Introduction

Subjective well-being has been recognized as an important predictor of health outcomes and health behaviours, and it has emerged as a significant point of discussion in the areas of health policy and public health intervention1,2,3,4,5,6. One dimension of well-being is one’s sense of belonging to community, which has been characterized as a shared sense of connection and identity to a community7 as well as attachment to and comfort within a community8. The concept of community belonging has been of longstanding research interest9 and has been discussed in the broader research context of place attachment, rootedness and social capital10. Prior research suggests that weaker sense of belonging to one’s community is associated with negative health indicators and outcomes, including self-rated health11,12, self-rated mental health13, unfavourable changes in health-related behaviour14, and unmet healthcare needs15.

Studies have demonstrated associations between various measures of social connection (e.g., social support, loneliness, social isolation) and downstream health outcomes, including cardiovascular disease16, prolonged hospitalization17, hospital-treated infections18, and mortality6,19,20. An important health outcome that is largely unexplored with respect to community belonging is avoidable hospitalization, which is often considered to be that related to ambulatory care sensitive conditions (ACSCs) that should not require acute hospitalization if appropriately managed in outpatient settings21.

Community belonging is a multi-dimensional concept that draws upon theories of social connections and place attachment10 from a variety of disciplines, including but not limited to sociology, psychology, and health promotion. Notably, Berkman et al. developed an overarching conceptual model that links social networks and health, and drew upon Émile Durkheim’s theories of social integration, John Bowlby’s attachment theory, and social network theory22. Their model suggests that a combination of social processes and psychobiological processes are involved, and can be extended to explain the relationship between community belonging and health-related outcomes, such as avoidable hospitalization. Conceptually, having a weak sense of community belonging may trigger social processes that influence health by: (1) reducing the community contacts available for receiving health-related information; (2) lowering transmission and reinforcement of healthy social norms, behaviours, and attitudes, leading to unfavourable behaviours such as delayed help-seeking and poor treatment adherence, and; (3) decreasing access to community services and resources23. These processes could also influence more proximal pathways, such as increasing one’s chronic stress responses. Ultimately, the combination of these social dynamics stemming from a lack of community belonging could elevate the likelihood of preventable illness requiring hospitalization11,14,24,25.

Aside from a study that found an association between weak community belonging and lower odds of diabetes-related hospitalization among Canadians with diabetes23, no studies have evaluated the relationship between community belonging and avoidable hospitalizations in the general Canadian population, and none have incorporated sex-stratified analyses. A focus on a broader range of upstream psychosocial factors has the potential to inform decision-making for our health and social systems to generate large-scale improvements in health outcomes across populations. The examination of this relationship is of particular interest for Canada, where the universal healthcare system minimizes financial barriers to healthcare but non-financial (e.g., psycho-social) barriers still exist26. The objective of this study is to evaluate the association between community belonging and avoidable hospitalization using a population-based cohort of adult men and women residing in Canada.

Methods

Study design and sample

We conducted a population-based cohort study of respondents of the Canadian Community Health Survey (CCHS)27, and included survey respondents aged 18–74 years at the time of interview from eight CCHS cycles between 2000 and 2014. Those under 18 were excluded because risk factors for pediatric preventable hospitalizations tend to differ from adult preventable hospitalizations. Consistent with the definition of ACSC hospitalization used in this study21, individuals aged 75 years and older were excluded due to the high prevalence of multiple comorbidities in older adults, posing challenges in distinguishing preventable hospitalizations from those that are not28. Respondents were excluded if they: (1) had a death date preceding interview date, which indicates either an erroneous date of death or linkage); (2) resided in Quebec at the time of interview, as the province of Quebec does not report to the hospitalization database used in this study29, and/or (3) were pregnant at the time of interview, due to the potential for deviation of health and behavioural characteristics (such as BMI and alcohol consumption30) from typical baseline values during pregnancy. We adhered to the principles outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement31. This study was approved by the University of Toronto’s Research Ethics Board (#41965).

