A systematic review and meta-analysis of 90 cohort studies of social isolation, loneliness and mortality

The associations between social isolation, loneliness and the risk of mortality from all causes, cardiovascular disease (CVD) and cancer are controversial. We systematically reviewed prospective studies on the association between social isolation, loneliness and mortality outcomes in adults aged 18 years or older, as well as studies on these relationships in individuals with CVD or cancer, and conducted a meta-analysis. The study protocol was registered with PROSPERO (reg. no. CRD42022299959). A total of 90 prospective cohort studies including 2,205,199 individuals were included. Here we show that, in the general population, both social isolation and loneliness were significantly associated with an increased risk of all-cause mortality (pooled effect size for social isolation, 1.32; 95% confidence interval (CI), 1.26 to 1.39; P < 0.001; pooled effect size for loneliness, 1.14; 95% CI, 1.08 to 1.20; P < 0.001) and cancer mortality (pooled effect size for social isolation, 1.24; 95% CI, 1.19 to 1.28; P < 0.001; pooled effect size for loneliness, 1.09; 95% CI, 1.01 to 1.17; P = 0.030). Social isolation also increased the risk of CVD mortality (1.34; 95% CI, 1.25 to 1.44; P < 0.001). There was an increased risk of all-cause mortality in socially isolated individuals with CVD (1.28; 95% CI, 1.10 to 1.48; P = 0.001) or breast cancer (1.51; 95% CI, 1.34 to 1.70; P < 0.001), and individuals with breast cancer had a higher cancer-specific mortality owing to social isolation (1.33; 95% CI, 1.02 to 1.75; P = 0.038). Greater focus on social isolation and loneliness may help improve people’s well-being and mortality risk. In a systematic review and meta-analysis of 90 prospective cohort studies, Wang et al. find a significant association of both social isolation and loneliness with increased risk of all-cause mortality.


Characteristics of the included studies
Supplementary Tables 5-9 present the characteristics of all included studies. Of the 90 included studies, 29 were conducted in the USA, and 61 were performed in other countries, such as the UK, Japan, Korea and Finland, with most (90%) conducted in developed countries. Publication dates ranged from 1986 to 2022, and the follow-up duration ranged from 2 to 24.4 years in general populations and from 6 months to 20.44 years in patient populations. Sample sizes ranged from 119 to 580,182 for SI and from 227 to 466,901 for loneliness. In total, 2,205,199 participants (1,485,469 for SI and 1,209,407 for loneliness) were included in this systematic review. All participants were aged 18 years or older, and most (70%) were aged 50 years or more. In most studies, controlling for confounders of conventional risk factors was carried out, including age, sex, body mass index, smoking and alcohol consumption. Other variables such as physical activity, education level, depression, cognitive function, diabetes and hypertension were also adjusted in some specific studies.
On the basis of the Newcastle-Ottawa scale, all studies had a low risk of bias in all components. A total of 43 studies had a rating of 7; 30 studies had a rating of 8, which was considered high quality; and 17 studies with a rating of 6 were considered medium quality. Supplementary Table 10 presents the details of the risk-of-bias assessment.

