Socioeconomic inequalities in risk of infection with SARS-CoV-2 delta and omicron variants in the UK, 2020-22: analysis of the longitudinal COVID-19 Infection Survey

Population and databases

We used the UK COVID-19 Infection Survey to examine the potential differential effects of the risk of infection with the SARS-CoV-2 virus in different occupations. The COVID-19 Infection Survey offered a unique opportunity to examine the longitudinal association between occupation sectors and the SARS-CoV-2 virus during the covid-19 pandemic. The COVID-19 Infection Survey, conducted by the Office for National Statistics and the University of Oxford, was a nationwide longitudinal survey to monitor SARS-CoV-2 infection in the community and is currently closed.13

Participants in the survey were recruited from randomly selected households to reflect the UK population (England, Scotland, Wales, and Northern Ireland). Sampling was grouped geographically to ensure that people from all local areas of the UK were represented; the response rate was 13.3% and the sample size was adjusted to account for varying response rates. The COVID-19 Infection Survey was an open cohort study where new participants were recruited over the study period and longitudinal data were collected from consenting existing participants. Each new participant was surveyed for five weeks initially, and then monthly.14 Data collected at each visit included nose and throat swabs, a blood sample, and questionnaire data. The data in this analysis were collected by the COVID-19 Infection Survey from 26 April 2020 to 31 January 2022. Only working aged adults (16-64 years) were eligible for the analysis.15 The website of the Office for National Statistics has more details on the design and methodology of the COVID-19 Infection Survey.16

Exposure

The main exposure of interest in this study was deprivation, measured by the index of multiple deprivation.17 18 The index of multiple deprivation is the official measure of relative deprivation in the UK and is calculated from 39 separate indicators, organised across seven distinct domains (income, employment, health deprivation and disability, education and skills training, crime, barriers to housing and services, and living environment). The index of multiple deprivation is an area level marker of deprivation based on the geographical location of residence and calculated for every Lower layer Super Output Area (LSOA), with each area's deprivation level ranked based on their relative scores. LSOAs comprise 400-1200 households, each with a resident population of 1000-3000 people. The index of multiple deprivation does not provide individual level estimates of deprivation for a person. The index of multiple deprivation was based on the residential address of participants in the study. For the purposes of this study, we used the index of multiple deprivation deciles as our marker of deprivation, which ranged from the most deprived 10% to the least deprived 10%.

Outcome

Variants of the SARS-CoV-2 virus of concern were limited to the delta and omicron variants because these were the dominant variants during the period of this analysis.19 We divided the cohort into a delta cohort (2 July 2020 to 19 December 2021) and an omicron cohort (on or after 20 December 2021), as implemented in previously published reports from the Office for National Statistics.20 SARS-CoV-2 diagnoses compatible with the delta variant were defined based on the gene patterns OR+S or N+S or OR+N+S, with a cycle threshold <30 in the delta cohort. Similarly, diagnoses compatible with the omicron variant were based on the gene pattern of OR+N with a cycle threshold <30 in the omicron cohort, as used in previous reports.21 The outcome variable was coded as a binary variable, denoting whether the gene pattern was compatible with the delta variant during the delta period. Similarly, during the omicron period, the outcome variable was coded as a binary variable if the gene pattern was compatible with the omicron variant. This approach meant that people infected with other variants were coded as not having the outcome. A polymerase chain reaction test result was used to collect the sample for testing.

Covariates

Our analysis included self-reported sociodemographic and clinical data collected from the COVID-19 Infection Survey, including age, sex, ethnic group, comorbid conditions, urban or rural home address, household size, patient or client facing job, and time of year (as quarters of the year). Age in years was calculated at the time of the participant's first visit. Sex was self-reported as either male or female sex. Ethnic group was self-reported based on the 18 UK category ethnic classifications and, for the purposes of this analysis, categorised as white or non-white participants because of the low numbers of individuals from ethnic minority groups. Comorbid conditions were measured by a binary variable for reporting of having any physical or mental conditions or illnesses lasting or expected to last ≥12 months. Household size was categorised into three groups: one person household, double person household, and household of three or more persons.

