The association between loneliness, social isolation and all-cause mortality in a nationally representative sample of older women and men

Abstract

Objectives

Individuals who feel lonely and those who are socially isolated have higher mortality risks than those who are not lonely or socially isolated. However, the importance of loneliness and social isolation for survival is rarely analysed in the same study or with consideration of gender differences. The aim was to examine the separate, mutually adjusted, and combined effects of loneliness and social isolation with mortality in older women and men.

Methods

Data from the SWEOLD study, a nationally representative sample of people aged 69+ years living in Sweden, was combined with register data on mortality and analysed using Cox regressions.

Results

Mortality was higher among older women and men with higher levels of loneliness or social isolation. Social isolation was more strongly associated with mortality than loneliness and the association remained when controlling for health. The combined effects of loneliness and social isolation did not surpass their independent effects.

Conclusion

Loneliness and social isolation is associated with an increased mortality risk, and social integration should be a prioritised target for activities and services involving older adults.

Keywords:

• Older people
• loneliness
• social contacts
• social activity
• quality of life/wellbeing

Introduction

There has been a growing focus on loneliness and social isolation among older adults in policy, with an acknowledgement of their negative effects on health and well-being. This has been further highlighted during the COVID-19 pandemic, when many governments have enforced restrictions on social contacts. People experiencing loneliness and social isolation are not only at risk of poor health and well-being but also of premature mortality (for reveiws including a broad age range, see Leigh-Hunt et al., 2017; Nyqvist et al., 2014; Rico-Uribe et al., 2018; Shor & Roelfs, 2015). It has been suggested that these mortality risks are of similar magnitude as well-established risk factors, such as obesity, substance abuse, physical inactivity and mental health problems (Holt-Lunstad et al., 2015). Despite an increasing number of studies, the respective associations of loneliness and social isolation with mortality have rarely been included in the same study and more research is needed (Holt-Lunstad et al., 2015). This study aims to examine the separate and relative associations of loneliness and social isolation with mortality in a nationally representative sample of older adults in Sweden.

Loneliness is a negative feeling, that is, a subjective evaluation of social relations. Loneliness has been defined as the discrepancy between an individual’s desired and achieved levels of social relationships (Perlman & Peplau, 1981), which may concern the quantitative aspects of the relationships, such as a desire for more social contacts or greater frequency of contacts, or the quality of the relationships, such as a desire for greater intimacy or trust in social relations. Social isolation on the other hand is an objective assessment of social relations and refers to infrequent or few contacts with family and friends and may also include living alone (Holt-Lunstad et al., 2015). There is a continuum from social isolation to social participation and integration in society (de Jong Gierveld et al., 2018; Victor et al., 2008). Feelings of loneliness are related to social isolation, although the association is only partial (e.g. Dahlberg et al., 2018; Leigh-Hunt et al., 2017; Taylor, 2020; Tilvis et al., 2012), that is, socially isolated people do not always experience loneliness and lonely people are not always socially isolated (Victor et al., 2008).

Previous research on loneliness, social isolation and mortality

In a meta-analytic review, Holt-Lunstad et al. (2015) included over 70 independent studies that covered a wide range of age groups. The results showed that, while controlling for a variety of factors such as age, gender, socioeconomic status, health, physical activity and smoking, the odds of premature death was 26% higher in the group experiencing loneliness, 29% higher in the socially isolated group and 32% higher in the group living alone. However, the authors concluded that so far there is ‘no evidence to suggest that one involves more risk than the other for mortality’ (Holt-Lunstad et al., 2015, p. 234). Thus, the results are mixed in existing studies that directly compare these predicting variables (Courtin & Knapp, 2017).

In another meta-analysis, also with a broad age range, the consequences of loneliness and social isolation for public health including mortality were examined (Leigh-Hunt et al., 2017). The results showed that both loneliness and social isolation have a significant association with increased risk of mortality. Even though significant associations of loneliness and social isolation with an increased risk for mortality were identified, the authors conclude that more research is needed to confirm whether loneliness and social isolation have a direct impact on mortality, or whether the association is indirect via the cardiovascular system and psychological health (Leigh-Hunt et al., 2017).

