Introduction

Loneliness (perceived discrepancy between current and desired social relationships [1]) and social isolation (lack of social activities [2]) are major threats to morbidity and longevity. For example, a recent meta-analysis [3] showed that both loneliness and social isolation were significantly associated with a greater risk of all-cause mortality: the pooled effect size for loneliness was 1.14 (95% CI: 1.08 to 1.20, p < 0.001) and the pooled effect size for social isolation was 1.32 (95% CI: 1.26 to 1.39, p < 0.001). Moreover, another meta-analysis [4] showed that poor social relationships were associated with a 32% (pooled relative risk: 1.32, 95% CI: 1.04 to 1.68) increase in stroke risk and a 29% (pooled relative risk: 1.29, 95% CI: 1.04 to 1.59) increase in coronary heart disease risk.

Both greater loneliness and greater social isolation are associated with a greater likelihood of having mental health disorders. For example, a former meta-analysis showed that loneliness had large effects on mental health outcomes (depression, anxiety, general mental health and suicidality) [5]. Large negative effects of social isolation on mental health (particularly amongst older people) have also been demonstrated [6].

A recent meta-analysis showed a pooled loneliness prevalence of 28.6% (95% CI: 22.9 to 35.0%) and a pooled social isolation prevalence of 31.2% (95% CI: 20.2 to 44.9%) amongst older adults aged 65 years and over (based on 15 countries of four continents: North America, South America, Asia and Europe) during the COVID-19 pandemic [7]. Higher pooled prevalence rates for loneliness were also identified amongst older adults (compared to young adults) in eastern European countries [8]. Moreover, two former meta-analyses showed higher loneliness levels in eastern and southern European countries, compared to northern European countries [8, 9].

Critical life events take place (e.g. loss of friends and relatives or health deteriorations) in later life, which can contribute to loneliness and social isolation amongst the oldest old (individuals aged 80 years and over) [10,11,12]. Moreover, social distancing during the pandemic can also have contributed to increased levels of loneliness and social isolation [13, 14]. To date, nine studies (e.g. [15, 16]) have examined the prevalence—and occasionally the correlates—of loneliness and social isolation amongst the oldest old. However, there has been no systematic review of studies (including meta-analysis and meta-regression) that systematically synthesises the present evidence. Therefore, our aim was to address this knowledge gap (by focussing on community-dwelling and institutionalised individuals aged 80 and over).

Specifically, the aim of this systematic review, meta-analysis and meta-regression was to identify the prevalence and correlates of loneliness and social isolation in the oldest old. Such knowledge is of great importance, particularly in view of the growing number of individuals in this age bracket. Additionally, our work may identify correlates of loneliness and social isolation. Furthermore, our work may clarify potential knowledge gaps and may thus inspire upcoming studies. Moreover, pooling of studies is possible by performing a meta-analysis. This can help deliver a more accurate overview compared to individual studies. A meta-regression can also assist in separating the influence of significant moderating factors (such as region in which the study was conducted, or tool used to quantify loneliness or social isolation).

Methods

Our current work satisfied the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [17]. A PRISMA checklist can be found in Additional file 1. Furthermore, our work has been registered in the International Prospective Register of Systematic Reviews (PROSPERO, registration number: CRD42022339013). No amendments were made. In January 2023, an electronic search was conducted (three databases: Medline, CINAHL, PsycINFO). In Table 1, our search strategy for Medline is displayed (for the other databases: please see Additional file 2). The suitability was assessed by two reviewers (AV, AH) based on two steps (1: title/abstract screening and 2: full-text screening afterwards). Moreover, we conducted a hand search (i.e. (1) we examined the references of included studies and (2) we examined studies that cited the included studies). Grey literature was not searched. When perspectives on inclusion of studies differed, we used discussions to resolve this (if needed, a third party (HHK) was used). The same procedure applied to assessment of study quality and extracting data.

Table 1 Search strategy (Medline)
Full size table

Our inclusion criteria were:

  • Cross-sectional and longitudinal observational studies identifying the prevalence of loneliness and social isolation amongst the oldest old (80 years and over), covering both, community-dwelling and institutionalized individuals

  • Studies adequately assessing these constructs

  • Studies published in peer-reviewed journals (German or English language)

It should be noted that the appropriate assessment of the constructs was strongly guided by the criteria of the COSMIN guidelines [18].

In contrast, studies exclusively focussing on samples with a certain disorder (e.g. individuals with Parkinson’s disease) were excluded. No restrictions were applied with respect to the time and place of the studies.

