantage – are patterned by ethnicity and may be part of the causal chain linking ethnicity to risk of exposure and illness from COVID-19.

Ethnicity has also been considered as a fixed, immutable characteristic. In fact, as many studies have shown, there is no genetic basis to commonly used broad categories of ‘race’ and ethnicity – and a person’s ethnicity will heavily depend on other external factors such as place, time and culture. There is a danger that studies looking singularly at ‘ethnicity’ and not taking broader factors that impact health into account might contribute to perpetuating inequalities, and that policy responses may take too narrow a view about possible solutions.

To help readers critically appraise the quality of emerging research findings we identify five key questions they should ask.

To understand what drives inequalities between ethnic groups, we need to be clear about what we mean by ethnicity. Whereas classification of age or the severity of a health condition may be relatively straightforward to define, classifying ethnicity is complex.

Ethnicity classifications used by data providers such as the Office for National Statistics (ONS) or as part of routine health care in the NHS are often based on a combination of attributes, such as country of birth, skin colour, culture and religion. The categories reflect how particular groups of people have been racialised – that is, how their ethnic identity has been shaped by current and historical processes and public policies rather than how a group would identify itself. These have often negatively affected people from ethnic minorities and often have deep-rooted inequalities at their centre. The meaning of a category may also depend on place and time. To take one example, the experience of a white Irish person in England is likely to be different now than it was in the 1980s because of the social and political context at that time. Broad ethnic categories are socially constructed and do not reflect genetic differences.

Because the ethnicity categories used in research are socially constructed, health differences between ethnic groups can only be understood and tackled if we consider that social construct in research– how political and historical contexts shape the lives of people in particular ethnic groups. This context includes the impact of wider determinants of health – for example access to and experience of education, housing, and employment. And underneath all of these is the experience of discrimination – so understanding how structural, institutional and interpersonal forms of racism together shape the experiences of minority ethnic groups is important to consider in research (Box 1).

There is already research evidence linking these forms of racism to poorer outcomes for people from minority ethnic backgrounds. Compared to the white British majority, many minority ethnic groups face discrimination in employment and earnings, the quality of their housing, access to neighbourhood amenities and treatment in the criminal justice system. The fact that disadvantage is experienced by many minority ethnic groups across not just one but many of these domains, points to what has been described as ‘entrenched structural and institutional racism’. Of course, studies cannot collect measurable quantitative data to model this directly and explicitly, as they cannot for other forms of discrimination. However, the interpretation of findings from studies, and ideally the methods used in analysis, should be informed by all the available relevant evidence. This includes evidence of racism and evidence from people with lived experience of being from a minority ethnic background.

If it is accepted that ethnicity classifications are crude and do not capture groupings that are natural or fixed, and that we need to consider existing evidence on the social and economic context (structures and processes) that perpetuate ethnic inequalities when interpreting study findings, what else do we need to look for in the emerging evidence on COVID-19?

This is a crucial question, because the prior assumptions researchers make about what could be driving differences between broadly classified ethnic population groups have a fundamental impact on the methods and data used in their research. Good studies should set out and justify hypotheses, the data used and methods. Several studies have used available data to test the contribution of social, economic and health factors in explaining differences in COVID-19 risk of exposure and outcomes between ethnic groups. Alternative explanations, such as cultural or biological factors, have not yet been tested with observed data.

The evidence shows that social and economic circumstances are important factors in exposure to COVID-19 and outcomes, and help to explain much of the differences between ethnic groups. An ONS report, for example, described differences between ethnic groups after controlling for region, population density, socioeconomic deprivation in the local area and in the household, household composition, housing tenure, occupational class of the head of household and educational attainment. This report showed that differences between ethnic groups were about half as big after making these adjustments.

Underlying health conditions also explain some of the differences in outcomes from COVID-19 between ethnic groups. Analysis from the OpenSAFELY project, for example, adjusted for 14 existing health conditions known or thought to be linked to COVID-19 risk and a measure of socioeconomic deprivation. Socioeconomic disadvantage is strongly associated with increased risk of long-term health conditions that put people at high risk of serious complications from COVID-19. The high rates of existing health conditions seen in many minority ethnic groups reflect long-term exposure to social and economic disadvantage. The OpenSAFELY analysis showed that differences between ethnic groups were smaller after adjustment for these existing health conditions and deprivation, although significant differences still existed after adjustment.

Cultural factors, such as living in a multi-generational household, have been put forward as another explanation, though there is currently no empirical evidence to support this. Another broad explanation put forward to explain remaining variations between ethnic groups is that these are due to inherent biological differences in populations, such as genetics and physiological characteristics. There is no evidence to date that demonstrates specifically the genetic mechanisms of the disease, although there is evidence that certain populations with a higher prevalence of some chronic diseases, such as diabetes and hypertension in South Asian groups, are more at risk. Again, unless the mechanisms at work are explicitly hypothesised and tested using data, speculation on which biological mechanisms underly differences in COVID-19 risk between broadly classified ethnic populations is unwarranted. Indeed, unfounded speculation relating to the mechanisms of the disease can be deeply damaging to communities already experiencing social and health disadvantage.

What many studies have shown is that adjustment for multiple social and economic factors and existing conditions contributes to explaining a substantial portion (although not all) of the differences in the risk and outcomes experienced by people from the white British majority group compared to people from minority ethnic groups. Studies that focus discussion on untested, individual explanations for the increased risk of COVID-19 in people from black, Asian and minority ethnic groups – without giving weight to social and economic factors – may ultimately fail to identify the key risk factors for the worse outcomes, and might even do more harm than good.

Many of these factors, however, are interrelated and the presence of one factor could exacerbate the effect of another. Much more research is required to understand the chains of events to help us form interventions and policy responses.

