1 Introduction

Medical accessibility can be defined as an individual or population’s capacity to enter the health system. It has spatial, organizational, and financial dimensions (Mudd et al., 2019). China gradually expanded financial support for medical accessibility through Citizenship Medical Insurance Programs and new Rural Cooperative Medical System. All provinces in China planned to establish Settlement of Non-local Medical Treatment within province, which meant individuals would have accessibility to hospital outside local area. Since national Settlement of Non-local Medical Treatment was based on provincial version and still on further planning, currently focus could be laid on provincial scope, not national scope. For people were not restricted only to local county hospital, higher attention needed paying to other types of accessibility like spatial accessibility while recognizing medical-shortage area.

Since 1980s, the use of spatial accessibility to medical resources has gained prominence and is applied extensively in international policy and research agendas despite being criticized for not considering neighboring effect enough (Radke & Mu, 2000). To address this, researchers and decision-makers increasingly developed Two-step Floating Catchment Area (2SFCA) and its various extensions (Luo & Wang, 2003; Wang, 2012). Although some accessibility analyses had been made in township unit in certain urban areas in China (Zhong et al., 2016), yet on provincial scope, medical accessibility and its disparity in township unit needed further study to provide medical facility at large to small scale on urgent bases to secure life and repose the disaster. Further, the pervious studies consider a larger area which may ignore the area characteristics at local level, where the scenarios are different and varied from larger to smaller area.

To aid policy making for reducing inequalities in spatial accessibility to medical resources after realization of Settlement of Non-local Medical Treatment in provincial China, this study was designed to show how spatial accessibility of township to medical resources distributed in provincial China after considering travel time to potential non-local medical resources. While analyzing medical accessibility of township on provincial scope, for large gap between urban and rural China, we also tried to discover differences of accessibility of township to medical resources in special age groups and among urbanization groups in provincial China. Based on distribution of accessibility and different population groups, this study further explored which townships had poor spatial accessibility and high medical demand measured by age and urbanization.

Compared to the previous spatial accessibility research, our study had the following strengths: (1) in E2SFCA computation, we introduced real travel time from township to county hospital under help of China’s leading online map supplier AutoNavi, and expected to provide a more accurate measurement of travel cost from township to medical resource; and (2) based on spatial-overlay method in prior research, integrated with cold and hot spot analysis, we provided more specific and accurate information on age and urbanization characteristics of township with poor spatial accessibility. The contribution made by this study is provision of real time travel from township to countries’ hospital with respect to spatial analysis and age groups in urban characteristics and time base analysis.

The paper was organized as follows. First, we briefly reviewed the literature on E2SFCA and the literature on recognition method and scale of medical-shortage areas. Then, we presented the study area and methods used. Third, we moved to our results including (i) global distribution of township accessibility, (ii) disparities of accessibility in and among demographic and economic groups, and (iii) recognition of medical-shortage townships in different types. In the discussion, we argued that the large gap between different social groups was noteworthy and needed efforts to address. A conclusion wrapped up the paper.

Generally, there were still some research gaps. Firstly, though some previous studies focused on medical spatial accessibility of township in China, yet the case regions were just certain urban areas and few were on provincial scope, so they might ignore townships in rural regions. Secondly, province, as primary unit of medical resource allocation in China, was facing medical resource allocation by patient population from both local and non-local counties and not only by patient population from local area. So medical spatial accessibility of township on provincial scope was helpful for medical resource allocation in province during application of Settlement of Non-local Medical Treatment within province. Thirdly, previous research found different medical demand from different population groups, i.e., as for one hospital bed, whose equipment would be different for the aged and children, yet they tended to integrate these differences into single medical demand index during calculation of spatial accessibility using E2SFCA. If medical-shortage townships were recognized by population group, medical resource supply would be higher pertinent to different demand from special population group.

1.1 Literature review

The application of spatial accessibility to medical resources can be traced back to Joseph, who proposed an improved gravity model in Hansen’s gravity model for accessibility calculation (Hansen, 1959; Joseph & Bantock, 1982). This model later became the main model for computing accessibility to medical resources. Based on gravity model, service demand and supply measurement, and the space decomposition method, 2SFCA was proposed (Luo & Wang, 2003). Since then, scholars have made further improvements in modeling supply and demand of medical resources and the distance function on different scales (Zhong et al., 2016; Apparicio et al., 2017; Paez, Higgins, and Vivona 2019).