Participants were asked about their employment status, and those who were employed or self-employed were asked to select their employment category from this list: teaching and education; healthcare; social care; transport (including storage and logistics); retail (including wholesale); hospitality (eg, hotel, restaurant, or cafe); food production and agriculture (including farming); personal services (eg, hairdressers and tattooists); information technology and communication; financial services (including insurance); manufacturing or construction; civil service or local government; armed forces; arts, entertainment, or recreation; and other (online supplemental figure S1). We used the most recent valid employment sector as reported by participants.

The COVID-19 Infection Survey also collected information on whether the individual's current job regularly involved direct (in-person) contact with patients or clients. To adjust for this information in the regression models, we further categorised participants into patient or client facing workers and non-patient or client facing workers. Calendar time was divided into quarters of the year to account for any seasonal fluctuations in the incidence of infections. These covariates were selected a priori based on expert opinion and an extensive literature review.

Statistical analysis

We compared baseline characteristics for the delta and omicron cohorts, with data presented as median (interquartile range) or number (percentage), unless otherwise stated. We also presented baseline characteristics for the delta and omicron cohorts by the most deprived group (group 1; using the index of multiple deprivation divided by deciles) and least deprived group (group 10).

We calculated the crude incidence rate, assuming a Poisson distribution, to examine the association between index of multiple deprivation and having a positive test result for the delta or omicron variant of the virus after adjusting for covariates. Multivariable Poisson regression models were fitted to estimate the adjusted incidence rate ratio after adjusting for age, sex, ethnic group, comorbid conditions, urban or rural residence, household size, patient or client facing job, and time (as quarters of the year). For the delta cohort, person months were calculated as the time between an individual's first registered study visit (index date) on or after 2 July 2020 and the earlier of the event date (ie, the study visit date where they first reported a positive swab test result compatible with the delta variant) and their final study visit or study end date (19 December 2021). The corresponding dates in the omicron cohort were 20 December 2021 (study start date) and the earlier of the outcome (event date) and the study end date (31 January 2022). We used the log of the follow-up time as the offset term, and robust variance for estimation of confidence intervals. We also included two separate interaction terms for index of multiple deprivation by occupation and index of multiple deprivation by sex in our models.

For analysis of the delta cohort, we included samples from 2 July 2020 to 19 December 2021, and removed participants with no information on health condition or ethnic group. For analysis of the omicron cohort, we included samples on or after 20 December 2021 to 31 January 2022. Figure 1 shows the details of the selection of participants.

Request permission

We did not apply any imputation for covariate missingness because of the small amount of missing covariate data (figure 1). Because we adjusted for a range of covariates in the model, we restricted our subgroup analyses (ie, by occupation sectors) when the outcome events were ≥50 in each of the 10 equal groups of index of multiple deprivation to ensure the statistical stability of our estimates. This approach meant that we were only able to use the work sectors, healthcare, manufacturing or construction, retail, and teaching and education. In a sensitivity analysis, we also used a multilevel Poisson regression model allowing for random effects at the country level (to account for possible clustering at the country level), and a second order polynomial for age and time variables (to allow for any potential non-linear associations).

Patient and public involvement

Patients were involved in the design stage of this research, with feedback implemented in the design and setup of the Office for National Statistics COVID-19 Infection Survey. The public can ask questions and provide feedback to the COVID-19 Infection Survey inbox. We thank the participants in the COVID-19 Infection Survey for taking part in the survey. Findings from this analysis and further analysis based on the survey have been shared with the public through publications from the Office for National Statistics.

Incidence rate

The crude incidence rate for the presence of the delta and omicron variants of the SARS-CoV-2 virus were higher in the most deprived group (delta cohort 4.33 per 1000 person months, 95% confidence interval 4.09 to 4.58; omicron cohort 76.67 per 1000 person months, 71.60 to 82.11) than in the least deprived group (3.18, 3.05 to 3.31 and 54.52, 51.93 to 57.24, respectively). We found no appreciable difference in incidence between men and women (online supplemental table S2 and table S3).

