The Efficiency of Primary Health Care Institutions in the Counties of Hunan Province, China: Data from 2009 to 2017

This study aimed to estimate the efficiency and its influencing factors of Primary Health Care Institutions (PHCIs) in counties in Hunan Province, China, and put forward feasible suggestions for improving the efficiency of PHCIs in Hunan Province. We applied the Input-Oriented Data Envelopment Analysis (DEA) method and the Malmquist Index Model to estimate the efficiency of PHCIs in 86 counties in Hunan Province from 2009 to 2017. Then, the Tobit model was used to estimate the factors that influence the efficiency of PHCIs. Since the implementation of the new health-care reform in 2009, the number of health resources in PHCIs in Hunan Province has increased significantly, but most counties’ PHCIs remain inefficient. The efficiency of PHCIs is mainly affected by the total population, city level, the proportion of health technicians and the proportion of beds, but the changes in per capita GDP have not yet played a significant role in influencing efficiency. In the future, the efficiency of PHCIs should be improved by increasing medical technology skills and enthusiasm of health technicians and by improving the payment policies of medical insurance funds.


Introduction
Achieving universal health coverage (UHC) is one of the targets of the World Health Organization (WHO). Primary Health Care (PHC) can meet the majority of a person's health needs over the course of their life and health systems with strong PHC are needed to achieve UHC [1]. In China, Primary Health Care Institutions (PHCIs), which are the providers of PHC, not only undertake general disease diagnosis and treatment but also undertake basic public health services [2]. The current hospital-centered health service system consumes a large number of health funds, but it cannot meet the ever-changing needs of people and does not have long-term sustainability [3]. Compared with hospitals, PHCIs are more suitable for the prevention and control of major diseases, reducing the incidence of diseases at the source and improving the health of residents. PHCIs in China consist of community health service centers, community health service stations, township health centers and village clinics [4]. Since the new health-care reform, initiated in 2009, China has introduced a series of policy measures in the areas of medical insurance policies, hierarchical medical systems and general practitioner systems to improve the quality of PHCIs' services and guide residents to first visit PHCIs [5][6][7].
In the past nine years, the new health-care reform has made some progress, but there are also some problems [8]. According to the Statistical Communique on the Development of China's Health Care Industry, the number of beds and the number of health technicians in PHCIs have increased by 38.98% and 36.66% from 2009 to 2017, respectively. However, their proportion of all medical institutions has decreased by 5.65% and 5.25%, respectively. Meanwhile, the number of outpatients/inpatients in PHCIs accounted for 61.75%/31.01% of the total outpatients/inpatients in 2009, yet decreased to 54.16%/18.21% in 2017 [9,10]. Affected by factors such as geography, the economy and policies, the provision and utilization of PHC in China varies from region to region, and between urban and rural areas [11][12][13]. With the construction of China's county medical alliances, it is extremely important to improve the efficiency of PHCIs in the counties. Therefore, it is necessary to measure the efficiency of PHCIs, which could provide a policy basis for the health administration departments, and better guide the distribution and utilization of health resources.
Data Envelopment Analysis (DEA) is a non-parametric approach that uses linear programming to build a piece-wise linear-segmentation efficiency frontier based on best practice [14]. DEA has become an effective tool for measuring the efficiency of health care since the mid-1980s, and has been widely used in PHC over the past two decades [15,16]. Scholars usually evaluate PHC from two perspectives-institutions and regions-and some scholars evaluate the specific services, such as oral health care in PHC [17][18][19]. In China, studies from the perspectives of institutions and provinces have shown that the efficiency of PHCIs is low [20][21][22][23]. However, there are few studies that have evaluated the efficiency of PHCIs from the perspective of counties.
In this paper, we focused on the efficiency of PHCIs in 86 counties in Hunan Province, China. Hunan Province is located in central China and has 18 county-level cities, 61 counties, seven autonomous counties and 36 municipal districts. Its population was 68.89 million in 2018, of which the urban population accounts for 56.02% [24]. In recent years, Hunan Province has been committed to the construction of PHCIs to meet the health needs of the residents. While much has changed in Hunan's PHCIs, little is known about the efficiency. This article studies the service efficiency and its influencing factors of PHCIs in Hunan Province from 2009 to 2017 in units of counties, and puts forward feasible suggestions for improving the efficiency of PHCIs in Hunan Province. The main objectives of this paper were: (1) To measure the efficiency of Hunan's PHCIs during the new health-care reform from 2009 to 2017; (2) to evaluate changes in productivity during the new health-care reform; (3) to determine the influencing factors of Technical Efficiency (TE); and (4) to make feasible suggestions for improving the efficiency of PHCIs.

