Background

China has launched a large set of reforms regarding health insurance and healthcare since the beginning of the twenty-first century. The Social Basic Medical Insurance (SBMI) system that integrates the urban and rural sectors has achieved nearly universal coverage in terms of participation rate. By the end of 2017, the SBMI had grown to cover over 97% of Chinese population [1]. However, considering the deductibles, copayments, and maximum allowable costs, it provides limited service provision and financial protection with reimbursement rates of range from 44 to 68% [2]. Therefore, the current SBMI system is in its early stage that can be characterized as “wide coverage, basic protection”. Previous literature showed inconsistent results on the relationship between the SBMI and health service utilization, i.e., some studies reported that medical insurance could help increase health service utilization and improve health outcomes [2,3,4,5,6,7], while some others yielded different outcomes [8, 9]. Consequently, whether the SBMI achieves equitable utilization of health services needs further investigation.

Currently, there are three programs in SBMI system: the Urban Employee Basic Medical Insurance (UEBMI) for the urban employed, the Urban Resident Basic Medical Insurance (URBMI) for all the non-UEBMI-eligible residents in the urban area, and the New Cooperative Medical Scheme (NCMS) for the rural population [10]. By the end of 2014, the UEBMI, URBMI, and NCMS covered 283.3, 314.5, and 736 million people at 2842.4, 1628.4, and 408 yuan annual premium cost per capita, respectively [23].

Geographic location (provinces)

Geographic location was measured by the province in which the respondent is located (including 28 provinces). The provinces estimated results were not reported in detail.

Access to care

Access to care was assessed by the number of medical facilities in the respondents’ village/community. Our study computed three proxy measures: number of hospitals (Hospitals: including general hospitals, specialized hospitals and Chinese medicine hospitals), number of community health care centers (Health centers: including community health care center and community health care medical post) and number of township/village clinics in the community/village (Clinics: including township health clinic and village medical post) [24].

Health insurance information

Respondents’ health insurance information was measured by a self-reported item. Respondents were asked if they have any kind of SBMI coverage, including the UEBMI, URBMI, or NCMS.

Health service utilization

Previous studies have examined health service utilization with different measures. Some studies simply used the frequency of health service utilization such as total outpatient or inpatient visits during a period of time to measure health service utilization [25, 26]. While some others used the costs of receiving health services within a certain period and weighted to costs to sum the resource intensity overall healthcare services [27]. In this study, we use the total cost of receiving health services to measure the health service utilization of the middle-aged and older adults.

Our study aims to deviate from the impact of the medical expenditure differences caused by different diseases, geographic locations and types of outpatient and inpatient services. Therefore, we further controlled five variables, namely, self-reported health, chronic condition, demand for inpatient services, different levels of hospital and geographic location. In this manner, the total medical expenditure can better measure the health service utilization of the middle-aged and older adults.

Health service utilization in our analysis reflect the medical expenditure of different utilization of health services. The health service utilization satisfies two conditions, namely, (a) medical expenditure of hospitalization in the past year preceding the survey date and (b) medical expenditure of outpatient visits in the last month preceding the survey date [2, 4, 5, 28]. We combined the results of the two surveys and computed their logarithm.

Quantile regression

To measure the differences across levels within groups, this study used a quantile regression approach [29]. Quantile regression obtains a different quantile function by examining the dependent variable distribution of different quantiles between (0, 1). Thus, a trajectory of the conditional distribution is formed. The quantile regression model can be expressed as follows [13]:

$$ {Q}_{\theta}\left(\left.y\right|x\right)=x\hbox{'}\beta \left(\theta \right), $$

where θ is the independent health service utilization of the samples; x is the explanatory variable; and regression coefficient β(θ) represents the interpretation of the independent variable y at the quantile level of the dependent variable. The corresponding β(θ) can be obtained by minimizing β in the following formula [29,30,31]:

$$ {n}^{-1}\sum \limits_{i=1}^n{\rho}_{\theta}\left({y}_i-{x}_i\beta \right) $$

with

$$ {\rho}_{\theta}\left(\mu \right)=\left\{\begin{array}{cc}\theta \mu & when\;\mu \ge 0\\ {}\left(\theta -1\right)\mu & when\;\mu \le 0\end{array}\right. $$

