What Are the Determinants of the Decision to Purchase Private Health Insurance in China?

The objective of this study was to assess the determinants of the decision to purchase private health insurance (PHI) in China. Nationally representative data from the fourth wave of the China Household Finance Survey from 2017 were used, and the dataset comprised 105,691 individuals aged 18 years or older. The Andersen health services utilization model was used to inform the research. Chi-square tests and logistic regression analyses were used to estimate the decision to purchase PHI. The proportion of the sample that had PHI was small, at 5.06%, but coverage for social basic medical insurance (SBMI) was 90.64%. Among PHI holders, the overwhelming majority (87.40%) also had SBMI. Logistic regression analysis demonstrated that predisposing factors (age, education, marital status, household size), enabling factors (household income, SBMI status, geographical factors, household medical expense, and medical debt), and needs-based factors (health status) were statistically significant determinants of the decision to purchase PHI. This study suggests that the socio-economic circumstances of households play a crucial role in the decision to acquire PHI. The findings may be used by the insurance industry to inform actions to enhance PHI coverage and by policy decision-makers that seek to improve equality in access to PHI.


Introduction
For more than three decades, the Chinese government has devoted considerable efforts to expanding universal health insurance coverage [1]. Three main programs provide insurance coverage for various groups in Chinese society. Firstly, the Urban Employee Basic Medical Insurance (UEBMI) scheme was piloted in 1994 and was launched nationwide in 1998. This scheme provides coverage to employees in formal sectors and is funded by both employees and their employers. Secondly, the New Cooperative Medical Scheme (NCMS) was set up in 2003 and provides coverage for rural residents. Households and central and local governments fund it. Thirdly, the Urban Resident Basic Medical Insurance (URBMI) scheme was piloted in 2007 and then implemented nationwide in 2009. It was designed to provide coverage to urban residents who were not covered by the UEBMI scheme [2]. Both URBMI and NCMS were funded jointly by those insured and by the government [3], while UEBMI was funded by those insured and their employers. Due to the disparities caused by the market segmentation of health insurance (UEBMI, NCMS, and URBMI) in China, different sections of the population face different degrees of financial protection [4,5]. Since 2016, the Chinese government has moved forward with reforms to the existing schemes. The first step was taken in 2016 to integrate URBMI

Conceptual Framework
The Andersen model of health service utilization provides the underlying conceptual basis for our empirical analysis. This conceptual framework has been used in the literature to account for variations in health services utilization [24][25][26]. The model advances three sets of factors to account for health services utilization and, in this paper, the determinants of the decision to purchase private health insurance. These three factors are predisposing, enabling, and needs-based factors [27,28]. Each of these determining factors are described below.
Predisposing factors: these represent factors that make it more likely that individuals will seek health services. These include demographic variables, such as age and gender, which influence a person's likelihood of purchasing PHI. Additionally, social variables, such as marital status, education level, and household size, are variables that predispose individuals to acquire PHI.
Enabling factors: these represent factors that support health-seeking behavior and, in our context, provide support for individuals to purchase PHI. These include household income, as the main unit of decision-making in Chinese society is the household unit, and variables that represent employment status, as each type of employment circumstance offers differential access to other types of health insurance that may either substitute for or complement the decision to acquire PHI. Moreover, place of residence and household registration status (i.e., Hukou status) provide individuals with different opportunities to access services and thereby influence their decision to purchase PHI. Household medical expense is also considered, as these expenses capture the scale of financial protection sought by individuals in their decision to purchase PHI. We also included medical debt status as a variable to support their seeking behavior.
Needs-based factors: these represent factors associated with health status that generate an underlying need to acquire health services and consequently result in a derived demand for PHI. These variables include perceived and evaluated health status, as these variables are prerequisites for health service utilization.

Data Source
Data were obtained from the fourth wave of the China Household Finance Survey (CHFS) in 2017. This survey was launched in 2009 by the Survey and Research Center for China Household Finance, which conducts nationwide surveys every two years. A three-stage stratified random sampling design method was used in the survey. First, counties were randomly selected based on GDP groups (ranked according to GDP and classified into groups) from 2585 counties in all provincial administrative regions excluding Xinjiang, Tibet, Macao, Hong Kong, and Taiwan. Second, within each country selected, there was random sampling of communities and then the random selection of 25-50 households within each urban community and 20 households for rural communities [29,30]. Finally, face-to-face interviews were conducted with adults in selected households. Details of the design of CHFS can be found on their website: https://chfs.swufe.edu.cn/science/family.html.
The CHFS represents the most comprehensive public source for household finances in China. As the decision to purchase PHI is strongly influenced by household financial status, the growth in household finance associated with the dramatic growth of the Chinese economy has resulted in a burgeoning of PHI. Over the past five years, with the steady growth in PHI, more attention has been paid to this sector by the central Chinese government. We use, in our analysis, the most recently available public data to identify the determinants of the decision to purchase PHI. The fourth survey used in our analysis covered 29 provinces, 355 counties, and 1428 villages, with a sample size of 40,011 households that in total comprised 127,012 individuals. Our research focused on individuals aged 18 years or older who were asked to complete information on their decision to purchase PHI. This resulted in an analysis sample of 105,691 individuals, with 99.54% reporting PHI purchasing information.

