Data collection
The data source for this study was the WHO Global AGEing and Adult Health (SAGE), a longitudinal study of health and ageing in China, Ghana, India, Mexico, Russia and South Africa. The aim of SAGE was to provide reliable data on aging people and their health status in low- and middle-income countries (20). The present study used data from Wave 1 (2007-2010) in the China. In the SAGE study, the elderly population was defined as aged 50 years and above. This age cut-off was based on the WHO definition of younger and older adults in Low- and middle-income countries (20). Data were collected from Guangdong, Hubei, Jilin, Shanxi, Zhejiang, Shanghai, Shandong and Yunnan with considering geographical and economic effects. The sample design followed a stratified multistage cluster principle. One district (urban) and one county (rural) was randomly selected from each province, and from each district/county, 4 communities/townships were picked out probability proportional to size. Similarly, 2 residential blocks/villages were selected probability proportional to size from each community/township. Afterwards, 84 households were randomly selected in each picked residential block/village: 70 were over 50 years households and 14 were between 18 to 49 years households(20). All the eligible households were included in conducting household interviews, and individual interviews were from the eligible respondents in the household roster. All these questionnaires were piloted as part of the pretest in 2005, then translated into local language from English followed those developed by the World Health Survey(WHS). A strict translation protocol designed by a team of bilingual experts, focused back-translations, and in-depth linguistic analysis had been used to ensure the cultural issues(21). At last, 10278 household surveys and 15050 individual surveys had been collected. Post-stratification weights were used in both household and individual level analysis to adjust for province, locality, age and sex distributions.
In this study, 237 people were excluded for not completing the survey, 1265 people were then excluded for analysis because their ages were under 50 years. Missing check was applied to select the respondents who answered all included questions(22). After the missing check, 10099 who completed the questionnaires were derived from 15050 of Chinese sample for future analysis (figure 1).
Variables selection
Variables selection from the individual questionnaire
The results were only for the people who had completed interviews (interview were conducted including body measurement, performance tests and blood sample). NCDs selection was based on the self-reported responses to presence of one or more following diseases: arthritis, stroke, angina, diabetes, chronic lung disease, asthma, cataract, and hypertension. Hypertension was identified as systolic blood pressure was ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg.
Social-demographic variables included age, sex, marriage status and educational level. People above 50 years were selected. Risk factors and health behavior variables contained body mass index (BMI), tobacco using, alcohol consumption, fruit/ vegetables intake as well as physical activity variables. BMI was derived from weight in kilogram (Kg) divided height in meter (m). China’s center for disease control guidelines set BMI<18.5 (kg/m2) to underweight, 18.5-23.9(kg/m2) to normal, 24-27.9(kg/m2) to overweight and >28(kg/m2) to obesity. The fruit/vegetables variable was defined as adequate with intake of fruit and vegetables>=5 servings daily(20). Physical activity was measured by using WHO Global Physical Questionnaires(23).
Variables selection from the household questionnaire
Household asset items were chosen for deriving wealth quintiles. Under household expenditure module, variables like general tax, mandatory health insurance, voluntary health insurance as well as out of pocket (OOP) were picked up. General taxes were consisted of property tax, vehicle tax, income tax and non-health related insurance. Public Health insurance would be asked about the premiums paid for health plans. In this SAGE survey, designers summed up UEBMI, URBMI and NCMS as mandatory health insurance, without classifying it into detailed groups. Out-of-pocket payments were made up of hospital care, drug purchases, diagnostic tests, dental care, ambulance fees and other costs with excluding the insurance reimbursement. The time frame for these four payments was set into 12 month.
The WHO SAGE study was approved by the Ethics Review Committee, World Health Organization, Geneva, Switzerland; the Ethics Committee, Shanghai Municipal center for Disease control and Prevention, Shanghai, China. All procedures were approval covered in this study. All participants got informed consent forms before started.
Data analysis
Wealth status was derived from household assets and Principal Component Analysis (PCA) was used to produce the wealth quintile(24). Factor test was a measurement to look at the adequacy of sampling, with using Bartlett’s Test and KaiserMeyer-Olkin(KMO)(24)(25). At last, component scores were sort into five categories to get the wealth quintile and ranked into poorest to riches with the value of KMO = 0.871, p-value=0.000, and the value of correlation matrix = 0.062. Chi-square test was applied for association analysis between socioeconomic risk factors and different health status. Partial proportional odds model was employed for comparing the contributions of each factors to different chronic conditions(26).
Progressivity analysis was applied for examining the vertical equity on healthcare payments(27). Measuring progressivity was to compare shares of general tax, mandatory health insurance, voluntary health insurance as well as out of pocket payment(OOP) contributed by cumulative wealth of households ranked by the share of ability to pay(ATP)(27). The fairness of healthcare expenditure was analyzed by using Kakwani index, which was one of the most widely used in measuring inequity in healthcare payment field(18). Kakwani index was formulated as the difference between concentration index for healthcare payments and Gini coefficient(18). Concentration index was known as the measurement of socioeconomic-related health inequality, and Gini coefficient was an indicator for measuring the degree of wealth inequity in a country(28). If health finance was in a vertical equity situation, poorer people would contribute less for the share of health payments than their share of ATP, which was also called progressive system. Whereas it was a regressive system if sharing of health payments were greater than ATP sharing(17)(27). As recommended by the World Bank, household assets were measured to reflect the ATP since assets indices could reduce the currency fluctuation comparing to income or expenditure, which have been widely used in developing countries(12)(27).
Ordinary least squares(OLS) regression was employed in estimation of CI and Gini coefficient in this study. It compared healthcare payment variables with household fractional ranking according to ATP distribution(27)(28). (See Equation 1 in the Supplementary Files)
where was a health payment variable for household i and was the mean. was the household fractional ranking according to ATP distribution was the sample variance of the fractional rank. in OLS was the estimation for CI or Gini coefficient.
The Kakwani index for healthcare payment was defined as the difference between concentration index for healthcare payments and Gini coefficient(18)(27).(See Equation 2 in the Supplementary Files)
where C was the concentration index (CI) for healthcare payments, ranging from -1 to 1. G referred to the Gini coefficient, with a range of 0 to 1. Values increased as inequality increased. The value of πk was from -2 to 1. The positive value indicated a progressive system in healthcare spending, to the graph, the concentration curve lied outside of the Lorenz curve, which meant the richer households paid more according to the ATP than poor households. While a negative value indicated a regressive spending, as the concentration curve lied inside of the Lorenz curve(18)(27).
As KI was the difference between CI and Gini coefficient, KI value could be computed by the following OLS regression(17)(27): (See Equation 3 in the Supplementary Files)
where was a health payment variable for household i and is the mean, was the ATP variable and was its mean. was the household ranking in ATP distribution and was a sample variance of fractional ranking. In this formula, was estimated for KI value.
Sampling weights were included for all the analysis. Statistical software STATA 14.0 was applied in this study.