Then, our research model can be expressed as follows:

$$ y{\left(\mathit{\log}(expenditure)\right)}^{\tau }={\beta_1}^{\tau}\times age+{\beta_2}^{\tau}\times gender+{\beta_3}^{\tau}\times edu+{\beta_4}^{\tau}\times srh+{\beta_5}^{\tau}\times \log (income)+{\beta_6}^{\tau}\times marital+{\beta_7}^{\tau}\times chronic+{\beta_8}^{\tau}\times functional\ status+{\beta_9}^{\tau}\times demand+{\beta_{10}}^{\tau}\times access\ to\ care+{\beta_{11}}^{\tau}\times insurance1+{\beta_{12}}^{\tau}\times insurance2+\mathrm{p} rovinces\times {\gamma}^{\tau }+{constant}^{\tau }+{error}^{\tau }. $$

In our research model, βτ is the fixed-effects coefficients for the τ th quantile, provinces are the random effects independent variables, and γτ is the random-effects coefficients for the τ th quantile.

Given that the sample is from many provinces in China, our study used Linear Quantile Mixed Models (LQMM) to test our model. We used the R language tool lqmm package for linear quantile mixed regression analysis [32]. And we reported the result in Table 2.

Results

Descriptive statistics

Figure 1 presented the medical expenditure distribution of the three programs of the SBMI. It showed that most medical expenditure of the three programs is 0. Given that the distribution of the dependent variable was highly-skewed, we chose the linear quantile mixed regression to fully understand the effect on the dependent variable at different quantiles. Considering that the dependent variable prior to the 0.75 quantile equals zero, our study only ran linear quantile mixed regression after 0.75 quantile.

Fig. 1
figure 1

Medical expenditure distribution density

Full size image

Table 1 summarized the descriptive statistics of the variables. We reported the frequency and percentage for each categorical variable and the mean and standard deviation for each continuous variable. In the total sample, 83.2% (n = 10,886), 12.2% (n = 1596), and 4.6% (n = 605) were enrolled in NCMS, URBMI, and UEBMI, respectively. This result showed that NCMS covered the largest number of people in China. Almost half of the participants evaluated their health status as fair (54.9%), followed by poor (22.4%), good (11.3%), very good (10.2%), and excellent (1.2%). Based on the self-reported health information, we concluded that the majority of older adults were not in good health. The samples were divided into three groups according to the type of SBMI. Then we used the Kruskal–Wallis test to assess differences among the groups [33], and the results showed that there were substantial differences among the three groups. The descriptive statistics of Provinces and Chronic were shown in Table S1 (see Additional file 1).

Table 1 Descriptive statistics
Full size table

Linear Quantile mixed models

To characterize the determinants of and differences in utilization of health services in various locations, our study performed linear mixed model (LMM) analyses followed by linear quantile mixed regression. We took NCMS as the reference level and examined the associations of URBMI and UEBMI with health service utilization.

For ease of illustration, Table 2 selected five representative quantiles, namely, 0.75, 0.8, 0.85, 0.9, and 0.95. In the case of controlling the demographic and control variables, a significant positive association between the URBMI and health service utilization at 0.75 (β = 1.608, p < 0.01), 0.8 (β = 1.578, p < 0.01), 0.85 (β = 1.473, p < 0.01), 0.9 (β = 1.403, p < 0.01) and 0.95 (β = 1.152, p < 0.01) quantiles was observed. These results demonstrated that URBMI was significantly associated with an improvement in health service utilization of the middle-aged and older adults, but at higher quantiles, the improvement in utilization of health services was smaller. Consequently, URBMI provided the most significant improvement in health service utilization for middle-aged and older adults with lower health service utilization. And in linear mixed model, URBMI was significantly positive associated with health service utilization for the middle-aged and older adults (β = 0.660, p < 0.01).

Table 2 Estimation results of health service utilization
Full size table

Meanwhile, UEBMI was significantly positively associated with health service utilization at 0.85 (β = 1.196, p < 0.01), 0.9 (β = 1.070, p < 0.01) and 0.95 (β = 0.736, p < 0.01) quantiles. These results showed that UEBMI was associated with an improvement in health service utilization of the middle-aged and older adults at these quantiles, but this utility in health service utilization was smaller at higher quantiles. Accordingly, UEBMI had the most significant utility in health service utilization for middle-aged and older adults with middle-level health service utilization. Therefore, a certain degree of health inequity was observed between the NCMS coverage population and the URBMI and UEBMI coverage population. This inequity becomes evident at the urban and rural levels. However, no significant association between UEBMI and health service utilization was observed in linear mixed model.

