Targeting vulnerable groups of health poverty alleviation in rural China— what is the role of the New Rural Cooperative Medical Scheme for the elderly population?

BACKGROUND : Based on the health poverty alleviation policy, we explored whether the New Rural Cooperative Medical System (NRCMS) had effectively reduced the economic burden of the rural elderly population and impoverished vulnerable groups with regard to medical expenses, providing further evidence for increasing the medical insurance system. METHODS: Data were obtained from the 2015 China Health and Retirement Longitudinal Study. The method was adapted from WHO to calculate the catastrophic health expenditure (CHE) and impoverishment by medical expense (IME). The treatment effect model was used to identify the determinants of CHE in the rural elderly residents. RESULTS: The incidence of CHE in rural China for the elderly is 19.65%, and the impoverishment by medical expense has reached 6.94%. The households enrolled in NRCMS suffered higher CHE (21.9%) and IME (8.0%), than unenrolled households (20.6%, 7.7%). The NRCMS did not provide sufficient economic protection from CHE for households with 3 chronic diseases, inpatients, or adults over 65, Risk factors for CHE included education levels, households with inpatients, people over 65 years old, disabilities, and so on. CONCLUSIONS: Although the NRCMS had reduced barriers to the use of household health services not Our research identifies the characteristics of vulnerable groups that the and risk falling through health

the three years from 1997 to 1999,every year about 8 million people solve the problem of food and clothing and poverty. China's rural areas ushered in the highest poverty alleviation since the 1990s. 9 The resulting rapid economic growth doubled the annual average poverty alleviation rate in rural areas----up to 10.41%. The fourth stage brought about the implementation of a consolidation-oriented comprehensive poverty reduction strategy (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012). At this stage, poverty-reducing factors were diversified. 10 The economy was no longer the only factor causing poverty among rural residents.
Aging and the high incidence of chronic and major diseases increased the economic burden on  13 The rural population had a serious economic burden. Impoverishment from medical expenses or returning to poverty due to illness had become one of the largest factors contributing to the poverty of the rural population.
It is estimated that the annual growth rate of medical expenses for the elderly in China will increase to 2.2% in 2010-2030, which is much higher than that of the United States and other Organization for Economic Cooperation and Development (OECD) countries (0.3%---0.5%) 14 . The number of poor people in rural China is three times that of urban areas. 15 At the end of 2017, China's rural population reached 576.61 million, of which the population aged over 60 reached 240.9 million, accounting for about 41% of the total rural population. Moreover, the elderly have always been susceptible to chronic diseases. 16,17 The economic burden of diseases caused by chronic illnesses accounts for 70% of the total disease economic burden. 18 Older people with chronic diseases must respond to their accompanying burden of disease through their remaining lives. 19 As the most direct and effective means of health poverty alleviation, the medical insurance system aims not only to protect the health of residents, but to also avoid economic risks caused by the utilization of medical services. 20 In 2017, the coverage rate of NRCMS had reached more than 98%. However, the NRCMS had gradually exposed some disadvantages in the development process. While the NRCMS had reduced the threshold for the utilization of medical services for the rural population, it also increased the economic burden resulting from diseases. In 2016, the urban employee medical insurance hospitalization reimbursement ratio reached 75%, which was 20% higher than the NRCMS reimbursement rate.
Although the coverage of NRCMS is the highest among the three basic medical insurances, it lagged behind the other two types of insurance in terms of depth of coverage and height. 21 Changes in the demographic structure have led to changes in household demand. Can health poverty alleviation accurately target the weakest people in the operational processes? As a carrier of health poverty alleviation, has the NRCMS really alleviated the economic burden of the rural population? Has the overall health poverty alleviation reached the expected goal? These questions still deserve a definite answer. Therefore, from a multi-dimensional perspective, we scanned the characteristics of povertystricken groups aged over 45 years in rural areas, and identify the key challenges that lead to the failure of the NRCMS.

