Loneliness is a multifaceted public health problem [1]. The burden of loneliness is high in some countries [2-9], and it is increasing in many other countries [10]. Loneliness is expected to rise because of lockdowns, quarantine, self-isolation, and social distancing measures that are being enforced in several countries to tackle the COVID-19 pandemic [11,12]. Therefore, tackling loneliness is imperative, and digital technology could play a major role in addressing loneliness [13].

Loneliness refers to an individual’s subjective feelings of a perceived discrepancy between actual and desired social relationships [14,15]. Although loneliness affects people of all ages [15,16], older, younger, and vulnerable people are affected more by it [7,17,18]. Risk factors of loneliness include demographic characteristics, social factors, and physical environments [17-19]. Loneliness enhances the risk of poor physical and mental health [14,20-23], dementia [24], premature mortality, and all-cause mortality [21], particularly in older adults [23]. In addition, the implications of loneliness include the high costs of health and well-being (eg, between £6429.00 [US $8074.80] and £9616.00 [US $12,077.70] per person per year in the United Kingdom) [25] as well as lost work days and productivity (eg, costing up to £2.5 billion [US $3.14 billion] per annum for employers in the United Kingdom) [26]. Therefore, it is imperative to tackle loneliness.

Loneliness is being addressed through a range of social [27] and technological interventions [28]. The latter type of interventions includes numerous and diverse types of digital apps, web-based social networking tools, sensors, and robots [29]. Although these tools use digital technology, they are heterogeneous in many aspects, including the means they provide to socially connect; the purposes for which they are used; the ways and methods of their application; the frequency of their use; and their users, who differ from each other in many traits such as demographic, social, and economic characteristics, and some may have physical and mental limitations. Therefore, these digital technology tools need to be systematically evaluated for their effectiveness in tackling loneliness.

Several published reviews have reported that digital technology interventions (DTIs) are effective in reducing loneliness [30-34]. However, some of these studies are weak and have a high risk of bias [35], and other studies have used a few selected technological interventions and covered literature published over a short span, such as the 3-year period from January 2010 to January 2013 [31].

There is limited evidence on the effectiveness of DTIs for loneliness [36], and there are calls for further research [32,33] to assess and identify the latest DTIs that are effective in reducing loneliness [34,36]. In addition, evaluation of the latest evidence on the effectiveness of DTIs in reducing loneliness is imperative from the perspectives of patients and their families and other stakeholders such as health and social care providers and health insurers [37].

The primary objective of this study is to assess the effectiveness of DTIs in reducing loneliness in adults. The secondary objective is to identify DTIs that are used to reduce loneliness in adults.

The main research question was “Are DTIs effective for reducing loneliness in adults?” The secondary question was “What DTIs are used for reducing loneliness in adults?”

The main outcome measure was loneliness. We extracted data on loneliness measured at both the baseline (before the intervention) and follow-ups (at least 3 months after the intervention) for the intervention groups and control groups, if any.

We undertook a systematic review and meta-analysis as suggested in the Cochrane Methods for Systematic Reviews of Interventions [38]. We have reported the findings in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [39].

We registered this systematic review and meta-analysis with the PROSPERO database on June 10, 2019 (registration ID: CRD42019131524) [40], and we published our protocol [37] before undertaking this study.

A patient and public manager affiliated with our research center reviewed the study protocol and provided suggestions that were incorporated into the protocol. We had no access to any patient diagnosed with loneliness; therefore, we could not include any patients or members of the public in the design and conduct of the study. However, the findings of this study will be disseminated as an open access publication that will be freely available to patients and everyone else globally.

We define the term DTI as an intervention that applies digital technology, that is, the technology, equipment, and apps that process information in the form of numeric codes, usually a binary code [41].

We selected studies that met our predefined eligibility criteria [37]. Study designs included interventional studies (randomized and nonrandomized) that investigated the effects of DTIs on loneliness. We included a range of DTIs, that is, computers, computer tablets, iPads, internet, web-based videos, communication, chatting, social groups, meetings, conferences and messages, sensors, social robots, smart mobile phones, social media tools, and the World Wide Web. We set 3 months as the minimum intervention duration and follow-up period. The research participants were adults, both male and female, aged 18 years or more. We included different settings, that is, residential dwellings, including private residences and care or nursing homes or centers in any country. The studies were limited to journal articles in English published from January 1, 2010, to July 31, 2019.

