This longitudinal quantitative study employed a balanced panel design to investigate the spatio-temporal patterns and determinants of persistent CHE in China from 2012 to 2020.

This study utilized data from five waves of the CFPS, conducted in 2012, 2014, 2016, 2018, and 2020. Launched in 2010 by the Institute of Social Science Survey at Peking University, the CFPS is a biennial nationwide household survey frequently used for estimating CHE and conducting related research (25, 27, 28). The baseline sample covers 25 provinces, autonomous regions, and municipalities, representing 94.5% of Mainland China’s total population. The sample is derived through a multi-stage probability sampling method that employs implicit stratification. CFPS subsamples were hierarchically drawn, starting at the county level, then the village level, and finally the household level. Detailed sampling protocols are available at www.isss.edu.cn/cfps/. Participants underwent biennial follow-ups, resulting in a total of six national survey periods to date (26). To maintain ethical standards, the Peking University Biomedical Ethics Review Committee granted approval (Approval number: IRB00001052-14010). Respondents were provided with a detailed statement outlining the study’s purpose, and their consent was obtained before data collection commenced.

The baseline cohort of the CFPS in 2010 included 14,960 households and 42,590 individuals. No new households were added to the panel subsequently. The success-tracking rates for the follow-up periods in 2012, 2014, 2016, 2018, and 2020 were 85, 89, 89, 86, and 89%, respectively. This study used the individual characteristics of household heads (designated as financial respondents) as representative of the family’s individual characteristics within the sample. To address potential selection bias from panel attrition, we utilized a balanced panel of households that participated in all five waves. After excluding observations with missing data, our analytic sample consisted of a balanced panel involving 2,544 households, comprising a total of 12,720 observations.

The selection of covariates was guided by the Andersen health service utilization model, providing a comprehensive framework for analysis (34). According to this model, health service utilization is influenced by predisposing, enabling, and need variables.

Predisposing variables primarily include household size, household residence (urban or rural), gender of the household head (male or female), age of the household head, marital status of the household head (living alone, married or partnered), job classification (unemployed, agricultural worker, and non-agricultural worker), years of education of the household head, and the number of family dependents (members aged >over 60 years or under<6 years). Enabling variables were measured using three indicators: type of health insurance held by the household head, geographic location (northeast, west, middle, central eastern areas), and socioeconomic group. The sample was categorized into five groups based on equivalent household capacity to pay, ranging from the most deprived 20% to the most affluent 20%. Health insurance options included the urban employee basic medical insurance scheme (UEBMI), the urban resident basic medical insurance scheme (URBMI), or new cooperative medical scheme (NCMS), and being uninsured (9). Regarding need variables, the study considered members with chronic diseases, self-rated health (ranging from worst to best), and hospitalization (yes or no).

Descriptive statistics were used to succinctly outline the characteristics of the study sample. Categorical data were presented through frequency and proportion, whereas continuous data were summarized using mean and standard deviation. This study explored the spatial and temporal distribution characteristics of CHE incidence, depth, and duration among sample families from 2012 to 2020. Additionally, this study employed global and local Moran’s I indices to examine the spatial patterns of the incidence, depth, and duration of CHE across 31 provinces in China. The Global Moran’s I index estimated the spatial agglomeration and divergence, with a value ranging from −1 to 1 (35). Global Moran’s I > 0 signifies positive spatial agglomeration, I < 0 indicates negative spatial correlation, and I = 0 denotes a random spatial distribution. The local Indicator of Spatial Association (LISA) is designed to further examine the spatial association between a local unit and its neighboring units. Local Moran’s I > 0 means positive correlation between an observation and its neighbors (high values surrounded by high values, or low by low). I < 0 indicates negative spatial dependence (higher-lower or lower-higher clusters). The statistical significance of the global and local Moran’s I indices is assessed using the z-value (36). The Tobit model is a regression model designed for situations where the dependent variable is censored or truncated.

Given the threshold limitations within the study context, a significant number of instances in the depth of CHE were observed with values of zero. Therefore, Tobit regression analysis was employed as a methodological approach to comprehensively examine the determinants of CHE depth. Furthermore, this study analyzed five-wave longitudinal data collected from the same households, which resulted in within-subject correlation due to the repeated observations. While conventional models for cross-sectional data (e.g., logistic regression) cannot account for this autocorrelation, the Generalized Estimating Equations (GEE) model addresses it by incorporating an appropriate working correlation matrix (37). Consequently, a GEE model was constructed under a Gaussian distribution to discern the factors influencing the duration of CHE. Statistical significance was set at p < 0.05, and 95% confidence intervals were reported. All statistical analyses were executed using Stata 16.0, and spatial analyses were conducted using ArcGIS Pro software.

