Risk factors for COVID-19 diagnosis, hospitalization, and subsequent all-cause mortality in Sweden: a nationwide study

The Public Health Agency of Sweden provided data from its SmiNet database on all cases of COVID-19 confirmed in Sweden until mid-September 2020. Reporting confirmed cases to SmiNet is required by law. We did not have information on the methods of testing used to diagnose the COVID-19 cases. Only the first date of diagnosis or positive test for each individual was provided.

We obtained a control population by requesting that Statistics Sweden (the agency of government statistics) randomly sample 5 non-diagnosed individuals for each COVID-19 case. Each control was residing in Sweden on January 1, 2020, and was alive on January 31, 2020. No matching was performed. Statistics Sweden also provided registry data on sex, year and month of birth, country of birth, highest level of completed education in 2018, disposable family income in 2018, home municipality (on December 31, 2019), and a pseudo-anonymized household identifier, indicating which persons lived at the same street address (on December 31, 2019).

The Swedish National Board of Health and Welfare provided registry data on deaths, diagnoses, hospitalizations, prescription medication use, residence in long-term care facility, and use of homemaker service. All deaths in Sweden are recorded in the Cause of Death Register [13]. Diagnoses and hospitalizations are registered in the National Patient Register, to which all health care providers have been required to report hospitalizations since 1987 and physician visits in secondary care (i.e., non-primary care) since 2001 [14]. Each hospitalization or visit is assigned a main diagnosis and one or more secondary diagnoses, which indicate the purpose(s) of the hospitalization or visit. Medical and surgical procedures are also recorded. The National Patient Register has been validated, showing positive predictive values of more than 90% for most diagnoses, although sensitivity is often lower [14]. For cancer, we also selected diagnoses recorded in the Swedish Cancer Register from 1964 through 2018, so as to obtain a longer lookback period than is possible with the National Patient Register. Health-care providers have been required to report new cancer diagnoses to the Swedish Cancer Register since 1958 [15]. Data on prescription medication use were obtained from the Prescribed Drug Registry, which records all prescription medications collected at pharmacies in Sweden since July 2005 [16]. Data on residence in a long-term care facility and use of homemaker service were available from the Register for Care and Services for the Elderly and for Persons with Impairments According to the Social Services Act. Local governments are required to report to this register, whose quality is considered adequate for publication by the National Board of Health and Welfare [17]. Homemaker services are domestic services provided to persons (primarily older persons) who live at home but need help with shopping, cleaning, meal preparation, and similar tasks. Local governments are responsible for determining eligibility for these services, although the services may be provided by private businesses.

Information about ICU care was obtained from the Swedish Intensive Care Registry. In 2019, 83 of Sweden’s 84 ICUs reported to this registry [18]. The data from the registries were linked using the Personal Identification Number that is issued to each resident of Sweden at the time of birth or immigration. We obtained data files in which Personal Identification Numbers had been replaced by pseudo-anonymized identifiers, generated by Statistics Sweden. The study was approved by the Swedish Ethical Review Authority, who waived the requirement of obtaining informed consent (number 2020–02552).

We investigated three different outcomes related to COVID-19 infection, each reflecting an increased severity of infection: COVID-19 diagnosis without hospitalization; non-ICU hospitalization with confirmed COVID-19 as the main diagnosis (International Classification of Diseases, 10^th Revision, Swedish Version, [ICD-10-SE] Code: U071); and ICU hospitalization for confirmed COVID-19 (ICD-10-SE: U071). ICU hospitalizations were traced in SmiNet and non-ICU hospitalizations were traced in the National Patient Register until October 1, 2020. These groups, along with the control group, were followed-up for all-cause mortality until October 1, 2020.

Table 1 shows the demographic variables, comorbidities, and prescription medications that were included in the analysis as potential risk factors for COVID-19. Although pneumonia is sometimes a complication of COVID-19, we included history of pneumonia as a potential risk factor because it may be a marker of frailty or impaired immune system. Variable definitions are available in Supplemental Tables 1 and 2. A composite variable was also constructed, which indicated the presence of at least one of the comorbidities or medications.

Multinomial logistic regression was used to estimate odds ratios for COVID-19 diagnosis or ICU or non-ICU hospitalization. Cox regression was used to estimate hazard ratios for mortality. In each model, persons were excluded if they had missing data on at least one of the included variables.

For non-hospitalized COVID-19 cases, baseline date was defined as the date of testing or diagnosis, whichever came first or was available. If both dates were unavailable, the case was excluded from the analysis. Baseline date for hospitalized cases was the date of ICU or non-ICU admission. For controls, we randomly sampled a date from the baseline dates among cases. Controls who had died by their sampled date were excluded from the analysis.

A subgroup analysis was conducted to examine the prevalence of risk factors for COVID-19 and subsequent mortality in the age group 0–19 years. We did not estimate odds ratios and hazard ratios in this age group because of the small number of deaths and hospitalized COVID-19 cases.

The incidence of COVID-19 was estimated by dividing the number of cases by the size of the Swedish population (in total and by age group and sex) on December 31, 2019 [19]. The Kaplan–Meier method was used to estimate mortality rates, with confidence intervals based on the log(survival) transformation. All statistical analyses were performed in RStudio (R version 3.6.2). Statistical significance was determined by 95% confidence intervals not overlapping 1.

