Health Reports
Social isolation and mortality among Canadian seniors

by Heather Gilmour and Pamela L. Ramage-Morin

Release date: June 17, 2020


The impact of social isolation and loneliness on health and well-being is recognized globally as a public health issue. The United Kingdom appointed a Ministerial lead on lonelinessNote 1 and the World Health Organization recognizes the impact of social isolation on disability and death.Note 2 Research generally demonstrates that social isolation is associated with increased risk of mortalityNote 3Note 4Note 5Note 6Note 7 on par with or greater than more traditional risk factors such as alcohol use, smoking and obesity.Note 8

Social isolation can be experienced at any age, although some circumstances relate more specifically to older ages.Note 9Note 10Note 11Note 12 These may include transitions to retirement and accompanying role loss, ill health, loss of a spouse or friends, mobility problems, vision and hearing loss, lower income, residential changes, and changes in access to transportation. At a broader social level, ageism may contribute to social isolation.Note 12 Social isolation experienced by marginalized populations such as the LGBTQ community can continue into the senior years, adding to other factors associated with aging.Note 13

While there may be an intuitive understanding of social isolation, measurement of this concept varies. Objective measures such as the size of social networks and the frequency of social participation reflect the structural aspects of social relationships while subjective feelings of social isolationNote 11 or lonelinessNote 14 embody the functional aspects of relationships.Note 3 Associations between both subjective and objective measures of isolation and risk of death are evident in some studies Note 5Note 7Note 15Note 16Note 17Note 18 although others report different associations depending on the measure.Note 19Note 20Note 21

The primary goal of this study was to examine an objective and a subjective measure of social isolation and their associations with mortality for Canadian seniors aged 65 or older. Prevalence estimates of social isolation using each measure are presented. Selected factors associated with social isolation were examined prior to their inclusion in the multivariate models. Note that data on transitions (e.g. retirement) and changes in life circumstances (e.g. loss of spouse) were not available in the cross-sectional data on which this study was based. Associations with survival were assessed with multivariate Cox proportional hazards models adjusting for selected sociodemographic and health-related characteristics. Structural equation models (SEM) were used to examine direct relationships between social isolation measures and death and indirect effects mediated through health status.

Social isolation was defined objectively as infrequent social participation (low participation) and subjectively as feelings of loneliness and a weak sense of community belonging (subjective isolation). Because of evidence showing different associations between measures of social isolation and mortality by sex,Note 6Note 22Note 23Note 24 men and women were analyzed separately.


Data sources

Canadian Community Health Survey—Healthy Aging (CCHS-HA)

The cross-sectional 2008/2009 CCHS-HA collected information about factors that contribute to healthy aging, targeting people aged 45 or older living in private dwellings in the 10 provinces. About 4% of the population were excluded: residents of the three territories, First Nations reserves, certain remote regions, institutions and Canadian Forces bases, and full-time members of the Canadian Forces. Data were collected from December 2008 through November 2009. Computer-assisted personal interviews were conducted for 94% of the sample; telephone interviews were permitted to accommodate the language needs of respondents. The overall response rate was 74.4%. Detailed documentation for the CCHS-HA is available at:

The Canadian Vital Statistics—Death Database (CVSD)

The CVSD is an administrative dataset that includes demographic and cause of death information for deaths that occur in Canada. Data are collected annually from provincial and territorial vital statistics registries. Deaths that occurred from December 2008 through December 31, 2017 that linked to CCHS-HA records were used in this analysis. Detailed documentation for the CVSD is available at:

Data linkage

Linkage approval (007-2018) was granted by the Chief Statistician of Canada and performed in accordance with the Directive on Record Linkage. CCHS-HA respondents who agreed to share and link their data were probabilistically linked to the Derived Record Depository (DRD) in the Social Data Linkage Environment (SDLE) at Statistics Canada. Probabilistic record linkage works with non-unique identifiers (e.g., name, sex, date of birth, and postal code) and estimates the likelihood that records refer to the same entity.Note 25 Only employees directly involved in the process have access to the information required for linkage and do not access health and/or death-related information. An analytical file without identifying information was created for this study.