Data sources

The CCHS is a cross-sectional survey administered by Statistics Canada, and contains data on community belonging as well as demographic, socioeconomic, behavioural, and health-related factors27. The survey collects health-related data from Canadians 12 years and older and is representative of 98% of the population. Data collection is completed every two years until 2007 and annually thereafter. Individuals are excluded from the survey if they: live on reserves or other Aboriginal settlements, are institutionalized, are a full-time Canadian Forces member or live in the Quebec health regions of Région du Nunavik and Région des Terres-Cries-de-la-Baie-James. The survey data were individually linked to the Discharge Abstract Database (DAD) up to 31 March 2018, using a generalized record linkage software (G-link) that utilizes deterministic and probabilistic linkage methods32. The DAD is maintained by the Canadian Institute for Health Information (CIHI) and collects information regarding hospital discharges in all provinces and territories in Canada except Quebec33.

Measures

Exposure: community belonging

Community belonging was captured by the CCHS, which includes a question asking respondents to rate their sense of belonging to their local community on a 4-point Likert scale, ranging from very strong, somewhat strong, somewhat weak, to very weak. This single-item measure has been used in prior studies and has been found to be significantly associated with social capital as well as health and mental health outcomes12,14,34. We re-categorized community belonging to a 3-level variable: very strong, intermediate (which includes somewhat strong or somewhat weak), very weak. This categorization focuses on those with the strongest and weakest reported belonging compared to those who felt less strongly about their belonging and were similar. As a sensitivity analysis, we also modelled the 4-level and 2-level categories.

Outcome: avoidable hospitalizations

The DAD was used to obtain information on avoidable hospitalizations from the CCHS interview date to the end of follow-up (31 March 2018). Avoidable hospitalizations were defined as acute care hospitalizations of individuals younger than 75 years old for an ACSC, which include the following: grand mal status and other epileptic convulsions, chronic lower respiratory diseases, asthma, diabetes, heart failure and pulmonary edema, hypertension, and angina21. ACSCs were identified using International Classifications of Disease 9th version (ICD-9) and International Classifications of Disease 10th version (ICD-10) codes (Supplementary Table 1).

Potential confounders

Potential confounders were selected a priori based on support from prior literature (Supplementary Table 2) and the Andersen Newman Framework for determinants of medical care35. The directed acyclic graph (DAG) can be found in the Supplementary Fig. 1. Overall, full adjustment included demographic and socioeconomic factors (age, ethnicity, urban/rural classification, newcomer status, marital status, household income quintile, household education level) as well as health and behavioural factors (alcohol consumption, smoking, physical activity, body mass index, and the presence of four major and common chronic conditions including chronic obstructive pulmonary disease/emphysema, cancer, diabetes, heart disease, operationalized as yes/no for each).

Statistical methods

Descriptive analyses of the full range of factors were stratified by sex, community belonging, and ACSC hospitalization status. Weighted Kaplan–Meier survival curves were created to compare ACSC hospitalization across levels of community belonging in women and men. Cox proportional hazard models were used to estimate the hazards associated with the three-level community belonging exposure and the outcome of future ACSC hospitalization for women and men separately. We defined the study time as being from the CCHS interview date to ACSC hospitalization, right-censoring for the study endpoint (March 31, 2018), 75th birthdate or death (maximum follow-up of 18 years). The results of three models are presented: unadjusted (Model 1), intermediately adjusted for survey cycle, demographic and socioeconomic variables (Model 2), and fully adjusted for survey cycle, demographic, socioeconomic, health and behavioural variables (Model 3). Survey weights from Statistics Canada were used to produce nationally representative estimates and account for complex sampling design and non-response bias. Bootstrap weights were incorporated to calculate variance estimates. Missing data were imputed using hot deck imputation methods for item non-response in large-scale surveys36. Imputation cell variables included age, sex, province or territory of residence, and urban/rural residence. Every variable with missing data had non-response rates of less than 5%. Analyses were conducted using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina).

