Study design and population

This study was designed as a cohort study of participants from the Danish National Health Survey, which was conducted in five regional stratified random samples and one national random sample [14]. Based on the availability of the applied survey items, we included 33,285 survey participants from one region collected during January 30 to April 30, 2013 (Central Denmark Region) and 129,319 survey participants from four regions collected during February 3 to April 30, 2017 (Central Denmark Region, North Denmark Region, Region Zealand, and Capital Region of Denmark). In total, the initial study population encompassed 162,604 individuals with a response rate of 57.5%. Subsequently, we applied unique identification numbers from the Danish Civil Registration System [15] to link with individual-level information in the national registers on public healthcare utilization, wage income, transfer payments, and covariates. We included 3,426 responses (2.1%) which were provided by individuals who by chance participated in both 2013 and 2017 but excluded 107 individuals (0.07%) with no register linkage at the time of survey participation, leaving a potential sample of 162,497 individuals. We subsequently excluded 1,071 (0.7%) individuals who emigrated and 2,456 (1.5%) individuals who died before the end of the calendar year following survey participation (2014 or 2018) for which their annual measures were incomplete. The analytical sample thus comprised 158,970 individuals. A flowchart illustrating the definition of the study population is provided in S1 Fig. To address non-response and selection probabilities, we applied inverse probability weights that have been made available with the survey data and are constructed by Statistics Denmark with a model-based calibration approach based on information from the national registers [14].

Ethics

The study was registered with the Danish Data Protection Agency at Aarhus University (No 2016-051-000001-2587) and approved by Statistics Denmark and the Danish Health Data Authority. According to Danish law, informed consent or ethical approval is not required for register-based studies in Denmark. For survey participants, information about the survey was provided to potential participants in writing. All survey participants were informed that participation was voluntary and that their survey data would be linked to the registers for research purposes. The respondents’ full or partial completion of the survey constituted implied consent. Linking the survey data and register data was done by Statistics Denmark. All data were pseudonymized and not recognizable at an individual level and analysed on the secure platform of Statistics Denmark.

Social connections

Loneliness was assessed in the Danish National Health Survey with the Danish version of the Three-Item Loneliness Scale [16,17], providing a score ranging from 3 to 9 with higher scores indicating greater loneliness. In 2017, the third item was slightly rephrased to enhance correspondence with the definition of loneliness, but the scale has demonstrated good internal consistency at both time points [18]. Corresponding to the most conservative dichotomisation of severe loneliness applied in the literature [19], we classified a score of 7 or higher as indicating loneliness. Social contacts were assessed in the survey by quantifying different areas of social contact with inspiration from the Berkman-Syme Social Network Index [20]. Four indicators of limited social contact were used to measure social isolation providing a score ranging from 0 to 4: whether an individual i) was living alone, ii) was unemployed and not enrolled in education, iii) had less than monthly contact with friends, and iv) had less than monthly contact with family outside of the household. We classified a score of 3 or higher as indicating social isolation. Social support was assessed in the survey as perceived emotional support with inspiration from the MOS Social Support Instrument [21]. The single-item “Do you have someone to talk to if you have problems or need for support?” with four response options: “Yes, always”; “Yes, mostly”; “Yes, sometimes”; and “No, never or almost never” was applied. We classified answers in the two last-mentioned response options as indicating low social support. To capture both structural and functional aspects of social disconnectedness, we additionally constructed a composite measure of either loneliness, social isolation, or low social support.

Economic outcomes

Healthcare costs included public reimbursements of healthcare services delivered by general practitioners and medical specialists from the Danish National Health Service Register [22], public cost averages for hospital-based services from the DRG National Patient Register and the DRG Psychiatric Patient Register, and public subsidies for redeemed prescriptions from the Danish National Prescription Registry [23]. For each individual, total healthcare costs were calculated by summing the costs of primary healthcare services, hospital-based services, and subsidies for redeemed prescriptions. Wage income and transfer payments were assessed using information from the Income Statistics Register [24] on the annual primary taxable wage income (i.e., excluding transfer payments) and a range of transfer payments (e.g., unemployment benefits, subsidized employment, and disability pension) which are primarily publicly funded. All economic outcomes were assessed in the calendar year following survey participation (2014 or 2018), and outcomes from 2014 were weighted to represent values in 2018 of Danish kroner using the gross domestic product deflator from the World Bank. Subsequently, we converted from Danish kroner to euros using the average exchange rate from the European Central Bank in 2018 (1 euro equalling 7.47 Danish kroner). Further data documentation is available at Open Science Framework (https://osf.io/f2vm5).

Covariates

Sex (registered legal sex), age when participating in the Danish National Health Survey, and country of birth (Denmark and Greenland vs. abroad) were obtained from the Danish Civil Registration System [15]. Information on hospital-diagnosed mental health conditions in 18 years preceding survey participation was obtained from the Danish Psychiatric Central Research Register [25], which since Jan 1, 1995, contains data on both outpatient services in psychiatric departments and admissions to inpatient and emergency psychiatric departments with the International Classification of Diseases, 10th revision (ICD-10) [26]. Organic disorders (ICD-10: F00–F09) and intellectual disabilities (ICD-10: F70–F79) were not included in the applied definition due to considerations regarding distinctive characteristics of these patient groups, but all other psychiatric diagnoses (ICD-10: F10–F69 & F80–F99) were included.