DEA
DEA is a nonparametric method for examining the relative efficiency of similar decision-making units (DMUs) with multiple inputs and outputs. It was first proposed by Charnes, Cooper and Rhodes in 1978 [25]. There are two important assumptions in the efficiency measurement of DEA [26]. One important decision to make when performing DEA is whether to use an input-orientation or output-orientation. The input-oriented model holds the current level of output constant and minimizes inputs, whereas an output-oriented model maximizes output while keeping the number of inputs constant. Another important theoretical hypothesis in DEA is whether the return to scale is constant or the returns to scale are variable. The first nonparametric models for efficiency estimation by Charnes, Cooper and Rhodes (CCR) assumed constant returns to scale (CRS) [25]. Later on, Banker, Charnes and Cooper (BCC) incorporated variable return to scale (VRS) to account for firms, which do not operate at their optimal scale [27]. At present, the choice of medical institutions in China depends mainly on patients and it is difficult to estimate the need for health services. Moreover, the government controls the input more than the output. Therefore, this paper chose to use the input-oriented BCC model to measure the annual TE from 2009 to 2017. The input-oriented BCC model is presented below: This relies on the fact that h ≥ 0 will be satisfied when the components of every X j and Y j are all nonnegative-as is the case for the observational data we are considering. This is a linear programming problem, the duality of which can be written as:

Malmquist Index Model
The Malmquist index model is a non-parametric DEA model based on the productivity index research method first proposed by Malmquist. In 1982, Caves et al. first introduced the Malmquist Index (MI) as an efficiency index into the field of production analysis [28]. The Malmquist index model measures the change in total factor productivity change (TFP) of DMUs over time. The TFP is divided into two parts: Technical change (TECHCH) and technical efficiency change (EFFCH). The technical efficiency can be further decomposed into the product of pure technical efficiency change (PECH) and scale efficiency change (SECH). The input-oriented MI consists of four input-oriented distance functions that functionally represent multiple-input and multiple-output production techniques. They can be used to characterize efficiency because the distance function is the reciprocal of the technical efficiency measure proposed by Farrell [29]. The distance function is a natural way of modeling production boundaries because the deviations and offset from this boundary represent changes in efficiency and technology, respectively. This paper uses the Malmquist index model to measure the efficiency changes in Hunan Province from 2009 to 2017. The change in productivity between period t (base period) and period t + 1 is defined as: where M 0 is the input-oriented Malmquist productivity index, and y represents the output vector that can be produced using input vector x. Among them, EFFCH = , respectively, indicate that efficiency improvement and technological progress occurred in the period t to t+1. Therefore, TFP can be expressed as follows: TFP = EFFCH × TECHCH. If the TFP > 1, indicates that the TFP of the t+1 period is higher than that of the t period; a TFP < 1, indicates a decrease in TFP; and TFP = 1 indicates that TFP is unchanged.

Tobit
In this study, the TE obtained by DEA is the dependent variable, and the main factors affecting the TE of PHCIs in the county are explored by constructing regression models. Since the efficiency value measured by DEA ranges from [0, 1]; that is, the value less than 0 and greater than 1 is punctured, the dependent variable with punctured feature belongs to the restricted dependent variable. Estimating the regression parameters using a regression model based on the ordinary least squares method is likely to cause an estimation bias due to incomplete data. In view of this, Tobin first proposed the puncturing regression model, or the Tobit model, in 1958 [30]. The formula for the Tobit regression equation is the following: where Y i is the TE value, X i is the explanatory variable, and β i is the coefficient to be evaluated.

