The Effect of Descending Resources Reform on Reallocating Healthcare Resources: An Analysis Based on the Difference-in-Differences Method

Abstract

Since 2010, China has been exploring descending resources reform in order to correct the imbalanced allocation of healthcare resources and promote coordinated economic development among regions. This paper for the first time estimates the impact this reform has had on the reallocation of healthcare resources by using prefecture-level cities panel data from Zhejiang Province, China, which implemented the reform province-wide in 2013. The time-varying difference-in-differences (DID) method was used to estimate the reform’s policy effects. The data used in this paper is from published statistical yearbooks and local governments, which include panel data from 11 prefecture-level and higher cities in Zhejiang Province as the treated group and 46 prefecture-level cities in Jiangsu, Henan, and Sichuan Province as the control group. The entropy weight method was used to construct the supply index and demand index to incorporate multiple inputs and outputs, and efficiency indicators were constructed using the ratio method. This research found that the reform has had a positive effect on outpatient visits in different prefecture-level cities with vast rural areas. However, this reform exerted no significant impact on inpatient services or supply-side or resource allocation efficiency. Several robust tests support the above conclusions, and one theoretical explanation is provided. The descending health resources reform can be a valuable reform path in promoting more balanced healthcare resource allocation; however, the resultant disparities in its effects should be considered when implementing it.

• What do we already know about this topic?

• Descending resources reform is the latest reform practice in China to correct interregional healthcare resources imbalances.

• How does your research contribute to the field?

• This paper for the first time estimates the impact of descending resources reform on the reallocation of healthcare resources by using time-varying difference-in-difference (DID) methodology and prefecture-level cities panel data from Zhejiang province, China.

• What are your research’s implications toward theory, practice, or policy?

• The descending health resources reform can be a valuable reform path in promoting more balanced healthcare resources allocation, however, the resultant disparities in its effects should be taken into account when implementing it.

Introduction

Unbalanced regional economic development is a major challenge for countries around the world.¹ This imbalance is reflected in interregional differences in the allocation of healthcare, education, and other resources. In turn, residents’ amenities tend to differ, which leads to the outflow of labor from lagging areas and further exacerbates interregional imbalances.² Previous regional policies used mainly fiscal transfer payments and differentiated macroeconomic policies to make up for fiscal deficits and re-attract capital to flow into more backward areas. However, for such areas, factors other than financial funds and capital, including human capital and technology, are needed to reduce developmental imbalances; regrettably, the previous policies were not always effective, and in some cases, were counterproductive.

In China, a new regional policy, partner assistance, has been introduced to correct interregional imbalances that have occurred since the 1980s. This policy incentivizes human capital and other factors from developed areas to be redeployed and embedded in backward areas in the short term. It does so through the organization and mobilization of central and/or higher government and compensation for the cost of the reform.³ After 2010, partner assistance was expanded to the healthcare sector by way of descending resources reform. Theoretically, the descending resource reform could embed human capital to low-level hospitals in backward areas and help them reattract patients, thereby realizing a balanced interregional allocation of healthcare resources.⁴ Although previous work revealed that this reform could incentivize patients to choose local low-level hospitals by using microlevel survey data,^5,6 the evidence on the effectiveness of this reform is insufficient. There is a research gap: we are still unclear if this reform generates resources reallocation effect with interregional macrolevel data and if it provides firm evidence of its effectiveness as a regional health policy.

Previous literature evaluates the reform effects from 2 perspectives: First, cross-sectional survey data is used to evaluate the impact of the reform on patients’ choices.^5,6 Because the goal of descending resources reform is to enhance the capability of low-level hospitals, whether patients are aware of this and choose low-level hospitals for treatment as a result would be helpful in evaluating the reform effect. Positive evidence is found that this reform can improve patient satisfaction and reattract patients to low-level hospitals.^5,7-10 It is also reported that several problems remain, such as weak patient incentives, insufficient capabilities of low-level hospitals, and competition among different hospitals involved in the reform.^11-15 However, survey-data research cannot fully reveal the effectiveness of this reform as a regional policy. Second, a few studies utilized the DID method to evaluate the marginal effects of healthcare reform with hospital-level data, with a focus on single cities such as Xiamen, Changsha, and Beijing.^16-18 Such studies usually support the effectiveness of this reform by encouraging patients to choose low-level hospitals^16,17; however, conflicting evidence has also been reported using survey data and the DID model.¹⁸ According to our survey, no macrolevel studies have been performed to demonstrate the effectiveness of descending resources reform as a regional policy. This study is the first attempt to use prefecture-level macro data and the DID model to explore the reform effect, which would help to fill the research gap.

In this paper, we take all prefecture-level and higher cities in Zhejiang Province as the treated group, and 46 prefecture-level cities in Jiangsu, Henan, and Sichuan Province as the control group. Zhejiang was the first province to implement the reform triggering for certain pilot cities in 2010 and province-wide in 2013, which could be viewed as a quasi-natural experiment. Meanwhile, the uneven allocation of healthcare resources is very challenging in Zhejiang Province. Most high-level (tertiary) hospitals are concentrated in Hangzhou, the capital of Zhejiang Province, whereas most other prefecture-level cities (and affiliated counties containing vast villages) were allocated only a few high-level hospitals. Therefore, we use data from Zhejiang Province to perform the DID estimation.

This paper stands out from previous studies for the following reasons: First, prefecture-level city panel data is used for the first time to evaluate the reform’s impact on healthcare resource reallocation by focusing on a pilot provincial sample. Second, the time-varying DID method is utilized to reflect significant differences in the startup time of the reform and more accurately identify the reform’s policy effects. Finally, to incorporate multiple inputs and outputs, the entropy weight method was employed to construct the supply and demand index, and the ratio method was subsequently used to provide efficiency indicators. This paper contributes to filling the research gap between China’s new healthcare reform and its regional policy effect and could generate new insights into correcting the imbalanced allocation of healthcare resources.

