A population-based cohort study of socio-demographic risk factors for COVID-19 deaths in Sweden

The study addresses which socio-demographic groups are most at risk of death from COVID-19, overcoming a paucity of high-quality microlevel data linking deaths to detailed background information. Early pandemic evidence, largely based on aggregated data or clinical cohorts, suggested higher risks among men, older individuals, racial/ethnic minorities, and those of lower socioeconomic status, despite narratives that the virus “does not discriminate.” Responding to calls for detailed linked data, the authors leverage Swedish national registers with complete population coverage to examine how COVID-19 mortality varies by age, sex, civil status, income, education, country of birth, and region of residence. Sweden’s comparatively less-restrictive approach and high per-capita mortality during the study period provide a context of international relevance for identifying vulnerable populations and informing policy as countries ease restrictions.

Initial studies and reports indicated elevated COVID-19 severity and mortality among men, the elderly, racial/ethnic minorities, and those in lower socioeconomic positions, but were limited by aggregated data and limited sociodemographic detail beyond age and sex. The paper situates its contribution against this backdrop, noting the need for microlevel analyses linking mortality to socio-demographic registries. The authors reference evidence on broader mortality inequalities by socioeconomic status and discuss the typical healthy migrant advantage observed for non-COVID causes, providing a contrast to their findings for COVID-19 mortality among immigrants.

Design: Nationwide, population-based cohort using linked Swedish administrative registers with complete coverage of residents aged 20+ living in Sweden in December 2019. Follow-up: March 13, 2020 (first recorded COVID-19 death) to May 7, 2020. Final cohort: 7,775,064 individuals after excluding those not resident in Sweden in the prior two years (N=147,557) and those with missing country of birth (N=8,370) or income (N=12,862). Total exposure: 1,189,484 person-years (mean follow-up 56 days). Outcomes: Deaths from all causes during follow-up, with COVID-19 deaths identified via the Swedish Cause of Death Register using emergency ICD codes U07.1, U07.2, or B34.2. Among 17,181 total deaths, 3,126 were COVID-19 related (2,988 underlying cause; 138 contributing). Exposures/covariates: Age (underlying time scale in Cox models: biological age in months), sex, civil status (married, never married, widowed, divorced), individual net income (tertiles across all adult residents), educational attainment (primary ≤9 years, secondary 10–12 years, postsecondary >12 years, missing), country of birth (Sweden; high-income countries [HIC]; low- and middle-income countries [LMIC] from Northern Africa and Middle East [MENA]; other LMIC), region (Stockholm County vs rest of Sweden). Data sources include the Total Population Register, Cause of Death Register, and LISA database. Statistical analysis: Cause-specific Cox proportional hazards models estimated hazard ratios (HRs) for risk of death from COVID-19 (censoring at non-COVID deaths) and from all other causes (censoring at COVID-19 death). Two main model sets: stratified by sex; stratified by age segment (working ages 20–65; retirement ages 66+), with continuous age within segments. Proportional hazards assumption tested via Schoenfeld residuals and log–log plots; some violations for sex, income, and education addressed with stratified models and robustness checks. Correlation matrices examined for multicollinearity (highest coefficient correlation R=0.3 between lowest income tertile and primary education). Sensitivity analyses included alternative tie handling and logistic regression models by gender (results similar to Cox). Analyses adjusted for Stockholm residence and conducted in Stata 16.

