2.1 Data and sample
Data were obtained from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative survey of individuals aged ≥ 45 that is administered every two years since 2011. The data included 17,708 participants distributed across 450 villages in 150 counties [16]. For cross-national comparisons of other international aging surveys, a Harmonized CHARLS was created to coordinate the CHARLS with the Health and Retirement Survey in the United States, for which detailed information is publicly available online [17].
Patients with hypertension were defined as respondents who self-reported high blood pressure in 2011 and were not covered by the NEPHSP in that year (hereafter, NEPHSP-uncovered). The final data set included 3,192 hypertensive patients who were not lost to follow-up and had no significant variables missing in 2011, 2013, and 2015 (Figure 1). This study reported the direct spending of outpatients and inpatients separately and used the data as the dependent variable in the models. All outpatient expenditure in the past month was recorded, including both treatment and medication costs. Direct medical costs were included in inpatient expenditures during the past year, while indirect medical costs were excluded. To identify total hospitalization expenditure in the past year, total medical cost of doctor visits, and amount paid by their insurance company, we used the following items from the CHARLS baseline questionnaire: “How many times have you received inpatient care?,” “How many times did you visit a medical facility?,” “What is your total hospitalization cost?,” and “What is your total outpatient cost?” If the respondent had two or more inpatient or outpatient treatments in the past year or month, then the respondent was asked to list the total medical costs for all visits.
The essential independent variable was the NEPHSP-covered participants. To determine whether hypertensive patients were covered by NEPHSP, we used the question “When did you have your last physical examination?” to identify responders who had received physical exams in 2011, 2013, or 2015. The question “Who paid for your last physical examination?” measured whether the medical costs were covered by the NEPHSP, given that the NEPHSP includes a free physical exam at least once a year for patients with hypertension aged ≥ 35. Respondents who chose “government” were the treatment group, and respondents who chose “non-government” had not received a free medical exam and were defined as the comparison group.
The covariate variables for this study consisted of individual socioeconomic characteristics and health information, such as age (≥ 65 or not), sex (male or female), marital status (living with or without a partner), occupation (farmer or not), education (1 = Less than lower secondary education, 2 = Elementary school, and 3 = Middle school). Household income classified by quartiles into four groups: poor (< RMB 3,335), low income (RMB 3,335–17,200), middle income (RMB 17,200–40,965), and high income (≥ RMB 40,965). Other characteristics were insurance (have or not), overweight and obesity (body mass index [BMI] ≥ 24.0 or not) [18], and self-reported health status (not poor or poor).
2.2 Statistical analysis
As indicated above, respondents covered by NEPHSP between 2013 and 2015 were defined as the treatment group, while the comparison group had not received this service. First, we found a gap in the socioeconomic and health characteristics between the two groups of hypertensive patients. The covariates of the two groups of hypertensive patients in 2011 were matched by propensity score matching (PSM) [18], which enables us to calculate weights based on the socioeconomic characteristics and health information of patients with hypertension that yield unbiased estimates of the impact of factors of interest [19]. We adopted the 1:4 neighbor matching method to match the treatment and comparison groups and retained the matched data for the final analysis.
Second, based on the matched data, a difference-in-differences (DID) method was used to analyze healthcare expenditure changes from before the NEPHSP (2011) compared to after (2013 and, separately, 2015). DID can effectively detect the intervention effect between the treatment and comparison groups and isolate the time trend unrelated to the intervention [20]. The effect of the NEPHSP is estimated by comparing the differences between two changes in outcomes: (1) changes between pre-and post-NEPHSP within the treatment group and (2) the pre-and post-intervention periods in the comparison group [21]. Given that the skewed distribution of health cost data (which contain many zero values) violated the normal distribution assumption of the ordinary linear model, a Tobit model is suitable for the analysis [22]. Therefore, a panel data Tobit regression model was employed for continuity outcomes to estimate NEPHSP’s effect on each group’s healthcare expenditure. More than 50% of the cost of patients with hypertension in our sample has a value of 0, and the values that are not 0 conform to the normal distribution; we reported the mean value when describing the sample. A P-value <0.1 was considered statistically significant. All analyses were performed using Stata software, version 16.
2.3 Ethical approval
All participants provided written informed consent and ethical approval for collecting data on human subjects was obtained from the appropriate Biomedical Ethics Review Committee.