Loneliness and sleep: A systematic review and meta-analysis

1In, Dr. Vivek Murthy, 19th Surgeon General of the United States, named loneliness as the most common pathology he had encountered in his 3 years of service. The statement made headlines, but the awareness of social isolation as a health risk dates back decades. synthesized research at the time to argue that social relationships affect health and called for work on the social, psychological, and biological processes mediating this relationship. Subsequent research has bolstered their conclusion: multiple meta-analyses indicate that social factors predict morbidity and mortality.

2However, there remains a dearth of research identifying mechanisms underlying the link between social relationships and health. Loneliness, in particular, can be defined as feeling separate from others, and has been identified as one key aspect of social factors that influence health. proposed that sleep disturbance is a mechanism through which loneliness influences health, citing two studies led by, ) in which lonely persons reported lower sleep quality and showed lower sleep efficiency and higher levels of wake time after sleep onset than non-lonely persons. argued that this sleep disturbance marked the loss of a fundamentally restorative behavior, thus affecting metabolic, neural, and hormonal processes. Evidence for this theory has subsequently been reviewed narratively (e.g. ) but not systematically.

3The present article aimed to provide a comprehensive review of the literature on the relationship between loneliness and sleep. A systematic review was conducted to describe the current literature on loneliness and sleep, with an emphasis on sampling and measurement. Meta-analytics were then used to quantitatively examine the cross-sectional literature on loneliness and sleep disturbance—defined as insomnia symptoms and subjective sleep quality—to generate mean effect sizes and assess for the presence of moderators (age and gender) and publication bias. This article represents a critical first step in synthesizing the current literature on the relationship between loneliness and sleep disturbance, thus seeking to address the 30-year-old question of the mechanisms that underly the connection between social relationships and health.

4This review was conducted according to the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines.

5Searches were conducted in PubMed and PsycINFO between 2 and 7 February 2018. Reference sections of eligible studies were then reviewed (see Supplementary Appendix A for full searches). Eligibility criteria were (1) analysis of the relationship between sleep and loneliness (summarized in the results or tables), (2) quantitative methodology, (3) peer-reviewed, (4) adult sample, (5) written in English, (6) original research, (7) not sample with sleep apnea, and (8) not study with manipulation of temporal cues or other extreme conditions. Titles and abstracts, then full-text articles, were screened to determine eligibility. The search, screening, and selection process were conducted by the first author (S.G.; see Figure 1 for PRISMA flow diagram) then replicated by a co-author (S.R.). The replication process generated one additional article that was deemed ineligible because it pertained to cancer-related loneliness rather than loneliness more broadly.

6The following information was extracted from all studies: study design, sample characteristics (mean age, age range, gender, race/ethnicity, nationality), sampling method, measurement of sleep and loneliness, adjustments to model, findings on the relationship between sleep and loneliness, and effect sizes. The following information was extracted for longitudinal studies: length of follow-up, time points, attrition rate, and handling of attrition bias. This process was conducted by the first author (S.G.) then replicated by a co-author (A.B.W.). Bias was assessed qualitatively due to evidence that current tools fail to discriminate between methodological issues and deficient reporting.

7The principal summary measure was the correlation coefficient (r) between loneliness and sleep. Other summary measures were transformed to r or generated where possible (bivariate β = r; conversion using online resources from or ). Confidence intervals (CIs) were estimated when not provided. Meta-analyses were conducted analyses in R x64 3.5.1 (following guidance from; packages from; ). Estimates were generated for sleep disturbance (defined as self-reported sleep quality and insomnia symptoms), self-reported sleep quality, and insomnia symptoms. Sleep quality was coded so that higher values represented greater problems with sleep quality. Sensitivity analyses were conducted, excluding studies with effect size estimates not meeting the r normality assumption and not excluding outliers. Risk of bias was assessed by visual examination of a funnel plot, the rank correlation test, and Egger’s regression test. Study homogeneity was assessed using the Q-statistic. Moderators included age and gender. There were too few studies on other sleep outcomes to make quantitative analysis informative; however, these studies were included in the systematic review and described qualitatively. Similarly, we did not synthesize summary statistics across the longitudinal relationship due to the size (k = 8) and heterogeneity of this literature. Input data file, R code, and output are available on the Open Science Framework (https://osf.io/s6kbt/?view_only=6de74e9fa6ed40cfa027004e2cb66bb9).

8The total estimated sample size at baseline was 45,177 (if multiple studies reported on the same sample, the larger sample size was used in this estimate). Several articles used data from a larger study (e.g. the Chicago Health, Aging, and Social Relations Study). There were three cases of sample overlap across articles. Sample characteristics for each study are summarized in Table 1 (see Supplementary Appendix B for the race/ethnicity of US samples).