Impacts of the types of medical insurance on health service utilization

In our study, the quantile coefficients at the 0.75–0.95 interval were calculated with a step-size of 0.025. Figure 2 presents a comprehensive picture of the linear mixed model results for important variables and the variation of coefficients with quantile in the linear quantile mixed regression. The figure presents the coefficients of URBMI on the utilization of health services, which fluctuated continuously between 0.75 and 0.80 quantiles and then increased to the peak at 0.83 quantile after which it could be witnessed a drop to the smallest coefficient at 0.97 quantile. This finding proved that a certain degree of inequity existed in the utilization of health services within URBMI.

Fig. 2
figure 2

Linear quantile mixed regression results. The long-dashed line in the figure is the estimated value of the coefficient of each variable linear mixed model. The short-dashed line indicates the confidence interval of the linear mixed model estimation. The solid line represents the estimated value of the linear quantile mixed regression coefficient of each variable, and the shaded part refers to the confidence interval of the linear quantile mixed regression (linear mixed model and confidence interval of linear quantile mixed regression is 0.95)

Full size image

Compared with NCMS, the coefficients of UEBMI were partly positive significant. Before 0.85 quantile, the coefficients of UEBMI gradually increased. And between 0.88 and 0.92 quantiles, the coefficients of UEBMI have fluctuated continuously. However, after 0.92 quantile, the coefficients of UEBMI decreased. This result proved not only the inequity utilization of health services among different types of SBMI but also the fact that a certain degree of inequity existed in the utilization of health services within a specific SBMI program. This finding also confirmed that a certain degree of health inequity was found between the middle-aged and older adults due to the different reimbursement benefits of the SBMI programs.

Impacts of insurance on outpatient and inpatient health service utilization

Our study used the total medical expenditure to measure the health service utilization, and the outpatient medical expenditure may be largely different from the inpatient medical expenditure. For providing a comprehensive understanding of the relationship between SBMI and health service utilization, our study further divided research samples into three groups: middle-aged and older adults without any health service utilization (without any medical expenditure), middle-aged and older adults with only outpatient health service utilization (with only outpatient medical expenditure), and middle-aged and older adults with inpatient health service utilization (with or without outpatient medical expenditure). Our study also summarized the proportions of three groups in Table 3. From Table 3, the three different health service utilizations of NCMS account for 72.3, 15.2, and 12.5%, respectively. As for URBMI, the proportions of different health service utilizations were 68.8, 14.7, and 16.5%. The different health service utilizations of UEBMI accounts for 71.4, 12.6, and 16.0%, respectively. Therefore, most middle-aged and older adults did not have any health service utilization.

Table 3 The proportion of different health service utilization in SBMI
Full size table

Considering that the dependent variable in the group without health service utilization equals zero, our study performed linear quantile mixed regression on middle-aged and older adults with only outpatient health service utilization or inpatient health service utilization. For convenience in presentation, our study selects five representative quantiles, namely, 0.1, 0.25, 0.5, 0.75, and 0.9. The estimation results for the group with only outpatient health service utilization were shown in Table 4, a significant positive association between URBMI and outpatient health service utilization at all quantiles was observed. The UEBMI was also significantly positive associated with outpatient health service utilization at 0.75 (β = 0.609, p < 0.01) and 0.9 (β = 0.692, p < 0.01) quantiles, and the results were consistent with the results in Table 2.

Table 4 Estimation results of samples with only outpatient health service utilization
Full size table

The estimation results for the group with inpatient medical expenditure were provided in Table 5, showing that the URBMI was significantly positive associated with inpatient health service utilization at all quantiles. These results demonstrated that URBMI was significantly associated with an improvement in inpatient health service utilization of the middle-aged and older adults, but this gain in health service utilization was smaller at medium quantiles. And a significant positive association between UEBMI and inpatient health service utilization was observed at 0.1 (β = 0.559, p < 0.01), 0.25 (β = 0.420, p < 0.05), 0.5 (β = 0.352, p < 0.05), and 0.75 (β = 0.306, p < 0.05) quantiles. These results showed that UEBMI was associated with an improvement in inpatient health service utilization of the middle-aged and older adults at most quantiles, but the gain in inpatient health service utilization was smaller at higher quantiles.

Table 5 Estimation results of samples with inpatient health service utilization
Full size table

Discussion

The purpose of this study was to investigate the association between China’s SBMI schemes and health service utilization of middle-aged and older adults at different quantiles. We used linear quantile mixed regressions to provide a comprehensive understanding of the relationship between the SBMI and health service utilization. Our study yielded two key findings.