We electronically searched PubMed, MEDLINE, CINAHL, Embase, and Web of Science and covered the publication period from January 1, 2010, to July 31, 2019. We used an a priori list of keywords prepared in our preliminary literature searches [37]. The keywords were of 2 categories: medical condition or problem (ie, loneliness, lonely, isolation, aloneness, disconnect*, solitude, singleness*, lonesomeness, solitariness, and remoteness) and intervention or technology (ie, digital, technolog*, sensor*, robot*, internet, social media, *phone*, online, iPad*, tablet*, computer*, electronic, web, video, and videoconference), as reported in our published protocol [37].

First, we searched the keywords in the subject headings such as MeSH major terms in PubMed or equivalent terms in other databases (for detailed search history, see Multimedia Appendix 1). Thereafter, we searched for keywords in the title and abstract fields in the selected databases using 3 Boolean operators: “OR,” “AND,” and “NOT.” In addition, we hand searched the reference lists of the shortlisted articles. We wrote emails to the authors of 2 studies requesting for full copies of their research articles [42,43], which were gratefully emailed to us. We contacted the authors of 2 further studies for missing or additional data [44,45]. We had a good response from the authors of both studies, and data were thankfully provided for 1 study only [45]. We sought support from an expert librarian at our library for running literature searches.

Literature searches retrieved 4939 articles, of which 965 duplicate articles were removed (Figure 1). Two researchers (SGSS and DN) independently screened the remaining articles (n=3974) by title, which was followed by reading the abstracts of 442 articles (Figure 1). This screening process led to the exclusion of 3876 articles and identification of 98 articles for full-text review. Three researchers (SGSS, DN, and VK) independently read the full texts of these 98 articles.

When recommendations differed between reviewers at the title, abstract, and full-text review stages, another reviewer (HCvW) reviewed these articles, and his recommendations to either include or exclude an article were final.

Finally, 92 articles were excluded, and the remaining 6 articles were included in the data extraction (Figure 1). All these 6 studies were included in the narrative synthesis, whereas 5 studies—all clinical trials involving an intervention group and a control group—were included in the meta-analysis (Figure 1). One study with a pre- and postintervention design involving only the intervention group was excluded from the meta-analysis.

For data collection, we used an a priori data extraction template (Tables 1 and 2), which comprised several columns: authors, year, and country of study; study aim or objectives; research design; settings; participants’ characteristics (age, gender, and ethnicity); health or medical condition; sampling method and sample size; participant attrition (numbers and percentages); research methods and data collection tools; interventions (eg, type and tool of digital technology); comparators (eg, alternative intervention, placebo, or care as usual); intervention duration (weeks or months); measurement stages (eg, baseline and follow-up: weeks or months after the baseline); outcomes, result, and findings (eg, loneliness scores, including statistics; eg, mean values, SDs, SEs, and CIs); and study authors’ conclusions [37].

SGSS and DN independently extracted data from all included studies (n=6) using the data extraction template (Tables 1 and 2) and resolved discrepancies in the extracted data with discussion and agreement. Data extraction forms were compared and contrasted, thereby avoiding bias and reducing errors in the data extraction process [55]. We extracted aggregated data at the study level as much as possible with respect to the intervention, which is imperative for the reproducibility of effective interventions [56,57]. Following suggestions for reporting data once from studies with duplicate and multiple publications [55], we extracted and reported data only once [50] from a research study with multiple publications [50,58].

We report both a narrative synthesis (narrative summary) and a statistical (quantitative) synthesis (meta-analysis) of our review, as suggested for reporting of a systematic review on effectiveness [59]. In the narrative synthesis, we have included all 6 studies and reported their characteristics, including the study design, settings, sample sizes, data collection methods, participants, interventions, comparators, outcome measurements, and study conclusions.

In the meta-analysis, we have included 5 studies and pooled extracted data on loneliness measured by continuous loneliness scales, that is, the University of California, Los Angeles (UCLA) loneliness scale [47,51] in 4 studies and the De Jong-Gierveld loneliness scale [49] in 2 studies (Table 1). Loneliness scores at baseline and follow-up were reported as the mean values and SDs in 5 studies, whereas 1 study reported mean scores with SEs. For the latter study, we calculated SDs from SEs using a formula suggested in the Cochrane guidelines [60].

We assessed the quality of research by applying the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach [70].

We assessed the risk of bias by focusing on 5 domains: the evaluation of sequence generation, allocation concealment, blinding (outcome assessors), incomplete data, selective outcome reporting, and assessing other biases using the Cochrane guidelines [60]. In a meta-analysis, publication bias can be assessed with a graphical method using funnel plots [60,71] and statistical methods such as the Egger test [60]; however, both methods require at least 10 studies in the meta-analysis [60]. When the number of studies is small, the Egger test has low power and fails to differentiate chance from real asymmetry [60]. Similarly, assessing publication bias using funnel plots with fewer studies would be of very limited usefulness because it would be difficult to spot the publication bias. As we had a maximum of 3 studies in a meta-analysis, we could not check the publication bias with either method.