SI or loneliness and all-cause mortality
In total, 38 and 45 papers, including 1,485,469 and 1,209,407 participants, investigated the association of SI and loneliness, respectively, with all-cause mortality in the general population. The total pooled effect estimate was 1.32 (95% CI, 1.26 to 1.39; P < 0.001) for SI and 1.14 (95% CI, 1.08 to 1.20; P < 0.001) for loneliness (Fig. 2). Substantial heterogeneity was observed among the included studies (I 2 = 77.8%, P < 0.001 for SI; I 2 = 91.1%, P < 0.001 for loneliness). Significant publication bias was detected using funnel plots ( Supplementary Fig. 1a) in the SI analysis, as well as in Egger's (P = 0.006) and Begg's (P = 0.019) tests. However, the pooled effect size was not substantially changed after adjustment using the trim-and-fill method (Supplementary Table 11). Little evidence of publication bias was observed in the loneliness analysis (P = 0.465 in Egger's test; P = 0.021 in Begg's test; Supplementary Fig. 1b). Additionally, five papers including 54,561 participants reported the combined effects of SI and loneliness on all-cause mortality when co-existing in the general population. The pooled effect estimate was 1.18 (95% CI, 1.05 to 1.32; P < 0.001) with significant heterogeneity (I 2 = 79.2%, P = 0.001) ( Supplementary Fig. 2a). No evidence of publication bias was detected (Egger's test, P = 0.849; Begg's test, P = 0.806) (Supplementary Table 11).
A total of 21 and 18 papers reported associations between SI and all-cause mortality among men and women, respectively, in the general population. The pooled effect estimate was 1.39 (95% CI, 1.27 to 1.51; P < 0.001; I 2 = 78.6%; P < 0.001; Egger's test, P = 0.003; Begg's test, P = 0.053) in men and 1.44 (95% CI, 1.28 to 1.61; P < 0.001; I 2 = 79.4%; P < 0.001; Egger's test, P = 0.306; Begg's test, P = 0.612) in women (Supplementary Figs. 2b,c and 3a,b and Supplementary Table 11). Additionally, 11 and 9 papers reported associations between loneliness and all-cause mortality in men and women, respectively. However, our meta-analysis found no evidence for these associations in men (1. In subgroup analyses according to country, follow-up time, education, depression, smoking, drinking, body mass index, ethnicity, chronic disease and the assessment methods of SI and loneliness, social network, having infrequent social contacts or possibly living alone 3 . By contrast, loneliness is a subjective feeling of distress, arising when there is a discrepancy between desired and actual social relationships 4 . Several possible pathways have been proposed to explain the adverse effects of SI and loneliness on health and mortality [5][6][7] . SI may promote unhealthy behaviours, such as malnutrition and physical inactivity 6 , which may increase the risk of death. SI has also been associated with health-related biological processes, such as higher C-reactive protein and lipid levels and poorer immune function 5,8 . In terms of loneliness, mechanisms including deleterious health behaviours, sleep disorders, and neuroendocrine and immune dysfunction have been identified as contributing to the negative association of loneliness and health 7,9 . The findings of epidemiologic studies on the associations between SI, loneliness and mortality are controversial. Some studies have found a higher risk of mortality from all causes, cardiovascular disease (CVD) and cancer associated with SI and loneliness [10][11][12][13] , whereas others have reported non-significant results [14][15][16][17] . The only meta-analysis 18 up to 2015 found that SI and loneliness were factors contributing to all-cause mortality, with SI increasing all-cause mortality by 29% (95% CI, 1.06 to 1.56) and loneliness increasing all-cause mortality by 26% (95% CI, 1.04 to 1.53). Many subsequent studies have been carried out on the relationship between SI or loneliness and mortality, but the results are inconsistent 12,14,[19][20][21] . Additionally, studies have shown that the link between social support and health is bidirectional, which could lead to a vicious cycle where poor health causes patients to lose social support (for example, SI) over time, but patients tend to need social support more than the general population. However, no systematic review of an association of SI or loneliness with the risk of mortality in patient populations (for example, individuals with CVD or cancer) has been conducted.
We therefore conducted an updated and more comprehensive systematic review and meta-analysis to summarize the associations between SI, loneliness and risk of mortality from all causes, CVD and cancer in the general population, as well as in a subset of individuals with CVDs or breast/colorectal malignancies. Our findings can provide vital clarification to help in promoting population health management and improving primary health services.

