Rapid systematic review of systematic reviews: what befriending, social support and low intensity psychosocial interventions, delivered remotely, are effective in reducing social isolation and loneliness among older adults? How do they work?

Search strategy

Searches of 11 bibliographic databases and online resources across the fields of health, social care, psychology and social science were carried out on 23^rd-24^th April 2020. We searched: Applied Social Sciences Index and Abstracts (ASSIA)(Proquest), Emerging Sources Citation Index (Web of Science), Database of Promoting Health Effectiveness Reviews (DoPHER), Epistemonikos, Medline (OVID), NHS Evidence, PsycInfo (OVID), Social Policy and Practice (OVID), Social Sciences Citation Index (Web of Science), Social Systems Evidence and Sociological Abstracts (Proquest).

The search terms reflected four concepts that needed to be present in each of the study citations:

An example search history for Medline is presented in the full report¹⁰ and as Extended data⁶⁹.

Inclusion and exclusion criteria

Inclusion and exclusion criteria were also based on the Population, Intervention, Comparator, Outcome and Study Design (PICOS) framework:

Population: We included reviews on ‘older’ adults age 50+ (see protocol for further details¹³). Participants could be located in a variety of settings in the community or residential care, although reviews of interventions delivered to older adults in hospital settings were excluded. Studies included older people who were socially isolated, lonely, or who were otherwise at risk of loneliness or isolation.

While we identified all reviews on older people, we only synthesised evidence from reviews focussed on diverse populations of older people. We did not synthesise evidence from reviews focussed exclusively on particular groups of older people, specifically older caregivers (see 10), although interventions including caregivers are well represented in the evidence presented. The descision to synthesise evidence from a subset of reviews was inline with the rapid timescales of the review (see (Figure 1).

Figure 1. Flow of studies through the review.

Intervention: Included reviews examined interventions that sought to reduce levels of social isolation or loneliness, through strengthening individuals’ social contacts and social relationships (e.g. befriending and social support), or through low intensity psychosocial interventions (e.g. internet-delivered CBT - iCBT), using remote methods and technologies. Interventions were delivered on a one-to-one basis (e.g. befriending), or as remote group-based interventions (e.g. discussion groups). We did not include interventions that examined the use of social robots, pets or virtual pets, or reviews solely focused on the use of technology for educational or training purposes.

Comparator/control: We included reviews that included studies with most forms of control group (randomised and non-randomised) and those without a control group (pre-post designs). Reviews on the implementation of interventions, including qualitative evidence syntheses were also included.

Outcomes: Included reviews measured social isolation or loneliness as a primary outcome. Based on previous reviews, we expected various measures of loneliness and social isolation to be reported, and we included both bepsoke measures and proxies, such as reports of social connectedness.

Study design: We defined systematic reviews as those that met at least four of the following criteria¹⁴:

We did not include any other reviews of reviews, but used these to identify additional systematic reviews. We included unpublished manuscripts. We did not place any restrictions on date of publication, although only reviews in English were selected.

Study selection and data extraction

We exported search records to EPPI-Reviewer web¹⁵ and de-duplicated the records. Title and abstract screening was undertaken independently by three reviewers (DK, EB, PH) following joint screening of 204 citations (10%) to ensure consistency. For records included for full-text screening, each record was examined in duplicate, and reviewers met online to reconcile any differences. Reasons for exclusion are reported in Figure 1.

Systematic reviews in this area often include a mix of eligible and ineligible interventions. In line with previous overviews¹⁶, and in addition to the criteria outlined above, systematic reviews were included if they:

Interventions in scope were befriending, social support, and low intensity psychosocial interventions, delivered remotely, to reduce social isolation and loneliness among older adults. We did not include reviews where only a single study within the review met our criteria. Individual studies reported within systematic reviews were identified as relevant, using the same inclusion criteria as above (albeit applied at the study, not review level) and after agreement of two reviewers.