A pretest of 100 titles/abstracts was performed before the final inclusion criteria were determined. However, our inclusion criteria remained unchanged. Data were extracted by one reviewer (AV) and cross-checked by another (AH). Study design, definition and operationalization of loneliness, social isolation, characteristics of the sample, statistical analysis and key outcomes were extracted as data. If data were missing, study authors were contacted.

The quality of the studies was assessed using the established Joanna Briggs Institute (JBI) standardised critical appraisal instrument for prevalence studies [19]. The score ranges from 0 to 9 (whereby higher values reflect higher study quality and a lower general risk of bias). Study quality was independently assessed by two reviewers (AV and AH). A cut-off score for excluding studies from meta-analysis was not applied.

With respect to the meta-analysis, in order to pool proportions across the included studies, we used random-effects models because heterogeneity across studies was expected. Following given recommendations, heterogeneity between studies was estimated using the I2 statistic, with I2 values between 25 and 50% considered as low, 50% and 75% as moderate and 75% or more as high heterogeneity [20]. The well-known Stata tool ‘metaprop’ [21] was used to conduct meta-analysis.

It should be noted that loneliness was grouped into “not lonely”, “moderately lonely” and “severely lonely” largely following the procedure proposed by Gardiner et al. [22]. For further details, please see the Additional file 3.

With regard to social isolation, the few single studies dealing with social isolation usually simply distinguished between the presence of social isolation and the absence of it. Therefore, we maintained this dichotomy for the meta-analysis. The detailed presentation of the dichotomization of social isolation in the single studies is provided in Table 2.

Table 2 Study overview and key findings
Full size table

Regarding meta-regression, we used the ‘meta regress’ command. More precisely, we performed a random-effects meta-regression with restricted maximum likelihood. Knapp–Hartung adjustment was applied for the standard errors [23]. The coefficients were recalculated. The reason is that the coefficients in the meta-regressions were initially scaled as double arcsine values (rather than proportions) (following Lipsey and Wilson [24]). Meta-regressions were conducted to identify the heterogeneity sources [25].

To detect a potential publication bias, a funnel plot as well as the Egger test (p < 0.05 indicates publication bias) were conducted [26]. Stata 16.1 (College Station, TX, USA) was used in our current study.

Results

Study overview

A flow chart is given in Fig. 1 [17]. More precisely, this figure illustrates the flow of information across the various stages of our systematic review and meta-analysis.

Fig. 1
figure 1

Flow chart

Full size image

After eliminating duplicates, a total of 6,906 studies underwent screening, specifically through the evaluation of titles and abstracts. During this initial phase, the most prevalent reason for exclusion was the absence of reported data on loneliness or social isolation prevalence amongst the oldest individuals. In the subsequent step, which involved assessing the full text of selected studies, some distinct reasons for exclusion were identified (e.g. not reporting prevalence data or not examining oldest old individuals). When different studies used the same dataset, we selected the study that used the most comprehensive dataset (see also: [27, 28]). Ultimately, our present systematic review incorporated a total of 22 studies [15, 16, 29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48], with all of these studies included in the meta-analysis. The hand search did not reveal additional studies. Table 2 describes important characteristics of the studies and key findings. Within the scope of our analysis, three studies investigated the factors associated with loneliness [16, 40, 43], and their adjusted results are presented in Table 2.

Data were from Europe (n = 12, two studies each from the Netherlands, Sweden, the United Kingdom, one study each from Belgium, Faroe Islands, Finland/Sweden, Germany, Greece, Portugal), South America (n = 1 from Brazil), Asia (n = 4, two studies from China, one study from Israel and one study from India), North America (n = 3, all from the United States) and Oceania (n = 2, both from New Zealand). In sum, 13 studies had a cross-sectional design, and nine studies had a longitudinal design. Large, representative surveys were used in 17 studies. Given the fact that data were collected amongst the oldest old, it has to be noted that twelve had a large sample size (with sample sizes in the higher three- or four-digit range, e.g. 600 or higher). The proportion of women ranged from about 60% to 80% in 15 studies and the average age, if reported, ranged from 80 to 90 years in nine studies. Overall, 15 studies used single item measures to quantify loneliness. The remaining four studies used different versions of the UCLA-tool and the De Jong Gierveld tool (11-item version). Three studies examining social isolation used the LSNS-6, whereas the remaining three studies used single-item measures and a self-developed tool (based on four activities). Additional file 4 displayed the frequency for the tools used to quantify loneliness and social isolation, respectively.

The studies were published between 1994 and 2022 and 14 out of the 22 studies were published in or after 2018 (3 times: 2018, 3 times: 2019, 3 times: 2021 and 5 times: 2022). Data collection took place during the first COVID-19 lockdown in one study [30]. In a second study, data collection took partly place during the COVID-19 pandemic (i.e. June to July 2020) [32]. Further details are shown in Table 2.

