key: cord-0910716-xz551lsd
authors: Nafilyan, Vahé; Islam, Nazrul; Ayoubkhani, Daniel; Gilles, Clare; Katikireddi, Srinivasa Vittal; Mathur, Rohini; Summerfield, Annabel; Tingay, Karen; Asaria, Miqdad; John, Ann; Goldblatt, Peter; Banerjee, Amitava; Glickman, Myer; Khunti, Kamlesh
title: Ethnicity, household composition and COVID-19 mortality: a national linked data study
date: 2021-03-24
journal: J R Soc Med
DOI: 10.1177/0141076821999973
sha: 3f59b3d6deb60e1309699c92d2ca4837b6009f3e
doc_id: 910716
cord_uid: xz551lsd

OBJECTIVE: To estimate the proportion of ethnic inequalities explained by living in a multi-generational household. DESIGN: Causal mediation analysis. SETTING: Retrospective data from the 2011 Census linked to Hospital Episode Statistics (2017-2019) and death registration data (up to 30 November 2020). PARTICIPANTS: Adults aged 65 years or over living in private households in England from 2 March 2020 until 30 November 2020 (n=10,078,568). MAIN OUTCOME MEASURES: Hazard ratios were estimated for COVID-19 death for people living in a multi-generational household compared with people living with another older adult, adjusting for geographic factors, socioeconomic characteristics and pre-pandemic health. RESULTS: Living in a multi-generational household was associated with an increased risk of COVID-19 death. After adjusting for confounding factors, the hazard ratios for living in a multi-generational household with dependent children were 1.17 (95% confidence interval [CI] 1.06–1.30) and 1.21 (95% CI 1.06–1.38) for elderly men and women. The hazard ratios for living in a multi-generational household without dependent children were 1.07 (95% CI 1.01–1.13) for elderly men and 1.17 (95% CI 1.07–1.25) for elderly women. Living in a multi-generational household explained about 11% of the elevated risk of COVID-19 death among elderly women from South Asian background, but very little for South Asian men or people in other ethnic minority groups. CONCLUSION: Elderly adults living with younger people are at increased risk of COVID-19 mortality, and this is a contributing factor to the excess risk experienced by older South Asian women compared to White women. Relevant public health interventions should be directed at communities where such multi-generational households are highly prevalent.

households. While several studies have examined whether socio-demographic and economic factors are driving ethnic inequalities in COVID-19, no study has sought to explicitly quantify the contribution of household composition to the elevated risk of COVID-19 mortality in ethnic minority groups.

Using retrospective data from the 2011 Census linked to Hospital Episode Statistics and death registration data for England, we examined the relationship between household composition and COVID-19 mortality risk among elderly adults (!65 years). Living in a multi-generational household was associated with an increased risk of COVID-19 death. The adjusted hazard ratios for living in a multi-generational household with dependent children were 1.17 (95% confidence interval [CI] 1.06-1.30) and 1.21 (95% CI 1.06-1.38) for elderly men and women. The hazard ratios for living in a multi-generational household without dependent children were 1.07 (1.01-1.13) for elderly men and 1.17 (95% CI 1.07-1.25) for elderly women. Using a causal mediation approach, we estimated whether the higher propensity to live in multi-generational households among ethnic minority groups contributed to their raised risk of COVID-19 death. We found that living in a multi-generational household explained around 11% of the elevated risk of COVID-19 death among elderly women from a South Asian background, but little for South Asian men or people in other ethnic minority groups.

Living in a multi-generational household is associated with an increased risk of COVID-19 infection and death. The increase in risk appears greater for elderly women than men living in a multi-generational household, and this is particularly the case when living with dependent children. It explains some of the excess COVID-19 mortality risk for women of South Asian background, but very little for men of South Asian background or people from other ethnic groups. Differences in household composition are therefore unlikely to be the main explanation of ethnic inequalities in COVID-19 outcomes, but may make an important contribution for some specific population subgroups, and may therefore be taken into account when prioritising vaccination. Relevant public health interventions (such as the provision of free accommodation to assist with self-isolation) should be considered to mitigate risks of infection spread within a household. Ensuring such interventions are accessible to communities where multi-generational households are highly prevalent (such as South Asian women) may be warranted.