Data were extracted by two reviewers and any differences agreed in online reconciliation meetings. We extracted the following data from reviews:

• Lead author and team;

• Year of publication;

• Number of primary studies included in the review;

• Primary study design(s) (e.g., RCT studies, qualitative studies);

• Aims of review and main topic focus; (e.g. if focussed on social isolation/loneliness);

• Target population (e.g., if focussed on particular group e.g. bereaved older people);

• Participant characteristics (e.g., age, gender);

• Intervention approaches in primary studies (e.g., type of remote intervention);

• Synthesised outcomes/key findings relating to social isolation and/or loneliness; secondary outcomes relating to implementation and adverse effects;

• Quality assessment characteristics and rating.

Critical appraisal

Included systematic reviews were critically appraised using AMSTAR-2¹⁷ by two reviewers (DK/BH and EB/PH). Criteria were summed and categories of quality created based on the AMSTAR-2 assessment (low risk of bias: equivalent to high confidence in AMSTAR-2; unclear: equivalent to moderate confidence; and high risk of bias: equivalent to low or critically low confidence).

Data synthesis

Descriptive analysis of reviews and studies. We produced textual descriptions of the reviews and their findings and presented this in tabular form to develop a preliminary understanding of the evidence. The results also helped to populate an evidence map (see later synthesis).

Narrative synthesis of the evidence. A narrative synthesis was conducted to examine review-level and study-level findings. The narrative synthesis focussed on the outcomes of befriending, social support, and low intensity psychosocial interventions delivered remotely. Building on the descriptive analysis, we followed guidance outlined elsewhere¹⁸.

Intervention Component Analysis and Qualitative Comparative Analysis. We drew on two complementary synthesis methods – Intervention Component Analysis (ICA) and Qualitative Comparative Analysis (QCA) – and applied these to primary studies contained within the reviews that presented quantitative results, to understand how interventions ‘worked’. The first approach, ICA¹⁹, is an inductive approach developed in response to the poor reporting of intervention processes. It involves (a) inductively coding the nature of intervention features (i.e. components) and (b) using trialists’ informally-reported experiences of implementing the intervention (i.e. information usually located in introduction and discussion sections of trial reports, which is usually not incorporated into analysis)¹⁹. This information is then used in conducting the QCA.

The second approach, QCA, is applied to numeric data and is based on set-theory²⁰. QCA is employed as a solution to the challenge of analysing data containing a small number of studies (known as cases in QCA terminology), each with an extensive array of factors that may trigger a given outcome²¹. This ‘small N-many variables’ challenge is similar to that faced by systematic reviewers, and Thomas and colleagues provide one of the first examples where QCA was utilised within a systematic review to understand configurations of intervention components that were aligned with ‘successful’ interventions²². Studies were eligible for QCA if they reported quantitative findings (see Results). We identified studies as belonging to both ‘condition sets’ (i.e. belonging to a distinct set of studies distinguished by the presence or absence of different characteristics or processes) and ‘outcome sets’ (i.e. belonging to a group of studies differentiated by whether they are considered most effective or least effective). Ultimately, we were interested in establishing which condition sets ‘overlapped’ with successful outcome sets. The goal of QCA is to identify the simplest expression of characteristics/processes that lead to effective interventions; to find the simplest expression we drew on Boolean minimisation. We followed standards of good practice that have been laid out elsewhere in conducting the QCA²³. Further explanation of the approach is provided alongside the results.

Review and study characteristics

The literature searches identified 2,715 citations. After duplicates were removed, 2,057 citations were screened at title and abstract level, identifying 75 possible studies for inclusion. Full texts were obtained for all 75 records, with nine potential reviews identified and five included for synthesisⁱ (see Figure 1). Not all of the primary studies within these five reviews met our inclusion criteria (see Methods) and from the 112 primary studies included across the five reviews, we identified 18 studies as eligible for synthesis.

Review populations. The reviews covered a range of populations, using different definitions and age thresholds for ‘older adults’, with a combined age range of 50–95. The settings were not always clearly stated, but were primarily older adults’ own homes, nursing homes, or supported living facilities, in North America, Europe and Taiwan. Whilst some reviews contained studies focused on the general older adult population, others included studies of people with multiple chronic conditions, specific conditions (such as Alzheimer’s Disease, or breast cancer), or in a particular geographical area.