9Studies examined different dimensions of sleep, to include sleep duration, sleep quality, insomnia symptoms, sleep satisfaction, and sleep adequacy; two studies measured sleep quality objectively. Loneliness was measured both directly (i.e. using the term “lonely”) and indirectly (avoiding the term “lonely” due to potential stigma). See Supplementary Appendix C for a summary of the measures for each study with a brief note on quality; see Supplementary Appendix D for psychometric properties of scales.

10The random-effects model of correlation between loneliness and sleep disturbance (where sleep disturbance is defined as impaired sleep quality and insomnia symptoms, number of studies (k) = 24, and number of participants (n) = 34,254) showed a medium-sized association, r =.27, 95% CI (.24,.30). The random-effects model of the correlation between loneliness and subjective sleep quality (k = 15, n = 24,018, r =.26, 95% CI (.22,.31)) and insomnia symptoms (k = 9, n = 10,236, r =.28, 95% CI (.24,.33)) similarly showed medium-sized associations. There was evidence of heterogeneity in effect size across all estimates (Q-statistics at p <.05). See Supplementary Appendix E for details on these analyses and see Figure 2 for forest plot.

11All but one study found a bivariate correlation between loneliness and subjectively measured sleep quality and insomnia symptoms. Both studies that measured sleep quality objectively (a: polysomnography;: actigraphy) found a bivariate association for a dimension of sleep quality. All studies that assessed sleep adequacy, sleep satisfaction, or change in sleep (b) detected a bivariate association between these variables and loneliness. None of the studies that examined sleep duration detected an effect. However, found that sleep duration mediated the relationship between loneliness and both diabetes and cardiovascular disease. See Figure 3 for a forest plot of effect sizes not included in analyses.

12Most studies did not control for potential confounds. See Table 2 for a narrative summary of findings on the relationship between loneliness and sleep with adjustments for confounds. Adjustments varied substantially across studies. A common adjustment was for depression, which attenuated the relationship between loneliness and both sleep quality and insomnia symptoms across all studies accounting for this factor.

13The Baujat plot identified the third study by as an outlier (see Supplementary Appendix F). The study was thus excluded from main analyses but included in sensitivity analyses.

14There was no evidence that mean age (Q(1) = 0.30, p =.58) or gender (Q(1) = 0.24, p =.63) moderated the association between loneliness and sleep disturbance.

15The risk of publication bias was examined for the cross-sectional literature using a funnel plot, rank correlation test, and Egger’s regression test. Visual inspection of the funnel plot (see Supplementary Appendix G), the rank correlation test (Kendall’s tau =.10, p =.50), and the Egger’s regression test (z = 1.05, p =.29) did not suggest publication bias.

16Excluding all estimates that did not meet the statistical assumptions of the r statistic did not appreciably change results. Including the study that was previously deemed to be an outlier inflated effect size estimates. See Supplementary Appendix E for details.

17Only eight studies evaluated loneliness as a risk factor for sleep difficulties. These studies varied considerably in terms of measures, sample, length of follow-up (3 months to 7 years), attrition rates (5.5%–56.1%), handling of potential confounds, and handling of attrition. Table 3 contains a narrative summary of the findings of each study, which varied. Two studies led by examined the inverse relationship—sleep problems as a risk factor for loneliness. In the first study, change in sleep did not predict loneliness. In the second, insomnia symptoms predicted loneliness 5 weeks later, though not when controlling for depression.

18This review aimed to synthesize the research on the relationship between loneliness and sleep. Studies examined different sleep outcomes, to include quality, duration, insomnia symptoms, adequacy, satisfaction, and change in sleep. Loneliness correlated with higher self-reported sleep disturbance (defined as impaired sleep quality and insomnia symptoms). Loneliness was also associated with sleep inadequacy and dissatisfaction, but not sleep duration. However, there was no evidence that the relationship between loneliness and sleep disturbance was moderated by age or gender. There is, as of yet, insufficient evidence to identify loneliness as a risk factor for sleep difficulties, due to the inconsistency of results and methodology across this small (k = 8) set of studies.

19A strength of the current body of literature is the wide array of samples, from agrarian Anabaptists to higher education students in Norway. The diversity of samples suggests that the association between loneliness and sleep is robust, and likely not limited to specific populations. However, this requires empirical examination: particularly of import in the United States is whether research can speak to the US population. None of the samples were representative of the US population, necessitating research using a nationally representative sample.

