Ethnic inequalities in COVID-19 infection, hospitalisation, intensive care admission, and death: a global systematic review and meta-analysis of over 200 million study participants

Summary Background COVID-19 has exacerbated existing ethnic inequalities in health. Little is known about whether inequalities in severe disease and deaths, observed globally among minoritised ethnic groups, relates to greater infection risk, poorer prognosis, or both. We analysed global data on COVID-19 clinical outcomes examining inequalities between people from minoritised ethnic groups compared to the ethnic majority group. Methods Databases (MEDLINE, EMBASE, EMCARE, CINAHL, Cochrane Library) were searched from 1st December 2019 to 3rd October 2022, for studies reporting original clinical data for COVID-19 outcomes disaggregated by ethnicity: infection, hospitalisation, intensive care unit (ICU) admission, and mortality. We assessed inequalities in incidence and prognosis using random-effects meta-analyses, with Grading of Recommendations Assessment, Development, and Evaluation (GRADE) use to assess certainty of findings. Meta-regressions explored the impact of region and time-frame (vaccine roll-out) on heterogeneity. PROSPERO: CRD42021284981. Findings 77 studies comprising over 200,000,000 participants were included. Compared with White majority populations, we observed an increased risk of testing positive for infection for people from Black (adjusted Risk Ratio [aRR]:1.78, 95% CI:1.59–1.99, I2 = 99.1), South Asian (aRR:3.00, 95% CI:1.59–5.66, I2 = 99.1), Mixed (aRR:1.64, 95% CI:1.02–1.67, I2 = 93.2) and Other ethnic groups (aRR:1.36, 95% CI:1.01–1.82, I2 = 85.6). Black, Hispanic, and South Asian people were more likely to be seropositive. Among population-based studies, Black and Hispanic ethnic groups and Indigenous peoples had an increased risk of hospitalisation; Black, Hispanic, South Asian, East Asian and Mixed ethnic groups and Indigenous peoples had an increased risk of ICU admission. Mortality risk was increased for Hispanic, Mixed, and Indigenous groups. Smaller differences were seen for prognosis following infection. Following hospitalisation, South Asian, East Asian, Black and Mixed ethnic groups had an increased risk of ICU admission, and mortality risk was greater in Mixed ethnic groups. Certainty of evidence ranged from very low to moderate. Interpretation Our study suggests that systematic ethnic inequalities in COVID-19 health outcomes exist, with large differences in exposure risk and some differences in prognosis following hospitalisation. Response and recovery interventions must focus on tackling drivers of ethnic inequalities which increase exposure risk and vulnerabilities to severe disease, including structural racism and racial discrimination. Funding 10.13039/501100000269ESRC:ES/W000849/1.


Introduction
Minoritised ethnic groups have suffered disproportionately from the COVID-19 pandemic, with higher rates of SARS-CoV-2 infection, hospitalisation, intensive care unit (ICU) admission, and death, compared to the ethnic majority group in a given population. [1][2][3][4][5][6][7] It is not known whether the inequalities in severe disease and deaths, seen in many countries among minoritised groups, relates to greater infection risk, poorer prognosis, or both. [8][9][10][11][12] Previous systematic reviews conducted early in the pandemic have demonstrated and quantified some of these risks. [13][14][15][16][17] However, a new review is warranted for several reasons: first, seroprevalence studies were lacking despite providing the best estimates of infection; second, broad ethnic categorisations were analysed and may hide important heterogeneity in risks between different ethnic groups; and third, now that COVID-19 has spread across the world and remains endemic in many areas, a much larger body of literature is available to address these issues. [13][14][15][16][17] We therefore sought to perform a systematic review and meta-analysis examining clinical outcomes in COVID-19, comparing people from minoritised ethnic groups to the ethnic majority group, incorporating data from lower-and middle-income countries that were not available in previous meta-analyses.

Methods
This systematic review and meta-analysis is reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, 18 and was registered with PROSPERO (CRD420212824981). Ethical approval and informed consent of participants were not required for this metaanalysis as no new data were collected.