People of ethnic minority background in the UK and the USA have been disproportionately affected by COVID-19 [1] [2] [3] [4] [5] compared to the White population, particularly Black and South Asian groups. While several studies have investigated whether adjusting for socio demographic and economic factors and medical history reduces the estimated difference in risk of mortality and hospitalisation, [6] [7] [8] the reasons for the differences in the risk of experiencing harms from COVID-19 are still being explored.

One important driver of these ethnic inequalities may be differences in household structure between ethnic groups. Household composition varies substantially between ethnic groups, with some ethnic minority populations more likely to live in large, multigenerational households. 9 While living in multi-generational households is associated with increased social capital, 10 which could have beneficial health effects, 11 it may also increase the risk of potential viral transmission. 12, 13 For older people, who are at greater risk of experiencing severe complications if infected, residing with younger people may represent an increase in exposure to infection, which could lead to an increased risk of hospitalisation and mortality from COVID-19. To the best of our knowledge, no study has yet examined whether the difference in household composition partly explains the elevated risk of COVID-19 mortality in ethnic minority groups.

In this study, we examined the relationship between household composition and COVID-19 mortality risk among elderly adults (!65 years) in England, with a focus on multi-generational households (elderly adults living with younger adults or dependent children). We then investigated how the propensity to live in a multi-generation household varies across ethnic groups, and whether this heterogeneity contributes to the raised risk of COVID-19 mortality among ethnic minority groups compared to the White population.

This retrospective cohort study was based on the 2011 Census of England linked to mortality registration data and Hospital Episodes Statistics (2017-2019). The study population included all usual residents of England aged 65 years or over in 2020, who had been enumerated in private households at the time of the 2011 Census (27 March 2011) , had not moved to a care home by 2019 (identified by linking to the NHS Patient Register) and were still alive on 2 March 2020. We further excluded individuals who entered the UK in the year before the Census due to their higher propensity to leave the UK prior to the study period, and those aged over 100 years at the time of the Census. Our study population consisted of 10,078,568 individuals aged 65 years or over at 2 March 2020 (see Table 3 in Appendix 1 for details on the number of participants at each stage of the sample selection).

To adjust for out-migration, we applied weights reflecting the probability of having remained in the country until March 2020 after being enumerated in March 2011, based on data from the NHS Patient Register and the International Passenger Survey (IPS). Further information on the data has already been published. 6 All the variables used in the analysis, including their definitions and sources, are detailed in Table 4 in Appendix 1.

Deaths involving COVID-19 included those with an underlying cause, or any mention, of International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) codes U07.1 (COVID-19, virus identified) or U07.2 (COVID-19, virus not identified). We analysed deaths that occurred between 2 March 2020 and 30 November 2020, which corresponds to the deaths that occurred during the first and second COVID-19 waves.

Household composition in 2020 was derived based the household composition at the time of the Census. We excluded people who died between 27 March 2011 and 1 March 2020 or had moved to a care home by 2019. To mitigate measurement error, we removed people aged 10 to 24 at the time of the Census because they were more likely to have moved out in 2020. We defined a multi-generational household to be a household in which someone aged 65 years or over on 2 March in 2020 co-resided with at least one other adult aged more than 20 years younger or with at least one child. Our household composition variable classified households in five categories: Single; Two elderly adults; Multi-generational household without dependent children; Multigenerational household with dependent children; three or more elderly adults. As sensitivity analyses, we removed people aged 10 to 19 instead of 10 to 24. We also defined a multi-generational household to be one in which someone aged 65 years or over in 2020 co-resided with at least one other adult aged more than 15 years (instead of 20 years) younger. We also used longitudinal data from the English Longitudinal Study of Ageing to estimate change in household composition among adults aged 65 or over between 2008-2009 and 2016-2017 (see Appendix for more details). We also compare estimates of the proportion of elderly adults living in multi-generational households by ethnic group in our linked data to estimates based on the 2019 Annual Population Survey.

In the mediation analysis, the exposure was selfreported ethnic affiliation based on a nine-group classification ( Table 4 in Appendix 1). The two mediators were binary variables for living in a multi-generational household with or without children.

