Data used in this study were derived from the China Health and Retirement Longitudinal Study (CHARLS) 2011-2015, and a detailed description of the data collection has been published (23). Briefly, the data were collected biannually from 2011 to 2015 through interviews with a nationally representative sample of adults aged 45 or above. Multistage cluster sampling was adopted for sample selection, and the overall response rate was 80.05% (24). The national baseline survey of CHARLS included 17,705 respondents from 10257 households. We included participants that had responded to at least two waves of surveys. Participants with missing data in both mental status and memory assessments or activities were excluded. A flowchart of the participant enrolment in this study is shown in Figure 1. Different from previous studies that only included participants aged above 65 years, we included participants of all ages and categorized them into three groups, respectively the middle-aged adults (45-54), the late middle-aged adults (55-64) and the older adults (65 or above). We aimed to examine the potential protective effect of activities against cognitive decline in different life phases.
Assessment of activities
The participants were interviewed about their engagement in activities in the last month, including “interacting with friends”, “playing Mah-jong, chess, cards or going to community club”, “providing help to family, friends, or neighbors who do not live with you for free”, “going to a sport”, “taking part in a community-related organization”, “doing voluntary or charity work”, “caring for a sick or disabled adult who does not live with you”, “attending a course”, “stock investment” and “using the Internet”. Frequency of doing the abovementioned activities was asked if the participant answered yes to any of the activities, and the frequency was categorized into “infrequent participation” and “daily participation”. We combined these two questions and re-categorized the frequency of doing each activity into “no participation”, “infrequent participation” and “daily participation”. Activities with “no participation” rate over 95% were excluded to ensure the activities have considerable acceptability within the studied population. A detailed description of the excluded activities is presented in Table S1. In the end, four activities were selected as independent variables of this study, respectively “interacting with friends”, “playing Mah-jong and other games”, “going to a sport” and “providing help to family, friends or neighbors”.
Assessment of cognitive functions
CHARLS adopted components of the Telephone Interview of Cognitive Status (TICS) battery (25) for cognitive assessment. The TICS evaluated respondents’ cognitive capacity in terms of episodic memory and mental status (26). In this study, we analyzed the protective effect of activities on these two constructs separately.
In CHARLS, memory was assessed by an immediate and a four-minute-delayed word recall of 10 Chinese words that were read to them (25, 27). The total score was calculated by averaging the number of words correctly recalled in the immediate and delayed tests, ranging from 0 to 10. Mental status was assessed by 11 questions: orientation was assessed by asking respondents the current year, season, month, day, and day of the weak; numeric ability was assessed by serial subtraction of 7 from 100 (up to five times); visuospatial ability was assessed by respondents to draw the figure shown to them (two overlapped pentagons). The total score on mental status was the number of correct answers, ranging from 0 to 11 (28). For both assessments, higher scores indicate better cognitive ability.
Assessment of covariates
Sociodemographic information, comorbidity and health behaviors were considered as covariates in this study. Sociodemographic information included age (years), gender (female and male), residence (urban and rural), highest education (illiterate, elementary and below, junior high and above) and marital status (married with spouse or not). Comorbidity included self-reported psychiatric problems (yes or no), hypertension (yes or no), heart disease (yes or no), stroke (yes or no), diabetes (yes or no), and memory disease (yes or no). Health behaviors included smoking (yes or no) and alcohol consumption (drinking more than once a month, drinking but less than once a month or none of these).
We fit multilevel growth modelling with an unstructured covariance matrix and random intercept and random slope to examine the association between time-varying activity frequency with changes in global cognition and individual cognition including mental status and memory. In our study, two-level linear growth models were used with time as our level-1 variable and respondents/individuals as the level-2 variable.
In our models, individual’s mental status and memory scores over the 3 waves were modelled as a function of time (years since the baseline including 1, 3, 5 ). The intercept (individual initial mental status or memory score at baseline) and slope (the yearly rate of change in both scores) were specified as random at level-2 (person level). The frequency of activity was the predictor variables, and mental status and memory were two separate outcome variables. Age, gender, residence, highest education, marital status, psychiatric problems, hypertension, heart disease, stroke, diabetes, memory disease, smoking, alcohol, and disabled were adjusted for the final multivariate model. Random forest imputation was applied to fill in the missing data on covariates.
First, we used the unconditional means model to estimate the Intraclass Correlation Coefficient (ICC). Second, we used the unconditional growth model to examine the unadjusted association between specific activities and mental status or memory scores. At last, we used a two-level multilevel growth model to model the change in mental status and memory scores over time for older people by different frequency of activity. In this model, level-1 model addressed how individual changes over time and a level-2 model that addressed how changes in mental status and memory scores differs between individual by frequency of activity. All statistical analysis was performed using R version 4.0.0. P < 0.05 (two-sided) was considered statistically significant.