Sample and data
Data were from an ongoing longitudinal study of the China Health and Retirement Longitudinal Study (CHARLS). The CHARLS collected a national representative cohort of Chinese residents aged 45 and older to gather a wide range of personal health information and social and economic data for geriatric and health policy research. Approximately 150 areas (counties) within 28 provinces in China were first selected proportional to population size, and 3 villages/communities were then chosen from each county as primary sample units (PSUs). Eighty households were chosen at random from each of the 450 PSUs, with 24 being investigated. If there were persons 45 years or older in the household, one of them was chosen as a respondent at random, and both the respondents and their spouses were assessed using structured questionnaires. The data currently include 4 waves, including baseline (wave 1: 2011) and follow-ups every 2 years at wave 2 (2013), wave 3 (2015), and wave 4 (2018). Eventually, a total of 17,708 participants from 10,257 households were recruited at baseline. A detailed description of the cohort information and methods has been reported elsewhere [14].
Of 17,708 participants at baseline, 12,683 participants were excluded because of (1) age less than 45 years old; (2) lack of data on PA intensity; (3) diagnosis of Parkinson’s disease, Alzheimer’s disease, brain atrophy, or stroke attachment; and (4) lack of cognitive and depressive symptom assessment data. Eventually, a total of 5,025 participants were eligible for the current analysis. A total of 3,617, 3,606, and 2,672 of the respondents provided cognitive function data during follow-up in wave 2, wave 3, and wave 4, respectively, while 4,198, 4,125, and 3,764 provided depressive symptom data, respectively. The flow chart of the current study is shown in Figure 1.
Measures
Physical activity intensity
PA intensity was categorized according to three levels, including vigorous PA, moderate PA, and light PA [15]. Vigorous PA was defined as activities that made individuals breathe much harder than normal, including digging, aerobics, heavy lifting, ploughing, quick cycling, and cycling with a heavy load. Moderate PA was defined as activities that made individuals breathe faster than normal, including cycling at a normal speed, carrying light items, practicing tai-chi, and fast walking. Light PA referred to walking for leisure or recreation, walking at work, and travelling between places. The respondents were asked whether they performed vigorous PA, moderate PA, or light PA for at least 10 minutes every week and how many days a week they took part in the corresponding type of PA. In this study, the frequency of vigorous PA, moderate PA, and light PA with a range of 0-7 days/week was included in the analyses.
Cognitive function
Following a previous study [16], cognitive function was investigated via the Telephone Interview of Cognitive Status (TICS, measuring mental intactness), with a score range of 0-9; a word recall test (assessing episodic memory), with a score range of 0-20; and a figure drawing test (assessing visuospatial abilities), with a score range of 0-1. The TICS included a serial 7s test (range 0-5) and date naming (range 0-4). In the serial 7s test, the participants were asked to subtract 7 from 100 for up to five consecutive times. In the date naming test, the participants were asked to correctly report the date (day, month, year) and day of the week. In the word recall test, the participants were asked to recall 10 different words both immediately and 4-10 min later. In the figure drawing test, the participants were asked to draw overlapping pentagons. Cognitive function was assessed based on the sum score of these tests, and the total score ranged from 0 to 30. A higher total score indicated better cognitive function.
Depressive symptoms
Depressive symptoms were assessed by the Chinese version of the 10-item Epidemiologic Studies Depression Scale (CESD), which has been validated and widely used in Chinese older adults [17]. The participants were asked, “How often did you feel this last week?” They were asked to rate this question on a 4-point scale from 1 (rarely or one) to 4 (most or all of the time, 5-7 days). The total score ranged from 0 to 40, and a higher score suggested a higher level of depressive symptoms.
Covariates
The covariates included baseline age, sex, educational attainment, marital status, current residence location, number of chronic diseases, and activities of daily living (ADLs). Educational attainment was categorised as “no formal education”, “elementary school and middle school”, and “high school and higher”. Marital status was grouped as “married/partnered” and “unmarried”. The current resident locations included “rural” and “urban”. The number of chronic diseases was used to assess self-reported chronic diseases that had been diagnosed (such as diabetes, cancer, heart attack, chronic lung diseases, and liver disease). ADLs were assessed to determine if the participants had difficulties in bathing, eating, dressing, getting in and out of bed, using the toilet, and controlling urine, with a score range of 0-6.
Statistical analysis
The demographic characteristics included the mean (standard deviation, SD), median (interquartile range, IQR), and number (percentage). To study the impact of PA intensity on trajectories of cognitive functioning and depressive symptoms across the four waves, multilevel mixed-effects generalised linear models (GLMs) with a Gaussian distribution and an identity link function were used. The repeated measurements of data nested within individuals were taken into consideration by these multilevel models. To determine if trajectories differed by level of different PA intensity (vigorous PA, moderate PA, and light PA), we employed interaction terms (i.e., vigorous PA * wave, moderate * wave, and light PA * wave). If a significant effect of the interaction term was to be determined, we then tested marginal effects and plotted the trajectories of cognitive functioning and depressive symptoms by this interaction term. Covariates reported at baseline were adjusted in all models. In the sensitivity analysis, we performed multiple imputations by chained equations (MICE) to handle missing data [18]. A total of 10 imputed data sets were obtained, and the results were pooled by Rubin’s rule. Similar results were found (Supplementary Table S1). Analyses were performed using R version 4.0.4.