First, our study found that URBMI and UEBMI were significantly positive associated with health service utilization compared with NCMS at most quantiles. These results were consistent with those of existing literature about the impact of medical insurance [4, 34]. And it could be explained as follows: compared with URBMI and UEBMI, the deductible of NCMS is generally high and the budget is relatively limited and thus the coverage is typically shallow [12, 35, 36]. Therefore, middle-aged and older adults from rural areas experience a certain degree of inequity in the health service utilization compared with middle-aged and older adults from urban areas.

Second, the results of the quantile regression analyses demonstrated that the associations between the SBMI and health service utilization largely differed at different quantiles. URBMI had the most significant gain in health service utilization for middle-aged and older adults with lower health service utilization, and UEBMI had the most significant utility in health service utilization for middle-aged and older adults with middle-level health service utilization. A likely account for this result was that middle-aged and older adults with high health service utilization might still cannot afford the medical expenses after reimbursements since the reimbursement rates of UEBMI, URBMI, and NCMS were 72, 50, and 40%, respectively [12]. And the limited utility of UEBMI for middle-aged and older adults with high inpatient health service utilization also confirmed this result. Due to the high deductibles for outpatient services, UEBMI had a limited impact on middle-aged and older adults with lower outpatient health service utilization. In addition, our results showed that the SBMI system had limited impact on the utilization of health services by the middle-aged and older adults who were overburdened with medical care. Because of the low national average premium per capita (2842.4, 1628.4, and 408 yuan under UEBMI, URBMI, and NCMS, respectively), middle-aged and older adults with lower health service utilization were more likely to benefit from the SBMI. This finding is consistent with previous literature [37, 38]. Previous studies have also proven that the rich people were more likely to benefit from medical insurance [39], and NCMS increased the utilization of health services. However, it increased the burden of medical expenses for the poor [40]. Thus, health inequity persisted in the middle-aged and older adults with various levels of health service utilization in China.

The results of this study provide insights that can contribute to policy design. First, due to the low reimbursement rates, the association between health service utilization and URBMI/UEBMI was limited for the middle-aged and older adults with high health service utilization. Second, reimbursement rates could be set according to the levels of medical expenditure. Third, health inequity still exists, especially in urban and rural areas. Future healthcare reforms in China should not only focus on expanding coverage, but also on improving the equity of distribution of healthcare benefits [41], and the gaps in the benefits package across the SBMI systems should be further reduced [18].

Expanding health insurance coverage is a critical step toward health equity. Patients are supposed to be able to access and afford healthcare [42]. However, various programs of the SBMI have different effects on the utilization of medical services among the middle-aged and older adults in China. The underlying reason is the design of the benefits package of basic social health insurance. In the near future, the gap in the utilization of health services among the middle-aged and older adults will continue to exist and should be properly settled.

We also made several contributions to the existing literature. Our study can be regarded as a continuation of the health equity research in the process of SBMI coverage in China, which expands the application scope of health equity. Meanwhile, using a quantile regression approach, we further explored the impact of different SBMI programs on the utilization of health services for the middle-aged and older adults in China, which expands the application scope of quantile regression.

Limitations

Our findings should be interpreted with caution because of the following limitations. First, we used self-reported survey data, which might suffer from measurement error. Second, our study only used medical expenditure to measure health service utilization, so the validity of measurement might be a little bit limited. Although we further controlled some variables, it could still be a potential problem. Third, the coefficients of UEBMI were smaller than URBMI, which might due to the fact that UEBMI beneficiaries were usually employed and younger and healthier than URBMI. It suggested that there might be some unobserved variables concerning differences in healthcare need that had not been included in this study. Fourth, our study only considered three SBMI programs in China and did not consider other commercial medical insurance systems, and it also ignored the middle-aged and older adults without the SBMI. Those samples can be further studied in future works.

Conclusions

We used the CHARLS dataset to explore the relationship between SBMI and health service utilization for middle-aged and older adults in urban and rural Chinese areas. Our study suggests that while accelerating the promotion of SBMI coverage, policymakers should reduce the gaps in the benefits package across different schemes. Policy efforts should further focus on adjusting social health insurance and optimizing healthcare resource allocation in order to enhance the effective utilization of healthcare services and control the increase of costs among middle-aged and older adults [41]. Furthermore, inequity can be gradually reduced through continuous adjustment of the medical insurance scheme, thereby effectively targeting economically disadvantaged enrollees [43].