Data source and sampling method
This study used the China Health and Retirement Longitudinal Study (CHARLS) database, which is a large-scale interdisciplinary survey project jointly conducted by the National Development Research Institute and the Social Science Research Center of the Peking University. In 2015, using multi-stage sampling and PPS sampling methods, CHARLS randomly selected 45-year-olds from the survey households that were, in turn, selected from 450 communities in 150 counties of 28 provinces (autonomous regions and municipalities) across China. After data cleaning (eliminating abnormal and incomplete data), 7,080 households and a total of 13,740 people remained for the calculation of catastrophic health expenditure (CHE). This database will be openly and freely available to academia one year after the survey is completed.

Statistical analysis· Catastrophic health expenditure calculation method
The method recommended by the World Health Organization (WHO) was used to calculate catastrophic health expenditure. The key variables in the algorithm included: Out-of-pocket health expenditure (OOP): The amount paid by a family member in cash when purchasing health care services. Household consumption expenditure (exp) refers to the currency and goods used in all goods and services consumed by the family. A household's capacity to pay (CTP) is the nonsubsistence spending of a household as a share of total household consumption expenditure.
When OOP exceeds 40% of the family's ability to pay, the family is considered to have catastrophic health expenditures. 22

Treatment-effect model and instrumental variables
This study selected whether a household was incurring catastrophic health expenditure as the main outcome variable. Since the interpreted variable is a binary variable and based on the nature of the voluntary participation in the NRCMS, the use of traditional ordinary least squares (OLS) regression leads to biased results. A binary selection model will more scientifically reflect the impact of medical insurance on CHE. Therefore, we choose the treatment effect model with the following model parameters.
The medical insurance scheme was included in the model as the primary factor in mitigating catastrophic health expenditures, but it had been identified by other studies as an endogenous variable that affects the OOP and CHE. If there is endogeneity, NRCMS cannot truly reflect the impact on CHE. In this regard, tool variables were introduced on the basis of the treatment effect mode model to correct the model and to prevent errors in the results.
The selection of the tool variables for this research primarily refers to the existing literature, selecting the same community participation rates of medical insurance schemes, self-assessment health, and provinces as the initial instrumental variables. Firstly, the weak tool variable was tested by the Stata software. When F>10, there is no weak instrument variable. After F-test, the F = 46.84 is greater than 10, so there is no weak instrument variable. Secondly, the redundant tool variable test was carried out, and the self-rated health and provinces were excluded according to the results of the redundancy test of the tool variables. As a tool variable for residents to participate in medical insurance, the rate of community participation is higher in life, and the probability of individuals participating in medical insurance is relatively high. However, the same rate of participation in the community does not generally affect the incidence of catastrophic health expenditures (CHE) in other households. Therefore, we can determine whether the participation rate and the explanatory variable are stored in the correlation, and if the variables are independent of the random error term of the CHE, and if the variables conform to the application condition of the instrumental variable method. Table 1 Sample characteristics

Basic information
The total sample comprised 7080 households and 13740 individuals, the final participation rate of the sample population was 85.8%. 53.1% of household heads were male, and about 32% of family members have 65-year-olds. Primary and junior high school education accounted for 66.5%, and more than half (62.5%) of elderly members had chronic diseases.  (Table 2): the prevalence rate of populations with three chronic diseases was the highest (17.2%). The higher demand for medical services was converted into higher health service utilization, with the highest hospitalization rate of 15.1% for populations with three chronic diseases. However, its hospital reimbursement rate was only 45.8%, nearly 7% lower than that of the populations with two chronic diseases. Populations with members over 65 years of age achieved high utilization while high demand and the hospitalization reimbursement rate had reached to 53.7%, which was 1.3 times than that of households without 65 years old members.
Disabled people also showed the same trend. Meanwhile, the prevalence rate of disease decreased with the increase of the educational level. Compared with populations with a high school education, populations who were illiterate had a higher utilization of health services, hospitalization rate (14.3%), and outpatient rate (20.6%), but its outpatient reimbursement rate (13.3%) was 10.6 percentage points lower. Compared with the central region, the health demand and utilization in the western region were both higher, with 5% prevalence and 3.2% hospitalization, respectively. Meanwhile, the western region had the highest reimbursement rate of the three regions. The prevalence of the uninsured population (15.9%) was higher than that of the insured population (14.8%), but whether outpatient or hospitalization, the uninsured populations were lower than the insured population, and the payment ability of the two was almost the same.