Demographic factors, geographical variables, socioeconomic characteristics and measures of pre-pandemic health are listed in Table 4 in Appendix 1. These covariates were generally considered to be confounders of the relationship between household composition and COVID-19 mortality risk, and mediators of the ethnicity-mortality relationship ( Figure 1 ).

We calculated age-standardised mortality rates (ASMRs) stratified by household composition and sex, separately for COVID-19-related deaths and other deaths. The ASMRs were standardised to the 2013 European Standard Population and can be interpreted as deaths per 100,000 of the population at risk during the analysis period.

We estimated Cox proportional hazard models to assess whether the risk of COVID-19-related death varies by household type (using living with one other older adult with as the reference category) after adjusting for the geographical factors, socioeconomic characteristics and measures of health listed in Table 4 in Appendix 1. These factors could confound the relationship between household composition and COVID-19-related mortality, as shown in Figure 1 . We estimated separate models for men and women, as the risk of death involving COVID-19 differs markedly by sex. 7 When fitting the Cox models, we included all individuals who died during the analysis period and a weighted random sample of those who did not (5% of White people, and 20% among ethnic minority groups), and applied case weights to reflect the original population totals. Standard errors were clustered at the household level. We assessed the proportional hazard assumption by testing for the independence between the scaled Schoenfeld residuals and time-at-risk.

We conducted a causal mediation analysis 14 to estimate the proportion of excess risk in ethnic minority groups which is attributable to living in a multigenerational household. As a measure of inequality in COVID-19 mortality, ethnicity-specific odds ratios for COVID-19 mortality were estimated using logistic regression models, fitted to men and women separately and adjusting solely for age in the baseline model. The proportion of the difference in COVID-19 mortality rates between ethnic groups mediated by living in a multi-generational household was then estimated as the Average Causal Mediated Effect as a proportion of the age-adjusted difference in the probability of COVID-19 mortality, using a non-parametric approach 15 (see Appendix 1 for more details). The mediator models and the full outcome model were adjusted for geographical factors (region, population density, urban/rural classification), socio-economic characteristics (IMD decile, educational attainment, social grade, household tenancy) and health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts), We did not adjust for overcrowding or housing type, as these are likely to be consequences of living in a multi-generational household rather than confounding factors ( Figure 1 ) In addition, among elderly adults, household size and overcrowding are strongly linked to multi-generational households, because most households with more than two people are multi-generational. Confidence intervals were obtained via bootstrapping, clustered at the household level, using 500 replications. All statistical analyses were performed using R version 3.5.

Characteristics of the study population are reported in Table 1 . In our study population of 10,078,568 individuals in England aged 65 years or over who were not in a care home in 2019 and were still alive on 2 March 2020, just over half (54.0%) were female, the mean age was 75.2 years, and 93.9% reported being from a White ethnic background ( Table 2 ). Over the outcome period (2 March 2020 to 30 November 2020), 39,419 (0.39%) died of COVID-19, and 227,041 (2.3%) died of other causes. Figure 1 . Directed Acyclic Graphs summarising the relationship between ethnicity, household composition and COVID-19 mortality. Note: When analysing whether household composition directly affects the risk of COVID-19 death, our effect of interest is A. In the mediation analysis, where we estimate the proportion of the ethnic disparity in COVID-19 that is explained by living in a multi-generational household, the effects of interest are A þ B. Compared with elderly adults living with one other older adult (n ¼ 5,538,963), people living by themselves (n ¼ 3,287,395) had a higher mean age, were more likely to be female and tended to be more deprived. Older people living in a multi-generational household without dependent children (n ¼ 987,306) and with dependent children (n ¼ 199,112) were on average younger and were more likely to be from an ethnic minority group, live in London and large urban conurbations, and tended to be more deprived than older people living with another older adult. Summary statistics stratified by ethnic groups are reported in Table 5 in Appendix 1. Figure 2 shows that household composition varied substantially between ethnic groups. Among older people, just over 10% of those of White background lived in a multi-generational household, compared to over half of Bangladeshi or Pakistani background (58.7% and 58.8%, respectively) and 45.8% of Indian background. The patterns were similar for men and women, although a larger proportion of women live by themselves ( Figure 5 in Appendix 1). Similar proportions are obtained when using data from the 2019 Annual Population Survey and applying the same definition of multi-generational households ( Table 6 in Appendix 1).

