Discussion
This is the first study to specifically describe the impact of South Asian ethnicity on the outcome of COVID-19 infection using highly characterised and accurate primary and secondary data from patients admitted to hospital in the UK. South Asians were significantly younger and twice likely to have diabetes than white patients, and when accounted for the age structure of the local population had a high admission and death rate. South Asians were also more likely to present with severe symptoms, but with no difference in the duration of symptoms and more likely to be admitted to ITU. Importantly, after adjusting for age, deprivation and multiple comorbidities, the effect of South Asian ethnicity on mortality was 42% higher.
Our study is in line with two population-based studies10 11 where South Asians were found to be at increased risk of death, with similar effect size with the study11 that considered similar covariates for adjusting (aHR 1.4 (1.2–1.8) vs 1.6 (1.4–1.8)). However, the findings of the two population-based studies could reflect infection rates rather than case fatality rates, and therefore differentiating this is important from a public health and research perspective. If increased deaths in South Asian patients reflected high infection rates, then our focus should be on looking at barriers and emphasising the need for adherence to current social distancing guidelines. If this reflects an increased susceptibility to poorer outcomes from SARS-CoV-2 infection, we need to urgently understand the reasons for severity of infection and mitigate risk or develop targeted treatments. While no firm conclusion can be drawn from the current data set, our study potentially supports the latter, with South Asian patients more likely to be admitted, more likely to present with severe symptoms and have an increased risk of mortality. This supports the call by the government and research community for urgent research on the reasons underpinning these observations. Our study was not sufficiently powered to report on other ethnic groups, particularly those of black ethnicity, and therefore the findings among these ethnic groups should be interpreted cautiously.
The excess age-adjusted mortality in COVID-19 is not solely attributable to a range of cardiovascular and metabolic risk factors that are over-represented in this ethnic group. In sensitivity analysis the HR for South Asian ethnicity using both Cox regression and propensity score matching was stable when ‘number of co-morbidities’ was exchanged for the presence of specific comorbidities including diabetes mellitus and hypertension. To place this in context, the effect of South Asian on mortality is significantly less than the effect of one or more comorbidities (present in 80.2% of all admitted patients) and approximates to the effect of ageing 10 years in the white population. In this study we did not observe an independent signal related to higher deprivation levels26 and poor outcomes in contrast to the population-based study,11 suggesting deprivation is likely to be related to high infection rates and thereby high mortality, rather than high rates of severe infection leading to increased mortality.
It is notable that in the subanalysis of patients admitted to QEHB, where we were able to immediately integrate a COVID-19-specific assessment into our EHR, South Asian patients appear to present with more severe disease, but there was no difference in the duration of symptoms prior to admission, suggesting disease severity was not caused simply by delayed presentation to medical services or differences in health service utilisation (although this cannot be fully excluded, given the unknown burden of COVID-19 in the community). Indeed, when comparing those still alive and those who had died at the end of the study, the patients who had died had a shorter history of symptoms prior to admission, suggesting a different disease course.
A significantly higher rate of admission to ITU in the South Asian ethnic group could relate to this more severe disease at presentation. It may also relate to patient-level differences in joint decision-making regarding ITU treatment, in patients who have higher levels of specific comorbidities such as dementia and COPD, groups that are significantly over-represented in the white ethnic group, which was also significantly older.
The limitations in our overall analysis need to be considered, specifically that 5% of patients remain in hospital at the time of the data lock and more patients have been admitted, so our findings will evolve. Since the proportion of patients presenting from different ethnic groups was stable across the course of data collection, any consequence for our main conclusion on the mortality risk in South Asians admitted to hospital is likely to be small. Data on the admission severity scoring did not include all patients, which is a limitation. This limitation reflects the real-world response within a global pandemic which includes designing a score to inform care escalation decisions and updating the UHB electronic health record to capture this information during the first wave of patient admissions. However, these data suggest further exploration of severity of disease at presentation is warranted.
It is also important to acknowledge that standardised admission and mortality ratios from Birmingham and Solihull use the most recent census data, but that these are from 2011. Estimates of the contemporary age structure do not however suggest a need to significantly qualify these findings. The UK has not undertaken widespread screening or diagnosis of patients in the community, and we are therefore unable to comment on the natural history of COVID-19 prior to admission to secondary care, irrespective of ethnic group. This testing regimen is likely to evolve with the development of capacity and methodology and will provide a more complete picture of COVID-19. A description of disease in the community will help build a clearer understanding of the apparent excess mortality following admission, for which there remain a number of possible explanations. A limitation of this study (and, arguably, of any observational study) is that it cannot exclude the possibility that another, unmeasured variable could account for the ethnicity effect described here. The assessment of comorbidities does not reflect the degree of severity of the condition, nor disease treatment or control, and the assessment of social deprivation might impact on chronic disorders. There is also the possibility that differences in work or home living arrangements might impact on potential transmission and this may be different across ethnic communities. However, a real strength has been the ability to study a highly curated and complete data set, without the inherent issues of significant undercoding seen with morbidity and ethnicity data when using a secondary care data set.
The biological basis of any difference in outcome can only be speculated on at present. There are reported differences in outcomes for non-white ethnic groups from ARDS even after adjusting for sex, age, disease severity, type of hospital and median household income.27 28 The worst clinical manifestations of COVID-19 appear to be associated with a cytokine storm syndrome. Here a hypercytokinaemia is seen,4 with predictors of mortality reflecting a virally induced inflammatory state which can be assessed using a scoring system including validated clinical laboratory tests.29 Candidate genes associated with ARDS have been identified in bioinformatic analyses, with a strong predominance of inflammatory pathways, including reactive oxygen species, innate immunity-related inflammation and endothelial vascular signalling pathways.30 Ethnicity may influence cytokine gene polymorphisms and inflammatory profiles following specific challenges,31 with some ethnic groups more prone to a heightened inflammatory response. Of note, socioeconomic factors might also impact on inflammatory pathways and gene expression.32 These factors remain poorly understood, and were a priority for our patient and public involvement group who were consulted for this study, and there is an urgent need to understand the genomic and associated phenomic and socioeconomic characteristics of patients who are susceptible or resistant to the severe manifestations of COVID-19 to understand this further.
Although our study includes only one NHS Foundation Trust, it covers an ethnically diverse contiguous population of 1.3 million people for which it is the sole provider of adult acute secondary care across four hospital sites. This provides for continuity of data, clinical protocols and access to therapy. The immediate availability of access to an electronic representation of a primary care record to support the care of admitted patients also supports the integrity of data collection, the quality of which might otherwise be more limited.33
Our findings, which describe and quantify the risk of COVID-19 in the South Asian population, are relevant to national policy and to understanding the underlying biological mechanisms in ‘at risk’ populations. Future studies will extend our observations and explore underlying epidemiology and biological mechanisms, to improve interventions based in the community, the emergency department, ward and ITU. Perhaps most importantly our findings inform the UK’s national discussion on at ‘at risk’ groups and the ensuing fear arising from uncertainty.