A comprehensive systematic search was conducted across eight electronic databases from inception to 30 June 2024. The following databases were queried: EMBASE, PubMed, the Cochrane Library, Web of Science, China National Knowledge Infrastructure (CNKI), China Science and Technology Journal Database (Weipu), WanFang Database, and China Biology Medicine disc (CBM). Search terms were developed by combining Medical Subject Headings (MeSH) and free-text keywords related to three domains: loneliness (e.g., “social isolation” or “loneliness” or “social exclusion” or “social exclusions” or “ostracism” or “social alienation”), old adults (e.g., “aged” or “elderly” or “oldest old” or “agenarians” or “nonagenarian” or “octogenarian” or “centenarian” or “old people or “old person” or “old adult”), and social support interventions (e.g., “social support” or “social care” or “psychosocial support” or “psychological support”). Citation chasing was performed by manually screening the reference lists of all included articles and relevant reviews to identify additional studies. Only studies published in English and Chinese were included, due to database coverage and the language proficiency of the review team. Grey literature was not included, as this study focused on peer-reviewed evidence.

Inclusion criteria for the studies were as follows: (a) individuals aged 60 years or older residing in the community; (b) the experimental group received social support interventions of any kind; (c) validated tools like the Los Angeles Loneliness Scale from UCLA (UCLA), the De Jong Loneliness Scale (DJLS), the PROMIS-L Social Isolation Scale (PROMIS-L), Ando-Osada-Kodama (AOK) loneliness scale; (d) comparisons were made against passive (usual care) or active (non-social support) control groups; (e) study design was randomized controlled trial (RCT); (f) publications were in English or Chinese. The exclusion criteria were (a) no full text; (b) studies that reported duplicate data.

The research selection and data extraction process involved three reviewers. Two reviewers autonomously chose all research papers based on their titles, abstracts, and full texts, following the specified inclusion and exclusion criteria. To address any inconsistencies or disputes that arose during the selection process, a third reviewer was brought in for consultation.

Data were extracted using astandardized form. Included in the data were basic information, participant details (including number, eligibility, age, and gender), intervention/exposure specifics (type, frequency, duration, content, comparison with control group, delivery format, provider information), outcome data (effect size, standard deviations, statistical significance), and instruments used. Data missing for relevant outcomes were obtained from original authors via email.

Two researchers independently evaluated the methodological rigor of the studies included, utilizing the Joanna Briggs Institute (JBI) critical appraisal checklists for randomized controlled trials (RCTs) (Munn et al., 2020). The checklist for randomized controlled trials includes 13 items that can be answered with “yes,” “no,” “unclear” or “not applicable,” and a “yes” response is worth one point. Referring to the JBI manual and previous research, RCTs with scores less than seven points were identified as weak quality studies. Any disagreements among reviewers were settled through consensus or conversation with a third reviewer.

Review Manager (RevMan 5.3) and R Studio softwares were used for quantitative analysis of selected studies. The research utilized a narrative approach that was not suitable for quantitative analysis in the meta-analysis. The inverse variance method was employed in meta-analysis to determine the weights assigned to the studies. Effect size was determined by calculating the mean and standard deviation (SD) of the difference between scores at the last follow-up and baseline. In the pooled analysis, mean differences (MD) or standardized mean differences (SMD) were utilized as effect estimates along with their corresponding 95% confidence intervals (CI). SMD is utilized when assessing the identical result using a distinct tool compared to the identical result assessed using the identical tool. Statistical heterogeneity was assessed using the Chi-squared test and I² statistic. A random-effects model was adopted when significant heterogeneity was detected (I² > 40% or p < 0.05); otherwise, a fixed-effects model was used. Sensitivity analyses were conducted by excluding lower-quality studies to examine the robustness of results. Sensitivity tests were performed to evaluate the impact of a low-quality study on combined results. Publication bias was evaluated visually via funnel plots and statistically using Egger’s regression test. An analysis of subgroups was conducted to investigate how various lengths and forms of social support impact results.

A total of 2,420 articles were retrieved in Figure 1. Out of the articles reviewed, 789 duplicates were removed using Endnote software, 1,490 articles were eliminated based on title and abstract, leaving 141 articles for full-text screening. This study ultimately incorporated 19 research studies (Banks and Banks, 2005; Chang et al., 2018; Choi et al., 2020; Cohen-Mansfield et al., 2018; Galinha et al., 2021; Gilbody et al., 2021; Gu et al., 2019; José. et al., 2022; Ilgaz et al., 2023; Kremers et al., 2006; Lai et al., 2020; Larsson et al., 2016; Ristolainen et al., 2020; Rodríguez-Romero et al., 2021; Saito et al., 2012; Wang et al., 2014; Wei et al., 2021; Yang et al., 2023; Zhou et al., 2014).