Scholars have theorised that although loneliness and social isolation are different constructs there are similarities regarding the mechanisms by which they lead to poor health (Cacioppo et al., 2015; Elovainio et al., 2017; Hawkley & Cacioppo, 2010; Victor et al., 2000). Therefore, it is plausible that synergistic effects of loneliness and social isolation for mortality exist, since both constructs are related to stress, decreased immune functioning and poor health behaviours, e.g. alcohol consumption, smoking and physical inactivity. However, loneliness and social isolation have rarely been included in the same study and there is therefore a lack of research on their combined effect on mortality (Holt-Lunstad et al., 2015). Still, two recent studies have examined this topic. A study from Ireland of community-dwelling adults aged 50 years and older showed that social asymmetry (the degree of overlap between loneliness and social isolation) and the combination of loneliness and social isolation were associated with an increased mortality (Ward et al., 2021). A study of middle-aged and older adults (mean age 60) in Germany showed synergistic effect between social isolation and loneliness on mortality (Beller & Wagner, 2018). While findings from these studies foremost applies to younger older adults, we know less about the oldest old.

The associations of loneliness and social isolation with mortality may also differ between women and men, since women have longer life-expectancy and are more likely than men to have disabling, non-lethal health conditions including functional limitations and depressive symptoms (e.g. Crimmins et al., 2011). In addition, loneliness is more prevalent in women than men (e.g. Dahlberg et al., in press). Women have also reported larger networks than men even in later life (McLaughlin et al., 2010), at the same time as older women, at least in Sweden, to a greater extent than men live in single households (Statistics Sweden, 2020). It is therefore important to address gender differences in the relation between loneliness, social isolation, and mortality.

Aim

The aim of this study is to examine the association of loneliness and social isolation with mortality in a nationally representative sample of older adults, including persons living in the community and in residential care. We include loneliness and social isolation in the same study to address aspects that are fairly rare in the current literature: 1) the separate and mutually adjusted associations of loneliness and social isolation with mortality; 2) the combined effects of loneliness and social isolation on mortality; and 3) gender differences in these associations.

Methods

Sample

This study is based on the Swedish Panel Study of Living Conditions of the Oldest Old (SWEOLD) (Lennartsson et al., 2014) in combination with the Swedish Cause of Death Register. SWEOLD is a national survey of the oldest old (born between 1892 and 1935) living in Sweden at the time of data collection. This article is based on the 2004 wave of data collection including older adults aged 69 years or older. Both older adults living in residential care and in the community were included in the analyses. In the 2004 sample, 13% of those aged 80 years and older were living in residential care facilities.

Materials

The dependent variable was all-cause mortality. Information on mortality (date of occurrence) was obtained from the Swedish Cause of Death Register, which maintains records of all death certificates in Sweden. Respondents were followed from 2004 through the end of 2009, thus resulting in a mortality follow-up period of 5 years. Exposure time was measured in days.

The two variables of main interest for this study were: loneliness and social isolation. Loneliness was measured by one direct question: ‘Are you ever bothered by feelings of loneliness?’, with four response categories: almost never, seldom, often, and nearly always. Information was imputed for a total of 22 respondents who had missing values on the loneliness variable. Seven of the respondents participated in the 2002 wave of data collection and the imputed values for these respondents were taken from the 2002 study. The remaining 15 respondents had their missing value imputed with the mean value of loneliness for women and men, respectively.

We measured social isolation by constructing a summary index adopted from Tanskanen and Anttila (2016). The index comprised four indicators: living alone; lack of social contacts with relatives and friends; lack of social contacts with children and grandchildren; and low level of social activity.

(1) Living alone was measured via the item ‘Do you live alone?’ (yes; no). Respondents living in old age care homes/institutions were considered to live alone.

(2) Lack of social contacts with relatives and friends was measured with four questions relating to: visiting relatives; having relatives over to visit; visiting friends; and having friends over to visit. These questions had three response alternatives: no; yes, sometimes; yes, often. Responding ‘often’ on at least one of these four questions and ‘sometimes’ on at least one question or ‘sometimes’ on at least three questions were considered as not socially isolated on this dimension.

(3) Lack of social contacts with children and grandchildren was based on two items regarding frequency of contacts with children and with grandchildren/great grandchildren, respectively: ‘How often do you usually meet and spend time with your child/children (or grandchildren/great grandchildren)?’ (daily; several times a week; few times a week; a few times a month; a few times a quarter; seldom or never). Those who did not meet and spend time with their children or grandchildren at least monthly were considered socially isolated on this dimension. Not having children and grandchildren were also classified as being socially isolated on this dimension.

