We use data from CHARLS,¹ a nationally representative longitudinal follow-up survey of individuals aged 45 years and older and their spouses in China. Detailed information related to the study design has been reported elsewhere (15). Concerning data collection, CHARLS applies a variety of methods to collect data, including face-to-face interviews to collect questionnaire data, physical examinations to obtain body measurements, and collection of blood samples to obtain biomarker data. Of these, for the collection of questionnaire data, the specific practice was for trained interviewers to use Computer Assisted Personal Interviewing to conduct face-to-face surveys, systematically presenting questions through a computer program in accordance with the questionnaire’s logical structure and recording respondents’ answers directly into the system. Following each data collection wave, CHARLS compiles and publicly releases complete survey data, to date, five waves of survey data have been publicly released. The CHARLS program was approved by the Biomedical Ethics Committee of Peking University (No. IRB00001052–11015) for all baseline and follow-up surveys and all participants signed an informed consent form before data collection.

In this study, we utilized data from five waves of CHARLS surveys in 2011 (baseline survey), 2013 (first wave of follow-up), 2015 (second wave of follow-up), 2018 (third wave of follow-up), and 2020 (fourth wave of follow-up) for analysis. Specifically, a total of 17,705 participants were enrolled in the baseline survey in 2011. We initially excluded participants aged < 45 years (n = 364) and those who did not participate in the social activity survey at baseline (n = 1,431), resulting in a final sample of 15,883 participants for the follow-up survey (Figure 1). During the follow-up period, a total of 1,509 participants (9.50%) were lost to follow-up. A comparison of the characteristics of those who were lost during follow-up with those who were not is shown in Supplementary Table S1.

Social participation was assessed by asking participants the following equally weighted question: “Have you done any of these activities in the last month” (16). The specific social activities include: (1) Interacted with friends; (2) Played Ma-jong, played chess, played cards, or went to community club; (3) Provided help to family, friends, or neighbors who do not live with you and who did not pay you for the help; (4) Went to a sport, social, or other kind of club; (5) Took part in a community-related organization; (6) Done voluntary or charity work; (7) Cared for a sick or disabled adult who does not live with you and who did not pay you for the help; (8) Attended an educational or training course; (9) Stock investment; (10) Used the Internet and (11) engaged in other activities. The numbers of participants in each social activity are detailed in Supplementary Table S2. For each participant, we calculated the total number of social participations by cumulatively summing the assigned values for the 11 activities described above. If the total number of social participations is 0, it means that the participant did not engage in social participation; conversely, a total of ≥ 1 means that the participant did engage in social participation (16, 17). To assess the effect of the level of social participation on all-cause mortality, we further categorized the total number of social participations into three categories based on previous literature, that is, totals of 0, 1, and ≥ 2 (7, 16).

Our interest outcome was all-cause death, which was determined based on the interview status (alive or dead) of the participants at the time of each follow-up survey (18, 19). Simultaneously, we also collected the survival time of the participants in the study. It is important to note that in the CHARLS dataset, the exact time of death of participants was only recorded in the Wave 1 Follow-up Survey and Wave 4 Follow-up Survey. If a precise time was recorded at the time of the participant’s death event, survival time was calculated as the interval from the date of the baseline investigation to the exact date of the participant’s death; Conversely, if the exact time to the death event was not available, survival time was estimated based on the median time from the date of the baseline survey to the wave in which the death was recorded (18, 19).

Based on previous literature (20, 21), we collected information on several covariates to adjust for confounders (Supplementary Table S3). The depressive symptom was measured using the 10-item Center for Epidemiologic Studies Depression Scale (CES-D) (22), which was categorized as with (CES-D ≥ 10 scores) or without (CES-D < 10 scores) depressive symptom based on depressive symptom score (23). Cognitive function was assessed using the Chinese version of the Mini-Mental State Examination (MMSE) for two related domains of cognitive function (situational memory and mental integrity) and was categorized as either cognitively impaired (MMSE < 11) or cognitively intact (MMSE ≥ 11) based on MMSE scores (24). The Activities of Daily Living (ADL) were assessed using the ADL scale, which covers 11 domains: eating, bathing, dressing, indoor moving, toileting, continence control, performing housework, shopping, cooking, taking medicine, and financial management (25). ADL disability was determined based on whether an individual is unable or has difficulty completing any of the 11 items (25). Physical frailty was measured using a modified version of the Fried physical frailty phenotype approach, which consists of five dimensions: shrinking, weakness, slowness, low physical activity, and exhaustion (26). If participants met three or more criteria they were defined as with and without physical frailty (27). Hypertension was defined as participant self-reported hypertension diagnosed by doctors, using antihypertensive medication, systolic blood pressure ≥ 140 mmHg, or diastolic blood pressure ≥ 90 mmHg (28). Diabetes mellitus was defined as participant self-reported diabetes diagnosed by doctors, use of anti-diabetic medications (including use of insulin or oral hypoglycemic agents), fasting blood glucose ≥ 126 mg/dL, or glycosylated hemoglobin ≥ 6.5% (29). Heart disease, stroke, and cancer were determined based on participants’ self-reported physician diagnoses.