Background

In the past 40 years, to satisfy the requirements of population mobility and cultivate competitive markets, China has witnessed a vast transformation from a government-led, publicly financed medical security system to a market-oriented, privately financed one.¹⁹ As a result, a large number of public hospitals have been transformed into commercial institutions, and healthcare services have been transformed from welfare-oriented offerings into efficiency-oriented ones.²⁰ Since 1992, while further decentralizing public hospitals and expanding their autonomy, China declared its market-oriented goal of healthcare system reform accompanied by the establishment of a socialist market economy.²¹ This market-oriented reform has helped China to rapidly expand medical resources—but it has also caused problems, such as the decline of government expenditures and medical insurance coverage and the increased costs of medical treatment.²² Khatri and Shuiyuan reported that China’s government health expenditures continued to decline from 1995 to 2001 when they reached a historic low.²³ Low-level public financing and high out-of-pocket spending ratios then became the most prominent problems to be solved.

Owing to insufficient government investment, these newly autonomous public hospitals are now responsible for achieving a cost-benefit balance and are motivated to increase profits by fueling market competition. High-level hospitals quickly win such competitions because of their scale and financial advantages, which in turn reshapes patient expectations regarding diagnosis and treatment capabilities and encourages them to choose high-level hospitals for treatment. The first-mover advantage of high-level hospitals is further strengthened by the inflow of human capital and more advanced medical facilities.²⁴ The profit-seeking motive encourages inflation in the cost of receiving medical treatment, and patients’ influx to high-level hospitals contributes to structural congestion.^25,26 Downstream, supply-induced demand and skimming behavior damages the reputation of hospitals and doctors and triggers increasing social instability and even healthcare workplace violence.^27,28 In addition, the underutilization of low-level hospitals and the outflow of human capital mean a weakened ability to deal with public health crises and an unbalanced allocation of healthcare resources.²⁹

The SARS (severe acute respiratory syndrome) crisis in 2003 highlighted the neglect of low-level hospitals during the previous reform. Since then, outsized investments in low-level hospitals have dominated China’s healthcare reforms. During this subsequent reform phase, increased government expenditures have focused mainly on material capital; however, they have not helped low-level hospitals obtain human capital inflows or improve their diagnosis and treatment capabilities. As a result, patients are still reluctant to choose low-level hospitals for treatment.

China’s national universal healthcare reform, launched in 2009, emphasized a return to government-led, people-centered changes. The early phase of this reform was a 3-year action plan for 2009 to 2011 that focused on expanding health insurance coverage and zero-markup medical price reform. Research shows that the 2009 reform had a positive effect on reducing patients’ economic burden and helped alleviate the high cost of medical treatment from a microeconomic perspective.^30,31 However, an unintended consequence followed: Because patients’ affordability constraints were relaxed,³² they could now afford to choose high-level hospitals for treatment.³³ Thus, structural congestion was not significantly alleviated and the efficiency of low-level hospitals did not improve.³⁴ Consequently, even after a series of healthcare reforms, low resource utilization and capability constraints of low-level hospitals remain a serious challenge for China.³⁵ Past demand-side reforms did not incentivize patients to choose low-level hospitals and may even have contributed to the imbalanced allocation of healthcare resources.

Starting in 2010, Zhejiang Province tested a new policy that focuses on recruiting doctors of high-level hospitals to work in low-level hospitals. In 2013, the Zhejiang provincial government implemented the reform province-wide. This new supply-side reform originated from the combination of China’s dominant public hospital system and past partner assistance policy. It emphasizes the positive effects of human capital spillover and brand implementation.³⁶ Human capital spillover, or externality of human capital, increases productivity via knowledge and technology transfer, peer interactions, and other mechanisms. Brand implementation, or rebranding, refers to an inferior entity (eg, a low-level hospital) operating with the brand of a higher-level entity (the descending hospital) and adding its name as a branch to attract more patients.

The goal of these reforms is to balance the allocation of medical resources on both the supply and demand sides. According to the People’s Government of Zhejiang Province,^37,38 the main elements of Zhejiang’s reform can be summarized in the following ways:

• Descending hospitals: (a) The 2013 document requires provincial/municipal tertiary hospitals to establish a fully managed cooperative relationship with at least 1 (county) low-level hospital, usually a secondary one. A 2015 document expanded the coverage of the reform. Provincial general and specialized hospitals are required to establish cooperative relationships with at least 4 and 2 county-level public hospitals, respectively. (b) Other tertiary hospitals in major cities are required to participate, but the number of paired hospitals is determined by each municipality itself. Among them, city-owned tertiary hospitals of Hangzhou are required to cooperate with more than 5 county-level hospitals. Tertiary hospitals owned by a prefecture-level city are required to establish a fully managed cooperative relationship with at least 1 county-level public hospital. The descending hospitals can receive custody fees from hospitals receiving descending doctors, comprising no less than 3% of the latter’s gross revenue.

• Hospitals receiving descending doctors: By the end of 2015, full coverage of urban high-quality medical resources should descend to all counties, and county-level hospital resources should descend to all towns.

• The scale of descending doctors and related incentives: Urban tertiary hospitals are required to recruit sufficient managers and doctors to work full-time in hospitals that are receiving descending doctors. The number of descending doctors should be no less than 5% of those with intermediate and higher titles in the descending hospital. In addition, 80% of descending doctors should have intermediate and higher titles. The descending doctors and managers usually work in and are managed by hospitals that are receiving medical resources for a few months to 2 years. In normal practice, the experience of descending doctors is listed as an essential condition for their promotion. All hospitals receiving medical resources as well as their local governments provide an economic incentive to descending doctors.

• Government cost compensation: A special provincial fund was set up to incentivize descending hospitals to partner with cooperative low-level hospitals. For provincial high-level hospitals, the subsidy is approximately 500 000 to 4 000 000 Yuan, based on the cooperative relationship. Hospitals receiving medical resources can also receive subsidies to compensate for their reform-related expenses. However, these financial funds are allocated according to the performance appraisal results in the following year, to provide additional incentives.