(4) Low level of social activity was measured via the same type of questions as social contacts with relatives and friends (see above), using the following list of activities: going to movies, theatre, concerts, museums, exhibitions; eating out at restaurants; going out dancing; participating in study circles or courses; going on trips or excursions; or other activities, such as exercise, playing boule, playing bridge, organisation/club activities or engagement in pensioner organisations. Respondents that did not engage in at least one social activity often or at least two social activities sometimes were regarded as socially isolated on this dimension.

The respondents were scored either 0 or 1 on these four indicators, where 1 indicated that they were socially isolated in that specific dimension; resulting in a summary measure with scores that ranged from 0 to 4, with higher scores indicating higher levels of social isolation. Principal component analysis (PCA) indicated that living alone; social contacts with relatives and friends; and social activity correlated and loaded highly into one dimension, while contact with children and grandchildren showed lower correlation with the other three indicators. Possibly because contacts with children often become more regular with deteriorating health whereas other social contacts and activities tend to decrease with poor health. Conceptually, however, contacts with children/grandchildren indicate whether the respondents are in fact socially isolated or not, and we therefore decided to keep all four dimensions in the social isolation summary index.

In order to separate the effects of social isolation and loneliness on mortality from the effects of poor health on mortality, four measures of health status previously found to be associated with social isolation and loneliness were included in our multivariable models: mobility limitations, self-rated health (SRH), psychological distress, and problems related to cardiovascular diseases.

Mobility limitations were measured via four self-reported items: ‘Can you walk 100 metres?’; ‘Can you walk 500 metres?’; Can you run 100 metres?’; and ‘Can you climb stairs without difficulties?’ (for all items: yes (0); no (1); scale range 0-5, where higher scores indicate greater mobility limitations).

Self-rated health (SRH) was assessed by the question: ‘How would you rate your general health?’ with the response alternatives: good (0), neither good or bad (1) and poor (2). Psychological distress was measured by two indicators: anxiety and depression. Respondents were considered to have psychological distress (coded 1) if they reported at least one severe problem or two slight problems. Problems related to cardiovascular diseases were measured with questions on whether or not they had chest pain, heart attack, stroke, heart failure, and high blood pressure in the last 12 months. Respondents were considered to have cardiovascular problems (coded 1) if they reported at least one severe problem or three slight problems with chest pain, heart problems, or high blood pressure, or a slight or severe problem with heart attack or stroke.

Additional covariates included gender (women (0), men (1)), age in years, and educational attainment (grade school, i.e. up to 7 years depending on year of birth and school catchment area (0), above grade school (1)).

Procedure

Telephone interviews were carried out as the main interview mode. Informed consent was obtained prior to each interview. A postal questionnaire was used if the respondent did not agree to or was unable to conduct an ordinary interview due to, for example, hearing problems. All items analysed in this article were identical for both interview modes. For individuals not able to complete an interview or questionnaire, proxy interviews were conducted with a relative or a member of staff. The most common reason for an indirect interview was dementia or frailty. The total response rate was 87.3% (n = 1180), of which 15.5% were proxy interviews. The high response rate, the inclusion of institutionalised people and the use of proxy informants ensure that the SWEOLD sample is highly representative of older adults in Sweden in 2004 in terms of gender, age and institutional living (see Lennartsson et al., 2014). Due to internal missing the analytical sample used in this study comprised of 1161 persons.

Analysis

Cox proportional hazard regression was used to estimate the effects of independent variables on mortality. The exponent of the regression coefficients—the hazard ratios—from these models were interpreted as the change in the hazard ratio associated with a unit change in the independent variable. All regression analyses were done on the total sample, and separately for women and men. In the multivariable analyses, the effect on mortality was first estimated separately for social isolation and loneliness and then in a model where the effects of social isolation and loneliness were mutually adjusted.

The combined effect of loneliness and social isolation was tested by including an interaction term of loneliness and social isolation. The interaction coefficients in a multiplicative model are conditional on the estimated value of the other variables in the model; therefore, it is uninformative to interpret the interaction coefficient as one point-estimate (see e.g. Brambor et al., 2006). It has been recommended to predict marginal effects of observed values on the independent variables of interest; we followed this procedure and predicted marginal effects from a regression model that included an interaction term between loneliness and social isolation (Ai & Norton, 2003). Furthermore, marginal effects are assessed on an additive scale, which more closely resembles the theoretical notion of synergetic effects in public health and social sciences (Rothman et al., 1980). All analyses were performed in R (version 3.6.2) and the cox regression models were fitted with the coxph command from the survival package.

