Systematic Review of Efficacy of Interventions for Social Isolation of Older Adults

Background: The social isolation of older people is a growing public health concern. The proportion of older people in society has increased in recent decades, and it is estimated that ~40% of the population will be aged 50 or above within the next few decades. This systematic review aims to summarize and renew knowledge of the effectiveness of existing interventions for alleviating social isolation of older adults. Methods: Relevant electronic databases, including Cochrane Library, CINAHL, SCOPUS, and Web of Science, were searched by a systematic evaluation method. Eligible randomized controlled trial (RCT) studies were published between 1978 and 2021 in English or Chinese. The primary and secondary outcomes were social isolation and loneliness. The quality of the included RCTs was scored by the Cochrane risk-of-bias tool to assess their quality. Two independent reviewers extracted data, using a standardized form. Narrative synthesis and vote-counting methods were used to summarize and interpret study data. Results: Twenty-four RCTs were finally included in this review. There was evidence of substantial heterogeneity in the interventions delivered. The overall quality of included studies indicated a low-to-medium risk of bias. Eighteen of 24 RCTs showed at least one dimension effect on reducing social isolation. The interventions with accurate targeting of clients in social and public places had more obvious effect. The interventions in which older people are active participants also appeared more likely to be effective. In addition, group intervention activities and individual intervention interviews were effective in improving structural social support; mixed intervention, and group intervention on training support significantly improved functional social support. Conclusions: This study suggests that group and mixed intervention targeting of older adults could be helpful for alleviating social isolation problems. The use of modern technology for remote services could also present good results. More well-conducted RCTs of the effectiveness of social interventions for alleviating social isolation are needed to improve the evidence base. Especially as the debating results of remote interventions, further research in this field should be conducted.

BACKGROUND Social isolation is a major threat to the health of older adults. There are many risk factors in social isolation in old age, including the lack of family members, rare or no daily communication with friends, depression, and a solitary lifestyle (Iliffe et al., 2007). Studies have indicated that social isolation and loneliness are common negative emotions, and social states among older adults that could lead, without timely intervention, to even more serious situations (Laursen and Hartl, 2013). Subsequently, research has indicated that 40% of adults over the age of 50 often felt lonely (Ferreiraalves et al., 2014). Although "loneliness" is often co-emergent and mutually influential with "social isolation, " they are two different concepts (Grenade and Boldy, 2008). Loneliness relates specifically to negative feelings of one about a situation. It may reflect social isolation or a sense of abandonment, resulting from an excessive gap between expectations and reality (Petersen et al., 2020) and increase with age (Li and Zhou, 2002), while the definitions of social isolation incorporate "structural" and "functional" social support (Lu et al., 2013). Social isolation is, therefore, multidimensional and includes the lack of structural and functional social support (Lubben and Gironda, 2003;Victor et al., 2010). In this research, social isolation was divided into two dimensions: "structural social support" and "functional social support." Structural social support is an objective evaluation of the scale or frequency of social support participation (Lubben and Gironda, 2003;Victor et al., 2010); and functional social support is a subjective judgment on the quality of social support, including feelings, tools, and information provided by the perceived responses of others (Hall et al., 2019). According to this definition, social isolation is a multidimensional concept, which results from the lack of quality and quantity of social support (Petersen et al., 2020). The current study adopted this definition as the basis for research.
Social isolation is an essential threat to the health of older adults, and many scholars have provided evidence for methods of alleviating this problem. A meta-analysis conducted in 2010 (n = 308, mean age = 64 years; Lunstad et al., 2010) indicated that social isolation of people with strong social relationships might decrease by 50%. The compound variables used to calculate "strong social relationships" included loneliness and social isolation. Specific studies assessing the relationship between social isolation and health have reached different conclusions. For example, social isolation can lead to increased mortality, worse self-rated health (Cornwell and Waite, 2009), more susceptibility to Alzheimer's disease (Fratiglioni et al., 2000), and an increased rate of disability in older adults (Lund et al., 2010). A recent study suggested that social isolation was negatively correlated with health-related quality of life and health status of older adults (Hawton et al., 2010). Much evidence has accumulated to indicate that social isolation can affect the health of an individual. Therefore, it is an important public health problem. Moreover, the results of interventions for social isolation must be scientifically evaluated to reduce its negative impact.
There are several systematic reviews of this topic. For example, one study summarized interventions for loneliness. However, it does not fully address the effectiveness of interventions for social isolation (Masi et al., 2011). In this article, data were integrated from heterogeneous samples and included out-ofschool children, homeless teenagers, and older adults. Moreover, the interventions included online chat rooms, exercise, social events, and training support. Although there is a debate about the appropriateness of meta-analysis of heterogeneous data, this kind of systematic evaluation of outcome research has seldom been reported. Recently, two systematic reviews have been conducted that included studies before 2016 (Stojanovic et al., 2016;Poscia et al., 2018). However, in these two systematic reviews, there was no quality evaluation process, and RCTs were not included. Moreover, they did not search the three main databases of PsycINFO, PubMed, and Proquest. Since then, many changes have taken place in the social environment. Remote services have been widely adopted, especially with the rapid development of information technology. Remote and other newtech intervention RCTs targeting social isolation in older adults have been published until 2021, which necessitates updating of current knowledge.
Outcomes regarding structural social support and functional social support are important indicators of effect in the multidimensional definition of social isolation used in our review. In addition, reporting on loneliness may also contribute to the understanding of intervention effects. Therefore, this systematic review was designed to summarize and update the current knowledge about the efficacy of existing interventions for alleviating social isolation and loneliness among older adults.