• Bidirectional spillover: Key managers and doctors from hospitals receiving medical resources are accepted into descending hospitals for all-staff training within 3 to 5 years. A mentorship training system for doctors working in low-level hospitals has been established. And some excellent doctors are permitted to do a rotation in descending hospitals.

If patients become aware of the above reform policy information, presumably they would reshape their expectations and more often choose treatment at local, low-level hospitals. Thus, patients’ biased interregional-choice behavior could be corrected. Meanwhile, if the capability and attractiveness of low-level hospitals in backward areas improved, more human capital inflow could be expected. Increasing local government support would also promote more balanced resource allocation. The allocation efficiency of medical resources would improve because underutilization of low-level facilities would tend to decrease. However, empirical evidence is needed to test this hypothesis. In this paper, we view Zhejiang’s reform as a quasi-natural experiment and use the DID method to estimate its impact on resource reallocation from the supply, demand, and efficiency perspectives.

Material and Methods

Empirical Model

Because the medical market is characterized by multiple inputs and outputs, we first constructed a supply index and a demand index using the entropy weight method. By using the supply/demand index, we could reduce the dimensions of supply/demand indicators and better measure changes in supply and demand as shown in the existing literature.³⁹ The entropy weight method can determine the weight coefficients of different indicators based on the amount of information provided by observations, and it enables us to reduce the information lost in the calculation process. After standardizing the supply (input) and demand (output) indicators for each region i with equation (1), we can calculate the supply index and demand index for each region i using the entropy weight method as follows:

$$P_{ij}=\frac{Y_{ij}}{{\sum }_{i=1}^n{Y}_{ij}}$$ (1)

$$E_j=-\frac{1}{ln(n)}\sum _{i=1}^n{P}_{ij}ln(P_{ij})$$ (2)

$$W_J=\frac{1-E_j}{k-{\sum }^{}{E}_j}$$ (3)

$$T_i=\sum _{j=1}^n{Y}_{ij}\times W_j$$ (4)

where ${\mathrm{Y}}_{\mathrm{ij}}$ represents the value of indicator j of region i, and ${\mathrm{E}}_{\mathrm{j}}$ represents the information entropy of indicator j. ${\mathrm{W}}_{\mathrm{j}}$ is the weight of indicator j, and ${\mathrm{T}}_{\mathrm{i}}$ represents the value of the supply index or demand index of region i.

Next, by using the ratio method to divide the demand index by the supply index, we derive the following measure of allocation efficiency of healthcare resources (MRAE):

$$MRA{E}_i=\frac{MR{D}_i}{MR{S}_i}$$ (5)

Since the descending resources reform could be viewed as a quasi-natural experiment, it is suitable to utilize the DID methodology and panel data in evaluating its effectiveness.^40,41 Based on the above indexes, we construct the following DID model:

$$y_{it}=\alpha +\delta treat\_yea{r}_{it}+\gamma X_{it}+{\mu }_i+{\eta }_t+{\epsilon }_{it}$$ (6)

where i denotes a prefecture-level city and t represents a year. The dependent variable y refers to resource allocation indicators. Treat_year is a dummy variable representing whether the reform occurs in year t for unit i. If unit i implements the reform in year t, $\mathrm{treat}\_{\mathrm{year}}_{\mathrm{it}}$ equals 1 and otherwise is 0. The model also includes individual-fixed effects ${\eta }_{\mathrm{t}}$ µ_i as well as time-fixed effects ${\epsilon }_{\mathrm{it}}$ , and ${\mathrm{P}}_2$ denotes the random error. δ is the DID coefficient that needs to be estimated, that is, the interregional differences in resource allocation between those regions that did or did not implement the reform. X is a vector of control variables that vary in both the individual and the time dimensions.

Data

The paper selected the prefecture-level cities in Zhejiang as the treated group. The descending resources reform of Zhejiang Province involves 11 prefecture-level and higher cities (Figure 1). Among them, as the capital of Zhejiang Province, Hangzhou was excluded from the treated group and studied separately. For the control group, Nanjing, the capital of Jiangsu Province, was removed for better comparability of data; Deyang, Ziyang, Garzê, and Liangshan in Sichuan Province were also excluded because of a lack of data. The control group contains panel data from 12 prefecture-level cities in Jiangsu Province, 17 prefecture-level cities in inland Henan Province, and 17 prefecture-level cities in Sichuan Province (Table 1). The sample period covers 2008 to 2016; in August 2016, the Chinese government announced a nationwide reform of descending medical resources to low-level hospitals through medical associations.

Figure 1.

The location of Zhejiang province and its 11 prefecture- and above level cities.

Table 1.

The Setting of Treated Group and Control Group.

	Province	City
Treated Group	Zhejiang	Ningbo, Wenzhou, Shaoxing, Huzhou, Jiaxing, Jinhua, Quzhou, Taizhou, Lishui, Zhoushan, Hangzhou
Control Group	Jiangsu	Wuxi, Xuzhou, Changzhou, Suzhou, Nantong, Lianyungang, Huai’an, Yancheng, Yangzhou, Zhenjiang, Taizhou, Suqian
	Henan	Zhengzhou, Kaifeng, Luoyang, Pingdingshan, Anyang, Hebi, Xinxiang, Jiaozuo, Puyang, Xuchang, Luohe, Sanmenxia, Shangqiu, Zhoukou, Zhumadian, Nanyang, Xinyang
	Sichuan	Chengdu, Zigong, Panzhihua, Luzhou, Mianyang, Guangyuan, Suining, Neijiang, Leshan, Nanchong, Meishan, Yibin, Guang’an, Dazhou, Ya’an, Bazhong, Tibetan Qiang Autonomous Prefecture of Ngawa

Source. The authors.