Research in context
Evidence before this study We searched PROSPERO for existing systematic reviews with 'ethnic' or 'race' in the title, from 1st December 2019 to 30th August 2022. Previous systematic reviews, conducted early in the pandemic, synthesised emerging evidence primarily from the United Kingdom (UK) and United States of America (USA), identifying an increased risk of infection, severe disease, and death amongst minoritised ethnic groups. However, it is not known whether ethnic inequalities in severe disease and deaths, seen in many countries, relates to greater infection risk, poorer prognosis, or both. Broad ethnic groups described in previous meta-analyses may have also obscured heterogeneity in risks of infection, severe disease and death between different minoritised ethnic groups.

Added value of this study
We searched MEDLINE, EMBASE, EMCARE, CINAHL, and the Cochrane Library from 1st December 2019 to 3rd October 2022. We assessed inequalities in both incidence and prognosis by separating population-based studies from studies including only COVID-19 cases. Over 200 million participants were included from 77 studies. Compared with White majority groups, we identified an increased risk of testing positive for infection for Black, South Asian, Mixed, and Other ethnic groups; an increased risk of infection for Hispanic people was observed only in seroprevalence studies.
In population-based studies, Black and Hispanic ethnic groups and Indigenous peoples had an increased risk of hospitalisation; most minoritised ethnic groups and Indigenous peoples had an increased risk of ICU admission; and Hispanic, Mixed, and Indigenous groups had an increased risk of mortality. Following hospitalisation, South Asian, East Asian, Black and Mixed ethnic groups had an increased risk of ICU admission, and the risk of mortality was greater in Mixed ethnic groups.

Implications of all the available evidence
We present the most comprehensive summary of ethnic inequalities in a range of outcomes relating to COVID-19, during the first few years of the pandemic, before widespread immunity. We demonstrate the presence of systematic ethnic inequalities relating to both COVID-19 infection and severe disease, but to varying extents across minoritised ethnic groups. In particular, we observed large differences in infection risk for minoritised ethnic groups. Our findings highlight the need for policy interventions to address ethnic inequalities in exposure to the virus. Ethnic inequalities in prognosis following hospitalisation were also noted, which may reflect poorer healthcare quality for minoritised ethnic groups or differential vulnerability to severe disease. Our findings are of vital public health importance and should inform strategic response and recovery policies.

Search strategy
The search strategy was developed and conducted by an academic librarian (PD), using the databases MEDLINE, EMBASE, EMCARE, CINAHL, and the Cochrane Library (supplementary materials: search strategy). Peerreviewed publications were searched for between 1st December 2019 and 3rd October 2022.

Eligibility criteria
The relevant outcomes were SARS-CoV-2 infection (laboratory confirmed infection through polymerase chain reaction [PCR] or evidence of previous infection through laboratory confirmed IgG/IgM for SARS-CoV-2 [seroprevalence]), hospitalisation, ICU admission, and mortality. Studies were eligible if they reported original clinical data (cross-sectional, case-control, cohort studies) on the relevant outcomes, disaggregated by ethnicity or closely related indicators of ethnicity (i.e., Indigenous status, race, country of birth, migrant status) (see Table S1 for detailed inclusion and exclusion criteria). Although we did not set out to make comparisons by country of birth or migrant status, these indicators were included as ethnicity data are not collected in some countries. 19 Population-based studies (individuals with and without confirmed SARS-CoV-2 infection) were eligible, as well as studies which examined prognosis (individuals with confirmed SARS-CoV-2 infection only). Ecological studies, prognostic modelling studies, animal studies, qualitative studies, and pre-prints, were excluded. Studies were also excluded if participants were recruited based on a specific physical or mental health condition or healthcare utilisation (other than for COVID-19), as were studies specifically in children (under the age of 16) and religious groups.

Study selection
Two reviewers independently screened titles, abstracts, and full texts (PI assessing all articles, with DK, HT, SVK, LB, DP and SS each assessing a proportion). Where there were disagreements in title and abstract screening, articles identified as potentially relevant by one reviewer were included for full text screening. Disagreements in full texts were resolved by discussion or consultation with the review team. The software, Covidence, 20 was used for screening following automatic de-duplication. To minimise the inclusion of duplicate data (i.e., participants from the same population assessing the same outcome), predefined criteria were used to determine which dataset to include (supplementary materials: criteria to minimise inclusion of duplicate data).

