Sex-specific associations of social isolation and loneliness with residual life expectancy at age 45 years among middle-aged and older adults in China

Abstract

Background

Globally, one-third of older adults experience social isolation or loneliness, making them critical public health priorities. However, the impact of social isolation and loneliness on life expectancy remains underexplored. This study investigated the separate and joint associations of social isolation and loneliness with residual life expectancy at age 45 years, alongside the sex-specific variations in these associations with the aim of informing targeted strategies for mitigating social health disparities in rapidly ageing populations.

Methods

This prospective cohort study was conducted using data from the China Health and Retirement Longitudinal Study. A total of 11,315 community dwellers (5,274 men; 6,041 women) aged 45 years and above included. After the baseline 2011 assessment year, participants from the 2013, 2015, 2018, and 2020 waves were followed up. The measurements included the following: social isolation, assessed on the basis of social network usage, activities, and engagements; loneliness, evaluated on the basis of the subjective feeling of loneliness; and a flexible parametric Royston–Parmar model to estimate hazard ratios (HR) for all-cause mortality and to predict residual life expectancy differences.

Results

In men, socially isolated and lonely individuals had the greatest reduction in residual life expectancy at age 45 years, losing 4.61 years (95% CI: 1.49–7.74) compared to those with neither experience. Social isolation alone (3.82 years lost) and loneliness alone (2.83 years lost) resulted in significant reductions. In women, social isolation alone was significantly associated with reduced residual life expectancy (3.11 years lost, 95% CI: 0.68–5.55), whereas loneliness alone and combined exposure did not show statistically significant effects.

Conclusions

Significant sex-specific differences were identified in the impact of social isolation and loneliness on residual life expectancy. Thus, implementing sex-specific public health strategies that focus on dual psychosocial intervention for men and social integration for women is crucial to mitigate premature deaths in ageing populations.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-23708-x.

Introduction

Social isolation and loneliness are two distinct yet interrelated psychosocial phenomena prevalent in ageing societies. Social isolation, defined as the objective lack of social relationships or contact [1], differs fundamentally from loneliness, which reflects a subjective dissatisfaction with social relationship quality [2]. Globally, one-third of older adults experience social isolation (33%) or loneliness (31.6%) [3, 4], making them critical public health priorities. Compelling evidence suggests that deficient social relationships independently elevate mortality risk, with effects comparable to smoking, obesity, and physical inactivity [5, 6]. However, critical knowledge gaps persist regarding their sex-specific health impacts and long-term consequences on population longevity.

The current understanding of sex-specific associations between social relationships and mortality remains fragmented, particularly for loneliness. Although a meta-analysis of 90 cohorts found no significant sex-specific effects (women: hazard ratio [HR] = 1.01, 95% CI, 0.98–1.05; men: HR = 1.09, 95% CI, 0.99–1.20) [7], another meta-analysis reported stronger mortality risks in lonely men compared to lonely women (men: HR = 1.44, 95% CI, 1.19–1.76; women: HR = 1.26, 95% CI, 1.07–1.48) [8]. While these studies mostly focus on Western countries, a study conducted in the Chinese cohort of the oldest-old (mean age 86) found no loneliness-mortality link in either sex (men: HR = 1.01, 95% CI, 0.97–1.05; women: HR = 1.01, 95% CI, 0.96–1.06) [9], potentially reflecting survival bias in extreme longevity cohorts. These contradictory findings underscore the need to examine earlier life stages, when social relationship patterns are established, and contextualise findings within cultural settings where family structures may buffer social impairments differently by sex.

Moreover, previous studies have primarily focused on relative risk metrics to assess the impact of social relationships on mortality (e.g., hazard ratios). While informative, these relative metrics lack translational value for public health policymaking as they fail to quantify population-attributable longevity losses. Life expectancy estimation addresses this gap by converting relative risks into absolute measures that directly reflects preventable years of life lost. Notably, despite growing recognition of psychosocial determinants of aging, rigorous quantification of life expectancy deficits attributable specifically to social isolation and loneliness remains scarce. Furthermore, although social isolation and loneliness represent distinct constructs with partial overlap (r = 0.10–0.20) [10, 11], few existing studies have examined their joint effect. This oversight is problematic, given emerging evidence that their combined presence may synergistically accelerate health decline [12, 13], particularly when cultural norms create a discordance between objective social networks and subjective satisfaction (e.g. obligatory family co-residence without emotional closeness).