The data used in this paper derive from published statistical yearbooks and local governments, obtained through open-data requests (Table 2). Based on available data, the demand-side indicators include outpatient visits, number of discharged patients, and bed utilization of inpatient services.^42,43 For the supply-side indicators, number of medical institutions and actual open beds are used to measure material inputs. Health expenditure is used to measure financial input, and a number of doctors and nurses are used to measure human resources input.^43,44 Their weight coefficients are shown in Table 2. We constructed the supply and demand index by multiplying demand-side/supply-side indicators by weight coefficients, respectively. We also introduced 3 control variables: resident population, fiscal self-sufficiency, and per capita gross domestic product. The above data were algorithmized, and their definition can be found in Table 2.

Table 2.

Definition and Variable Name of Supply and Demand Evaluation Index.

Variable	Evaluation index	Symbol	Definition	Weight	Sign
Supply index (MRS)	Material input	Institiutions	Number of medical institutions	0.154	+
		Beds	Number of actual open beds	0.189	+
	Finance input	Medexp	Health expenditure	0.225	+
	Human resource input	Doctors	Number of doctors	0.195	+
		Nurses	Number of nurses	0.238	+
Demand index (MRD)	Outpatient service	Patients	Outpatient visits (10 000)	0.514	+
	Inpatient service	Beduse	Bed utilization (%)	0.016	+
		Outpatients	Number of discharge patients (10 000)	0.470	+
Efficiency	Input-output ratio	MRAE	Supply index/demand index
Control	Resident population	SIZE	Number of residents
	Finance self-sufficiency	GFC	Budget revenue/ budget expenditure
	Per capita GDP	GDP	GDP/ number of residents

Source. The authors.

Result

Trends in Healthcare Resources Allocation in the Study Regions

As shown in Table 3, the mean value of the supply index is 0.159 for all prefecture-level cities (range: 0.017, 0.444). The mean value of the demand index is 0.179 (range: 0.026, 0.487). The supply and demand index intervals for the control group are (0.008, 1.000) and (0.011, 1.000), respectively. It can be seen that the control group samples are comparable with the treated group. Meanwhile, the interval of the efficiency indicator for the treated group (0.740, 1.533) is included within the control group efficiency interval (0.221, 4.294). This characteristic is also reflected in the different indicators used to construct the supply and demand indices.

Table 3.

Descriptive Analysis of Main Variables.

Variables	Max		Min		Mean		S.D.
	Treated	Control	Treated	Control	Treated	Control	Treated	Control
Efficiency	1.533	4.294	0.740	0.221	1.154	1.051	0.156	0.422
Supply index	0.444	1.000	0.017	0.008	0.159	0.182	0.107	0.126
Demand index	0.487	1.000	0.026	0.011	0.179	0.182	0.111	0.126
Health visits (10 000)	9341.020	12 762.240	571.473	239.716	3200.507	2600.755	2114.447	1786.000
Discharge patients (10 000)	110.275	390.775	7.916	9.213	48.793	70.258	25.755	48.876
Bed utilization (%)	99.990	113.470	80.310	14.990	91.917	88.630	4.882	9.723
Health expenditure (108 Yuan)	88.720	113.760	4.210	2.140	24.037	23.760	18.560	17.680
Medical institutions	5568.000	9853.000	193.000	393.000	2073.556	3370.548	1462.962	1922.166
Actual open beds	35 688.000	128 058.000	3663.000	2615.000	16 940.700	22 089.190	8443.425	15 842.570
Doctors	24 976.000	54 718.000	2296.000	1370.000	10 362.610	9676.778	5757.893	6939.923
Nurses	23 714.000	66 724.000	1887.000	769.000	9395.478	8983.374	5479.935	8104.291
Resident population (10 000)	919.700	1591.760	106.100	88.640	457.080	506.545	243.769	264.202
Per capita GDP (10 000)	11.066	14.556	2.205	0.681	5.649	3.874	1.961	2.646

Source. The authors’ collection.

In Figure 2, we report the evolution of the supply, demand, and efficiency indicators as well as the number of outpatient visits for Hangzhou, Ningbo, Wenzhou, and Zhoushan. It can be seen that the dynamics of related indicators in Ningbo and Wenzhou are relatively consistent. Specifically, outpatient visits in Ningbo and Wenzhou experienced a rapid rise in 2011 before continuing a smoother upward trend. The demand and supply indices in the 2 cities kept increasing, while a significant increase in the growth rate could be observed from 2010 to 2014. However, the 2 cities had differences in changes of efficiency indicators. The efficiency indicator of Wenzhou tended to decline before 2010 but increased in 2011 and then remained stable with slight fluctuations in subsequent years; in contrast, that of Ningbo declined from 2010 to 2012 and then showed a stable trend with small fluctuations after 2013.

Figure 2.

Evolution of Health Care Resources Allocation in Selected Cities of Zhejiang Province.

Zhoushan is an island region with the weakest medical resources in Zhejiang Province owing to its remoteness and high commuting costs. Although Zhoushan’s number of outpatient visits and supply and demand indices also showed an upward trend, its value was far lower and its growth trend weaker than other cities. As for Hangzhou, the capital of Zhejiang Province, its number of outpatient visits grew at a faster rate before 2012, slowed down between 2012 and 2016, and accelerated in subsequent years. Its supply and demand indices were higher than those of the other cities. However, Hangzhou’s efficiency indicators grew before 2012 but declined in subsequent years with fluctuations and have approached the levels of Wenzhou and Ningbo.

Table 4 reports changes in the supply and demand indices for all the prefecture-level cities of the treated group. It can also be seen that the supply and demand indices of Hangzhou were significantly higher than those of other cities. In addition to Ningbo and Wenzhou, healthcare resource allocation in Jiaxing, Huzhou, Shaoxing, Jinhua, and Taizhou was significantly lower than that of Hangzhou but considerably higher than that of Zhoushan; in Lishui and Quzhou, it was only slightly higher than that of Zhoushan. These results reveal the great interregional imbalances of healthcare resources in Zhejiang Province, which is consistent with the control group. Table 4 also provides the time points of reform triggered in different prefecture-level cities, defined as the month when descending resources reform was first reported in official documents. The current year is set to 1 if the time point of reform occurs before July 1, which means that the current year is the shock year; if the time point of reform occurs after July 1, the following year is the shock year and set to 1. We did this to take into consideration the time lag needed to organize, mobilize, and implement the reform.

