Regional differences, evolutionary trends, and influencing factors of the level of medical assistance protection across Chinese provinces

Abstract

Background

Since the launch of China’s new round of health system reform in 2009, substantial progress has been made in expanding medical security coverage and strengthening assistance for vulnerable populations. However, provincial disparities in medical assistance protection remain insufficiently understood, which may challenge the equity and sustainability of the medical security system. This study assessed the level of medical assistance protection across Chinese provinces, examined its regional differences and evolutionary trends, and identified influencing factors.

Methods

Panel data from 31 Chinese provinces from 2019 to 2023 were used. The entropy-weighted TOPSIS method was applied to construct a comprehensive index of medical assistance protection. Dagum’s Gini coefficient decomposition and kernel density estimation were used to examine regional differences and evolutionary trends. Interrupted time series analysis was employed to assess pandemic-related changes, and a panel Tobit model was used to identify influencing factors of medical assistance protection.

Results

Medical assistance protection declined sharply in 2020 and then gradually recovered through 2023. The western region consistently showed higher protection levels than the eastern and central regions. Overall, interregional, and intraregional disparities declined during the study period, while the dominant source of disparity shifted from interregional differences to hyper-variable density, and intraregional disparities became more important than interregional disparities by 2023. Kernel density estimation indicated divergent polarization patterns, with intensification in the eastern region, gradual weakening in the central region, and no clear polarization in the western region. Fiscal investment and hospitalization rates were positively associated with protection levels. Economic development and education levels were negatively associated with protection intensity, suggesting a possible welfare paradox.

Conclusion

The findings reveal an atypical regional pattern in which economically less-developed western provinces showed higher medical assistance protection levels than more developed eastern provinces. This pattern suggests that targeted fiscal transfers and policy support may play a more direct role than economic growth alone in strengthening medical assistance protection. Region-specific strategies are needed to optimize fiscal investment, improve service utilization, and strengthen primary healthcare capacity, thereby promoting a more equitable and sustainable medical assistance system.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12889-026-28620-6.

Introduction

Medical assistance is a targeted safety-net arrangement within China’s medical security system. It subsidizes vulnerable groups’ participation in social health insurance and provides additional reimbursement after basic medical insurance and catastrophic medical insurance payments [1]. As such, it serves as an important supplementary protection mechanism for disadvantaged populations.

Since the launch of China’s new round of health system reform in 2009, substantial progress has been made in expanding health insurance coverage [2, 3]. Nevertheless, broad insurance coverage does not necessarily translate into sufficient financial protection for vulnerable populations. Out-of-pocket payments, catastrophic health expenditure, and medical impoverishment remain important concerns, particularly among low-income households and populations with high healthcare needs [4]. Assessing the protection provided by medical assistance is therefore essential for understanding whether the medical security system effectively protects disadvantaged groups from medical financial risk.

Medical assistance protection may vary substantially across Chinese provinces. Such variation may be associated with unequal fiscal capacity [5], uneven healthcare resource supply [6], differences in health service utilization [7], and variation in policy implementation capacity [8]. In China, local governments play an important role in financing and implementing social assistance and medical security policies [9]. Provincial disparities therefore deserve particular attention. Although recent reforms have aimed to improve equity in medical security, the spatial distribution and temporal evolution of provincial medical assistance protection remain insufficiently understood.

Previous studies have examined the benefit equity of social health insurance, the distribution of government health insurance subsidies, catastrophic health expenditure, and provincial or regional differences in health financial risk in China [10–13]. However, three issues remain insufficiently addressed. First, provincial medical assistance protection has rarely been systematically measured using a multidimensional framework. Second, its regional differences and evolutionary trends remain underexplored. Third, the influencing factors of provincial medical assistance protection require further empirical examination within a unified panel-data framework.

To address these gaps, this study analyzes panel data from 31 Chinese provinces from 2019 to 2023. A multidimensional evaluation framework is constructed from five dimensions: resource input, coverage equity, resource accessibility, protection effectiveness, and system sustainability. The entropy-weighted TOPSIS method is applied to measure provincial medical assistance protection. Dagum’s Gini coefficient decomposition and kernel density estimation are used to examine regional differences and evolutionary trends. Interrupted time series analysis and panel Tobit regression are further employed to assess pandemic-related changes and influencing factors of medical assistance protection. This study provides empirical evidence for improving the equity and sustainability of China’s medical assistance system.

Literature review

Medical assistance protection and health financial risk protection

Research on China’s medical assistance system has focused primarily on health service utilization, medical financial burden, and poverty alleviation. Existing studies have shown that different medical assistance benefit packages affect health service utilization among the rural poor, indicating that benefit design may influence whether disadvantaged groups can access needed healthcare services [14]. Based on evidence from poor rural populations in four Chinese counties, Ma et al. (2011) suggested that medical financial assistance functions as a form of social health assistance, although its implementation, coverage, and actual financial protection remain constrained [15]. Liu et al. (2017) further examined whether medical assistance effectively supplements social health insurance for low-income households. Their findings showed that although medical assistance provides additional protection, its effect on reducing medical financial risk among low-income households remains limited [1]. These studies suggest that medical assistance evaluation should consider not only formal coverage, but also service utilization and actual financial burden reduction.

Recent research has extended the discussion of medical assistance beyond financial compensation to broader issues such as health improvement, poverty governance, and multitiered medical security. Gao et al. (2023) analyzed the policy effects of medical assistance from the perspectives of health improvement and medical financial burden, showing that medical assistance is associated not only with cost compensation, but also with healthcare accessibility among disadvantaged groups [16]. Dai and Xu (2020) examined the poverty-alleviation effect and constraints of rural medical assistance, finding that medical assistance can help alleviate health-related poverty, although its effectiveness is constrained by resource input, assistance standards, and beneficiary identification [17]. Tao and Zhang (2025) investigated rural medical assistance from the perspective of multidimensional poverty, further highlighting the relationship between medical assistance and poverty governance [18]. In addition, evidence from 667,055 low-income households showed that basic medical insurance, catastrophic medical insurance, and medical assistance jointly constitute a multitiered medical security system for disadvantaged groups. This system contributes to reducing catastrophic health expenditure and medical impoverishment, although inequalities in financial protection remain across population groups [19].

Regional differences and evolutionary trends of medical assistance protection

Regional disparity is an important perspective for understanding medical assistance protection in China. Provincial capacity to provide medical assistance may vary because of differences in economic development, local fiscal capacity, population structure, healthcare service supply, and policy implementation conditions. Existing studies have identified regional and socioeconomic differences in health financial risk protection in China. For example, Fu (2022) compared catastrophic health expenditure and related inequality between urban and rural households and found differences in exposure to medical financial risk [20]. Yuan et al. (2021), through a systematic review and meta-analysis, found that catastrophic health expenditure varied across urban–rural groups and regions in China, with a higher incidence among rural residents and poorer households [21]. Liu et al. (2021), using national panel data from 2010 to 2018, found that the incidence and intensity of catastrophic health expenditure declined over time, but remained higher among rural households than urban households; inpatient service use was also identified as a key determinant of catastrophic health expenditure [22]. These studies suggest that the protective effects of medical security policies cannot be evaluated solely on the basis of national averages; regional, urban–rural, and service-utilization differences also require further examination.