The normal distribution is expressed as mean ± standard deviation for continuous variables, median (interquartile range) for non-normally distributed continuous variables, and frequency (percentage) for categorical variables. The group comparisons of baseline characteristics were compared using the t-test or Mann–Whitney U test for continuous variables and the χ²-test for categorical variables.

Person-years of follow-up were calculated for each participant from the date of the baseline survey until the date of the occurrence of the endpoint event (death), loss of follow-up, or end of follow-up, whichever came first. The Cox proportional hazards model with hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) was employed to assess the relationship between social participation and all-cause mortality. Three models were applied to minimize the role of potential confounders, including model 0 unadjusted for any covariates, model 1 adjusted only for age and sex, and model 2 based on model 1 further adjusted for the place of residence, marital status, educational attainment, annual household living expenditures, smoking status, alcohol consumption status, depressive symptom, cognitive function, physical frailty, disability in activities of daily living, hypertension, diabetes mellitus, heart disease, stroke, and cancer. To further assess the relationship between the level of social participation and the risk of all-cause mortality, the amount of social participation was converted to a three-categorical variable by 0, 1, and ≥ 2 for inclusion in the Cox model and tested for a trend. Meanwhile, the restricted cubic spline (RCS) regression model was used to assess the potential nonlinear association between social participation and all-cause mortality using the 10th, 50th, and 90th percentiles of amounts of social participation as nodes and the 10th percentile as reference (30).

In addition, considering that age and sex characteristics may be influencing the association between social participation and all-cause mortality, subgroup analyses were conducted by age (< 60 years, ≥ 60 years) and sex (male, female) (16), with potential interactions assessed by adding interaction terms. Finally, two sensitivity analyses were carried out to assess the robustness of the primary results. First, to minimize potential reverse causality, we repeated the main analysis after excluding participants with less than 2 years of follow-up (31). Second, to fully utilize the available sample data, we performed multiple imputations of chained equations (MICE) for data missing covariate characteristics and repeated the main analysis using the dataset with multiple imputations (32, 33).

All the above analyses were performed using R version 4.0.5. All the hypothesis tests were performed using two-sided tests, and P values less than or equal to 0.05 were considered statistically different.

A total of 15,883 participants, a median age was 58 (52, 66) years, and a percentage of males 47.66%, were included in the final analysis sample of this study (Table 1). There were 8,027 (51.67%) participants in the study with social participation at the time of the baseline survey. In comparison to the participants of the study who were socially engaged and not socially engaged, a statistically significant difference was observed in age, gender, place of residence, education, annual household living expenditure, smoking status, alcohol consumption status, depressive symptom, cognitive function, physical frailty, disability in activities of daily living and diabetes mellitus characteristics (P < 0.05).

The median follow-up of the participants in the study was 9.01 (8.67, 9.01) years, with a cumulative follow-up of 128,789 person-years, and deaths occurred in a total of 2,175 participants (Table 2 and Supplementary Tables S4, S5). Among them, the participants without social participation at baseline developed 1,205 deaths with an incidence density of 18.99/1,000 person-years, while those with social participation at baseline developed 970 deaths with an incidence density of 14.84/1,000 person-years. After adjusting for all potential confounders, results from Cox proportional hazards regression modeling showed that participants who participated in social activities at baseline had an all-cause risk of death of 0.88 (HR = 0.88, 95% CI: 0.78, 0.99) compared with participants who did not participate in social activities at baseline.

Further assessing the relationship between the extent of social participation and all-cause mortality (Table 2 and Supplementary Table S6) showed that after adjusting for all potential confounders, compared with participants with no social participation, those with 1 item of social participation had an all-cause mortality risk of 0.93 (HR = 0.93, 95% CI: 0.81, 1.06), and those with ≥ 2 items of social participation was 0.77 (HR = 0.77, 95% CI: 0.64, 0.94). The results of the test for trend showed that the participant’s risk of all-cause mortality decreases progressively as the extent of social participation increases (P for trend = 0.011).

In addition, an RCS regression model was used to assess the nonlinear relationship between social participation and all-cause mortality (Figure 2), which showed that there was insufficient evidence to support a nonlinear relationship between the amount of social participation and all-cause mortality after adjusting for all potential confounders (P _nonlinear = 0.355).

The results of subgroup analyses by age showed that after adjusting for all potential confounders, the association between social participation and the risk of all-cause mortality was statistically significant only among those ≥ 60 years of age (HR = 0.88, 95% CI: 0.80, 0.97), but no significant effect modification was observed (P _interaction = 0.571). Similarly, the analysis of the results by sex subgroup showed that the association between social participation and risk of all-cause mortality was statistically significant in both males (HR = 0.90, 95% CI: 0.82, 0.98) and females (HR = 0.84, 95% CI: 0.72, 0.99), but no significant effect modification was observed either (P _interaction = 0.440) (Table 3).

The results of the two sensitivity analyses showed that both sensitivity analysis 1 (excluded participants with less than 2 years of follow-up) and sensitivity analysis 2 (utilized the multiple imputations dataset) presented results that remained generally consistent with the main results (Table 2) in terms of the magnitude and direction of the association effect, suggesting that our results are robust (Supplementary Table S7).