Ernst, M.eyt al. (2022). Loneliness before and during the COVID-19 pandemic: A systematic review with meta-analysis. American Psychologist, May 2022. http://dx.doi.org/10.1037/amp0001005
Abstract: The COVID-19 pandemic and measures aimed at its mitigation, such as physical distancing, have been discussed as risk factors for loneliness, which increases the risk of premature mortality and mental and physical health conditions. To ascertain whether loneliness has increased since the start of the pandemic, this study aimed to narratively and statistically synthesize relevant high-quality primary studies. This systematic review with meta-analysis was registered at PROSPERO (ID CRD42021246771). Searched databases were PubMed, PsycINFO, Cochrane Library/Central Register of Controlled Trials/EMBASE/CINAHL, Web of Science, the World Health Organization (WHO) COVID-19 database, supplemented by Google Scholar and citation searching (cutoff date of the systematic search December 5, 2021). Summary data from prospective research including loneliness assessments before and during the pandemic were extracted. Of 6,850 retrieved records, 34 studies (23 longitudinal, 9 pseudolongitudinal, 2 reporting both designs) on 215,026 participants were included. Risk of bias (RoB) was estimated using the risk of bias in non-randomised studies—of interventions (ROBINS-I) tool. Standardized mean differences (SMD, Hedges’ g) for continuous loneliness values and logOR for loneliness prevalence rates were calculated as pooled effect size estimators in random-effects meta-analyses. Pooling studies with longitudinal designs only (overall N = 45,734), loneliness scores (19 studies, SMD = 0.27 [95% confidence interval = 0.14–0.40], Z = 4.02, p < .001, I 2 = 98%) and prevalence rates (8 studies, logOR = 0.33 [0.04–0.62], Z = 2.25, p = .02, I 2 = 96%) increased relative to prepandemic times with small effect sizes. Results were robust with respect to studies’ overall RoB, pseudolongitudinal designs, timing of prepandemic assessments, and clinical populations. The heterogeneity of effects indicates a need to further investigate risk and protective factors as the pandemic progresses to inform targeted interventions.
Public Significance Statement: This synthesis of international research with a focus on longitudinal study designs shows small, but robust increases in loneliness during the COVID-19 pandemic across gender and age groups. As loneliness jeopardizes mental and physical health, these findings indicate that public health responses to the continuing pandemic should include monitoring of feelings of social connectedness and further research into risk and protective factors.
Keywords: COVID-19, loneliness, mental health, pandemic, social isolation
Discussion
The main aim of this study was to summarize the most recent high-quality evidence for changes in loneliness in association with the COVID-19 pandemic in a systematic and rigorous way. The statistical synthesis focused on longitudinal study designs. The robustness of the results was tested and predictors of change in loneliness were also explored. Based on the pooled effect sizes of 19 studies, an overall increase in loneliness since the start of the pandemic (SMD = 0.27 [0.14–0.40] for continuous measures) was found. This constitutes a small (Cohen, 1992; Ferguson, 2009) effect, which was also heterogeneous. An exploratory metaregression was modeled to statistically explain the observed variation.
The confidence in the finding that there has been an increase in loneliness—albeit small—during the pandemic is strengthened by the results of the sensitivity analyses, the inclusion of only high-quality and longitudinal research in the meta-analyses, the relatively large number of studies with a pooled sample of 45,734 participants, and the lack of any indication of publication bias.
A previous rapid review and meta-analysis (Prati & Mancini, 2021) reported small increases in mental distress (overall g = 0.17) based on longitudinal studies. It also included three studies concerning loneliness conducted in spring 2020 (Luchetti et al., 2020; Niedzwiedz et al., 2021; Tull et al., 2020), only one of which (Niedzwiedz et al., 2021) could be included in the main analyses of this review (another one (Luchetti et al., 2020) was included in a sensitivity analysis). Their synthesis showed no statistically significant change in loneliness (g = 0.12, p = .34). The present study expands on this rapid review by including more original studies from different countries with assessments later in the pandemic.
