Settings and study population
The present study is a registry-based study with data from the SSIA, the Swedish National Board of Health and Welfare, and Statistics Sweden, based on the unique Swedish personal identification number. It presents data on sick leave due to Covid-19 that started between March 1 and August 31, 2020, with follow-up for 4 months. A patient has been involved in the study process as a partner in research.
Inclusion criteria for the study population were the following: being registered with sickness benefits due to Covid-19, which were defined as the International Statistical Classification of Diseases (ICD) [17] code U07, starting within the inclusion period. Codes U (U00-U49) are used by WHO for provisional assignment of new diseases of uncertain etiology (Please see this link: https://icd.who.int/browse10/2019/en#/U07.1).
The SSIA is the public authority in Sweden that makes decisions on sick leave and pays sickness benefits, and it provided the study population and sick leave data for the present study. All working people in Sweden are eligible for sickness benefits from the SSIA if deemed to have reduced work ability due to sickness, regardless of citizenship or place of residence. Furthermore, self-employment, parental leave, and unemployment (after previous employment) also entitle one to sickness benefits. The employer provides sick pay during the first 2 weeks of sickness absence; thereafter, the SSIA pays sickness benefits. If a person is unemployed, sickness benefits are provided by the SSIA from the start. In the present study, receiving sickness benefits regardless of amount is defined as sick leave.
The National Board of Health and Welfare provided data on date of death during the study period from the Cause of Death Register, which records all cases of death that have been registered in Sweden. Data from the National Patient Register, which includes all inpatient care in Sweden, were used to investigate hospital stay due to Covid-19.
Statistics Sweden holds registries of all people registered in Sweden. Statistics Sweden provided data on sociodemographic variables to the study.
Variables
The sick-leave period in the present study includes at least one period of sickness benefits due to Covid-19 diagnosis. Other predefined related diagnoses were merged with the Covid-19 sick leave if the gap of non-registration between sick leaves was ≤2 weeks. The related diagnoses are shown in additional Table 1 and included, for example, unspecified virus infections, fever, and postviral fatigue syndrome, but also a second sick-leave registration for Covid-19 diagnosis. In cases of sick pay provided by the employer, these were also merged with the sick-leave period. The sick-leave period could comprise a maximum of 122 days (4 months of follow-up). For predictive analyses, sick leave was dichotomised into sick leave ≥1 month (≥30 days): yes/no, and sick leave for long Covid (≥ 12 weeks, in line with the WHO definition [16]): yes/no.
Sick leave prior to Covid-19 was defined as either being on sick leave for one period of at least 28 days between March 1, 2019, and the date of first Covid-19 sick leave registration, or being on sick leave at least six times during the same period of time.
The SSIA register includes the employment status for which a person receives the sick leave. The types of employment comprise employment (including parental leave, and combined employment and self-employment), self-employment, and unemployment (including studies).
Educational level was categorised as primary school (≤ 9 years), secondary school (10–12 years), short university education (13–14 years), or long university education (≥15 years). The educational level registered in 2019 was used. The income variable was the disposable income for each person during 2019, presented in thousands of Swedish krona SEK (1 Euro = 10.16 SEK, March 4, 2021). Income was categorised in tertiles of low, medium, and high income. Country of birth was presented as Sweden, Nordic countries except for Sweden, European countries except for the Nordic countries, and Countries outside of Europe. For marital status in 2020, married and registered partnership were both classified as married. Likewise, divorced and widow/widowed meant a change from either marriage or registered partnership. Inpatient care due to Covid-19 was classified as being registered with a hospital stay of > 1 day with a registration of any of the Covid-19 diagnoses U07. The primary diagnoses are presented in additional Table 2 in the cases where U07 was not registered as the primary diagnosis.
Statistical methods
The data were processed and analysed using IBM SPSS Statistics 25. Data are presented as number and percentage (%), mean and standard deviation (SD), and median and interquartile range (IQR). The significance level (alpha) was set to 5%. To compare differences between groups, the Mann-Whitney U test and Fisher exact test were used.
To graphically present cumulative incidence of sick leave over time, Kaplan-Meier curves were used. There was no censoring, as cases of death during the study period were treated with a worst-case-scenario approach and were set at sick leave for the maximum number of days, and there was no other loss to follow-up.
Multiple logistic regression was used for predictive analysis. The regression analyses were performed on three separate populations: the total study population, the participants receiving inpatient care for Covid-19, and the participants not receiving inpatient care. Two different dependent variables were used in different models: sick leave ≥1 month and sick leave for long Covid. The independent variables were chosen based on clinical and theoretical reasoning: age, sex, educational level, income, country of birth, sick leave prior to Covid-19, employment status, marital status, and inpatient care due to Covid-19. The results are presented as odds ratio (OR), 95% confidence interval (95% CI), and p-value in forest plots. The ordinal or continuous independent variables were tested for multicollinearity using the Spearman correlation test, with values < 0.3 being acceptable. To test the accuracy of the models, receiver operating characteristics (ROC) curves were constructed. An area under the ROC curve > 0.70 indicate acceptable accuracy [18].