Data sources
Data from the China Health and Retirement Longitudinal Study (CHARLS). The CHARLS collected high-quality data that represent elderly individuals over 45 years old and their families in China. These samples are followed up every two years. We excluded observations with missing values: n = 15457 in wave 1 (2011), 16576 in wave 2 (2013), and 16486 in wave 3 (2015).
Variables
Health outcomes
Each new survey respondent was queried, “Have you ever been diagnosed with diabetes by a doctor?” In follow-up interviews, participants were asked, “Our records from your last interview show that you have had/not had diabetes, is this right?” and “have you been diagnosed with diabetes by a doctor since your last interview in the last 2 years?” Participants recorded yes or no responses to all questions. Respondents who answered “yes” to any questions were required to provide medical or hospital records. People who answered“yes” were classified as having diabetes.
Other variables
Ten categories of factors, which may be related to the prevalence of diabetes, were used in this study. Demographic characteristics, such as age, gender, marital status, educational level, income, urban/rural location, region, body mass index (BMI), smoking and drinking.
Socioeconomic status
To measure inequalities in the prevalence of chronic disease among people with different standards of living, data on household assets and housing characteristics were used to construct a proxy index to measure living standards [16]. In this study, durable consumer goods (containing owning an automobile, electric bike, motorcycle, fridge, washing appliance, television, computer, sound system, video camera, camera, air conditioner, mobile phone, furniture, musical instrument, valuable decorative items, jewelry, collectibles, precious metals, or art works), and housing characteristics (including the type of structure of residence, a one-story or multilevel building, and having a toilet, electricity, running water, bathroom facilities, coal gas or natural gas, heating, a source of cooking fuel, a telephone, and an internet connection) were combined into an index of SES to measure household living standards.
Principal component analysis (PCA) was used to measure the socioeconomic status of households. PCA is a standard factor analysis method used to describe variation in a set of variables as linear combinations of the original variables, in which each continuous linear combination is derived, to explain variation in the original data as much as possible, while being uncorrelated with other linear combinations. To perform PCA on the variables related to SES, qualitative categorical variables were re-coded as binary variables. Then, all the variables and other continuous variables were entered into the model. SES was classified by weighting the first factor of PCA [17]. In the case of PCA, the wealth index Ai for individual i is defined as follows:
[Please see the supplementary files section to view the equation.]
Statistical analysis
The concentration index was used to quantify revenue-related disparity in the self-reported incidence of diabetes. The scope of concentration index was -1 to 1, where 0 represents no income-related inequality. A positive concentration index means that health inequality is more pronounced among rich people; a negative concentration index means that health inequality is more pronounced among poor people [18]. The formula for calculating the concentration index is as follows:
[Please see the supplementary files section to view the equation.]
where CI represents the concentration index, y is whether an individual has diabetes, μ represents the mean of the prevalence of diabetes, and r represents the fractional rank of income distribution.
- Decomposition of the concentration index
The concentration index was decomposed to determine the selected contributors to inequality in diabetes. In this study, we defined as the socioeconomic factor related to diabetes prevalence. Thus, the linear regression analysis model of diabetes prevalence and related factors is as follows [19]:
[Please see the supplementary files section to view the equation.]
where Y denotes the prevalence of self-reported diabetes; the are related to socioeconomic factors, and is the error term. A generalized linear model (GLM) with a binomial distribution and an identity link, which was used to calculate the element associated to the prevalence of diabetes, as Y is a binary variable.
The concentration index may consist of contributions of individual factors to diabetes prevalence inequality. And each contribution is the product of the sensitivity of diabetes prevalence related to the factor and the degree of inequality in that factor. The concentration index decomposition was calculated as follows [20]:
[Please see the supplementary files section to view the equation.]
All data preparation and analyses were performed in SAS version 9 (SAS Institute Inc., Cary, NC, USA). The concentration index and the 95% confidence interval were calculated using the bootstrap method. Furthermore, concentration curves in Figure 1 were obtained using Stata 12.0.