This study intent to use two methods to analyze the reimbursement effect of SIMIS.
First, descriptive analysis was carried out on the inpatient situation of the serious illness patients, describing the different types of inpatient expenses and reimbursement situation in the year, and analyzing the reimbursement effect. In this paper, the hospitalization types could be divided into five categories: County hospitalization refers to the inpatient in Jinzhai County, City hospitalization refers to the inpatient outside Jinzhai County and in Lu'an City, Provincial hospitalization refers to the inpatient outside Lu'an City and in Anhui Province, and outside Anhui Province hospitalization refers to the inpatient outside Anhui Province, and cross regional hospitalization refers to the inpatient cross regional.
Second, the regression discontinuity (RD) method [5,6,7] was used to analyze the relief of medical expenses of rural residents after the implementation of SIMIS. RD method made use of the discontinuous characteristics of policy, that was, when the specific index of the research object was greater than the critical value specified by the policy, it would be treated by the policy, and the critical value was the breakpoint (C). As the policy of SIMIS in Jinzhai was to compensate the inpatients whose annual serious illness expenses were more than CNY 20,000, we could choose C = CNY 20,000 as the breakpoint in this study. The subjects whose annual serious illness expenses was more than CNY 20,000 would be included in the experimental group, and the subjects whose annual serious illness expenses was less than or equal to CNY 20,000 would be included in the control group. It could be considered that the basic situation of the objects near the breakpoint was similar. Whether they could enjoy the compensation treatment of SIMIS was the result of the random allocation of policies and systems. It could be regarded as a quasi-experiment. Because of the existence of random grouping, the average treatment effect of SIMIS near the breakpoint can be estimated.
In this study, STATA 15.0 RD statistical software package was used for data statistical analysis. The analysis process was as follows:
- The average and frequency indexes were used to describe the inpatient characteristics and payment status of serious illness patients.
- Determine the optimal bandwidth (H). The best distance from the breakpoint location. Generally speaking, the smaller the H was, the smaller the deviation of objects on both sides of the breakpoint was, but it may lead to less observation objects, resulting in excessive variance; otherwise, the larger the H was, the smaller the variance was, but it included objects far away from the breakpoint, resulting in excessive deviation of objects. Therefore, this study uses the method proposed by imbens and kalyanaraman (IK method) [8] to select the optimal bandwidth by minimizing the mean square error of two regression functions at the breakpoint.
- RD analysis. The dependent variables were actual medical insurance payment proportion (AMIPP), inside medical insurance payment proportion (IMIPP), inside medical insurance self-payment proportion (IMSPP) and outside medical insurance self-payment proportion (OMISPP). The independent variable (grouping variable) was Serious illness expenses. Covariates were age, hospital stay, total medical expenses, gender and inpatient type. Table1 showed variable definition and basic information. In the two intervals (C-H, C] and (C, C + H), the weighted least square method was used for linear regression, and the weight was determined by the trigonometric kernel function. The difference between the estimates of dependent variables of the two functions at point C was called local average treatment effect (LATE), which was also known as "local Wald estimator" (lwald).
- Validity test. When doing RD, we should also pay attention to the possibility of "endogenous grouping". For example, the patients whose serious illness expenses were less than CNY 20,000 had known the grouping rules in advance, they may take the initiative to make their serious illness expenses reached CNY 20,000 and enjoy the compensation policy, resulting in endogenous grouping rather than random grouping of patients near the breakpoint.
For the possibility of endogenous grouping, this study used the method proposed by McGrary (2008) [9]to test whether the density function of grouping variable was discontinuous at the breakpoint. First, the grouping variables were subdivided equidistantly on both sides of breakpoint C, the group distance was B, the center position of each group was noted as variable Xj, and then the standardization frequency of each group was calculated, which was noted as Yj. By using trigonometric kernel and local linear regression on both sides of breakpoint C, the estimated value and standard error of density function could be obtained according to the value of grouped variable. By comparing the estimated values of the density function at the breakpoint, we could judge whether the density function was continuous at the breakpoint.
In addition, if the conditional density function of covariates at breakpoint C also had a jump, it was not appropriate to attribute all policy effects to the implementation of policies. In fact, the implicit assumption of RD was that the conditional density of covariates was continuous at the breakpoint. In order to test this hypothesis, we took each covariate as the dependent variable and the serious illness expenses as the independent variable, and then carried out RD again to investigate whether there was a jump in its distribution at the breakpoint.
Table1 Basic information of variables
Variable
|
Variable definition
|
C≤20,000(n=5984)
|
C>20,000(n=1370)
|
Mean
|
SD
|
Mean
|
SD
|
Dependent variable (%)
|
AMIPP
|
Proportion of total medical insurance payment to total medical expenses
|
48.16
|
13.52
|
51.93
|
11.47
|
IMIPP
|
Proportion of total medical insurance payment to total inside medical insurance expenses
|
59.48
|
13.19
|
61.39
|
11.93
|
IMSPP
|
Proportion of inside medical insurance self-payment to total expenses
|
32.69
|
11.96
|
32.81
|
11.14
|
OMISPP
|
Proportion of outside medical insurance self-payment to total expenses
|
19.15
|
10.99
|
15.26
|
10.71
|
Independent variable (grouping variable)
|
Serious illness expenses
|
Total of ceiling self-payment,Clinical Necessary treatment cost,self-payment under the NRCMS
|
1.38
|
0.27
|
2.41
|
0.28
|
Covariates
|
Age
|
Age of inpatients
|
49.82
|
18.39
|
50.45
|
17.15
|
Hospital stay
|
Length of hospital stay
|
23.26
|
29.38
|
26.46
|
36.34
|
Total expenses
|
Total annual inpatient medical expenses
|
4.78
|
2.42
|
7.49
|
3.27
|
Gender
|
Female=1
|
0.58
|
0.49
|
0.57
|
0.49
|
Inpatient type
|
Inpatient grade(1-5)
|
4.06
|
1.06
|
4.24
|
1.16
|
Sample
The research data came from the medical insurance management center of Jinzhai County, Anhui Province, covering the individual annual hospitalization reimbursement data of NRCMS from 2013(n=73,042) to 2016 (n=73,571 in 2014, n=75330 in 2015, n=71928 in 2016), the use of data had been approved by Jinzhai County. Case information included: basic information of inpatients, hospitalization and medical expenses payment. In order to better analyze SIMIS reimbursement effect, we merged the four-year data (2013-2016, n=293871) for analysis.
According to the research idea, RD focused on the objects near both sides of the breakpoint. In order to ensure the same span on both sides of the breakpoint, we selected the patients from the merged data of 2013-2016 whose serious illness expenses was between CNY 10,000 and CNY 30,000 as the analysis objects (n=1553 in 2013, n=1797 in 2014, n=2147 in 2015, n=1856 in 2016, total n = 7353). And descriptive analysis of SIMIS reimbursement objects (n=468 in 2013, n=535 in 2014, n=831 in 2015, n=886 in 2016, total n = 2720) from the merged data of 2013-2016. Table 2 showed the distribution of serious illness expenses from 2013 to 2016.
Table 2 Distribution of serious illness expenses in Jinzhai County from 2013 to 2016
Expenses
|
Frequency
|
Frequency ratio(%)
|
0.0 ~
|
285501
|
97.15
|
1.0~
|
5983
|
2.03
|
2.0 ~
|
1370
|
0.47
|
3.0 ~
|
493
|
0.17
|
4.0 ~
|
221
|
0.08
|
5.0 ~
|
303
|
0.10
|