Literature search
We identified 14,358 records (9,328 for SI and 5,030 for loneliness) through electronic database searches of PubMed, Web of Science and Embase. After we excluded 14,176 articles that did not meet the selection criteria, 182 records remained (118 for SI and 64 for loneliness). Following a full-text review, 81 articles were further excluded for the following reasons: 28 studies did not consider SI or loneliness as an independent variable, 17 studies did not analyse mortality as an outcome variable, 20 studies had duplicated data, 13 studies had insufficient data, 2 studies enrolled individuals with HIV and HIV-infected/uninfected older veterans, and 1 study was conducted among older patients in the intensive care unit. Additionally, 18 articles analysing both SI and loneliness were retrieved repeatedly and later removed, and 7 studies were added after a manual search and review of the reference lists. The independent retrieval processes are shown in Fig. 1.

SI or loneliness and CVD mortality
Fifteen and eight papers, including 1,265,108 and 487,982 participants, examined the association of SI and loneliness, respectively, with the risk of mortality from CVD or circulatory disease in the general population. The pooled effect estimate was 1.34 (95% CI, 1.25 to 1.44; P < 0.001; I 2 = 63.3%; P < 0.001) for SI and CVD or circulatory disease mortality, with non-significant publication bias in Egger's (P = 0.160) and Begg's tests (P = 0.363) (Fig. 3a, Supplementary Fig. 5a and Supplementary  Table 11). But the pooled effect estimate was non-significant for loneliness (1.14; 95% CI, 0.97 to 1.35; P = 0.118; I 2 = 77.3%; P < 0.001; Egger's test, P = 0.012; Begg's test, P = 0.371) (Fig. 3b, Supplementary Fig. 5b and Supplementary Table 11). There was no significant heterogeneity observed for sex, country, follow-up duration or adjustment for confounding, except for education level (P = 0.024; Supplementary Table  15) in the subgroup analysis of SI and CVD mortality.

SI or loneliness and cancer mortality
Thirteen and three papers examined the association of SI and loneliness, respectively, with the risk of mortality from cancer in the general population. A total of 1,171,644 and 476,404 participants were analysed, and positive associations with cancer mortality were observed for SI ( Table 11). The subgroup analysis of SI and cancer mortality did not reveal sources of heterogeneity (Supplementary Table 16).

SI or loneliness and all-cause mortality in individuals with CVD
The association of SI and loneliness with all-cause mortality in individuals with CVD was examined in eight and four papers, including 487,722 and 50,936 participants, respectively. The pooled effect estimate was 1. 28

SI and mortality in individuals with cancer
Among individuals with cancer, only those with breast or colorectal cancer were analysed for the relationship between SI and mortality in original studies. On the basis of 7 papers including 21,913 individuals with cancer, there were significant associations with pooled effects for all-cause mortality (1.47; 95% CI, 1.33 to 1.63; P < 0.001) and for cancer-specific mortality (1.26; 95% CI, 1.04 to 1.52; P = 0.016) (Supplementary Fig. 8a Excluded by title and abstract review on SI (n = 9,210) and loneliness (n = 4,966) Full-text reports retrieved for detailed evaluation on SI (n = 118) or on loneliness (n = 64) Full-text articles excluded on SI (n = 68) or on loneliness (n = 13) Articles included in the meta-analysis (n = 90) Manually retrieved articles (n = 7) Loneliness was not an independent variable (n = 1) Mortality was not the outcome variable (n = 3) Duplicated reports (n = 3) Insu icient information, di erent forms of outcome data, unable to extract the e ect size (n = 6) Articles on loneliness and mortality identified through database searching (n = 5,030): PubMed (n = 1,141) Embase (n = 2,126) Web of Science (n = 1,763) SI was not an independent variable (n = 27) Mortality was not the outcome variable (n = 14) Duplicated reports (n = 17) Insu icient information, di erent forms of outcome data, unable to extract the e ect size (n = 7) The subjects were patients with HIV and HIV-infected/uninfected older veterans (n = 2) The subjects were elderly critically ill patients in intensive care units with a history of social isolation exposure (n = 1) Articles included in the meta-analysis on SI and mortality (n = 50) or on loneliness and mortality (n = 51) Duplication: articles on both SI and loneliness and mortality (n = 18)

Fig. 1 | Studies included in the present review.
The retrieval processes shown found 90 studies that met the selection criteria and were relevant to this meta-analysis.