Table 3:CHE, and IME in different households
The above data showed that the highest incidence of CHE were concentrated in households with members that have three chronic diseases (38.0%), inpatients members (31.0%), or members over 65 years old (30.5%). As can be seen in Table 3, the capacity to pay medical expenses in households with three chronic disease was the weakest, but they also had the heaviest economic disease burden.
The NRCMS is insufficient for the reimbursement of households that have various chronic diseases, resulting in the highest incidence of CHE (38.0%), and the households members with 65 members were more at risk to incur impoverishment by medical expense; the incidence of IME (11.1%) was with 2 persons or less, participation in insurance had increased the utilization of health services, but their OOP payment did not increase. Among the groups with disabled people, the prevalence of households participating in insurance (17.1%) was lower than that of uninsured (17.7%). Their hospitalization and outpatient utilization rates were higher than the latter, especially the outpatient utilization rate (20.8%) which was 2% higher than the uninsured households. Participation in insurance reduced the threshold of health service utilization and provided corresponding economic protection with the OOP payment 0.89 times higher than the uninsured. It can be seen in the families of 2 or less, although the NRCMS made insured households with lowered health demand access higher utilizations, with the outpatient rate (20.1%) 3.9% higher than the uninsured family. Households with lower educational levels also showed the same trend. In the context of high health service utilization, the OOP payment of illiterate population (N = 276.49) who participated in insurance was far lower than the 406.03 yuan of uninsured people. Among the wealthiest groups, although the OOP payment of uninsured households was 1158 yuan higher than the insured households, their utilization of health services was also higher. The hospitalization rate of uninsured was 17.3% higher than that of the insured households, which demonstrated that medical expenses increased with the availability of health services, so the choice of participating in the insurance had little effect on the wealthiest families.

Figure 1. CHE and IME under different insurance schemes
From the perspectives of different insurance schemes (Figure 1), except for the sub-poor households, regardless of the incidence of CHE or IME, the insured households were higher than the uninsured households. The riskiest households were the insured households with three types of chronic diseases (39.44%), inpatients (32.64%), and had members above 65 (30.9%). Among the households with three chronic diseases, the incidence of CHE from those enrolled in medical insurance was 8.58% higher than the uninsured; for the households with inpatients, the incidence of CHE for enrolled households was 1.4 times than that of the latter; for the households with 65-year-olds members, the incidence of CHE for households participating in medical insurance was 2.82 times higher than the latter, but even for the households with 65+ members but no participating in medical, the incidence of CHE was still 1.2 times than that of households with two chronic diseases enrolled in medical insurance. Secondly, for the sub-poor families, regardless of whether they had medical insurance, the incidence of IME were both high (22.97% and 22.14% respectively), but it is worth noting that the incidence of CHE in the sub-poor households was not the highest (23.93/26.96%) the second highest incidence of IME was the households with 3 chronic diseases, 15.8% for insured families and 13.58% for uninsured families; and finally, 11.71% of the insured families with inpatients occurred the IME 1.4 times higher than that of inpatients (7.95%) who did not have insurance. Regardless of whether the household head had a university level education and was insured or not, the incidence of CHE and IME were both zero. And it can be clearly seen that CHE and IME decreased with the degree of education for both CHE and IME.

