Impact of physical activity on the depression and self-care ability among Chinese older adults during the COVID-19 pandemic: propensity score matching analysis | BMC Geriatrics

Study design and sample selection

This study employed a cross-sectional design using data from the China Longitudinal Aging Social Survey (CLASS) conducted in 2021. The CLASS survey is designed to collect social, economic, family, and health information from Chinese older adults aged 60 years and above. It aims to explore the challenges older adults face in the aging period and to evaluate their personal growth history, family history, occupational history, health, and physical activity. The survey meets the requirements of multiple cohort studies from a life course perspective. It employs a longitudinal design that follows participants over time, collecting data regularly to observe changes and developments in their health status. This approach allows for the examination of the long-term effects of physical activity on depression and self-care ability among older adults [33, 34]. We utilized propensity score matching (PSM) to analyze the effects of different intensities of physical activity on these outcomes, which helped to reduce potential selection bias and confounding factors inherent in observational studies. In this study, three methods were used for sample matching, including nearest neighbor matching with caliper, radius matching, and kernel matching, to ensure the robustness of the estimation results. The estimation was considered robust when the results obtained from different matching methods were consistent. The data were retrieved in January 2022, with an initial sample size of 11,398 participants. After data cleaning, which involved the elimination of participants with incomplete or irrelevant data, a total of 8,483 valid samples were used for the analysis. The stratified multi-stage probability sampling method selected county-level areas as the primary sampling units (PSUs) and villages/residential councils as the secondary sampling units (SSUs). The final sample included participants from 476 villages/residential committees in 30 provinces/autonomous regions or directly administered cities in China.

Study participants

The participants recruited in the study were Chinese citizens aged 60 and above (without an upper age limit) living in the study area. The interviewers read the the questionnaire items and recorded responses from the participants. The survey comprised 476 villages/residents’ committees in 30 provinces/autonomous regions or directly administered cities. The Capital University of Physical Education and Sport Ethics Committee approved the study protocol (approval number: ChiCTR-IOR-ChiCTR2200063177).

Table 1 summarizes the descriptive statistics of the study’s dependent, independent, and control variables. The dependent variables (DV) include the depression score (DS), instrumental activities of daily living (IADLs), and activities of daily living (ADLs). These are presented as continuous variables, with mean ± standard deviation reported. Independent variables, such as physical activity levels—vigorous (VPA), moderate (MPA), and light (LPA)—are presented as categorical variables using percentages. Control variables (CV), such as gender, marital status, and education level, are also categorical. Continuous control variables include family numbers (FN, representing the count of family members), sedentary time per week (SEDT, in minutes), daily sleep time (SLPT, in minutes), and age (in years), and these are reported using mean ± standard deviation. Satisfaction variables, including self-rated life satisfaction (SWL), self-rated health (SRH), and peers’ relative health (PRH), are presented as categorical distributions, with responses distributed across levels of satisfaction. Continuous control variables, such as FN, SEDT, and SLPT, are also used as match variables (MV) in the propensity score matching analysis.

Physical activity

For the propensity score matching (PSM) analysis, participants were divided into treatment and control groups based on their physical activity levels as assessed by the International Physical Activity Questionnaire Short Form (IPAQ-SF). The treatment group included participants who engaged in one of three intensity levels of physical activity, namely vigorous, moderate, and light, according to the International Physical Activity Questionnaire Short Form (IPAQ-SF). Vigorous physical activity was defined as activities that cause a significant increase in breathing or heart rate, such as running or heavy lifting. Moderate physical activity included activities that cause a moderate increase in breathing or heart rate, like brisk walking or gardening. Light physical activity refers to activities that cause a slight increase in breathing, such as casual walking or light household chores. The control group comprised participants reporting no engagement in any physical activity during the same period. To ensure comparability between groups, participants were matched based on demographic characteristics (age, gender, marital status, education level) and baseline health indicators (sedentary time and sleep duration). The IPAQ-SF, whose reliability and validity have been demonstrated in the Chinese population [35,36,37], has shown high reliability with intraclass correlation coefficients (ICC) for test-retest reliability ranging from 0.812 to 0.999 and validity correlations between 0.608 and 1.000 for vigorous activity and 0.916 for walking activities [38, 39]. This tool was used to evaluate the physical activity (PA) of older adult participants over the past seven days. Out of the 8,483 valid samples, 456 participants (5.38%) engaged in vigorous physical activity (VPA), 1,385 participants (16.33%) engaged in moderate physical activity (MPA), and 6,327 participants (74.58%) engaged in light physical activity (LPA). These proportions demonstrate appropriate sample sizes across physical activity intensity levels for conducting PSM analysis, which enables systematic comparison of outcomes between matched groups.

Depression score

DS was determined using the DS Scale (CES-D9) to evaluate the DS levels of participants [40,41,42]. The scale has three items that represent positive mood (in a good mood, having a good time, and having lots of fun), two items that indicate negative mood (lonely and sad), two items that indicate emotional marginalization (not being useful anymore and having nothing to do), and two items that represent somatic symptoms (not wanting to eat and not sleeping well). Participants who reported any of the DS items in the past week were required to indicate the frequency of the mood on a scale of 0 (none), 1 (sometimes), or 2 (often). The total score indicated the level of DS, with DS ranging from 0 to 18, whereby higher scores indicated higher DS levels. Previous studies have demonstrated good psychometric properties of the CES-D9 scale among Chinese older adult populations, with internal consistency reliability (Cronbach’s α) ranging from 0.729 to 0.88 [39, 43,44,45,46]. Additionally, the scale exhibits a sensitivity of 0.85 and a specificity of 0.83 [46, 47]. In this study, the Cronbach’s α coefficient of the scale was 0.853, indicating good internal consistency.