Search Strategy
databases were slightly different. Therefore, we also searched through the reference lists of systematic review articles on social isolation.

Review Strategy
According to the research topic and summary, two researchers (FT and CLY) made a preliminary identification of study criteria. The third researcher (FF) read the abstract of the indeterminate literature and determined the specific discussion about the disagreement. A pair of independent raters selected abstracts for full review based on inclusion/exclusion criteria. Two independent reviewers extracted data, using a standardized form. Due to the heterogeneity of different outcome indicators (e.g., family ties increased, feeling of social support, social relationship), meta-analysis is not suitable for use. According to the analysis method of the three previous evaluations (Díaz and García, 2016;Canedo-García et al., 2017;Li et al., 2018), narrative synthesis and vote-counting methods were used to summarize and interpret study data. The current review was reported in accordance with the latest PRISMA guidance (Page et al., 2021).

Inclusion and Exclusion Criteria
The primary and secondary outcomes are social isolation and loneliness. All papers selected for final inclusion met the following criteria: (i) older adults over 50 years of age with no mental illness or cognitive impairment; (ii) the purpose of the intervention was to alleviate social isolation or loneliness; (iii) the results of social isolation intervention were reported; (iv) there were randomized controlled trials but no drug trials; and (v) the paper was written in Chinese or English. Exclusion criteria for the study: (i) study samples aged younger than 50 years; (ii) not used a randomized controlled trial (RCT); (iii) drug intervention was used; and (iv) outcomes reporting only on loneliness but no social isolation.

The Quality Evaluation of the Research
Because of the heterogeneity of the intervention types and results of the trial, quantitative analysis of data was not used in the review, so the method of narrative synthesis was applied to analyze the effect of interventions. In the quality evaluation of open randomized controlled trials, we chose not to use the Jadad standard (Berger, 2006) as this is focused on blind and random sequences; therefore, the Cochrane risk-of-bias tool was deemed more appropriate (Ma et al., 2012). In this paper, based on the Cochrane risk of bias, the quality of the randomized control trial and the bias risk level were identified, and the grading principle of JADAD was used to evaluate the overall research quality. The Cochrane bias-risk tool evaluation principle involves six aspects: selection bias, implementation bias, measurement bias, data bias, publication bias, and other bias (Higgins et al., 2011). The system evaluation report is based on the PRISMA (preferred reporting items for systematic reviews and meta-analyses) standard (Page et al., 2021).