Review study designs. RCTs, quasi-experimental cohort studies, survey studies, and qualitative (semi-structured interviews and focus groups) were all represented. Three of the five reviews conducted quality appraisals on the included studies^25–27, one evaluated only the effectiveness of the technologies within the studies, not the quality of the study itself²⁸, and one did not report any quality appraisal²⁹ⁱⁱ.

The reviews contained studies reporting interventions using various technologies to deliver remote befriending, social support or low intensity psychosocial interventions including those in scope (e.g. video-communication and telephone befriending) and those out of scope (e.g. computer training and internet training). There was a range of different outcome measures within the reviews, although all contained some measure of loneliness or social isolation.

Risk of bias assessment of included reviews. All of the reviews were deemed to be of low or critically low quality (displayed as having a high risk of bias in Table 2). Although all had reasonably clearly defined PICO components and had conducted reasonably comprehensive search strategies, the majority had failed to prepare a protocol, and many failed to justify the choice of study selection. This latter concern was particularly problematic where authors had included studies of various designs.

Primary study characteristics. Befriending, social support and low intensity psychosocial interventions reported in the 18 primary studies fell into five categories reflecting modes of delivery:

Further primary study characteristics, including their populations, details of implementation, methodological details including how the outcome was measured, and outcomes as found in Table 1.

Narrative synthesis of findings

Supported videoconferencing to alleviate loneliness. Four reviews included a total of three qualitative studies, three quantitative studies, and one mixed-methods study, on supported video-communication interventions. Five studies involved supporting older adults to communicate with family members^35–38,51, with the other two reporting on the videoconferencing element of the ACTION service in Sweden and Norway^34,46. The qualitative evidence suggests the interventions were generally regarded positively by older adults, with increased feelings of connection with their family members³⁵. The quantitative evidence showed some evidence of decreases in feelings of loneliness and increases in social support scores. Two quantitative studies^37,38 found reduced feelings of loneliness at one week, three months and 12 months, although this achieved statistical significance in only one studyⁱⁱⁱ³⁷. Torp et al.’s⁴⁶ mixed-methods pilot cohort study, employing questionnaires and focus groups, also found that the video phone was important for building and maintaining relationships. It is important to note that all interventions included ongoing support to use the technology. See Table 1 for contextual details of reviews and studies on videoconferencing.

Telephone befriending to reduce social isolation. Two reviews included a total of one qualitative and one quantitative study reporting on forms of telephone contact, one of which was a study of telephone befriending. Cattan et al.³⁹ reported on the Call in Time intervention, with qualitative findings from 40 participants. Telephone calls were made to older adults by volunteers, with a project co-ordinator managing the process. Findings included reduced feelings of social isolation, loneliness, depression and anxiety; improved state of mind, contentment with life, confidence level, and physical health (less pain). This study built on an earlier evaluation report that presented data used for the QCA⁵²; this evaluation report was not directly included in any of the reviews, but quantitative data presented within this report suggested that participants had lower wellbeing and social support after the intervention, albeit with a number of caveats.

The only other included study to incorporate telephone contact was Gustafson et al.⁴⁰, from the Morris et al.²⁷ review, where one element of the intervention was to match up participants with peer advocates, who engaged in weekly phone calls. This was not a telephone befriending service, as the peer advocate had a different role to that of a befriender. Findings showed that, of those who used a peer advocate 77.3% felt somewhat or very much connected with their peer advocate, and 81.6% felt that the peer advocate helped them cope (somewhat or very much so) with their breast cancer. Perceived social support increased significantly over the four months, but the intervention included more elements than just telephone support (computer and internet training, discussion group, ‘ask an expert’ service and written guidelines).