Statistical Analysis
The dependent variable was binary and indicated whether an individual bought PHI or did not purchase such insurance. The independent variables were summarized previously under the three sets of factors suggested by the Anderson model, namely predisposing, enabling, and needs-based factors, and these variables are listed in Table 1.
Logistic regression analysis and chi-square tests were used to assess the determinants of the decision to purchase PHI empirically. First, descriptive statistics and chi-square tests of independence were calculated in order to assess the whole data set (Tables 2 and 3). Second, univariate logistic regression analysis was applied to examine the crude relationship between each independent variable and the decision to purchase PHI (Table 4, Model 1). All the results of model fit from univariate logistic regression are reported in Appendix A. Third, multivariable logistic regression analysis was performed based on the results of the univariate analysis. Backward stepwise logistic regression analysis was performed with a significance level of 0.10 to enter variables into the multivariate model (Table 4, Model 2). The results of Model 2 were statistically significant (χ 2 = 5783.55, p = 0.000), indicating that the model was able to distinguish between respondents who purchased PHI and those who did not purchase PHI. Other models were also performed to assess model robustness (Appendix A Table A1).
All of the logistic regression results were reported as odds ratios (ORs) and 95% confidence intervals (CIs). An OR above one indicates that the specified determinant was more likely to be associated with PHI purchasing. STATA 15.0 was used to perform the data analysis. it is a kind of medical security regime that provides free or nearly free medical services for the staff in government or public institutions. It was set up in 1952 and reformed in 1998. While it has been gradually phased out since 1998, there were still a small number of employees with FMS in 2017. c Household medical expenses refer to the total household medical spending, including the insurance claim amounts and the out-of-pocket payment. Table 2 reports the descriptive statistics for the study variables. Of the 105,691 respondents, 5344 (or 5.06%) reported that they had purchased PHI. The sample was almost equally divided between men and women (49.77% vs. 50.23%). The proportion of individuals in each age group was relatively balanced, and the largest proportion was in the group aged 65 or over (21.61%). More than 79% of the respondents were married. Around 65.18% of the sample were from urban regions, and 48.73% were from the eastern area of China. Chi-square tests of independence were used to analyze the difference between the characteristics of those that purchased PHI and those that did not make such purchases. Table 2 reports the preliminary analysis of chi-square tests of independence and demonstrates, as expected, that there were significant differences in all variables (except the continuous independent variable, household size).   Table 3 reports the descriptive information on respondents with both PHI and SBMI. The overwhelming majority (90.64%) of the total sample had one of the four primary forms of SBMI. The overwhelming majority (87.4%) of individuals who purchased PHI also had some form of SBMI. For those with both forms of insurance, almost half (46.6%) also had UEBMI.  Table 4 reports the results of the logistic regression analysis. The univariate logistic regression suggested that all the potential factors were significantly associated with PHI purchasing (Model 1, Table 4). Men, married respondents, urban residents, and those living in the eastern area of China tended to report a higher likelihood of purchasing PHI than the reference group. Education level and household income were also positively related to this likelihood. Household size and medical debt were negatively related to the likelihood.