Data extraction
One reviewer (PI) completed 100% of the data extraction from each eligible article, all of which were independently checked by additional reviewers (DK, HT, DP, SS).
Data on study and participant characteristics, ethnicity, outcomes, and covariates, were extracted to an Excel file. Ethnicity measures were extracted and evaluated to determine the comparison group for meta-analyses. When describing ethnic groups, we have used terminology from the included studies. Where US studies reported both race and ethnicity, we chose the categories that were most amenable to meta-analysis (i.e., comparable to other studies).

Risk of bias and conceptualisation of ethnicity
Two reviewers (PI completing 100% and DK, HT, LB, DP and SS a proportion) independently assessed the risk of bias of each included study, using an adapted Joanna Briggs Institute (JBI) tool 21 (supplementary materials: adapted JBI critical appraisal tool). Scores were calculated as a total out of the maximum number of applicable questions, then standardised. Studies with a score of 80-100% were considered low risk of bias, 60-79% medium risk of bias, and 0-59% high risk of bias. One item from the JBI, regarding the description of study participants, was adapted to critically evaluate how ethnicity was conceptualised and measured. Studies which measured ethnicity through self-report or another reliable indicator (e.g., country of birth registered at birth or migration) and used disaggregated descriptions of ethnicity scored positively on this item of the risk of bias tool. No studies were excluded based on critical appraisal scores.

Meta-analysis
To determine ethnic inequalities in COVID-19 health outcomes, we prioritised extracting age-and sexadjusted results which are important confounders of ethnic inequalities, 22 but considered other variables to be likely mediators and therefore should not be adjusted for. 8 We contacted authors to request data, if unavailable (supplementary material: data manipulation methods). Unadjusted or over-adjusted models were extracted if age-and sex-adjusted models were not available (highlighted in forest plots and reported in Table 1).
We conducted meta-analyses on age-and sex adjusted data to determine the risk of each outcome across disaggregated minoritised ethnic groups (guided by the critical appraisal) compared to the White majority. The Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach was used to evaluate the certainty of the overall evidence for adjusted analyses. 23 The overall certainty estimates were categorised as high, moderate, low, or very low. In keeping with GRADE guidance for prognostic studies, observational studies started as high certainty evidence and were rated down for risk of bias, imprecision, inconsistency, indirectness, and publication bias (supplementary materials: GRADE criteria). 23 First author Outcomes To disentangle ethnic inequalities in risk of testing positive for infection from SARS-CoV-2, from ethnic inequalities in prognosis once infected (i.e., hospital admission, ICU admission, mortality), we separated studies by their denominator in the following way: (1) general population studies including individuals with and without SARS-CoV-2 infection; (2) individuals with confirmed SARS-CoV-2 infection; and (3) individuals hospitalised with COVID-19.
As some studies used country of birth or migrant status as an indicator of ethnicity, and the majority ethnic group was not White in some studies, we conducted unadjusted meta-analyses to identify the risk of each outcome for minoritised ethnic groups (combined within each study) compared with the ethnic majority group, which varied depending on which ethnic group was the majority in the study country. Crude numbers were used to calculate unadjusted risk ratios (RR) for minoritised ethnic groups versus the ethnic majority group. We conceptualised majority groups in terms of power and privilege, and minority if the ethnic or Indigenous group meets one or more of the following criteria: the group is numerically smaller than the rest of the population; it is not in a social, economic, or politically dominant position; and it has a culture, language, religion or ethnicity that is distinct from that of the majority. This guided our decisions when selecting the ethnic majority group (for the unadjusted analyses), as we did not always use the reference group of the study or the statistical majority. This analysis is reported in supplementary materials.
We performed meta-analyses using the DerSimonian and Laird random effects model 24 to determine the pooled effect sizes with corresponding 95% Confidence Intervals (CIs). Levels of statistical heterogeneity were determined using the I 2 statistic. 25 Where there were sufficient data (n > 10), meta-regressions were used to explore whether region (low-and middle-income countries [LMIC]) versus high-income countries [HIC]) and time frame (before widespread vaccine roll-out versus after widespread vaccine roll-out) were drivers of heterogeneity. Publication bias was explored through visual funnel plots, Egger's test of asymmetry for metaanalyses with 10 or more studies. 26 All analyses were conducted using Stata SE 15. 27 Synthesis without meta-analysis (SWiM) Where data were not amenable to meta-analyses, for example, where it was not possible to extract or calculate effect sizes, we provide synthesis without meta-analysis (SWiM) in supplementary materials. Effect direction plots and sign tests were conducted to assess evidence of associations. The quantitative synthesis is reported in line with the SWiM guidelines. 28