Self-care ability of daily living

In this study, Activities of Daily Living (ADLs) and Instrumental Activities of Daily Living (IADLs) scales were utilized to evaluate self-care ability, which indicates older adults’ physical health and function [33].

Activities of Daily Living (ADLs) are basic self-care tasks necessary for fundamental functioning, reflecting the essential self-care ability of older adults. The ADLs scale comprises six items, including eating, dressing, transferring, toileting, bathing, and managing incontinence [48, 49]. Each item is rated on a three-point ordinal scale: “do not need others’ help” (score 1), “need some help” (score 2), and “cannot do anything at all” (score 3). The total ADL score is calculated by summing the item scores, with higher scores indicating greater dependency.

Instrumental Activities of Daily Living (IADLs) assess the ability of older adults to live independently within the community, involving more complex activities than ADLs. The IADLs scale comprises nine items, using the telephone, shopping, cooking, managing finances, taking public transportation, doing housework, lifting weights, walking up and down stairs, and preventing falls [49, 50]. Each item is rated on a three-point ordinal scale: “no help required” (score 1), “some help required” (score 2), and “not able at all” (score 3). The total IADL score is calculated by summing the item scores, with higher scores indicating greater dependency.

Both the ADL and IADL scales have been validated and shown to be reliable for use among Chinese older adult populations [51, 52]. In this study, the Cronbach’s α coefficient for the ADL scale was 0.892, and for the IADL scale was 0.891, indicating high internal consistency.

Control variables and matching variables

In the present study, four demographic variables, namely gender, age, marriage status, and education level, were selected as control variables. Average daily sedentary time, average daily sleep time, self-rated life satisfaction, self-rated health satisfaction, and peers’ comparative health satisfaction were used as matching variables in the propensity score matching (Fig. 1).

Frequency distribution of study variables

Statistical analyses

Six items in ADL scores and nine items in IADL scores are categorical variables. However, ADLs and IADLs total scores were treated as approximate continuous variables for statistical analysis, which is a common practice in similar studies [36]. The three categories of PA at different intensities and DS, ADLs, and IADLs were subjected to t-tests to evaluate the differences between the treatment and control groups. The logit model was used to determine the probability of occurrence of participation in different PA among older adults. PSM was used to calculate the average treatment effect on the treated. The endogeneity problem resulting from the individual selectivity bias of older adult participants was eliminated to ensure the robustness of the model results using a propensity value matching analysis model (PSM), which classified the sample into treatment and control groups.

PSM is based on the principle of transforming a multivariate variable X into a one-dimensional propensity score through a functional relationship (Propensity score) \(\:ps\left({\text{X}}_{i}\right)\). Matching is then performed based on the propensity score. The propensity score is an observable characteristic for \(\:{\text{X}}_{i}=x\) probability of an individual receiving a disposition.

The same propensity scores for individuals in the control group \(\:{\text{D}}_{i}=0\) and the treatment group \(\:{\text{D}}_{i}=\:1\:\)implies that the mean values of the observable features of individuals in the two groups are equal. Calculating the Average Treatment Effect on the Treated (ATT) based on PSM given mostly observable features 𝑥 reduces to calculating the treatment effect given the one-dimensional propensity score 𝑝𝑠(𝑥). The ATT measures the average effect of the treatment among those who received it, compared to matched controls who did not receive the treatment.

In the formula: \(\:{\text{Y}}_{i}\left(1\right)\)and\(\:{Y}_{i}\left(0\right)\:\)are dependent variables representing DS, IADLs, and ADLs in the two groups of participation and non-participation in different PA among the older adult participants, respectively. \(\:{\text{D}}_{i}\) represents the critical independent variable indicating participation or non-participation in PA, \(\:{\text{D}}_{i}=1\) was used for participation and \(\:{\text{D}}_{i}=0\) for non-participation. \(\:{\text{X}}_{i}\) represents a set of characteristic variables determined by whether the participants participate in PA or not; \(\:{\text{Y}}_{i}\left(1\right)\) and \(\:{Y}_{i}\left(0\right)\) represent the three dependent variables of older adult participants in the treatment and control groups, respectively.

PA at different intensities was evaluated for DS, IADLs, and ADLs. The probability of PA participation among older adults was predicted using logit models, and the treatment groups were matched using PSM to estimate the average treatment effect (ATT).

In this study, three methods were used for sample matching, including nearest neighbor matching with the caliper, radius matching, and kernel matching, to ensure the robustness of the estimation results. The estimation was considered robust if the results from the methods were similar. The principle ε ≤ 0.25≤σ̂_pscore was used for the propensity score matching (PSM). The caliper range for both nearest neighbor matching and radius matching was set to 0.001. The psmatch2 package in Stata16 was used to conduct the statistical analysis.

Equilibrium analysis

The balancing hypothesis was tested to verify the quality of matching and the reliability of the estimation results. This involved testing matching variables for balance to determine if there was a significant reduction in individual differences after matching. The standardized deviation was used as the primary indicator for balance determination, with successful matching indicated by standardized deviations controlled within 5% and no significant differences between groups after matching.