Data and Selection of Variables
Considering the integrity and availability of data, this study included 86 counties (county-level cities) in Hunan Province from 2009 to 2017; one county was not included because of partial missing data. The data were collected from the National Health Statistical Information Report System during the period 2009-2017, and the Hunan Statistics Yearbook.
In this paper, the input and output variables were selected by reviewing the relevant literature and based on the availability of data in the National Health Statistical Information Report System [16,[31][32][33]. Regarding the input variables, both human resources and facilities were considered important for health care services. We selected the number of health technicians instead of human resources, and took the number of beds and the amount of equipment valued at more than 10,000 RMB to replace the facilities. The output indicators selected in this paper are the number of outpatients and emergency visits, and the number of discharged patients (refer to the standard explanation, Table 1). In the Tobit regression analysis, influencing factors that affect PHCI's efficiency indirectly were the total population, city level, per capita GDP, the proportion of health technicians, the proportion of beds (refer to the standard explanation, Table 1).
The input and output variables were descriptively analyzed using SPSS 22.0 (International Business Machine, New York, America). The software DEAP 2.1 (University of New England, Armidale, Australia) was used to calculate the value of Technical Efficiency (TE) and Total Factor Productivity (TFP) and used the software STATA 12.0 (StataCorp LLC, Texas, TX, USA) for Tobit regression analysis. Table 2 shows the statistics of the input and output variables.

Results of the DEA Model
As shown in Table 3

Results of the DEA Model
As shown in Table 3, for the

Results of the Malmquist Index Model
The Malmquist index model was applied to analyses on the changes in productivity over the 2009-2017 period, and the year 2009 was taken as the technical reference. Table 4 presents the Malmquist index summary of annual geometric means. On average, TFP decreased by 3.1%, among which, TECHCH decreased by 6.5% and EFFCH increased slightly by 3.7%. During the period 2009-2017, 24 (27.91%) counties had TFP score greater than 1, indicating growth in TFP; 61 (70.93%) counties had TFP score less than 1, indicating a deterioration in TFP. Figure 2 shows the distribution of TFP in Hunan Province. The most efficient counties are Xiangyin County (30) and Lengshuijiang City (77). The TFP of each county is shown in Table A1, Appendix A.

Results of Tobit Regression
In this article, Tobit regression was used to analyze the influencing factors of TE during 2009-2017. The results are presented in Table 5. In order to reduce heteroscedasticity, we have performed a logarithmic conversion on the two independent variables of population and per capita GDP. Regarding environmental factors, Ln (total population) (p < 0.001), city level (p < 0.001), the proportion of health technicians (p < 0.001), the proportion of beds (p = 0.002) exhibited significance, indicating that they are significant in influencing the efficiency of PHCIs. However, Ln (per capita GDP) (p = 0.977) was statistically insignificant.

Results of Tobit Regression
In this article, Tobit regression was used to analyze the influencing factors of TE during 2009-2017. The results are presented in Table 5. In order to reduce heteroscedasticity, we have performed a logarithmic conversion on the two independent variables of population and per capita GDP. Regarding environmental factors, Ln (total population) (p < 0.001), city level (p < 0.001), the proportion of health technicians (p < 0.001), the proportion of beds (p = 0.002) exhibited significance, indicating that they are significant in influencing the efficiency of PHCIs. However, Ln (per capita GDP) (p = 0.977) was statistically insignificant.