Sensitivity analysis

Provinces and territories transitioned from ICD-9 to ICD-10 over a six-year period beginning in 200137. To assess if the change in coding classification had an impact on the primary analysis, a sensitivity analysis was run limiting the study start date to April 1st, 2004, the date in which all provinces and territories (except for Quebec) implemented ICD-10. To address concerns of selection bias and account for ACSCs at baseline that might impact community belonging, we also conducted an analysis in which ACSC hospitalizations were included if they occurred at least two years after the CCHS interview date. To evaluate the potential impact of bias resulting from hot deck imputation, a model was run utilizing a non-imputed cohort incorporating ‘unknown’ categories for all covariates with missing or unknown responses. A survival model that included self-rated general health as a potential confounder was also run to examine the possibility of residual confounding by self-rated general health.

Results

The final cohort consisted of 456,415 respondents, representing a weighted population of N = 34,332,000 (Supplementary Fig. 2). Both men and women exhibited similar trends in baseline characteristics across levels of community belonging (Table 1). Compared to those who reported very strong belonging, those who reported very weak community belonging were more likely to be younger, be of visible minority status, live in urban areas and report being in the lowest income quintile. Adults with very weak community belonging were also less likely to be married/common-law, have more than secondary school education, be regular drinkers, and be physically active.

Table 1 Weighted* distribution of all baseline characteristics from CCHS respondents surveyed from 2000 to 2014 and followed up to 2018, stratified by sex and community belonging (N = 34,332,000).
Full size table

Overall, 2.2% of the population experienced one or more ACSC hospitalizations after survey interview. For women, 3.3% of those with weak sense of belonging were hospitalized for an ACSC, whereas only 2.4% of those with strong sense of belonging were hospitalized for an ACSC. Among men, 3.3% of those with weak sense of belonging were hospitalized for an ACSC, while 3% of those with strong sense of belonging were hospitalized for an ACSC (Table 1). Those who experienced ACSC hospitalizations during follow-up were more likely to be older, be widowed, separated, or divorced, report being in the lowest income quintile, be current light smokers, and have poor self-rated health—compared to individuals who did not experience ACSC hospitalizations during follow-up. Adults who experienced ACSC hospitalization were also less likely to: be visible minorities, have more than secondary school education, be physically active, and be of normal BMI status (Table 2).

Table 2 Weighted distribution of baseline characteristics from CCHS respondents surveyed from 2000 to 2014 and followed up to 2018, stratified by sex and avoidable hospitalization status (N = 34,332,000).
Full size table

Graphical assessment of the differences in time to ACSC hospitalization across levels of community belonging using Kaplan–Meier survival curves showed that women with very weak belonging appear to have the least favourable outcomes in terms of time-to-hospitalization, followed by women with very strong belonging (Fig. 1). Men with very weak belonging and men with very strong belonging appear to have similarly less favourable outcomes in terms of time-to-hospitalization, compared to men reporting somewhat strong/weak community belonging. Table 3 presents sex-stratified Cox proportional hazard models for the association between community belonging and time to first ACSC hospitalization after survey response. In unadjusted models (Model 1), both very weak and very strong community belonging, compared to intermediate levels of belonging, were associated with a significant increase in risk of avoidable hospitalizations in women (very weak: HR 1.82, 95% CI 1.60, 2.07; very strong: HR 1.44, 95% CI 1.30, 1.59) and in men (very weak: HR 1.44, 95% CI 1.25–1.68; very strong: HR 1.42, 95% CI: 1.29–1.56).

Fig. 1
figure 1

Kaplan–Meier curves stratified by sense of community belonging for women and men.