Table 4.

Descriptive Statistics of Demand and Supply Index in Zhejiang and the Time points of reform.

City	Demand index				Supply index				Time points of reform
	Max	Min	Mean	S.D.	Max	Min	Mean	S.D.
Hangzhou	0.737	0.333	0.559	0.154	0.646	0.277	0.438	0.123	2013.11
Ningbo	0.487	0.264	0.376	0.076	0.444	0.208	0.336	0.087	2013.12
Wenzhou	0.404	0.173	0.288	0.084	0.436	0.147	0.293	0.099	2011.9
Jiaxing	0.242	0.121	0.175	0.040	0.184	0.085	0.137	0.033	2011.12
Huzhou	0.166	0.073	0.122	0.034	0.130	0.059	0.092	0.026	2013.10
Shaoxing	0.256	0.106	0.183	0.052	0.238	0.098	0.152	0.045	2010.12
Jinhua	0.315	0.127	0.225	0.071	0.297	0.104	0.200	0.070	2013.9
Quzhou	0.111	0.039	0.072	0.025	0.124	0.032	0.068	0.030	2013.8
Zhoushan	0.058	0.026	0.042	0.011	0.057	0.017	0.037	0.014	2014.1
Taizhou	0.292	0.146	0.221	0.054	0.280	0.108	0.185	0.060	2013.9
Lishui	0.111	0.049	0.083	0.024	0.143	0.041	0.091	0.035	2013.10

Source. The authors’ calculation.

DID Estimation Result With Samples Excluding Hangzhou

As Table 5 shows, DID estimation results were insignificant in the supply index, demand index, and efficiency models. Because descending resources reform in Zhejiang influences patients’ choice mainly by the spillover effect of human capital and rebranding, the reform has a weak short-term impact on material and financial inputs in each prefecture-level city; therefore, its insignificant influence on the supply index is logical. However, the goal of this reform is to encourage patients to choose hospitals in the prefecture-level city they reside in and is expected to exert an impact on the demand side. Consequently, our results showing that this reform does not have a significant effect on the demand index were surprising.

Table 5.

Estimation Results of DID Model With Samples Excluding Hangzhou.

Variables	Efficiency	Supply index	Demand index	Outpatient visits	Discharge patients
DID	−0.028 (0.069)	−0.010 (0.016)	0.007 (0.013)	0.031* (0.018)	−0.018 (0.011)
Financial self-sufficiency	−2.449*** (0.624)	0.057 (0.082)	−0.068 (0.078)	−0.070 (0.093)	−0.075 (0.079)
Resident population	1.856** (0.751)	0.305 (0.211)	0.420** (0.188)	0.375* (0.205)	0.483** (0.206)
Per capita GDP	−0.001 (0.027)	0.010 (0.007)	0.011 (0.006)	0.014* (0.007)	0.007 (0.006)
Cons	−8.924* (4.543)	−1.745 (1.323)	−2.373** (1.178)	−2.116 (1.285)	−2.767** (1.272)
N	504	504	504	504	504
Time fixed	Yes	Yes	Yes	Yes	Yes
City fixed	Yes	Yes	Yes	Yes	Yes
R 2	.537	.939	.954	.950	.947

Note. [1] The values in brackets are standard errors. *** represents 1% significance levels. [2] a, b, and c indicate that the time point of reform is set 1 to 3 years in advance, respectively.

If we further break down demand into outpatient services and inpatient services, DID regression analysis shows that the reform has had a positive impact on outpatient visits at a significance level of 10%, but this significant positive effect does not apply to inpatient services. One explanation is that the reform has led to doctors descending from high-level hospitals to local, low-level hospitals, which could have influenced residents’ behavior to seek outpatient treatment. That is, the reform has helped changed residents’ habit of going to high-level hospitals for treating minor illnesses, and more of them have chosen local low-level hospitals to obtain their first diagnosis.

However, to change the demand for inpatient services, which are usually for more severe illnesses, a longer time and greater spillover effect are needed to improve the quality of services; also, patients’ expectations and habits are more difficult to change, causing insignificant results in the inpatient services model. The coexistence of the positive effect on outpatient visits and the insignificant results for the inpatient model result in an insignificant estimate of the reform’s effect on the demand index.

Robustness Test Results

To test the robustness of the above empirical results, we adopted the following testing strategy: First, a parallel trend test was performed to test whether the DID model was appropriate, to ensure that the trends for outpatient visits in the treated and control groups were parallel before the reform. Second, the placebo test was used to test whether the DID coefficient remained significant in the fictitious policy time, and if it was, the existing estimation results would be biased. Finally, 2 additional robustness tests were carried out by eliminating outliers and replacing explained variables.

The parallel trend test was performed first. The event study method was used for testing: Let t−1, t−2, t−3 and t, t + 1, and t + 2 represent 1 to 3 years before the current year and 1 to 2 years after the time point of reform, respectively. We introduced dummy variables that indicated the difference between the treated group and the control group in a given year. If the interactive terms of the treated dummy and the fake policy shock-year dummy were insignificant, we concluded that no significant differences exist between the treated and control groups prior to the reform. As Figure 3 shows, the regression coefficients of the interactive terms did not differ from 0 for 1 to 3 years prior to the reform; however, from year t to t + 2, the regression coefficients all significantly differed from 0, indicating that the treated and control groups satisfy the parallel trend assumption.

Figure 3.

Parallel trend test results for samples excluding Hangzhou.

The placebo test was then performed to see whether the significant policy effect originated from other policy changes or from random factors. We set a fake policy year at 1 to 3 years earlier than the shock year and constructed an interactive term for the virtual policy year and the treated group; we then used the DID model to perform the regression. If the interactive terms were insignificant, this indicated that the estimated marginal effects originated from the descending resources reform; if the interactive terms were significant, then the effect of other factors could not be excluded and the existing results were not robust. As Table 6 shows, the regression coefficients of the interactive terms were insignificant, indicating that the increase in outpatient visits could be attributed to descending resources reform. The results of the outpatient model are therefore robust.