Studies on medical assistance also indicate that protection outcomes may be shaped by spatial accessibility and local conditions. Based on data from medical assistance recipients in Shaanxi Province, Ren et al. (2020) found that inpatient health service utilization varied by geographic remoteness and accessibility conditions, suggesting that disadvantaged groups may still face spatial barriers to healthcare access even when an assistance system is in place [23]. This finding indicates that medical assistance protection depends not only on eligibility and funding, but also on whether beneficiaries can access needed healthcare services in practice.

Measurement methods and influencing factors of medical assistance protection

Medical assistance protection is multidimensional, and a single indicator cannot fully capture its actual protection capacity. Indicators such as total medical assistance expenditure or the number of beneficiaries alone cannot simultaneously reflect resource input, coverage equity, service accessibility, protection effectiveness, and sustainability. Existing research shows that medical assistance evaluation has gradually shifted from single-dimensional cost compensation to multidimensional issues such as beneficiary identification, demand estimation, fund sustainability, and policy optimization. Zhu and Hu (2019) examined target identification and funding demand estimation in China’s medical assistance system, showing that medical assistance protection involves not only financial input, but also beneficiary identification and demand matching [24]. Zhou and Zhu (2025) conducted expenditure forecasting and policy optimization under a stratified classification model, emphasizing that medical assistance should balance fund sustainability with targeted assistance [25].

In terms of comprehensive evaluation methods, the entropy-weighted TOPSIS method has been widely used in medical security, health service capacity, and public health evaluation because it combines objective weighting with distance-based ranking. This method determines indicator weights according to the information dispersion of each indicator and calculates a comprehensive score based on the distance between each evaluated unit and the ideal and negative-ideal solutions, thereby reducing subjective weighting bias. Previous studies have applied entropy-weighted TOPSIS to the evaluation of multitiered medical insurance systems, primary healthcare service capacity, and health policy implementation, indicating that this method is suitable for multi-indicator comprehensive evaluation [26–28].

Overall, existing studies have examined medical assistance in relation to health service utilization, financial burden reduction, poverty alleviation, multitiered medical security, and policy optimization. However, limited attention has been paid to the systematic measurement, regional pattern, and evolutionary trends of medical assistance protection at the provincial level. In particular, it remains unclear whether medical assistance protection exhibits distinct eastern, central, and western regional differences, whether regional gaps have widened or narrowed over time, and whether disparities mainly originate from within-region or between-region differences. Therefore, this study constructs a provincial comprehensive evaluation index and combines regional difference decomposition, evolutionary trend analysis, and panel regression models to systematically examine medical assistance protection levels and their influencing factors in China.

Methods

Research methods

Entropy-weighted TOPSIS method

The entropy-weighted TOPSIS method combines the entropy weighting method and the technique for order preference by similarity to an ideal solution (TOPSIS). The entropy weighting method was used to determine the objective weight of each evaluation indicator, while TOPSIS was used to rank provinces according to their relative distance from the ideal solution [29]. This approach reduces the influence of subjective weighting and is suitable for multi-indicator evaluation. In this study, the entropy-weighted TOPSIS method was applied to calculate the comprehensive index of medical assistance protection across Chinese provinces. The procedure was as follows.

First, the original indicators were standardized. For positive indicators, the standardized value was calculated as:

For negative indicators, the standardized value was calculated as:

where denotes the original value of indicator for province , and denote the maximum and minimum values of indicator , respectively, and denotes the standardized value.

Second, the proportion of province under indicator was calculated as:

Third, the entropy value of indicator  j was calculated as:

where denotes the number of provinces included in the study, namely (n = 31).

Fourth, the information redundancy of indicator  j was calculated as:

Fifth, the weight of each indicator was calculated as:

Finally, the comprehensive index of medical assistance protection was calculated using weighted summation:

where . A larger value of indicates a higher level of medical assistance protection in province .

Dagum Gini coefficient method

The Dagum Gini coefficient method was used to measure regional disparities and decompose their sources. Compared with the traditional Gini coefficient, the Dagum method can account for within-region differences, between-region differences, and the contribution of transvariation intensity caused by overlapping distributions between regions [30]. Therefore, this method was applied to examine the relative regional disparities in medical assistance protection across China. The overall Dagum Gini coefficient is expressed as:

where  G denotes the overall Gini coefficient. A larger value of  G indicates greater regional disparity in medical assistance protection.  k denotes the number of regions, and  n denotes the total number of provinces. and denote the medical assistance protection levels of province  i in region  j and province  r in region  h, respectively. and denote the numbers of provinces in regions  j and  h, respectively, and denotes the national average level of medical assistance protection.

Kernel density estimation

Kernel density estimation was used to describe the dynamic distribution of medical assistance protection levels. This nonparametric method estimates the probability density function of a random variable and can reveal changes in distribution location, dispersion, peak shape, and polarization. In this study, kernel density estimation was applied to examine the evolutionary trends of medical assistance protection at the national level and across the eastern, central, and western regions.

Assuming that  X is a random variable with density function , the kernel density estimator is defined as:

where  n denotes the number of observations, denotes the observed value, denotes the kernel function, and  h denotes the bandwidth. The Gaussian kernel function was used in this study:

Panel Tobit model and robustness strategy

Given that the dependent variable is a comprehensive index of medical assistance protection bounded between 0 and 1, a random-effects panel Tobit model with lower and upper bounds at 0 and 1 was adopted as the benchmark specification. Compared with a conventional linear panel model, this specification accounts for the limited range of the dependent variable.

Let denote the observed medical assistance protection level of province  i in year  t, and let denote the corresponding latent protection capacity. The model was specified as follows:

where , , , , and denote per capita fiscal input for medical assistance, fiscal autonomy, rural subsidized insurance coverage, hospitalization rate among assistance recipients, and the number of health technicians per 1,000 population, respectively. denotes the m-th control variable, including real GDP per capita, urbanization rate, population aging, and education level. captures unobserved province-specific heterogeneity, and is the idiosyncratic error term.

The random-effects specification was used because the data consist of repeated observations for 31 provinces from 2019 to 2023. A likelihood-ratio test was conducted to assess whether the random-effects Tobit model was preferable to a pooled Tobit model. To further examine model robustness, several sensitivity analyses were conducted, including linear fixed-effects regression, pooled Tobit regression, replacement of the dependent variable, replacement of core explanatory variables, adjusted-sample estimation, IV-2SLS estimation, and fractional logit regression.

Interrupted time series analysis

Interrupted time series analysis (ITS) was used to isolate the potential impact of the COVID-19 pandemic from underlying temporal trends. ITS is a quasi-experimental design commonly used to evaluate the effects of policy shocks or external disruptions. The model was specified as follows:

where denotes the national average medical assistance protection level at time ; is a continuous time variable, with 2019 coded as 1, 2020 as 2, and so forth; is a dummy variable coded as 0 before the intervention and 1 after the intervention beginning in 2020; and counts the time elapsed since the intervention.

The counterfactual trend was projected using the pre-pandemic trajectory, under the assumption that the linear trend would have continued in the absence of COVID-19. The difference between the observed and counterfactual values was used to estimate the attributable pandemic-related change.

Data sources and indicator selection

Data sources

This study used panel data from 31 Chinese provinces from 2019 to 2023. Data were obtained from the China Statistical Yearbook, China Health Statistics Yearbook, China Civil Affairs Statistical Yearbook, China Medical Insurance Statistical Yearbook, provincial statistical yearbooks, and official websites of medical insurance bureaus. Missing values, mainly occurring in 2023 because of reporting delays, were imputed using linear interpolation. This method is appropriate for short-panel fiscal and health indicators with strong temporal continuity. As a robustness check, three imputation approaches were compared: linear interpolation, last observation carried forward (LOCF), and trend extrapolation. The results showed that coefficient signs and statistical significance remained consistent across specifications (Table S1). Monetary indicators were adjusted using the provincial GDP deflator, with 2019 as the base year, to account for price changes.

Index selection

Based on the Andersen Behavioral Model of Health Services Use [31] and the Opinions on Reforming and Improving the Social Assistance System [32], medical assistance protection was defined as the capacity to respond to potential health needs through resource input, generate actual health benefits, and maintain long-term system sustainability. Accordingly, an evaluation index system was constructed with five primary dimensions: input of assistance resources, fairness of assistance coverage, accessibility of assistance resources, effectiveness of assistance protection, and sustainability of the assistance system (Table 1).

Table 1.

Evaluation index system and weights of medical assistance protection in China

Primary Indicator	Secondary Indicator	Tertiary Indicator	Attribute	Weight
Input of Assistance Resources	Fiscal Input	Total Expenditure of Medical Assistance Fund (10,000 CNY)	Positive	0.055
		Per Capita Fiscal Input in Medical Assistance (CNY/person)	Positive	0.052
	Institutional Resources	Coverage Rate of Designated Primary Medical Institutions (%)	Positive	0.018
		Bed-to-Nurse Ratio in Primary Medical Institutions (beds/person)	Negative	0.021
Fairness of Assistance Coverage	Population Coverage	Coverage Rate of Subsidized Insurance Participation (%)	Positive	0.036
		Connection Rate for Non-Household-Registered Population (%)	Positive	0.086
	Urban-Rural Gap	Urban-Rural Disparity in Subsidized Insurance Coverage (%)	Negative	0.005
Accessibility of Assistance Resources	Accessibility of Medical Resources	Number of Beds per Thousand Population in Medical Institutions (beds/1,000 people)	Positive	0.023
		Number of Health Technicians per Thousand Population (persons/1,000 people)	Positive	0.028
		Number of Primary Medical Institutions per Ten Thousand Population (institutions/10,000 people)	Positive	0.035
Effectiveness of Assistance Protection	Cost Reduction	Proportion of Out-of-Pocket Expenses for assistance recipients (%)	Negative	0.053
	Hospitalization Treatment	Hospitalization Rate of assistance recipients (%)	Positive	0.033
Sustainability of Assistance System	Fiscal Sustainability	Fund Protection Rate for Medical Assistance Recipients (%)	Positive	0.115
		Annual Growth Rate of Per Capita Subsidized Insurance Intensity (%)	Positive	0.411
	Sustainability of Health Input	Proportion of Total Health Expenditure in GRP (%)	Positive	0.028

For the input of assistance resources, four indicators were included based on Zheng and Shen [33] and public finance theory [34]: total expenditure of the medical assistance fund, per capita fiscal input, coverage rate of designated primary medical institutions, and the bed-to-nurse ratio in primary medical institutions. Fiscal investment reflects the scale of available funds and the capacity to finance assistance benefits. The bed-to-nurse ratio was treated as a negative indicator because a higher ratio indicates relatively insufficient nursing resources and weaker service capacity.

For fairness of assistance coverage, three indicators were included based on Bai [35], Hui [36], and horizontal equity in healthcare [37, 38]: subsidized insurance participation rate, connection rate for non-household-registered populations, and urban-rural disparity in subsidized insurance coverage. A higher connection rate reflects stronger capacity to include migrant populations in the medical assistance system. In contrast, urban-rural disparity was treated as a negative indicator because larger gaps indicate residence-based inequity in assistance coverage.

For accessibility of assistance resources, three indicators were included based on Zhang et al. [39] and health service accessibility theory [40]: number of beds per thousand population, number of health technicians per thousand population, and number of primary medical institutions per ten thousand population. These indicators reflect the availability of healthcare resources and the practical conditions under which assisted populations can access medical services.

For effectiveness of assistance protection, two indicators were included based on Zhang et al. [39] and the concept of financial protection in universal health coverage [41]: proportion of out-of-pocket expenses for recipients and hospitalization rate of assistance recipients. The proportion of out-of-pocket expenses directly reflects the degree of financial burden relief and was treated as a negative indicator. The hospitalization rate captures whether medical assistance is translated into actual inpatient service use.

For sustainability of the assistance system, three indicators were included based on the requirements of the 14th Five-Year Plan for National Medical Security regarding basic protection, sustainability, and prevention of fund operation risks [42]: fund protection rate, annual growth rate of per capita subsidized insurance intensity, and proportion of total health expenditure in GRP. These indicators reflect the adequacy of fund reserves, the growth of per capita subsidy intensity, and the relative priority of health expenditure in regional development.

The entropy-weighted TOPSIS method was then used to standardize indicator values, determine indicator weights, and calculate the comprehensive index through weighted summation. Higher values of positive indicators indicate stronger medical assistance protection, whereas higher values of negative indicators indicate weaker protection.

Explanatory variable selection and measurement

To identify the influencing factors of provincial medical assistance protection, this study selected core explanatory variables based on public finance theory and Andersen’s behavioral model of health service utilization, while also drawing on the studies of Liu et al. [1], Hao et al. [14], Ma et al. [15], and Ren et al. [23]. The selection of variables also considered theoretical relevance, measurability, comparability, and provincial-level data availability. Accordingly, per capita fiscal input for medical assistance, fiscal autonomy, rural subsidized insurance coverage, hospitalization rate among assistance recipients, and the number of health technicians per 1,000 population were selected as the core explanatory variables. In terms of measurement principles, this study prioritized per capita indicators, ratios, and coverage rates to reduce the influence of differences in provincial population size and to enhance comparability across provinces and years. Monetary variables were adjusted for price changes where necessary.

Specifically, public finance theory emphasizes the importance of government fiscal input and public resource allocation capacity for the operation of social assistance and medical security systems [43]. Therefore, per capita fiscal input for medical assistance and fiscal autonomy were used to capture fiscal resource input and local fiscal support capacity, respectively. Andersen’s behavioral model of health service utilization emphasizes that healthcare utilization is influenced by institutional accessibility, service needs, and healthcare resource supply [44]. Therefore, rural subsidized insurance coverage, hospitalization rate among assistance recipients, and the number of health technicians per 1,000 population were used to reflect institutional coverage, actual healthcare utilization, and healthcare resource supply capacity, respectively.

In terms of specific measurement, per capita fiscal input for medical assistance was measured as annual medical assistance fund expenditure divided by the year-end resident population. Fiscal autonomy was measured as the ratio of general public budget revenue to general public budget expenditure. Rural subsidized insurance coverage was measured as the number of rural residents receiving subsidized insurance participation through medical assistance divided by the number of minimum living allowance recipients. The hospitalization rate among assistance recipients was measured as the proportion of hospitalized recipients among all medical assistance beneficiaries. The number of health technicians per 1,000 population was used to measure provincial healthcare service supply capacity.

In addition, real GDP per capita, urbanization rate, population aging, and education level were included as control variables to reduce potential estimation bias caused by differences in provincial socioeconomic development and population structure. Real GDP per capita was used to control for regional economic development and was adjusted using the price index. The urbanization rate was measured as the share of the urban population in the total resident population. Population aging was measured as the proportion of residents aged 65 years and above. Education level was measured as the natural logarithm of the number of college students per 100,000 population.

Results

National and regional trends in medical assistance protection levels

Figures 1 and 2 show the national and regional trends in medical assistance protection levels from 2019 to 2023. At the national level, the mean index showed a V-shaped pattern. It decreased from 0.266 in 2019 to 0.166 in 2020, and then gradually increased to 0.205 in 2023, although it remained below the 2019 level. By contrast, the median remained stable at 0.192 throughout the study period. The shift in the relationship between the mean and median suggests a change in the distribution of provincial protection levels. In 2019, the mean was higher than the median, indicating a relatively right-skewed distribution. After 2020, the mean was lower than the median, suggesting that a small number of provinces with relatively low scores pulled down the national average.

Fig. 1.

National trend in medical assistance protection levels in China, 2019–2023

Fig. 2.

Regional trends in medical assistance protection levels in China, 2019–2023

At the regional level, all three regions experienced a clear decline in 2020, followed by gradual recovery from 2021 to 2023. The western region consistently showed the highest medical assistance protection level, with values of 0.300 in 2019, 0.180 in 2020, and 0.215 in 2023. The eastern and central regions showed relatively similar trajectories, but their ranking varied slightly across years. In 2023, the western region remained the highest, followed by the eastern region and then the central region. This pattern differs from the usual economic development gradient, suggesting that medical assistance protection may be shaped not only by regional economic development, but also by targeted fiscal support, policy priorities, healthcare costs, and population needs. Despite the recovery after 2020, regional gaps persisted, and protection levels in all regions remained relatively modest, indicating room for further improvement.

Pandemic-related changes in medical assistance protection

To account for the potential influence of the COVID-19 pandemic, interrupted time series analysis was performed (Fig. 3). The observed medical assistance protection level in 2020 was substantially lower than the counterfactual value, indicating an apparent pandemic-related decline. The observed value decreased to 0.166 in 2020, whereas the counterfactual trajectory projected a higher level for the same year. Although the observed index gradually increased after 2020, it remained below the counterfactual trajectory through 2023, suggesting that the recovery was incomplete during the study period.

Fig. 3.

Interrupted time series analysis of medical assistance protection in China

The phased regression results further support this pattern (Table S2). Compared with the reference period, namely Phase 1 (2019), both Phase 2 (2020) and Phase 3 (2021–2023) showed significantly lower protection levels, with coefficients of -0.114 and − 0.108, respectively (both p < 0.01). The Wald test showed no statistically significant difference between Phase 2 and Phase 3 (χ² = 0.03, p = 0.86), suggesting that the post-2020 recovery remained limited rather than returning to the pre-pandemic trajectory.

Provincial-level medical assistance protection

Figure 4 presents the average medical assistance protection levels across 31 provinces from 2019 to 2023. The results show clear provincial heterogeneity. Gansu recorded the highest average protection level, followed by Qinghai, Shaanxi, Shanghai, Zhejiang, Beijing, Jiangsu, Guizhou, and Xinjiang, all of which showed relatively high levels of medical assistance protection. By contrast, Liaoning, Tianjin, Shanxi, Jilin, Hainan, Fujian, Henan, and Inner Mongolia remained at relatively low levels.

Fig. 4.

Average medical assistance protection levels across 31 Chinese provinces, 2019–2023

The provincial distribution suggests that medical assistance protection does not simply follow the conventional pattern of regional economic development. Several western provinces ranked among the higher-scoring provinces, whereas some economically developed eastern provinces did not show clear advantages. This pattern may be related to targeted fiscal support, policy orientation, healthcare cost pressure, and the needs of vulnerable populations. Overall, the results indicate marked provincial differences in medical assistance protection, suggesting that provinces with relatively low protection levels still require further policy attention.

Dagum Gini coefficient decomposition results

Dagum’s Gini coefficient decomposition was applied to analyze regional disparities in medical assistance protection levels. The overall Gini coefficient declined from 0.155 in 2019 to 0.104 in 2023 (average annual decrease 7.75%), indicating narrowing regional disparities (Table 2; Fig. 5). The primary source of disparity shifted from inter-regional to hyper-variable density. Inter-regional disparity contributed 44.6%, 41.9%, and 42.2% in 2019, 2021, and 2022, respectively. However, hyper-variable density became the dominant contributor in 2020 (36.0%) and increased to 40.5% by 2023, while intra-regional disparity (31.8%) exceeded inter-regional disparity (27.6%). This shift suggests disruption of the rigid hierarchical regional pattern and highlights the growing influence of outlier provinces on overall disparity.

Table 2.

Dagum Gini coefficient and decomposition results of medical assistance protection in China, 2019–2023

Year	Overall	Intra-regional Disparity			Inter-regional Disparity			Contribution Rate (%)
		Eastern	Central	Western	Eastern-Central	Eastern-Western	Central-Western	Intra-regional	Inter-regional	Hyper-variable Density
2019	0.155	0.133	0.093	0.169	0.11865	0.16609	0.16098	32.315	44.625	23.060
2020	0.109	0.070	0.110	0.106	0.09584	0.11773	0.10294	31.853	32.111	36.036
2021	0.102	0.098	0.060	0.101	0.08874	0.10446	0.10168	31.649	41.898	26.453
2022	0.112	0.124	0.078	0.087	0.11002	0.11209	0.10218	30.328	42.242	27.430
2023	0.104	0.120	0.049	0.095	0.09982	0.11099	0.09018	31.847	27.621	40.532

Fig. 5.

Magnitude and contribution rates of regional disparities in medical assistance protection in China. a Overall and intra-regional disparities; b inter-regional disparities; c sources of disparities and their contribution rates

Intra-regional disparity decreased from 0.133 in 2019 to 0.120 in 2023 (9.8% decline). The central region showed the largest reduction, falling from 0.093 to 0.049 (47.3%), followed by the western region, which declined from 0.169 to 0.095 (43.8%). The eastern region decreased from 0.133 to 0.120. Despite improvements, intra-regional disparity remained evident across regions. The western region showed the largest intra-regional disparity in 2019, while the eastern region became more prominent in 2022 and 2023.

Inter-regional disparity was largest between the eastern and western regions, followed by the central and western regions, and smallest between the eastern and central regions. Disparities decreased significantly in 2020 before gradually returning to previous levels. By 2022, the east-central difference (0.110) exceeded the central-west difference (0.102), suggesting the central region narrowed its gap with the western region during post-pandemic recovery while widening its gap with the eastern region.

Evolutionary trends of medical assistance protection in China

Evolutionary trends at the national level

Figure 6a illustrates the dynamic distribution of national medical assistance protection levels from 2019 to 2023. The center of the kernel density curve shifted left from 2019 to 2020, reflecting a decline in protection levels, and then gradually moved right after 2020, indicating a gradual recovery. The narrowing right tail after 2020 suggests that gaps between provinces decreased to some extent. The height of the main peak dropped markedly in 2020, while the distribution width first expanded and then narrowed, indicating widened differences during 2020 followed by partial convergence. The distribution evolved from a bimodal pattern in 2019 to a single wide peak in 2020, and then to a single-peak long-tail distribution after 2021, without clear polarization. These patterns suggest temporary convergence in 2020, followed by renewed differentiation. The increase in the number of provinces with relatively high protection levels after 2020 suggests that some lower-level provinces gradually improved, although geographical, fiscal, and resource disparities continued to limit short-term equilibrium.

Fig. 6.

Dynamic distribution of medical assistance protection at the national and regional levels. a National; b eastern region; c central region; d western region

Evolutionary trends at the regional level

Figures 6b–d illustrate the dynamic trends across the three regions. In the eastern region, the curve center shifted sharply left from 0.25 in 2019 to 0.15 in 2020, and then recovered to around 0.20 after 2020, indicating a substantial decline followed by recovery. The distribution changed from relatively dispersed in 2019 to more concentrated in 2020, and then became more dispersed again from 2021 to 2023. The peak pattern evolved from bimodal in 2019 to a single wide peak in 2020, and then to a main peak with a side peak after 2021. This suggests initial convergence followed by renewed divergence between high-level provinces, such as Shanghai and Zhejiang, and lower-level provinces, such as Tianjin and Hainan.

In the central region, the leftward shift was relatively small, followed by a steady rightward movement after 2020, suggesting a smaller decline and more stable recovery. The peak remained relatively stable, while the distribution width gradually expanded. The curve showed a consistent “main peak + side peak” pattern, with the side peak decreasing over time, indicating persistent gaps but weakening polarization.

In the western region, the curve shifted in both directions but remained at a relatively high position, with limited fluctuations. The distribution first converged and then diverged slightly, while the peak height first decreased and then increased, and the distribution width narrowed substantially. The western region maintained a single-peak pattern, with the main peak shifting right and increasing in height after 2020, suggesting no clear polarization and a gradual movement toward higher protection levels.

Inter-provincial heterogeneity within regions

Table S3 presents inter-provincial heterogeneity within each region based on medical assistance protection levels. The eastern region showed the highest internal variation, with a mean spatial coefficient of variation of 0.200. It was stratified into three tiers: a high-level cluster, including Shanghai (0.244), Zhejiang (0.238), Beijing (0.231), and Jiangsu (0.228); a middle group, ranging from Guangdong to Fujian (0.167–0.192); and a low-level cluster, including Hainan, Tianjin, and Liaoning (< 0.170). The inter-tier gap narrowed temporarily in 2020 (0.063), but expanded again to 0.097 by 2023.

The central region showed the lowest internal disparity, with the spatial coefficient of variation declining from 0.183 in 2019 to 0.113 in 2023, indicating a converging trend in medical assistance protection. In terms of protection levels, Jiangxi showed the largest improvement (+ 29.8%), whereas Shanxi declined substantially (− 43.7%, from 0.250 to 0.141).

The western region, despite having the highest regional mean protection level (0.222), showed substantial internal stratification, with a spatial coefficient of variation of 0.160. Upper-tier provinces, ranging from Gansu (0.299) to Guizhou (0.240), maintained markedly higher levels than lower-tier provinces, ranging from Inner Mongolia (0.174) to Ningxia (0.191), with an internal gap of 0.125. Temporal stability also varied considerably across western provinces, ranging from Shaanxi (temporal coefficient of variation = 0.753) to Tibet (temporal coefficient of variation = 0.067).

Influencing factors of medical assistance protection in China

To further examine the influencing factors of medical assistance protection, a panel Tobit model was estimated using the comprehensive index as the dependent variable. Table 3 reports the benchmark regression results for the national sample and the three regional subsamples.

Table 3.

Tobit regression results of influencing factors on medical assistance protection at the national and regional levels

Variable	(1)National	(2)Eastern	(3)Central	(4)Western
Per Capita Fiscal Input in Medical Assistance (fin)	0.001** (0.0004)	0.002*** (0.001)	0.002*** (0.0006)	0.001 (0.0006)
Fiscal Autonomy (bud)	0.204*** (0.053)	0.170** (0.068)	0.013 (0.066)	0.107 (0.157)
Rural Subsidized Insurance Coverage Rate (cover)	0.125 (0.090)	0.099 (0.373)	-0.160 (0.277)	0.144 (0.121)
Hospitalization Rate of Assistance Recipients (hosp)	0.630*** (0.103)	0.682*** (0.203)	0.566*** (0.116)	0.416** (0.184)
Number of Health Technicians per Thousand Population (med)	0.011* (0.005)	0.150*** (0.004)	0.000 (0.009)	0.022* (0.013)
Real Per Capita GDP (gdp)	-0.0001*** (0.0001)	-0.0001** (0.0001)	-0.0001** (0.0001)	-0.0001*** (0.0001)
Urbanization Rate (urban)	-0.075 (0.098)	0.035 (0.126)	-0.217 (0.207)	-0.125 (0.159)
Population Aging Rate (old)	-0.252 (0.199)	-0.171 (0.191)	0.333 (0.490)	-0.003 (0.420)
Higher Education Level (edu)	-0.065** (0.027)	-0.097*** (0.028)	-0.044 (0.038)	-0.023 (0.045)
Con	0.525*** (0.200)	0.634** (0.261)	0.536* (0.300)	0.266 (0.316)
sigma_u	0.019*** (0.006)	0.000 (0.008)	0.008 (0.006)	0.018 (0.012)
sigma_e	0.411*** (0.027)	0.029*** (0.003)	0.020*** (0.003)	0.046*** (0.005)
Obs	155	55	40	60

*p < 0.1, **p < 0.05, ***p < 0.01; robust standard errors in parentheses

Benchmark regression results

1. National-level regression results

The likelihood-ratio test confirmed the suitability of the random-effects panel Tobit model (p < 0.01). At the national level, per capita fiscal investment in medical assistance was significantly and positively associated with protection levels (β = 0.001, p < 0.05), indicating that fiscal investment remains an important basis for improving medical assistance protection. Fiscal autonomy also showed a significant positive association (β = 0.204, p < 0.01), suggesting that greater local fiscal capacity may support more flexible resource allocation and policy implementation. The hospitalization rate among assistance recipients was positively associated with protection levels (β = 0.630, p < 0.01), indicating that actual inpatient service utilization is closely related to the effectiveness of medical assistance. Health technician density per 1,000 population showed a marginally significant positive association (β = 0.011, p < 0.10), suggesting that stronger health workforce capacity may contribute to improved medical assistance protection.

Among the control variables, real GDP per capita showed a significant negative association with medical assistance protection (β = -0.0001, p < 0.01). This counterintuitive result may reflect a “welfare paradox,” whereby the relative role of medical assistance becomes less prominent in economically developed areas with more comprehensive basic medical insurance coverage. Higher education level also showed a significant negative association (β = -0.065, p < 0.05), which may be related to greater health literacy, stronger disease prevention awareness, and lower reliance on medical assistance among more educated populations. Urbanization rate (β = -0.075) and population aging rate (β = -0.252) were not statistically significant, indicating that their associations with medical assistance protection were not evident in the current model.

• (2) Regional-level regression results

Given the spatial heterogeneity of medical assistance protection across China, regional regressions were conducted for the eastern, central, and western regions. A random-effects panel Tobit model was used, and the results are presented in columns (2)–(4) of Table 3.

Per capita fiscal investment in medical assistance was significantly and positively associated with protection levels in the eastern region (β = 0.002, p < 0.01) and the central region (β = 0.002, p < 0.01), but no statistically significant association was observed in the western region (β = 0.001, p > 0.05). This regional difference may be related to the fact that fiscal resources in western China are often directed toward basic medical and public health services, which may weaken the observable marginal association between medical assistance investment and protection levels. Fiscal autonomy showed a significant positive association only in the eastern region (β = 0.170, p < 0.05), suggesting that stronger fiscal autonomy may be more easily translated into medical assistance protection in economically developed areas. No significant associations were observed in the central region (β = 0.013, p > 0.05) or western region (β = 0.107, p > 0.05).

The hospitalization rate among assisted populations was significantly and positively associated with medical assistance protection in all three regions (eastern: β = 0.682, p < 0.01; central: β = 0.566, p < 0.01; western: β = 0.416, p < 0.05). This suggests that inpatient service utilization is an important channel through which medical assistance may be transformed into actual protection. Health technician density per 1,000 population showed significant positive associations in the eastern region (β = 0.150, p < 0.01) and western region (β = 0.022, p < 0.10), but no significant association in the central region (β = 0.000, p > 0.05). These results suggest that health workforce capacity may support medical assistance protection, although its role differs across regions.

Among the control variables, real GDP per capita showed consistently negative associations with medical assistance protection across all regions (eastern: β = -0.0001, p < 0.05; central: β = -0.0001, p < 0.05; western: β = -0.0001, p < 0.01), consistent with the national-level result. Higher education attainment was significantly and negatively associated with medical assistance protection only in the eastern region (β = -0.097, p < 0.01), while no significant associations were observed in the central region (β = -0.044, p > 0.05) or western region (β = -0.023, p > 0.05). This may indicate that the “welfare paradox” is more evident in economically developed regions. Population aging showed a positive but non-significant association in the central region (β = 0.333, p > 0.05), while no significant associations were observed in the eastern region (β = -0.171) or western region (β = -0.003). Urbanization rate was not significantly associated with medical assistance protection in any region (eastern: β = 0.035; central: β = -0.217; western: β = -0.125; all p > 0.05).

Variable definitions: fin = per capita fiscal input in medical assistance; bud = fiscal autonomy; cover = rural subsidized insurance coverage rate; hosp = hospitalization rate of assistance recipients; med = health technicians per thousand population; gdp = real per capita GDP; urban = urbanization rate; old = population aging rate; edu = higher education level.

Robustness tests

To examine the reliability of the benchmark Tobit estimates, several robustness checks were conducted from four aspects: model specification, variable measurement, sample composition, and potential endogeneity (Table 4 and Table S4).

Table 4.

Robustness checks

Variable	(1) Linear FE	(2) Pooled Tobit	(3) Replace DV	(4) Replace Vars	(5) Adjust Sample	(6) IV-2SLS
fin	0.001*** (0.000)	0.001*** (0.000)	0.002*** (0.000)	—	0.001** (0.000)	0.001*** (0.000)
bud	-0.166 (0.124)	0.203** (0.046)	0.109*** (0.000)	0.141** (0.058)	0.196** (0.050)	0.007 (0.032)
cover	0.622*** (0.196)	0.135 (0.075)	0.824*** (0.000)	0.145 (0.100)	0.109 (0.085)	0.268*** (0.047)
hosp	0.581*** (0.108)	0.566** (0.098)	0.533*** (0.000)	0.829*** (0.109)	0.653*** (0.089)	0.210*** (0.061)
med	-0.007 (0.238)	0.012** (0.004)	0.010*** (0.000)	—	0.009* (0.005)	0.006* (0.003)
Med-new	—	—	—	-0.016* (0.010)	—	—
Fin-new	—	—	—	0.007*** (0.002)	—	—
gdp	-0.000 (0.000)	-0.000** (0.000)	0.000*** (0.000)	-0.000*** (0.000)	-0.000** (0.000)	0.000*** (0.000)
urban	-2.237** (0.784)	-0.094 (0.080)	0.075*** (0.000)	0.004 (0.095)	-0.064 (0.094)	-0.152** (0.054)
old	-0.284 (0.923)	-0.239 (0.158)	0.460*** (0.000)	0.014 (0.287)	-0.201 (0.194)	0.150 (0.132)
edu	-0.018 (0.068)	-0.047** (0.022)	0.043*** (0.000)	-0.054* (0.028)	-0.081* (0.027)	-0.024 (0.017)
Constant	1.467*** (0.365)	0.383** (0.156)	-0.490*** (0.000)	0.531*** (0.201)	0.657** (0.189)	0.257** (0.120)
sigma\_u	0.303	—	—	0.215	0.216	—
sigma\_e	0.033	0.029	—	0.041	0.034	—
rho	0.988	—	—	0.187	0.29	—
First-stage F	—	—	—	—	—	79.29**
DWH test	—	—	—	—	—	2.22
Observations	155	155	155	155	155	124

*p < 0.1, **p < 0.05, ***p < 0.01; robust standard errors in parentheses

First, alternative model specifications were estimated. The linear fixed-effects model and pooled Tobit model produced results that were broadly consistent with the benchmark random-effects Tobit model, indicating that the main findings were not driven solely by the baseline panel Tobit specification. Second, the dependent variable and core explanatory variables were replaced. After recalculating the medical assistance protection index and using alternative measures for fiscal input and medical resource supply, the main coefficient directions remained generally stable. Third, the sample was adjusted by excluding extreme observations, and the key findings were not substantially changed.

Fourth, potential reverse causality was further examined using IV-2SLS estimation with lagged fiscal input as the instrumental variable. The first-stage F-statistic was 79.29, exceeding the conventional weak-instrument threshold, and the Durbin-Wu-Hausman test did not reject the exogeneity assumption. The IV estimate of fiscal input remained positive and statistically significant, suggesting that the positive association between fiscal investment and medical assistance protection was robust after considering potential endogeneity.

Finally, a fractional logit model was estimated as an additional robustness check for the bounded dependent variable (Table S4). Unlike the Tobit model, fractional logit does not rely on a censoring assumption and is suitable for fractional response variables within the interval [0,1]. The results were broadly consistent with the benchmark Tobit estimates: per capita fiscal input, fiscal autonomy, hospitalization rate, and health technician density remained positively associated with medical assistance protection, whereas real per capita GDP and higher education level showed negative associations. These findings further suggest that the main conclusions are not driven solely by the Tobit specification.

Discussion

This study examines medical assistance protection levels across 31 provinces in China from 2019 to 2023, utilizing the entropy-weighted TOPSIS method, Dagum Gini coefficient decomposition, kernel density estimation, and panel Tobit model. Regional differences, evolutionary trends, and influencing factors were systematically analyzed. The findings reveal a V-shaped trajectory in medical assistance protection levels, initially declining and then rising during the study period. Although regional disparities have narrowed, an absolute gap persists. The western region consistently leads in protection levels, while the eastern region experiences significant polarization. Meanwhile, the central region shows steady improvement, gradually closing the gap with more advanced areas. Factors such as per capita fiscal input for medical assistance, fiscal autonomy, hospitalization rate among assistance recipients, and health technician density per 1,000 population significantly enhance protection levels. Conversely, economic development and higher education levels negatively affect medical assistance protection levels. To interpret these findings theoretically, this discussion draws on China’s fiscal decentralization framework and the developmental stage theory of social assistance systems.

This study broadens the current literature in several ways. Primarily, past research has concentrated on the overall fairness [45, 46] and efficiency [47–49] of the medical security system, often neglecting regional disparities within the medical assistance system, which functions as a safety net. For the first time, this research systematically illustrates the regional disparities in China’s medical assistance protection levels, uncovering an atypical gradient pattern: levels decrease from the western to eastern, and then to central regions. This finding contrasts sharply with the traditional socioeconomic development pattern, where the eastern region typically leads [50]. To interpret this anomaly, this discussion considers the interplay between fiscal decentralization and the developmental stage theory of social assistance systems [51]. Two main factors may explain this pattern. First, the central government has directed fiscal transfer payments toward the western region, particularly through concentrated investments of special medical assistance funds during poverty alleviation campaigns [52]. Under China’s fiscal decentralization framework, such targeted transfers enable the western region to achieve relatively high protection levels that are driven mainly by central government policies and funding, rather than by local economic development. This financial support has helped establish a high-standard medical assistance system in the western region. Second, high medical costs and severe population aging in the eastern region have diluted the relative strength of medical assistance protection [53]. For instance, in Gansu Province, located in the western region, the combination of low medical service demand and consumption expenditure, along with policy preferences and fiscal subsidies from basic medical insurance, serious illness insurance, and medical assistance, results in relatively sufficient protection for residents’ medical expense compensation [39]. The finding that some economically developed provinces in the eastern region have relatively low protection levels aligns with existing research on the efficiency of medical assistance [54]. Currently, there is a misalignment between the efficiency of medical assistance across regions and China’s socioeconomic development levels. Specifically, the overall efficiency of medical assistance in the economically developed eastern region is lower than in the central and western regions, which have relatively lower economic development levels. This difference can be explained by the different developmental stages of each region’s medical assistance system. The eastern region has moved into a consolidation phase focused on efficiency and sustainability, while the western region remains in an expansion phase, where institutional growth naturally leads to higher comprehensive indices [55–57]. This phenomenon reflects the regional differentiation of the medical assistance system at various development stages, in line with dynamic socioeconomic changes. From the perspective of medical assistance demand, there is generally a negative correlation between total demand and economic development level. In regions with higher economic development, residents have more secure incomes, better health [58], and greater access to medical resources, leading to lower overall demand for safety-net medical assistance. Consequently, the output of medical assistance is relatively lower under the same investment. Additionally, as a public good, medical assistance exhibits diminishing marginal effects in its input-output relationship. Due to lower economic development, resident income levels, and accessibility to medical resources and health knowledge in the central and western regions, the demand for safety-net medical assistance is higher. Comparatively, the marginal output from medical assistance investment in the eastern region is slightly lower than in the central and western regions. Overall, the regional differentiation in medical assistance demand in China is a typical phenomenon of socioeconomic development, shaped by the country’s unique regional economic and population dynamics. It represents a normal stage in the evolution of China’s medical assistance system. Regarding financial investment intensity in medical assistance, as economic levels rise and residents’ per capita disposable incomes increase, their capacity to bear disease diagnosis expenses also strengthens. Consequently, local governments consider these factors when formulating medical assistance fiscal expenditure budgets, adjusting fiscal spending to align with local economic development levels [59].

This study identifies a significant interaction between regional disparities in medical assistance protection levels and the pandemic’s impact. The results reveal that external public health crises affect regional medical security systems differently, depending on their foundational protection levels. Using kernel density estimation for dynamic evolution analysis, the results show that medical assistance protection in the eastern region was most severely affected by the pandemic but recovered the fastest, following a trajectory of “dispersion–concentration–redispersion.” In contrast, the western region experienced minor fluctuations, with a notable narrowing of internal gaps, while the central region maintained a steady catch-up trend. However, this recovery should be interpreted with caution. Interrupted time series analysis reveals that protection levels remained significantly below pre-pandemic levels through 2023, showing no substantial improvement compared with 2020. This means the pandemic’s impact persisted—the system did not automatically return to its original trajectory, but remained at a lower equilibrium. Viewed through the lens of institutional resilience, regions responded differently to the pandemic. The eastern region showed strong resilience and recovered quickly, while the central region experienced weak resilience and stalled recovery, indicating that the pandemic trapped certain regions in a persistently low level of protection [60]. These findings align with existing literature, suggesting that regions with better medical resources exhibit stronger risk resistance and recovery resilience. Insights from experts highlight that chronic stressors, such as underfunding, staff strikes, and governance challenges, consistently challenge health systems. These stressors can exacerbate acute crises such as pandemics, implying that well-resourced systems recover more robustly, whereas poorly resourced ones may worsen public health crises [61]. Furthermore, the importance of good governance and accessible medical resources in building resilience in fragile environments suggests that these elements are crucial for efficient medical system operation and resilience during crises [62]. They help withstand external shocks, narrow regional development gaps, and achieve rapid system repair and sustainable resilience. Post-2020, regional disparities in China’s medical assistance protection levels showed a convergence trend. The hyper-variance density contribution rate rose, indicating a breakdown of the absolute hierarchical pattern between regions. From a theoretical perspective, this shift can be explained by “institutional path dependence”—the idea that past conditions shape future outcomes [63]. The pandemic served as a turning point: provinces that were already strong recovered faster, while weaker ones lagged, making intra-regional polarization worse. This shift may be attributed to the centralized leadership of the Communist Party of China and the rapid response of medical assistance policies [64]. Special transfer payments from the central government to struggling regions during the pandemic effectively eased local financial pressures, preventing further widening of regional disparities [65, 66]. Additionally, the pandemic increased reliance on medical assistance among vulnerable groups such as the elderly and low-income populations, prompting local governments to enhance medical assistance systems, ultimately boosting overall protection levels. However, this quantitative expansion occurred within the context of the depressed protection levels identified above, highlighting a tension between broader coverage and adequate depth in the post-pandemic system. This highlights an important theoretical distinction: emergency fiscal support can prevent regional disparities from worsening in the short term, but without concurrent efforts to build institutional capacity, regions may remain stuck in a persistently low level of protection—unable to recover to pre-pandemic levels.

This study quantitatively assessed the mechanisms of various influencing factors, with the hospitalization rate among assistance recipients being the most significant. Among adjustable input and demand factors, per capita fiscal input for medical assistance and fiscal autonomy showed significant positive effects. However, there was notable regional variation: financial investment significantly affected the eastern and central regions but not the western region, while fiscal autonomy was only positively significant in the eastern region. This highlights a contradiction between “input and efficiency” in the western region. Despite increased financial investment, the region’s underdeveloped economy prioritizes basic medical and public health services, limiting the marginal effect of medical assistance. In contrast, eastern region governments enjoy greater fiscal flexibility, allowing them to adjust medical assistance policies to meet local needs, thereby enhancing resource allocation efficiency. The study also identified a negative correlation between economic development and medical assistance protection levels, a phenomenon known as the “welfare paradox.” This aligns with Mackenbach’s findings that health inequality in highly developed welfare countries, such as those in the Nordic region, is not necessarily less than in countries with less comprehensive welfare systems, such as the UK or Southern European nations [67]. However, institutional analysis indicates that this paradox fundamentally reflects regional disparities in provincial fiscal burdens and beneficiary access thresholds under fiscal decentralization [24, 68]. Zhu and Hu (2019) found that economically developed eastern provinces, such as Shanghai and Beijing, set substantially higher income thresholds for low-income eligibility than less developed western regions, such as Shaanxi [24]. Similarly, criteria for identifying illness-induced poverty in wealthy provinces were 3.6 times more stringent than in poorer areas, such as Liaoning. This suggests that despite their greater fiscal autonomy [68], eastern provinces effectively restrict coverage through stricter beneficiary entry barriers. By contrast, western provinces rely heavily on central transfer payments [36], which account for 50–60% of their medical assistance funds. During the poverty alleviation campaign, these provinces adopted inclusive policies covering both registered poor households and marginal low-income groups [36, 69]. Consequently, despite lower economic development, these regions show higher levels of medical assistance protection [69]. One reason could be that high-income groups in economically developed areas benefit more from medical insurance, leading residents to rely more on basic medical insurance rather than medical assistance, resulting in a relatively lower level of medical assistance protection [70]. Additionally, the negative effect of higher education levels on medical assistance protection may stem from residents in highly educated regions having better health literacy and disease prevention awareness, thus reducing their need for medical assistance. These findings suggest that improving medical assistance protection requires more than economic growth; it demands targeted institutional design and policy intervention.

This study has significant public health implications for enhancing China’s medical assistance system and promoting health equity. The findings reveal that while medical assistance protection levels in China are increasing, they remain relatively low overall, with marked regional disparities. To address these inequalities, it is crucial to develop targeted policies for regions with low protection levels. This includes increasing financial investment in medical assistance, optimizing fiscal transfer payment mechanisms, and improving the integration of multi-level medical insurance to strengthen foundational protection. Furthermore, regional differences in fiscal autonomy and investment efficiency exacerbate inequalities in meeting individual assistance needs. Policymakers must implement interventions at both regional and individual levels to mitigate these disparities. Special attention should be given to the sustainable development of medical assistance in central and western regions, where fiscal capacity is constrained. Establishing a central adjustment system for medical assistance funds can help balance fiscal burdens across regions. Additionally, the hospitalization rate among assistance recipients positively affects all regions, suggesting that improving the actual utilization rate of the medical assistance system is an effective strategy for enhancing protection levels. Policy formulation and implementation should consider the regional socioeconomic context. Strengthening primary healthcare services, optimizing the medical assistance process, and promoting “one-stop” settlements can lower the threshold for medical service utilization among assistance recipients, thereby minimizing protection inequality.

Limitations

This study has three main limitations. First, the analysis relies on provincial statistical yearbooks, which precludes the inclusion of micro-level indicators, such as individual reimbursement ratios and out-of-pocket expenses, and limits the analysis of within-province disparities across prefectures or between urban and rural areas. Moreover, the traditional east-central-west classification may obscure intra-provincial heterogeneity, such as variations in fiscal capacity across prefectures, and inter-provincial homogeneity, such as similar reliance on central transfers across certain central and western provinces. Future research should incorporate household survey data to assess individual-level protection and adopt refined spatial clustering or urban agglomeration perspectives, such as the Yangtze River Delta, Pearl River Delta, and Chengdu-Chongqing, to better characterize spatial heterogeneity.

Second, while IV-2SLS addresses endogeneity, linear IV estimation is theoretically second-best to IV-Tobit for bounded dependent variables. However, given that the outcome variable ranges between 0.1 and 0.3, well away from the 0–1 boundaries, this approximation is reasonable. Whether 2023 levels represent genuine recovery or a new lower equilibrium remains uncertain. Future studies should extend observation beyond 2023 to distinguish temporary pandemic shocks from structural shifts.

Third, certain indicators rely on proxy variables, such as bed-to-nurse ratios for service quality, and may be affected by anomalous fluctuations in 2020 data. Qualitative comparative case studies are warranted to explain the “welfare paradox” and the institutional mechanisms underlying regional disparities.

Conclusion

In conclusion, while regional disparities in medical assistance protection in China are narrowing, significant absolute gaps and structural contradictions persist. The western region’s high level of medical assistance protection contrasts sharply with the internal variations in the eastern region. To enhance fairness and sustainability in the medical assistance system at both regional and individual levels, multi-departmental collaboration is essential. This includes optimizing fiscal investment structures, improving the efficiency of fund utilization, strengthening primary medical service capabilities, and refining the multi-level medical insurance connection mechanism. Future research should focus on clarifying the institutional mechanisms behind these disparities, exploring effective integration paths between medical assistance, basic medical insurance, and catastrophic medical insurance, and examining the impact of digital transformation on service accessibility. Such efforts will provide evidence-based support for advancing the high-quality development of medical security.

Authors’ contributions

Qing Li: Writing - review & editing, Methodology, Supervision, Writing - original draft, Conceptualization. Xiaojing Huang: Visualization, Data curation, Writing - original draft, Formal analysis, Writing - review & editing, Methodology, Conceptualization.

Funding

This study had no funding.

Data availability

The data used in this study were publicly available from official sources, including the China Health Statistics Yearbook, the China Statistical Yearbook, the China Civil Affairs Statistical Yearbook, the China Healthcare Security Statistical Yearbook, the National Bureau of Statistics (https://www.stats.gov.cn/sj/ndsj/), the National Healthcare Security Administration (https://www.nhsa.gov.cn/col/col7/index.html), as well as provincial statistical yearbooks available on the websites of provincial bureaus of statistics. All data covered the period from 2019 to 2023.

Declarations

Ethics approval and consent to participate

The raw data is collected from the China Statistical Yearbook, China Health Statistics Yearbook, China Civil Affairs Statistical Yearbook, China Healthcare Security Statistical Yearbook, National Bureau of Statistics website, National Healthcare Security Administration website, provincial statistical yearbooks from 2019 to 2023, therefore, ethical proof does not applicable to this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.