Online discussion groups/forums to reduce social isolation and/or loneliness, or to improve/maintain social connectedness. Two reviews contained eight quantitative studies and one mixed-methods (questionnaires and focus groups) on online discussion groups and forums. The studies included synchronous and asynchronous communication: real-time chat discussions, instant peer-messaging, email contact with professionals, and discussion boards. Interventions were designed to support women with chronic illness or breast cancer^40,44,47,50 ; people with diabetes or heart transplant recipients^41–43; and caregivers of people with dementia or stroke survivors^45,46. The qualitative evidence suggested that discussion groups helped older adults to build social networks and friendships and to feel more familiar with people through regular connections^45,46. The quantitative evidence showed mixed results with regard to loneliness and social isolation. The majority of studies showed increases in social support, but only two showed reductions in loneliness, with four studies not measuring loneliness at all. The asynchronous chat room ‘Koffee Klatch’ in Hill et al.’s⁴⁴ primary study provided a forum for women with chronic illnesses to share their feelings, concerns, life experiences and provide support to each other over 22 weeks, resulting in significant improvement in social support, but not in loneliness, compared to the control group. The Sharing Circle in Weinert et al.⁴⁷ provided the same opportunities, with the addition of discussion of self-study units and internet-based health information. This study saw statistically significant improvements in loneliness, but not in social support, compared to the control group. See Table 1 for details of reviews and studies on online discussion groups/forums.

Supported use of social networking sites for mitigating social isolation and loneliness. Two reviews included the same study on social networking sites⁴⁸. The authors of this qualitative study found that the utilisation of a bespoke social networking site had the potential to reduce loneliness in older people, as there were positive impacts on temporal loneliness (especially in the evening) and on connectedness. Review authors suggested that older adults were more interested in a smaller number of strong relationships mediated through the internet, than they were in a larger network with weak ties. They report that perceived value could have been an issue for older adults, which may have been more obvious through supported social networking service interventions such as that reported by Ballentyne et al.⁴⁸.

Multi-tool interventions (PC, training, internet use, messaging, chat groups) to reduce loneliness and/or social isolation, or increase social connectedness. Three reviews included a total of four quantitative studies on multi-tool interventions. van der Heide et al.⁴⁹ report on the Care TV package for people receiving home care in The Netherlands. This video and voice network allowed clients to communicate round-the-clock with a nurse practitioner. They received a ‘Good Morning/Goodnight’ call and could use the video facility to call family members. Average feelings of loneliness decreased substantially, with social and emotional loneliness showing pronounced decreases. The three other studies reported on web-based discussion groups in the Women-to-Women programme, with mixed results regarding levels of loneliness and social support. Weinert et al.⁴⁷ reported on an RCT of a web-based discussion groups, with a peer-led online support group and self-study units supported by an Advance Practice Nurse. Improvements were found in loneliness, but there was no significant difference in social support between the intervention and comparison groups, following the 11-week intervention. Weinert et al.⁵⁰ found significant increases in both loneliness and social support, compared to the control group, over the 22-week intervention. Hill et al.⁴⁴, found statistically significant effect on both social support and loneliness after 22 weeks. See Table 1 for contextual details of reviews and studies on multi-tool interventions.

Intervention Component Analysis and Qualitative Comparative Analysis

QCA and ICA were undertaken to help us further identify the processes and mechanisms that were common across the interventions described in Table 1 and the narrative synthesis. To undertake QCA, we first conducted ICA to understand the nature of the interventions. We inductively coded the nature of intervention features (i.e. components) and used trialists’ informally reported experiences of implementing the interventions (e.g. author reflections reported in introduction and/or discussion sections) to understand the importance and underlying mechanisms of particular features²⁰.

Theory selection and setting up the QCA. A fundamental element of QCA is the selection of an appropriate theory to base the analysis on, and to help identify suitable evidence to extract as part of the ICA. To understand which processes might be important to incorporate in interventions – regardless of specific mode of delivery (i.e. videoconferencing or internet chat group) – we drew on Robert Weiss’s⁵³ ‘Fund of Sociability’ theory^iv. The theory is intended to capture assumptions, content, and functions of social ties that can help to support developing social relationships. The theory specifies six characteristics of social interactions and relationships that are necessary for well-being and the avoidance of loneliness⁵³. Table 3 outlines the six categories, their definitions and how we interpreted them in relation to the interventions in the QCA.

Our QCA built on the earlier descriptive and narrative synthesis, and addressed the question: ‘Do the characteristics of social interactions and relationships stated in the fund of sociability theory explain differences between remotely delivered interventions found to be effective compared to those found to be ineffective?’ To gain familiarity with the studies and attempt to gain ‘deep case knowledge’, we started by reading and re-reading the studies.