Table 1 outlines the comprehensive characteristics of the surveyed households from the years 2012, 2014, 2016, 2018, and 2020, with a particular focus on the 2,544 households that consistently participated in all surveys. The average household size decreased from a mean of 4.007 members in 2012 to 3.816 members in 2020. The majority of household heads resided in rural areas (with the percentage declining from 75.28 to 71.05%), and most lived with a spouse or partner, decreasing from 92.02 to 86.79%. The proportion of householders working in agriculture saw a slight decrease from 41.97% in 2012 to 39.81% in 2020, while the number of family dependents increased from 0.85 to 1.01 over the same period. The number of households with UEBMI rose from 409 in 2012 to 478 in 2020, whereas those with URBMI or NCMS decreased from 1956 to 1879. The prevalence of chronic diseases among household members increased from 0.372 to 0.401, and the percentage of householders who perceived themselves to be in good health decreased from 78.774% in 2012 to 75.977% in 2020.

Figure 1 illustrates the levels and variations in the depth and incidence of persistent CHE in China from 2012 to 2020. The incidence of CHE has demonstrated a fluctuating pattern, characterized by a sequence of “decrease, increase, decrease” over the years, with recorded rates of 16.27, 14.50, 16.98, 16.23, and 13.88% from 2012 to 2020, respectively. Although the fluctuations in CHE among urban households have not been statistically significant, there has been a marginal increase from 10.17% in 2012 to 10.77% in 2020. The overall trend for rural households closely mirrors the national average, showing a decline from 18.27% in 2012 to 15.19% in 2020. The persistent depth of CHE has consistently followed an ascending trajectory, increasing from 0.025 in 2012 to 0.069 in 2020. At the urban level, the persistent depth of CHE has experienced an elevation from 0.015 in 2012 to 0.049. Similarly, at the rural level, the persistent depth of CHE has decreased from 0.029 in 2012 to 0.077 in 2020.

Figure 2 depicts the distribution of households experiencing different durations of CHE in 2020 as a percentage of the total number of households. The prevalence of households affected by persistent CHE has increased notably, with an average annual growth rate of 2.39% over a five-year period. In 2020, the proportion of households affected by persistent CHE reached 14.32% in 2020. When comparing urban and rural households, the percentages were 10.36 and 16.09%, respectively, in 2020. This disparity indicates that rural families are more vulnerable to the prolonged effects of CHE than urban counterparts.

To visually represent the spatial patterns of CHE, this study created spatial distribution maps of CHE measurement indices (Figure 3). At the national level, CHE indices in 2012 showed a general trend of increasing from east to west. By 2020, however, this trend had changed to an increase from south to north. In urban areas, spatial disparities have diminished, and by 2020, a pattern of increasing indices from east to west became apparent. In rural areas, initially, the higher CHE indices were concentrated in the northeast, but by 2020, there was a shift towards a continuous increase in indices from south to north.

The global spatial autocorrelation analysis in Table 2 revealed that the spatial clustering of CHE duration was statistically significant only in 2012 (Moran’s I = 0.294, p < 0.05). From 2012 to 2016, Moran’s I of persistent CHE showed a significant decrease, changing from positive to negative values. Subsequently, from 2016 to 2020, Moran’s I rebounded from negative to weakly positive values but remained substantially lower than the initial level observed in 2012.

To more intuitively illustrate the local spatial correlation of persistent CHE, this study plotted LISA maps for the depth and duration of CHE across the years 2012, 2014, 2016, 2018, and 2020 (Figure 4). As for depth of CHE, in 2012, a high-high cluster was observed in Shandong province. By 2014, however, the spatial pattern had diversified: Guangxi, Jiangxi, and Fujian formed low-low clusters, Chongqing emerged as a high-low outlier, and Shaanxi was identified as a low-high outlier. Inner Mongolia was consistently recognized as a low-high clustering region in 2016 and 2018. In 2020, low-low clusters were identified in Guangxi, Yunnan, and Hunan, and high-low outliers were observed in Heilongjiang and Shaanxi. Concerning the duration of CHE, Shanxi was identified as a low-high cluster in 2012, while a significant high-high cluster was observed in Gansu and Ningxia. In 2016 and 2018, the spatial pattern shifted notably, with Inner Mongolia being classified as an LH outlier. By 2020, Gansu and Shaanxi formed a high-high cluster; Guangxi was identified as a low-low cluster; and Fujian, Yunnan, and Hunan were characterized as high-low clustering pattern.

Table 3 presents factors influencing persistent CHE using Tobit and GEE models. Tobit regression indicated that household size was associated with a reduction in the depth of CHE (OR = 0.973), whereas protective factors included higher education levels of household heads (OR = 0.996), marital status (OR = 0.955), employment in agriculture or non-agriculture sectors (OR = 0.969 and 0.944), residence in the eastern region (OR = 0.977), and possession of UEBMI insurance (OR = 0.969). In contrast, factors contributing to an increased depth of CHE were the presence of family dependents (OR = 1.047) and chronic disease (OR = 1.013).