Leveraging 10-year data from the China Health and Retirement Longitudinal Study (CHARLS), this study fills these gaps by quantifying the sex-specific associations of social isolation and loneliness (individually and jointly) with life expectancy at age 45 within the unique Asian familial context, informing targeted strategies for mitigating social health disparities in rapidly ageing populations.

Materials and methods

Study population

The CHARLS, established in 2011, is an ongoing, nationally representative longitudinal survey of individuals aged 45 years or above. The survey assesses physical and mental health, socioeconomic status, family dynamics, employment, and retirement. In the 2011 wave, 17,708 participants from 10,257 households, 150 counties or districts, and 450 villages in 28 provinces of China were recruited using multistage stratified probability proportionate sampling methods [14]. After the baseline assessment, participants from the 2013, 2015, 2018, and 2020 waves were followed up. Detailed information on CHARLS has been published previously [14]. The CHARLS survey project was approved by the Biomedical Ethics Committee of Peking University (approval number: IRB00001052-11015) in accordance with the Declaration of Helsinki, and all participants signed informed consent forms. A total of 11,315 participants were eligible for the main analyses (age range: 45–101 years). The sample selection process is shown in Supplementary Figure S1.

Measures

Definition of social isolation

An index of social isolation was created on the basis of social networks, activities, and engagements. In line with previous literature [12, 15, 16], the index included 4 indicators: living alone, having fewer than once a month contact with children, participating in social activities, and currently unmarried. The detailed definitions of each indicator are provided in the Methods section of the Supplementary Material. The participants were assigned 1 point if they met each of the abovementioned criteria. The total social isolation score ranged from 0 to 4, with higher values representing greater isolation levels. Considering that it would be positively skewed, the social isolation score was further classified it into non-isolated (score < 2) and isolated (score ≥ 2) [12, 15].

Assessment of loneliness

Loneliness was assessed by asking the residents how often they felt lonely. The 4-point response categories were as follows: almost never, seldom, often, and always. This measure is strongly correlative with multi-item loneliness scales and widely used in studies [17, 18]. Loneliness was treated as a dichotomous variable (1 = lonely and 0 = not lonely), where the responses of ‘often’ or ‘always’ were classified as ‘lonely’ and those of ‘never’ or ‘seldom’, as ‘not lonely’ [17, 18].

Assessment of mortality

Mortality was determined using death certificates or by interviews with informants during follow-up. In the event of death, the survival time was measured from baseline until the participants’ exact time of death or until the median time from the first interview to the wave with a recorded death. For participants with no death record over the study period, their survival time was calculated as the interval between the date of the 2011 wave and the date of the last available wave.

Confounders

Potential confounders were identified from the literature [5, 7, 8]. All confounders were measured at baseline and treated as fixed characteristics in the analysis, consistent with the primary objective of estimating residual life expectancy based on initial exposure status. These confounders included sociodemographic (educational level, residential status, and employment status), lifestyle (smoking, alcohol intake, sleep duration, and body mass index [BMI]), and health-related factors (self-rated health status, comorbidities, depression, and activities of daily living [ADL]).

Educational level was categorised as not educated, primary school, middle school, high school, or above. Residential status before the age of 16 years was classified as rural or urban. Employment status was classified as currently employed, unemployed/on the welfare system/retired.

Participants who reported not smoking or had quit smoking at least three years before were classified as the low-risk group for smoking [19]. The low-risk group for alcohol intake included non-regular drinkers and daily light-to-moderate drinkers (daily < 30 g of pure alcohol for men and < 15 g for women) [20]. Adequate sleep was defined as 7–9 h of sleep per night. Additionally, participants with a BMI of 18.5–23.9 kg/m² were defined as having normal BMI.