GRADE assessment
Supplementary Table 11 shows the GRADE assessments of the overall certainty of the evidence for the associations between SI, loneliness and risk of mortality from all causes, CVD and cancer in the general population and in individuals with CVD or cancer. Owing to the observational study design, most evidence for these pooled effect estimates was graded as low (n = 4) or very low (n = 14), all downgraded because of inconsistency or publication bias.

Discussion
With a sample size of more than two million, this systematic review and meta-analysis provides a large set of evidence on the associations between SI and loneliness and mortality from all causes, CVD and cancer. We found that in the general population, both SI and loneliness were significantly associated with an increased risk of all-cause and cancer mortality. SI also increased the risk of CVD mortality. No positive combined effect of SI and loneliness on all-cause mortality was observed.
The findings also highlight that socially isolated individuals with CVD or breast cancer had increased all-cause mortality, and individuals with breast cancer had higher cancer-specific mortality due to SI. A recent representative study in 20 high-, middle-and low-income countries on five continents showed that SI increased the risk of all-cause mortality by 26% (95% CI, 1.17 to 1.36) 11 . Those researchers held the view that SI is correlated with increased mortality on a par with or greater than traditional risk factors such as alcohol use, smoking and obesity 42,101 . By contrast, in a study by Hoogendijk et al. 53 , SI was not found to be significant (P = 0.41; HR = 1.06; 95% CI, 0.92 to 1.21)  Weights are from random e ects analysis.
Identifying and understanding gender differences in the relationship between SI or loneliness and all-cause mortality may help in improving gender equity and promoting population health status. Ward et al. 16 observed that SI increased the risk of all-cause mortality in women (HR = 1.66; 95% CI, 1.08 to 2.54) but not in men (1.18; 95% CI, 0.79 to 1.75). Gronewold et al. 28 found the opposite result, with SI positively increasing all-cause mortality in men (HR = 2.45; 95% CI, 1.54 to 3.87) but not in women (1.11; 95% CI, 0.77 to 1.62). Similarly, Holwerda et al. 88 found that loneliness was associated with increased mortality only in men (P < 0.05; HR = 1.3; 95% CI, 1.04 to 1.62). Generally, the prevalence of SI or loneliness varies by gender 22 . Owing to differences in occupation, social position and physiological specificity, women are more susceptible to feelings of loneliness than men 104 , but women have larger social networks than men, even later in life 105 . Living alone and having a lack of interpersonal connections seem particularly predominant and detrimental in men 106 . Additionally, considerable differences in other exposures closely related to SI and loneliness (such as smoking, unhealthy diet, physical inactivity and health care utilization by gender subgroups) are objectively common, especially in some countries with a low sociodemographic index. Studies may be hindered by limited statistical power, and discrepant results produced in a single study may be attributed to adjustment for different confounding factors. For example, Henriksen et al. 75 did not adjust for physical activity, depression or alcohol consumption in their research, and Novak et al. 17 did not adjust for a history of diabetes or hypertension. We found a significantly increased risk for all-cause mortality with SI in both sexes in the meta-analysis. Although marginal non-significance within groups was observed, loneliness was associated with an obvious increase in all-cause mortality when combining 45 cohorts, which provides reliable evidence regarding this complex relationship.
Because a small number of respondents experienced high levels of both loneliness and SI, the included literature lacked studies analysing the combined effect or interaction of SI and loneliness. Loneliness has been shown to predict mental health 32,107 (for example, depression), and SI has been shown to predict cognitive and physical health 107 . Beller and Wagner 50 found a synergistic effect on mortality; however, the combined effect (1.18; 95% CI, 1.05 to 1.32) of SI and loneliness was not larger than the separate effects (1.32 for SI and 1.14 for loneliness) according to our meta-analysis, and SI alone most strongly influenced premature mortality. A similar inference has been put forth by Thoits 108 , who stated that people who are lonely but not socially isolated have mental health stress but might be resilient to it because of their social networks. Ward et al. 16 also suggested that study participants reporting low loneliness and high SI showed a higher mortality risk than the group with high loneliness and low SI. Socially isolated individuals should therefore be given higher priority than individuals experiencing loneliness in terms of social attention.