Across occupations, we found differences in incidence between the most and least deprived groups. In the least deprived group, the highest incidence rate in the delta cohort was in the teaching sector (4.07, 95% confidence interval 3.65 to 4.53); in the most deprived group, the highest incidence was in the manufacturing or construction sector (5.41, 4.40 to 6.65) (table 2). In the least deprived group, the highest incidence of the omicron variant was in the manufacturing or construction sector (71.89, 61.36 to 84.23); in the most deprived group, the highest incidence was in the healthcare sector (97.47, 78.29 to 121.35) (table 2).

Adjusted incidence rate ratio

The adjusted incidence rate ratio for a positive test result for the delta and omicron variants gradually increased with increasing levels of deprivation. We found the highest incidence rate ratio in the most deprived group compared with the least deprived group (delta cohort adjusted incidence rate ratio 1.37, 95% confidence interval 1.29 to 1.47; omicron cohort adjusted incidence rate ratio 1.34, 1.24 to 1.46) (online supplemental figure S2 and figure S3). Similar patterns were found when the cohorts were examined by sex (figure 2).

Request permission

Request permission

The adjusted incidence rate ratios for a positive test result for the delta variant were higher (≥1.50) in the healthcare and manufacturing or construction sectors, when comparing the most deprived group with the least deprived group (figure 3); for the omicron variant, the adjusted incidence rate ratios were higher (≥1.43) in the healthcare and teaching and education sectors (figure 3) when comparing the most deprived group with the least deprived group. Results from the sensitivity analyses did not differ from the main results (online supplemental table S4 and table S5).

Main findings

Based on this large, nationally representative UK community based survey, we found that a positive test result for the delta and omicron variants of the SARS-CoV-2 virus was associated with area level deprivation, with a higher incidence and higher incidence rate ratio in the most deprived group compared with the least deprived group. Results did not differ between men and women. Similar patterns were seen for some occupations, where a positive test result for the SARS-CoV-2 delta and omicron variants was higher in the most deprived group than in the least deprived group for the healthcare, manufacturing or construction, and teaching and education sectors.

Comparison with previous literature

Our findings are in agreement with previous evidence indicating that individuals from more deprived areas had a higher risk of SARS-CoV-2 infection and long covid.4 12 22 Our investigation, however, also assessed whether specific variants of the virus had different incidence rates across occupations. Previous evidence indicated that occupational exposure to the virus might account for some incidences of SARS-CoV-2 infection, especially in healthcare and in people facing occupations.4 9 23–25 In our analysis, we have provided more detail, however, by showing that the manufacturing or construction sector had the highest incidence of the delta variant whereas the healthcare sector had the highest incidence of the omicron variant.

Although previous research has reported that occupation and deprivation level are independently associated with the risk of SARS-CoV-2 infection,22 23 we have extended this observation by quantifying individual and combined associations, and showed that the pattern of increased incidence of infection in individuals from more deprived areas was generally consistent across all occupations. We found, however, that socioeconomic inequality differed by occupation. The risk of infection with the delta variant in the most deprived group compared with the least deprived group was highest in the healthcare and manufacturing or construction sectors, whereas the risk of infection with the omicron variant between these two deprivation groups was highest in the healthcare and teaching sectors. Possible explanations for the almost dose-response way in which the incidence rate ratio for a positive test result increased with increasing levels of deprivation in both men and women might be that hierarchies are formed in the workplace, with individuals in lower status roles at increased risk (eg, in patient or public facing occupations or not having the opportunity to work remotely).26

Examining intersectionality between sociodemographic factors is important because it allows us to assess whether risk decreases, remains the same, or increases across more granular social categories (eg, deprivation and occupation), rather than within independent categories leaning towards a single axis framework (eg, deprivation or occupation).27 Also, recent reports from the UK suggested that in some sections of the healthcare workforce, shortages or poor fitting personal protective equipment was a problem, with people from ethnic minority groups or from more deprived backgrounds being most affected.25 28 29 Healthcare workers with less access to personal protective equipment were reported to be more likely to have a positive test result for SARS-CoV-2.25 The increased risk in manufacturing or construction and teaching and education sectors in our analysis, however, could be related to other factors, such as whether participants were more likely to be infected or tested, or both, depending on policies specific to their occupation, wider government policies on covid-19, and the timing of covid-19 restrictions.