1.1.1 Scale

The geographical variations in health existed at many scales, from global to local (O'Boyle, Henly, and Larson 2001). Neighborhoods or contexts to express health-related measures could be defined in different ways. However, in any way, because all persons residing within the same administrative district -socio-economically different as they may be- were assigned equal levels of accessibility, accessibility measured at different level of administrative districts might obfuscate micro-level accessibility problems of social groups and neighborhoods (Omer, 2006), and ignore important sources of social inequality including income barriers (Wang & Luo, 2005), diverging health needs (Roeger et al., 2010) and rural–urban type (McLafferty et al., 2011). Despite there were a few scholars trying to use individual’s healthcare data to measure accessibility and aggregated effect of different scales (Cabrera-Barona et al., 2018), the most practical way was to use minimum aggregated unit-census area such as census block and census tract (McGrail & Humphreys, 2009; Shah et al., 2017).

Yet, for China, the largest developing country, for data availability and processing capacity, analysis based on the smallest administrative unit: towns and streets (hereinafter referred to as townships), mainly focused on urban areas of large cities, such as Taipei and Chengdu (Song et al., 2019; Tang et al., 2017), or a typical county such as Xiantao, Deqing (Liu et al., 2017; Luo et al., 2017). There were few studies on the accessibility of township to medical resources in state-level administrative region. Wang et al. (2018) took Sichuan Province, China as an example to study disparity of accessibility to primary health care between different counties and population groups. Although they measured spatial accessibility of county to medical resources on the unit of 2 km*2 km population grid, yet they averaged on county unit to match social development indexes of county (GDP, ethnicity, et al.). So average accessibility on county-level might obscure smaller area variation in accessibility on township-level and was hard to provide a sufficiently robust description of how access to health care was distributed among small administrative units (Kwan & Weber, 2008). And it might be hard to make clearer allocation of medical resources which generally was based on administration unit. Besides, this study did take road networks into account, yet seem to miss using matured commercial online map to obtain travel time considering real traffic condition.

1.1.2 Identification medical-shortage areas

Single index and multiple indexes methods were the 2 main ways of identifying medical-shortage areas. The single index could be classified further into 2 categories: absolute and relative standards. Since spatial accessibility based on 2SFCA was measured by the rate of supply and demand weighted by distance, some scholars directly used standard physician-to-population ratio (PPR) to identify areas with poor medical resources; Daly et al. (2019) identified area with a PPR of less than 1:3500 as a medical-shortage area. Absolute standard lacked flexibility and could not be adjusted in time about regional differences. For accessibility based on 2SFCA, results calculated by different distance functions or parameter settings differed to an appreciable extent. By adopting absolute standard, results of recognition would be ambiguous, so most research chose a relative standard. For example, Gautam et al. (2014) applied 3 accessibility methods to 3 kinds of regional units to form 9 accessibility-calculation results and obtained an average value of regional accessibility using the weight—population proportion of each unit, which was used as a threshold value to identify medical-shortage areas. Furthermore, some scholars had used principal component analysis or an entropy method to calculate an integrated deprivation index of medical resources, weighted by non-spatial and spatial factors, respectively, to measure the degree of lack of medical resources for each unit.

Multiple indexes derived from measurement of medical demand when some researchers paid special attention to the needs of different population groups. Recent progress in this field had been in identifying medical-shortage areas by integrating more non-spatial factors, i.e., multiple indexes, such as socio-economic status. The most common synthesis method of multiple indexes was spatial overlay. Typically, Gatrell and Wood (2012) identified medical-shortage areas by the spatial overlay of units in 3 different types: Firstly, they identified statistical units with accessibility to one hospital for more than half an hour. Secondly, they used cancer mortality as a proxy variable for medical needs and filtered out units above the average cancer mortality rate. Thirdly, they recognized units with a poverty index above the average. Finally, they overlaid these 3 types of units, determining common parts as medical-shortage areas. Similarly, in census tract-based unit in Georgia, US, Yin (2019) set a unit which had both a standard deviation of accessibility below the average and a standard deviation of health-resource demand above the average as medical-shortage area. Generally, scholars select average accessibility as threshold to recognize medical-shortage areas, yet as to medical demand, different age groups and physical status groups may have different medical demands because different major illnesses exist in these groups (Zhi & Xu, 2018). Though in total quantity, medical supply is no less than demand, yet the supply may fail to satisfy diversified demand of these population groups.

