Study design and population
The CLHLS is a large-scale, ongoing prospective cohort study being conducted across half of the counties or municipalities in 22 of China’s 31 provinces. The primary objective of the CLHLS is to investigate the factors influencing healthy aging, longevity, and mortality within the Chinese population. Initiated in 1998, the study has been carried out in multiple waves, with follow-up data collected every 2–3 years. The dataset provides individual – level information on demographics, health indicators, socioeconomic characteristics, and social and behavioral risk factors. Rigorous evaluations have ensured the dataset’s high integrity, including low random attrition, reliable measurement scales, and accurate age reporting. A detailed description of the CLHLS is available elsewhere [14, 15].
For this analysis, we utilized longitudinal data from participants initially recruited during the 2002 (2002 wave), 2005 (2005 wave), 2008/2009 (2008 wave), 2011/2012 (2011 wave), and 2014 waves, with follow-up data extending through the 2018 wave (2017–2019). The 1998 wave was excluded from the analysis due to substantial differences in how the frequency of playing cards/mahjong was classified in 1998 compared to the other survey waves. Between March 2002 and November 2014, a total of 35,467 participants were interviewed. To focus on older adults, we applied exclusion criteria, removing participants younger than 65 years (n = 600), those lost to follow-up (n = 5,465), cases with incorrect death dates (death preceding baseline interview, n = 144), and individuals with missing data on playing cards/mahjong (n = 4). A flowchart detailing the study sample inclusion process is presented in Fig. 1. Ultimately, 29,254 participants were included in the final analysis. This study was approved by the Biomedical Ethics Committee of Peking University (IRB00001052-13074).
Flowchart of the included study population
Assessment of playing cards/mahjong
Participants were asked about their current frequency of playing cards/mahjong at baseline, with responses categorized as: (1) never, (2) sometimes (at least once a month or sometimes), (3) often (at least once for a week), or (4) almost every day. This variable was treated as a categorical variable in the analysis.
Outcome assessment
The primary outcome of this study was all-cause mortality. Deaths were ascertained through follow-up interviews conducted every two to three years, and the date of death was recorded based on family reports and official death certificates when available. For those still alive at the end of follow-up, data were censored at their last interview date.
Covariates assessment
Our model comprehensively integrated baseline sociodemographic factors, lifestyle behaviors, and health status indicators. The sociodemographic variables included age, gender (male or female), education level (no formal education, 1–6 years, or more than 6 years), residential setting (rural or urban), marital status (married or other, including divorced, widowed, or never married), living arrangements (living with family, living alone, or in an institutional setting), and economic status (economically dependent or independent). Economic status was defined based on primary financial support sources: participants were classified as “economically independent” if their primary income came from pensions or their own labor/work, and “economically dependent” if their primary support came from other sources (e.g., spouse, children, grandchildren, other relatives, or government/community assistance). Lifestyle behaviors were evaluated across multiple domains, including smoking status (never, current, or former), alcohol consumption (never, current, or former), and regularity of exercise (never, current, or former). Dietary habits were assessed by the frequency of vegetable and fruit intake (daily, frequently, occasionally, rarely, or none), as well as the consumption frequency of meat, fish, and eggs (categorized as daily, weekly, monthly, occasionally, rarely, or none).
Health characteristics included body weight, the number of natural teeth (< 10, 10–20, or ≥ 20), denture use, limitations in activities of daily living (ADL) (yes or no), cognitive function, and self-reported, physician-diagnosed conditions such as hypertension, heart disease, cerebrovascular disease, diabetes, respiratory diseases (e.g., pneumonia, bronchitis, emphysema, asthma), and cancer (all recorded as yes or no).
Body weight measurements were taken by trained staff following standard protocols, with participants dressed in light indoor clothing and without shoes. ADL were assessed using the Katz Index scale [16], which evaluates six self-care tasks: bathing, dressing, eating, indoor transferring, toileting, and continence. Disability was defined as requiring assistance or experiencing difficulty in performing one or more of these six activities [17]. Participants were categorized into two disability levels: Low level disability: requiring assistance with 1 to 4 ADL items. High level disability: requiring assistance with 5 to 6 ADL items [18]. Cognitive function was assessed using the Chinese-adapted version of the Mini-Mental State Examination (MMSE; Supplementary Table 1). The MMSE evaluates six cognitive domains: time/place orientation, word registration, attention/calculation, memory, visual construction, and language, with total scores ranging from 0 to 30. To define cognitive impairment, education-adjusted cutoff scores were applied as follows: (1) MMSE < 18 for illiterate participants; (2) MMSE < 21 for those with 1–6 years of education; and (3) MMSE < 25 for individuals with > 6 years of schooling [19, 20].
Statistical analysis
Baseline characteristics of the study population were summarized as means and standard deviations (SDs) for continuous variables and as percentages for categorical variables. Missing data were addressed using Multiple Imputations by Chained Equations with predicted mean matching, enhancing the robustness of our analysis by generating five imputations. Detailed information on missing variables is provided in Supplementary Table 2.
We assessed the proportional hazards assumption using the Schoenfeld residual test, confirming no violations. Kaplan-Meier survival curves were generated to compare unadjusted survival probabilities across the four card/mahjong-playing frequency groups. Log-rank tests were used to assess differences between groups. Cox proportional hazard models were applied to determine the association between playing cards/mahjong and all-cause mortality. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated across different frequencies of playing cards/mahjong, with “never” as the reference group. Model 1 adjusted for age and sex, Model 2 added adjustments for variables such as marital status, education, residence, living arrangement, economic status, smoking, drinking, regular exercise, body weight, and dental health. Model 3 further adjusted for dietary factors (fruit, vegetable, meat, fish, egg intake) and pre-existing conditions (cognitive function, hypertension, heart disease, cerebrovascular disease, diabetes, respiratory disease, and cancer). The crude incidence rate (IR) of all-cause mortality was calculated per 1000 person-years. Since data on card/mahjong playing frequency at follow-up were unavailable, we excluded: Participants who died during the first follow-up (n = 12,949); Those with missing follow-up exposure data (n = 61). We then analyzed participants with repeated exposure assessments during follow-up. To minimize bias from unmeasured changes in card/mahjong playing frequency, we restricted analyses to individuals with stable exposure patterns (i.e., no change in frequency between baseline and follow-up). Additionally, we assessed mortality risk among participants who transitioned between playing frequencies. Interaction and subgroup analyses were conducted based on age, sex, marital status, residence, living arrangement, economic status, and cognitive function using a multivariable-adjusted model.
Sensitivity analyses were performed to ensure robustness. First, to address potential bias from missing data, we conducted a complete case analysis (n = 26411 after excluding participants with any missing values in covariates variables). Second, individuals with prevalent major chronic diseases (heart disease, diabetes, cerebrovascular disease, respiratory disease, or cancer) at baseline were excluded to address potential reverse causality. Third, to address potential reverse causality and better align with the study’s focus on cognitively/socially engaging activities, we excluded individuals with ADL limitations or cognitive impairment at baseline. Fourth, to account for potential age and cohort effects, we included birth decade (e.g., < 1899, 1900–1909, 1910–1919, 1920–1929, etc.) and enrollment wave (2002, 2005, 2008, 2011, 2014) as covariates in additional sensitivity analyses. Lastly, we excluded deaths within the first year to mitigate the impact of short-term follow-up, as the effects of playing cards/mahjong are likely to manifest over a longer period.
All analyses were performed using R (version 4.2.2), with statistical significance set at P < 0.05.