• Study period outcomes: 17,181 total deaths; 3,126 COVID-19 deaths in 1,189,484 person-years.
• Age and sex: Strong age gradient; men had higher COVID-19 mortality than women.
• Civil status: Compared to married, never married, divorced, and widowed had ~1.5–2.0 times higher COVID-19 mortality for both sexes.
• Education (adjusted for income): Relative to postsecondary education:
  • Men: secondary HR 1.25 (95% CI 1.09–1.43); primary HR 1.24 (1.07–1.43).
  • Women: secondary HR 1.38 (1.17–1.62); primary HR 1.51 (1.28–1.79).
• Income (adjusted for education): Relative to highest tertile:
  • Men: tertile 2 HR 1.51 (1.29–1.78); tertile 1 HR 1.76 (1.49–2.09).
  • Women: tertile 2 HR 0.99 (0.78–1.25); tertile 1 HR 1.26 (1.01–1.58).
• Country of birth (vs Sweden):
  • HIC: men HR 1.19 (1.01–1.39); women HR 1.08 (0.92–1.26).
  • LMIC other: men HR 2.20 (1.81–2.69); women HR 1.45 (1.12–1.90).
  • LMIC MENA: men HR 3.13 (2.51–3.90); women HR 2.09 (1.52–2.89).
• Region: Living in Stockholm County associated with ~4.5-fold higher COVID-19 mortality versus rest of Sweden (e.g., men HR 4.51; 95% CI 4.08–5.00). By contrast, for all other causes, Stockholm showed only modestly higher mortality (men HR 1.24; 95% CI 1.16–1.31).
• Age-segmented results:
  • Working ages (20–65): Stronger socioeconomic gradients. Income tertile 1 vs 3 HR 5.40 (3.51–8.35); primary education HR 2.62 (1.65–4.16); secondary HR 2.22 (1.46–3.37). Civil status: significant excess only for never married HR 1.48 (1.04–2.10).
  • Retirement ages (66+): Socioeconomic differentials less pronounced; civil status differences stronger: widowed HR 1.48 (1.34–1.63), divorced HR 1.62 (1.46–1.80), never married HR 1.65 (1.44–1.89), all vs married. Immigrant excess mortality ~2x vs Swedish-born in both segments. Stockholm excess similar across age segments.
• Comparison with all-cause mortality: Many sociodemographic gradients align between COVID-19 and other causes, but country of birth and region differ markedly. For all-cause mortality, immigrants often show a healthy migrant advantage (e.g., LMIC MENA men HR 0.82; 95% CI 0.68–0.99), unlike COVID-19 where immigrant groups have higher risk.

The study demonstrates clear socio-demographic inequalities in COVID-19 mortality in Sweden. Beyond age, being male, unmarried, lower educated, and lower income are independent risk factors. The excess risk among immigrants from LMIC, particularly MENA, persists after adjusting for socioeconomic status and residence, contrasting with their typical advantage in all-cause mortality. Socioeconomic factors are more salient among working-age adults, while marital status (and associated living arrangements and care dependencies) plays a larger role among retirees. Geographic exposure, notably residence in Stockholm during the study period, had a substantial effect specific to COVID-19. These findings directly address the research question by quantifying the independent associations of key socio-demographic variables with COVID-19 mortality, informing targeted public health strategies and resource allocation, including enhanced outreach and healthcare support for disadvantaged and immigrant communities.

Using complete population registers, the study quantifies socio-demographic risk factors for COVID-19 mortality in Sweden, showing higher risks for men, the unmarried, lower income and lower educated individuals, and immigrants from LMIC, and a strong regional effect for Stockholm. The distinct pattern for immigrants relative to all-cause mortality underscores COVID-19’s interaction with social determinants and exposure pathways. Policy implications include prioritizing prevention and healthcare resources for disadvantaged groups and immigrant-dense communities, and addressing socioeconomic vulnerabilities, especially among working-age populations. Future research should elucidate mechanisms behind socioeconomic and migrant disadvantages (e.g., comorbidities, occupations, housing density, neighborhood characteristics, transmission vs severity pathways) and investigate the role of elderly living arrangements, care homes, and home-help services in shaping mortality risks.

• Short follow-up (<2 months) limits time-to-event leverage and may bias Cox estimates if spatiotemporal spread is unaccounted; however, logistic regressions yielded similar results.
• Possible misclassification of COVID-19 as underlying vs contributing cause; preliminary coding.
• All-cause mortality comparisons occur in the presence of the pandemic and may capture indirect effects (e.g., care-seeking changes), potentially underestimating the true pandemic mortality burden when focusing only on confirmed COVID-19 deaths.
• Lack of emigration data for early 2020 could slightly underestimate mortality rates; likely minor impact given age distribution of emigration.
• Proportional hazards assumption violated for some covariates (sex, income, education) in certain models; addressed via stratification and robustness checks, but residual bias cannot be fully excluded.