2 Methods

2.1 Study area

We choose Anhui Province, China as the study area, which is located in central China (Fig. 1). The main reasons to select this area were as follows. Firstly, the plain and mountainous regions of Anhui provide more insights to China. The northern portion of the province is occupied by the North China Plain as an immense level surface that periodically has been flooded by its northern major Huai River of China. The southern section of the province, the Yangtze River valley, is separated from the northern plain by a series of mountains that stretch roughly from west to east. The Dabie Mountains as an eastern extension of the Qin (Tsinling) range lying to the north of the Yangtze form a convex curve of steep slopes facing east and northeast on the southwestern Hubei-Anhui border. The Baiji Mountains lie south and east of the Yangtze and extend to the southeastern border between Anhui and Zhejiang, including the Huang Mountains which rise to a height of about 1,800 m and have become one of the most popular tourist destinations in China for their massive shapes and lush vegetation.

Fig. 1
figure 1

Location of Anhui Province, China

Full size image

Secondly, for high outmigration intensity, townships in rural Anhui were suffering severe imbalance of age groups, i.e., a large quantity of left behind children and elders. Though Anhui is geographically closer to China's developed eastern seaboard, it shares with inland provinces low levels of economic development, a large volume of surplus rural labor, and a relatively long history of labor out-migration (Wang & Fan, 2006). In 2018, gross domestic product per capita (GDPP) was 47,712 Yuan, much lower than the national level of 64,644 Yuan, ranking it in the 22nd place among 31 provincial administrative districts in mainland China (National Bureau of Statistics, 2019). Coupled with poor economic development, Anhui also had low urbanization rate and high outmigration intensity. In 2018, with population of 63.24 million, its urbanization rate was 54.69%, much lower than 59.58% of the whole nation, being the 2nd last in 6 central provinces (National Bureau of Statistics, 2019). As a result of rural–urban outmigration in recent decades, Anhui’s rural demographics had transformed dramatically, especially in imbalance of gender and age groups (Dong, 2019), e.g., Gini coefficients for a proportion of age group 0–14 in total population of township were 0.15 in 2010, higher than other 5 central provinces, ranking the 10th in state-level China.

Thirdly, Anhui had limited medical resources for poor economics and faced special challenges in satisfying the medical demand of left behind children and elders in rural. In 2018, the numbers of health technicians, doctors and assistant doctors, registered nurses, medical institutions, and hospital beds per 1000 population were reported to be 5.27, 2.01, 2.37, 0.39, and 5.19 in Anhui, 1.55, 0.58, 0.57, 0.32 and 0.83 lower than national average, ranking the last, 2nd last, 3rd last, 3rd last and 6th last in 31 provincial administrative districts in mainland China (National Bureau of Statistics, 2019). Besides, there were great disparity of medical resources among prefecture-level cities in Anhui. Taking Gini coefficient for hospital beds per 1000 of prefecture-level city (except for provincial capital) by province in 2017 as an example, it ranked Anhui in the 11th in total 28 provinces and in the 2nd in 6 central provinces.

Fourthly, Anhui was a pilot province in rounds of health policy reforms for its poor medical conditions. Rapid changes in industrial and agricultural sectors had taken place in Anhui province over recent years, along with various kinds of health and environmental issues, e.g., Anhui was one of the provinces with a high incidence of birth defects in China (Tao et al., 2013). Because of poor equality of medical resources, when China started the new round of medical reforms in 2009 and the reforms of county-level public hospitals in 2012, Anhui was one of the 4 pilot provinces of the reform (Li, Vo, et al., 2017; Li, Wang, et al., 2017). In 2019, Anhui realized settlement of Non-local Medical Treatment within province. As most less-developed provinces had similar inequality of medical resources, we selected Anhui Province as the focal point like most studies and expected that findings in Anhui would echo, if not represent, observations in other provinces in similar situations.