Study sample

The study was based on CCHS-HA respondents aged 65 or older with data linked to the CVSD for those who died (Appendix Table A). Follow-up ranged from 8 to 9 years, from the CCHS-HA interview date to December 31, 2017. The study sample comprised 13,037 individuals (5,408 men, 7,629 women); 4,953 respondents (2,175 men, 2,778 women) died in the follow-up period between their CCHS-HA interview and December 31, 2017.


Social isolation measures

Low participation was based on the social participation module that questions respondents about eight community-related activities. Respondents were asked how often in the past 12 months (at least once a day/week/month/year or never) they participated in each activity. Individuals whose overall participation was less than weekly were classified as having low participation versus high participation (one activity or more on a daily or weekly basis). The “weekly” benchmark was selected based on earlier studies.Note 13Note 26

Although low participators did not participate in any of the eight activities on a weekly basis, 4% of them participated in 4 or more activities on a monthly basis.

Subjective isolation was a composite of two measures capturing loneliness and sense of community belonging. The three-Item Loneliness Scale was based on the Revised UCLA Loneliness Scale.Note 27 Respondents were asked “How often do you feel: that you lack companionship? left out? isolated from others?” Response category values (1 = hardly ever; 2 = some of the time; 3 = often) were summed. Respondents who scored 4 to 9 were categorized as lonely versus a score of 3. Sense of community belongingwas determined with one question: “How would you describe your sense of belonging to your local community? Would you say it is very strong? Somewhat strong? Somewhat weak? Very weak?” Individuals who were lonely and reported a “somewhat” or “very” weak sense of community belonging were considered isolated.


Age in years was grouped (65 to 74, 75 or older) for prevalence estimates and the number and percentage of deaths, and entered as a continuous variable in the multivariate analyses. Seniors refers to the study population, that is, those aged 65 or older. The highest level of educational attainment by any household member (less than post-secondary, post-secondary graduation or more) was selected as a measure of socioeconomic status. It represents a family resource that is often correlated with levels of income and wealthNote 28 and is suitable when transitions from employment to retirement are likely in the population. Place of residence was dichotomized as urban or rural. A combination of marital status and living arrangements classified individuals as married/common-law versus not married/common-law (i.e. single, widowed or divorced); the latter was further divided into those who were living alone versus with others where “others” could be a child, friend, sibling or other. A dichotomous variable classified respondents as married or common-law versus neither for the path analysis (SEM).

The Health Utility Index Mark 3 (HUI3)Note 29 was used to assess functional health status in eight domains: vision, hearing, speech, mobility, dexterity, cognition, emotion, and pain and discomfort. Overall scores were categorized into levels of disability for reporting prevalence: no/mild disability (0.89 to 1.00), moderate disability (0.70 to 0.88), or severe disability (less than 0.70).Note 30 Continuous HUI3 scores were used in multivariate models – higher scores equate to better health status.

Smoking status was categorized as current (daily or occasional), former, or never smoked. For the path analysis current and former smokers were grouped together and compared to those who had never smoked.

Physical activity was based on the Physical Activity Scale for the Elderly (PASE). It captures self-reported occupational, household and leisure activity over the previous 7 days.Note 31 Higher scores indicate higher levels of physical activity. Using scores from the population (weighted), respondents’ activity levels were classified into quartiles as least active (score: < 58), low to moderate (58 to 99), moderate to high (100 to 143), or most active (> 143). Continuous PASE scores were used in the multivariate analyses.

Analytical techniques

Cross-tabulations were used to estimate the prevalence of low participation and subjective isolation, and the number and percentage of deaths by social isolation measures and selected characteristics. Associations between social isolation and mortality were examined using Cox proportional hazards models with subjective isolation and low participation entered into models simultaneously. These social isolation variables were moderately correlated (0.28 for men and 0.38 for women), suggesting that they are related, but distinct, concepts. Variance inflation factors (⋜ 2.9) and tolerance estimates (⋝0.2) demonstrated that multicolinearity was not a problem. Adjusted models controlled for potential confounders measured at the time of the CCHS-HA interview. The first model controlled for age; the second added sociodemographic variables (household education; marital status and living arrangements; urban/rural residence) and the final model included health-related characteristics (health status, smoking status, and physical activity). The selection of covariates was based on the literature and availability in the CCHS–HA. The proportional hazards assumption was tested by visual examination of SAS PROC LIFETEST plots.

