Data source
The data used for analysis in this study were collected by the China Health and Retirement Longitudinal Study (CHARLS) in 2018. CHARLS aims to set up a high-quality public micro-database that can provide a wide range of information ranging from the socioeconomic status to the health status of residents aged 45 years and over. The CHARLS questionnaire contains basic information and behavioral data on the interviewees and their families, as well as personal health status and healthcare utilization. By using a stratified multi-stage PPS random sampling strategy, a total of 19816 individual were sampled from 450 communities or villages in 150 counties or districts[37]. As a result, the data are highly representative.
The target population of this study is elderly patients with chronic diseases. In accordance with the World Health Organization definition of the elderly, we included 8,640 patients with chronic diseases aged 60 years and above[38]. After excluding the samples with missing relevant variables, such as healthcare utilization and demographic characteristics, the final sample size was 7,243.
Variables and definitions
Dependent variables
Healthcare utilization includes outpatient visit and inpatient care. Therefore, our survey selected two indicators to measure healthcare utilization: frequency of outpatient visits and frequency of inpatient visit. Participants were asked the following: (1) How many times did you visit a doctor for outpatient care during the last month? (2) How many times did you receive inpatient care during the past year? In this study, healthcare utilization was considered as a continuous variable.
Independent variables
Our study divided independent variables into health need factors and socioeconomic factors. The proxies of the health need factors were age, gender and self-rated health status (SRH). Age was categorized into three levels: 60−69, 70−79 and 80+. Gender included male and female. SRH was divided into five levels: very good, good, fair, poor and very poor. Socioeconomic factors were represented by education, marital status, number of children, residence, insurance coverage and per capita household outcome. The first four independent variables were defined as binary variables, and the others were defined as multi-category variables. Table 1 shows the detailed coding of each variable.
Table 1
Definitions of independent variables
Variable
|
Codes
|
Health need factors
|
|
Age
|
1 = 60−69, 2 = 70−79, 3 = ≥80
|
Gender
|
1 = male, 2 = female
|
Self-rated health status
|
1 = very good, 5 = very poor, from 1 to 5, with 1 being very good to 6 being very poor
|
Socioeconomic factors
|
|
Education
|
1 = below lower secondary education
2 = upper secondary and vocational training or above
|
Marital status
|
1 = partnered,2 = single
|
Number of children
|
1 = ≤2, 2 = ≥3
|
Residence
|
1 = rural, 2 = urban
|
Insurance coverage
|
1 = basic health insurance
2 = other health insurance
3 = non−coverage
|
Per capita household outcome
|
1 = lowest, 5 = highest
|
Given that our study focused on income-related healthcare utilization inequity in the elderly with chronic disease, economic status is an important independent variable. In accordance with recent research[39], we chose to use per capita household consumption to quantify economic status. The data on the per capita household consumption were directly obtained from the Harmonised CHARLS database, which is funded by the National Institute on Ageing (R01 AG030153, RC2 AG036619, 1R03AG043052) and is a user-friendly version of a subset of the CHARLS interviews, to increase the accessibility of the data to researchers and to facilitate comparisons[40]. We divided the per capita household consumption into five levels on the basis of their ranking, with 1 being the lowest and 5 being the highest.
Statistical method
Firstly, we used LCA on all samples to identify the potential classes of the elderly with chronic diseases. We considered the Akaike information criterion (AIC), Bayesian information criterion (BIC), adjusted Bayesian information criterion (aBIC) and entropy index to determine the number of latent classes. If the percentage of latent class was less than 10%, it could represent chance findings and be a false indication for the number of latent classes[41]. Then, the differences, including differences in dependent variables and independent variables, between categories were examined through analysis of variance and \({x}^{2}\) test.
Secondly, the CI was calculated to quantify the total inequities in healthcare utilization[42]. The CI ranged from − 1 to 1. The CI equal to zero is indicative that there is no inequity; the positive CI is indicative of the disproportionate concentration of healthcare utilization in rich individuals and the negative CI is indicative of the disproportionate concentration of healthcare utilization in poor individuals[43]. The formula for CI can be written as:
$$C = \frac{2}{\mu } {cov}_{w}({y}_{i} , {r}_{i})$$
where \(C\) is the CI;\(\mu\) is the (weighted) mean healthcare utilization; \({cov}_{w}\) is the weighted covariance and \({y}_{i}\) and \({r}_{i}\) are the healthcare utilization and the scoring rank of the per capita household outcome of an individual \(i\)(\(i\)=1 for the poorest and \(i\)=N for the richest), respectively[44].
Thirdly, the CI is further decomposed into the contributions of health need factors and the socioeconomic factors[12,44]. Multiple linear regression was used in this study, given that the frequency of frequency of healthcare utilization, the dependent variable, is a continuous variable:
$${y}_{i} = \alpha +{\sum }_{v}{\beta }_{v}{x}_{vi}+{\sum }_{j}{\beta }_{j}{x}_{ji}+{\epsilon }_{i}$$
where \({y}_{i}\) is the frequency of healthcare utilization; \({x}_{vi}\) represents the health need factors; \({x}_{ji}\) represents the socioeconomic factors; \({\beta }_{v}\) and \({\beta }_{j}\) are the marginal effects of each variable and \({\epsilon }_{i}\) is the error term.
The decomposition of the total CI can be written as:
$$C = \alpha +{\sum }_{v}\frac{{\beta }_{i}{\stackrel{-}{x}}_{i}}{\mu }{C}_{y}+{\sum }_{j}\frac{{\beta }_{j}{\stackrel{-}{y}}_{j}}{\mu }{C}_{j}+\frac{G{C}_{\epsilon }}{\mu }$$
where \(C\) denotes the total CI; \({\sum }_{v}\frac{{\beta }_{i}{\stackrel{-}{x}}_{i}}{\mu }{C}_{y}\) is the contribution of the health need factors, \({\sum }_{j}\frac{{\beta }_{j}{\stackrel{-}{y}}_{j}}{\mu }{C}_{j}\) is the contribution of the socioeconomic factors and \(\frac{G{C}_{\epsilon }}{\mu }\) is the contribution of \(\epsilon\). In this study, health need factors include age, gender and SRH, and socioeconomic factors include education, marital status, number of children, residence, insurance coverage and per capita household outcome.
Fourthly, HI indicates the inequity of healthcare utilization in individuals with equal health demands. Wagstaff proposed calculating CI as CI minus the contribution of health need factors[44] with the formula:
$$HI = C-{\sum }_{v}\left(\frac{{\beta }_{i}{\stackrel{-}{y}}_{i}}{\mu }\right){C}_{y}$$
The range of HI is (− 1, 1), and the meaning of the value is similar to the that of CI. A value close to zero is indicative of low inequity. A value within (− 1, 0) indicates pro-poor inequity, and that within (0, 1) indicates pro-rich inequity.
We used MPLUS8.0 to perform LCA. Other analyses were performed with Stata14.0 with statistical significance at p < 0.05.
Ethics statement
Data collection from CHARLS received ethical approval from the Institutional Review Board (IRB) of Peking University. The IRB approval number for the main household survey, including anthropometrics, is IRB00001052-11015. Respondents were asked to sign two copies of the informed consent in the survey.