2.2 Background of the study

After Settlement of Non-local Medical Treatment within provincial scope, population could be freer to visit non-local county hospital, so travel time would be more important in decision of hospital. E2SFCA was employed for considering travel time by introducing distance-decay function in the calculation of 2SFCA. Centered on spatial accessibility of township, regression model was taken to quantitively analyze disparity of different age groups and urbanizing groups. For showing townships on different severe level of medical shortage while suggesting policymakers, cold and hot spot analysis was selected to identify low medical accessibility townships and high medical demand townships on different significance levels.

2.2.1 Accessibility calculation based on E2SFCA

E2SFCA mainly used the distance-decay function in the calculation of 2SFCA and still followed two steps to compute accessibility. For county hospital was top hospital in one county mainly dealing with chronic and severe diseases, referred to Pfeffer (1973), number of beds in a county hospital was used to present medical resources.

The first step was to calculate beds-to-population ratio, Rj, for county hospital j, as in formula (1):

$${R}_{j}=\frac{{S}_{j}}{{\sum }_{i\in \mathrm{D}}{e}^{-\frac{{d}_{ij}^{2}}{\upbeta }}{P}_{i}}$$
(1)

where Sj was the number of beds in a county hospital j; Pi was population of township i; D was the county hospital service radius; dij was the minimum travel time from township i to county hospital j; β was the friction coefficient.

The second step was to calculate accessibility Ai of township i, as in formula (2):

$${A}_{i}={\sum }_{j\in \mathrm{D}}{e}^{-\frac{{d}_{ij}^{2}}{\upbeta }}{R}_{j}$$
(2)

where dij, D, and β were the same variables as in formula (1); Rj was beds-to-population ratio for county hospital j.

Some research highlighted that the friction coefficients were different in city and rural areas, the range of city being usually 0.9–2 (Song et al., 2016), yet Gharani et al. (2015) found the best value of the friction coefficient to be 0.705, via regression analysis of data in Idaho, USA. Since township population on provincial scope was considering here, so friction coefficient was set to be 0.7. Similarly, the service radius setting needed to consider the urbanization status and hospital grade. In urban areas, since the medical-resource density is high, the service radius is small; while in rural areas, the medical-resource density is low, and the service radius is large. Regarding hospital grade, when the grade is low and suitable for minor diseases, distance greatly influences the choice of medical resources, resulting in a small service radius. In contrast, when hospital grade is high, dealing with serious disease, distance has little influence on hospitalization. Generally, the service radius of the secondary hospital was in 0.5–1.5 h and that of the tertiary hospital was over 1.5 h (Zhong et al., 2016). As county hospitals were generally in top grade and serving population from both urban and rural areas, so the service radius was set as 2 h.

2.2.2 Measuring spatial accessibility to county hospital

Based on the China County Statistical Yearbook (volume of township) 2016, the population of townships in Anhui Province were adjusted according to the total population of counties in the Bulletin of the 2015 Population Sample Survey of Anhui Province. The location of a township, set as the position of the township administrative office, was from the website http://www.3edata.com/, corrected by AMap and the administrative territorial entity of the People’s Republic of China. Totally 1,601 townships with population data were analyzed.

In China, county hospital still sits in the core of urban and rural medical linkage and is the most important medical resource for rural population. Medical supply was measured by the total number of hospital beds in urban area of one county, including those in maternal and child health care hospitals and special disease prevention and treatment institutes as defined in the Anhui Statistical Yearbook 2016. Since our research scope was provincial and analysis unit was township, considering hard to obtain habituated country population in each township, population of township was assumed to wholly reside at township administrative site. This analysis mainly concerned accessibility of township to county hospital in urban area of one county in a provincial context, so it was acceptable to set location of the top hospital in each county (generally the People’s Hospital of the county hereafter represents the county hospital) as medical-resource location in whole urban area of one county. A total of 103 county hospitals were finally included in the analysis.

Travel time by car from township to county hospital in general working hours was obtained using AutoNavi Application Programming Interface of AMap, in the form of an origin–destination (OD) matrix. Age-grouped population data of townships were from the 2010 China Population Census, and urbanization status data of townships were from the China County Statistical Yearbook (volume of township) 2016.