Table 6.

Placebo Test Results for Samples Excluding Hangzhou.

Variables	Dependent variables: outpatient visits
	(1)	(2)	(3)
1 year in advance	0.031 (0.019)
2 years in advance		0.025 (0.019)
3 years in advance			0.022 (0.021)
Control	Yes	Yes	Yes
Year	Yes	Yes	Yes
City	Yes	Yes	Yes
R 2	.950	.949	.949
N	504	504	504

Note. The values in brackets are standard errors.

In addition, we report estimation results excluding the Zhoushan sample as outliers (Table 7). It can be seen that the main empirical results are robust, that is, the estimated DID coefficients in efficiency, supply index, demand index, and inpatient models remain insignificant. However, the DID coefficient of the outpatient model is significant at the level of 5%, convincing us of the marginal policy effect of the descending resource reform in boosting outpatient visits in different cities.

Table 7.

Estimation Results of DID Model With Samples Excluding Outliers.

Variables	Efficiency	Supply index	Demand index	Outpatient visits	Discharge patients
DID	−0.010 (0.071)	−0.004 (0.015)	0.014 (0.013)	0.040** (0.017)	−0.014 (0.011)
Finance self–sufficiency	−2.500*** (0.638)	0.054 (0.082)	−0.072 (0.078)	−0.072 (0.092)	−0.081 (0.079)
Resident population	1.849** (0.755)	0.316 (0.210)	0.429** (0.187)	0.386*(0.204)	0.491** (0.205)
Per capita GDP	0.003 (0.028)	0.012* (0.007)	0.012* (0.006)	0.016** (0.007)	0.009 (0.007)
Cons	−8.918* (4.583)	−1.823 (1.320)	−2.442** (1.175)	−2.195* (1.288)	−2.827** (1.270)
N	495	495	495	495	495
Year	Yes	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes	Yes
R 2	.538	.940	.955	.952	.948

Note. The values in brackets are standard errors. *, **, and *** represents 10%, 5%, and 1% significance levels, respectively.

Finally, we replaced the dependent variables to perform the robustness test. We used the data envelopment analysis (DEA) method to construct a new efficiency measure for estimation. We first calculated comprehensive and scale efficiency using the DEA method, and such indicators could be used directly as dependent variables (Model 1) for estimation. However, since the effective DEA unit is represented by 1 and the ineffective unit is represented by a number of (0, 1), which are inadequate to measure the difference between the effective and ineffective unit, we further introduced dummy variables of comprehensive efficiency and scale efficiency (Model 2). In these variables, the comprehensive efficiency or scale efficiency is taken as 1 for the effective unit, and otherwise is 0.

Results for replacing the dependent variables are reported in Table 8. The regression results of Model 1 are still insignificant; however, the estimation results of Model 2 show that the DEA efficiency significantly improved after the time point of reform. Compared with the previous empirical results, Model 1 in Table 8 provides evidence of its robustness. However, the results of Model 2 provide more optimistic evidence that the reform helps to increase the efficiency of healthcare resource allocation.

Table 8.

Estimation Results of DID Model After Replacing the Dependent Variable.

Variables	Comprehensive efficiency		Scale efficiency
	Model 1	Model 2	Model 1	Model 2
DID	0.005 (0.020)	0.294***(0.110)	0.008 (0.016)	0.261**(0.112)
Control	Yes	Yes	Yes	Yes
Year	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes
R 2	.600	.533	.532	.515
N	504	504	504	504

Note. The values in brackets are standard errors. *, **, and *** represents 10%, 5%, and 1% significance levels, respectively.

DID Estimation Results With the Hangzhou Sample

We separately report the empirical results of the Hangzhou sample (Table 9). It shows that, except for the efficiency model, the other DID regression coefficients were all positive at the significance level of 1%. It seems that demand was boosted after the reform shock, which went against the goal of the reform. The explanation may lie in the rapidly growing population in Hangzhou and the release of pent-up demand that was previously suppressed by congestion. On the supply side, the reform’s positive effect can be explained by the investment incentives of high-level hospitals stimulated by easing congestion during the same period; abundant high-level hospitals in Hangzhou received substantial government financial subsidies and incentives to participate in the reform, which contributed to the increase in investment in medical resources.

Table 9.

Estimation Results of DID Model with the Hangzhou Sample.

Variables	Efficiency	Supply index	Demand index	Outpatient visits	Discharge patients
DID	−0.034 (0.060)	0.094*** (0.020)	0.139*** (0.017)	0.205*** (0.018)	0.074*** (0.019)
Finance self-sufficiency	−2.750*** (0.760)	0.101 (0.090)	−0.033 (0.089)	0.001 (0.100)	−0.077 (0.096)
Resident population	1.782** (0.771)	0.317 (0.214)	0.445** (0.192)	0.408* (0.215)	0.501** (0.208)
Per capita GDP	−0.006 (0.032)	0.015** (0.008)	0.014** (0.007)	0.018** (0.007)	0.010 (0.007)
Cons	−8.464* (4.701)	−1.858 (1.346)	−2.569** (1.210)	−2.375* (1.353)	−2.896** (1.291)
N	423	423	423	423	423
Year	Yes	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes	Yes
R 2	.550	.946	.957	.954	.948

Note. The values in brackets are standard errors. *, **, and *** represents 10%, 5%, and 1% significance levels, respectively.

To determine the robustness of the empirical results for the Hangzhou sample, the placebo test was again performed for regression. Table 10 shows that the interactive items were all significant at a level of 1%, indicating that all resource allocation indicators were not robustly impacted by the reform. Thus, Hangzhou’s growth in supply and demand could be attributed to factors other than the reform. This again suggests that it is logical to remove the Hangzhou sample from the previous estimation.

Table 10.

Placebo Test Result of the Hangzhou Sample.