Discussion
Since the new health-care reform in 2009, the government has invested a lot in monetary, equipment, buildings and health technicians in PHCIs, but the growth of these resources has not brought about an increase in the use of PHCIs, residents' choice of health care services still tends to tertiary hospitals [8,35].
In May 2019, the National Health Commission of People's Republic of China issued a policy document requiring the overall performance of the county-level health system and the ability of PHCIs to be improved, and the visit rate of PHCIs in the county is required to reach 65% [36]. However, in recent years, the visit rate of PHCIs has been declining. This paper can provide a reference for improving the efficiency of PHCIs by studying the efficiency of PHCIs in 86 counties of Hunan Province and their influencing factors.
The BCC model is a method used to analyze the efficiency of a single period. This study used the BCC model to measure the annual efficiency of PHCIs in Hunan Province each year from 2009 to 2017. It was found that the average TE of PHCIs in Hunan Province in the past nine years was lower than 0.8 and most counties were ineffective, which indicates that the efficiency of PHCIs needs to be improved. Besides, the TE and the counties with effective TE both increased from 2009 to 2015 but declined in 2016. While we cannot know the trend of efficiency after 2018, a potential explanation is that the dissolution and consolidation of PHCIs in Hunan Province during 2015-2016 that affected the continuity of PHC and according to the general choice of patients and increasingly convenient traffic conditions, patients who previously went to the revoked PHCIs may not go to the retained PHCIs, but are more likely to a secondary or tertiary hospital [37][38][39]. Compared to SE, TEvrs was lower, which indicates that while improving the scale of PHCIs, we must pay more attention to improving efficiency by improving the medical technology of PHCIs. Compared with other studies using the same method, the average TE of PHCIs in Hunan Province was higher than that in the Community Health Centers of Jiangsu, which is 0.15-0.40 [40], but lower than China's total medical services efficiency, which is 0.904 [41]. This difference may be due to different regions or research subjects.
The Malmquist Index Model measures the dynamic efficiency change of the DMU over multiple periods. The results of the Malmquist Index Model verified that PHCIs in Hunan experienced a significant decline in TFP from 2009 to 2017, which was consistent with the results of Zhang et al. [23]. The trend of TECHCH and TFP was consistent, so it can be considered that the change of TFP was mainly affected by TECHCH changes, which is consistent with the findings of Leng et al. [4]. Chen et al. pointed out that technological progress determines the key factors of the TFP of community health services in China [42]. The main reason for the low TECHCH is the lack of excellent health technicians, and the underutilization of medical equipment in PHCIs.
Some studies indicate that health resources in areas with low economic levels can be better utilized, which may increase efficiency [18,43,44]. In this study, the Tobit regression results indicated that the impact of the per capita GDP on the growth of TE of PHCIs is not significant. This is consistent with the findings of Leng et al. who found that this is mainly due to the guiding role of government policies and the implementation of health-care reform policies that have narrowed the gap in regional health resource allocation [4]. Therefore, the level of economic between regions has a small impact on TE.
The positive impact of the total population on TE in this study is consistent with the findings of Tan et al. [45]. The number of permanent residents at the end of the year reflects the potential health care service needs in the region, which will affect the efficiency of the hospital. In addition, the increase in China's population is accompanied by the implementation of the two-child policy and the development of an aging population, which has also increased the demand and utilization of health services [46,47]. However, research by Chen et al. suggested that a large population will increase the burden on the hospital and make health care service inefficient, and this may be caused by different research objects [48].
This study shows that the city level has a negative impact on TE. At the same time, Zhang et al. showed that the impact of urbanization on the efficiency of PHCIs in the rural and urban areas is different [49]. With the advancement of urbanization in China, the number of poor people has decreased. Zhang et al. showed that wealthy people are more likely to use hospitals with adequate resources, and poorer people are more likely to use poorly funded PHCIs for health care services [50]. Meanwhile, due to the relatively concentrated population of urbanization and the convenience of modern transportation, which has further improved the geographic accessibility of health care services, residents may be more inclined to seek higher quality health care services at secondary or tertiary hospitals, which has a negative impact on the efficiency of PHCIs [51].
While the new health-care reform has increased the number of health technicians and beds in PHCIs, TE has decreased as their proportion increased. In the long run, a lack of qualified health technicians and inadequate use of medical equipment may reduce technical efficiency [52]. PHCIs still have problems such as severe maldistribution of quality human resources, low education of health technicians and the detrimental elements of health-care reform policies [53]. These problems have led to a lack of patients' trust in the quality of PHC. Therefore, patients' demands for high-quality health care services make the gradient design of the health insurance reimbursement rate insufficient to guide patients to PHCIs for treatment [20,54]. Meanwhile, due to the small development space, low wages and lack of sufficient performance incentives of PHCIs, the enthusiasm of health technicians has been affected, limiting the improvement of PHCIs' efficiency [5,55,56]. After the implementation of the new health-care reform, the government provided a lot of advanced medical equipment for PHCIs, but the health technicians in the PHCIs lacked operating skills and could not use the equipment well [57,58]. As a result, a large amount of equipment and beds were idle, which hindered PHCIs in improving efficiency.
This study conducted a horizontal and vertical analysis of the efficiency of PHCIs from 2009 to 2017 from the perspective of the county, and explored its influencing factors. However, there are some limitations. First, the input-output variables and influencing factors in this paper were selected based on the literature and the availability of data, which may lead to deviations in the research results. Second, the TE of PHCIs may be affected by other factors not considered in this article, such as government health expenditure, the education level of the residents, life expectancy, et al. Third, bias adjustments of TE and TFP scores were not carried out due to the limitation of the basic DEA approach. Fourth, due to the availability of data, we only use the data from 2009 to 2017 for analysis, which cannot be compared with the efficiency of PHCIs in other periods. Despite these limitations, this study can be considered a useful attempt to measure the TE and TFP of PHCIs at the county level in China.

