Study design and sample
This study analyses the influence of the 2009 Chinese healthcare reform on CMHS utilisation in Shaanxi Province. Shaanxi Province, in the west of China, was selected as the study area because of its predominantly rural character and a high proportion of poor in the population, the type of region that the policy is intended to target. By the end of 2013, there were roughly 37.60 million population with the per capital Gross Regional Product (GRP) 42,692 Chinese yuan in Shaanxi Province; the birth rate was 10.01% and the natural growth rate was 3.86%, 51.31% of residents were living in urban areas and 48.69% in rural areas [20]. The National Health Service Survey (NHSS) is a population-based cross-sectional nationally representative survey commissioned by the China’s National Health Commission every five years [21-23]. Based on the structure of Chinese administrative districts and the imbalanced population distributions among the different provinces, a multi-stage stratified cluster randomized design was used to provide a representative sample in each province. Data collection for the fourth (2008) and fifth (2013) NHSS were conducted in Shaanxi Province represent time points before and after the healthcare reform of interest (2009).
In each survey round, face-to-face interviews were collected by the investigators trained by China’s National Health Commission using a household health questionnaire that mainly included open-ended questions (see Supplementary Questionnaire S1 and S2 online). Data on maternal socio-economic status (including area, age, education, health insurance, annual personal expenditure, employment) as well as chronic disease, parity, antenatal visits, hospital delivery and postnatal visits from pregnancy to 42 days after delivery were collected in the interview. During data collection, experts provided supervision and revisited 5% of the sampled households to check the accuracy of data recorded by interviewers. They asked 14 key questions again to check the consistency of the information recorded and the consistency should be at least 95%. The Myer’s Blended Index was used to assess the representativeness of the sample (1.67 in the 4th NHSS and 1.62 in 5th NHSS), indicating that in both surveys there was no significant difference between the sampled age distribution and the overall age distribution of Shaanxi Province [24, 25].
In brief, 44 counties in the fourth NHSS and 32 counties in the fifth NHSS were randomly selected. 18,290 household members in the 2008 NHSS were collected, more household members (57,529) were collected in the 2013 NHSS because of the expansion of the investigation site. For our study, women who had at least one delivery were selected as the sampling unit of interest in the fourth NHSS. From the fifth NHSS, only women whose last delivery occurred after January 2010 were selected, considering the official inception date of the health system reform (September 2009). This gave us a sample of 638 women in the fourth NHSS and 1,694 women in the fifth NHSS in this analysis (Figure 1).
Indicators
China’s 2009 healthcare reform in this study refers to a series of measures introduced and implemented after China’s 2009 healthcare reform to strengthen women’s maternal health care, mainly including the national important and basic public health service. According to WHO level, the utilisation of CMHS is categorized as: women who received ≥4 prenatal visits, hospital delivery and ≥3 postnatal visits from pregnancy to 42 days after delivery [26]. In the level of China, the utilisation of CMHS is categorized as: women who received ≥5 prenatal visits, hospital delivery and ≥1 postnatal visits from pregnancy to 42 days after delivery [27, 28]. Considering the difference of rural and urban population sample in terms of income, education and health service utilisation, the two groups were analysed separately and compared to look at geographic equity [29].
Statistical analysis
In this study, sample data has been checked for missing data and outliers and cleaned prior to data analysis. Descriptive analysis was performed to show the demographic information of maternal women in the sample and their status CMHS. A generalised linear mixed model (GLMM) including both fixed and random effects were used in this study to show the association between China’s 2009 healthcare reform and CMHS utilisation when controlling for other confounding factors. The healthcare reform was specified as fixed effects, women’s family code as a random effect; maternal women’s age, education, employment, annual personal expenditure, health score, health insurance, chronic disease and parity were included as covariates. The model we used was as following:
[Due to technical limitations, the formulas could not be displayed here. Please see the supplementary files section to access the formulas.] (1)
In equation (1), the linear prediction η is the combination of the fixed and random effects excluding the residuals. is the rate of CMHS utilisation. is a constant, represents the effects of on , and is a random error. The link function is binomial.
Concentration curve, concentration index (CI) and horizontal inequity index (HI) were used to measure the equity of CMHS utilisation. Before to measure equity, inequality should be measured first. Concentration curve and CI were used to measure the extent of income-related inequality of CMHS utilisation. It is calculated as twice the area between the concentration curve and the line of equality and changed from -1 to 1 [30]. A positive concentration index means that high-income women utilize more CMHS utilisation than their low-income counterparts and negative one means the low-income group utilizes more CMHS utilisation than their rich counterparts, the formula is as following:
[See supplementary files.] (2)
where C stands for concentration index, is CMHS utilisation index, μ is the mean of CMHS utilisation index, and is the fractional rank of annual personal consumption expenditure distribution.
Inequality can be further explained by decomposing the concentration index into its determining components, then horizontal inequity index (HI) can be computed by subtracting the contribution of need variables (such as women’s age, health score and chronic disease) from the concentration index of CMHS utilisation; it is a summary measure of the magnitude of inequity in the dependent variable [31]. These determinants are selected according to previous researches but constrained by the variables collected in the investigation [15, 32]. A probit regression model was used to indirectly standardize the CMHS utilisation since the outcome variable is binary. As the standardization of health utilisation holds for a linear model of healthcare, we applied the linear approximation to the probit model to extract marginal effects of each determinant on observed probabilities of the outcome variable. The formula for the concentration index decomposition can be written as follows:
[See supplementary files.] (3)
G is functional transformation, is the dependent variable, are needs variables, and are control variables. Then the standardized need was estimated using the following equation:
[See supplementary files.] (4)
where is standardized continuum of maternal health service utilisation, n is sample size. The more CMHS allocated to the population with greater need, the less inequity of CMHS utilisation.
The statistical analyses were performed using STATA statistical software version 12.0 (StataCorp LP, College station 77845, USA). A two-tailed P value < 0.05 was considered statistically significant.