Logistic Regression Model
The multivariable logistic regression analysis was performed by using backward stepwise regression methods (Model 2, Table 4). This analysis identified the variables that were statistically significant determinants of the decision to purchase PHI. We also checked the collinearity of the independent variables. The variance inflation factor (VIF) for all the independent variables in Model 2 ranged from 1.02 to 4.08, with all values below the conventional threshold value, and the maximum VIF was below 10 [31]. This result indicated no serious collinearity problems. We applied the Anderson model as the conceptual framework to guide our empirical work and listed independent variables under each of the three sets of factors, which were expected to be associated with the decision to purchase PHI. The findings are discussed in line with each of these factors. In the case of predisposing factors, namely age, education, marital status, and household size, each was seen to be a significant independent variable in driving decisions to purchase PHI. Individuals with higher levels of educational achievement exhibited higher odds of purchasing PHI. Compared with the reference group, individuals aged 25-54 years old were more likely to buy PHI. Being married increased the likelihood that PHI would be purchased by 18.0% (OR, 1.180; 95% CI, 1.047-1.330). While marital status was significantly related to the decision to purchase PHI, household size was negatively associated with the same decision. The results demonstrated that a unit increase in the number of household members was associated with a reduction in the odds of purchasing PHI by 10.9% (OR, 0.891; 95% CI, 0.866-0.916).
For enabling variables, we were not able to find a statistically significant role played by either Hukou type or employment status in the decision to purchase PHI. In contrast, we did find a critical role placed by geographic factors. Specifically, residents from eastern regions had a higher propensity to acquire PHI than those in the middle regions of China, with the lowest propensity to purchase PHI occurring in western regions. Residents in western regions were 18.2% less likely to acquire PHI than those in eastern regions (OR, 0.818; 95% CI, 0.743-0.901). Rural residents were 11.0% less likely to buy PHI than urban residents (OR, 0.890; 95% CI, 0.797-0.994). For those who were currently working, the type of work unit influenced the odds of purchasing PHI. Employees of state-owned/collective enterprises and private/foreign-owned enterprises were more likely to acquire PHI than those in other types of enterprises. Income was an essential factor in the purchasing decision. The results showed that the likelihood of purchasing PHI increased with household income. SBMI is a mandatory form of health insurance in China, but there are limits to its medical service coverage and the level of reimbursement. When the government liberalized the commercial health insurance market, there was a growth in the decision to purchase PHI in order to obtain more and better coverage, even when individuals were still holding SBMI. The likelihood of purchasing PHI varied by the type of SBMI that individuals held. Individuals were 16.1% less likely to acquire PHI, relative to those covered by the UEBMI scheme, if they were covered by the NCMS (OR, 0.839; 95% CI, 0.745-0.994), but they were 36.1% more likely to purchase PHI if they were covered by URRBMI (OR, 1.361; 95% CI, 1.092-1.698). Individuals without any form of SBMI were 67.1% more likely to purchase PHI (OR, 1.670; 95% CI, 1.458-1.913). Moreover, because the decision to purchase insurance is correlated with financial risk perception, our expectations were confirmed when the decision to purchase PHI was significantly associated with decisions by individuals who had bought other types of private non-health insurance. Those that made such purchases were much more likely-indeed, nine-fold more likely-to buy PHI than those that had not made such decisions (OR, 10.222; 95% CI, 9.395-11.122). It was also shown that an increase in medical expenditure increased the likelihood to purchase PHI. Specifically, when household annual medical spending was in the 2000-4999 RMB and 5000-9999 RMB groups, the decision to purchase PHI was 12.9% and 16.4% more likely, respectively, than for those in the 0-1999 RMB group (OR, 1.129; 95% CI, 1.027-1.242 and OR, 1.164; 95% CI, 1.037-1.306). In more extreme expenditure situations-i.e., where expenditure exceeded 10,000 RMB-there was no statistically significant relationship to report, as such levels of expenditure might drive the household into debt. Indeed, where respondents reported medical debt, the likelihood that they would acquire PHI was significantly reduced, by 29.2%, compared to those without such debts (OR, 0.718; 95% CI, 0.548-0.942).
Needs-based factors were also shown to be significant determinants of the decision to purchase PHI. Individuals with poor self-evaluated health were less likely to buy PHI than individuals reporting good or fair health. Existing studies indicated that healthier individuals had higher risk aversion, and purchasing PHI was more suggestive of risk aversion [21]. This might be the reason that healthier individuals were more likely to buy voluntary PHI.
We also conducted other multivariable logistic regression models in order to acquire robust results. First, a multivariable model with all the potential predictors included was performed, and the results are reported in Appendix A (Table A1, Model 4). Second, considering potential bias caused by data where employer type was missing, the multivariable model with all the potential predictors except employer type was performed (Table A1, Model 3). The main results were consistent with our results above. We also performed models with each set of factors individually to show the model improvements when considering all three sets of factors (Table A2, Model 5-7).