Variables	DID	Control	Year	City	R 2	N
MRS	0.090a,*** (0.019)	Yes	Yes	Yes	.9457	423
	0.079b,*** (0.020)	Yes	Yes	Yes	.9451	423
	0.073c,*** (0.020)	Yes	Yes	Yes	.9447	423
MRD	0.154a,*** (0.016)	Yes	Yes	Yes	.9591	423
	0.173b,*** (0.018)	Yes	Yes	Yes	.9607	423
	0.175c,*** (0.018)	Yes	Yes	Yes	.9601	423
Outpatient visits	0.238a,*** (0.017)	Yes	Yes	Yes	.9579	423
	0.278b,*** (0.019)	Yes	Yes	Yes	.9620	423
	0.286c,*** (0.019)	Yes	Yes	Yes	.9612	423
Discharge patients	0.071a,*** (0.018)	Yes	Yes	Yes	.9479	423
	0.067b,*** (0.019)	Yes	Yes	Yes	.9478	423
	0.065c,*** (0.019)	Yes	Yes	Yes	.9476	423

Note. [1] The values in brackets are standard errors. *, **, and *** represents 10%, 5%, and 1% significance levels, respectively. [2] a, b, and c indicate that the time point of reform is set 1 to 3 years in advance, respectively.

Discussion

Achieving a balanced allocation of healthcare resources among regions in China is a major target of equality in healthcare that will promote amenities and the inflow of factors to backward areas, thereby contributing to coordinated regional development overall. Unlike the vertical and horizontal fiscal transfers and differentiated fiscal and monetary policies emphasized in previously established practices, in recent decades, China has been experimenting with a new regional policy, partner assistance, to promote balanced development among regions. Descending resources reform is the latest attempt at expanding the partner assistance policy to the healthcare sector to remedy intra-provincial imbalances.³ This reform works by recruiting doctors for the short term from urban, high-level hospitals to county, low-level hospitals using the superior government’s incentives and organization and mobilization strategies. In theory, the partner assistance policy aims to reconstruct the production function of healthcare service in backward areas. It is expected to generate human capital spillover and the implementation of new branding to reshape patient choice among different level hospitals and to reallocate interregional resources. By taking Zhejiang’s reform as a quasi-natural experiment, this paper performs for the first time a prefecture-level city panel data–based DID estimation from supply, demand, and efficiency perspectives simultaneously.

Our empirical research shows the existence of differentiated policy effects of descending resources reform. This reform did not exert a significant impact on the supply side and allocation efficiency of healthcare resources. The characteristics of the reform, which focus on descending doctors and their spillover effect, did not directly change resource inputs in the recipient regions.⁴⁵ However, robust evidence showed that the reform’s positive policy effects occurred mainly in expanding outpatient services for local, low-level hospitals, while its impact on the demand index and inpatient services was insignificant. Therefore, it can be concluded that the reform helped drive patients to choose local, low-level hospitals for outpatient services more often, but the goal of shifting inpatient demand to local, low-level hospitals was not realized.

To understand our empirical results, we now provide a simple theoretical analysis. Since medical services are credence goods,⁴⁶ in order to overcome the buyer’s information disadvantage, patients’ choice among different service providers first comes from their a priori expectations. For example, when they have experienced a medical consultation, they will adjust their expectations based on their personal experience and all available information to optimize their treatment-choice behavior.

Assuming no government regulations exist, patients can freely choose service providers. However, differences exist in the diagnosis and treatment capabilities of hospitals at different levels.⁴⁷ Assume that there are 2 types of service providers: (urban) high-level hospitals and (county) low-level hospitals, and suppose that doctors can be divided into 3 categories: L, M, and H, and L < M < H according to their diagnosis and treatment capabilities. Assuming that there are only 2 types of doctors, M and H, in high-level hospitals, and 3 types of doctors in low-level hospitals, the probability of meeting M-type doctors in high-level hospitals is P3′, and the probability of meeting L- and M-type doctors in low-level hospitals is ${\mathrm{P}}_3$ and ${\mathrm{P}}_3^{\prime }<{\mathrm{P}}_2+{\mathrm{P}}_3$ , respectively. Accordingly, without loss of generality, ${\mathrm{P}}_1$ . Additionally, suppose that the probability of choosing low-level hospitals is $(1-{\mathrm{P}}_1)$ , the corresponding diagnosis and treatment expenses are c, and the generalized transportation costs incurred in travel, queuing, and other activities are t. In such a case, the probability of going to a high-level hospital is ${\mathrm{R}}_{\mathrm{L}}$ , and the corresponding generalized transportation cost and diagnosis and treatment costs are T and C, respectively. The result is the game structure shown in Figure 4.

Figure 4.

The game structure of patients’ choice for different levels hospitals.

Let ${\mathrm{R}}_{\mathrm{H}}$ and ${\mathrm{U}}_{\mathrm{R}}$ represent the expected net benefits of selecting low- and high-level hospitals, respectively. ${\mathrm{U}}_{\mathrm{H}}$ and ${\mathrm{L}*\mathrm{P}}_2+{\mathrm{M}*\mathrm{P}}_3+\mathrm{H}*\left(1-{\mathrm{P}}_2-{\mathrm{P}}_3\right)<\mathrm{M}*{P^{\prime }}_3+\mathrm{H}*\left(1-{P^{\prime }}_3\right)$ are the corresponding utility of health benefits. The expected net benefits are calculated by the difference between utility and cost, so we derive the following equation (7):

$$\{\begin{array}{c}E\left(R_L\right)=U_L\left[\begin{array}{l}L*P_2+M*P_3+H\\ *\left(1-P_2-P_3\right)\end{array}\right]-\left(t+c\right)\\ E\left(R_H\right)=U_H\left[M*{P^{\prime }}_3+H*\left(1-{P^{\prime }}_3\right)\right]-\left(T+C\right)\end{array}$$ (7)

Since L < M < H, $\mathrm{T}\gg \mathrm{t}$ is established. Considering that high-level hospitals are generally located in major cities, on average T > t, and patients’ hospital selection behavior depends on their expectation of quality differences rather than total costs between different levels hospitals.⁶ Assuming ${\mathrm{P}}_1$ in the initial state, and that the diagnosis and treatment expenses of hospitals at different levels are given, the equilibrium state is that patients will choose low-level hospitals (with probability $(1-{\mathrm{P}}_1)$ ) for treatment. This is because the transportation cost of going to high-level hospitals is much greater than low-level hospitals, so it is sufficient to compensate for the expected lower-quality probability of the latter. Only those patients with lower transportation costs preferentially choose high-level hospitals (with probability ${\mathrm{P}}_1$ for treatment. This is the initial equilibrium state dominated by transportation costs. Of course, if the disease state and condition exceed the capability of the low-level hospital after the first visit, patients will redo their cost-benefit analysis and make referral choices.