Selection of studies (cases)^v for the QCA. We focussed on studies that met our criteria for the QCA including that they (a) presented quantitative results, (b) were remotely delivered, (c) focussed on older people, and (d) actively sought to strengthen social relationships or prevent/offset loneliness. From the 18 primary studies described above, 12 met these criteria.

Developing a data table. QCA is based on set-theory with sets differentiated as belonging to a successful and unsuccessful set on the basis of their outcome. The outcome can be based on an objective measure or subjective or quality measure⁵⁴, and on a single measure or a composite indicator⁵⁵. The allocation of studies into a successful set and unsuccessful set can follow different strategies. Firstly, success may be defined through the observation of clinically or statistically significant change in the outcome (for example⁵⁶). A second approach is for the researcher to set thresholds for determining success. A third strategy is ensure (approximately) equal representation of un/successful cases by ranking cases according to their effectiveness and allocating studies into un/successful outcome sets. A fourth strategy is to use a more qualitative approach where additional characteristics besides the outcome value are considered to ensure a representation of studies in the un/successful outcome set⁵⁷. In a systematic review, stratifying studies by their characteristics to ensure a diversity in study size or study design among successful studies may ensure a more informative solution is produced.

To generate our outcome sets, and group interventions as being ‘successful’ or ‘unsuccessful’, we calculated an estimated effect size for each study. Effect sizes are used differently within QCA as opposed to meta-analysis; i.e. as a guide to allocating studies to successful (most effective) or unsuccessful (least effective) outcome sets, rather than to provide a pooled estimate of effect with precision. Most effect sizes were based on measures of social support, which we regarded as a proxy measure for social isolation. The exceptions were Schwindenhammer³⁶ and O'Connor, et al.⁴⁵ where a measure of loneliness was the only suitable outcome available. However, we did attempt to express the effect sizes in a common rubric where possible, e.g. prioritising post-test measures for studies that involved randomisation of participants or clusters (five studies), and change measures where these data were not available. For those studies with a comparator group (eight studies), effect sizes were calculated in the standard way see Thomas et al., 2017⁵⁸; for those studies that employed a pre- and post- evaluation design an effect size was estimated based on changes in the pre- and post- individual scores divided by the standard deviation at pre-test^59,60; in some cases this involved using mean differences as proxy information and other approximations.

Using the effect size for indicative purposes, we grouped interventions into those that were ‘successful’ (studies with effect sizes over 0.5), ‘partially successful’ (studies with effect sizes between 0.2 and 0.5) and ‘not successful’ (studies with effect sizes under 0.2 or suggested negative impacts) based on thresholds suggested by Cohen⁶¹ for interpreting effect sizes. However, combining the different study designs, and particularly those with and without a comparison group, using the same approach could lead us to overstate the effectiveness of studies without a comparison group. To mitigate this possibility we also present the results of a sensitivity analysis, where we imposed an additional ‘penalty’ on studies without a control group – studies with effect sizes of 0.5 and over were deemed to provide partial evidence of success (0.66); studies with effect sizes between 0.2 and 0.5 were deemed to provide weak evidence of being ‘not successful’ studies (0.33); while studies with values lower than 0.2 were deemed to provide strong evidence of being ‘not successful’ (0). This is akin to adding in additional ‘qualitative’ information – in this case on study design – to distinguish studies as belonging to a successful and unsuccessful outcome set⁵⁷. We also examined the potential impact of omitting these four studies, although this is not a preferred option given that QCA models typically need 10 or more cases as a minimum.

To create our data table, a coding scheme was developed to determine whether the conditions reflecting the fund of sociability processes were actually present in the studies (see 10). The results of this coding and the data table are presented below in Table 4.