Sensitivity analyses
As several studies reported country of birth, nationality, or migrant status, as an indicator of ethnicity, sensitivity analyses were conducted, replicating the unadjusted analyses with these studies excluded. To further explore differences across regions (LMIC versus HIC), the adjusted meta-analyses were stratified by region, where there were sufficient data (n > 10). Additional sensitivity analyses were conducted, excluding studies with a high risk of bias (as determined using the JBI), to explore the impact on the main findings. The findings of the sensitivity analyses are presented in supplementary materials.

Role of the funding source
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. All authors had access to the data and critically reviewed and approved the manuscript as submitted.  S1). Most study designs were cohort (N = 50, 65%) or cross-sectional (N = 23, 30%) ( Table S2). Details of the outcomes, comorbidities, and covariates included in each study are reported in Table 1.
The summary of scores for the JBI risk of bias items are presented in supplementary materials (Fig. S2) and the standardised scores for each study are presented in Table S2. Whist an indicator for ethnicity was reported in all 77 studies, only 29 (38%) studies measured ethnicity (or closely related indicators) in a valid and reliable way, and it was unclear how ethnicity was measured in 17 (22%) studies. These 29 studies also disaggregated broad ethnic groups, which we considered important to determine whether there was heterogeneity in COVID-19 outcomes when using a more granular categorisation of minoritised ethnic groups (e.g., South Asian and East Asian, as opposed to an aggregated Asian ethnic group).
Five population-based studies investigated the risk of hospitalisation (Fig. 3A). Black, Hispanic, and Indigenous people had an increased risk of hospitalisation  We assessed prognosis following infection. Four studies reported the risk of hospitalisation among confirmed COVID-19 cases (Fig. 4A). Hospitalisation risk was increased for Asian and Indigenous people,   (Fig. 4B). However, three of the four studies had a high risk of bias. The risk of death among confirmed COVID-19 cases was reported in four studies, identifying no increased risk for any minoritised ethnic group, compared to the White majority, observing reduced risks for South Asian, East Asian, Mixed, and Other ethnic groups (Fig. 4C). When assessing prognosis following hospitalisation, we noted that South Asian, East Asian, Asian (aggregated in studies, including South and East Asian people), Black, and Mixed groups were all more likely to be admitted to ICU compared to White majority participants. Those from Hispanic and Other ethnic groups were not at an increased risk, across four studies (Fig. 5A). In a synthesis of 11 studies, we observed an increased risk of mortality for Mixed ethnic groups (K = 6, aRR = 1.06, 95% CI: 1.02-1.09, I 2 = 97.0), compared to the White majority, with trends towards an increased risk for Indigenous peoples (K = 3, aRR = 2.03, 95% CI: 0.99-4.14, I 2 = 99.8) (Fig. 5B). Egger's test indicated no evidence of publication bias (p = 0.626).  The findings of the unadjusted analyses are reported in supplementary materials. Minoritised ethnic groups were found to have an increased risk of infection, seropositivity, hospital admission (population-based studies only), ICU admission (population-based studies and studies of hospitalised patients with COVID-19 only), but not mortality (Figs. S3-S7). Sensitivity analyses replicated the unadjusted analyses, excluding studies which reported country of birth, nationality, and migrant status. In these analyses, minoritised ethnic groups were not at an increased risk of infection or hospital admission, though all other original findings remained the same (Figs. S8-S12).
Meta-regressions explored whether region (LMIC versus HIC) was associated with heterogeneity, where there were sufficient data (n > 10). Region did not explain heterogeneity for the pooled unadjusted or     We provide SWiM for the findings of studies that were not amenable to meta-analysis. Across all outcomes, 16 studies were excluded from the unadjusted analyses, and 37 studies that either contained certain ethnic groups that were not included (i.e., if only one study reported an effect for that ethnic group), or could not be included at all (i.e., if the reference group was not White). The SWiM reports mixed findings, which may reflect the heterogeneity of the studies (supplementary materials: Tables S4-S6).

