Data sources
Stockholm’s administrative organization (Region Stockholm), responsible for all healthcare within the region, manages VAL (Swedish: Vård Analys Lager, the Stockholm Regional Healthcare Data Warehouse), a data warehouse of healthcare utilization. VAL contains complete hospital inpatient and hospital outpatient data, and primary care information, including consultations and diagnoses at the individual level. The coverage in VAL for inpatient care is over 99% [8] and the validity of the diagnostic coding is 85–95%, depending on the diagnosis [9]. The VAL database includes linkage to the Swedish National Tax Agency (death dates) and was also further linked to national population registers held by Statistics Sweden. This included the Total Population Register “RTB” (sex, age, birth country) [10]. RTB also contains Household Register data on geographical information, including street address and apartment data (size of household) and the Integrated Database For Labor Market Research ”LISA” (educational level, income, and occupation) [11]. In addition, we used data from SmiNet which is a national electronic surveillance system for reporting communicable diseases [12]. Since February 1, 2020, it is mandatory for the Swedish laboratories to report all PCR-confirmed cases of COVID-19 to SmiNet. All register linkage used the unique personal identity number given to each Swedish citizen [13].
Study population
The study population consisted of all individuals 18 years and older, residing in Stockholm County during a calendar year from March 1st, 2020, to February 28th, 2021, based on data from the Population Register [10]. Individuals permanently staying in nursing homes were excluded since they were mainly treated for COVID-19 in their facilities and hence did not contribute risk for hospitalization. Also, those with home-care services were excluded due to increased risk of infection, uncoupled to sociodemographic status.
Variables used as exposure and confounders in statistical models
Educational level: separated into low (pre-secondary education), medium (secondary-), and higher education (post-secondary education) based on LISA data.
Income: household disposable income (LISA data) was separated into quintiles, from the 20% with the lowest income to the 20% with the highest income defined as income including welfare, after taxation.
Work: using the “standard for Swedish occupational classifications” (SSYK), based on the “international standard classification of occupations” (ISCO), we dichotomized individuals based on the ability to work from home or not. Additionally, we have analyzed healthcare workers, and adults not working (full-time students, unemployed, on long-term sick leave, or retired) separately. These classifications were made by individually assessing the work characteristics of the different occupations.
Living area: the greater Stockholm area was divided into 164 neighborhoods with an average of 14,000 inhabitants. They were ranked after death due to COVID-19 per 10,000 people (excluding those living at nursing homes) and then grouped into quintiles. In other analyses, we introduce them separately, as baseline hazards.
Living condition: measured as the size of household (the number of people in the household).
Country of origin: data on country and region of origin is available, but for these analyses, we divided subjects into those born in Sweden and not born in Sweden.
Co-morbidities: the following ICD codes from the VAL database were chosen, based on previous risk factor publications [14, 15] heart failure (I50), ischemic heart disease (I20-25), diabetes (E10-14), obesity (E66), chronic kidney disease (N18), chronic obstructive pulmonary disease (J44), cancer (C00-97) and liver disease (I85–I85.9; I98.2; K79–K71, K71.3–K77.8; R16–R18.9; Z52.6; Z94.4). All individuals with one of these diagnoses within the last 5 years (or 2 years for cancer) or one hospitalization due to cardiovascular disease have been classified as having a co-morbidity.
Outcomes
The primary outcome was 30-day all-cause mortality after laboratory-confirmed COVID-19 infection. The secondary outcome was hospitalization with confirmed COVID-19 infection. Hospitalization as outcome was verified via SmiNet and validated if inpatient treatment included the emergency ICD10 codes for COVID-19 (issued by the WHO): U07.1–U07.2 as the main diagnosis.
Observation period
The observation period ran from the 1st of March 2020 to the 28th of February 2021. Follow-up ended at loss to follow-up (emigration from the Stockholm Region), end of study, or the date of outcomes.
Statistical analyses
Multivariate logistic regression model was used, fitted for each outcome. The modeling strategy consisted of analyzing a selection of individual covariates first, followed by the same individual covariates and a selection of area-level covariates. To avoid collinearity in the latter part of the modeling, each area-level covariate was included one at a time, always including the whole set of individual covariates. Area-level covariates outside of area percentages of children, percentages of elderly, density, and inhabitants born outside of Sweden were included in a principal component analysis to generate a composite neighborhood deprivation score (NDS), divided into three levels, from least (NDS 1) to most deprived (NDS 3).
In Additional file 1: Table S1 associations were estimated using a Cox proportional hazards model with income as exposure and adjusted for confounder variables as categorical variables and stratified for area effects, thus allowing for different baseline hazards in each area. In this analysis, individuals were censored at the date of death from other causes, emigration from the region, or at the end of follow-up, whichever came first. The potential confounders were added sequentially in order to show the confounding impact of different domains of the sociodemographic factors.
Ethical approval
The study has been approved by the Regional Ethical Review Board, Stockholm (2021–00810). All data were analyzed in a pseudonymized format and confidentiality was always maintained. Reporting follows the STrengthening the Reporting of OBservational studies in Epidemiology and the REporting of studies Conducted using Observational Routinely-collected health Data statements [16, 17].
Data sharing
Swedish privacy law prohibits us from making registered data publicly available.