RESULTS
About 746 items were found in the related research, with 452 duplicates removed, 268 of the studies excluded as they did not meet the selection criteria. Two studies were excluded because of high-bias risk. Twenty-four studies were eventually included (Figure 1).

Inclusion of the Study
A total of 24 randomized controlled trial studies were included with a total of 4,078 subjects, each involving 26-708 cases. Table 1 (including two high-bias risks) introduces the characteristics of these studies in accordance with the principle of PICOS, including: population, intervention, comparison, outcomes, and study type (Methley et al., 2014).
Of all the studies, there were only two studies from mainland China (Hang et al., 2011;Yi et al., 2012), and one of them    belonged to high risk of bias. The rest of the studies were from Hong Kong, Europe, and other developed countries. The United States occupied 10 studies, while the low-risk bias research was mostly from Finland (Ollonqvist et al., 2008;Routasalo et al., 2009). In terms of the intervention forms, there were three main categories: group intervention, individual intervention, and mixed intervention. Fourteen studies were conducted using group interaction interventions (e.g., Harris and Bodden, 1978;Constantino, 1988;Lökk, 1990;Ollonqvist et al., 2008;Routasalo et al., 2009), seven studies used individual interventions (e.g., Schulz, 1976;Heller et al., 1991;Brennan et al., 1995;MacIntyre et al., 1999;Yi et al., 2012), and five studies combined the above two approaches (Drentea et al., 2006;Hang et al., 2011). The three intervention types could be subclassified into seven subtypes: intervention activities provided, group intervention training support, group intervention in remote service, face-toface individual intervention, individual interventions in remote service, mixed interventions in remote service, and mixed interview intervention. Among them, seven items were group intervention activities-provided studies (e.g., Harris and Bodden, 1978;Constantino, 1988;Lökk, 1990;Ollonqvist et al., 2008;Routasalo et al., 2009), eight items were group intervention training support studies (Fukui et al., 2003;Savelkoul and de Witte, 2004;Kremers et al., 2006;Bøen et al., 2012;Saito et al., 2012), one item was a group intervention in remote service study (White et al., 2002), three items were face-to-face individual intervention studies (Schulz, 1976;MacIntyre et al., 1999;Yi et al., 2012), four items were individual interventions in remote service studies, two items were mixed interventions in remote service studies (Mountain et al., 2014;Czaja et al., 2017), two items were mixed interventions in remote service studies (Drentea et al., 2006;Hang et al., 2011), and three items were mixed interview intervention studies.
With regard to the time and frequency of intervention, most of the intervention frequency was regular, one time or two times per week. Most interventions lasted from 6 weeks to 1 year, and there was one study that lasted 5 years (Drentea et al., 2006); one study did not elaborate on the intervention frequency information (Heller et al., 1991). Among them, the primary recipient of the intervention included caregivers, disabled people, family members, older adults, and older adults living alone in the community. Only 50% (13/26) of the studies were specifically focused on social isolation or isolation (e.g., Harris and Bodden, 1978;Savelkoul and de Witte, 2004;Routasalo et al., 2009;Black et al., 2014;Chan et al., 2017), while the rest of the studies were secondary or indirect observations of variables. Intervention practitioners were health commissioners or professional social workers (e.g., Lökk, 1990;Savelkoul and de Witte, 2004;Ollonqvist et al., 2008;Routasalo et al., 2009;Saito et al., 2012), teachers (White et al., 2002;Czaja et al., 2017), students (Schulz, 1976;Constantino, 1988;MacIntyre et al., 1999), or experts. One study involved all of the above staff (Bøen et al., 2012), and one study did not specify the identity of the intervention practitioner (Harris and Bodden, 1978). In studies that featured control conditions, the control or comparison intervention included setting the control group (e.g., Constantino, 1988;Kremers et al., 2006;Ollonqvist et al., 2008;Routasalo et al., 2009;Black et al., 2014), conventional care, and waiting-list control; four studies used a variety of cross interventions (Schulz, 1976;Savelkoul and de Witte, 2004;Slegers et al., 2008;Mountain et al., 2014); and six studies conducted remote interventions (e.g., Heller et al., 1991;Brennan et al., 1995;Morrow-Howel et al., 1998;Slegers et al., 2008;Mountain et al., 2014). In addition, between 6 months and 3 years after the baseline review, seven studies conducted only one followup (e.g., Schulz, 1976;Harris and Bodden, 1978;MacIntyre et al., 1999;Ollonqvist et al., 2008;Black et al., 2014). Thirteen studies included two to four follow-up visits in 2 years after the intervention (e.g., Constantino, 1988;Lökk, 1990;Routasalo et al., 2009;Chan et al., 2017). One of the studies collected followup data 11 times during the 5 years of the study (Drentea et al., 2006).

