Interventions Associated With Reduced Loneliness and Social Isolation in Older Adults

1Older adults (generally defined as those aged ≥65 years),, are more vulnerable than younger adults to loneliness and social isolation.,,, The COVID-19 pandemic has exacerbated this phenomenon. Loneliness is described as the subjective perception of missing social contacts or a desired companion, while social isolation is the objective lack of social contact with other persons. Loneliness and social isolation are associated with morbidity and mortality.,,, In the US, one-third of adults aged 45 years and older report loneliness and nearly one-quarter of adults aged 65 years and older are considered socially isolated.

2Loneliness in older adults can be mediated by supportive social networks, physical mobility, and living arrangements. Previous systematic reviews (SRs) on interventions targeting loneliness showed that a multitude of interventions,,, can be associated with reduced loneliness in older adults, including physical exercise,,, reminiscence therapy, and technological interventions., Several reviews,, incorporated social support as an outcome, suggesting that multiple mechanisms may improve the social milieu of older adults. Although more recent studies have divided interventions by subtype, previous SRs were limited to a specific intervention (eg, exercise or technology),,,,,,, the absence of meta-analyses, and an older search date.,,, The most recent meta-analysis searched the literature to 2009 and included 20 randomized clinical trials of only loneliness outcomes. There has since been increasing awareness of loneliness and social isolation by clinicians, researchers, and policy makers, with calls to centralize evidence and best practices. The National Academies Press additionally highlights the importance of assessing both social isolation and loneliness. The aim of this SR and meta-analysis was to update and broaden the knowledge base on the interventions associated with a reduction in loneliness and social isolation in older adults.

3This SR and meta-analysis was registered with PROSPERO (CRD42020178836). We reported according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline.

4We included randomized clinical trials of adults aged 65 years and older that reported a validated quantitative outcome measurement of loneliness, social isolation, or social support or network in the English language. There was no exclusion based on prerequisite loneliness and/or social isolation. Theses and protocols were searched to identify subsequently published studies and were included if there was a peer-reviewed journal publication. Social support was defined as a multifaceted concept encompassing the type of support being received, and the perception of having accessible and quality social ties, with social needs being met., Low social support and loneliness are interrelated constructs that are associated with poorer quality of life.,

5The search was conducted on March 2020. We searched OVID, CINAHL, CENTRAL, Embase, PsychINFO, Web of Science, and Scopus databases. Citations of included SRs were hand searched. We included 5 concepts and their associated MeSH, EMTREE, or PsychINFO terms: older adult, social isolation or loneliness or social support, social intervention, technology, and music therapy or animal therapy. Details of the search strategy can be found in eTable 1 in the Supplement.

6Titles, abstracts, and full-text articles were reviewed in duplicate for inclusion or exclusion. Full texts of intervention studies targeting loneliness or social isolation in older adults were included. Discrepancies were resolved by discussion, and if required, a third reviewer. The κ statistic was used to determine reviewer agreement (eTable 2 in the Supplement) for abstract selection.

7Data were extracted and entered into an Excel version 16.64 (Microsoft Corp) template by independent pairs of authors (J.M. and S.M., C.V.T. and P.H., K.R. and H.S., C.W. and P.H., and K.M. and P.H.). One author extracted and entered the data while the second author confirmed accuracy. The following data were extracted: author, year, country, setting, study design, number of participants, attrition, demographics (mean or median age and percentage female), inclusion and exclusion criteria, loneliness or social support scale used, study outcomes, and a description of the study groups. Authors were contacted to obtain missing study data. Long-term care (LTC) was defined as participants who required institutional living (eg, nursing home); this excluded assisted and retirement living. For studies with multiple outcome measurement time points, the final measurement was extracted. Studies were grouped by intervention (eg, animal therapy, psychotherapy or cognitive behavioral therapy [CBT], exercise, social interventions, and information and communications technology), similar to recent SRs., Combination or multicomponent interventions were defined as studies containing multiple different interventions (eg, exercise and CBT).,,, Intervention types were assessed independently by 1 author (P.H.) and reviewed by 2 other authors (J.M. and J.A.K.). Discrepancies were resolved by consensus.

8Risk of bias was assessed independently by pairs of authors (J.M. and S.M., C.V.T. and P.H., K.R. and H.S., C.W. and P.H., and K.M. and P.H.) using the revised Cochrane risk of bias tool for randomized trials. Discrepancies were resolved by consensus or a third reviewer.