Conclusions
In this study, we utilized a panel dataset of 86 county-level units of Hunan during the period 2009-2017 to describe the allocation of health resources and the provision of health care services in PHCIs, to compare the efficiency of the regional primary health care systems, and to explore the determinants of TE. The main conclusions of this study were as follows: Firstly, the efficiency of PHCIs in Hunan Province during 2009-2017 was low, and most counties are inefficient. Secondly, the results of the Malmquist index showed that the PHCIs in Hunan Province experienced significant decline in TFP from 2009 to 2017, which mainly resulted from the low TECHCH. Third, the factors affecting the TE of PHCIs are the total population, city level, the proportion of health technicians and the proportion of beds in PHCIs.
According to the conclusions, we propose the following suggestions: First, we need to improve the medical technology skills of health technicians, strengthening the training of clinical skills, health management skills and equipment operation skills of health technicians, and creating policies on job title and continuing education could help to attract excellent health technicians to work in PHCIs. Second, improving the performance policy of PHCIs would allow PHCIs to independently determine internal performance wages and distribution methods, and increase the proportion of incentive performance wages on the premise of guaranteeing the payment of basic wages, thereby enhancing the enthusiasm of health technicians. Third, improving the payment policy of medical insurance funds, by increasing the total annual budgetary payment of PHCIs in the county medical insurance fund. The increase should be higher than the evaluation increase of the total annual budget payment of local medical institutions, and should shape the medical insurance payment policy to guide residents who have common diseases or frequently occurring diseases to go to PHCIs for health care service.
Author Contributions: Conceptualization, L.C. and K.Z.; methodology, K.Z.; validation, S.C., K.Z. and H.C.; formal analysis, K.Z., S.C. and F.L.; resources, H.C. and F.L.; data curation, F.L.; writing-original draft preparation, K.Z.; writing-review and editing, L.C. and S.C.; supervision, L.C.; project administration, L.C.; funding acquisition, K.Z. All authors have read and agreed to the published version of the manuscript. Acknowledgments: Thanks to Miss Xinyin Wu for suggestions on this article. Thanks to the Health Commission of Hunan Province for providing data.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.