The GEE model was constructed with the duration of CHE as the dependent variable. It revealed that older household heads (OR = 1.003), a greater number of family dependents (OR = 1.057), chronic disease (OR = 1.015), poorer health status (OR = 1.010), and recent hospitalization (OR = 1.082) were correlated with a longer duration of CHE. Larger household size (OR = 0.960) and higher education levels of household heads (OR = 0.995) acted as protective factors. Urban residence and marriage were associated with reduced sensitivity to the duration of CHE (OR = 0.939 and 0.886), while households with heads employed in agricultural (OR = 0.958) and non-agricultural (OR = 0.923) sectors experienced shorter durations compared to those with no employment. Furthermore, families in Quintiles III (OR = 0.962) and IV (OR = 0.970) experienced shorter durations compared to those in the most deprived families.

Sensitivity analyses for the 10 and 25% thresholds indicated an increase in the severity and persistence of CHE from 2012 to 2020 (p < 0.05), mirroring a similar trend observed when the threshold was set at 40%. The associations between the duration and depth of CHE and the independent variables remained unchanged. However, the statistical significance of certain factors varied (see Supplementary Table 1 for further details).

This study indicated that the incidence of CHE decreased from 16.27% in 2012 to 13.88% in 2020. This observed decline may be attributed to the synergistic impact of economic growth and various policy interventions, particularly the expansion of universal health insurance coverage and increased subsidies for vulnerable groups (28). In line with prior research utilizing comparable CHE measurements, our findings broadly align with these studies, indicating an overall incidence of CHE around 14% and demonstrating a declining trend (24, 25). However, the number of households impacted by persistent CHE saw a significant rise, with an average annual increase of 2.39% over a five-year period. Additionally, a consistent upward trend in the persistent depth of CHE, increasing from 0.025 in 2012 to 0.069 in 2020, underscoring a growing financial crisis for households. Families facing conditions like cancer or uremia may not be excluded from the medical insurance system but can still enter a cycle of “treatment-debt-retreatment,” leading to chronic financial exhaustion. The cause may lie in the design of medical insurance schemes. Despite expanded coverage, medical insurance schemes with high OOP ratios, low reimbursement caps, and restrictive drug formularies offer inadequate protection against the long-term financial impact of major chronic diseases (38).

Given the pronounced disparities in economic levels, social security systems, and health resources between urban and rural areas in China, the incidence, persistent depth, and duration of CHE in rural households exceed those in urban households (24, 27, 28). This disparity stemmed from deeper, systemic vulnerabilities that can be analyzed across three key dimensions. First, the concentration of high-quality medical resources in urban areas forces rural residents to bear higher indirect costs (e.g., transportation, accommodation) for non-local treatment, exacerbating their direct financial burden. Second, despite the integration of urban and rural resident health insurance, substantive gaps remain compared to UEBMI in terms of reimbursement rates, coverage ceilings, and drug formularies, resulting in shallower financial protection for rural residents. Third, rural households typically have more singular income sources, weaker savings, and a less robust social safety net, inherently lowering their financial resilience (24, 39).

Our regional investigation has revealed significant inter-regional disparities in the duration of CHE. At the national level, CHE indices in 2020 indicated a shift from the southern to the northern regions. This uneven distribution of CHE across China was affected by differences in financial contributions, treatment packages, and reimbursement rates among various health insurance policies (40, 41). Further global autocorrelation analysis revealed that the spatial pattern of persistent CHE in China underwent a general shift from “clustered” to “relatively dispersed” between 2012 and 2020, which was consistent with findings from others (42). The Global Moran’s I for CHE duration was statistically significant only in 2012 (p < 0.05). The lack of statistical significance in Moran’s I for CHE duration during 2014–2020 can be attributed to three factors. First, following the expansion of universal health insurance coverage and targeted increases in reimbursement rates for rural and low-income groups, the regional gap in financial protection against CHE narrowed. Second, the sustained migration of rural working-age populations to urban areas altered the demographic structure of high-risk groups for persistent CHE. Third, the increasing prevalence of chronic diseases shifted the disease burden from historically high-prevalence regions to more geographically dispersed areas. Local Indicators of Spatial Association (LISA) further supported this transition. Our results identified high-high clusters of CHE duration in Gansu and Ningxia provinces in both 2012 and 2020. This spatial persistence stems from regions’ fragile economic foundations, worsened population aging from young adult outmigration, and an underdeveloped healthcare insurance system unable to provide robust safeguards. Therefore, it is crucial to improve health insurance coverage, support capacity building in county and primary medical institutions, and create local jobs to retain young talent. In contrast, low-low clusters were mainly observed in regions such as Guangxi, potentially due to broader health insurance coverage and policy support, as well as more severe CHE issues in neighboring provinces. Due to its resource-based economy, central fiscal transfers, and lighter demographic burden, Inner Mongolia consistently showed a low-high clustering pattern. For such areas, fostering cross-regional medical cooperation and addressing internal development disparities are recommended. Yunnan and Hunan were high-low clusters in CHE duration in 2020 but low-low in depth, meaning households faced mild shocks but lacked recovery capacity. Addressing this requires strengthening chronic disease management and enhancing household economic resilience.