Self-rated health was initially recorded at five levels (very good, good, fair, poor, and very poor) and then categorised into dichotomous variables: good (very good and good) and poor (fair, poor, and very poor) [21]. Comorbidities were defined as self-reported diagnoses of two or more chronic diseases by a physician. Depression was assessed using the 10-item Centre for Epidemiologic Studies Depression scale [22]. To avoid redundancy, items related to loneliness and sleep quality were excluded, resulting in an eight-item scale. The total score ranged from 0 to 24, with lower scores indicating fewer depressive symptoms. Cronbach’s alpha for the reduced depression scale was 0.74, indicating acceptable internal consistency. ADL were evaluated on the basis of self-rated performance in six basic activities, and ADL disability was defined as difficulties in one or more activities [23].

Statistical analysis

Baseline variables were presented as means (standard deviation) and frequencies (percentages). Statistical differences in characteristics were compared using t-tests and χ² test for continuous and categorical variables, respectively. HR and 95% CIs of all-cause mortality were estimated using a flexible parametric Royston–Parmar proportion-hazards model, with age as the time scale. The Royston–Parmar model, which utilises restricted cubic splines, offers flexibility in modelling the baseline hazard for censored survival data [24]. In the model, the baseline log cumulative hazard was modeled using a spline with 4 degrees of freedom, corresponding to 3 internal knots placed at default percentiles of the log time distribution.

To calculate sex-specific losses of life expectancy, first, the survival curve was calculated on the basis of the parametric model predicted for each individual and averaged for all. Next, residual life expectancy was estimated as the area under the survival curve by integrating the curve up to the age of 100 years, conditional on survival at ages 45–100 years (1-year intervals). Thirdly, losses of residual life expectancy and 95% CIs were calculated as the difference between the areas under the survival curves given the absence and the presence of exposure categories of interest, respectively. We estimated 95% confidence intervals for the losses of residual life expectancy using the bootstrap method. We constructed the above model separately for men and women. First, we examined social isolation and loneliness as separate exposures. Second, we combined them into a 4-level categorical variable (no isolation/no loneliness, isolation only, loneliness only, both) to assess their joint effects.

In the sensitivity analysis, we also confirmed the joint impact of isolation and loneliness on residual life expectancy at ages 65 and 85 years. To minimize potential reverse causality, we excluded participants who died during the first follow-up wave in 2013. In addition, we tested for statistical interactions between sex and social isolation/loneliness by including interaction terms in the survival models.

All the analyses were adjusted for confounding factors. All statistical analyses were performed using Stata (version 17.0). Graphs were plotted using R software (version 4.4.1).

Results

Participants’ characteristics

In the median follow-up period of 9.0 years (interquartile range [IQR] 8.92–9.08; 91,802.83 person-years), 11,315 participants (men: 46.61%) were included in the study, with a mean baseline age of 58.40 (9.57) years. All-cause mortality was observed in 1385 individuals (849 men and 536 women).

The sex-specific baseline characteristics are presented in Table 1. Women were generally more socially isolated and lonely, whereas men were generally older, more educated, and more likely to be unemployed. They also smoked and drank more frequently than women did. The subgroup characteristics based on social isolation and loneliness are shown in Supplementary Table S1.

Table 1.