Discussion
We identified systematic inequalities experienced by minoritised ethnic groups, but to a varying extent across ethnic and Indigenous groups. We found that Black, South Asian, Mixed, and Other ethnic groups had a greater risk of testing positive for infection. The findings demonstrate large differences in exposure risk, which may be driving ethnic inequalities in severe outcomes. Almost all minoritised ethnic groups were at an increased risk of hospital admission and ICU admission, in population-based studies, yet these findings attenuated when examining outcomes among confirmed COVID-19 cases only. Additionally, Hispanic people were more likely to be seropositive, compared to the White majority. Seropositivity to any SARS-CoV-2 protein within populations that have not yet been vaccinated highlights a history of past infection; therefore, it may be that Hispanic groups had reduced access to testing early in the pandemic.

Articles
We also observed differences in prognosis following hospitalisation, with South and East Asian, Black and Mixed ethnic groups being more at risk of ICU admission, and Mixed ethnic groups being more likely to die from COVID-19. Finally, we found that the unadjusted risk of severe disease (ICU admission and mortality) among hospitalised COVID-19 cases was greater for minoritised ethnic groups (versus the majority ethnic group) in LMIC compared to HIC. This could be due to HIC having more universal health care systems (excluding the USA). Our work is the most comprehensive summary of risk for minoritised ethnic groups globally, by using multiple markers for key clinical outcomes (molecular testing and serology for infection, and hospitalisation, ICU admission, and death for severe disease). Future work will likely include cohorts with differing levels of immunity (from previous infections, or heterogenous vaccine regimens) to different SARS-CoV-2 variants.
In agreement with previous meta-analyses, our data clearly demonstrates that the COVID-19 pandemic has exacerbated existing socioeconomic inequalities that disproportionately affect the health of minoritised ethnic and Indigenous groups. 7,13,14,16,17,136 Structural racism (discrimination embedded within systems) drives socioeconomic inequalities that increase the risk of exposure to COVID-19 infection. 137 During the pandemic, when multiple countries implemented strict lockdown measures, minoritised ethnic groups were more likely to be employed in sectors with increased exposure and were less likely to be able to self-isolate or work from home, due to economic precarity. 8,138,139 Minoritised ethnic groups were also more likely to live in overcrowded households with reduced access to open spaces (a consequence of racism and socioeconomic inequality), which could lead to frequent and prolonged exposure to airborne pathogens. 140 Among populationbased studies, we observed an increased risk of severe disease for Black, Hispanic, South Asian, East Asian,  Mixed and Indigenous ethnic groups, but this attenuated once we examined prognosis following infection, demonstrating that differences in exposure risk may have driven the greater number of people experiencing severe disease. When assessing prognosis following hospitalisation, we observed ethnic inequalities in ICU admission and mortality, which potentially reflect poorer healthcare quality, or barriers to adequate healthcare (e.g., language barriers, migrant status, mistrust, disparities resulting from highly marketised healthcare in the US), 141 resulting from institutional racism. 136 Going forwards, unless significant effort is made to address these inequalities, it is likely that minoritised ethnic groups will continue to have increased exposure to respiratory viruses as the world learns to live with COVID-19.
Despite the importance of racism in relation to clinical outcomes, as captured by ethnicity, we found that the quality of ethnicity recording in studies to be suboptimal. Approximately a quarter of studies did not  describe how they recorded ethnicity, despite investigating its relation to clinical outcomes. The most common method was using routinely recorded data, yet evidence shows that ethnicity is often miscoded, and this affects minoritised ethnic groups disproportionately. 