Full size image
Table 3 Weighted sex-stratified cox proportional hazard models evaluating community belonging and ACSC hospitalization for pooled study participants from CCHS cycles 2000/2001–2014, followed from time of CCHS interview to first ACSC hospitalization, death, or end of study (March 31, 2018) (N = 34,332,000).
Full size table

Adjusting for demographic and socioeconomic variables (Model 2) and additionally for health and behavioural factors (Model 3) attenuated associations between community belonging and avoidable hospitalization. Compared to women with intermediate sense of belonging, women who report very weak community belonging have a 29% greater risk of avoidable hospitalization (HR 1.29, 95% CI 1.12, 1.47) while women who report very strong community belonging have a 15% greater risk for avoidable hospitalization (HR 1.15, 95% CI 1.03, 1.27). However, these associations were not observed in men (very weak: HR 1.12, 95% CI 0.97, 1.29; very strong: HR 1.08, 95% CI 0.98, 1.19).

The sensitivity analyses consisting of the lagged ACSC hospitalization outcome (Supplementary Table 3), adjustment for self-rated general health (Supplementary Table 4), restriction to outcomes ascertained using ICD-10 (Supplementary Table 5), and a non-imputed cohort that included missing variables as a separate category (Supplementary Table 6) resulted in similar effect sizes to the primary analysis, and no changes in the direction of risk. In addition, we re-ran the analysis using both the 4-level and 2-level variations of community belonging and while direct comparisons are not possible given the different referent group, we observed the same findings of increased risk for low community belonging among females (Supplementary Tables 7–8).

Ethics approval, and consent to participate in the survey was obtained by Statistics Canada. All analyses were conducted under project number 21-MAPA-UTO-7020 at the University of Toronto site of the Canadian Research Data Centre Network, a secure laboratory which provides access to micro-data holdings of Statistics Canada and has in place a detailed protocol to protect the confidentiality of respondents. Consistent with this protocol, all frequencies have a rounding base to the nearest five respondents, and tabulations resulting in cell-counts under 30 individuals were not released.

Discussion

Our study examined the relationship between community belonging and avoidable hospitalizations using a population-based cohort of Canadian adults. After adjusting for potential confounders, both very strong and very weak community belonging, compared to intermediate sense of community belonging, were associated with an increased risk in ACSC hospitalization, however the association is more pronounced for very weak community belonging compared to very strong. After fully adjusting for confounders, these associations remained conclusive among women, but not men. A previous study reported an 80 per cent increased odds of diabetes hospitalization among adults 45 years and older with pre-existing diabetes who reported weak, compared to strong, community belonging23. This is consistent with our findings related to those reporting very weak community belonging. Our findings are also in line with studies that demonstrate the importance of social connections for favourable hospitalization outcomes17,38,39. A recent study that found individuals who reported low life satisfaction had almost three times the risk of ACSC hospitalization compared to those who reported the highest level of life satisfaction3, while another found that weaker sense of belonging was associated with longer length of stay in hospital—most strongly among older adults17.

Current studies generally report a dose–response relationship between stronger community belonging and favourable health outcomes12,14,40, for which we find partial support. In addition to observing higher risk of avoidable hospitalization among those with very weak belonging, we found that those reporting very strong belonging were also at higher risk of avoidable hospitalization. In spite of our efforts to account for a variety of potential demographic, socioeconomic, health and behavioural factors, our findings could be vulnerable to residual confounding, although alternative explanations for the negative health impacts of social capital have been proposed41. A strong sense of community belonging could place more demands, responsibility, and obligation on the individual to contribute to their community, potentially leading to elevated levels of stress. In much the same way as individuals can adopt and emulate others’ beneficial health behaviours, those closely connected to a community with unhealthy or harmful health practices may likewise adopt these behaviors, and in turn, face greater risk of adverse health consequences. Regardless, this finding warrants further investigation into this relationship and other risk factors and conditions that could be driving this pattern.