The direct and indirect impact of subjective isolation and low participation on survival were assessed using SEM with Stata/MP 14.2. Indirect paths were assessed with functional health status (HUI3) as a single hypothesized mediator. HUI3 underwent an arc sine transformation to approximate a normal distribution. The SEM analyses controlled for covariates found to be significantly associated with mortality in the fully-adjusted Cox proportional hazards models (men: age, smoking status, physical activity; women: age, smoking status, physical activity, household education, living arrangement). An arc sine transformation was also applied to the outcome variable, time to death. The model goodness of fit was assessed using the Standardized RMR (SRMR) and the coefficient of determination (CD).

Sampling weights were used to account for the survey design and non-response, and to adjust for differences in agreement to link and share. The use of sampling weights is essential to account for unequal probabilities of selection and to reduce the potential for bias resulting from differing response, share, and agreement to link rates. Bootstrap weights were applied using SAS-Callable SUDAAN 11.0 to account for underestimation of standard errors due to the complex survey design.Note 32 The significance level was set at p < 0.05.


Baseline characteristics of the study population

The weighted study sample (n=13,037) represented 4.2 million people aged 65 or older with a mean age of 74 years in 2008/2009. Almost half (45%) were men, and most (63%) were married or common-law; 54% lived in households where at least one person was a postsecondary graduate; 79% lived in an urban setting; 10% were current smokers, 56% former smokers and 35% had never smoked; mean physical activity (PASE) score was 106 and mean HUI3 score was 0.81.

Subjective isolation and low social participation

From the 2008/2009 CCHS-HA, an estimated 525,000 people (12%) age 65 or older felt isolated in that they reported feelings of loneliness and a weak or somewhat weak sense of community belonging. At 15%, women were more likely than men (10%) to report subjective isolation (Table 1). Over 1 million (1,018,000) older Canadians (24%) reported low participation with no difference in the percentages for men and women. Seniors who were not married or living with a common-law partner were more likely than partnered individuals to report low participation and subjective isolation. Non-partnered seniors who lived with others were particularly likely to be low participators – 41% of men and 35% of women. Lower household education was associated with low participation for women and subjective isolation for men. Age group and place of residence (urban or rural) were not associated with either measure of social isolation.


An estimated 33% of men and 26% of women died during the follow-up period (Table 2). Low participators in 2008/2009 were more likely than those who participated regularly to die, and the same was true for men and women who reported subjective isolation compared to those who did not feel isolated.

There was a greater likelihood of death for older people and residents of lower-education households. Individuals who were married or in a common-law relationship were less likely than those who were not in such a relationship to die over the follow-up. More severe disability, being a current or former smoker, and lower levels of physical activity were associated with an increased likelihood of death.

Low participating men and women had, on average, shorter survival times than individuals who participated more frequently in community-related activities (Table 3). For example, the average survival time for women who were low participators was approximately 8 months shorter than those who were high participators. There was no difference in average survival times for men based on subjective isolation. However for women, there was a significant association between feeling isolated and shorter survival times.

Low participation and subjective isolation were further examined in multivariate models that included both measures. Low participation was significantly associated with death for men and women even when the potential confounding effects of subjective isolation, socio-demographic characteristics, health status, and health behaviours were considered (Table 4). For women, subjective isolation was associated with mortality when low participation and socio-demographic characteristics were considered (HR 1.4), but lost significance in the final model when controlling for health status and behaviours. Subjective isolation was not associated with death in any of the models for men. As expected, increasing age, smoking and lower physical activity scores were significantly associated with mortality; higher functional health scores were protective. For women only, lower household-level education and not being married or having a common-law partner were associated with death in the fully adjusted model.