If patients see their transportation costs decrease due to the improvement of infrastructure, the increase of capabilities, and the decrease of congestion in high-level hospitals, then T tends to decrease. And, compared with the initial equilibrium state, an increase in patients preferring high-level hospitals will be observed, indicating that $1-{\mathrm{P}}_1$ will decrease while ${\mathrm{P}}_1$ will increase. Similarly, if doctors’ income is related to the probability of patients choosing low- or high-level hospitals and positively related to the number of patients, then the decline of $1-{\mathrm{P}}_1$ corresponds to decreasing patients and income in low-level hospitals; in contrast, doctors’ income in high-level hospitals would go up with an increase in ${\mathrm{P}}_1$ . The income difference between the 2 hospital types will form and expand, encouraging the flow of doctors to high-level hospitals.⁴⁸ It is reasonable to assume that H-type doctors have the greatest mobility, followed by M-type doctors. For low-level hospitals, the loss of H-type and possibly M-type doctors will weaken their diagnosis and treatment capabilities; consequently, patients will perceive this and reshape their expectations. In contrast, with the inflow of H-type doctors, the diagnosis and treatment capabilities of high-level hospitals will improve and the expected capability difference will widen. As a result, ${\mathrm{U}}_{\mathrm{L}}$ will further decrease. The above comparative static analysis implies a self-reinforcing one-way factor flow and describes patients’ biased hospital-choice behavior. This is the way that China’s distorted incentive structure and structural congestion worked before the implementation of descending resources reform.

According to the game structure shown in Figure 4, the main mechanism of descending resources reform is to recruit short-term H- or M-type doctors from high-level hospitals and descend them to low-level hospitals. Its effect is twofold: First, it increases the probability of seeing H- and M-type doctors in low-level hospitals to promote patients’ expected $\mathrm{L}*\Delta {\mathrm{P}}_2+\mathrm{M}*\Delta {\mathrm{P}}_3+\mathrm{H}*(1-\Delta {\mathrm{P}}_2-\Delta {\mathrm{P}}_3)$ . The change of probability equals $\mathrm{C}\gg \mathrm{c}$ . Since L < M < H in a comparative static sense, improving patients’ expectation for low-level hospitals’ capabilities will push them to give more priority to such hospitals. Second, the descent of H- and M-type doctors will invite human capital and technology spillover effects.⁴⁵ This will tend to impact the structure but not the quantity of doctors in low-level hospitals, which means that the proportion of L-type doctors could decrease and that of M-type doctors could increase. This would have the same effect as reversing patient choice between hospitals. The above analysis could also explain the positive effect of the reform on outpatient visits. However, owing to the possible time lag in human capital spillover, more time is needed for the reform to exert a positive effect on inpatient services. Considering the limitations of DID methodology and the data, we cannot capture the reform’s impact on inpatient services and efficiency, which should be left for future research.

One could suggest that if the government permitted free adjustment of medical prices, it would play the same role as increasing transportation costs and would incentivize patients to choose low-level hospitals by $\mathrm{C}\gg \mathrm{c}$ .⁴⁹ However, political opposition and the unpredictable cost-shifting of medical expenses would hinder its implementation. Although some differentiated reimbursement policies for medical insurance have been under consideration, no evidence has shown that price adjustment by itself can reshape patients’ expectations and choice, which occurred in past decades in China.

Some may question whether the goal of this reform could be achieved by market-oriented mechanisms rather than by government-led reform. According to microeconomic theory, if the government can subsidize the inflow of H- and M-type doctors, the same effect will be generated.¹⁶ However, local governments in backward areas usually cannot afford to do this. And, because of amenity differences between urban and rural areas, even if subsidies were provided, it would still be difficult for backward areas to attract an inflow of human capital. Market-oriented inflows of human capital are less advantageous in reshaping patient expectations than descending resources reform since the reform can rapidly deploy doctors and implement rebranding from high-level hospitals at the same time. Even so, one should not conclude that market-oriented cultivation and the inflow of human capital are less important for low-level hospitals.

Conclusion

The goal of descending resources reform is to rebalance the allocation of healthcare resources and then promote coordinated development. This paper, for the first time, estimates the impact of the reform on resource reallocation by using the DID method and data from prefecture-level panel data for Zhejiang, China. Findings show that the reform had positive effects on outpatient visits to different prefecture-level cities within vast rural areas. However, this reform exerted no significant impact on supply-side and allocation efficiency. The above robust conclusions imply that descending resources reform is one effective regional policy that can reallocate resources driven by demand, but more time is needed to extend its effects to inpatient visits, supply side, and efficiency.

This study faced several limitations. First, because this reform focuses mainly on county low-level hospitals, if future research could extend to the county level when data are available, we could better understand the impact of this reform on resource reallocation. Second, because of the time lag that occurred before the reform could fully generate its effects, future studies should focus more on measuring human capital spillover and capability changes in county low-level hospitals in the context of descending resources reform. Finally, richer evidence of its heterogeneous impact could be achieved if more provinces and counties were incorporated into the empirical estimations. Even so, our findings suggest that descending resources reform can be a valuable reform path, albeit with implied complexity, and that it can provide an alternative for those operating public hospital systems and facing severe interregional inequalities in healthcare resources allocation.