Strengths and limitations

Our study had several strengths. We used data from a nationally representative community based survey and adjusted for a range of covariates in our models to estimate the independent effects of the index of multiple deprivation on our outcomes. We also examined intersectional inequality by examining inequality by sex, social deprivation, and occupation. The COVID-19 Infection Survey provided uniquely rich, contemporaneous, and longitudinal data on occupation and employment, job status, covid-19 status, and deprivation level.

Our study had some limitations. Comorbid conditions were self-reported, and were not validated against an objective diagnosis. We assumed that the potential measurement errors, however, would be non-differential for the index of multiple deprivation group. The index of multiple deprivation is an ecological area level measure of deprivation and, therefore, the findings might not be applicable at the individual level. Also, the number of infections in each deprivation group for some occupations were low and hence these occupations were excluded to ensure the statistical stability of our estimates.

Data on vaccination status were not available in this study, which is relevant to susceptibility to the SARS-CoV-2 virus after 8 December 2020 (date of first vaccination in the UK). This limitation is important because vaccination has been shown to reduce transmission.30 Also, some sectors were prioritised (eg, healthcare staff) for vaccination at the beginning of the vaccine rollout, which could have biased our results, while some sociodemographic groups also reported a lower uptake of the vaccine.31 32 However, the effect of vaccination should be non-differential for all individuals who received a vaccine during our study, while accounting for time will take into consideration potential changes in vaccination uptake.

Because our outcome was specific for the variants of concern (delta and omicron), infection rates of other variants might have affected the estimates of the incidence rate ratio in our analyses. Hence our incidence rate ratios might have been even higher if we had included all infections in our outcome, suggesting possible underestimation of our results. Our analysis, however, could not determine associations between deprivation or occupation and less prevalent variants of the SARS-CoV-2 virus circulating at the time of our study.

An observational analysis cannot establish causality and our study also lacked precise data on lockdowns or whether individuals were working from home. These factors might have varied by occupation and individual situations. Nevertheless, some degree of residual confounding might still exist.

Potential non-response bias could cause uncertainty in the data, which might not be fully mitigated by the methods used to adjust for this bias in the original survey design. The sampling method ensured that the UK population was well represented, however, and a higher number of households were invited to take part in the survey to account for attrition and non-response bias. Although the COVID-19 Infection Survey sample was nationally representative, the response rate was relatively low. Once recruited, however, the attrition rate was generally low; based on a definition of formally withdrawing from the study or not attending the three most recently scheduled follow-up visits, the attrition rate among enrolled survey participants was <1% in 2021.33 Nevertheless, participants in the most deprived groups might have been less likely to take covid-19 tests. If this is true, our results are conservative estimates of the true incidence and rate ratios.

Lastly, we could not determine if the source of infection was at a person's workplace (eg, people could have been working from home). Therefore, the risk estimates reported in this study are a weighted average for the whole occupational sector (ie, those who worked from home and those who worked on site).

Conclusions

Our analysis showed that differences existed between occupations when the risk of a positive test result for the SARS-CoV-2 virus was compared between the most and least deprived groups. Also, when the results were not grouped by occupation, we found a pattern of increasing incidence and rate ratios for SARS-CoV-2 from low to high deprivation groups, with findings similar in men and women. These results will help inform employers and health policy in conducting evidence based risk assessments and in allocating potentially limited resources to those at greatest risk of covid-19 in different occupation sectors.