Few studies have conducted sex-based analyses of the relationship between social connections and down-stream health outcomes with which to compare our findings, and those that exist have produced mixed results. A Finnish study of 206 older men and women found that low social support was associated with higher mortality risk for women, but not men19. The authors of this study suggested this difference could have been related to the fact that women in the study were more likely than men to be widowed and living alone, and thus more dependent on social connections. In contrast, an older study of Finnish men and women demonstrated a graded association between social connections and mortality for men, but not for women20. With respect to the relationship between self-rated health and sense of community belonging, we observed little difference between men and women12. The potential for community belonging to improve health outcomes for women versus men may also vary based on the aspects or types of social connections under consideration, as one study of community belonging and depression suggests42. The relationship between a variety of risk factors (such as comorbidity, socioeconomic status, and physical activity) and ACSC hospitalizations are known to differ for women and men in Canada43. Furthermore, other sex- and gender-based differences in health (i.e. biological/social factors, traditional gender roles) and behaviours (e.g. healthcare seeking patterns) could impact the relationship between subjective well-being and healthcare utilization. More research on the mechanisms behind such sex-based differences is needed to inform health and social policy, and suggests such policies need to be tailored to men and women.

Strengths and limitations

Our study has multiple strengths. First, the use of record-linked data provides a unique opportunity to evaluate the individual-level association between hospitalizations and community belonging—a measure that is not commonly captured in administrative data. Second, we aimed to reduce bias and improve accuracy by accounting for time at risk and adjusting for a variety of demographic, socioeconomic, health and behavioural factors. Third, we conducted several sensitivity analyses accounting for potential bias arising from variations in administrative data collection, missing data, and residual confounding, all of which provided evidence for the robustness of our primary results.

Our study also has limitations. First, there may be risk of measurement error in the covariates given the self-reported nature of the CCHS. This is a particularly important consideration for the exposure, as community belonging is a subjective concept that may be understood differently across different subgroups,11,25 including sex44. Furthermore, the CCHS and DAD captures data on sex, but not gender, so we were limited in our ability to explore gender-based variations in belonging and how differences in socialization might impact the relationship between community belonging and avoidable hospitalization. However, this is still a growing area of research, and at present, this single-item measure of community belonging has been shown to be an efficient and parsimonious measure that captures community belonging as well as related aspects such as social capital and attachment to place10,34. A second and related limitation is that we were not able to account for change in residence across the follow-up period. Moving to a different neighbourhood could lead to changes in one’s sense of belonging to their local community, as social connections are presumably lost in one place and built in another. We hypothesize however that such variations over time would bias our results toward the null. That we still detected significant associations without accounting for neighbourhood changes underscores the potential strength of the relationship between community belonging and avoidable hospitalizations. A third limitation is that we were only able to ascertain death for those with a discharge disposition indicating death in the DAD, thus, there is a risk of incomplete censoring for those who died outside of hospital or before the collection of discharge disposition data in ICD-10. Fourth, although we adjusted for many potential confounders, we were limited to the information available on the survey and administrative data and our results remain vulnerable to residual confounding. Lastly, while our findings are not fully generalizable to all of Canada (as the DAD does not capture data from Quebec and the CCHS has several population exclusions), our study does capture a large sample from the majority of Canadian provinces and territories and utilizes survey weights to account for non-response bias and the sampling strategy of the survey.

Conclusion

We observed that women with very strong and very weak self-reported sense of belonging to their local community had increased risk of avoidable hospitalization, while no conclusive associations by community belonging were found for men. These findings add to the current growing literature on the importance of subjective well-being to health, provide a foundation for further research on the association between community belonging and hospitalization, and draws attention to the importance of sex to variations in subjective well-being and health. Future studies should build upon this work and consider examining the impacts of life stage on community belonging and its relationship to sex-specific population health outcomes. This study may help inform decision-makers regarding strategies and policies aimed to reduce avoidable hospitalizations in Canada.