Baseline characteristics of the participants

	All (n = 11,315)	Men (n = 5,274, 46.61%)	Women (n= 6,041, 53.39%)	P value
Age, mean (SD), years	58.40 (9.57)	59.21 (9.32)	57.70 (9.73)	< 0.001
Educational level, n (%)
Not educated	5,134 (45.37)	1,621 (30.74)	3,513 (58.15)	< 0.001
Primary school	2,548 (22.52)	1,450 (27.49)	1,098 (18.18)
Middle school	2,363 (20.89)	1,412 (20.77)	951 (15.74)
High school or above	1,270 (11.22)	791 (15.00)	479 (7.93)
Residential status, n (%)				0.659
Urban	992 (8.77)	469 (8.89)	523 (8.66)
Rural	10,323 (91.23)	4,805 (91.11)	5,518 (91.34)
Employment status, n (%)				< 0.001
Employed	10,112 (89.37)	4,629 (87.77)	5,483 (90.76)
Unemployed	1,203 (10.63)	645 (12.23)	558 (9.24)
Smoking, n (%)				< 0.001
Low risk	7,564 (66.85)	1,911 (36.23)	5,653 (93.58)
High risk	3,751 (33.15)	3,363 (63.77)	388 (6.42)
Alcohol intake, n (%)				< 0.001
Low risk	10,149 (89.66)	4,202 (79.67)	5,947 (98.43)
High risk	1,166 (10.34)	1,072 (20.33)	94 (1.57)
Sleep duration, n (%)				< 0.001
Low risk	5,168 (45.67)	2,508 (47.55)	2,660 (44.03)
High risk	6,147 (54.33)	2,766 (52.45)	3,381 (55.97)
BMI, n (%)				< 0.001
Normal	5,991 (52.95)	3,125 (59.25)	2,866 (47.43)
Abnormal	5,324 (47.05)	2,149 (40.75)	3,175 (52.57)
Self-rated health, n (%)				< 0.001
Good	2,567 (22.69)	1,335 (25.31)	1,232 (20.39)
Poor	8,748 (77.31)	3,939 (74.69)	4,809 (79.61)
Comorbidities, n (%)				< 0.001
Yes	4,290 (37.91)	1,882 (35.68)	2,408 (39.86)
No	7,025 (62.08)	3,392 (64.32)	3,633 (60.14)
Depression, mean (SD)	16.43 (3.88)	15.83 (3.62)	16.95 (4.03)	< 0.001
ADL disability, n (%)				0.834
Yes	466 (4.12)	215 (4.08)	251 (4.15)
No	10849 (95.88)	5,059 (95.92)	5,790 (95.85)
Lonely				< 0.001
Lonely	1,901 (16.79)	729 (13.82)	1,171 (19.38)
Not lonely	9,415 (83.21)	4,545 (86.18)	4,870 (80.62)
Social isolation				0.004
Isolated	1,493 (13.19)	644 (12.21)	849 (14.05)
Not isolated	9,822 (86.81)	4,630 (87.79)	5,192 (85.95)
Joint effects				< 0.001
Not isolated and not lonely	8,436 (74.56)	4,125 (78.21)	4,311 (71.36)
Lonely and not isolated	1,386 (12.25)	505 (9.58)	881 (14.58)
Isolated and not lonely	979 (8.65)	420 (7.96)	559 (9.25)
Isolated and lonely	514 (4.54)	224 (4.25)	290 (4.80)

Data are presented as the mean (SD) for continuous variables and numbers (percentage) for categorical variables

BMI Body mass index, ADL activities of daily living

P-values were calculated using t-test for continuous variables and 𝜒2 test for categorical variables

Sex differences in the associations of social isolation and loneliness with All-Cause mortality

After adjusting for potential confounders, socially isofated men exhibited a significantly higher risk of all-cause mortality compared to non-isolated men (HR, 1.39, 95% CI, 1.17–1.65). Similarly, men reporting loneliness had an elevated risk of mortality compared to those without loneliness (HR, 1.31, 95% CI, 0.09–1.56). In the joint analysis, men experiencing both social isolation and loneliness had the highest mortality risk (HR, 4.61, 95% CI, 1.49–7.74) compared to those with neither experience. Men who were lonely but not socially isolated (HR = 2.83, 95% CI: 0.30–5.37) and those who were socially isolated but not lonely (HR = 3.82, 95% CI: 1.32–6.32) also showed increased mortality risks, although to a lesser extent (Table 2).

Table 2.

Separate and joint associations of social isolation and loneliness with all-cause mortality and loss of residual life expectancy (95% CI) at age 45 years