Truth table. As we had a limited set of cases for the number of conditions, our analytical strategy involved first creating a ‘truth table’ based on six conditions, and then producing a reduced truth table containing four conditions and minimised solution²⁴. A ‘truth table’ sorts cases according to the configuration of conditions they exhibit. Although we noted that both ‘availability’ and ‘control’ were conditions generally only observed in successful intervention studies, they did not appear to be as critical to outcome success as the other conditions, appearing in fewer studies. Our reduced truth table thus contained four conditions (intimacy, interaction, support and nurturance) with five of a possible 16 configurations represented (Table 5). Two configurations are observed as triggering a successful outcome; in one, supported by five studies, all four conditions are present; in the second, supported by two studies, three of four conditions are present. On the right side of the table is a column marked consistency; this indicates the strength to which studies that belong to the condition set are also a subset of the outcome set. A value of 1 indicates perfect consistency; all cases in the configuration are strong members of the condition set and the successful outcome set; and there is strong evidence that these intervention characteristics trigger successful outcomes. A value of 0 indicates perfect inconsistency and there is no evidence that these intervention characteristics trigger successful outcomes. Values in between indicate some degree of ambiguity, which was expected given that we used a “fuzzy-set coding scheme” which allowed studies to be partial members of sets (using a value of 0.85 to denote membership).

Boolean minimisation and formation of a solution. We applied Boolean minimisation to obtain the simplest expression of those conditions (intervention processes) that were associated with triggering a successful intervention. We developed a complex solution based on the observed data only, and found that those interventions that ensured the following processes took place were those in the successful outcome set:

INTIMACY and INTERACTION and SUPPORT

Within QCA, information from unobserved configurations (logical remainders) can be used to simplify the solution and check the quality of the solution. We incorporated these logical remainders to develop two further solutions^vi, although incorporating logical remainders in this model did not help to simplify the solution above. Our model and details of its fit are presented below (Table 6). The high consistency value for the solution suggests that when this configuration of conditions is observed in an intervention, it is generally sufficient to trigger a successful intervention (i.e. a substantial change in social support). The coverage statistic suggests that the model broadly accounts for the successful interventions observed.

Sensitivity and additional technical quality checks. Using an alternative measure of effect size that incorporates a ‘penalty’ reflecting the greater uncertainty around pre-post studies, we re-ran the analysis described above. The truth table (Table 7) with this alternative outcome showed one configuration of successful studies. This suggested that studies which incorporated all three processes observed earlier, as well as nurturance, were those that triggered a successful outcome (using a slightly lower consistency value of 0.825, which is still well within recommended thresholds²⁴). Coverage was slightly lower for this solution, although the solution still provided a comprehensive explanation of why some interventions were successful (Table 8). The replication of the same three core conditions provides a degree of triangulation that our main solution identified in Table 6 provides a robust account; the inclusion of nuruturance as an additional condition below is not contradictory, but suggestive that as a condition it may help to distinguish a smaller pool of studies as successful.

We also undertook quality checks to understand whether our solution, or the assumptions we made in its derivation, could also predict unsuccessful outcomes, and found little evidence that this was a possibility. We also explored whether focussing only on the 10 studies that measured social support would change our interpretation, and again found little evidence that this would influence the model. Similarly, focussing only on studies that had a comparison group showed a similar pattern descriptively^vii.

Interpretation of the solution. The successful outcome set contained those interventions that: (i) supported participants to form ‘intimate’ relationships and express their feelings freely without self-consciousness between peers; (ii) ensured that there were shared characteristics between participants and their peers (beyond a single experience, and beyond geography alone); and (iii) included some form of pastoral care or support (e.g. light-touch oversight of a discussion forum by professionals or opportunities for participants to contact professionals for advice). This configuration explained the majority of the successful outcomes we observed.

Taken together, albeit with some caveats, these characteristics can form a set of design principles for future interventions that are delivered remotely which aim to increase support available to older adults and offset the risks of social isolation and loneliness. The interventions that were not in the successful outcome set did not provide evidence that all three processes had been part of the interventions, and some indicated that processes to the contrary had taken place.

Summary of findings

In this rapid review of reviews, narrative synthesis showed that supported video-communication interventions are regarded positively by older adults and can have positive effects on loneliness and social support. However, the quantitative evidence remains uncertain and, although they were placed in the effective set of studies in QCA analysis, uncertainty about effectiveness is a shared conclusion in other similar reviews⁶². Evidence about online discussion groups and forums also demonstrated mixed results, with increases in social support, but less evidence for improvements in loneliness. Telephone befriending has not been widely researched, but qualitative evidence suggested this intervention model may be helpful in addressing loneliness and social isolation, although quantitative evidence did not show this. The evidence for social networking sites was weak. Multi-tool interventions showed decreases in loneliness, but not always increases in social support. Clearly, these interventions vary greatly, so it is difficult to isolate the effective elements. Similarly, conceptualisations of loneliness and social isolation vary, making comparisons and conclusions challenging.