The characteristics of the included studies are displayed in Table 1. China (n = 7) and United States (n = 4) accounted for the majority of the studies conducted. The 16 studies included 1,809 subjects, with sample sizes varying from 14 to 392 older adults. The average age of the population fell between 60.00 and 84.17 years. The types of interventions were divided into single intervention (n = 12) and multiple intervention (n = 7). The single intervention were various, including group counseling (Gu et al., 2019; Wang et al., 2014), tele-support (Choi et al., 2020; Gilbody et al., 2021), social activity (Galinha et al., 2021; Larsson et al., 2016), animal-assisted therapy (Banks and Banks, 2005), professional support (Zhou et al., 2014), peer support (Lai et al., 2020), Group intervention of Technology (Wei et al., 2021), support from themselves (Kremers et al., 2006), and I-SOCIAL intervention (Cohen-Mansfield et al., 2018). The interventions in control group were usual care (n = 12), follow-up visits (n = 2), no treatment (n = 2), and tele-support (n = 3). The intervention duration ranged from 4 weeks to 3 years. Half of the studies conducted follow-up after intervention, but other studies did not describe the follow-up period. Follow-up duration varied between 1 week and 6 months. The level of loneliness was assessed through various scales including UCLA (n = 12), DJLS (n = 4), PROMIS-L (n = 1), AOK loneliness scale (n = 1), and loneliness scale for the elderly (n = 1). There are numerous variations of the instruments, with UCLA having five versions and DJLS having two.

The summary of the bias risk in the 19 studies was provided in supplementary material. 17 RCTs met 7–13 items. No randomized controlled trials achieved complete blinding of participants, providers, and evaluators at the same time. Five randomized controlled trials failed to meet the five criteria of the JBI critical appraisal checklist. Twelve randomized controlled trials did not clearly report allocation concealment, did not conduct intention-to-treat analysis, and did not blind participants, implementers, and outcome assessors.

A meta-analysis was performed on 19 studies in Figure 2 to investigate how social support can reduce loneliness. The analysis included 1809 participants, with 892 in the experimental group and 917 in the control group. The primary analysis showed a significant impact of the intervention compared to the control group in decreasing feelings of isolation [SMD = −0.60, 95%CI (−1.00, −0.20), P = 0.003, random effect model], suggesting a substantial effect with a wide yet meaningful confidence range. The studies showed a significant amount of diversity (I ² = 93%, P < 0.001), suggesting a substantial level of variation among them. Therefore, further subgroup analysis was conducted to examine potential factors explaining the heterogeneity among these studies.

Given the potential impact of intervention duration on effectiveness, this study categorized duration into three subgroups. The findings indicated that participants in the experimental group experienced notably reduced feelings of lonely in comparison to those in the control group within less than 3 months. The combined results were [SMD = −0.68, 95%CI (−1.31, −0.06), I ² = 93%, P = 0.03, random effect model]. No significant variances were noted between the intervention and control groups when the intervention lasted 3–6 months or more than 6 months [SMD = −0.66, 95%CI (−1.50, 0.18), I ² = 96%, P = 0.13, random effect model] and [SMD = −0.17, 95%CI (−0.41, 0.06), I ² = 0%, P = 0.52, random effect model] in Figure 3.

It stratified interventions into two subgroups based on the potential impact of intervention types on efficacy, including single intervention and multiple intervention. The findings indicated that the multiple intervention group had notably reduced feelings of loneliness in comparison to the control group. The pooled results were [SMD = −1.26, 95%CI (−2.20, −0.32), I ² = 97%, P < 0.001, random effect model]. In contrast, the subset that received single treatment did not display a notable distinction between the groups that received intervention and those that did not [SMD = −0.22, 95%CI (−0.55, 0.12), I² = 82%, P = 0.200, random effect model] in Figure 4.

According to the funnel plot analysis and the Egger test conducted using R Studio on 19 included papers in this study (t = −1.43, df = 17, p = 0.171), suggesting low probability of publication bias in Table 2 and Figure 5.

There are multiple advantages to this research. This study explored the effects of different social support on elderly loneliness. Analyzing various durations and types of interventions further assessed the impact of social support on loneliness in elderly individuals. Additionally, the present research has its own constraints. Due to inconsistent classification of social support in the included studies is confused, it is not possible to analyze subgroup based on a certain classification and determine which specific type of social support is more effective for elderly loneliness. Although the included studies provided valuable insight, some contextual factors such as gender, socioeconomic status, and rural–urban differences were rarely reported or inconsistently measured, precluding subgroup analysis. The instruments of outcomes included in the study were validated tools, there were too many versions available. It may contribute to statistical heterogeneity. Despite efforts to reach out to the authors, numerous studies were excluded during the literature screening process due to insufficient data. This may lead to a potential bias about our results. Variations in national healthcare systems and community care structures may influence the implementation and sustainability of social support interventions. Additionanlly, differences in intervention providers can lead to inconsistencies in quality and intensity, contributing to heterogeneity. Future research should further examine how these factors affect intervention outcomes.