SI and loneliness have been reported to have a critical role in the mechanisms of CVD incidence, progression and mortality 109 , including hypothalamic-pituitary-adrenocortical (HPA) activation 110 , sympathetic nervous system hyperactivity 111 , parasympathetic dysfunction 112 and pro-inflammatory immune response 113 . Smith et al. 114 indicated that the risk of death without hospital admission was substantially higher in the most socially isolated individuals than in the least isolated individuals for coronary heart disease and stroke. A meta-analysis in 2016 reported an increased risk of coronary heart disease and stroke in individuals reporting SI 115 . However, most included studies reported effect sizes on a small scale, and the meta-analysis evidence was limited by publication bias. It remains unclear whether these associations are independent of biological, behavioural, psychological, health and socio-economic factors. On the basis of the fully adjusted effect sizes from the original studies, our results confirmed that SI can significantly increase CVD mortality by 34% in the general population. However, because SI and loneliness have been associated with many of the conventional factors, such as low education, obesity, smoking and pre-existing chronic illness, the lack of these adjustments may lead to an underestimation of the true effect size. For example, the present meta-analysis of the effect of SI on CVD mortality was influenced by education (P = 0.024), possibly because SI is more common among less-educated and unemployed individuals 11 . Additionally, Hakulinen et al. 27 identified a 1.5-fold risk associated with SI for all-cause mortality in individuals after acute myocardial infarction or stroke. Similar results were found for SI in our meta-analysis, but not for loneliness after adjustment for publication bias. It thus seems that the deleterious effect of SI on mortality after cardiovascular events is stronger than that of loneliness. Further prospective studies and cumulative meta-analyses are needed to confirm this association.
By analysing data from the UK Biobank cohort study, Elovainio et al. 32 revealed a significant association (HR, 1.32; 95% CI, 1.22 to 1.43) between SI and cancer mortality. However, Alcaraz et al. 22 observed a positive correlation only in female and white participants in the USA; no significant association was observed in some other studies 16,37 . According to the summarized data from 11 studies conducted in five countries, our results showed a significant positive association (1.22; 95% CI, 1.18 to 1.27). The only three studies 16,30,32 on loneliness all showed meaningless results, but the pooled results revealed a weak positive correlation (1.09; 95% CI, 1.01 to 1.17). Hence, the risk of SI and loneliness for cancer-related mortality should not be ignored.
In the Nurses' Health Study among 2,835 postmenopausal women with breast cancer, Kroenke et al. 41 found that socially isolated women were twice as likely to die of breast cancer than socially integrated women. However, in the Women's Health Initiative Study 116 , Kroenke et al. found that associations between social networks and mortality outcomes depended on levels of social support and the burden in relationships. Accordingly, the difference may be explained by the degree of acceptance women experience in their social roles or self-perceived stress (such as loneliness or overburden) 17 . We summarized four studies including 18,955 individuals with breast cancer; the results showed that SI increased the risk of all-cause (1.51; 95% CI, 1.34 to 1.70) and cancer-specific mortality (1.33; 95% CI, 1.02 to 1.75). However, similar results were not found in patients with colorectal cancer. From the perspective of risk factors, breast cancer is more closely related to women's psychological factors, SI and social stress 117,118 , but colorectal cancer is more closely related to dietary and physical activity factors 119,120 . It is therefore essential to improve the social relationship status of individuals with cancer, especially those with breast cancer, to prolong survival time.