	Pearson-years	No. of deaths/no. at risk	HR (95%CI)	P value	Loss of residual life expectancy [95% CI], 45 y
Men
Separate effects
Social isolation
Not isolated	36,753.58	671/4,630	1.00 [Reference]		1.00 [Reference]
Isolated	4,602.38	178/644	1.39 (1.17, 1.65)*	< 0.001	3.64 (1.56, 5.71)*
Loneliness
Not lonely	35,903.13	677/4,545	1.00 [Reference]		1.00 [Reference]
Lonely	5,452.83	172/729	1.31 (0.09, 1.56)*	0.004	2.91 (2.81, 5.01)*
Joint effects
Not isolated and not lonely	32,885.75	566/4,125	1.00 [Reference]		1.00 [Reference]
Not isolated and lonely	3,867.83	105/505	1.29 (1.04, 1.61)*	0.023	2.83 (0.30, 5.37)*
Isolated and not lonely	3,017.38	111/420	1.41 (1.15, 1.74)*	0.001	3.82 (1.32, 6.32)*
Isolated and lonely	1,585	67/224	1.52 (1.17, 1.97)*	0.002	4.61 (1.49, 7.74)*
Women
Separate effects
Social isolation
Not isolated	43,287.54	368/5,192	1.00 [Reference]		1.00 [Reference]
Isolated	6,364	168/849	1.38 (1.14, 1.68)*	0.001	3.01 (0.92, 5.10)*
Lonely
Not lonely	40,219.75	186/4,870	1.00 [Reference]		1.00 [Reference]
Lonely	9,431.80	54/1,171	0.99 (0.82, 1.23)	0.939	−0.07 (− 1.97, 1.82)
Joint effects
Not isolated and not lonely	36,021.38	290/4,311	1.00 [Reference]		1.00 [Reference]
Not isolated and lonely	7,266.17	78/881	0.95 (0.73, 1.24)	0.721	−0.44 (− 2.84, 1.96)
Isolated and not lonely	4,198.38	109/559	1.40 (1.11, 1.77)*	0.005	3.11 (0.68, 5.55)*
Isolated and lonely	2,165.63	59/290	1.31 (0.97, 1.77)	0.079	2.50 (− 0.46, 5.47)

The model was adjusted for sociodemographic (educational level, residential status, and employment status), lifestyle (smoking, alcohol intake, sleep duration, and body mass index), and health-related factors (self-rated health status, comorbidities, depression, and activities of daily living)

HR Hazard ratio, CI Confidence interval

^*P<0.05

Among women, social isolation was associated with a higher mortality risk compared to non-isolated women (HR, 3.01, 95% CI, 0.92–5.10). However, loneliness did not significantly increase mortality risk compared to that among individuals without it (HR, 0.99, 95% CI, 0.82–1.23). In the joint analysis, compared to women who experienced neither loneliness nor social isolation, only social isolation without loneliness was significantly associated with mortality (HR, 3.11, 95% CI, 0.68–5.55). Neither loneliness without social isolation (HR, − 0.44, 95% CI, − 2.84–1.96) nor the combination of both loneliness and social isolation (HR, 2.50, 95% CI, − 0.46–5.47) showed significant associations with mortality.

Sex differences in the associations of social isolation and loneliness with residual life expectancy at 45 years

Social isolation and residual life expectancy

Socially isolated men had a significantly shorter residual life expectancy at age 45 (29.18 years, 95% CI: 26.01–32.36) compared to non-isolated men (32.82 years, 95% CI: 30.03–35.62), corresponding to a loss of 3.64 years (95% CI: 1.56–5.71). Similarly, socially isolated women showed reduced residual life expectancy at age 45 (35.07 years, 95% CI: 31.77–38.38) versus non-isolated women (38.08 years, 95% CI: 35.02–41.15), with a loss of 3.01 years (95% CI: 0.92–5.10) (Fig. 1A; Table 2).

Fig. 1.

Life expectancy and years of life lost based on social isolation and/or loneliness (A) Estimated life expectancy at age 45 years based on social isolation (B) Estimated life expectancy at age 45 years based on loneliness (C) Estimated life expectancy at age 45 years based on social isolation and loneliness The model was adjusted for sociodemographic (educational level, residential status, and employment status), lifestyle (smoking, alcohol intake, sleep duration, and body mass index), and health-related factors (self-rated health status, comorbidities, depression, and activities of daily living)

Loneliness and residual life expectancy

Loneliness in men was associated with a shorter residual life expectancy at age 45 (29.96 years, 95% CI: 26.81–33.11) compared to non-lonely men (32.87 years, 95% CI: 30.06–35.68), resulting in a loss of 2.91 years (95% CI: 2.81–5.01). In contrast, loneliness had no meaningful impact on women’s residual life expectancy at age 45. Lonely (37.27 years, 95% CI: 34.08–40.47) and non-lonely women (37.20 years, 95% CI: 34.23–40.17) showed nearly identical estimates, with a non-significant loss of − 0.07 years (95% CI: −1.97–1.82) (Fig. 1B; Table 2).