In the analysis of risk factors for COVID-19 diagnosis and hospitalization, 1264 controls were excluded because they died prior to their assigned baseline date. Furthermore, 2411 non-hospitalized COVID-19 cases were excluded because no date of testing or diagnosis was available.

Risk factors for COVID-19 diagnosis and hospitalization are presented in Tables 1 and 2. The majority of non-hospitalized COVID-19 cases were women, but the majority of hospitalized cases were men. After adjustment for other risk factors, the odds of non-ICU hospitalization increased until the age group 40–49 years, after which there was no observable trend. Both the adjusted and unadjusted odds of ICU admission increased until the age group 60–69 years and decreased thereafter.

Being born in Sweden was associated with lower odds of COVID-19 diagnosis and hospitalization (Table 2). Higher levels of education were associated with higher odds of COVID-19 diagnosis but with lower odds of hospitalization. After adjustment for other risk factors, family disposable income was positively associated with COVID-19 diagnosis, but there was no clear association with hospitalization.

After adjustment for other risk factors, residence in a long-term care facility was associated with increased odds of diagnosis and, albeit to a lesser degree, non-ICU hospitalization (Table 2). Residence in a long-term care facility was associated with lower odds of ICU admission. Use of homemaker service was associated with both COVID-19 diagnosis and hospitalization (ICU and non-ICU).

Approximately 90% of ICU and non-ICU hospitalized patients had at least one of the investigated comorbidities or medications, and this was associated with more than twice the odds of ICU and non-ICU hospitalization after adjustment for demographic factors (Tables 1 and 2). The two comorbidities most strongly associated with ICU and non-ICU hospitalization after adjustment for other risk factors were diabetes and Down syndrome. Other comorbidities associated with both ICU and non-ICU hospitalization were hypertension, immune disorder, asthma, influenza, and pneumonia. Comorbidities significantly associated with increased odds of either ICU or non-ICU admission, but not with both, were autoimmune disease, chronic obstructive pulmonary disease (COPD), renal failure/chronic kidney disease, sepsis, solid organ transplantation, and liver disease. Human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS) and glomerular disease were non-significantly associated with both ICU and non-ICU hospitalization. Alcohol intoxication was associated with lower odds of COVID-19 diagnosis and hospitalization after adjustment for other risk factors.

Cancer was not associated with increased odds of hospitalization after adjustment for other risk factors (Table 2). The unadjusted association was explained by sex, age group, hypertension, cardiovascular disease, corticosteroid use, and opioid use (Supplemental Table 3). Cancer in the past year, however, was associated with COVID-19 diagnosis and non-ICU hospitalization, but not with ICU hospitalization, after adjustment for all variables in Table 2 other than Any Comorbidity/Medication (adjusted odds ratio for diagnosis 1.17, 1.09–1.26; adjusted odds ratio for non-ICU hospitalization 1.30, 95% CI 1.20–1.42; adjusted odds ratio for ICU admission 0.90, 95% CI 0.72–1.12).

Cardiovascular disease was associated with slightly increased odds of non-ICU hospitalization, but with lower odds of ICU admission, after adjustment for other risk factors (Table 2). The unadjusted association of cardiovascular disease with ICU admission was explained by age group, sex, hypertension, and diabetes (Supplemental Table 4).

Antithrombotics, proton-pump inhibitors, corticosteroids, and opioids were associated with both ICU and non-ICU hospitalization for COVID-19 after adjustment other risk factors (Table 2). Antivirals and lipid-modifying agents were not associated with either ICU or non-ICU hospitalization after controlling for other risk factors. The unadjusted association for lipid-modifying agents was explained by age, sex, hypertension, diabetes, and cardiovascular disease (Supplemental Table 5). Immunosuppressants were associated with increased odds of ICU hospitalization and slightly increased odds of non-ICU hospitalization, although the associations were not significant.

In the analysis of all-cause mortality, an additional 45 non-hospitalized COVID-19 cases were excluded because they died prior to the date of testing or diagnosis. The median number of days from baseline until the end date of the analysis (October 1, 2020) was 119 days (interquartile range, 98–155 days).

Excess mortality was observed in hospitalized and non-hospitalized COVID-19 cases after adjustment for other risk factors (Fig. 1, Table 3, Table 4). The only demographic factor not associated with all-cause mortality was birth in Sweden (Table 4). Most comorbidities were associated with mortality in excess of any risk conferred by COVID-19 (Table 4).

Male sex was not associated with COVID-19 diagnosis or ICU or non-ICU hospitalization in persons aged 0–19 years (Supplemental Table 6). Hospitalized COVID-19 cases were less often born in Sweden, and they more often had hypertension, cancer, diabetes, COPD, sepsis, pneumonia, and corticosteroid use than did general-population controls and non-hospitalized COVID-19 cases. Asthma and proton-pump inhibitors were more common in non-ICU hospitalized and diagnosis-only COVID-19 cases than in general-population controls. There were too few deaths to evaluate excess mortality among COVID-19 cases aged 0–19 years (Supplemental Table 7).