Correlates of loneliness

Three studies examined the correlates of loneliness [16, 40, 43]. All three studies found that living alone and the presence of depression are associated with a greater likelihood of loneliness [16, 40, 43]. Two (out of two) studies found that being widowed is also associated with a greater likelihood of loneliness [16, 40].

It may be worth noting that some other studies used loneliness as independent variable. They found that greater loneliness is associated with a greater likelihood of depression [29], lower quality of life [30], lower life satisfaction [39], and poor self-rated health [45].

Correlates of social isolation

The correlates of social isolation were not examined by any of the studies. In contrast, one study used social isolation as independent variable and found that the occurrence of social isolation was associated with an increased likelihood of experiencing self-care problems (OR: 1.92, 95% CI: 1.01–3.65), and pain/discomfort (OR: 2.01, 95% CI: 1.16–3.48) over time [34].

Meta-analysis and meta-regression

The estimated prevalence of severe loneliness was 27.1% (95% CI: 23.7–30.4%; Fig. 2). There was significant heterogeneity between studies (I2 = 98.7%, p < 0.001). The estimated prevalence of moderate loneliness equalled 32.1% (95% CI: 15.8–48.4%, Fig. 3; I2 = 98.6%, p < 0.001). Moreover, the estimated prevalence of social isolation was 33.6% (95% CI: 28.9–38.2%, p < 0.001; I2 = 88.7%, p < 0.001).

Fig. 2
figure 2

Meta-analysis (severe loneliness)

Full size image
Fig. 3
figure 3

Meta-analysis (moderate loneliness)

Full size image

With regard to sex-stratified prevalences for loneliness: The estimated prevalence of severe loneliness was 33.6% amongst women (95% CI: 6.6–60.7%, I2 = 99.4, p < 0.001), whereas it was 22.7% amongst men (95% CI: 3.0%–42.4%, I2 = 99.0%, p < 0.001; see Additional file 5 for meta-analysis stratified by sex).

Furthermore, our meta-regression analysis indicated that the assessment tool for loneliness and the continent in which the study took place did not significantly influence the prevalence of loneliness (Table 3; a model that considers each category of the two variables individually can be found in Additional file 6). Please note: The coefficients of the regression reflect the predicted change in the logit given a 1-unit change in the moderator variable.

Table 3 Meta-regression analysis of factors affecting heterogeneity (severe loneliness)
Full size table

In a robustness check, we also added the risk of bias scores in a meta-regression (to compare findings from studies at lower and higher risk of bias) [49]. However, this factor also did not achieve statistical significance (p = 0.18). We refrained from doing a meta-regression analysis with prevalence of social isolation due to the small number of studies included. We therefore conducted meta-analysis with prevalence of social isolation for subgroups (by region; by tool used to quantify social isolation; by risk of bias). These findings can be found in Table 4.

Table 4 Subgroup analysis of the pooled prevalence of social isolation
Full size table

The funnel plot (Fig. 4) suggested a potential asymmetry (for loneliness). However, the Egger test (p = 0.75) suggested no potential data asymmetry; this indicates the absence of potential publication bias (Fig. 5).

Fig. 4
figure 4

Meta-analysis (social isolation)

Full size image
Fig. 5
figure 5

Funnel plot

Full size image

Quality assessment/risk of bias assessment

Table 5 shows the quality assessment/risk of bias evaluation. The scores varied from 2 to 8 (mean score: 6.4, SD: 1.6), indicating a moderate to good level, and a comparably low danger of bias. The most common limitation was that the response rate was not clearly displayed/unclear handling of low response rate (all studies).

Table 5 Quality assessment/risk of bias assessment
Full size table

Discussion

The goal of this present work was to identify the prevalence of loneliness and social isolation amongst the oldest old. High prevalence rates were identified. Heterogeneity was observed amongst the studies. The Egger tests indicated the absence of potential publication bias. Meta-regressions conducted to explore the sources of heterogeneity found that neither the assessment of loneliness nor the study continent could be attributed as significant factors contributing to the observed heterogeneity.

The prevalence of loneliness and social isolation amongst the oldest old was high compared to younger individuals. For example, Röhr found a prevalence of social isolation of 12.3% (95% CI: 11.6–13.0) based on the LSNS-6 amongst individuals aged 18 to 79 years in Leipzig Germany about ten years ago (data collection took place between Summer 2011 and Winter 2014) [50]. The prevalence of social isolation identified in this work is, for example, comparable to the prevalence amongst the frequently marginalised group of transgender individuals. This recent study identified a prevalence of 34.4% for objective social isolation based on the LSNS-6 amongst transgender individuals [51]. The data collection took place between April to October 2022. The overall high prevalence rates (for both loneliness and social isolation) of the oldest old may be attributed to the wide variety of mental and somatic disorders which are linked to this very high age [52]. Moreover, individuals aged 80 years and over have a high need for long-term care [53] (e.g. due to functional impairment). A high need for care is associated with higher loneliness levels [54]. Such individuals with a high care need may face difficulties coping with everyday life. For example, mobility restrictions could make it difficult to stay in contact with other people.