Elderly people living by themselves were more likely to have died from COVID-19 over the study period than those living with another adult. For men, the ASMRs were 632 (95% confidence interval [CI]: 618 -646) and 465 (95% CI 456-473) per 100,000 of the population for those living by themselves and those living with another adult, respectively ( Figure  3 , Panel A). For women, the ASMRs were 309 (95% CI 302-316) and 236 (95% CI 230-243) per 100,000, respectively. A similar pattern is observed for deaths from other causes (Figure 3, Panel B) .

There was a positive association between the risk of COVID-19 death and living a in multi-generational household. Both elderly men and women living a multi-generational household without school-age children were more likely to die from COVID- 19 ( Figure 3 ). There was no clear relationship between living in a multi-generational household and mortality from other causes. Adjusting for individual-and household-level characteristics (including age, geographical factors, socioeconomic characteristics and measures of prepandemic health) reduced the estimated differences in COVID-19 mortality rates between elderly adults living in different types households (Table 2) . Adjusting for socioeconomic factors, such as IMD decile, household deprivation, educational attainment, social grade and household tenancy, had the strongest effect of the hazard ratio ( Table 7 in Appendix 1). However, even after adjusting for these characteristics, living in a multi-generational household, especially with children, remained associated with an increased risk of COVID-19-related death. Compared to living with another elderly adult aged 65 years or above, the rate of COVID-19-related death was 1.16 (95% CI 1.07-1.25) and 1.21 (95% CI 1.06-1.38) times greater for elderly women living in a multi-generational household without and with children, respectively. For elderly men, after adjusting for individual and household characteristics, living in a multi-generational household without children was associated with a 1.07 (95% CI 1.01-1.13) times greater risk of COVID-19-related death and living in a multi-generational household with children with a 1.17 (95% CI 1.06-1.30) times greater risk.

Using the ASMRs for women living with another elderly adult as the baseline risk, these hazard ratios imply that living in a multi-generational household without children would increase the ASMRs from 236 to 274 death per 100,000 people, and living in a multigenerational household with children to 286 death per 100,000 people. For men, living in a multigenerational household without children is expected to increase the ASMRs from 465 to 498 death per 100,000 people, and living in a multi-generational household with children to 544 death per 100,000 people.

The rate of COVID-19-related death was also 1.13 (95% CI 1.09-1.17) times greater for elderly men living alone than for those living with another older adult, and 1.10 (95% CI 1.06-1.15) times greater for elderly women. The results were similar in the sensitivity analyses using different definitions of household composition ( Table 8 in Appendix 1).

We tested the proportional hazard assumption by testing for the independence between the scaled Schoenfeld residuals and time-at-risk. For men, the test failed to reject the independence for the exposure (p ¼ 0.112 for men), suggesting that the proportional hazard assumption was unlikely to be violated. However, for women, the test suggested that the proportional hazard assumption was likely to be violated (p ¼ 0.002), However, as shown by the smoothed Schoenfeld residuals for each group ( Figure 6 in Appendix 1), the deviation from the estimated loghazard ratio is small, and the 95% confidence intervals around the smoothed Schoenfeld residuals always included the estimated log-hazard ratio, suggesting that violation of the proportional hazard assumption was unlikely to substantially affect our main results.

Living in a multi-generational household as a mediator for the disparity in COVID-19 death between ethnic groups Living in a multi-generational household did not explain much of the difference in COVID-19 mortality rates among elderly men. It accounted for 6.9% (95% CI 2.7-11.3) of the excess risk of COVID-19 mortality for men of Pakistani background, 6.0% (95% CI 2.3-9.9) for men of Indian background and 6.3% (95% CI 2.6-10.2) for men of Pakistani . Decomposition of odds ratios for COVID-19 mortality among elderly adults (aged ! 65 years) across ethnic groups, stratified by sex. Note: The overall height of the bar corresponds to the odds ratio (OR), relative to the White population, based on a logistic regression model adjusted for age. Error bars are 95% confidence intervals. The proportion of the age-adjusted ORs explained by living in a multi-generational household were calculated through a mediation analysis. The unexplained part corresponds to the ORs from a model adjusted for age, geographical factors (region, population density, urban/rural classification), socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy), and health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts, number of hospital admissions, total days spent in hospital).

background. It did explain a larger proportion the difference in risk between elderly women of South Asian background and White elderly women Living in a multi-generational household accounted for 11.6% (95% CI 5.9-18.0) of the excess risk of COVID-19 mortality for women of Indian background, and 11.5% (95% CI 4.0-20.0) and 11.1% (95% CI 4.8-17.5) for women of Bangladeshi and Pakistani background (see Table 9 in Appendix 1 for full results). The results were similar in the sensitivity analyses using different definitions of household composition ( Table 10 in Appendix 1).