2.2.3 Accessibility differences in special age groups and among urbanization groups

To examine whether there was significant accessibility difference in special age group, based on model in Ikram et al. (2015), a weighted regression model was proposed in formula (3):

$${Y}_{ik}={a}_{\text{k}}+{b}_{\text{k}}*Fla{g}_{ik}, Fla{g}_{ik}=\left\{\begin{array}{c}0 {A}_{i}\le {\sum }_{i=1}^{\mathrm{n}}\frac{{A}_{i}}{\mathrm{n}}\\ 1 otherwise\end{array}\right.$$
(3)

where Yik was percent of age group k population in township i; Flagik was 0 when accessibility of township i was below the average accessibility of case region and 1 otherwise; Ai was accessibility of township i; n was the number of townships in case region; ak was mean percent of population of age group k in townships with Flagik being 0; bk was the difference between Yik when Flagik was 1 and Yik when Flagik was 0. In regression, the weight of population of this age group was introduced to give a greater weight to the error items of the more populous township.

It was meaningful to investigate geographic variation and urban–rural gradient in accessibility of townships to county hospital in urbanizing China. However, only the area of an administrative region, number of residents, residents in urban area, employees, employees in the secondary and tertiary industries, and total industrial output value were available on township scale in China. Besides, the total industrial output value and number of residents had many zeros. Considering data integrity and representativeness, we finally selected the proportion of secondary and tertiary industry employees to total employees (hereafter referred to as the urbanization level) to characterize townships regarding their rural and urban status. Referring to Walker et al. (2017), we first carried out a logarithmic transformation of accessibility. Then, based on division of urbanization stages in Chen et al. (2011), we classified townships into 3 groups by urbanization level: Above 0.7 (group 1 and reference group), 0.3–0.7 (group 2), and below 0.3 (group 3). Finally, a logarithmic linear model, with accessibility as the dependent variable and dummy variables about urbanization level as independent variables, was established in formula (4):

$${\mathrm{ln}A}_{i}={\mathrm{b}}_{0}+{\mathrm{b}}_{1}{x}_{i1}+{\mathrm{b}}_{2}{x}_{i2}$$
(4)

where Ai was the accessibility of township i; b0 was mean accessibility logarithms of group 1; b1 was difference in accessibility logarithms between group 2 and group 1; b2 was difference in accessibility logarithms between group 3 and group 1; When township i belonged to group 2, xi1 = 1, otherwise xi1 = 0; When township i belonged to group 3, xi2 = 1, otherwise xi2 = 0.

To support spatial targeting for distribution of medical resources, cold and hot spot analysis Gi* was used to identify low medical accessibility townships and high medical demand townships on different clustering significance levels. Its calculation followed formula (5) (Ord & Getis, 1995).

$$\begin{array}{*{20}c} {G_{i}^{*} = \frac{{\sum\nolimits_{j = 1}^{{\text{n}}} {w_{i,j} x_{j} } - \overline{X} \sum\nolimits_{j = 1}^{{\text{n}}} {w_{i,j} } }}{{S\sqrt {\frac{{{\text{n}}\sum\nolimits_{j = 1}^{{\text{n}}} {w_{i,j}^{2} } - (\sum\nolimits_{j = 1}^{{\text{n}}} {w_{i,j} } )^{2} }}{{{\text{n}} - 1}}} }},} \\ {\overline{X} = \frac{{\sum\nolimits_{j = 1}^{{\text{n}}} {x_{j} } }}{{\text{n}}},\,S = \sqrt {\frac{{\sum\nolimits_{j = 1}^{{\text{n}}} {x_{j}^{2} } }}{{\text{n}}} - (\overline{X} )^{2} } } \\ \end{array}$$
(5)

where xj was the accessibility of township j, proportion of population in special age group in township or urbanization level of township j; wij was the spatial relationship between township i and j; n was number of townships participating in the calculation.

3 Results

Figure 2 showed that accessibility of townships decreased with distance from urban. Generally, from north to south, 4 township clusters of high accessibility were Bengbu-Huainan districts, Hefei district, Maanshan-Wuhu-Tongling-Chizhou-Anqing districts and Huangshan district. Townships with second highest were mainly located in Fuyang, Suzhou, Chuzhou, Luan and Xuancheng districts. Townships in Bozhou and Huaibei districts had the third highest accessibility. Townships with the least accessibility could be divided into 2 categories: mountainous townships which were mainly located in Dabie Mountains and mountains represented with Huang Mountain in southern Anhui; and townships with poor transportation which located in the hole of highway network and far from highway.