As discussed earlier, studies sometimes describe differences between ethnic groups after adjustment for multiple factors, including region, socioeconomic deprivation, urbanisation, occupation, overcrowding, housing tenure and pre-existing health conditions. To do this, researchers use regression models to quantify the relationship between COVID-19 risk and ethnicity, while statistically controlling for other factors. A ‘kitchen sink’ approach to regression models, in which numerous variables are thrown together without a guiding theory, obscures the role and interaction of upstream and downstream factors. Complex pathways link macroeconomic, social and public policies with an individual’s risk of, for example, deprivation, worklessness, overcrowding and long-term health conditions.

Analyses that ‘control out’ the effect of factors on the causal pathway can be highly misleading. In doing so, analysts may fail to identify key areas for intervention to reduce inequality and improve health outcomes for individuals. Rather than controlling out intermediary factors, an analysis that models causal chains can build understanding of where preventive action could be targeted. A well-designed study would explain (ideally using a theory) how and why variables are selected, where variables lie on the causal pathway, and the limitations of the analysis (such as quality and limitations of the data sources).

One example of a theory-led approach is a study of ethnic differences in diabetes. This study tested a model linking the wider socioeconomic and environmental determinants (eg neighbourhood social and physical disorder and access to health care) to intermediate outcomes, including lifestyle (eg physical activity and diet) and clinical risk factors (eg blood pressure) to establish diabetes risk. Being of black ethnicity was linked to having greater socioeconomic and environmental risk and these factors in turn were associated with having greater lifestyle and clinical risk. Lifestyle and clinical factors were strongly and directly associated with diabetes risk. From this kind of model, it is evident that focusing only on downstream interventions (eg blood pressure-lowering medication) would not address the long-standing structures and policies that result in black, Asian and minority ethnic people being more likely to live in neighbourhoods with higher levels of social and physical disorder. A similar approach was used in a study of ethnicity and likelihood of testing positive for COVID-19. That study tested whether socioeconomic factors, comorbidities, and lack of social distancing are part of the causal chain that could explain associations between ethnicity and COVID-19 risk.

Another example is a report from Public Health England (2018) that set its analysis of ethnic inequalities in health and social outcomes within a theoretical framework of structural racism as an overarching cause. The report concluded:

‘The major determinants of ill-health are largely the same across all ethnic groups. However, ethnicity is a salient social [our italics] identifier in modern Britain, shaping people’s networks of association and their social and economic opportunities. Further, minority ethnic identities continue, in many circumstances, to be stigmatised and subject to exclusionary forces. Therefore, without explicit consideration of ethnicity within health inequalities work, there is a risk of partial understanding of the social processes producing poor health outcomes and ineffective, or even harmful, intervention.’

Early reports on ethnicity and the direct effects of COVID-19 examined COVID-19 deaths. These studies do not help us understand whether there are ethnic differences in the risk of acquiring the infection, the risk of poor outcomes once infected, or both. More recently, studies of people with COVID-19 in the community and in primary care, and an additional analysis of published data from Public Health England, show a higher risk of infection for people from minority ethnic groups.

The recording of ethnicity can also be problematic. Ethnicity is incompletely recorded in hospital records and on death certificates in Scotland – and not recorded at all on death certificates in England, Wales and Northern Ireland. In the recent disparities report from Public Health England, ethnicity could not be established for around 10% of COVID-19 hospitalisations. Additionally, the ONS reports on deaths involving COVID-19 could not include around 10% of deaths in their analysis of ethnicity because of missing data. There is no perfect solution to handling missing data, but studies should provide information about the extent of missing data, and the sociodemographic characteristics and outcomes for people with complete and incomplete ethnicity data so that the reader can make their own judgement about any possible bias.

Given that the study of ethnicity and health is a complex area with the potential to do both harm and good, it is good practice to consider whether and how any affected communities were involved from the start of a study. In line with our earlier questions, it is useful for readers to consider whether people from affected communities contributed to shaping why the study is looking at ethnicity, how ethnicity is conceptualised, the data collection and analytical approach, and interpretation of the findings.

Existing social norms influence the questions we ask and the way that data is collected and interpreted. These need to be tested with the people whose health the study is intended to benefit. Although meaningful engagement with affected populations is not straightforward and can require us to confront tensions and balance interests, as well as acknowledge our own unconscious bias, it is part of a well-designed study. The challenges of engagement are even greater during the pandemic, but support is available for NHS and clinical researchers to do this. Work undertaken in partnership with representatives from the communities experiencing these inequalities has the potential to improve their health outcomes, increase public trust and make better use of public resources.

The five questions set out should help readers to evaluate the emerging evidence on COVID-19 and ethnicity. In evaluating this evidence, it is important to remember that all analysis to test explanations for ethnic differences in COVID-19 risk should start with a set of hypotheses, with the assumptions underpinning these made explicit. In addition, a theory-led approach is vital, as well as appropriate data to test the theory, and an inclusive research practice to help frame the research questions, design and interpretation appropriately.

It is commendable that the research and analytical community has acted quickly to develop an evidence base on ethnic inequalities in COVID-19 risk and outcomes, but it is also a field in which research can do harm as well as good. The five questions discussed are aimed at consumers of research, but rigorous standards are also needed for those carrying out studies on ethnicity and health. Standards, however, are starting to emerge and these include guidance for researchers, journals and reviewers and detailed checklist guidance to help authors and reviewers improve the quality of published research. Until these standards are more widely adhered to, the onus will be on readers to make their own careful evaluations of these studies.

Note: This long read was amended on 2 September 2020 to correct a reference to the OpenSAFELY project.