Results

Characteristics of the sample

Table 1 shows the demographic and social characteristics of the total sample and for women and men separately. Of the analytical sample, 59.1% were women and 51.2% had basic education, that is, grade school. More women than men had basic education. The mean age was 78.7 years among women and 77.9 years among men. Of the total analytic sample, 54.7% had weekly contact with children, while 14.4% were childless, which means that among those with children, 63.9% had weekly contact with them. About half of the sample considered their health good and 12.1% poor. The proportion who considered their health to be poor was higher among women than men. Of the total analytic sample, 22.3% had no mobility problems, with considerable better mobility among men than women. Of the total analytic sample, 15.5% had problems related to cardiovascular diseases, the proportion was higher for women than men, and 12.3% had psychological distress. Nearly twice as many women as men had psychological distress problems.

Table 1. Distribution of demographic and social characteristics and prevalence of loneliness and social isolation of total sample, women and men

Variable	Category	All (n = 1161)		Women (n = 686)		Men (n = 475)
Age	Min	69		69		69
	Max	100		100		100
	Mean	78.4		78.7		77.9
		n	%	n	%	n	%
Education	Grade school	594	51.2	370	53.9	224	47.2
	>Grade school	567	48.8	316	46.1	251	52.8
Contact with	No children	167	14.4	96	14.0	71	14.9
children	Seldom	359	30.9	202	29.4	157	33.1
	Weekly	635	54.7	388	56.6	247	52.0
Self-rated	Good	569	49.0	327	47.7	242	50.9
health	In-between	452	38.9	264	38.5	188	39.6
	Poor	140	12.1	95	13.8	45	9.5
Mobility	0 no problems	259	22.3	117	17.1	142	29.9
	1	312	26.9	181	26.4	131	27.6
	2	159	13.7	105	15.3	54	11.4
	3	174	15.0	115	16.8	59	12.4
	4	257	22.1	168	24.5	89	18.7
Cardiovascular	No	981	84.5	568	82.8	413	86.9
problems	Yes	180	15.5	118	17.2	62	13.1
Psychological	No	1018	87.7	583	85.0	435	91.6
distress	Yes	143	12.3	103	15.0	40	8.4
Loneliness	Almost never	806	69.4	423	61.7	383	80.6
	Seldom	225	19.4	169	24.6	56	11.8
	Often	93	8.0	66	9.6	27	5.7
	Nearly always	37	3.2	28	4.1	9	1.9
Social	0 Not isolated	219	18.9	91	13.3	128	26.9
isolation	1	335	28.9	203	29.6	132	27.8
	2	284	24.5	185	27.0	99	20.8
	3	200	17.2	132	19.2	68	14.3
	4	123	10.6	75	10.9	48	10.1
Mortality	Death	352	30.3	186	27.1	166	34.9

Loneliness and social isolation

Table 1 shows that 11.2% of the respondents ‘often’ or ‘nearly always’ were bothered by feelings of loneliness. Women were bothered by feelings of loneliness more often than men: 13.7% of all women and 7.6% of all men were often or nearly always bothered by feelings of loneliness. Nearly 19% of the respondents were not socially isolated at all, that is, they had no points on the social isolation index. This means that they were cohabiting; had contacts with children or grandchildren (or had no children/grandchildren); had contacts with relatives or friends; and were engaged in at least one social activity often or at least two social activities seldom. While more men than women were not socially isolated at all, it was equally common for women and men to have four points on the social isolation index, that is, to be socially isolated on all four dimensions.

The correlation between loneliness and social isolation was close to moderate (Table 2). This implies that loneliness only partly can be explained by the quantity of social relations and social activity and may also be based on the quality and expectations of social relationships. Moreover, some combinations of loneliness and social isolation were rare. For example, of those persons scoring low on social isolation (0 or 1 point on the social isolation index), few were often or nearly always bothered by feelings of loneliness.

Table 2. Cross-tabulation between loneliness and social isolation (%).

		Social isolation
		0	1	2	3	4	Total (n)
Loneliness	Almost never	21.3	22.2	13.7	8.9	3.4	806
	Seldom	1.6	7.0	5.4	4.1	1.3	225
	Often	0.1	1.8	2.6	2.6	0.9	93
	Nearly always	0.0	0.2	1.2	1.0	0.8	37
	Total (n)	266	362	266	193	74	1161

Spearman’s rank correlation coefficient: 0.323.