Fig. 2
figure 2

Distribution of accessibility in whole Anhui

Full size image

Table 1 reported that for the whole Anhui, except for 15–64 age group, percent of the other 2 groups (65 + and 0–14 age) in townships with below-average accessibility were both higher than those in townships with above-average accessibility significantly. It was also evident that the spatial accessibility to county hospital decreased as urbanization level descended. In the whole province, from group 1 to 2 and 3, mean accessibility decreased by 27.39% and 51.32% significantly.

Table 1 Difference of accessibility in special age groups and among urbanization groups
Full size table

The population of aged 65 + or 0–14 had a higher demand for medical resources, so townships with low accessibility to county hospital and high proportion of the population aged 65 + or 0–14 to total population (proportion of the population aged 65 + or 0–14 thereafter for short) were defined as physiological medical-shortage townships. Figure 3a mapped distribution of hot and cold spots of accessibility. Cold spots of accessibility were mainly at the boundaries of northern Anhui, and in the Dabie Mountains in western Anhui and mountainous southern Anhui. Figure 3b and c showed areas with higher proportion of the population aged 65 + were mainly concentrated in middle and mountainous southern Anhui, in sharp contrast to those with higher proportion of the population aged 0–14 mainly distributing in northern Anhui. Figure 3d showed 34 townships, as both cold spots of accessibility and population-proportion hot spots of aged 65 + or 0–14 on 99% confidence (see Online Resource 1: Supplementary Table 1), were mainly distributed in northern and mountainous southern and western Anhui. In detail, 14 townships, as both cold spots of accessibility and population-proportion hot spots of aged 65 + , were in mountainous western and southern Anhui, while 20 townships, as both cold spots of accessibility and population-proportion hot spots of aged 0–14, were located near provincial boundary in northern Anhui.

Fig. 3
figure 3

Distribution of hot and cold spots (a, b and c) for accessibility, proportion of the population aged b 65 + and c 0–14, and d distribution of physiological medical-shortage townships

Full size image

Since rural population were usually unlikely to attract substantial private investment in public infrastructure and more likely to suffer poverty (Ramachandran, 2014), also limited by available economic data of township, urbanization level of township further served as proxy of economic poverty. Like recognition of physiological medical-shortage townships, townships as both cold spots of accessibility and cold or hot spots of urbanization, were defined as economic medical-shortage townships; of these, hot or cold spots of urbanization were defined as economic-affluence or economic-poverty medical-shortage townships. Figure 4a showed distribution of cold and hot spots of urbanization. It was salient that almost all hot spots of urbanization were around the urban areas, and the cold spots were located at the county or provincial boundaries. Figure 4b mapped 13 economic medical-shortage townships, mainly in mountainous western and southern Anhui, and all of them were economic-poverty medical-shortage townships.

Fig. 4
figure 4

a Distribution of cold and hot spots of urbanization level and b recognition of economic medical-shortage township

Full size image

4 Discussion

Spatial accessibility of township to county hospital had great difference between with and without Settlement of Non-local Medical Treatment within province. Based on the Settlement of Non-local Medical Treatment, it was possible for patients to visit a doctor in a non-local county. Maybe because they lived far from their local county hospital, or maybe because another county had a famous specialist, or some other reasons, patients did not have to go to the local hospitals, so each county didn't need to have the same number of health resources per capita. Accessibility equity was defined by need, indicating the more need, the more health resource. If the government allocated more health resource to a county with large population, but most of the population visited a non-local hospital, the local health resource might be wasted. By E2SFCA, this paper considered travel time when population was not confined to only visit their local county hospital and calculated spatial accessibility of township to county hospital not just in local county. In this way, we could allocate more medical resources to those county hospitals serving more actual population not just reside in the same county with those hospitals. Suppose population was confined to only visit local county hospital, the number of beds per capita in the Xiangshan or Sanshan district was lower than those in Guoyang and Xiao counties. In contrast, if population was free to visit non-local county hospital, medical accessibility of Xiangshan or Sanshan district was not lower than those in Guoyang and Xiao counties. The reason was that because of its proximity to large urban area where medical resources existed and could be shared with population of other counties, after considering travel time, actual medical resources including those in nearby urban area were more in Xiangshan or Sanshan district than Guoyang and Xiao counties. If more medical resources were allocated to Xiangshan or Sanshan district rather than Guoyang and Xiao counties, there would be waste. By introducing spatial accessibility using E2SFCA, this waste could be avoided.

