The China Health and Retirement Longitudinal Study (CHARLS) is a nationally representative study of Chinese adults aged ≥ 45 years. The CHARLS is designed to describe the dynamics of retirement and its impact on health, health insurance, and economic well-being. The baseline survey was conducted in 2011-12 among 17,708 participants from 150 counties of China’s 28 provinces , and data on socioeconomic status, lifestyles, medications, health status and functioning assessments were collected. Details on the study design, sampling procedure, and data collection have been described in previous publications . Briefly, the CHARLS participants were recruited through a four-stage, stratified, cluster random sampling method. The CHARLS participants were followed biennially to obtain updated information. The CHARLS data are available for the baseline survey in 2011-2012, the first follow-up survey in 2013-2014, and the second follow-up survey in 2015-2016. The Biomedical Ethics Committee of Peking University approved this study, and all participants provided written informed consent.
The current analyses focused on individuals aged 50 years and older. As shown in supplementary Figure S1, a total of 7,011 participants lacked measurements of cognitive function (n=2,655) or health status and functioning (n=112), were younger than 50 (n=4,019), or had dementia or Parkinson’s disease (n=225) at baseline. Those participants were excluded from the current analyses. In addition, a further 3,454 participants were excluded due to loss to follow-up (n=854), missing data on cognitive function during follow-up (n=1,642) or incident dementia or Parkinson’s disease (n=129). Therefore, the current analyses were carried out among 7,243 participants aged ≥50 years who had complete cognitive tests assessed at baseline, the first follow-up, and the second follow-up survey.
Social and intellectual activities
The frequencies of leisure time activities in the past month were measured through self-reports. Leisure time activities included seven items and were classified as either social activities (interacting with friends; going dancing, exercising, or practicing Qigong; participating in community-related organizations; and doing voluntary charity work or assisting others) or intellectual activities (playing Mahjong, cards or chess; attending an educational or training course; investing in stock or surfing the internet) [10, 11]. The frequency of each activity was measured as “never”, “not regularly”, “almost every week” or “almost daily”. We coded each item as 1 = “almost every week” or “almost daily” and 0 = “never” or “not regularly”. The composite scores for social activities (ranging from 0 to 4) and intellectual activities (ranging from 0 to 3) were calculated as the sum of their corresponding activity scores.
In accordance with previous studies using data from the CHALRS [19, 20], cognitive function, including episodic memory, orientation/attention and visuospatial skills, was assessed at baseline and during the two follow-up visits by three tests: word recall, the Telephone Interview of Cognitive Status (TICS), and figure drawing.
The word recall test evaluated episodic memory. Examiners read a list of 10 random words, and participants were instructed to recall as many words as possible immediately afterward (immediate recall). The number of correctly recalled words was scored and indicated the participant’s immediate recall. Ten minutes later, the participants were asked to recall the same list of words (delayed recall). Episodic memory scores were calculated as the average number of immediate and delayed word recalls and ranged from 0 to 10.
The original TICS measures an individual’s mental status. In the CHARLS, ten questions from the TICS were used, including questions on serial subtraction of 7 from 100 (up to five times), the date (month, day, and year), the day of the week, the season of the year, and figure drawing. The questions primarily measure orientation/attention; the total score was calculated as the sum of correct answers to the ten questions and ranged from 0 to 10.
The figure drawing test assessed visuospatial abilities. Participants were shown a picture of two overlapping pentagons and were asked to draw a similar picture. Participants who drew correctly received a score of 1, and those who did not correctly draw the picture received a score of 0.
Overall cognitive function was calculated as the sum of the scores of the three tests and ranged from 0 to 21. To make the cognitive function scores compatible across the surveys, the scores for the first and second follow-up surveys were standardized to the baseline scores using the mean and standard deviation (SD) of the cognitive scores at baseline. Specifically, the cognitive function scores from the follow-up surveys were transformed to z scores by subtracting the mean score at baseline and dividing by the SD at baseline. In addition, a composite global cognitive z score was calculated for each participant by averaging the z scores of the three tests and re-standardizing them to baseline z scores using the mean and SD of the baseline global cognitive z score.