9Articles with quantitative outcomes were included in meta-analysis when possible. Social support and social isolation were analyzed separately from loneliness. The standardized mean difference, Cohen d, and associated 95% CIs were estimated for studies with available data. We used the compute.es package in R version 1.3.1056 (R Project for Statistical Computing) to estimate Cohen d when sufficient information was available. Random-effects models using generic inverse variance methods were performed to pool the overall effect size (ES) by intervention. Statistical heterogeneity was evaluated with the I2 statistic and estimating prediction intervals. Prediction intervals were estimated using an equation in Higgins et al that uses a t distribution with K − 2 degrees of freedom (where K represents number of studies). When heterogeneity was observed, we used the find.outliers function in R to identify which studies may be contributing the most influence to the heterogeneity, then removed these for a subsequent analysis. Sensitivity analyses were performed excluding multicomponent interventions (eg, combined Tai Chi and CBT) from the main analysis, and studies without active controls. We separately analyzed community and LTC settings, as studies suggest benefit for interventions targeting loneliness in LTC, where loneliness is highly prevalent., Mixed settings that included LTC were not included in the meta-analysis. Heterogeneity was qualitatively assessed given the heterogeneity in study design and methods, and the limited statistical power to perform meta-regression. Funnel plots were produced to assess for potential publication bias (eFigure 1 in the Supplement). Statistical analyses were completed using RStudio, version 1.3.1056 (R Project for Statistical Computing). Estimating methods for Cohen d can be found in eTable 3 in the Supplement. The BMJ Best Practice Grading of Recommendations Assessment, Development and Evaluation of Evidence Tool was applied to assess the quality of the evidence. Two-sided P <.05 was considered significant. Data were analyzed from November 2021 to September 2022.

10The search resulted in 16 229 citations, with 15 460 excluded after title and abstract screening. Eight hundred sixty studies were included for full-text review, of which 790 were excluded. The PRISMA flow diagram of the search results is shown in Figure 1. Seventy studies with 8259 participants met the criteria for inclusion in the SR (Table and eTable 4 in the Supplement). Articles were published between 1985 and 2020. Most studies were conducted in the US (25 studies). Forty-three studies enrolled community-dwelling individuals, and 12 were conducted in LTC settings. Study sizes ranged from 8 to 741 in the SR, with participants being predominantly female (range, 0%-100%) and between the ages of 55 to 100 years. Loneliness was measured using the UCLA Loneliness scale (33 studies), followed by the De Jong Gierveld Loneliness Scale (13 studies). Social isolation was primarily measured using the Lubben Social Network Scale (3 studies).

11Fourteen studies were excluded from the meta-analysis for lack of reported outcomes or LTC combined with a community setting (eTable 5 in the Supplement).,,,,,,,,,,,,, Forty-four studies were included in the loneliness outcome meta-analysis (33 in community; 11 in LTC) (Figure 2, Figure 3, and Figure 4; eFigure 2 in the Supplement). The social support outcome meta-analysis (8 studies), all set in the community, is found in eFigure 2 and eTable 6 in the Supplement.,,,,,,, Five studies set in the community measured social isolation (eFigure 2, eTable 6 in the Supplement).,,,, The total number of participants eligible for loneliness quantitative analysis was 3535 in community and 1057 in LTC (706 participants for social isolation [community] and 932 participants for social support [community]). Meta-analyses are presented for loneliness outcomes unless otherwise specified. Forest plots with 2 or fewer studies can be found in eFigure 2 in the Supplement. Seven studies had a loneliness enrollment prerequisite. Overall study quality was very low (eTable 7 in the Supplement). The overall risk of bias of the included studies was high (eFigure 3 and eTable 8 in the Supplement), associated with the effect of adhering to the intervention (eg, participants’ and researchers’ awareness of the intervention), unreported adherence outcomes, and repeated outcome measurement.

12Six studies,,,,, were included in the meta-analysis, 2 in the community (Figure 4) and 4 in LTC with an ES of −0.41 (95% CI, −1.75 to 0.92; I2 = 87%; P =.005) and −1.05 (95% CI, −2.93 to 0.84; I2 = 95%; P <.001), respectively. Upon excluding a study comparing group to individual animal therapy, the effect size was −1.86 (95% CI, −3.14 to −0.59; I2 = 86%; P <.001). Generally, participants interacted with living dogs or robotic animals (seal or dog). One study provided a bird in the participant’s room for the study duration.

13Five studies,,,, were included in the meta-analysis, 2 in the community (Figure 4) and 3 in LTC. The ES was −0.67 (95% CI, −1.13 to −0.21; I2 = 0%; P =.704) in community and −0.53 (95% CI, −0.86 to −0.20; I2 = 57%; P =.099) in LTC. Interventions included exercise with arts and crafts, home care with nursing outreach and educational resources, Tai Chi and CBT, and pain management programs. Six studies,,,,, were included in social support meta-analysis (all community-dwelling), with an ES of 0.29 (95% CI, 0.15 to 0.43) and low heterogeneity (I2 = 0%; P =.66).

