SARS-CoV-2 infections in migrants and the role of household overcrowding: a causal mediation analysis of Virus Watch data

Background Migrants are over-represented in SARS-CoV-2 infections globally; however, evidence is limited for migrants in England and Wales. Household overcrowding is a risk factor for SARS-CoV-2 infection, with migrants more likely to live in overcrowded households than UK-born individuals. We aimed to estimate the total effect of migration status on SARS-CoV-2 infection and to what extent household overcrowding mediated this effect. Methods We included a subcohort of individuals from the Virus Watch prospective cohort study during the second SARS-CoV-2 wave (1 September 2020–30 April 2021) who were aged ≥18 years, self-reported the number of rooms in their household and had no evidence of SARS-CoV-2 infection pre-September 2020. We estimated total, indirect and direct effects using Buis’ logistic decomposition regression controlling for age, sex, ethnicity, clinical vulnerability, occupation, income and whether they lived with children. Results In total, 23 478 individuals were included. 9.07% (187/2062) of migrants had evidence of infection during the study period vs 6.27% (1342/21 416) of UK-born individuals. Migrants had 22% higher odds of infection during the second wave (total effect; OR 1.22, 95% CI 1.01 to 1.47). Household overcrowding accounted for approximately 36% (95% CI −4% to 77%) of these increased odds (indirect effect, OR 1.07, 95% CI 1.03 to 1.12; proportion accounted for: indirect effect on log odds scale/total effect on log odds scale=0.36). Conclusion Migrants had higher odds of SARS-CoV-2 infection during the second wave compared with UK-born individuals and household overcrowding explained 36% of these increased odds. Policy interventions to reduce household overcrowding for migrants are needed as part of efforts to tackle health inequalities during the pandemic and beyond.

The direct effect was the association between migration status and SARS-CoV-2 infection due to all other causes not accounted for in the model.It was derived by comparing the proportion of migrants with evidence of infection with the counterfactual proportion of UK-born individuals who would have had evidence of infection if they had the same distribution of household overcrowding status as the migrant individuals i.e., household overcrowding status was kept constant.To estimate the direct effect, confounders of both the exposure-mediator, mediator-outcome and exposureoutcome mediator must be adjusted for.Based on the DAG in Supplementary Figure 1, the minimally sufficient adjustment set for the direct effect comprises age at baseline, sex at birth, ethnicity, clinical vulnerability, total household income at baseline and occupation.
The indirect effect was the association between migration status and SARS-CoV-2 infection due to differences in household overcrowding status.It is derived by comparing the proportion of migrants who had evidence of SARS-CoV-2 infection with the counterfactual proportion of migrants who would have had evidence of infection if they had the same distribution of household overcrowding status as UK-born participants, while adjusting for the same confounders that were adjusted for when estimating the direct effect.The total effect is estimated by summing the indirect and direct coefficients on the log scale.
Model checks for multicollinearity, particularly in the case of migration status and ethnicity, were conducted using logistic regression controlled for age, sex and ethnicity with the glm command in R version 4.1.2with the family set to binomial and a logit link.The generalised variance inflation factor (GVIF) was calculated using the vif command in the car package [4] with adjustment for the number of coefficients in the variables [5].GVIFs adjusted for the number of coefficients in the variables were all below 10 (Table below), which suggests a lack of evidence of multicollinearity [6].
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Supplementary Box 4: Variables used for multiple imputation
The following variables were included as predictor variables: • Index of multiple deprivation An observational study can generally be conceptualised as a conditionally randomised experiment if four key identifiability conditions are met[1].We outline these conditions and how they relate to our study in the table below.
However, a study in the US found higher rates of COVID-19 mortality in urban areas with more immigrants who had traditional/multigenerational family structures.Being a migrant is associated with an increased odds of infection and as SARS-CoV-2 is infectious, living in an area with more migrants could mean increased risk of infection for UK-born too.Consistency (i.e. the effect of an exposure is the same for all individuals who receive that exposure)YesMigration status is based on whether someone reports a UK or non-UK country of birth.It is not a composite variable.

Table : GVIFs adjusted for the number of coefficients in the variable for a regression model representing the total effect
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) J Epidemiol Community Health doi: 10.1136/jech-2022-220251 -655.

Table 2 :
Whether an individual had a positive test in the study period BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) Cohort demographic and household characteristics BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)