Mortality

Between the baseline 2004 interview and the end of the observation period (December 2009), 30.3% of the sample died (Table 1). The time-dependent hazard of mortality showed a constant mortality rate over time. A test of the proportionality assumption in Cox regression indicated that the hazard rate was proportionate over time (supplementary material Figure 1).

Multivariable analyses

In Table 3, the Cox proportional hazard regressions of loneliness and social isolation on mortality are presented. In model 1, the associations between loneliness and mortality adjusted for age, gender and education are shown. The findings reveal that feelings of loneliness were associated with higher mortality. Those who were often or nearly always bothered by feelings of loneliness had a significantly higher mortality rate than those who were almost never bothered by loneliness. In model 2, the equivalent analyses for social isolation show that socially isolation was also associated with an increased risk of mortality and the association showed a linear pattern. Those who had 2 points or more on the social isolation index had an increased risk of mortality than those who were not socially isolated.

Table 3. Cox proportional hazard regressions of loneliness and social isolation on mortality between 2004 and 2009. (n = 1161, deaths = 352).

	Model 1: Loneliness				Model 2: Social isolation				Model 3: Loneliness and social isolation				Model 4: Loneliness social isolation and health
	HR	LCI	UCI	p	HR	LCI	UCI	p	HR	LCI	UCI	p	HR	LCI	UCI	p
Loneliness (ref = 0)
1	0.893	0.675	1.180	0.429					0.842	0.637	1.114	0.229	0.820	0.619	1.085	0.165
2	1.421	1.015	1.990	0.041					1.182	0.842	1.660	0.335	1.017	0.711	1.453	0.929
3	1.894	1.207	2.970	0.005					1.408	0.893	2.220	0.140	1.201	0.740	1.949	0.459
Social isolation (ref = 0)
1					1.313	0.889	1.940	0.171	1.335	0.902	1.974	0.148	1.150	0.775	1.706	0.487
2					1.530	1.023	2.287	0.038	1.524	1.016	2.286	0.042	1.330	0.884	2.002	0.171
3					2.885	1.946	4.276	0.000	2.808	1.881	4.192	0.000	1.966	1.295	2.986	0.002
4					3.394	2.154	5.348	0.000	3.271	2.065	5.182	0.000	2.544	1.585	4.084	0.000
Age	1.122	1.104	1.140	0.000	1.104	1.086	1.123	0.000	1.104	1.086	1.123	0.000	1.098	1.079	1.118	0.000
Gender	1.625	1.308	2.010	0.000	1.713	1.383	2.121	0.000	1.707	1.371	2.125	0.000	1.848	1.484	2.302	0.000
Education	0.975	0.941	1.010	0.158	0.976	0.942	1.012	0.191	0.979	0.944	1.015	0.240	0.992	0.956	1.029	0.661
Self-rated health													1.255	1.054	1.494	0.011
Mobility													1.205	1.102	1.317	0.000
Psychological distress													0.870	0.620	1.223	0.423
Cardiovascular problems													1.317	1.018	1.704	0.036

HR = hazard ratio, LCI = lower confidence interval, UCI = upper confidence interval, p = significance.

To establish the relative association of loneliness and social isolation with mortality we mutually adjusted for these two risk factors (model 3). When including both loneliness and social isolation in the same model, the association of social isolation with mortality remained, while the association of loneliness with mortality was attenuated and become statistically non-significant.

When controlling for health, the coefficients for both social isolation and loneliness diminished and having two point on the social isolation scale was no longer significantly associated with mortality ( Table 3). Respondents who had 3 or 4 points on the social isolation index had significantly higher risks (HR 1.97 and HR 2.54, respectively) to die within five years than those who were not socially isolated.

The results also showed that being older, female, having a poor self-rated health, mobility impairment and problems related to cardiovascular diseases were all significantly associated with an increased risk of mortality.

All regression analyses were performed separately for women and men (see supplementary material Tables 1 and 2). Overall, the results from these analyses showed similar results for both women and men and these results were in line with the main results presented in Table 3. Thus, for both women and men, social isolation had a stronger association with mortality than loneliness and remained significant in the full model, while the association between loneliness and mortality was attenuated and non-significant in the full model. Yet, some associations differed between women and men. The association between loneliness and mortality was somewhat stronger for men than for women. In contrast, social isolation showed a slightly stronger association at each increased level of social isolation below the highest level for women compared to men. Furthermore, an additional test of interaction terms between gender and the independent variables in the main analysis were performed and showed no significant results (results not shown).