The GEE and Tobit models have illuminated that larger households, those situated in urban and eastern regions, and households with higher levels of education, employed members, and married household heads are associated with a decreased risk of incurring persistent CHE. The correlation between enhanced education levels and lower incidences of CHE suggests a plausible link to increased awareness of disease prevention and superior healthcare knowledge among families with higher education (43). In larger families with married household heads, the distribution of economic risks is apparent, indicative of a greater capacity to afford healthcare costs (25, 43). Additionally, the stability in income and social security for employed household heads further contributes to mitigating the risk of CHE (28). These factors primarily guarded against household financial distress and facilitate rapid recovery by enhancing ex-ante prevention and immediate payment capacity, as well as by relying on the household’s long-term financial resilience, social capital, and continuous risk-sharing capacity.

Compared to URBMI/NRCMS, UEBMI demonstrated a significant reduction in the duration and depth of CHE among the sampled families. This finding aligns consistently with existing literature, which indicates that UEBMI are capable of protecting some households from the impact of CHE (23, 44). UEBMI predominantly serves urban employed individuals, who generally enjoy more stable incomes and superior social security coverage. Conversely, the URBMI/NRCMS extends coverage to both urban and rural residents, some of whom may have lower incomes and less comprehensive social security coverage, thereby potentially increasing their vulnerability to CHE (20). Furthermore, UEBMI provides higher reimbursement rates, thus facilitating a more effective reduction of the financial burden of household medical expenses compared to URBMI/NRCMS (45).

Families experiencing overlapping vulnerabilities in their economy, health, and social conditions were more susceptible to falling into the cycle of poverty and illness (46, 47). Our study identified several risk factors for chronic household economic burden, including that households with a higher number of family dependents, members with chronic diseases, those living in low economic conditions and rural areas, those utilizing hospitalization services, and those with a low-level health status of the household head. For families managing chronic illness or disability, healthcare costs represent a persistent financial drain. If economic hardship forces them to compromise treatment, it can precipitate a cycle of deteriorating health and escalating medical expenses. To cope, households often resort to strategies like selling productive assets or taking on informal loans, which severely erode their long-term capacity to generate income and escape poverty (48). This situation is exacerbated when the ill individual or their caregiver is the primary earner, leading to a sustained loss of household income. The concurrent shock of rising expenses and falling income constitutes a core mechanism that deepens and perpetuates CHE. Consequently, it is crucial to implement effective, targeted interventions to disrupt this cycle.

Several factors contribute to the limitations of the findings in this study. Firstly, it’s important to note that OOP medical expenses recorded in the CFPS survey only encompass direct medical costs, excluding non-medical expenses such as transportation, accommodation, and income losses. This exclusion may result in an underestimation of both the frequency and severity of CHE. Consequently, future studies should focus on the mechanisms that influence these factors. Secondly, considering that this study relies on self-reported health information provided by household heads, there is a possibility of reporting errors. Such errors could potentially affect the accuracy and reliability of the findings.

The findings underscore the need for a more equitable and sustainable health financing system to mitigate the persistence of CHE. First, the adoption of a progressive universalism framework, which prioritizes low-income and high-risk populations, is central to achieving equitable universal health coverage and optimizing financial protection (49). Second, expanding reimbursement rates and outpatient and chronic-disease coverage, while dynamically adjusting benefit packages and national reimbursement drug lists, would reduce the persistent financial burden on households with chronic conditions or rapidly deteriorating health status. Meanwhile, narrowing institutional and regional disparities requires integrating urban–rural insurance schemes, raising pooling levels, and harmonizing benefit packages between the UEBMI and URBMI/NRCMS systems (50). Finally, establishing a targeted assistance and early-warning mechanism, underpinned by persistent monitoring of CHE, is instrumental in breaking the “illness–poverty” cycle. Cross-sectoral collaboration among medical insurance, social welfare, and financial services can facilitate this shift. By enabling stratified assistance and low-interest medical loans, coupled with chronic-disease management and primary-care follow-up, the system can transition from ex-post financial compensation to a paradigm of proactive, pre-emptive prevention (49, 51).