With regard to the meta-regressions, it was surprising for us that the loneliness prevalence neither varied by tool used nor the continent in which the study was conducted. This may suggest that loneliness is a general phenomenon amongst the oldest old, and may not be limited to areas or regions where, for example, individuals aged 80 years and over do not live directly or in the immediate vicinity of relatives, and where family cohesion is perhaps also differently pronounced.

Given a sufficient number of studies and data availability, future meta-analyses in this area could further explore other causes of heterogeneity. Those causes could be, for example, educational level, morbidity level or social factors (e.g. social engagement, owning a pet, grandchild care, private care receipt or spousal caregiving) [55,56,57,58]. Moreover, cultural differences, such as differences between individualistic and collectivistic societies, may be a source of heterogeneity and thus should be further explored [59].

It should be noted that the correlates of loneliness amongst the oldest old seem to be comparable to the identified correlates amongst individuals in old age [60]. For example, previous systematic reviews based on cross-sectional studies demonstrated the importance of marital status for both loneliness [60] and social isolation [61] amongst older adults.

However, great caution is required due to the overall very low number of studies investigating the correlates of loneliness amongst the oldest old. Indeed, in view of the limited number of studies examining the correlates of social isolation amongst the oldest old, it is not possible to compare it with prior findings in other age brackets (or other groups) [61].

With regards to study quality, the studies included in this meta-analysis generally exhibited a moderate to high level of methodological rigour. However, some common shortcomings were identified, such as a lack of description regarding response rates/unclear handling of low response rate. For example, a low response rate may reflect the fact that severely impaired individuals (e.g. functional or cognitive impairment) have a lower likelihood of participation; as is commonly found in cohort studies (e.g. [62]). Such impaired individuals often report higher loneliness and isolation scores compared to less impaired individuals. Thus, it may be the case that the prevalence rates reported in this work underestimates the true prevalence rates. Moreover, only very few longitudinal studies have been undertaken and even fewer have exploited the longitudinal data structure using, for example, FE estimates (which provide consistent estimates based on weak assumptions [63]). Overall, one should be very cautious about the causal interpretability of the findings, based on the available evidence.

Our systematic review and meta-analysis showed several gaps in present knowledge. There is a need for a greater number of longitudinal studies to identify the determinants of loneliness, and particularly social isolation amongst the oldest old population. In particular, we recommend the use of techniques to explore causal analysis relationships when dealing with observational data. Such techniques can include, for example, Mendelian randomisation [64, 65], matching approaches such as entropy balancing [66], cross-lagged panel models with fixed effects [67, 68], or difference-in-difference estimators [69].

Additionally, more studies based on more sophisticated tools (e.g. De Jong Gierveld tool or LSNS-6) are required. Moreover, additional studies from neglected geographic areas (particularly: Eastern Europe, South America (except for Brazil), Western Asia, South Asia, East Asia (except for China) and Africa) are required. Lastly, future studies should provide clear reporting of the response rate and should conduct a dropout analysis if necessary. Furthermore, studies are required examining the prevalence of loneliness and social isolation amongst the oldest in times of the challenging COVID-19 pandemic.

We would emphasise some strengths and shortcomings of our own work. It should be noted that this is the first systematic review/meta-analysis synthesising the prevalence and correlates of loneliness and social isolation exclusively amongst the oldest old. Additionally, important procedures were conducted by two reviewers independently. An additional hand search was performed. Furthermore, a meta-analysis and a meta-regression was conducted. A potential shortcoming is that we included solely peer-reviewed articles which may lead to the exclusion of appropriate studies. However, we decided to do so to assure a certain quality of the studies. Whilst most of the studies were published in the past few years, many more studies conducted during the COVID-19 pandemic and investigating individuals aged 80 years and over are needed. Furthermore, we restricted our search to three important databases. However, it may be the case that this choice may lead to an exclusion of studies that may be relevant. If available, other databases (e.g. Embase) should be included in the future research.

Conclusion

Loneliness and social isolation are important problems in the oldest old. In this age bracket, further studies are required from regions outside Europe. Additionally, longitudinal studies are required to investigate the determinants of loneliness and social isolation amongst individuals aged 80 years and over. Furthermore, studies using more pronounced tools to quantify loneliness and social isolation are required.