Finally, we examined whether there was a combined effect of experiencing loneliness and being social isolated, that is, if the mortality risk was higher in people who were both socially isolated and bothered by feelings of loneliness. This was tested by including an interaction term between loneliness and social isolation, the results from this analysis are presented in supplementary material Table 3. The interaction term was close to 1 and not significant (HR: 0.928, p-value: 0.227). In order to assess the possible synergetic effects across the entire range of observed values we estimated average marginal effects for both loneliness and social isolation (see supplementary material Figures 1 and 2). These analyses indicated no increased effects across higher levels of either variable. Thus, the mortality risk was not further elevated beyond the additive effect of each separate variable when people were simultaneously bothered by feelings of loneliness and socially isolated.

Discussion

This study focused on the separate, mutually adjusted, and combined associations of loneliness and social isolation with mortality in a nationally representative sample of older (+69) women and men.

The results showed that mortality was higher among older women and men who were often or nearly always bothered by feelings of loneliness, when controlling for age, gender and education, whereas there was no significant association between less frequent feelings of loneliness and mortality. The mortality risk also increased with higher levels of social isolation, and the association showed a linear pattern. This result is in line with the findings from a meta-analysis concluding that most research suggests a linear association between social isolation and mortality (Holt-Lunstad et al., 2015) and a Finnish study also concluded that isolation, even from only a few spheres of social relationships, can have an adverse effect on mortality (Tanskanen & Anttila, 2016).

Consistent with previous research (Steptoe et al., 2013; Tanskanen & Anttila, 2016), this study found that the association between social isolation and mortality remained when loneliness was included in the same multivariate model while the association between loneliness and mortality became insignificant when including social isolation in the model. Thus, social isolation seems to be of higher relative importance for mortality than feelings of loneliness. The association between social isolation and mortality remained, although to some degree attenuated, when four different health measures were adjusted for. Health problems can be considered as both mediators and confounders in the association between loneliness, social isolation, and mortality. There is evidence that poor health such as limited functional ability is a risk factor for loneliness (Dahlberg et al., in press) and social isolation (National Academies of Sciences, Engineering, and Medicine , 2020). However, loneliness and social isolation also increases the risk of declining health that in turn increases the risk of death (Holt-Lunstad et al., 2015). The design of this study does not allow us to disentangle the exact pathways of these complex relationships. Nonetheless, a substantial and significant association remained, indicating independent effects between social isolation and mortality that could not be fully explained by other variables included in the models.

In contrast to the findings by Beller and Wagner (2018), and in line with Tanskanen and Anttila (2016) our study found no combined effects of loneliness and social isolation with mortality. This means that among socially isolated individuals, the risk of mortality did not further increase if they experienced loneliness. Furthermore, although Ward et al. (2021) demonstrated that the combination of loneliness and social isolation is most harmful for premature mortality, a discordance between the two measures was also found to be associated with mortality. Their findings also indicated that social isolation was more strongly associated with premature mortality than loneliness, since the group reporting low loneliness and high social isolation had a higher mortality risk than the group with high loneliness and low social isolation. These findings indicate that social isolation and feelings of loneliness might operate in different ways and have independent pathways to mortality (Tanskanen & Anttila, 2016). However, given the conflicting results, more research is needed to fully understand the complex nature of these relationships.

With separate analyses for women and men, this study contributes to research by addressing gender differences. The separate analyses for women and men did, however, not deviate by any substantial degree from the findings for the total sample. Thus, for both women and men, social isolation was of more relative importance for mortality than feelings of loneliness (cf. Steptoe et al., 2013). However, this does not mean that the association between feelings of loneliness and mortality should be ignored. Previous research has found that loneliness, not social isolation, is associated with mortality (Holwerda et al., 2012; Iecovich et al., 2011) and there are indications of direct effects between loneliness and mortality in younger age groups (Holt-Lunstad et al., 2015). There is thus a need for further research on the association between social isolation, loneliness and mortality in older adults.

Strengths and limitations

A key strength of this study is that it is based on a nationally representative sample of older adults living in Sweden. The inclusion of people living in the community and in residential care facilities, the use of proxy informants for people unable to be interviewed directly, and a high response rate ensure the representativeness of this age group in Sweden. The data are, thus, highly representative of the population, including frail and cognitive impaired older adults (Lennartsson et al., 2014). Furthermore, our data on mortality is drawn from a national register with complete coverage of the population.