This paper makes two contributions to the research on COVID-19. First, we find that among elderly adults, household composition is associated with COVID-19 mortality, even after adjusting for a range of sociodemographic factors and measures of health. Our results indicate that compared to those living in a two older adult household, elderly adults, especially women, living in a multi-generational household are at greater risk of COVID-19 death. Living alone is also associated with elevated COVID-19 mortality. Second, we find that living in a multi-generational household explains around 11% of the excess COVID-19 mortality risk for women of South Asian background, but little for men or people from other ethnic groups.

Our results are consistent with emerging evidence that household size is associated with the risk of infection, 16, 17 and that elderly adults tend to be at greater risk of household transmission. 18, 19 Older people living in large household tend to live in multi-generational households, co-habiting with younger adults and children. There is some evidence that, among elderly adults, living with dependent children is not strongly associated with the risk of COVID-19 infection or adverse outcomes. 20 While our results indicate that elderly adults living in a multi-generational household are at greater risk of COVID-19 death compared to those living with another older adult, we find little difference in risk between older people living in a multi-generational household with or without young children. Our results are consistent with a recent study using Swedish data, which show that for elderly adults, living with a working-age adult was associated with increased COVID-19 mortality. 21 Several studies have analysed ethnic differences in COVID-19 infection and mortality. 3,4,6-8 Although we focus on elderly adults only, we find that almost all ethnic minority groups were at higher risk of COVID-19 deaths compared to the White population, and that the differences were attenuated once we adjusted for a range of geographical factors, sociodemographic characteristics and co-morbidities. We improve the existing evidence on ethnic inequalities in COVID-19 mortality by using a causal mediation approach to quantify the importance of living in a multi-generational household.

Our results suggest that older people are placed at increased exposure to infection by living with younger adults rather than young children. After adjusting for confounding factors, we find that the risk of COVID-19 death is similar among elderly adults living in a household with young children and those living in a household with younger adults only. This could be due to schools having been closed for a substantial proportion of the period at risk. The increased risk is likely to be driven by co-residing with younger adults, who have a higher risk of infection than older people. 17 Younger adults are likely to be at increased risk of exposure because of work, as evidence suggests that in England people who are working were at greater risk of infection compared to people not in employment, especially if they were working in patient or client-facing occupations. 17, 22, 23 The interaction between job characteristics and household composition is likely to account for some of the elevated mortality among ethnic minority groups in the USA. 24 Elderly adults living by themselves were also found to be at greater risk of COVID-19 death than those living with another older adult. During the COVID-19 pandemic, older people living alone were more likely to have received help from carers, including informal helpers, than people living with another older adult. 25 These frequent contacts with people from different households could increase the risk of being exposed to the virus.

We find that living in a multi-generational household explains around 11% of the excess COVID-19 mortality risk for women of South Asian background, but little for men, despite a similar proportion of them living in a multi-generational household. Women spend more time at home than men and still do the majority of unpaid housework, 26 which could increase the risk of household transmission. However, further research would be needed to understand the mechanisms driving our results.

The primary strength of our study lies in the use of a unique linked population-level dataset which combine the 2011 Census with death registration data and hospital records. Unlike data based solely on health records, our study dataset contains a broad range of information on demographic, socio-economic, and household characteristics, including occupation. Unlike sample survey data, it contains millions of observations covering the entire population of interest, allowing us to examine both the association between household composition and COVID-19 mortality and also whether living in a multi-generational household explains some of the disparity in COVID-19 mortality between ethnic groups. We were able to examine differences between disaggregated ethnic minority groupings rather than high-level categories of South Asian, Black and Other.