142 Poor data may obscure the true extent of ethnic inequalities in COVID-19 health outcomes. In contrast to previous systematic reviews, 13,14,16,17 when we further disaggregated Asian groups, we found that South Asian people were at increased risk of infection, whereas East Asian people were not, and studies using an aggregated Asian group observed a decreased risk of infection. Our analysis demonstrates the importance of granularity in collecting ethnic categories to describe the   outcomes. If we only examined 'Asian' as one ethnic group, we may have missed the increased risk of infection among South Asian people, since the decreased risk of infection in all Asian people may have nullified this effect. We call for health systems to make a concerted effort to record self-identified ethnicity and for research to use disaggregated ethnic groups. 143 Our systematic review has limitations. There was a large decrease in the number of studies and participants when we excluded studies with duplicate data, which was necessary to ensure rigour. This meant that in some analyses, estimates for certain ethnic groups were obtained from a small number of studies. Furthermore, although we set out to conduct a global synthesis, there were regions with no or limited data (e.g., Australia). Although there was some indication of publication bias in the unadjusted analyses, there was no evidence of publication bias in the adjusted analyses. Relatedly, the certainty of evidence for each outcome ranged from moderate to very low, mainly due to inconsistency within ethnic groups across studies. Additionally, heterogeneity between studies was high. However, we note that this is common with observational studies as I 2 is calculated as the proportion of total variation which is attributable to between-study variation, meaning studies with large sample sizes (i.e., small within-study variation), are likely to show inflated heterogeneity. 144 These limitations may result from the inclusion of observational studies from a range of regions. Nevertheless, we sought to present the most comprehensive work illustrating the disproportionate impact COVID-19 has had on minoritised ethnic groups. Clearly, this will have implications for future pandemics, especially if future pathogens have similar transmission dynamics and structural determinants.
In conclusion, we found clear evidence of systematic inequalities in COVID-19 health outcomes, experienced by minoritised ethnic groups, but to varying extents across ethnic groups during the first two years of the pandemic. We highlight the need to recognise and determine that pathways that lead to differing risks to COVID-19, before and after vaccine rollout periods. We observed large differences in exposure risk, particularly for Black, South Asian, Mixed, and Other ethnic groups. Hispanic groups may have had limited access to molecular testing early on in the pandemic, as reflected by the greater risk of seropositivity. Almost all minoritised ethnic groups being at an increased risk of severe outcomes among population-based studies (which attenuated when assessing outcomes only in confirmed cases), demonstrating the need for policy interventions to reduce exposure to infection. The differences in prognosis following hospitalisation may reflect poorer healthcare quality, illustrating the need for services and clinicians to ensure equitable care. 145 The COVID-19 pandemic has exposed and exacerbated ethnic inequalities in health, therefore response and recovery should focus on tackling the drivers of inequalities, including structural racism and racial discrimination. 146 Contributors PI, DK, LB, HT, SA, and SVK drafted the study protocol. DP, SS, PD, LJG, LBN, and MP provided critical feedback on the protocol. PD conducted the literature searches. PI screened all records, and DK, DP, SS, SK, LB, HT, and SVK contributed to the screening process and selection of included studies. PI initially extracted data, which were subsequently verified by a second reviewer (DK, DP, SS, SVK, HT) who also completed independent risk of bias scores. PI completed the data analysis, and all authors had access to the data. EK created the visual map and both EK and LJG supported the analyses. All authors critically reviewed and approved the manuscript as submitted.

Data sharing statement
The study protocol is published on PROSPERO: CRD42021284981. All extracted data and analytical codes are available from the corresponding author are available upon request.

Declaration of interests
SVK was co-chair of the Scottish Government's Expert Reference Group on Ethnicity and COVID-19 and a member of the Scientific Advisory Group on Emergencies (SAGE) subgroup on ethnicity. MP reports grants from Sanofi and Gilead Sciences and personal fees from QIA-GEN, outside the submitted work.