Although the use of proxy informants is a strength in relation to representativeness it can also be a limitation, since there is a risk that indirect interviewees report more health problems and poorer quality of life in the older person than older adults themselves would do (Graske et al., 2012; Moyle et al., 2012). However, other research has shown that responses from a person with good knowledge about the living conditions of the older person demonstrate good concordance with responses from older adults themselves when measuring, for example, social isolation (Boyer et al., 2004) and quality of life (McKee et al., 2002). In addition, the exclusion of those who are not able to conduct an interview by themselves would underestimate many difficulties and problems (Kelfve et al., 2013).

A potential limitation of this study is the use of a single-item measure of loneliness. Still, while there may be advantages of using validated instruments, single items of loneliness are frequently used and accepted in research on loneliness in old age (e.g. Jylhä & Saarenheimo, 2010; Luanaigh & Lawlor, 2008; Victor et al., 2008). While being less sensitive than scales, single items are regarded to have good face validity, as they ‘present an everyday life concept that is routinely used in daily interactions’ (Victor et al., 2008, p. 65).

Although this study included a broad set of indicators to measure social isolation there are aspects of social contacts and social activity not covered, which may have implications for the analysis of the association of social isolation and mortality.

We tested whether there was a combined effect of loneliness and social isolation that was larger than the separate effects. The results indicated no effect of the interaction between these two variables. However, only a small number of respondents experienced both high levels of loneliness and high levels of isolation. Therefore, there is a need for these findings to be complemented by further research examining the possibility of a combined effect of loneliness and social isolation on mortality in samples where these conditions are more prevalent.

Finally, results from observational studies may be affected by reverse causality. Health problems limit individuals’ potential to participate in society across various life domains. Thus, social isolation and/or loneliness may be more common among older adults with health problems and the risk of mortality may be higher because of higher levels of health problems, not because of social isolation and/or loneliness.

Implications for policy and practice

This study adds to previous literature by showing that social isolation and loneliness have consequences for mortality. In addition to the risk of mortality, social isolation and loneliness are important for health and well-being. Interventions to combat these conditions in older adults should, thus, be highly prioritised. Such risks are particularly important to consider at times such as the COVID-19 pandemic, when many governments have restricted physical social contacts.

Currently, there is a lack of evidence for effective interventions to reduce loneliness and social isolation (Fakoya et al., 2020; Victor et al., 2018). To support evidence-based practice, there is a need for studies of high quality that examines the separate effects of interventions on loneliness and social isolation. In the design of interventions, it is important to address risk factors of these conditions separately, and to target vulnerable groups of older adults identified in previous research.

Further research to disentangle the association between social isolation and loneliness and how these, in turn, are associated with various health outcomes would give insight into how older adults may benefit from different interventions (Newall & Menec, 2019).

Older adults with poor health and functional limitations are at an increased risk of loneliness and social isolation and for those who receive social care services, policies and strategies to prevent and reduce these conditions could be an integrated part of already existing social services. In Sweden, this is supported in the Social Service Act, which states that social services should support older adults in leading an active life together with other people. However, due to financial pressure on these services, in practice work to address health needs is often prioritised over work to prevent and or reduce social isolation and loneliness.

As social isolation has implications for mortality, the health care sector also needs to take older adults’ social situation into account, for example, when assessing health care needs. A previous study found that socially excluded individuals did not receive higher levels of social care than socially included individuals, but that they had an increased risk of health care visits, which offers an opportunity to identify this group and refer them to appropriate social care providers and/or other providers of support for social integration (Dahlberg & McKee, 2016).

Conclusions

For both women and men, social isolation showed stronger associations with mortality than loneliness in mutually adjusted models. Interventions to reduce or prevent social isolation are therefore likely to have greater direct benefits for mortality than interventions targeting loneliness. It is of utmost importance that social integration and social relations are a prioritised target for social services and activities involving older adults.

Ethical approval

Ethical approval was obtained from the Stockholm Regional Ethical Review Board (reg.no. 2004-314/5; 2010/403-31/4).

Disclosure statement

The authors declare that they have no conflict of interest.

Additional information

Funding

This work was supported by Swedish Research Council for Health, Working Life and Welfare (FORTE), grant numbers 2015-00440, 2016-07206, and 2017-00668.