Besides, noteworthy gap between different social groups was shown and most significant medical-shortage townships were recognized for policymakers. Particularly, average percent of 65 + and 0–14 age groups in townships with below-average accessibility were both higher than those in townships with above-average accessibility. Since the aged and children demand more medical resources than adults of 15–64 age group. So poor spatial accessibility for the aged and children needed more urgent concerns. For different urbanization groups, medical accessibility decreased greatly along with fall of urbanization level. This reflected large gap of medical resources between urban and rural China. Due to the lack of an official standard for the designation of medical-shortage area, hot spot analysis was employed to identify townships with poor spatial accessibility and high medical demand on specified significant level. Though only medical-shortage areas on 99% confidence were listed in this paper, policymakers could choose different significant levels to adjust the shortage township list, based on available health resources and planning purposes. In addition, since spatial accessibility depended on supply and demand locations, transportation infrastructure, and characteristics of travelers, each township with poor spatial accessibility should be investigated individually to find out its own cost-effective way for improvement.

Since medical-shortage townships mainly distributed in high outmigration northern Anhui and mountainous western and southern Anhui, this paper supported policies of encourage rural–urban migration (migration from rural to urban) with emigrant’s own children and old parents out of these populous or mountainous rural areas for better urban life. In short run, more medical resources needed in recognized shortage areas. There had been a great contribution from geography and location science to spatial planning of medical resources (Wang, 2012). Several accessibility-based location-allocation models had been proposed as regional planning tools to maximize or equalize spatial accessibility to services (Li, Vo, et al., 2017; Li, Wang, et al., 2017; Song et al., 2016; Tao et al., 2014). It needed noting that most of the existing spatial optimization models were focused on allocating medical resources to match rural residents and overlooked rural–urban migration in urbanizing China. Therefore, it needed considering rural labor force's free migration to any city with their old parents and young children, and then allocating medical resources (Zhao & Li, 2009).

This study was subjective to several limitations in data and methodology. Though township, the smallest administrative unit in China, served as analysis unit in this paper, yet as all studies using aggregate data, our study still suffered from the modifiable areal unit problem (MAUP) and ecological fallacy. So, it was helpful for policymakers to examine medical-shortage townships. Additionally, due to difficulty in obtaining data for medical-institution locations and their beds, as well as the relative lack of research results in accessibility on township scale in a provincial context, this paper only took the position of the best hospital in each county as the location of medical resources, so uncertainty was introduced in accessibility estimation by this strong assumption. Besides, limited to data available, this paper analyzed only the disparity of accessibility in special age groups and among urbanization groups, not considering other non-spatial factors like insurance type, the proportion of the population with a disability lasting more than one year, the levels of illiteracy, and the proportion of families without tap water. After release of China's 3rd National Agricultural Census, it was expected to have alternative access to more economic data of township. Lastly, with China’s rapid urbanization, township population, transportation infrastructure, and medical institutions were changing. This paper analyzed the spatial distribution of accessibility with cross-sectional data and did not explore the dynamic characteristics of accessibility.

5 Conclusion

It was helpful to reassess medical accessibility of township to county hospital after application of Settlement of Non-local Medical Treatment within province. Based on spatial accessibility from E2SFCA, this paper assessed variations of spatial accessibility to county hospital in special age groups and among urbanization groups in provincial China by township for 2015, recognized medical-shortage townships using hot and cold analysis and divided them by age and urbanization. The results showed that for the whole Anhui, percent of 65 + and 0–14 age groups were both higher in townships with below-average accessibility than those in townships with above-average accessibility, and accessibility to county hospital decreased significantly as urbanization level descended. On 99% confidence, 34 physiological medical-shortage townships recognized located in northern and mountainous western Anhui, and 13 economic medical-shortage townships were mainly found in mountainous western and southern Anhui. According to the results, in the short run, more medical resources needed allocating to these areas, yet in the long run, we suggested policymakers paying more attention to protect labor force's full right of free migration with family.