The mechanism related to SI, loneliness and mortality may be as follows. First, the HPA axis is the main producer of glucocorticoids 121 .
Article https://doi.org/10.1038/s41562-023-01617-6 There is clear evidence that SI and loneliness can lead to activation of the HPA axis in animals and humans 110,111,122 , which results in the release of cortisol, a physiological state of glucocorticoids. For example, the separation of pair-bonded prairie voles from their partners leads to an elevation of circulating corticosterone concentrations, indicating an activation of the entire HPA axis 110 . In human studies, socially isolated individuals among healthy middle-aged men and women in the Whitehall II cohort were found to show a greater cortisol arousal response and a greater total cortisol output over the course of the day 123 . In addition, lonely individuals have greater morning cortisol increases 124 , elevated circulating cortisol concentrations and impaired glucocorticoid receptor sensitivity, suggesting that loneliness also causes overaction of the HPA axis 125 . Continuous activation of the HPA axis and glucocorticoids affects a wide range of physiological functions, including glucose regulation, metabolism and inflammatory control; it also has cardiovascular, reproductive and neuronal effects 126 and increases the risk of CVD, cancer and mortality 122 . Second, both SI and loneliness are independently associated with negative mental health outcomes later in life, including higher rates of depression and cognitive decline 127,128 . Third, harmful behaviours associated with increased mortality are more common among people who are lonely or isolated, such as smoking, alcohol drinking, unhealthy dietary choices, and lower likelihoods of exercising 129 and adhering to prescription medications 85 . Finally, socially isolated individuals are less likely to receive emergency and routine medical care owing to their smaller social network 85 . Poor care provided by health care professionals who perceive this group as difficult to treat or time consuming further exacerbates adverse health outcomes 130 .
This study systematically evaluated the association of SI, loneliness and mortality from all causes, CVD and cancer. This study possesses the following strengths. First, owing to the prospective design of the included studies, recall bias and selection bias were effectively avoided. Second, the summarized analyses included the largest number of participants to date. Third, potential confounding was controlled to the greatest extent because only effect sizes from fully adjusted models in the original studies were included. Fourth, most included studies were of high quality, with a low risk of bias. Our meta-analysis also has some limitations. First, the source of heterogeneity is difficult to identify owing to a large number of internal adjustment factors in different original studies and the inconsistency of adjustment factors among studies. Potential unknown confounders may still exist in present analysis. Second, the measurement tools of SI and loneliness in the original studies are not completely consistent, which may affect the accuracy of our assessment. Third, most included studies were from high-income countries; additional results from other regions are required to comprehensively understand this topic. Fourth, SI was noted to be associated with an increased risk of suicide, which is also associated with premature death. Some of the original studies 23,25 controlled for this effect by excluding people who died prematurely if their outcome occurred within a year, but not all studies took this approach. It was therefore inevitable that our findings would be influenced by premature deaths from suicide, self-harm or other causes.
SI and loneliness are critical factors associated with an increased risk of all-cause, cardiovascular and cancer mortality. SI and loneliness may be exacerbated by post-cancer or CVD stress and less social contact as well as limited access to care, especially informal care from friends and family. Strategies are urgently needed to address this public health concern, such as raising awareness about the adverse health effects of SI and loneliness among health care professionals and the general public. It is also critical to develop interventions based on innovative technologies that mobilize resources from family members and community networks to address SI and loneliness. The health care system should also develop methods to identify SI and loneliness in health care settings so that health care professionals can provide appropriate clinical and public health interventions.