Statistical analysis

To avoid introducing selection bias [27], multiple imputation by chained equations were conducted to include 19,675 (12.4%) individuals with partially missing survey data (S1 Text). Means, standard deviations (SDs), and proportions were computed to describe the baseline characteristics of the cohort. All estimates were calculated with inverse probability of participation weights and pooled multiple imputed data using Rubin’s Rules.

All analyses were performed separately for each indicator of social disconnectedness (loneliness, social isolation, low social support, and the composite measure). We applied linear regression models with Taylor-linearized variance estimation and 95% confidence intervals (CIs) to compare each economic outcome between individuals who were socially disconnected at baseline and individuals who were not. Adjustments were made for sex, age, year of survey participation, and country of birth. The total of each economic outcome was estimated applying marginal standardization to average over the distribution of the covariates for individuals who were socially disconnected. Additionally, we extrapolated the results to the entire population in the corresponding Danish regions with the inverse probability weights and calculated the attributable share of total healthcare costs and transfer payments, respectively. As the first sensitivity analysis, we applied quantile regression to explore whether economic differences according to social disconnectedness were consistently found when modelling the first, second, and third quartile distribution of each economic outcome instead of the mean [28]. We also repeated the main analysis stratified based on sex and age groups in 10-year intervals. Furthermore, we analysed interaction with pre-existing hospital-diagnosed mental health conditions by estimating the deviation from additivity, also known as the interaction contrast. As the second sensitivity analysis, we repeated the interaction analysis with a broader definition of pre-existing mental health conditions additionally including self-reported information, redeemed prescriptions of psychopharmaceuticals, and consultations with private practicing psychiatrists. Additionally, we repeated the interaction analysis stratified by sex. Details are provided in S1 Text.

Statistical analyses were conducted in Stata version 18.0 using the svy and mi suite of commands on register data provided by Statistics Denmark between June 3 and September 8, 2022. A preregistered analysis plan and the code used for data management and statistical analysis are available at Open Science Framework (https://osf.io/mb2ua).

Comparison to prior findings

Our findings of excess healthcare costs and transfer payments for individuals who were socially disconnected are in line with several previous studies [7–9]. Nevertheless, a Dutch population-based study involving 341,376 individuals found, contrary to our results, reduced costs of somatic inpatient and outpatient services among individuals who were lonely [7]. A major difference in this comparison is that our findings are based on models with basic demographic adjustments to obtain descriptive estimates, whereas this study included adjustments for several variables that we conceptualized as potential mediators (i.e., socioeconomic and lifestyle-related factors, self-perceived health, and psychological distress). In a Swedish population-based study based on 53,920 individuals aged 20–64 years, they reported a higher risk of receiving disability pension for individuals who were socially isolated, except for men aged 40–64 years [8]. Similarly, a Dutch study predicted that fewer workers with high social support would take long-term sickness leave [9]. To our knowledge, our study is the first to explore the interaction with mental health conditions on economic outcomes.

Strength and limitations

This study benefitted from utilizing rich survey data, thus allowing us to assess three distinct aspects of social disconnectedness with a validated measure of loneliness [17] and a measure of social isolation created with inspiration from a well-known index [20]. Another strength is the assessment of several economic outcomes—a range of healthcare costs, wage income, and transfer payments—using high-quality administrative register data with nearly complete records. The study also benefitted from studying a population-based sample allowing us to explore variation across subgroups, while applying inverse probability weights that have been shown to account well for primary healthcare utilization [29] and mental health conditions [30]. Not least, multiple imputation was applied to avoid introducing selection bias when dealing with missing survey data.

However, the study also has important limitations. First, the study design does not allow inferences regarding whether social disconnectedness is a cause or an effect of the observed economic differences, or whether these differences are in fact due to other confounding factors. Second, the measure of social support and social isolation have not been validated, and the measure of social isolation did not include associational activities and voluntary work, whereas the measure of social support focused solely on perceived emotional support [31]. Third, despite the application of inverse probability weights based on administrative registers, the representativeness of the study participants cannot be confirmed with regard to factors not recorded in the registers. Notably, our finding that lower wage income was partially counterbalanced by excess transfer payments illustrates the relatively high level of welfare benefits available in Denmark. Accordingly, these findings may have limited generalizability for countries with different healthcare systems, economic structures, and social policies.

Potential explanations

Our findings of excess healthcare costs point towards three different reasonings. First, the higher healthcare costs may be attributed to a higher prevalence of medical conditions among individuals who are socially disconnected, consistent with numerous studies [2–5]. Second, individuals who are socially disconnected may face a greater severity of medical conditions, as observed for depression [32] and multimorbidity [33]. Third, the results could be explained by higher cost of treatments for comparable medical conditions among individuals who are socially disconnected (e.g., due to complications or readmissions). For instance, an association between low social support and poor self-management behaviours has been identified [34]. Our findings of lower wage income and excess transfer payments may be attributed to both the proportion of individuals who are employed and receiving transfer payments, as well as the average wage income among those employed and the average transfer payments among recipients. The former indicates differences due to health disparities, as discussed above, while the latter could reflect broader social inequalities, potentially caused by a wide range of factors.