14Six group-based studies,,,,, in community-dwelling participants were included in the meta-analysis. Interventions included bereavement counseling and instructor-led group support programs. The ES was −0.80 (95% CI, −1.96 to 0.36); heterogeneity was substantial (I2 = 97%; P <.001). When excluding Alaviani et al, the ES was less pronounced (−0.19; 95% CI, −0.35 to −0.03), with no heterogeneity (I2 = 0%; P =.48).

15Eight studies,,,,,,, were included in the meta-analysis, 5 in the community and 3 in LTC. All but 1 study was performed in a group setting (dance, yoga, Tai Chi, and strength and balance training). The ES was −0.15 (95% CI, −0.44 to 0.15) and heterogeneity was low (I2 = 35%; P =.19) in community and −0.53 (95% CI, −0.86 to −0.20; I2 = 57%; P =.10) in LTC. The ES was similar when excluding studies without active controls (−0.45 in community; −0.32 in LTC). Three studies,, were included in social support meta-analysis; ES was 0.17 (95% CI, −0.07 to 0.41) with low heterogeneity (I2 = 0%; P =.48). There was a potential for small study publication bias on the community funnel plot (eFigure 1 in the Supplement). One study set in the community assessed social isolation, with an ES of −0.12 (95% CI, −0.55 to 0.31).

16One community study measured loneliness with an ES of −0.34 (95% CI, −0.55 to −0.13). Two community studies, were included in the meta-analysis for social isolation. The ES was −0.11 (95% CI, −0.57 to 0.35) and heterogeneity was low (I2 = 0.0%; P =.37). Interventions included group rhythm instruments and a choir program.

17Two community-based studies, were included in the meta-analysis (Figure 4). Interventions included occupational therapist-guided technology or assistive devices training. The ES was −0.63 (95% CI, −1.96 to 0.71) with substantial heterogeneity (I2 = 90%; P =.002). Larsson et al, which enrolled persons who were lonely at baseline, was the only study that had a significant outcome in reducing loneliness.

18Two studies, set in LTC were included in the meta-analysis (Figure 4). Interventions included volunteer-led individual reminiscence sessions to group sessions that included sharing memories and identifying goals. The ES was −0.40 (95% CI, −1.98 to 1.17); heterogeneity was substantial (I2 = 95%; P <.001). The study by Moieni et al, which was set in the community, showed similar findings (ES −0.70; 95% CI, −1.17 to −0.22).

19Five studies,,,, set in the community were included in the meta-analysis. Interventions included befriending a volunteer, formation of social groups with discussion topics, and intergenerational programming. The ES was −0.02 (95% CI, −0.21 to 0.17) with low heterogeneity (I2 = 7%; P =.37). Excluding studies without active controls (ES = 0.10; 95% CI, −0.27 to 0.48) did not alter the conclusions. One study set in the community showed significant improvement in social support (ES = 1.02; 95% CI, 0.13 to 1.91).

20Nine studies,,,,,,,, were included in the meta-analysis, 7 in the community and 2 in LTC (Figure 4). Interventions included computer training (in-person or online), videoconferencing (either with family or a trained interviewer), and pedometers to track and provide fitness goals. The ES was −0.19 (95% CI, −0.51 to 0.14; I2 = 59%; P =.03) in community and −1.40 (95% CI, −2.37 to −0.44; I2 = 70%; P =.07) in LTC. Attrition was highest in studies in LTC (up to 44%)., Exclusion of multicomponent interventions (community) did not change the ES (−0.15; 95% CI, −0.53 to 0.24). Social support meta-analysis (2 studies), set in the community-dwelling, had an ES of 0.62 (95% CI, −0.07 to 1.31; I2 = 78%; P =.03)., One community-based study measured social isolation (ES, −0.18; 95% CI, −0.43 to 0.08).

21Four studies,,, set in the community were included in the meta-analysis with an ES of −0.52 (95% CI, −1.21 to 0.17), provided by trained personnel (eg, psychotherapist, doctoral students) in individual and group sessions. There was considerable heterogeneity (I2 = 83%; P <.001). Upon excluding studies without active controls, the ES remained similar at −0.46 (95% CI, −1.39 to 0.46). One study measured social support in the community (ES, 0.41; 95% CI, 0.10 to 0.72). Parry et al also measured social isolation, with an ES of 0.16 (95% CI, −0.06 to 0.38).