Another recent systematic review (Buecker & Horstmann, 2021), which did not synthesize its findings meta-analytically, reported based on 12 studies (three of which were included in this review (Bu et al., 2020; Heidinger & Richter, 2020; van Tilburg et al., 2020)) that most longitudinal investigations found increases in loneliness during the pandemic, which corresponds to the present findings. Studies showing decreasing loneliness had overwhelmingly relied on prepandemic assessments conducted shortly before the implementation of physical distancing, while those with comparison data from months or years before the pandemic had observed increased loneliness during the pandemic.
The present study extends previous knowledge on changes in loneliness during the pandemic; however, the observed increase needs to be interpreted with caution: On the one hand, loneliness can be considered a normal, nonpathological reaction to changing circumstances and many people experience it at some point in their lives. On the other hand, previous research has shown that particularly sustained or chronic loneliness jeopardizes mental and physical health (Cacioppo et al., 2015; National Academies of Science, Engineering, & Medicine, 2020), and the ongoing pandemic and associated restrictions could compromise lonely individuals’ efforts to reconnect with others (Qualter et al., 2015).
Furthermore, the overall pooled effect in this study was small and the effect sizes reported by the individual studies were heterogeneous. The numerical values of effect size indices often provide limited understanding of the real-world significance of those effects, as even statistically small effects can be of high importance (e.g., Meyer et al., 2001). Interestingly, the most rigorous analysis (the sensitivity analyses that included only longitudinal study designs and studies with moderate RoB) showed a larger pooled effect size than the main analyses. This mirrors findings of the metaregression, in which studies’ higher RoB was negatively associated with the observed effect sizes. Taken together, these results suggest that the pooled effect in the present study might underestimate effects in at-risk populations.
The heterogeneity of effects might stem from the diversity of study characteristics included in prior research (e.g., age groups, healthy and clinical populations, regions, study designs, and loneliness measures). However, the fact that the metaregression accounted for less than a third of observed variance suggests that other factors may influence the different trajectories of loneliness in the pandemic context. As some original studies failed to report on previously identified vulnerable groups (e.g., individuals living alone), these could not be tested as predictors. Hence, more high-quality studies that assess risk and protective factors are needed so that their relevance can be assessed across samples. This is an important step to inform targeted prevention efforts.
The metaregression identified age, clinical populations, and studies’ overall RoB as predictors of increases in loneliness, but only overall RoB had statistically significant effects. However, the analysis might have been underpowered as it was not possible to test all predictors of interest simultaneously. While neither of two other available reviews conducted a metaregression to explore characteristics associated with changes in loneliness (Buecker & Horstmann, 2021; Prati & Mancini, 2021), Prati and Mancini (2021) explored, using metaregression, predictors of increases in mental health symptoms during the pandemic. They found no effects of mean age, gender, or study design, either. More research is needed to better understand the mechanisms underlying observed changes in loneliness. They could include response biases such as social desirability or the perceived de-stigmatization of loneliness: learning that loneliness is an experience shared by many during the pandemic might make it easier to acknowledge and disclose one’s social needs.