Joint effects of social isolation and loneliness

The combined exposure to social isolation and loneliness was linked to the greatest residual life expectancy reduction at age 45. Men with both conditions had a residual life expectancy of 28.76 years (95% CI: 24.91–32.60), representing 4.61 additional years lost (95% CI: 1.49–7.74) compared to those with neither condition. Social isolation alone (29.55 years, 95% CI: 26.05–33.05; 3.82 years lost, 95% CI: 1.32–6.32) and loneliness alone (30.53 years, 95% CI: 27.07–34.00; 2.83 years lost, 95% CI: 0.30–5.37) also showed significant losses.

For women, only social isolation without loneliness was significantly associated with reduced residual life expectancy (34.84 years, 95% CI: 31.31–38.37; 3.11 years lost, 95% CI: 0.68–5.55). Neither loneliness alone (− 0.44 years lost, 95% CI: −2.84–1.96) nor the combination of both conditions (2.50 years lost, 95% CI: −0.46–5.47) reached statistical significance (Fig. 1C; Table 2).

Results from sensitivity analyses are shown in Supplementary Table S2-S5. The resident life expectancy estimates remained robust using sex-specific HRs at the age of 65 and 85 years. After excluding deaths that occurred in the first follow-up period, the results remained largely unchanged. In addition, we conducted the interaction test to analysis the interaction between loneliness and social isolation with sex, while no multiplicative interaction was found.

Discussion

To the best of our knowledge, this study is the first to quantify the disparities among social isolation, loneliness, and mortality using life expectancy metrics, focusing on sex differences, in a nationally representative sample of middle-aged and older adults. This study found sex-specific patterns in loss of residual life expectancy associated with social isolation and loneliness. Men who experienced both conditions showed the greatest residual life expectancy loss (4.61 years), demonstrating a synergistic effect exceeding individual exposures, whereas women exhibited significant reduction only with social isolation (3.11 years). The striking disparity by sex suggests that cultural shaping of gender roles in relationship maintenance and stress coping mechanisms may differentially modulate the health consequences of psychosocial deficits.

Moreover, a culturally nuanced sex disparity was observed in the association between loneliness and mortality among middle-aged and older adults in China, with loneliness significantly reducing residual life expectancy at age 45 in men but not in women. This finding contrasts with a meta-analysis of predominantly Western countries, which reported that loneliness was associated with increased mortality risk in both sexes, with a stronger effect observed in men [8]. It also differs from a previous Chinese study, which reported no significant association between loneliness and mortality for either sex [9].

These discrepancies may be attributed to differences in both the population age composition and cultural context examined in these studies. The previous study in China focused on an older cohort, with a mean age of 86 years, substantially higher than the mean age of 58 years in the current study. Individuals who survive to a very old age likely represent a ‘survivor cohort’ with inherently stronger health or social resources, which may buffer against the mortality risks associated with loneliness earlier in life. In contrast, younger-old adults may be more susceptible to the adverse health consequences of loneliness, leading to a stronger association with reduced residual life expectancy.

Furthermore, the male-specific vulnerability observed in this study may be shaped by cultural dynamics unique to China. In Chinese society, older men often rely heavily on workplace and spousal relationships for social support, both of which become vulnerable to disruption after retirement or widowhood, potentially exacerbating mortality risks [25]. By contrast, traditional gender roles in Confucian societies assign women caregiving responsibilities, fostering broader kinship and social networks that provide a buffer against loneliness through sustained social engagement. Additionally, cultural norms stigmatising male emotional expression (e.g. the ‘strong silent male’ ideal) may lead to chronic internalisation of loneliness. This psychological strain has been biologically linked to elevated cortisol levels in Chinese men experiencing loneliness [26], whereas women who are socially permitted to express emotions more freely may benefit from stress alleviation through social sharing. Finally, lonely men are more likely than women to engage in unhealthy behaviours, including poor nutrition, physical inactivity, and inadequate chronic disease management [27], further increasing their risk of premature death. These findings highlight the necessity of contextualising loneliness-mortality relationships within cultural and age-specific resilience frameworks. Future research should further explore the mechanisms underlying these sex differences to inform targeted public health interventions.