22This SR and meta-analysis aimed to determine which interventions are associated with a reduction in loneliness and social isolation in older adults and, to our knowledge, is the largest and most comprehensive such study to date. We additionally separately analyzed outcomes of loneliness, social isolation, and social support. Overall, we found that animal therapy (accounting for one study of group compared to individual animal therapy) and technological interventions in LTC had a large ES. However, given the small number of studies in each intervention and high heterogeneity, cautious interpretation of the ES’s magnitude is required.

23Many interventions are associated with a reduction in loneliness, and all encouraged socialization in some form. Expectations and perceptions of an intervention may influence its effectiveness. For example, studies with a specific goal, particularly in a group setting, may build social relationships that are associated with less loneliness (eg, exercise, multicomponent interventions, animal therapy, technology, psychotherapy, and CBT). However, group interventions are not necessarily beneficial., Interventions that target coping strategies (eg, psychotherapy, counseling, CBT, or reminiscence) may modify individual and environmental factors that can influence social behavior and self-efficacy, thereby reducing loneliness and improving socialization.,,

24Perhaps surprisingly, social interventions were not significant. Although social prescribing is thought to be a potentially effective intervention, in some it may cause social anxiety.,, Additionally, meaningful friendships may not always result from prescribed interventions. Studies of longer duration, required to develop strong friendships, or high value relationships (eg, family and close friends) should be considered and prioritized. Although this may explain why online interactions (eg, videoconferencing) with family had a large ES, only 2 studies were included. Accordingly, it is important that socially prescribed interventions are tailored to an individual’s unique needs. Interventions in LTC may have shown greater ESs for several reasons, including a higher prevalence of loneliness, and being accustomed to group living with shared programs and activities.,,, A number of interventions among community-dwelling participants were not associated with reduced loneliness, potentially related to a highly heterogeneous population., Interventions with social support meta-analysis generally had a smaller ES. This may reflect the subjective nature of these experiences and that an intervention may not change an individual’s social network despite reducing their loneliness.,

25Our SR found similar findings to previous studies. Animal therapy has generally shown positive outcomes for loneliness, potentially mediated by previous pet ownership (particularly dogs).,, Multicomponent interventions generally found success, which is not unexpected given the advantage of incorporating multiple interventions. Due to the multicomponent design of the LTC studies of exercise, the outcomes of exercise could not be isolated, despite the consistent negative effect size. Moreover, only 3 studies were identified. Similarly, reviews of exercise have found conflicting evidence on social support and loneliness.,, Reviews that included reminiscence therapy found benefit, but few studies were identified.,, Reviews of technological interventions generally found mixed ESs on loneliness.,,,, Although a review from Chipps et al identified benefit for videoconferencing, we have shown that this is potentially associated with studies in LTC. We similarly found no benefit for technological training programs (community). Similar to other reviews, there was moderate to substantial heterogeneity across most studies, which may be due to the complex and individual nature of loneliness.,

26Interventions targeting negative self-thought were not consistently associated with reduced loneliness, though evidence suggests such interventions may be beneficial. Future RCTs should aim for more equitable representation of sex/gender and culture, and incorporate measures of loneliness and social isolation. Social isolation and social support require further study in LTC. Future studies should consider identifying important contextual components in LTC associated with a reduction in loneliness. Methodological quality can be improved by adherence outcomes and its analyses, reporting adverse events, and implementing active controls.

27This study has limitations. Sample sizes were small, and few studies used active controls, the latter potentially confounding the intervention effect. The majority of studies enrolled community-dwelling older adults and may affect generalizability. A number of studies of specific interventions shared study authors and participant recruitment geographical locations, which may increase the risk of bias: all studies of exercise and multicomponent in LTC (Tse et al),,, 3 of 4 studies of animal therapy in LTC (Banks et al),,, and all studies of technology in LTC (Tsai et al)., The latter 2 authors reported no overlap in study samples in our correspondence. Although Tse et al reported separately collected cohorts, the possibility of participant overlap remains. The small number of studies per meta-analysis limited conclusions on sources of heterogeneity and the ES’s magnitude. The sustainability of the effect of the interventions cannot be concluded on the basis of our results. Only English language studies were included. Several studies required estimating Cohen d for pooling based on data provided in the original study (eg, P values and 95% CIs).

28In this SR and meta-analysis, exercise and technological interventions in the community had the highest precision with small ESs, whereas animal therapy in LTC had the largest ES when accounting for one study comparing group to individual therapy. When exercise is combined with other interventions (eg, CBT), the benefit may be strengthened. These results require cautious interpretation due to high heterogeneity and a small number of studies, particularly with respect to the ES’s magnitude.