Methodological Quality
In order to evaluate study quality and risk of bias, the Cochrane risk-of-bias tool was applied (see Table 2). Seven studies were classified as low risk of bias (e.g., Savelkoul and de Witte, 2004;Ollonqvist et al., 2008;Routasalo et al., 2009;Bøen et al., 2012;Chan et al., 2017), two studies were classified as high risk of bias (Schulz, 1976;Hang et al., 2011), and the rest of the 17 studies were rated as moderate risk of bias. Two studies with high-bias risk will not be discussed further. The remaining 24 studies will be discussed in the following.
The overall quality of the study continued to improve over time. Seven out of the eight intervention studies before 2000 (e.g., Schulz, 1976;Harris and Bodden, 1978;Constantino, 1988;Lökk, 1990;MacIntyre et al., 1999) were rated as moderate risks. Since 2000, 10 of the 18 studies were moderate bias risk; among which, seven were low bias risk.

Intervention Characteristics and Effects
Overall, outcomes labeled with "Y" means the intervention had significant effect on this variable, while "N" indicates no significant effect. Nineteen of the 24 intervention studies showed improvement in social isolation in at least one dimension (e.g., Harris and Bodden, 1978;Constantino, 1988;Lökk, 1990;Routasalo et al., 2009;Black et al., 2014). There was a diversity of definitions and methods of measuring social isolation where it was unclear on how best to categorize all outcomes that were grouped as "social isolation." Where there was sufficient information about type of a social isolation outcome being measured, studies were categorized as structural social support (such as emotional or psychological support) or functional social support (such as instrumental support) ( Table 3).

Intervention Effects According to Intervention Methods
Generally, according to the classification of different intervention methods, there are 14 group interventions, 8 of them were group activities, 5 of them focus on social support training, and 1 was conducted in a remote manner. Moreover, there are six individual interventions; two of them are interviews, and the rest of four are remote service. In addition, there are four mixed interventions, and two of them are remote service.
As to intervention methods, six of the eight group intervention activities improved structural social support (Harris and Bodden, 1978;Constantino, 1988;Lökk, 1990;Routasalo et al., 2009;Black et al., 2014;Lai et al., 2020), while various forms of outcome measures were conducted. For instance, one low-risk physical exercise study showed no obvious improvement in loneliness and structural social support (Ollonqvist et al., 2008), while Taiji physical exercise showed great effect (Black et al., 2014;Chan et al., 2017). Three group interventions focused on functional social support reported significant improvements (Fukui et al., 2003;Bøen et al., 2012;Saito et al., 2012). Four structural social support studies (Fukui et al., 2003;Savelkoul and de Witte, 2004;Kremers et al., 2006;Saito et al., 2012) reported that two out of four patients had no improvement effect or the effect disappeared over time, while a few studies reported significant effects (Fukui et al., 2003;Saito et al., 2012). A moderate risk bias group intervention conducted earlier with remote services found no improvement (White et al., 2002), while another mixed remote service intervention had effects on structural social support (Mountain et al., 2014). Two individual studies that involved face-to-face interviews showed significant improvement in structural social support (MacIntyre et al., 1999;Yi et al., 2012). One of the four older (before 2010) individuals involved in a remote service study (Heller et al., 1991;Brennan et al., 1995;Morrow-Howel et al., 1998;Slegers et al., 2008) showed improvement effects on structural social support, but the follow-up effect was very short (Morrow-Howel et al., 1998). Two studies with moderate risk bias conducted by mixed intervention showed improvement in functional social support (Czaja et al., 2017;Lai et al., 2020).