The main limitation of our study is that household composition is likely to be imprecisely measured. While household composition is based on a detailed and accurate measurement taken in 2011, we could only identify changes since then due to death of household members or a move to a care home. While we took several steps to limit the measurement error, such as focusing on elderly adults, including only adults aged 25 or over and children aged 0 to 9 at the time of the census in our definition of household composition, our household composition measure may not reflect current living circumstances of everybody in our population of interest. To mitigate concerns about measurement error, we showed that our results are robust to using different definitions of household composition. Nonetheless, measurement error is likely to attenuate the explanatory power of household composition in our models. In addition, while we have used a causal mediation approach, our analysis remains based on observational data and therefore residual confounding is likely. Another limitation is that our statistical approach assumes that the effect of living in different types of household composition is the same across ethnic groups.

Elderly adults living in multi-generational households are at elevated risk of experiencing harms from COVID-19 compared to elderly adults living with people of the same age. However, there has been little focus on implementing effective interventions (such as creating plans to effectively isolate and improving ventilation within the home) to reduce transmission risk within the household. 27 Relevant public health interventions should be directed at communities where multi-generational households are highly prevalent. Living in a multi-generational household explains some of the excess COVID-19 mortality risk for women of South Asian background, but little for men or people from other ethnic groups. Further research is needed to explain the difference in COVID-19 mortality between ethnic groups. 

The Average Causal Mediated Effect (ACME) of living in a multi-generational household for ethnic group k compared to the White group, (k) is estimated as:

where MGH indicates if individual i lives in a multigenerational household, Ethnic is the ethnic group and X is a vector of factors likely to confound the relationship between the mediator and the outcome. X includes geographical factors (region, population density, urban/rural classification), socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy) and health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts, number of hospital admissions, total days spent in hospital).

To estimate each component of ðkÞ, we use predicted probabilities based on logistic regression models.

The total estimated difference in probability of COVID-19 death between ethnic group k and the White ethnic group is given by

The proportion of the difference in the probability to die from COVID-19 between ethnic groups that is mediated by living in a multi-generational household is given by the ACME, ðkÞ, as a proportion of the total effect ðkÞ. We then used this proportion to decompose the age-adjusted odds ratios. Household deprivation is defined according to four dimensions: employment (at least one household member is unemployed or long-term sick, excluding full-time students); education (no household members have at least Level 2 education, and no one aged 16-18 years is a full-time student); health and disability (at least one household member reported their health as being 'bad'/'very bad' or has a long-term health problem); and housing (the household's accommodation is overcrowded, with an occupancy rating À1 or less, or is in a shared dwelling, or has no central heating). For people aged over 75 years at time of the 2011 Census, approximated social grade was imputed based on household tenure. Key worker type is defined based on the occupation and industry code. 'Exposure to disease' and 'proximity to others' are derived from the O*NET database, which collects a range of information about individuals' working conditions and day-to-day tasks of their job. To calculate the proximity and exposure measures, the questions asked were as follows: (i) How physically close to other people are you when you perform your current job? (ii) How often does your current job require that you be exposed to diseases or infection? Scores ranging from 0 (no exposure) to 100 (maximum exposure) were calculated based on these questions using methods previously described by the ONS. Note: Def A1: we derived household composition in 2020 based on the number of adults aged 20 years (instead of 25 years); Def A2: multigenerational household as household with someone aged 65 years or over in 2020 co-resided with at least one other adult aged more than 15 years (instead of 20 years) younger. Hazard ratios compared to living in a household with one other older adult. Fully adjusted Cox regression models include geographical factors (region, population density, urban/rural classification), ethnicity, socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy), health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts, number of hospital admissions, total days spent in hospital), a measure for overcrowding and property type. Figure 6 . Smoothed Schoenfeld residuals for household composition from the fully adjusted model, stratified by sex. Note: Fully adjusted Cox regression models include geographical factors (region, population density, urban/rural classification), ethnicity, socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy), health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts, number of hospital admissions, total days spent in hospital), a measure for overcrowding and property type. Dotted line shows the log-hazard ratio from the model. Residuals are smoothed with a generalised additive model. Confidence intervals are at the 95% level. Note: Main definition of multi-generational household: someone aged 65 years or over on 2 March in 2020 co-resided with at least one other adult aged more than 20 years younger (and at least 25 in 2011), or with at least one child. Def A1: we derived household composition in 2020 based on the number of adults aged 20 years (instead of 25 years); Def A2: multi-generational household as household with someone aged 65 years or over in 2020 co-resided with at least one other adult aged more than 15 years (instead of 20 years) younger. Proportion of difference in COVID-19 mortality between ethnic group mediated by living in a multi-generational household is estimated as the Average Causal Mediated Effect (ACME) as a proportion of the age-adjusted difference in the probability of COVID-19 death. The ACME is derived based on models that adjust for geographical factors (region, population density, urban/rural classification), socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy) and health (self-reported health and disability from the Census, pre-existing conditions based on hospital contacts).