Methods
The findings of this systematic review and meta-analysis are reported in accordance with the Reporting Checklist for Meta-analyses of Observational Studies (MOOSE) 131 (Supplementary Table 1) and Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 132 (Supplementary Table 2). The protocol of this study was registered with PROSPERO on 5 February 2022 (reg. no. CRD42022299959) at https://www.crd.york.ac.uk/PROSPERO/.

Data sources and database search
We searched PubMed, Web of Science and Embase for studies published in English between database inception and October 2022. Supplementary Table 3 provides the details of the search terms used. Manual retrieval and a review of the reference lists of retrieved studies were also performed to avoid missing eligible studies.

Inclusion and exclusion criteria
All article titles and abstracts were reviewed independently by two authors (Y.G. and Z.H.). The study inclusion criteria were as follows: (1) studies with a prospective observational design including adults aged more than 18 years; (2) studies reporting effect sizes using HRs and relative risk with the corresponding 95% CIs, for the association of SI or loneliness as the exposure of interest and mortality from all causes, CVD or specific cancers as the outcome (diagnoses were classified according to the World Health Organization International Classification of Diseases 133,134 ); and (3) if multiple studies reported data from the same population, only the study with the largest sample size was included. The following studies were excluded: (1) commentaries, editorials, letters, reviews, unpublished studies or meta-analyses; (2) studies in which SI or loneliness was not considered as an independent variable; (3) studies performed on individuals with HIV; and (4) studies reporting effect sizes of mixed risk factors for disease and mortality.
Because there was no standardized assessment method, all original studies using differing measures to assess SI or loneliness were included, for greater statistical power. The details are shown in the Supplementary Information (Supplementary Table 4). Subgroup analysis and sensitivity analysis were used to assess the impact of the differential measures in the meta-analysis.

Data extraction
The data were extracted independently by two researchers (Y.G. and Z.H.). Any disagreement was resolved in discussion or by a project supervisor. The following data were extracted from each publication: name of the first author, publication year, study design, location of the study, age range, sex, cohort size, duration of follow-up, methods used for the assessment of SI or loneliness, and effect sizes of comparison results (for categorical variables (SI versus non-SI and loneliness versus non-loneliness) or graded analyses with the lowest level as the reference) together with 95% CIs and confounding variables adjusted in the statistical analyses. In cases where the study population consisted of only male (or female) participants, or when the results of a mixed population were analysed separately by sex, we considered each analysis as a distinct study. If the included studies reported both crude and multivariable-adjusted effect estimates, we extracted the most fully adjusted outcomes. The extracted data were sorted in Excel 2016 (ref. 135).

Statistical methods
We summarized HRs and relative risks (along with 95% CIs) for comparisons of SI with non-SI and loneliness with non-loneliness or graded variables in the original studies to calculate pooled effect estimates for the association of SI or loneliness and mortality outcomes (significant at an α level of 0.05). Heterogeneity was determined using Cochran's Q test (significant at P < 0.10) and quantified with the I 2 statistic (range, 0-100%). We took I 2 ≥ 50% to indicate high heterogeneity and used a random-effects model (DerSimonian and Laird); we considered I 2 < 50% Article https://doi.org/10.1038/s41562-023-01617-6 to indicate low heterogeneity and used a fixed-effects model. We examined the possibility of publication bias by using funnel plots and conducting Egger's and Begg's tests (each significant at P < 0.05). When there was publication bias, we used the trim-and-fill method 136 to adjust the influence of bias. Subgroup and meta-regression analyses were conducted using predefined criteria, including study country, duration of follow-up, and SI or loneliness assessment tools, and statistically controlling for confounders (such as education, depression, smoking, drinking, body mass index and chronic diseases). A polynomial contrast procedure was used to analyse the changing trends in social network index grade and risk of mortality. Additionally, we conducted a sensitivity analysis, in which each prospective cohort study was excluded in turn, to examine the influence of that study on the overall estimates. All statistical analyses were performed using Stata statistical software version 14.2 (StataCorp). All statistical tests were two-sided.