Intervention Effects According to Intervention Environment
The external environment of intervention, such as the intervention practitioner, the place of intervention, and the client, was also an important factor in the effect of intervention. Five of the six interventions provided by experts showed improved outcomes (Heller et al., 1991;Fukui et al., 2003;Kremers et al., 2006;Mountain et al., 2014;Czaja et al., 2017). Seven of the 10 interventions provided by health or social workers were also effective (Lökk, 1990;Ollonqvist et al., 2008;Saito et al., 2012;Black et al., 2014;Chan et al., 2017;Lai et al., 2020;Ristolainen et al., 2020). Four interventions provided by teachers or students of the education community presented improved results (Constantino, 1988;MacIntyre et al., 1999;White et al., 2002;Slegers et al., 2008). In addition, one study multiple types of intervention practitioners (Bøen et al., 2012), another study didn't specifically described the information of intervention providers (Harris and Bodden, 1978).

Intervention Effects According to Duration of Effect
Among the 14 studies reporting structural social support effect, three of them used social support as the outcome (Saito et al., 2012;Chan et al., 2017;Czaja et al., 2017), two studies observed the change of new friend number as an outcome (Routasalo et al., 2009;Bøen et al., 2012). One study showed that 45% of the participants made new friends in 1 year (Routasalo et al., 2009), while another showed that 40% of the participants made new friends in 1 year (Bøen et al., 2012), two studies reported using social contact as an outcome (Harris and Bodden, 1978;. And all studies reporting functional social support effect took completely different indicators during 6-12 months.

DISCUSSION
This study found substantial heterogeneity in the interventions delivered, and the overall quality of included studies indicated a low to medium risk of bias. Also, group intervention activities and individual intervention were effective in improving structural social support; mixed intervention and group intervention on training support significantly improved functional social support. We found that the interventions with accurate targeting of clients in social and public places had   more obvious effect. Interventions in which older people were active participants also appeared more likely to be effective. In addition, professionals were better than teachers and students in conducting intervention. The findings provide a tentative indication of the potential benefits of specific types of intervention for improving loneliness/social isolation, advancing theory-informed development of interventions and improving design of evaluation studies. The remote service interventions were debatable, as the recent studies have showed improvement in structural social support, but no effect on older studies. Because of the contradictory results, more research is needed to examine the complexity of "remote interventions" from the perspective of process evaluation. Interventions conducted in social and public places had better effects, and interventions with accurate targeting of clients had more obvious effects. Studies evaluating interventions delivered by professional practitioners appeared to yield better outcomes than those where the intervention was delivered by non-professionals.
Effective intervention for older adults in isolation not only improved structural social support, functional social support, and mitigation of loneliness but also promoted the health of older adults.
In the experimental studies, there were a variety of interventions on social isolation. Although experimental design is not always feasible or accepted by participants, this kind of study can provide a scientific and normative reference for the implementation process and assessment report, promote the utilization of randomized control trials, improve the design level, standardize the research process, improve the quality of evidence, and provide a reference for policy-making. We advocate professionals to provide face-to-face intervention in the field of daily life rather than in the home environment and recommend that more efficient remote interventions within smart terminals be developed to achieve better results.
In real life, the environment preference of older adults has an obvious effect on their social interaction. Older adults who enjoy being alone are more likely to be socially isolated. The incidence of social isolation among older adults in different living conditions was also different, with those who were widowed, had low income, and in poor health, more likely to feel lonely and socially isolated. In addition, as age increases, older adults can be more dissociated from social interaction and prefer to be isolated (Lu et al., 2013). Therefore, when we design social isolation interventions, it is essential to consider personal preference, living status, and physiological characteristics of older adults and adjust measures accordingly so as to promote the effectiveness of the intervention. In addition, well-designed remote intervention system, such as personal reminder information and social management (PRISM) system, has the potential to change attitudes toward technology and increase technology self-efficacy.
At the policy level, the establishment of social support systems is imminent (Liu and Ni, 2002). With the advancement of family planning policy, such as China, the aging of the population is becoming more and more serious, and the "4-2-1" or "4-2-2" family pattern (4-grandparent, 2-parent, and 2-or 1-child) has gradually formed (Nan and Dong, 2019). Family support functions have greatly weakened, and, especially, the needs for social interaction and spiritual comfort are not satisfied. Therefore, we must establish a community-based pension support service platform, develop professional social work vigorously, cooperate with research institutes to obtain scientific evidence in order to address the problem of social isolation of older adults, improve their physical and mental health, as well as quality of life, and promote the healthy aging of the population.