20A second strength of the cross-sectional body of literature is the use of both subjective and objective sleep measures. Subjective and objective measures tap into different parts of the sleep experience, and thus are both important to collect. Moreover, the use of different types of measures helps to mitigate the potential bias stemming from the limitations of each measure. Measures of loneliness varied across studies, ranging in quality from single-items to scales with strong psychometric properties. Scales also varied in whether they assessed loneliness directly or indirectly. Direct measures of loneliness use the term “lonely,” whereas indirect measures circumvent this word (e.g. “Do you feel alone?”). The advantage of direct measures of loneliness is that they tap directly into the construct of interest, whereas indirect measures are more likely to tap into related constructs, such as social support. However, responses on direct measures of loneliness could be biased by the stigmatization of loneliness. The use of both direct and indirect measures is thus a strength of the current body of literature: if findings replicate across both types of measure it is likely that the findings do not stem from the bias of either.

21This review found a medium-sized effect for the cross-sectional association between loneliness and sleep disturbance, both in terms of lower sleep quality and higher insomnia symptoms. Controlling for other factors, especially depression, attenuated the association between loneliness and sleep disturbance, indicating that the association is not independent of depression. However, simply controlling for depression does not speak to the interplay between depression, loneliness, and sleep disturbance: further research is necessary to examine how loneliness and sleep disturbance occur in the context of depression and other factors, such as age, race, and gender.

22There was no evidence that either age or gender moderated the association between loneliness and sleep disturbance. The finding that gender does not moderate the connection between sleep and loneliness is consistent with previous meta-analyses examining psychosocial factors—to include social relationships, isolation, loneliness, and social support—and health, which consistently fail to identify gender as a moderator. However, this literature is mixed in terms of the identification of age as a moderator: a meta-analysis by showed that the link between mortality, loneliness, isolation, and living alone was stronger in younger samples; a meta-analysis by showed that the link between mortality and social support was stronger in older samples; and a meta-analysis by did not find differences in the strength of the connection between mortality and social relationships according to age. These apparent contradictions in the literature—and the fact that the present review did not detect age as moderator—likely stem from the constellation of risk and resilience factors in later adulthood. Older adulthood is marked both by increased strengths—via enhanced coping strategies—and increased vulnerabilities—via decreased ability to recover from the sustained arousal accompanying stressors. As such, it is possible that both protective and detrimental processes are occurring in older adulthood, which in turn shape the connection between loneliness and sleep disturbance in opposing directions.

23Eight studies evaluated loneliness as a risk factor for sleep difficulties, with differing conclusions, as well as two studies that assessed the inverse relationship, sleep disturbance as a risk factor for loneliness. These studies differed in terms of their outcome of interest, length of follow-up, measures, samples, analyses, and handling of attrition and potential confounds. The variability in methodology, in tandem with the paucity of studies, makes it difficult to pinpoint which factors are driving the differences in results.

24The present systematic review and meta-analysis must be interpreted in light of its limitations. First, the review does not include gray literature or unpublished findings. However, there was no evidence of publication bias (as examined by visual inspection of a funnel plot, the rank correlation test, and Egger’s regression test). Second, the review is limited in its ability to speak to the relationship between sleep disturbance and other aspects of social relationships, in particular isolation. The correlation between loneliness and isolation is modest yet their effects on mortality are comparable. As such, loneliness and isolation likely influence health through different, though potentially overlapping, mechanisms. This review speaks to one potential pathway through which isolation could affect health: isolation leads to loneliness, which in turn influences sleep disturbance; additional research is necessary to pinpoint other pathways.

25Third, this review only examines two potential moderators: age and gender. However, there was evidence of study heterogeneity, suggesting the presence of moderators. Further research is necessary to identify and evaluate these moderators, which could include socioeconomic status, comorbid health conditions, depression, anxiety, and race. Fourth, this review speaks only to loneliness as a correlate and risk factor for sleep disturbance. However, these two lines of evidence are necessary but not sufficient to establish sleep disturbance as a mechanism through which loneliness deteriorates health. The present review was limited to observational literature; experimental research is necessary to speak to causality and therefore mechanisms.

26There is a medium-sized correlation between loneliness and sleep disturbance, but not sleep duration, across a wide array of measures and samples. Accounting for other factors—in particular depression—attenuated this association across all studies that assessed for potential confounds. The literature on loneliness as a risk factor for sleep disturbance is inconclusive due to variability in methodology and findings. This review indicates that loneliness is associated with impaired sleep quality and insomnia symptoms. Further research is necessary to determine directionality (i.e. whether loneliness precedes sleep disruption or the reverse), assess how other factors such as depression play into this association, and speak to causality using experimental design. The inconclusiveness of the current literature precludes the ability to draw conclusions as to whether sleep disturbance is a mechanism for the connection between loneliness and health. However, this review represents a critical step in organizing and evaluating the current research with bearing on sleep disturbance as a mechanism, thus seeking to fill a research gap that has remained largely unaddressed for the past 30 years: how social relationships shape health.