Risk-of-bias assessment and GRADE assessment
Because all eligible studies were designed with a cohort study, the Newcastle-Ottawa quality assessment scale for cohort studies was used to evaluate the methodological quality 137,138 of the included studies. The GRADE 139 method was used to assess the quality of evidence and generate a profile that ranks the evidence as high, moderate, low and very low certainty (the details are in the Supplementary Methods). Two authors (Y.G. and Z.H.) independently conducted a risk assessment of bias for each study and a GRADE evaluation of the results of the meta-analyses, and any inconsistencies between them were resolved by regular group discussion or adjudicated by a project supervisor. The ratings were conducted in GRADEprofiler 140 .

Patient and public involvement
There was no patient or public involvement in setting the research question or the outcome measures, in developing the study plans or in interpreting or writing up the results. There are no plans to involve patients in the dissemination of the research results. Our findings will be accessible to the general public as an open-access published article.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
All included literature is available in PubMed (https://pubmed.ncbi. nlm.nih.gov/), Web of Science (https://www.webofscience.com) and Embase (https://www.embase.com). The data that support the findings of this study are available from the corresponding author upon request.

March 2021
Corresponding author(s): Maoqing Wang and Yashuang Zhao Last updated by author(s): Mar 14, 2023 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy All included literature is available in PubMed, Web of Science, and Embase.

nature portfolio | reporting summary
March 2021

Human research participants
Policy information about studies involving human research participants and Sex and Gender in Research.

Reporting on sex and gender
The present study is a systematic review and meta-analysis, and all the information was collected from original studies. There was no patient or public involvement in setting the research question or the outcome measures, nor in developing the study plans or in interpreting or writing up the results. But, we collected the information of gender from original studies, and conducted meta-analysis by different genders.

Population characteristics
We collected the population characteristics from original studies. There was no patient or public involvement in setting the research question or the outcome measures, nor in developing the study plans or in interpreting or writing up the results.

Recruitment
There was no patient or public involvement in setting the research question or the outcome measures, nor in developing the study plans or in interpreting or writing up the results.

Ethics oversight
There was no patient or public involvement in setting the research question or the outcome measures, nor in developing the study plans or in interpreting or writing up the results. There are no plans to involve patients in the dissemination of research results. Therefore, there are no ethical issues involved.
Note that full information on the approval of the study protocol must also be provided in the manuscript.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf Behavioural & social sciences study design All studies must disclose on these points even when the disclosure is negative.

Study description
Systematic review and meta-analysis

Research sample
We searched PubMed, Web of Science, and Embase for all studies published in English and included all studies that met the inclusion criteria

Sampling strategy
Prior to literature search, we formulated clear inclusion and exclusion criteria and included all studies that met the criteria. According to the set search terms for automatic retrieval, and then, using the "snowball" method for manual retrieval.

Data collection
Before literature retrieval, we used excel to make a literature feature extraction table to collect the required data. Data were extracted independently by two researchers (Y.G. and Z.H.) from original studies. Any disagreement was resolved in discussion or by a project supervisor.Because the article type was systematic review and meta-analysis, there was no blinding involved.

Timing
From April 2022 to October 2022.

Data exclusions
The exclusion criteria were: 1) commentaries, editorials, letters, reviews, unpublished studies or meta-analyses; 2) studies in which SI or loneliness was not considered as an independent variable; 3) studies performed patients with HIV; 4) studies reporting effect sizes of mixed risk factors for disease and mortality.

Non-participation
Following a full-text review, 81 articles were further excluded for the following reasons: 28 studies did not consider SI or loneliness as an independent variable, 17 studies did not analyze mortality as an outcome variable, 20 studies had duplicated data, 13 studies had insufficient data, two studies enrolled patients with HIV and HIV-infected/uninfected older veterans, and one study was conducted among elderly patients in the intensive care unit.

Randomization
Since the types of articles are systematic review and meta-analysis, the included and excluded articles have clear inclusion and exclusion criteria, so randomization was not applicable to the study.
Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.