We checked heterogeneity, that is, variation in study outcomes or intervention effect sizes between studies, by the Cochran Q test with a significance level of ρ<0.10 [72,73] because of the low power of the test in a meta-analysis with very few studies or studies with small sample sizes [74]. We calculated I² statistics to determine the magnitude of heterogeneity (ie, the proportion of variance in the true effect sizes) between studies [28]. We considered I² values of 25%, 50%, and 75% as low, moderate, and high heterogeneity between studies, respectively [75].

We report the findings of meta-analyses using SMDs with 95% CIs as a statistical summary, with the forest plots [60].

Findings about the characteristics of the studies, including the study designs, settings, participants, interventions, comparators, sample sizes, participant attrition, and data collection methods or tools used, are presented in Table 1. The interventions, comparators, follow-up durations, outcomes or measurement scores, results, and conclusions of the included studies are given in Table 2.

We conducted 3 meta-analyses, 1 each for follow-up measurements at 3, 4, and 6 months, involving 3, 2, and 2 studies, respectively.

The risk of bias assessment, that is, the risk of bias graph and risk of bias summary are presented in Figure 5 and Figure 6, respectively. A high risk of bias was noted in the attrition bias and other biases; an unclear risk of bias was detected in the blinding of outcome assessment, allocation concealment, and blinding of participants and personnel; and a low risk of bias was observed, especially, in the random sequence generation and selective reporting (Figure 6). In addition, most studies reported only within-group changes and not between-group comparisons of change, which may suggest a weak quality of the reporting of results and the analysis in these studies (Table 2).

The quality of evidence was moderate, very low, and moderate in meta-analyses involving 3 [46,50,54], 2 [53,54], and 2 studies [45,50], with follow-up at 3, 4, and 6 months, respectively (Figure 7).

To determine whether DTIs are effective in reducing loneliness in adults, we appraised peer-reviewed empirical research involving the application of DTIs in adults with loneliness. Our systematic review provides a narrative summary (qualitative synthesis) as well as a meta-analysis (statistical synthesis) of the findings. The narrative summary of 6 studies included in our review showed a reduction in loneliness in the intervention groups at follow-up compared with baseline (Table 2). However, our meta-analysis of 5 clinical trials with follow-up measurements at 3, 4, and 6 months showed no statistically significant pooled effect estimates as SMDs, the preferred method for summarizing effects on continuous outcomes such as loneliness. Although not statistically significant, the summary effect size at the 4-month follow-up (Figure 3) was better than the effect size at the 3-month follow-up (Figure 2) and the 6-month follow-up (Figure 4).

Our meta-analysis also revealed that CIs of the summary effects of 2 studies, that is, the studies by Larsson et al [50] and Tsai et al [46], were very wide, and the SMDs from these studies were more in favor of the control group than the intervention group (Figure 2). Thus, the wide range of CIs of the summary effects in these studies leave room for uncertainty about the beneficial effect of DTIs on measures of loneliness.

Overall, the findings of our meta-analysis showed no evidence supporting the effectiveness of DTIs in reducing loneliness in older adults.

The quality of evidence of the included studies was very low to moderate (Figure 7), and there was a high heterogeneity between studies [53,54] (Figure 2). All the included studies had a high proportion of female participants. Most notably, the total number of participants was low, especially in 2 studies [50,54], and the sample sizes were reduced further due to a high attrition rate in some studies [45,53]. The types and methods of DTIs varied between studies (Table 2), which were conducted in diverse settings (Table 1). Studies were conducted in 6 different countries: the United States, the United Kingdom, Sweden, the Netherlands (these four countries have individualistic cultures), Taiwan, and South Africa (these two countries have collectivist cultures; Table 1). In addition, despite our inclusion criteria of age 18 years and above, the selected studies more commonly involved older people with an average age of 70 years and above.

Loneliness is influenced by culture [77-79], gender [78], and age [5,78], and these factors could have contributed to the pooled estimates being not statistically significant in our meta-analysis. In addition, differences in participants, especially in terms of age, gender, and culture as well as varied types of DTIs, could have contributed to the heterogeneity observed, especially in the meta-analysis with the 4 month follow-up involving 2 studies [53,54], which differed from each other on different parameters, especially the study designs, settings, participants, interventions, and loneliness measurement scales used (Tables 1 and 2).