Baseline measurements of age, sex, education level, marital status, location of residence, household income level, smoking, drinking, self-perceived health status, physician-diagnosed chronic diseases, disability, self-reported visual and hearing impairments, depressive symptoms, and body mass index (BMI) were included as covariates in the current analyses. Educational level was categorized as “no formal education”, “primary school”, “middle school”, or “high school or above”. Marital status included “married” and “others”. Location of residence was divided into “rural” and “urban”. Household income was categorized into tertiles and coded as “low”, “medium”, and “high”. Self-perceived health status was reported as “good”, “fair” or “poor”. Physician-diagnosed chronic diseases included hypertension, diabetes mellitus, dyslipidemia, heart diseases, stroke, lung disease, arthritis, and cancer. Disability was defined as having limitations in any of the five activities of daily living, including bathing, dressing, eating, getting into/out of bed, and toileting . Depressive symptoms were assessed using the 10-item version of the Center for Epidemiologic Studies Depression Scale, and a score of ≥ 10 indicated the presence of depressive symptoms . BMI was calculated as the weight in kilograms divided by the square of the height in meters. BMI was categorized as follows: <18.5, 18.5-23.9, 24.0-27.9, and ≥ 28.0 kg/m2 .
Group-based trajectory models (GBTMs) implemented through the “traj” plugin procedure in Stata were used to identify the trajectories of the global cognitive z scores during the 4 years of follow-up  based on the assumption of a censored normal distribution . A maximum of five trajectory groups was set a priori. We fitted the models from one group trajectory to five group trajectories, and the follow-up time was used as a time scale. The Bayesian information criteria (BIC) and Akaike’s information criterion (AIC) were used to identify the best fitted model. Furthermore, an average posterior probability of assigning each participant to a group of approximately 70% or higher was indicative of a good ﬁt, and models with greater than 5% membership in each trajectory group were selected. As shown in supplementary Table S1, the AIC and BIC were similar across models with 3, 4, and 5 trajectory groups. To be comparable with findings from the Whitehall II cohort study , in which three trajectory groups were identified for cognitive function, we decided to choose the three-group trajectory model. Then, we compared the model fit with different forms (i.e., linear, quadratic, and cubic). Because the CHALRS had 3 waves of data collection, the models were tested for linear and quadratic trends for each trajectory group. In the final model, the overall global cognitive z scores had 3 trajectory groups, which were in linear, quadratic, and quadratic forms. The mean probabilities of final group membership were above 83% for all cognitive function measures. The final three trajectory groups had low, intermediate, and high cognitive function. We repeated the above analysis for the individual cognitive domains, including episodic memory, orientation/attention, and figure drawing.
The baseline characteristics of the study participants were described according to the three trajectory groups of overall cognitive function. Continuous variables were summarized as the mean (± SD), and categorical variables were expressed as frequencies and percentages. Continuous variables were checked for normality and log-transformed if necessary. Differences between groups were compared using one-way analysis of variance (ANOVA) for continuous variables and the χ2 test for categorical variables.
A multinomial logistic regression model was used to estimate the association of social and intellectual activities with the trajectories of the cognitive function measures, including overall cognitive function, episodic memory, and orientation/attention. The high trajectory was set as the reference group for each cognitive outcome. Odds ratios (ORs) and the corresponding 95% confidence intervals (CIs) associated with a 1-unit increase in social and intellectual activities scores were reported. We built five sets of models for each cognitive outcome. Model 1 adjusted for age and sex. Model 2 additionally adjusted for socioeconomic status variables, including education, marital status, location of residence, and household income. Model 3 adjusted for all variables in model 2 plus health behaviors, including smoking, drinking, and BMI. Model 4 additionally adjusted for health conditions, including self-perceived health status, depressive symptoms, disability, visual and hearing impairment, hypertension, diabetes mellitus, dyslipidemia, heart disease, stroke, lung disease, arthritis, and cancer. Model 5 was the fully adjusted model, in which both social and intellectual activities were added.
We also investigate the joint associations of social and intellectual activities with cognitive trajectories by adding an interaction term between social and intellectual activities in model 5. To make the interpretation easier, we recoded both social and intellectual activity variables, with 1 = more than 1 activity, which was labeled “frequent”, and 0 = less than 1 activity, which was labeled “rare”. Then, we created a combined variable with four categories: rare intellectual and rare social activities, rare intellectual activities but frequent social activities, frequent intellectual activities but rare social activities, and frequent intellectual and frequent social activities. For each cognitive trajectory, the model was adjusted for socioeconomic status variables, health behaviors, and health conditions.
All analyses were performed with Stata version 15.1 (StataCorp, College Station, TX). A two-sided p-value less than 0.05 was considered statistically significant.