Table 5 :The result of treatment-effect model
Since the instrumental variables must meet the two main indicators of effectiveness and relevance, according to the regression results of the first part of the treatment effect model, the invalid tool variables self-evaluation health (p = 0.383) and provinces (p = 0.377) were excluded, and the community participation rate of medical insurance was finally selected. The participation rate (p = 0.000) was finally used as a tool variable. Moreover, the OLS test demonstrated that it was a reliable tool variable (F = 46.84). Since there was only one endogenous variable and one instrument variable, it was just recognized, so there was no over-recognition and under-recognition.
From the results of the treatment effect model (Table 5), we can see that of the education level of the head of the household, households with inpatients, people over 65 years old, disabilities, family size, disease types, as well as whether participating in the insurance are all determinants affecting catastrophic household expenditure for the elderly population in this study. Among them, households with hospitalization, elderly people over 65 years old, disabled members and chronic disease types as the most significant (p < 0.001) factors impacting CHE. In rural China, income seemed to have no significant association with the occurrence of CHE and IME for the elderly.
The CHE decreased by 2.94% with the increase of education level. The increase in education level had a protective effect on the family. Inpatients in the family increased the incidence of CHE by 13.11%. it has also increased the cost burden of groups of three chronic diseases, 65-year-olds, and inpatients, increasing the risk of falling into poverty. Participating in insurance makes the self-paying households with three kinds of chronic diseases pay 1.32 times more than the uninsured, the rate of poverty caused by illness is 1.5% higher than that of uninsured families, and the risk of poverty among the elderly over 65 years old is as high as 11.18%. Chronic diseases are the main factor of disease burden in the elderly. More and more elderly people with chronic diseases increase their health care expenditures because the use of health care resources increases with age and the number of chronic diseases. 28 The elderly and families of chronically ill patients have increased the demand for medical systems and medical expenditures. If these families lack financial protection for health risks, high medical expenditures may plunge them into poverty. 29,30 Although the NRCMS has achieved better protection in groups with one or two chronic diseases, the protection is still insufficient in the superposition of various chronic diseases.The incidence of IME in the superimposed family reached 15.08%, which was 2.73 times that of families without chronic diseases. The OOP ratio for health services for the older adults over 45 years old in China was 55%, which was higher than the residents OOP 30% in 2016, 31 and much higher than the national average of 18.6% published by WHO in 2015. Therefore, in order to truly provide the economic protection of the NRCMS for rural middle-aged and older people over 45 years old in China, it is necessary to give further policy coverage to chronic diseases, hospitalized patients, and high-risk families over 65 years old by increasing the reimbursement rate and expanding the coverage of benefit packages, which not only involves direct medical expenses but also considers the compensation for indirect medical expenses such as time loss due to long-term care of chronic diseases. Secondly, the design of the NRCMS does not give enough coverage to people in lower socio-economic groups, which is one of the reasons for the poverty of the disabled or illiterate households with lower capacity to pay. The increase in education levels reduces the incidence of CHE by 2.9%, and the ability to pay increases with the increase of education level. The rate of IME decreases with the increase in education level.
0% of the high-education groups in universities and above also showed the protective effect of education on poverty caused by disease. Compared to ordinary families, disability increases the risk of CHE by 4.5%, and the disabled households had a weaker capacity to pay, and their incidence of CHE is higher than that of uninsured families (27.64%). At least 500 million of the world's 650 million people with disabilities are the poorest. 32 Disability is a complex situation that affects not only an individual but also their family, and it reduces the income available to individuals and families. 33 The risk of poverty due to illness decreases as the size of the family increases and the rate of IME in a family of one person reaches 9.24%. The risk of CHE in a larger family is also reduced by 2.9%.
Therefore, based on our summary of the key aspects of the failure of the NRCMS, it is found that the crux of insurance failures lies in the severe problem of identification and locking of the poor.
The NRCMS did not accurately identify and lock the characteristics of the poor in terms of system design. As an important area under China's precision poverty alleviation, health poverty alleviation constitutes an institutional means to promote health and poverty alleviation through the integration of basic medical insurance, major medical insurance, medical assistance, and bottom-up protection.
However, health alleviation only builds on the basis of the original poor population and ignores other potentially poverty-stricken people. Measuring the poverty based on income seems to be the most common, but based on the above research, it is found that the rural poor are not only originally economically disadvantaged, but also those with high demand for many health services, families with low economic and educational factors, and low risk of mutual aid, which may cause the un-poor family into a poverty-stricken situation because of the cost of medical care. In the process of connecting the basic medical system with medical assistance, the NRCMS is of considerable significance to the accurate docking of the poor to maximize the effectiveness of health poverty alleviation.

Conclusion
As the basic medical insurance, the NRCMS must accurately and comprehensively target the characteristics of the poor in the primary stage, comprehensively cover all the vulnerable people who are at risk due to the disease, and minimize the economic burden of the disease in the basic medical insurance stage, which can reduce the risk of people living in poverty due to illness. For those who are still in poverty and cannot afford high medical expenses, further in-depth protection is provided http://www.wordbank.org/en/topic/poverty/overview.     Figure 1 CHE and IME under different insurance schemes