ONS. Updating Ethnic Contrasts in Deaths Involving the Coronavirus (COVID-19), England and Wales: Deaths Occurring

Disparities in the Risk and Outcomes of COVID-19

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

Ethnic and socioeconomic differences in SARS-CoV-2 infection: prospective cohort study using UK Biobank

The impact of ethnicity on clinical outcomes in COVID-19: a systematic review

Ethnic minority groups in England and Wales -factors affecting the size and timing of elevated COVID-19 mortality: a retrospective cohort study linking Census and death records

OpenSAFELY: factors associated with COVID-19 death in 17 million patients

Ethnic differences in COVID-19 infection, hospitalisation, and mortality: an OpenSAFELY analysis of 17 million adults in England

ONS. Ethnicity Facts and Figures: Families and households

Living with parents or grandparents increases social capital and survival

Social capital, social cohesion, and health

Descriptive analysis of social determinant factors in urban communities affected by COVID-19

Covid-19: breaking the chain of household transmission

A general approach to causal mediation analysis

Identification, inference and sensitivity analysis for causal mediation effects

Socio-demographic heterogeneity in the prevalence of COVID-19 during lockdown is associated with ethnicity and household size: results from an observational cohort study

Antibody prevalence for SARS-CoV-2 in England following first peak of the pandemic: REACT2 study in 100,000 adults

Household secondary attack rate of COVID-19 and associated determinants in Guangzhou, China: a retrospective cohort study

The household secondary attack rate of SARS-CoV-2: a rapid review

Association between living with children and outcomes from COVID-19: an OpenSAFELY cohort study of 12 million adults in England

Residential context and COVID-19 mortality among adults aged 70 years and older in Stockholm: a population-based, observational study using individual-level data

Occupation and risk of severe COVID-19: prospective cohort study of 120,075 UK Biobank participants

Risk of hospital admission with coronavirus disease 2019 in healthcare workers and their households: nationwide linkage cohort study

COVID-19 and racial/ ethnic disparities in health risk, employment, and household composition

Older and 'staying at home'during lockdown: informal care receipt during the COVID-19 pandemic amongst people aged 70 and over in the UK

Gender divisions of paid and unpaid work in contemporary UK couples

Reducing risks from coronavirus transmission in the home-the role of viral load

Department for Communities and Local Government. The English Indices of Deprivation 2015. Department for Communities and Local Government

Welsh Index of Multiple Deprivation (WIMD) 2014. Welsh Government

Socioeconomic characteristics: ethnicity, IMD decile, household deprivation, educational attainment, social grade, household tenancy; pre-pandemic health: self-reported health and disability from the Census, pre-existing conditions based on hospital contacts, number of hospital admissions, total days spent in hospital; overcrowding; Housing: overcrowding and property type. Occupational exposure: Maximum of 'exposure to disease' score within each household

Proportion of difference in COVID-19 mortality rates between ethnic groups mediated by living in a multi-generational household, stratified by sex

Multi-gen with young children Multi-gen without young children Multi-gen (total)

Note: Proportion of difference in COVID-19 mortality between ethnic group mediated by living in a multi-generational household is estimated as the Average Causal Mediated Effect (ACME) as a proportion of the age-adjusted difference in the probability of COVID-19 death. The ACME is derived based on models that adjust for geographical factors (region, population density, urban/rural classification), socioeconomic characteristics (IMD decile, household deprivation, educational attainment, social grade, household tenancy), and health: (self-reported health and disability from the Census, preexisting conditions based on hospital contacts)

Acknowledgments: None.