Consistent with the survival analysis, results of path analysis showed that low participation was significantly associated with survival time for men and women (Figure 1). In addition to the direct effects, there were significant indirect effects mediated by health status. That is, men and women who were low participators in 2008/2009 had an increased risk of mortality (direct effect) as well as poorer health status (lower HUI3 score) which was associated with shorter survival time (indirect effect). The direct effect accounted for 89% of the total effect for men and 85% for women. There were no direct effects of subjective isolation on survival for men or women, only indirect effects through health status. Assessment of the fit for path analysis models were considered acceptable with SRMR=0.000 and CD=0.257 for men, SRMR=0.000 and CD=0.298 for women.


Mortality risk in community dwelling seniors was estimated prospectively in relation to objective and subjective social isolation using linked population-based survey and administrative data. About 12% felt isolated and 24% were low participators in 2008/2009. Survival and path analyses revealed that low participation was associated with mortality, while subjective isolation was only related to mortality indirectly via health status.

The lack of a gold standard for defining and measuring social isolation makes comparisons with other studies a challenge.Note 11Note 33Note 34Note 35Note 36 The term is often used interchangeably with “loneliness”, a related but distinct concept.Note 9Note 11 Loneliness has been defined as mismatch between the actual and desired quality and quantity of social connectionsNote 14 whereas social isolation is a broader concept, encompassing an individual's place among social networks and not just feelings of inadequate personal connections.Note 34Note 37Note 38 Zavaleta et al.Note 34 describe it succinctly as “a deprivation of social connectedness” (pg. 367). Nonetheless, a body of research using a variety of measures (e.g., social contacts, network size, loneliness, social support, or composite indexes that combine several aspects of social isolation) provides robust evidence that a lack of social connectedness negatively impacts longevity.Note 3Note 6Note 8Note 16Note 39Note 40Note 41

The combination of loneliness and weak sense of community belonging measures, used successfully in a previous study,Note 42 captures what WeissNote 43 refers to as emotional loneliness - the absence of close ties or personal relationships - and social loneliness – the absence of connections to a broader social network such as friends and community groups. Similar to the discrepancy theory of loneliness,Note 35 the intersection of loneliness and weak sense of community belonging permits the identification of those who subjectively appear most vulnerable, those who feel isolated both from close personal relationships and the broader community. Objective isolation represents a lack of social contacts measured in a quantitative manner – in this study by infrequent participation in community-related activities. The subjective and objective measures are conceptually distinctNote 44 but also linked – the larger the network the more likely a person’s need for close relationships will be met, reducing feelings of loneliness.Note 45

Our study, which investigates subjective and objective social isolation as separate albeit related entities, joins othersNote 18Note 20Note 21Note 39Note 40 in identifying the effect of each on mortality accounting for the effect of the other. Consistent with these previous studies, our results indicate an independent association between objective isolation and mortality that did not persist for subjective isolation.Note 20Note 21Note 39Note 40 Beller and WagnarNote 18 further identified a synergistic effect whereby the interaction between loneliness and an objective measure of social interactions was significant and the higher level of one, the larger the effect of the other on mortality. The current study did not find a significant interaction between subjective and objective isolation in relation to mortality. EllwardtNote 21 found that the association between subjective isolation (emotional and social loneliness) and mortality did not persist when mental health was added to the model and hypothesized that the relationship with mortality was indirect. In this study, we hypothesized that the pathway through which social isolation is associated with mortality is through functional health status (which includes an emotional health domain). Path analysis supported this conclusion. Together, these results emphasize the importance of including both subjective and objective measures of isolation in analyses.

Our study treats marital status and living arrangements – combined into a single covariate – as a potential confounder in the relationship between social isolation and mortality rather than as a measure of social isolation itself. KlinenbergNote 46 points out that since the mid-1900s, living alone has become increasingly common and that it is not synonymous with loneliness or social isolation. Certainly the loss of a spouse may lead to social isolation but so too may an unhappy marriageNote 3Note 45 or becoming a caregiver for an ailing partner.Note 47Note 48 Our study revealed that for women, not having a partner (married or common-law) was a risk factor for mortality over and above the impact of social isolation; for men, the apparent association was better accounted for by health status and behaviours.