Using QCA, we looked beyond specific models of intervention to explore which intervention processes are aligned with being in an effective intervention set. We have shown that the following processes are enabled in effective interventions including (i) supporting development of intimate relationships; (ii) supporting interactions through ensuring participants share experiences/characteristics; (iii) supporting participants through pastoral guidance.

Discussion

Gaps in the evidence. Despite our extensive searches, we found only one study of telephone befriending³⁹ included within a single systematic review. This was also the one of the few studies that made use of volunteers. There was no information provided about the training and support provided to the volunteers, as the focus was on the experience of older adults receiving the service. Similarly, we found little information about training and support provided to staff members supporting other forms of intervention. Information, communication, support, moderation and mediation was provided to older adults by research staff and health professionals (nurses, psychologists, advance nurse practitioners) within the primary studies, but there was little detail about how staff (paid and volunteer) were trained or supported to provide these. The evidence identified in the QCA finds that successful interventions are effective because they are able to ehance complex psychosocial processes and abilities, highlighting that staff may need specialist training in delivering interventions successfully. In addition, support and training is likely to be important for managing the wellbeing of those delivering the intervention. Guidelines published elsewhere suggest volunteers or staff members should receive high-quality training and regular supervision to be competent⁶³, yet the call for NHS Volunteer Responders to make telephone calls to isolated older adults did not include any offers of training or support⁴.

Most of the studies included in this review involved some form of new technology, with just two involving an intervention delivered through (traditional) telephone. No study examined an intervention delivered through a smartphone. Similarly, our inclusion criteria could have theoretically allowed other forms of remotely delivered interventions to be included, such as letter writing, although no such study was identified. These forms of interventions could be purposively considered in future reviews, with a recent intervention involving cross-generational letter writing suggestive of positive impacts for older and younger people alike⁶⁴. There may be scope in the future for inter or cross-generational interventions that can help to provide both befriending, and technological support, while maintaining the principles outlined earlier.

We found few studies reporting on low intensity psychosocial interventions, which could be due to our focus on loneliness and/or social isolation as outcomes of interest. In the broader literature, whilst some studies have demonstrated positive impacts on depression, wellbeing and general mental health of delivering therapies through remote means⁶⁵, several of these interventions may not specifically address loneliness and are not targeted at older adults⁶⁶.

Empowering and supporting older adults involved in remote interventions. Overall, the results here suggest that older adults can be empowered to support each other through online discussions and forums. In the narrative synthesis we found reviews containing several studies with peer support, provided through synchronous and asynchronous messaging, chat rooms and discussion forums. This challenges the assumption that older adults must always be on the receiving end of an intervention to address social isolation and loneliness. When we moved to study-level synthesis, we also found that studies that enabled older people to feel that their contributions could improve the outcomes of others (i.e. improved levels of self-worth^{40–42,46,50}) tended to be successful interventions^viii. As the mobilisation of thousands of volunteers takes place to support older people who are currently shielding in the COVID-19 pandemic, recognising that older people can be both providers and recipients of support simultaneously is likely to be an important principle to adhere to in the design of activities.

Strengths and limitations. The strengths of this rapid systematic review of reviews include the transparent and robust approach to searches, data extraction, review quality appraisal and analysis, ensured through pre-publication of a protocol on the EPPI-Centre website. Despite the rapid nature of this review process, we have conducted the review according to systematic review methodology⁶⁷. In this case, the rapid element of the review was primarily reflected in the decision to exclude reviews focussed on caregivers from the synthesis; other stages were conducted according to standard systematic reviewing practice. A further strength was the diversity of synthesis approaches conducted, including QCA and ICA.

Searching for systematic reviews means that we may have missed some more recent primary studies in this area, but it ensured that our review was achievable within the four-week timeframe required for a timely response during the COVID-19 crisis. We applied the AMSTAR2 quality appraisal tool to the included reviews, although the reviews included in the synthesis were found to have a low quality rating. In addition, we did not conduct any quality assessment of the primary studies that we looked at in more detail. Some of these had been assessed by the review authors, but many had not. There were very few identifiably robust primary studies that met our inclusion criteria. Only one primary study was identified by review authors as ‘strong’, with others rated as ‘weak’ or with no quality appraisal at all. The poor, or lack of, quality rating for many of the included studies means that findings should be considered with caution. In addition, few of the studies considered potential adverse impacts of the interventions. However, this is the case for many reviews in this research area and is not unique to our rapid review.

Owing to the rapid nature of this review, we focussed on reviews addressing interventions to mitigate loneliness or social isolation on the general older adult population. This meant we excluded reviews identified through the searches focussing exclusively on caregivers that may have provided additional insights. Other limitations included our treatment of primary studies in the QCA, where neither the precision of the effect size, study design, or quality were included in the model or the allocation into different outcome sets in our main model. Studies with weaker designs, and effect sizes derived from these, were treated in the same way as those with more robust designs in our main model. Although this is not uncommon in QCA practice, further synthesis could be conducted focussing on only those studies with a more robust design in future.

Further research and conclusions. Loneliness and social isolation are extremely complex phenomena⁶, and require a deep understanding and deliberative treatment that was beyond the remit of this rapid review. The risk of running unsuccessful interventions may be higher than many triallists appear to recognise, and a failure to ensure that the processes identified as important in effective interventions are incorporated into intervention design may have adverse impacts for older people, for example in heightening their feelings of alienation⁶⁸. Our findings do not lead us to recommend one particular mode of delivering befriending, social support, or low-intensity psychosocial interventions over another (e.g. videoconferencing, telephone calls, chat rooms or forums), and all may be of benefit, but our findings do suggest that the principles highlighted from the QCA should be incorporated into the delivery of an intervention.

We were surprised by the identification of only one systematic review including a telephone befriending intervention. Given the UK Government’s interest in encouraging volunteers to make phone calls to physically isolating and shielding older adults, under the ‘stay at home’ guidance, a systematic review of telephone befriending interventions is needed, to identify evidence to inform policy in this area. A review by Sharma et al.⁶⁹ suggested that a large portion of such interventions may be found in grey literature. In the current context of the COVID-19 pandemic, a number of befriending interventions are being delivered by a variety of organisations, and there is scope to incorporate learning from these in future systematic reviews in this area.

As the training and support components of the technology-mediated interventions were unreported in the reviews and studies that we synthesised, there is a need to search for these elsewhere. Evaluations of existing telephone befriending and psychosocial support services, often found in the grey literature, could act as a starting point. Additional valuable information could be obtained through contact with voluntary sector and NHS organisations delivering befriending, peer support and low-intensity psychosocial interventions. A review of these training and support components could add valuable insight for policy-makers and service providers to ensure that volunteers are well trained, empowered and supported in delivering interventions adhering to the principles outlined earlier. Although we believe all of the intervention modes in scope here have the capacity to include the processes found to lead to more successful interventions (supporting the development of intimate relationships; supporting interactions through ensuring participants share experiences/characteristics; provide pastoral guidance), a more encompassing piece of research is needed in order to identify which mode is most effective, or has the greatest potential, for changing outcomes.

Underlying data

All underlying data as published on the Open Science Framework

Open Science Framework (OSF): Rapid systematic review of systematic reviews: what befriending, social support and low intensity psychosocial interventions, delivered remotely, are effective in reducing social isolation and loneliness among older adults? How do they work? https://doi.org/10.17605/OSF.IO/VS2UX¹³

This project contains the following underlying data:

Extended data

Open Science Framework (OSF): Rapid systematic review of systematic reviews: what befriending, social support and low intensity psychosocial interventions, delivered remotely, are effective in reducing social isolation and loneliness among older adults? How do they work? https://doi.org/10.17605/OSF.IO/VS2UX¹³

This project contains the following extended data:

Reporting guidelines

Open Science Framework (OSF): PRIMSA checklist for “Rapid systematic review of systematic reviews: what befriending, social support and low intensity psychosocial interventions, delivered remotely, are effective in reducing social isolation and loneliness among older adults? How do they work?”. https://doi.org/10.17605/OSF.IO/VS2UX¹³

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).