There are limited published meta-analyses on technological interventions for tackling loneliness, and a few existing studies have covered literature published up to 2009 [28] and 2011 [30]. Our review and meta-analysis included the latest evidence published between January 1, 2010, and July 31, 2019. We did not replicate the findings of earlier meta-analyses that reported evidence suggesting that technological interventions resulted in decreased loneliness [28,30]. For example, a meta-analysis by Choi et al [30] reported statistically significant evidence suggesting that the internet and computers reduce loneliness. However, they [30] focused on older adults with depression and included the internet and computers only as technological interventions, whereas we included different types of DTIs, and our population of interest was adults of all age groups (≥18 years). In addition, the meta-analysis by Choi et al [30] included studies (n=5) with different follow-up periods (3-6 months), but they did not report which follow-up measurements were included in their meta-analysis. In our meta-analysis, we conducted separate meta-analyses for measurements at different follow-up periods, that is, 3, 4, and 6 months, as suggested by the Cochrane guidelines [60].

A meta-analysis by Masi et al [28] also reported that technological interventions reduce loneliness, which was more in pre-post studies and nonrandomized studies than in RCTs. However, they included studies with technology and nontechnology-based interventions [28], whereas we focused on studies with DTIs only. In addition, Masi et al [28] did not report how they analyzed measurements at different follow-up periods, whereas we did not combine measurements at different follow-up times, as suggested by the Cochrane guidelines [60]. Nonetheless, Masi et al [28] concluded that technology is yet to be capitalized for loneliness.

Interestingly, our findings provide new insights about DTIs and loneliness. Our meta-analysis showed no statistically significant reduction in loneliness in the intervention groups compared with the control groups at the 3 -, 4-, and 6-month follow-ups. Thus, our findings show no evidence supporting the effectiveness of DTIs in reducing loneliness in older adults, which goes beyond the findings of a recent Cochrane review that reported no evidence of video calls being effective in reducing loneliness in older adults [73].

In addition, our findings refute and contradict a commonly held view that digital technology can solve the problem of loneliness, especially in older people. Nonetheless, digital technologies provide tools and means that facilitate social connection [80], which may help in reducing loneliness for a limited period because the effects of DTIs are short-lived [81]. This may be because digital technologies do not provide real human interaction [80] and cannot replace human contact [45]; thus, they do not reduce social disconnectedness in real life [82] on a long-term basis.

Nonetheless, a review has reported that some nontechnological interventions are effective in reducing loneliness in older people [83], but these interventions require a meta-analytic evaluation. In addition, a recent meta-analysis [84] reported moderate evidence of the effectiveness of a range of social, emotional, and psychological interventions, delivered through technological and nontechnological means, in reducing loneliness in young people aged 3-25 years; however, the analyzed studies had limitations. Therefore, further research is required.

Our study had some limitations: the inclusion of only 6 studies with heterogeneous sets of results and the minimum intervention duration of 3 months, which could have resulted in the inclusion of a small number of studies and possible exclusion of potential studies that would have provided useful evidence.

In addition, we could not conduct subgroup and meta-regression analyses due to the very limited number of studies (n=5) in the meta-analysis and lack of data on loneliness by participants’ demographic characteristics. In addition, our study might be narrow because we excluded some studies [44,85-91], which met the technology criterion such as the use of robots, sensors, digital speakers, and apps but did not meet other selection criteria. Thus, our study may be limited to studies about social interactions and connectedness using digital technology tools.

Moreover, another limitation of our review could be the use of a meta-analysis based only on follow-up data. For example, a study by Tsai et al [46] in the 3-month follow-up meta-analysis had an SMD of 0.06 with a 95% CI of −0.8 to 0.65 (P=.03; Figure 2), which may suggest that these studies may have had higher power to show a difference compared with baseline loneliness.

As recommendations for future research, we suggest that researchers involved in trials agree on a common measure of loneliness and consider reporting of results in a standardized way, which will allow pooling of baseline-adjusted estimates of the treatment effect rather than differences in follow-up means.

Our meta-analysis showed no evidence supporting the effectiveness of DTIs in reducing loneliness in older adults. Therefore, there is a need for further research involving RCTs [50] with larger sample sizes and longer duration of interventions and follow-up measurement periods. Future research may apply inclusive research designs using a combination of digital apps, including robots, sensors, and social connecting apps, by involving adults aged 45-65 years, as this segment of the population is more likely to be more technology savvy and digital interventions might be more effective in this age group. Future research might also target ethnic minority communities and specific groups such as lesbian, gay, bisexual, and transgender people where loneliness is common [8,92].