Data come from the four follow-up data for older adults in China: the China Health and Retirement Longitudinal Research Study (CHARLS), a longitudinal survey of individuals aged 45 years and older [25]. The national baseline survey was conducted in 2011–2012, with wave 2 in 2013, wave 3 in 2015, and wave 4 in 2018, with multistage sampling and probability proportional to size sampling approaches. To ensure sample representativeness, the CHARLS baseline survey covered 150 countries/districts, 450 villages/urban communities, across the country, involving 17,708 individuals in 10,257 households, reflecting the mid-aged and older Chinese population collectively. The response rate for each wave was over 80%. We constructed panel data using the four waves of the CHARLS survey. Our sample was Chinese community-dwelling older adults aged 60 years and older in 2011, with those who chose to live in nursing homes, died, or were lost to follow-up in 2013–2018 excluded. Finally, 7542, 6386, 5087, and 4052 older adults were included in 2011–2018, respectively (Fig. S1).
Informal care need is assessed using receipt of informal care, informal care sources, and informal care intensity [26]. Receipt of informal care was a dichotomous categorical variable comprised of the categories receiving informal care and not receiving informal care. There were five forms of care sources from which CHARLS respondents with a long-term disabled condition could choose, including care from a spouse, children, parents, other relatives, and the community, which resulted in many caregiving combinations. In this study, informal care sources referred to the number of care sources that older adults received, ranging from 0 to 5, and higher number of care sources indicated that older people need more human resource to provide assistance with ADL. Informal care intensity was measured by the number of care hours received in a week from their children, and more care hours indicated more caregiving time.
Age, education level, marital status, place of residence, economic status, internal and external care resource provision (informal family support network, provision of resources outside the home such as community-based care), and personal health status all influence older adults’ utilization of informal care services [27, 28]. As one of the most classic models for health service utilization research, Andersen’s model provides an excellent summary of inter-individual differences in the health service utilization behavior. This model is applicable to the study of long-term care, life quality of patients with chronic diseases, and health cost [29].
Under the guideline of Andersen’s model, in which an individual’s use of healthcare services and associated outcomes are viewed as a function of predisposing, enabling, and need factors, we selected independent variables. Firstly, sociodemographic traits and social structure were considered predisposing factors since they reflected an older individual’s sociocultural preferences and influence their informal care service utilization behavior. Secondly, enabling factors included the family resources available to an individual adult, which have an effect on the accessibility and availability of informal care services, which in this study referred to annual household income and regular financial support provided by children. Most researchers agree that the higher older adults’ objective purchasing power, the larger their ability to transform potential need into actual need, and the lesser their proclivity for informal care. Thirdly, the need factors were measured by objective health status and subjective health status, affecting the likelihood of seeking and utilizing informal care. The objective health status was subdivided into two components: the prevalence of chronic diseases and the independence in ADL function. The functional disability basic ADLs was divided into four health statues: healthy (0 ADL), mild disability (1–2 ADL), moderate disability (3–4 ADL), and severe disability (5–6 ADL), using the Katz Index where six ADLs were bathing, dressing, grooming, transferring, eating, and toileting [30]. The subjective health status was measured by the self-perceived loneliness and the self-perceived health based on two CHARLS entries: “I feel lonely” and “How do you feel about your health”.
STATA version 15.1 was used for data cleaning and R version 4.1.2 was used for data analyses. Predisposing factors, enabling factors, need factors, receipt of informal care, and the most common combinations of informal care sources and the different degrees of informal care intensity in 2011–2018 were shown in Table 1. Besides, bivariate analyses were conducted to examine group differences (gender and place of residence) across all variables at the baseline (Table S1).
The Generalized Linear Mixed Models (GLMMs, models 1–2) was applied to analyze receipt of informal care among selected participants. The Linear Mixed Models (LMMs, models 3–6) was used to perform informal sources and intensity among selected participants. Both GLMMs and LMMs are designed to analyze longitudinal data due to their merits in adjusting the random effects from repeated measures on the same subject, and the within-subject and between-subject variability. During the modeling, GLMMs and LMMs could capture the effect of both time-invariant factors (e.g., gender) and the time-variant factors (e.g., residence type). The CHARLS survey timepoints (2011, 2013, 2015, 2018) was also included in the data analysis to model change in informal care need. In addition to exploring what factors determine informal care needs, we were interested in the effects of gender, residence, and financial support from children on estimates of informal care needs over time. Therefore, this study evaluated the interaction terms between these three independent variables and the survey timepoint in the GLMMs (model 2) and LMMs (model 4,6), so as to model the effects of the three primary variables on the change in informal care needs among community-dwelling older adults between 2011 and 2018 in China.
The Akaike Information Criterion (AIC) was compared to estimate the model fit, and a lower number indicates a better model fit. In addition, the full model was tested among each sub-group based on gender (female and male), and place of residence (urban and rural) to explore the group-specific associations. For the mixed models, the sampling probabilities varied exogenously by design. In this study, both the weighted and unweighted coefficients were consistent, but the weighted results tended to be less precise (e.g. larger standard errors). Therefore, it only reported the unweighted results in this paper.