Strengths and limitations

A strength of this study is the large sample, representative of the senior household population in 2008 to 2009. It includes an extensive follow-up period of eight to nine years with linkages to quality vital statistics death data. The inclusion of an objective and subjective measure of social isolation permits the examination of their relative contributions to mortality. In addition, the large sample allows the associations between social isolation and mortality to be analyzed separately for men and women.

There are a number of limitations to note. Although functional health status was conceptualized as being on the causal pathway between social isolation and mortality, these exposures were measured at the same time in the cross-sectional CCHS-HA. While this study posits social isolation as a risk factor for poor health, the opposite may also be true. Social isolation, health status, behaviours and other covariates were measured only at baseline and therefore it is not possible to establish if they varied during the follow-up period. Data on transitions (e.g. employed to retired) and changes in life circumstances (e.g. loss of spouse) that may be associated with social isolation were not available in the cross-sectional data on which this study is based. The CCHS-HA does not include residents of long-term care facilities. Proxy respondents (2.2% of the CCHS-HA sample) were excluded from the analysis. While this has the potential to introduce bias, so too would inclusion of survey records completed on behalf of infirm individuals.Note 49 An analysis of selected characteristics of the CCHS-HA data indicated that proxy respondents were more likely to be men, older, and in worse health than self-reporting respondents.Note 42 CCHS–HA data are self-reported and not verified by any other source. Some variables of potential relevance (for instance, medication use and history) were not available. Probabilistic linkage was used to match survey records to death information; the possibility of false links or missed links exists. The CVSD includes death information for events occurring predominantly in Canada; respondents who died outside of Canada account for less than 0.2% of the linked data. The primary goal of the study was to examine associations between social isolation and mortality. As such, only selected factors associated with social isolation, health status and mortality were included for use in the multivariate models. Future research could examine whether the relationships between social isolation and mortality differed for specific populations identified by characteristics such as LGBTQ, Indigenous and minority language status.


Our study identifies and quantifies direct and indirect associations between social isolation measures and mortality. Mechanisms through which social isolation impacts health status and mortality are likely many and varied. For example, low social participation could reflect deficits in social networks, mobility problems, lack of transportation, geographic isolation or other factors such as hearing or vision loss that present challenges to social participation. Feelings of isolation could be the result of loss or change in relationships through death or separation, physical and mental health problems, ageism and other factors that leave people feeling lonely and detached from the community.

Our results support screening and initiatives such as social prescribing by primary care cliniciansNote 50 and Age-Friendly Communities (AFC).Note 51Note 52 In the former, physicians screen and refer their patients to non-clinical community services that address the social determinants of health, including social isolation, taking individuals’ needs and interests into consideration.Note 50Note 53 The AFC approach,Note 54 which has been implemented globally including in some Canadian communities, provides guidelines to adapt structures and services to better respond to the needs of an aging population in several domains including social participation. Addressing social isolation is an upstream approach with the potential to help improve quality of life and delay morbidity and mortality.

Currently many Canadians are experiencing social isolation as a result of the COVID-19 pandemic; seniors in particular are advised to physically isolate themselves to reduce the risk of infection. While this study examines associations between social isolation and mortality over a nine-year period, it also points to the importance of seniors returning to their activities and interactions following the pandemic to prevent the development of long-term social isolation.


Statistics Canada thanks all participants for their input and advice during the development of the 2008/2009 CCHS-HA. The survey content was developed by the Health Statistics Division at Statistics Canada in consultation with Health Canada, the Public Health Agency of Canada, and experts conducting the Canadian Longitudinal Study on Aging (CLSA), a major initiative of the Canadian Institutes of Health Research. Consultations included stakeholders from Human Resources and Social Development Canada and provincial and territorial health ministries. The addition of 5,000 respondents aged 45 to 54 was funded by the CLSA.


Date modified: