Can health poverty alleviation policies simultaneously mitigate current and future medical-related poverty? Evidence from four waves (2011–2018) of the China health and retirement longitudinal study (CHARLS)

Abstract

Background

Health shocks were a major cause of poverty in low and middle income countries, including China. China has implemented a series of Health Poverty Alleviation Policy (HPAP) policies aimed at reducing poverty driven by health related factors. This study aims to systematically evaluate the short term and long term effects of China’s HPAP on preventing health induced poverty.

Methods

Four waves of data from the China Health and Retirement Longitudinal Study (CHARLS), involving 11,380 respondents, were used in this study. We employed two indicators to estimate both current and future health-related poverty risks at the household level: the Catastrophic Health Expenditure (CHE) index and the Health Poverty Vulnerability index, calculated using a three-stage feasible generalized least squares (FGLS) method. In the research design, households that answered “yes” to the survey question “Is your household registered as a poor household?” were classified as the treatment group, while those that answered “no” were classified as the control group. A difference-in-differences (DID) model was applied to assess the impact of the Health Poverty Alleviation Policy (HPAP) on the incidence of CHE. Given that households with different levels of risk may respond differently to the policy, we further employed a quantile difference-in-differences (QDID) model to evaluate the heterogeneous effects of HPAP on health poverty vulnerability.

Results

The average household health poverty vulnerability index was 0.48 ± 0.02. Longitudinal analyses indicated a declining trend over time in both the treatment and control groups. However, the treatment group consistently exhibited higher vulnerability levels than the control group across all survey waves. Furthermore, the incidence of CHE in the treatment group showed a sustained downward trajectory. difference-in-differences (DID) analysis demonstrated that the HPAP significantly reduced the incidence of CHE. However, its impact on overall health poverty vulnerability was limited, with only modest improvements observed among high-risk subpopulations.

Conclusions

The HPAP policy achieved notable effects in alleviating the direct economic burden caused by illness, but its impact on the long-term health poverty vulnerability of uncompensated low-to-medium-risk households remains limited. This disparity in effects primarily stems from the policy’s focus on medical cost control, which, while helpful in reducing CHE, lacks systematic investment in health capacity building. Future policy design should strengthen risk identification and dynamic monitoring mechanisms, expand the scope of coverage for preventive care to achieve sustainable mitigation of health poverty vulnerability.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13690-025-01821-y.

Text box 1. Contributions to the literature
• This study is the first to systematically evaluate the long-term impact of China’s Health Poverty Alleviation Policy (HPAP) on health poverty vulnerability using nationally representative data.
• Given that poverty induced by illness is a dynamic process, we further applied a quantile difference-in-differences (QDID) model to explore heterogeneous effects across risk subgroups.
• By integrating static poverty indicators with forward-looking risk estimates, this study fills a critical empirical gap in the dynamic evaluation of health poverty alleviation efforts.
• It shows that HPAP significantly alleviates the direct economic burden of illness but only modestly reduces long-term health poverty vulnerability, informing the refinement of risk-oriented health poverty alleviation policies.

Background

Poor health has long been recognised as a significant risk factor for poverty. Prior to 2020, globally, elevated healthcare costs had pushed or further entrenched over half a billion people into extreme poverty [1]. This trend was particularly pronounced in China, where data from the State Council Leading Group Office of Poverty Alleviation revealed that illness-related impoverishment remained alarmingly prevalent [2]. As such, addressing the health challenges faced by impoverished households continues to be central to the development of poverty alleviation policies at both the national and regional levels.

In 2015, the Chinese government formally implemented the Targeted Poverty Alleviation (TPA) policy, which sought to precisely identify the remaining impoverished population and address the diverse causes of poverty through a series of comprehensive measures, including industrial development, relocation, ecological compensation, educational support, and social security programs. The policy aimed to lift poor households out of economic hardship. To more effectively identify the extremely poor, TPA shifted the focus of anti-poverty efforts from region-based initiatives to household-level precision targeting, and established a poverty household registry system (“Jian Dang Li Ka”) based on multidimensional indicators [3]. This registry recorded the causes and needs of each household’s poverty, thereby facilitating targeted assistance for registered poor families concentrated in impoverished areas [4].

According to the registry data, more than 40% of China’s impoverished population had fallen into poverty due to illness, and over 14% due to disability [5]. Recognizing health as a key factor contributing to the persistence of poverty, the government subsequently developed and implemented the Health Poverty Alleviation Policy (HPAP) as a complementary initiative to TPA. The HPAP primarily covered households that had been registered as poor under the TPA’s “Jian Dang Li Ka” system and identified as medically vulnerable populations. Building upon the precision identification mechanism, this policy established a systematic health registry nationwide, providing tailored medical protection and targeted health support for households impoverished by illness [6]. Details of the policy framework are presented in Fig. 1.

Fig. 1.

The linkages between the Targeted Poverty Alleviation Policy (TPA) and the Health Poverty Alleviation Policy (HPAP)

The core strategies of the Health Poverty Alleviation Policy (HPAP) are structured around four key interventions [6]: First, medical insurance coverage is expanded through integrated schemes, with reimbursement rates being increased and a “diagnose first, pay later” system being implemented; Second, tiered treatment strategies are provided for rural poor populations suffering from major and chronic diseases, while basic medical, public health, and health management services are extended; Third, healthcare infrastructure in impoverished areas is strengthened, with facilities and staffing in county hospitals, township health centers, and village clinics being upgraded; Fourth, chronic, infectious, and endemic diseases are targeted through improved public health interventions, with prevention efforts being intensified.

Although these policies have effectively alleviated the economic burden of disease in the short term, the risk of poverty induced by health shocks remains a dynamic, long-term, and concealed process. Such risks do not occur as isolated events but rather evolve over time. At any given point, households may fall into poverty due to unexpected health shocks [7]. These shocks often increase medical expenditures, thereby depleting household capital, reducing total labor capacity, and lowering per capita resources [8]. The degree of deterioration in health and welfare after a shock depends heavily on household-specific characteristics [9]. Households with stronger socioeconomic foundations tend to recover more easily, while those in disadvantaged settings are more likely to become trapped in a vicious cycle of “disease and poverty” [10].

Traditional poverty measurement approaches primarily capture static economic burdens and thus fail to reflect the forward-looking risks of poverty. In contrast, probabilistic risk indicators offer an ex-ante analytical framework to quantify the potential damage to households’ income-generating capacity under health risks. This dynamic approach enables the depiction of poverty transition trajectories over time, providing valuable guidance for the optimization of health-oriented poverty alleviation policies (Figure 2).

Fig. 2.

Health poverty dynamic process display chart

Within this framework, health poverty vulnerability has been identified as a crucial, forward-looking indicator for measuring the risk of illness-induced impoverishment. It is defined as the probability that a household will experience asset depletion or a decline in living standards below the societal threshold due to health shocks [11]. Chaudhuri’s seminal work first operationalized the concept of vulnerability by estimating the probability of future poverty, while health poverty vulnerability specifically measures susceptibility to medical impoverishment [12]. This indicator serves both as a predictive tool and an early warning mechanism for recurrent health-related poverty [13]. By shifting from static poverty evaluation to dynamic vulnerability analysis, policymakers can transition from reactive compensation to proactive prevention, thereby enhancing the efficiency of poverty governance.

However, existing research has mainly focused on assessing the short-term economic effects of HPAP, with limited attention paid to the temporal dynamics of health poverty vulnerability. Most studies rely on cross-sectional data and lack longitudinal empirical evidence capable of capturing the long-term impacts of HPAP on poverty risks (Figure 3).

Fig. 3.

Schematic representation of the study’s methodological framework

To address these gaps, this study adopts a longitudinal and spatiotemporal perspective to systematically analyze the evolution of health poverty vulnerability before and after the implementation of HPAP. By integrating cross-sectional benchmark indicators of illness-induced poverty with dynamic vulnerability measures, this research compares the short-term and long-term impacts of HPAP, offering a comprehensive understanding of the dynamic transformation of health-induced poverty. The findings provide theoretical and empirical evidence for optimizing China’s health poverty alleviation strategies, mitigating the risk of illness-induced impoverishment, and establishing a sustainable, health-oriented poverty prevention mechanism.

Literature review

The core objective of the Health Poverty Alleviation Policy (HPAP) is to reduce the financial burden of medical expenses on impoverished populations and to improve the accessibility of healthcare services, thereby preventing poverty caused or aggravated by illness. Previous studies have primarily focused on the short-term effects of HPAP in alleviating the economic burden of medical care. Zhang et al. (2019) found that HPAP achieved a moderate level of success in the surveyed regions, significantly reducing patients’ financial pressure [14]. Lu et al. (2021), using internal data from Chifeng City’s government spanning 2017 to 2020, demonstrated that the incidence and intensity of catastrophic health expenditures decreased by 17.1% and 31.2%, respectively, with more pronounced effects among households in poor health and extreme poverty [15]. Li et al. (2022) further reported that in the year following implementation, out-of-pocket expenses decreased by 72.8% for outpatient care and by 89.39% for inpatient care [16]. Collectively, these findings confirm the positive short-term impact of HPAP in mitigating the medical financial burden of poor households.

However, several limitations persist within the existing literature. Unlike direct fiscal subsidies and other short-term interventions, the effects of HPAP are often latent and delayed, making its true impact neither immediately observable nor easily quantifiable. Moreover, most studies rely on cross-sectional data and primarily emphasize the reduction of current medical expenditures, paying insufficient attention to how HPAP influences the long-term evolution of poverty risks among households. In addition, empirical analyses exploring the spatiotemporal dynamics of health-induced poverty risks across different time periods and regions remain limited. In other words, current research has largely documented the policy’s immediate relief effects while neglecting its sustained mechanisms in preventing poverty recurrence and promoting long-term poverty alleviation.

In light of these gaps, this study adopts health poverty vulnerability as an analytical framework and employs nationally representative data to comprehensively examine both the short- and long-term effects of HPAP on poverty risk. By integrating static and dynamic indicators, the study systematically investigates the dynamic transformation process of health-induced poverty, thereby providing both theoretical and empirical support for the continued optimization of HPAP and the establishment of a sustainable, health-oriented poverty prevention mechanism.

Methods

Data sources

Data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal survey targeting individuals aged 45 and above to address aging-related challenges, were utilized in this study. CHARLS was jointly administered by Wuhan University and Peking University, with its baseline survey initiated in 2011, covering 150 counties, 450 villages, and approximately 10,000 households (17,000 individuals). Four waves of data (2011–2018) were analyzed, encompassing multidimensional metrics: demographic characteristics, health status, healthcare utilization, and socioeconomic indicators [17].

Missing values and outliers were systematically addressed using cross-sectional linkages between datasets for multilevel imputation. For irreparable within-wave gaps, adjacent-year imputation was applied if variables exhibited stability (p > 0.05 in annual comparisons). Remaining anomalies underwent rigorous exclusion. Final analyses employed a balanced longitudinal sample (2011, 2013, 2015, 2018) comprising 11,380 individuals, with sampling stratification detailed in Fig. 4.

Fig. 4.

Flowchart of Sample Selection from the China Health and Retirement Longitudinal Study (CHARLS)

Measurements

To comprehensively assess the impact of health issues on poverty, a dynamic and static analytical framework was adopted in terms of measurement methods. In the dynamic dimension, Chaudhuri’s concept of expected poverty vulnerability (VEP) was employed, and the probability of households falling into poverty due to health issues in the future was measured using a three-stage feasible generalised least squares (FGLS) method [12]. This method corrected for heteroskedasticity and accounted for heterogeneity in household-level consumption differences. This approach enabled a forward-looking assessment of poverty vulnerability under health risks. In the static analysis dimension, catastrophic health expenditure (CHE) was used as a cross-sectional benchmark indicator for health-induced poverty, calculated according to World Health Organisation standards. This dual-perspective measurement framework allowed both the direct economic burden of healthcare and the long-term risk of falling into poverty due to health shocks to be captured.

Measurement of dynamic indicators

Health poverty vulnerability refers to the probability that an individual or household will fall into a low-welfare state due to health-related risks, reflecting both the likelihood of falling into poverty in the future and the persistence of current poverty conditions [18]. The concept of Vulnerability to Expected Poverty (VEP) proposed by Chaudhuri was employed, and the probability of households falling into poverty due to health issues was estimated using the Three-Stage Feasible Generalized Least Squares (FGLS) method [12]. According to the VEP framework, the health poverty vulnerability index is a continuous variable truncated between 0 and 1, where a higher value indicates a greater degree of vulnerability under health risks. The specific steps were as follows:

1

The vulnerability to poverty of household i at time t was represented by , the welfare level of household i at time t + 1 was represented by , the poverty line was represented by , and the probability that the welfare level (income or expenditure) of household i at time t + 1 was below the poverty line was represented by. The poverty line defined the minimum standard of living required to escape poverty. According to the criteria established by the National Bureau of Statistics of China, the official poverty line was calculated based on two components: basic food needs and non-food needs. During the period of targeted poverty alleviation, the basic food requirement referred to the minimum dietary expenditure necessary to maintain human health, including approximately 500 g of rice or flour, 500 g of vegetables, and 50 g of meat or one egg per person per day, which provided about 2,100 kilocalories of energy and 60 g of protein daily. The non-food component covered essential living expenditures such as clothing, housing, education, healthcare, transportation, and communication. To ensure the poverty line reflected a consistent standard of living over time, annual adjustments were made to account for changes in price levels. In this study, the official poverty line set by the government was adopted, with thresholds of 2,536 yuan in 2011, 2,736 yuan in 2013, 2,855 yuan in 2015, and 2,995 yuan in 2018 [19].The future welfare level of the household was the focus of the study.

In this paper, household income was chosen as a proxy, and it was assumed that income remained stable over a certain period. The current characteristics of the household were used to predict future income levels, and an income prediction equation was established. Its basic characteristics mainly consisted of four aspects: family structure, socioeconomic status, health burden, and social support [20].

Building on the traditional health poverty vulnerability framework, this study incorporates health-related variables into the model to better capture the health dimension of poverty vulnerability. Specifically, the health burden variables include the share of medical expenditure in total household income, incidence of chronic diseases, presence of disabled members, and frequency of hospitalizations. By embedding these indicators, the model measures health poverty vulnerability, that is, the probability that a household will fall into or return to poverty as a result of adverse health shocks or insufficient health protection.

2

The logarithm of the income of household i at time t + 1 was represented by, and some basic characteristic variables of household i at time t + 1 were represented by. The fluctuation term of household income level was further estimated. Weighted regression was used to perform weighted regression on the logarithm of income and the sum of squared residuals, and the generalized feasible least squares estimates of income expectation and variance were obtained. It was assumed that income followed a normal distribution, and a formula for poverty vulnerability was derived:

3

The cumulative probability density function of the standard normal distribution was represented by .The poverty line was represented by.The expected value of the logarithm of household income was represented by .The expected value of the logarithm of per capita income was represented by.

Measurement of static indicators

Catastrophic health expenditure (CHE) incidence was assessed in accordance with World Health Organization guidelines using the following methodology [11]. Household capability to pay (CTP) was operationalized as total consumption expenditure minus essential food expenditure. A catastrophic health event was identified when a household’s out-of-pocket (OOP) medical payments exceeded 40% of its CTP threshold [21].

The CHARLS questionnaire includes detailed questions related to OOP expenditures. Outpatient OOP expenses were obtained from Section C, which asks: “How much was the total cost of your outpatient visits in the past month (including both out-of-pocket and reimbursed parts)?” and “How much did you pay yourself?” The reported monthly outpatient OOP expenditure was multiplied by 12 to obtain the annual value. Similarly, inpatient OOP expenditure was obtained from the questions: “How much was the total cost of your inpatient stay in the past year (including both out-of-pocket and reimbursed parts)? Please include only payments made to the hospital, excluding caregiver wages, transportation, and accommodation costs, but including ward fees,” and “How much did you pay yourself?” The total annual household OOP medical expenditure was then calculated as the sum of all members’ outpatient and inpatient OOP costs.

This dichotomous outcome variable was coded as 1 for households experiencing CHE and 0 for non-affected households. The computation procedure comprised three sequential steps: First, CTP was calculated by deducting basic nutritional costs from total household expenditure. Subsequently, the CHE threshold ratio was determined by dividing OOP health payments by the derived CTP value. Finally, households were classified as experiencing CHE when this ratio surpassed 0.4, indicating disproportionate healthcare spending relative to available economic resources. The calculation process is as follows:

4

Other independent variables

The independent variables examined in this study are organized into four conceptual domains, encompassing eight key predictors: demographic structure, socioeconomic status, health burden, and social security. Within the demographic structure domain, family size was categorized into three groups (1–3, 4–5, and ≥ 6 members). The socioeconomic status domain included the gender and marital status of the household head, annual household income (log-transformed to mitigate the influence of extreme values), and the number of household members aged 65 years or older, which reflects the old-age dependency burden. The health burden domain consisted of the number of family members with chronic diseases (0, 1, ≥ 2 members), the number of disabled family members (0, 1, ≥ 2 members), and medical expenditure. In accordance with previous CHARLS-based studies [16, 22], medical expenditure was defined as the proportion of total household out-of-pocket (OOP) medical expenses to total household expenditure. Finally, the social security domain was measured by whether any household members received financial assistance (coded as 1 = yes, 0 = no). Detailed operational definitions and coding schemes for all variables are systematically presented in Table 1, ensuring methodological transparency and reproducibility.

Table 1.

Coding of health poverty vulnerability and other independent variables

Variable	Definition
Demographic structure
Family size	Number of persons in the household	1 = households with 1–2 members 2 = households with 3–5 members 3 = households with 6 or more members
Socioeconomic status
Gender	Gender of head of household	1 = male,0 = female
Marital status	Marital status of head of household	1 = married,0 = unmarried
Household income	Income of family members	Continuous variables
The burden of old age	Whether there are people over 65 years of age	1 = yes,0 = no
Health burden
Chronic diseases	Number of members with chronic diseases	0 = 0,1 = 1,2 = ≥ 2
Disability status	Number of disabled members	0 = 0,1 = 1,2 = ≥ 2
Medical expenditure	Out-of-pocket medical expenses as a proportion of household expenditure	Continuous variables
Hospitalizations per year	Number of hospital visits in a year	0 = 0,1 = 1,2 = ≥ 2
Social security
Economic support	Whether financial support received	1 = yes,0 = no

 Statistical method

The Difference-in-Differences (DID) method, which was recognized as a quasi-experimental analytical framework, was designed to estimate causal effects of interventions by comparing outcome trajectories between treatment and control groups across pre- and post-intervention periods [23]. As a cornerstone of policy evaluation in public health research, this approach was employed by researchers to quantify the impact of interventions such as health policy reforms, program implementations, or regulatory changes [24].

In this study, a DID framework was utilized to evaluate the Health Poverty Alleviation Policy’s impact on household health poverty vulnerability and catastrophic health expenditures. Aligned with the policy’s rollout timeline, 2016 was designated as the intervention year, with 2011, 2013, and 2015 established as pre-intervention observation points and 2018 as the post-intervention assessment period [25].In the research design, households that answered “yes” to the survey question “Is your household registered as a poor household?” were classified as the treatment group, while those that answered “no” were classified as the control group.

A Quantile Difference-in-Differences (QDID) framework was further incorporated into the empirical strategy to investigate distributional heterogeneity in policy effects [26]. In contrast to the conventional DID approach, which estimates average treatment effects, the QDID method was designed to allow treatment impacts to vary across the conditional distribution of the outcome variable. This feature was particularly relevant in the context of poverty vulnerability, where households located at different risk quantiles may have responded differently to policy interventions.

Specifically, the QDID model enabled treatment effects to be estimated at multiple vulnerability quantiles, thereby capturing differential policy impacts among low-, moderate-, and high-risk households. By modeling these quantile-specific effects, the framework provided a more comprehensive assessment of whether and to what extent the intervention disproportionately benefited households at the upper end of the vulnerability distribution.

Considering that the policy effects experienced by each household might have varied depending on household-level characteristics, such as awareness of medical assistance programs, attitudes toward health, and access to healthcare services—household-specific unobserved heterogeneity was controlled for by incorporating household fixed effects in the regression models.

For model identification, a Linear Probability Model (LPM) was employed to estimate the likelihood of households incurring catastrophic health expenditures. Compared with nonlinear models such as Logit or Probit, the LPM provided a more intuitive interpretation of regression coefficients, as they directly represented the marginal effects of explanatory variables on the probability of occurrence. Moreover, the LPM offered computational efficiency and simplicity of interpretation when handling large-scale panel data. Therefore, the adoption of this model enhanced the interpretability of the results and aligned with empirical practices commonly adopted in public health policy evaluation research [27].

The following DID and QDID models were constructed:

5

6

In this study, “"was used to represent a household, and “” was used to represent the year. " was used to denote the outcome variable, which included household health poverty vulnerability and catastrophic health expenditure. “"was used to represent the conditional quantile of the outcome variable""at the “τ “quantile.was a dummy variable, which was set to “1” if a household was registered as a poor household and was considered to be affected by the policy, and “0” otherwise. “was a time dummy variable, which was set to “0” before the policy implementation and “1” after the policy implementation. “” was used to represent individual fixed effects, “"was used to represent time fixed effects, and"” was the error term.

Sensitivity method

To assess the robustness of the empirical findings to alternative thresholds for catastrophic health expenditures (CHE), a sensitivity analysis was conducted using a lower CHE threshold. Although the main analysis adopted the 40% threshold recommended by the World Health Organization (WHO), existing literature indicates that thresholds ranging from 10% to 60% have been widely applied, with 10% and 40% being the most commonly used [28]. Following established research practice [29], a 10% threshold was employed in the sensitivity test to re-estimate the effect of the Health Poverty Alleviation Policy (HPAP) on the incidence of CHE. This analysis was designed to examine whether the policy effect remained consistent under a more stringent threshold. All statistical analyses in this paper were conducted using STATA16 software, with statistical significance set at p < 0.05.

Results

Descriptive statistics

Across all survey waves, the average health poverty vulnerability score was 0.48 ± 0.02. As shown in Table 2, in Wave 1, the control group had an average score of 0.48, indicating an average 48% probability of falling into poverty due to health-related factors, while the treatment group exhibited a slightly higher probability (49%). A higher health poverty vulnerability score indicates a greater likelihood of a household falling into poverty as a result of poor health conditions.

Table 2.

Descriptive characteristics of baseline variables in the control and treatment groups

	Wave1		Wave2		Wave3		Wave4
	Control	Treat	Control	Treat	Control	Treat	Control	Treat
Dependent variables
Vulnerability	0.48	0.49	0.49	0.50	0.47	0.49	0.46	0.48
CHE	191(85.65)	32(14.35)	304(85.63)	51(14.37)	399(86.93)	60(13.07)	442(87.18)	65(12.82)
Independent variables
Age	58.57	59.72	60.57	61.72	62.57	63.72	65.57	66.72
Married	5462(84.15)	463(76.78)	5314(81.90)	452(74.96)	5139(79.20)	429(71.14)	4793(73.88)	385(63.85)
Others	1026(15.85)	140(23.22)	1174(18.10)	151(25.04)	1349(20.80)	174(28.86)	1695(26.12)	218(36.15)
Older persons over 65years of age
Yes	3576(55.10)	370(61.36)	3659(56.38)	410(68.00)	4213(64.90)	425(70.48)	4511(69.52)	442(73.30)
No	2912(44.90)	233(38.64)	2829(43.62)	193(32.00)	2275(35.06)	178(29.52)	1977(30.48)	161(26.70)
Family size
≤ 2	3496(53.84)	307(50.91)	3478(53.61)	312(51.74)	4758(73.35)	457(75.79)	5090(78.46)	491(81.43)
3–5	2001(30.84)	200(33.17)	2033(31.34)	200(33.17)	1400(21.58)	121(20.07)	965(14.87)	87(14.43)
≥ 5	991(15.32)	96(15.92)	977(15.05)	91(15.09)	330(5.07)	25(4.15)	433(6.67)	25(4.15)
Household income per ca. pita	3587.57	2368.07	4513.27	3254.21	6171.75	4521.36	9119.31	6739.06
Number of chronic patients
0	1353(20.85)	103(17.08)	1107(17.06)	83(13.76)	643(9.91)	54(8.96)	485(7.48)	35(5.80)
1	3300(50.86)	315(52.24)	3311(51.04)	312(51.74)	3219(49.61)	298(49.42)	3154(48.61)	294(48.76)
≥ 2	1835(28.29)	185(30.68)	2070(31.90)	208(34.49)	2626(40.48)	251(41.63)	2839(43.76)	274(45.44)
Number of disability patients
0	4382(67.54)	351(58.21)	3711(57.20)	263(43.62)	2976(45.87)	193(32.01)	2518(38.81)	141(23.38)
1	1791(27.60)	204(33.83)	2278(35.11)	274(45.44)	2745(42.31)	307(50.91)	2932(45.19)	324(53.73)
≥ 2	315(4.86)	48(7.96)	499(7.69)	66(10.95)	767(11.82)	103(17.08)	1038(16.00)	138(22.89)
Number of hospitalizations per year
0	5807(89.50)	541(89.72)	5459(84.14)	499(82.75)	5343(82.35)	478(79.27)	5076(78.24)	402(66.67)
1	628(9.68)	54(8.96)	908(14.00)	89(14.76)	1014(15.63)	104(17.25)	1195(18.42)	155(25.71)
≥ 2	53(0.82)	8(1.33)	121(1.86)	15(2.49)	131(2.02)	21(3.48)	217(3.34)	46(7.63)
Economic support
Yes	471(7.26)	52(8.62)	1630(25.12)	171(28.36)	2137(32.94)	385(63.85)	1697(26.16)	155(25.71)
No	6071(93.74)	551(91.38)	4858(74.88)	432(71.64)	4531(69.84)	218(36.15)	4791(73.84)	448(74.29)

Values represent means for continuous variables and numbers (percentages) for categorical variables. The numbers in parentheses indicate the proportion (%) of individuals or households within each category

Figure 5 further illustrates the distributional characteristics and intergroup differences in health poverty vulnerability across the four survey waves. The estimation of health poverty vulnerability was influenced by multiple factors, including household structure, socioeconomic status, health burden, and social support. Since the control group accounted for approximately 92% of the total sample and exhibited considerable internal heterogeneity, its estimates showed greater dispersion and a wider distribution (Supplement Fig. 1). In each survey wave, the treatment group consistently demonstrated significantly higher vulnerability scores than the control group, with the differences being statistically significant (p < 0.001). Although both groups experienced a gradual decline in vulnerability after Wave 2, households in the treatment group remained relatively more vulnerable. In Wave 4, following the full implementation of the policy, the vulnerability level of the treatment group slightly declined but returned to a level similar to that observed in Wave 1, indicating that the policy’s effectiveness in reducing household vulnerability remained limited.

Fig. 5.

Overall distribution of health poverty vulnerability

Meanwhile, the incidence of catastrophic health expenditure (CHE) in the treatment group was found to have remained almost unchanged between the first wave (14.35%) and the second wave (14.37%), after which it was observed to have declined gradually to 13.07% in the third wave and 12.82% in the fourth wave. This pattern indicates that the Health Poverty Alleviation Policy (HPAP) may have contributed to a modest but consistent reduction in health-related financial risks over time.

In terms of demographic characteristics, it was found that the heads of households in the treatment group were significantly older, and the proportion of unmarried members was higher, reflecting a relatively weak social support system. In terms of health burden, the prevalence of chronic diseases (≥ 2 conditions) in the treatment group was found to have increased annually, and the annual number of hospitalisations (≥ 2 times) was found to have tripled compared to the baseline level. Compared to the control group, the treatment group was found to face a heavier health burden. Economically, despite policy interventions, the per capita income of households in the treatment group was found to remain low, with a high degree of reliance on external support, highlighting that their structural vulnerabilities had persisted even after the intervention.

Impact of HPAP on household catastrophic health expenditures

Table 3 presents the results of the DID regression analysis for the impact of the HPAP policy on the incidence of catastrophic health expenditures (CHE) in households. The coefficient of the interaction term (Post*Treatment) is found to be statistically significant (p < 0.05), indicating that the incidence of CHE in households in the treatment group was significantly reduced by approximately 2.08% compared to the control group following the implementation of the policy. It is suggested that the risk of CHE may have been effectively mitigated through targeted support measures, such as subsidizing health insurance costs for vulnerable groups, which likely alleviated the trend of worsening health outcomes and reduced the risk of households falling back into poverty due to major illnesses [30].

Table 3.

Catastrophic health expenditures DID regression results

Variable	Coef.	P	[95% Conf.	Interval]
Post*Treatment	−0.0208	0.017	−0.038	−0.0037
Marital status
1	0.0047	0.556	−0.011	0.0205
Family size
2	−0.0316	< 0.001	−0.0388	−0.0243
3	−0.0525	< 0.001	−0.063	−0.0421
Household income	−0.011	< 0.001	−0.0141	−0.0078
Disability status
1	−0.0028	0.433	−0.01	0.0043
2	0.0131	0.027	0.0015	0.0247
Chronic diseases
1	−0.0012	0.838	−0.0131	0.0106
2	−0.0011	0.879	−0.0157	0.0134
Hospitalizations per year
1	0.2601	< 0.001	0.2533	0.2669
2	0.4179	< 0.001	0.4014	0.4343
Economic support	0.0116	< 0.001	0.0055	0.0177
Household fixed effect	Yes
Year fixed effect	Yes

Impact of HPAP on poverty vulnerability

To further compare policy effects between low-risk and high-risk populations, this study focused on examining the heterogeneous impacts across different levels of health poverty vulnerability. As shown in Supplement Fig. 2, the overall distribution exhibited a pronounced right-skewed pattern, with the majority of households concentrated in the lower vulnerability range and relatively few in the high-vulnerability range. Accordingly, the 10th and 90th percentiles were selected to represent, respectively, the lowest-risk and highest-risk groups. Based on these percentile thresholds, the sample was divided into four percentile-based groups (Q1–Q4): Q1 represented the bottom 10% (lowest risk), Q2 the next 25% (relatively low risk), Q3 the subsequent 75% (relatively high risk), and Q4 the top 90% (highest risk).

To gain a clearer understanding of the dynamics of HPAP policy effects across these groups, QDID regression analyses were conducted. The results (Table 4; Fig. 6) showed a statistically significant negative effect of HPAP policy for Q4 households (− 0.0073*, p < 0.05), suggesting that the intervention yielded a pronounced reduction in health poverty vulnerability among the highest-risk group. In contrast, the insignificant coefficients observed for Q1–Q3 indicated limited policy effects among households with lower vulnerability.

Table 4.

Health poverty vulnerability QDID regression results

Variable	Q1	Q2	Q3	Q4
Post*Treatment	0.0013	0.0035	−0.0029	−0.0073*
Gender	0.0002	0.0007*	0.0001	−0.0008
Marital status	0.0774***	0.0576***	0.0128***	0.0133***
Chronic diseases	0.0203***	0.0170***	0.0149***	0.0152***
Disability status	0.0140***	0.0144***	0.0161***	0.0155***
Hospitalizations per year	−0.0012	0.0035***	0.0106***	0.0199***
Household fixed effect	Yes
Year fixed effect	Yes

* denotes p< 0. 05, ** denotes p< 0. 01, *** denotes p< 0. 001

Fig. 6.

Health Poverty Vulnerability QDID regression results box plot

Furthermore, marital status remained positively and significantly associated with vulnerability across all groups, implying that married households systematically faced greater vulnerability risks. Healthcare utilization also demonstrated a clear gradient: hospitalization frequency was insignificant in Q1 but strongly positive in Q4, highlighting that repeated inpatient care imposed disproportionate economic pressure on the most vulnerable group.

Common trend assumption

The application of DID relied on a basic assumption, the parallel trend assumption, which implies that in the absence of policy changes, the trends in outcomes for the treatment and control groups would have been parallel. This study used four survey waves (2011, 2013, 2015, and 2018), of which 2011, 2013, and 2015 were collected before the policy implementation and constituted the pre-treatment period, while 2018 was collected after the policy implementation and constituted the post-treatment period. In the event-study specification, 2011 was omitted and treated as the baseline period to avoid perfect multicollinearity, so that the estimated coefficients for 2013 and 2015 represented the dynamic effects relative to the pre-policy baseline.

After the policy implementation, any divergent trends observed between the treatment and control groups were attributed to the impact of the policy. As shown in Fig. 7, before the policy was implemented in 2016, the differences in the incidence of catastrophic health expenditure between the treatment and control groups were not statistically significant. In contrast, after the policy, the differences became statistically significant, confirming the validity of the parallel trend assumption.

Fig. 7.

Results of parallel trend tests for catastrophic health expenditures

In addition, to further verify this assumption, a joint F-test was conducted on the interaction terms prior to the policy implementation (Supplement Table 1). The null hypothesis assumed that all pre-policy coefficients were equal to zero (H₀: βinter1 = βinter2 = βinter3 = 0). The test results showed an F-statistic of 2.20 with a corresponding p-value of 0.0855. At the 5% significance level, the null hypothesis could not be rejected, indicating that the trend differences between the treatment and control groups before the policy implementation were not statistically significant. This finding provides additional support for the validity of the parallel trend assumption.

Sensitivity test

To verify the robustness of the estimated policy effects, a sensitivity analysis was conducted using an alternative CHE threshold of 10%. The regression results based on this stricter threshold are presented in Table 5. Consistent with the main DID estimates obtained under the 40% benchmark, the direction and significance of the HPAP policy effects remained largely unchanged. This stability indicates that the reduction in the incidence of CHE attributable to the HPAP policy is not sensitive to the choice of threshold and that the findings are robust across different CHE definitions.

Table 5.

DID regression results on the occurrence of CHE at 10% thresholds

Variable	Coef.	P	[95%	Interval]
Post * Treat	−0.022	0.043	−0.044	−0.001
Marry
1	−0.002	0.855	−0.022	0.018
Family size
2	−0.040	0.000	−0.049	−0.031
3	−0.079	0.000	−0.092	−0.066
Household income	−0.016	0.000	−0.020	−0.012
Disability status
1	−0.002	0.640	−0.011	0.007
2	0.017	0.024	0.002	0.032
Chronic diseases
1	0.008	0.317	−0.007	0.023
2	0.031	0.001	0.013	0.050
Hospitalizations per year
1	0.509	0.000	0.501	0.518
2	0.625	0.000	0.604	0.646
Economic support	0.009	0.021	0.001	0.017
Household fixed effect Year fixed effect	Yes Yes

Discussion

This study draws on the China Health and Retirement Longitudinal Study (CHARLS) and employs a quasi-experimental difference-in-differences (DID) design to examine the heterogeneous effects of China’s Health Poverty Alleviation Policy (HPAP) on household health poverty vulnerability and catastrophic health expenditure (CHE). The findings indicate that while HPAP significantly reduces the incidence of CHE, its impact on long-term vulnerability to health-related poverty remains limited. This divergence suggests that the policy primarily focuses on mitigating immediate financial burdens rather than strengthening households’ long-term health resilience.

Our findings echo evidence from international health financing reforms. Vietnam has expanded social health insurance coverage, increased public health expenditure, piloted alternative provider payment methods, and implemented targeted protections for vulnerable groups [31]. India’s PM-JAY and Ecuador’s health protection reforms have similarly reduced medical out-of-pocket payments among low-income populations by increasing benefit packages, broadening service coverage, and optimizing payment mechanisms, thereby alleviating socioeconomic inequities in healthcare utilization [32, 33]. Consistent with this global evidence, the present study shows that China’s HPAP achieves measurable short-term reductions in CHE through enhanced insurance compensation and targeted subsidies.

The short-term effectiveness of HPAP can be attributed to several mechanisms. First, the coordinated expansion of basic medical insurance, critical illness insurance, and medical assistance achieved full coverage among impoverished households, providing essential financial protection. Second, reforms such as the “treat first, pay later” policy reduced financial barriers to hospitalization and minimized delays in seeking care due to upfront payment requirements [34]. Third, the policy strengthened critical illness insurance by lowering deductibles by 50%, increasing reimbursement rates by 5% points, and gradually raising or removing ceiling limits. In parallel, medical assistance programs were enhanced to ensure that approximately 70% of households’ out-of-pocket medical expenditures were reimbursed within the annual assistance quota, with additional preferential support for groups in extreme hardship [35]. Collectively, these demand-side interventions alleviated the direct economic burden of healthcare for poor households.

Despite these improvements, the observed decline in CHE was modest. International studies indicate that substantial reductions in CHE typically require long-term sustained implementation of health financing reforms. Given structural constraints in healthcare resource allocation, limited follow-up duration, and a relatively short policy exposure period, the current findings likely reflect only the initial stage of HPAP’s policy effects.

In contrast, the policy yielded minimal improvements in long-term health poverty vulnerability among low- and medium-risk households, with only marginal reductions observed among high-risk groups. This limited impact stems from HPAP’s emphasis on cost containment rather than addressing the underlying determinants of health vulnerability. Critical areas—such as chronic disease management, rehabilitation services, preventive care, and health literacy interventions—remain underdeveloped. Additionally, imprecise risk stratification and fragmented beneficiary identification mechanisms result in the exclusion of households located at the margins of eligibility. The limited use of big-data analytics further constrains the system’s ability to conduct early prediction and dynamic monitoring of health-related poverty risks. International evidence similarly suggests that cost-control policies alone, when not complemented by investments in preventive care and long-term capacity building, rarely succeed in breaking the cycle of “poverty caused by illness and illness caused by poverty” [36].

These gaps highlight a fundamental challenge: although HPAP has achieved short-term financial risk protection, it has not substantively enhanced the healthcare system’s supply capacity. Cost-control mechanisms alone cannot eliminate structural inequities in access to essential healthcare services, nor can they prevent vulnerable households from repeatedly falling into poverty. A strategic shift from single-dimensional financial protection to comprehensive capacity strengthening is therefore essential.

Future policy iterations should prioritize systemic reforms. First, risk monitoring should be enhanced through digital health platforms and community-based screening systems to improve precision and responsiveness. Second, public investment should be directed toward underserved regions, particularly in chronic disease management and rehabilitation services. Third, preventive and early intervention services should be incorporated into insurance benefit packages to reduce long-term health expenditures. Fourth, stronger intersectoral collaboration is needed to integrate health policies with social protection, education, and nutrition programs, thereby addressing non-medical determinants of health poverty. Only by simultaneously improving financial protection, service accessibility, and risk identification capacity can health poverty vulnerability be sustainably reduced.

Several limitations should be noted. First, due to data constraints, only one post-policy observation (2018) was available. Although the DID framework supports causal inference, the findings should be interpreted as reflecting short-term policy effects. Similar to evaluations of large-scale international health reforms, comprehensive assessment of long-term impacts requires multi-wave follow-up data. Second, household income used to calculate vulnerability was partially based on self-reported information, which may be subject to recall bias and could lead to underestimation of policy effects. Future research with verified income data and extended observation windows will further strengthen causal identification.

Authors’ contributions

All authors contributed to the study conception and design. XX contributed to the writing of the manuscript, data processing, and submission of the paper. RC contributed to the proposal of the paper. HC contributed to the overall design of the paper. YM contributed to the editing of the paper. CY and HS reviewed the manuscript. All authors read and approved the final manuscript.

Funding

A Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions.

Data availability

The data that support the findings of this study are available from “https://charls.pku.edu.cn”.

Declarations

Ethical statement and consent to participate

All cycles of the CHARLS survey were approved by the Biomedical Ethics Committee of Peking University. Informed consent was obtained from all subjects and/or their legal guardian(s).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.