The findings suggest that social isolation is linked to mortality and residual life expectancy in both men and women among Asian countries, which complements the results of meta-analyses conducted mostly in Western countries [7, 28]. This study further indicates a synergistic effect of isolation and loneliness on residual life expectancy only in men, indicating that in socially isolated men, residual life expectancy further decreases if they experience loneliness. Social isolation predominantly affects physical and cognitive health [29], whereas loneliness primarily affects psychological health, such as increasing risks of depression and anxiety [30]. When loneliness and social isolation coexist, the combined psychological stress and impaired physical health may exacerbate premature deaths risk.

Our findings have important implications for policy and practice. They highlight distinct intervention priorities for extending life expectancy across sexes: reducing social isolation emerges as the key strategy for women, whereas dual mitigation of both loneliness and social isolation is critical for men. Objective social isolation (e.g. limited relationship quantity and infrequent social contact) presents a more modifiable target than subjective loneliness, given its responsiveness to structural interventions. This underscores the urgency to develop sex-specific psychosocial programs, particularly for men. Interventions should commence before advanced age, as health benefits initiated earlier in life demonstrate comparable protective effects in later years. First, community nurses and public health workers should incorporate social isolation and loneliness assessments into routine check-ups, identify at-risk individuals early, and develop care plans that address these issues to avoid further health-related concerns. Second, community organisations can develop programmes that foster social integration and alleviate loneliness. Leveraging innovative technologies can help connect individuals who are physically isolated through virtual social media platforms and telehealth services. Third, governments should prioritise social connectivity in public health agendas, develop policies to support and fund programmes aimed at reducing social isolation and loneliness, and launch public awareness campaigns regarding their dangers.

This study has several strengths. First, it analysed extensive and well-characterised cohort data, including a large sample size, long follow-up, and a high number of documented deaths, which helped obtain more reliable sex-specific effect estimates for residual life expectancy. Second, it filled the evidence gap regarding the associations among isolation, loneliness, and residual life expectancy, and sex-specific differences in these associations among populations of less-developed economies and with various socioeconomic characteristics.

This study also has some limitations. First, the sample was from only one country; therefore, further investigations are required in other populations (e.g. other cross-cultural countries, oldest-old groups, and nursing home residents). Second, social isolation and loneliness were assessed only at baseline, yet both are dynamic and potentially fluctuating conditions over time. Reliance on a single time-point measurement may have led to exposure misclassification, potentially underestimating or overestimating the true associations with mortality. Future studies incorporating longitudinal assessments of social relationships could provide a more accurate understanding of how temporal patterns of isolation and loneliness influence life expectancy. Third, the possibility of residual confounding cannot be ruled out, although several potential covariates identified in the theoretical frameworks and previous research were rigorously controlled. Finally, information on the causes of death was unavailable after the 2015 wave, restricting the analysis of the association between loneliness and isolation concerning cause-specific mortality and limiting the exploration of the potential mechanisms underlying the findings.

Conclusions

The results of our large-scale lifespan analysis of middle-aged and older adults in China suggest that social isolation is significantly associated with loss of residual life expectancy in men, and its effect is exacerbated in the presence of loneliness. Social isolation alone was significantly associated with reduced residual life expectancy in women. These findings underscore the need for targeted sex-difference psychosocial interventions in health policies and community programmes. Further research is warranted to explore the mechanisms underlying these sex differences and to refine interventions that promote social integration and improve health outcomes.

Authors’ contributions

M. Z.: Conceptualisation, Funding Acquisition, Methodology, Data Curation, Writing– Original Draft, Writing– Review & Editing; X. H.: Writing– Review & Editing; H. Z.: Data Curation, Writing– Review & Editing; C. W.: Writing– Review & Editing; S. P.: Data Curation, Funding Acquisition; S. S.: Data Curation; M. L.: Supervision, Writing– Review & Editing; K. W.: Supervision, Writing– Review & Editing. All authors have read and approved the final manuscript submitted for publication.

Funding

This work was supported by the National Natural Science Foundation of China (grant no. 82404388); Humanities and Social Science Department of the Ministry of Education Planning Fund (grant number 23YJCZH314); Social Science Foundation of Shandong Province (grant number 21DSHJ05); and Key Natural Science Research Project of Higher Education Institutions in Anhui Province (grant number 2024AH050977).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

CHARLS received ethical permits from the Biomedical Ethics Committee of Peking University (approval number: IRB00001052-11015).

Competing interests

The authors declare no competing interests.