Discussion
This study represents the first examination of the determinants of the decision to purchase PHI in the context of China. We found that 5.06% of the study sample had purchased PHI, and 87.4% of those with PHI also had other forms of insurance coverage through SBMI. The results demonstrate the usefulness of the Anderson model, which highlights the role of predisposing factors (age, education, marital status, household size), enabling factors (household income, SBMI status, geographical factors, household medical expense, and medical debt), and needs-based factors (health status) as statistically significant determinants of the decision to purchase PHI.
Health and household financial status were significant determinants of the decision to buy PHI. Individuals with higher health status were more likely to acquire PHI. This finding is consistent with US studies that have shown that those with better health status are more likely to buy private Medigap insurance [32,33]. Higher socioeconomic status was also correlated with the acquisition of PHI [34]. Household financial status affects the decision to purchase health insurance, which in turn might impact out-of-pocket household health expenditure. For example, several studies have discussed catastrophic health expenditure (CHE), because CHE directly leads to medical debt and there is still a high percentage of CHE in Chinese households [35]. It was found that NCMS increased out-of-pocket medical expenditure by 12.3% [14], and it failed to prevent CHE and medical impoverishment [15]. In response to this situation, the Chinese government introduced, in 2016, the critical illness insurance (CII) program, which provides coverage to alleviate economic burdens among those with NCMS and URBMI coverage [36]. However, it is still unknown whether this program has achieved its intent [35]. To the best of our knowledge, few studies have addressed the effects of PHI on CHE in the context of China.
It is hard to say that PHI will provide financial protection for most of the people in China because of the complicated relationship between PHI and SBMI. PHI was expected to expand universal healthcare insurance coverage and alleviate disease burden, but there has been significant controversy regarding the relationship between PHI and SBMI. As for how SBMI affected PHI, Zhang et al. (2018) pointed out that the impacts of SBMI on PHI were different in terms of insurance penetration and density [11]. Studies in the context of other countries revealed that there was a negative (or crowding-out) relationship. For example, evidence from the US using Medicaid data indicated that the expansion of public health insurance was associated with lower private market insurance premiums [37]. Our study indicated that different types of SBMI might impact the demand for PHI in the context of China. However, we must pay attention to the differences in the health system context between China and other countries. More than 95% of the residents of China are enrolled in the SBMI, which is a mandatory form of insurance coverage, and only a small portion of those individuals also voluntarily acquired PHI. Among SBMI holders (including UEBMI, NCMS, URBMI, URRBMI, and FMS), 4.83% of them purchased PHI. The main point is that the health insurance system is different from the dual insurance system in countries such as Germany and Chile, where individuals only acquire either PHI or social health insurance [38]. In China, government encouragement of the expansion of PHI is not for the purpose of covering individuals who are unable to acquire SBMI. Rather, PHI in China is a form of supplementary coverage for individuals who are already enrolled in the SBMI scheme [8]. Our study revealed that PHI purchasing was significantly associated with household financial status, including household income and medical spending. As a result, the acquisition of PHI did not provide financial protection for those in financial need [23] but rather may best be seen as a mechanism to increase access to, and probably perceived quality of, health services. Consequently, it is limited in alleviating CHE and decreasing impoverishment in China.
Moreover, PHI has the potential to exacerbate health inequalities in China. Health insurance schemes were examined to have an impact on healthcare utilization and costs [39], and different health insurance schemes may be associated with healthcare utilization inequality [40]. There are demands in China to distribute access to healthcare services more equitably, in line with the population's needs rather than their ability to pay [41]. Our findings demonstrate the importance of household financial status, geographical factors, and SBMI type as associated with the decision to purchase PHI. Different SBMI schemes caused variation in inequalities in accessing healthcare [40], and the acquisition of PHI might further exacerbate this inequality. This warrants study in future work.
The results of this study demonstrated that household financial status was the critical determinant of the decision to purchase PHI. Our study has two contributions. Firstly, it can contribute to the existing literature by providing new findings about the determinants of PHI purchasing in a different relationship context between private and public health insurance. Different from PHI in Germany and other developed countries, where PHI is a substitute for public health insurance, PHI in China is supplementary of public health insurance. Secondly, the results of this study provide a reference for government to adjust health policies. The results indicate that there are signs of increasing health inequities. Due to the complicated relationship between PHI and SBMI in China, PHI might not play the role that it was expected to play. Government and policymakers need to pay more attention to this and make sure that the expansion in PHI coverage is consistent with national health policy goals.
There are several limitations associated with this study. Firstly, we used cross-sectional data in order to identify an association with the decision to acquire PHI. The PHI market is still at an embryonic stage of development and is evolving rapidly in China. The overall level of population coverage is quite small but growing. There are very few databases that offer information on the decision to acquire PHI and so our choice of data was limited to the China Household Finance Survey (CHFS). Secondly, some critical factors might have been excluded from our analysis because they were absent from the CHFS, especially data on some needs-based factors, such as physician visits, comorbid conditions, and disease related spending. Finally, the decision to purchase PHI was considered as a binary variable in our research; we did not examine the amount or form of coverage. Such details and complexities would be interesting areas of future research and also potentially open to explanation using the Anderson model of health service utilization.

Conclusions
This study contributes to our understanding of the determinants of the decision to purchase PHI in China. The socio-economic circumstances of households play a vital role in the decision to acquire PHI. Notably, household financial variables are critical determinants of the decision to acquire PHI. Moreover, it is important to monitor this area of the health services marketplace as the growth in PHI may have implications for health inequities in China. It might also be useful to both policy makers and insurance industry leaders to better understand the factors that influence decisions to acquire insurance as well as the associated depth of coverage. Policy makers may wish to monitor developments to ensure that growth in PHI is compatible with national health policy goals. OR refers to odds ratio; 95% CI refers to 95% confidence intervals. *** p < 0.01, ** p < 0.05, * p < 0.10