LIMITATIONS AND FUTURE RESEARCH DIRECTIONS
In this study, the inclusion literature was defined as older adults over age 50, who have been in isolation or loneliness. However, the relevant research on the concept of social isolation does not use a standardized and unified definition, so inclusion bias may have been incorporated. Although the inclusion criteria were designed to reduce social isolation or loneliness, only 14/24 studies specifically addressed the problem (e.g., Constantino, 1988;Li and Zhou, 2002;Savelkoul and de Witte, 2004;Kremers et al., 2006;Chan et al., 2017). The study may also have the potential risk that the assessment of social isolation or loneliness was due to other characteristics of the target client (Liu and Ni, 2002). Restricting the study language to English and Chinese may have increased the inclusion bias. The quality and expression of the research in the historical period also limited the quality of this study. Some studies conducted a qualitative report rather than quantitative data. It is not appropriate to use quantitative methods as well as meta-analysis due to the heterogeneity of the study subjects.
At the same time, we found that most pieces of randomized controlled trial research in this field were from developed countries. Future research not only needs to enrich the original evidence from all over the world but especially from developing countries. In addition, most of the pieces of research from Finland, Norway, the United States, and other developed countries were different from developing countries due to legal or volunteer service organizations; thus, the applicability and the effectiveness of the evidence are worth discussing further. Moreover, more refined subgroups of systematic review can be done in the near future; for example, systematic review could be used to quantify the effect of intervention on a certain type of intervention.

CONCLUSION
The findings provide a tentative indication of the potential benefits of specific types of intervention for improving loneliness/social isolation, advancing theory-informed development of interventions, and improving design of evaluation studies.
Firstly, this study suggests that group and mixed intervention targeting of older adults could be helpful for alleviating social isolation problems. The use of modern technology for remote services could also present good results. Moreover, our systematic review has identified a need for well-conducted studies to improve the evidence base regarding the effectiveness of social interventions for alleviating social isolation. However, more well-conducted RCTs of the effectiveness of social interventions for alleviating social isolation are needed to improve the evidence base.
Because of the debating results, further research is needed to examine the effect of remote interventions from the perspective of process evaluation.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
FT is responsible for analyzing data and writing the draft of the results. CLY is responsible for analyzing the data. FF is responsible for revising the manuscript and submitting the paper. LSW is responsible for writing the literature review. IC is responsible for providing suggestions for revising the paper. All authors contributed to the article and approved the submitted version.