Another question that should be addressed is whether changes in loneliness are primarily driven by changes in perceived relationship quality or quantity, and if this differs according to individual characteristics or in subpopulations (e.g., age groups). As a consequence, efforts aimed at preventing or reducing loneliness could pursue different strategies. For example, individuals who are lonely because they are socially isolated and have few contacts might benefit from programs fostering exchange, ideally across different living contexts and between generations. Previous research has shown positive effects of interventions enhancing social support (such as buddy-care programs; Masi et al., 2011). Within the pandemic context, these types of interventions could be carried out digitally or within small “social-support-bubbles.” Others might not feel that they have too few contacts overall, but instead be dissatisfied with their close relationships. Research has shown that people in conflictual relationships feel lonelier than those who perceive their relationships as supportive (Hsieh & Hawkley, 2018; Selcuk & Ong, 2013). As the pandemic implicates a myriad of stressors affecting relationships, interventions could target the quality of partner relationships, parent–child relationships, or other configurations in which people live together, for example, through better communication (about feelings and worries, needs for support, etc.). Further approaches at the individual level might also focus on strategies to modify maladaptive social cognitions (which Masi et al. (2011) found to be the most effective). As individuals differ with respect to their ability to adapt to new situations, some might benefit from interventions aimed at changing attitudes and expectations regarding social contacts during a pandemic (e.g., regarding availability, spontaneity, and modality).
In general, prevention and intervention programs should address particularly vulnerable groups such as older individuals without internet access. Concerns have been raised about their lack of representation in large-scale, longitudinal investigations of loneliness (Dahlberg, 2021), so care must be taken to ensure that preventive measures address the needs and reach the breadth of the population instead of focusing on those who are most likely to be research participants. It should also be a research desideratum to include the most hard-to-reach members of the community.
Strengths and Limitations Including Constraints on Generality
The present study synthesized substantially more original reports than previous rapid and systematic reviews. The meta-analyses’ focus on longitudinal study designs is another strength. Besides peer-reviewed publications, this review included studies identified via other sources, for example, preprint servers (but no unpublished studies). In addition to longitudinal studies, pseudolongitudinal studies were included in the narrative synthesis and in the exploratory metaregression. However, the informative value of the metaregression was still hampered by the limited number of predictors that could be tested on the basis of the available studies (which also necessitated a stepwise procedure).
The lack of control samples unaffected by the pandemic weakens possible causal inference, making it more difficult to attribute the increase in loneliness to the pandemic. Furthermore, an alternative explanation for increases in loneliness in the population was recently provided by Buecker et al. (2021) who reported linear increases in emerging adults over the last decades. The discussion of underlying period and/or cohort effects included more flexible social (including romantic) relationships, use of communication technology, and occupational instability. At the same time, some of these trends resulting in individuals having many, but weak social ties may have particularly come into effect in the pandemic context.
RoB assessments revealed that most original reports had a serious RoB in at least one domain, for example, regarding the measurement of loneliness (including the use of untested single items or adaptations of questionnaires originally intended to measure other constructs). Although sensitivity analyses supported the results’ robustness with respect to studies’ overall RoB, the metaregression suggested that it could have led to an underestimation of the magnitude of changes in loneliness.
Further, some variables could only be included in the analyses in ways that reduced the complexity of original study designs/dynamic situations: First, the duration between loneliness assessments was often a range and not a concrete number of days/months. The present analyses used the respective midpoint of this range. For the duration of pandemic-related restrictions, the same procedure was employed. Restriction measures were coded based on official mandates, however, this might have been imprecise if measures differed between regions and/or if the assessment spanned a period in which these rules changed. There was also little information available regarding participants’ adherence to restrictions. Thus, in summary, the study design was not suited to determine effects of (specific) restrictions on loneliness. Furthermore, as the pandemic progressed differently around the world, we used regional cutoffs to distinguish whether study assessments had taken place before or during the pandemic, but individuals might also have been affected by restrictions outside their place of residence (e.g., travel bans). However, a sensitivity analysis confirmed the results’ robustness regarding findings of studies whose “prepandemic” assessment overlapped with the introduced cutoffs.
As included studies mainly derived from the U.S. and Europe, whereas South America, Asia/Oceania, and Africa were underrepresented, the present findings might not be generalizable to populations not conforming to the WEIRD (Western, educated, industrialized, rich, and democratic) stereotype (Henrich et al., 2010). Further, the original investigations might have omitted specific groups, such as immigrants not speaking the country’s official language, people with mental and/or physical disabilities, and those without regular internet access, if conducted online.
Posts
