Authors: Yazhuo Liu, Yunlong Song, Yongjian Xu, Jia Wu, Hui Li, Jiaxin Sun, Xueyan Han, Shi An, Yinjun Tan, Yongbing Cheng
Categories: Research, Long-term care insurance, Financial risk protection, Catastrophic health expenditure, Older adults, Quasi-experiment, Difference-in-differences, Mediation analysis, China
Source: BMC Geriatrics
Authors: Yazhuo Liu, Yunlong Song, Yongjian Xu, Jia Wu, Hui Li, Jiaxin Sun, Xueyan Han, Shi An, Yinjun Tan, Yongbing Cheng
Protecting households from financial risk is widely accepted as a desirable objective of health system, and the same goes for Long-Term Care Insurance (LTCI). However, LTCI’s potential effect on household financial risk protection, still remains unclear. This study aims to systematically evaluate the effects of LTCI and its different benefit compensation strategies on financial risk protection, examining the mechanisms.
A quasi-experimental design was employed using nationally representative data from the China Health and Retirement Longitudinal Study in 2015 and 2018. Financial risk protection was measured by catastrophic health expenditure (CHE) among households aged 60 and above. A total of 3,733 households were included. A difference-in-differences (DID) analysis was performed to identify the LTCI effects, with robustness checked by coarsened exact matching. Bootstrap-based mediation analysis was conducted to test the mechanisms, including substitution effect (outpatient and inpatient service utilization) and the income effect (per capita household income).
The CHE incidence decreased by 6.62% before and after LTCI implemented in the pilot group (18.82% vs. 12.20%), whereas it increased by 3.25% in the non-pilot group (15.16% vs. 18.41%). DID analysis found that a reduction in CHE incidence was statistically significant after LTCI was piloted (OR: 0.384, P < 0.01). The mixed benefit compensation (combining service and cash benefits) statistically reduced CHE incidence (OR: 0.402, P < 0.01), whereas the service-only benefit compensation did not (OR: 0.290, P > 0.05). Mediation analysis showed that the indirect effects were statistically significant in outpatient service utilization (− 0.026, P < 0.001), inpatient service utilization (0.031, P < 0.05), and per capita household income (− 0.011, P < 0.05), mediating the association between LTCI and CHE.
LTCI enhanced financial risk protection for households with older adults in China through the dual pathways of substitution effect and income effect. Furthermore, its protective role was particularly significant in mixed benefit compensation strategy. The study suggests that universal coverage for LTCI should be prioritized on the agenda and policymakers should strive to adopt flexible benefit compensation strategy to enhance the role of LTCI in protecting households against financial risks.
The online version contains supplementary material available at 10.1186/s12877-026-07382-1.
Population aging has emerged as a major global public health challenge, with China experiencing a particularly accelerated demographic transition [1]. By the end of 2024, more than 310 million people in China had been aged over 60 years, accounting for 22.0% of the total population [2]. This aging process coincides with a shift in disease patterns, characterized by a high prevalence of non-communicable chronic diseases (NCDs) and functional disabilities among older adults [3, 4]. The growing demand for long-term care (LTC) imposes heavy direct medical costs and indirect opportunity costs on households, frequently triggering catastrophic health expenditure (CHE), situation where health costs exceed a household’s payment capacity, threatening its financial stability [5–7].
Traditionally, informal care provided primarily by family members (e.g., spouses and children) served as a vital “informal financial buffer” [8]. Yet this buffer has weakened in China as urbanization one-child policy have reshaped household size and intergenerational living arrangements [9, 10]. From a household economics perspective, the heavy reliance on informal care can bring a double burden. It may fail to meet professional care needs while simultaneously crowding out caregivers’ labor supply, thereby reducing household income and increasing financial vulnerability [11, 12]. Consistent with this concern, approximately one-third of older adults with limitations in activities of daily living (ADL) report unmet care needs [13], which may accelerate health deterioration and amplify future medical spending.
In response to the limitations of informal care, many countries have expanded formal care (e.g., institutional and community-based services) to reduce households’ LTC-related financial burden [14]. Long-term care insurance (LTCI) serves as an effective formal care response, providing sustainable coverage for care service costs, such as Japan and Germany [15–17]. In China, LTCI was first piloted in 15 cities in 2016 and had expanded to 49 cities by 2020 [18]. The scheme mainly targets individuals with severe functional disabilities who are enrolled in the Urban Employee Basic Medical Insurance (UEBMI). The policy employs a multi-channel financing and typically reimburses around 70% of eligible costs. Additionally, pilot cities adopted different benefit compensation strategies. Most implemented the service-only benefit compensation, whereas about 30% introduced a mixed benefit compensation combining services with cash benefits.
A growing body of studies has assessed the LTCI effects, with findings that are mixed across three primary domains. First, evidence on healthcare utilization and spending increasingly emphasizes structural changes rather than total costs only. Some studies suggest that LTCI may release previously unmet medical demand and increase overall expenditures [19, 20]. However, more detailed analyses revealed a structural substitution, whereby LTCI significantly reduced “social hospitalization” and unplanned readmissions through maintenance and preventive care [9, 21, 22]. Second, studies of health outcomes generally reported benefits beyond mortality reduction, including improved self-rated health and slower deterioration in ADL among older adults [23–25]. Furthermore, professional care has also been associated with fewer depressive symptoms and higher subjective well-being, particularly among people with NCDs and those in rural areas, potentially narrowing health inequalities [26–28]. Third, studies on family spillovers highlighted both time and financial effects. LTCI may ease caregivers’ time constraints and increase labor supply, especially among women [11, 29, 30]. Despite these advances, studies remain underdeveloped in two key respects. On the one hand, most studies focus on individuals (e.g., medical spending, labor supply) and pay limited attention to the household as the central unit of financial risk. In family-based societies, functional disability often produces a shared financial shock, however it remains unclear whether the observed structural substitution translates into meaningful protection against household-level financial risk. On the other hand, the studies lack an integrated theoretical framework to explain how LTCI generates financial risk protection and why different benefit compensation designs might bring different outcomes. In many studies, LTCI is treated as a “black box”, with limited theoretical separation of the channels linking it to household financial risk.
To address these gaps, this study develops an analytical framework grounded in Grossman’s health capital model [31] and consumer demand theory [32]. Grossman’s model conceptualizes that health is a depreciating capital stock that requires ongoing investment through medical and care services. Within this framework, LTCI can be viewed as an exogenous economic shock that shifts both the shadow prices of health inputs and the household budget constraint, thereby shaping financial risk protection through two parallel substitution effect and income effect. The substitution effect operates through relative price changes. By lowering the effective price of formal care compared with hospital-based services, LTCI can encourage a shift from costly curative care (repairing depreciated health capital) to more cost-effective maintenance care (slowing depreciation). Evidence from Canada and the UK similarly suggests that subsidized long-term care can substitute for prolonged hospital stays (including social hospitalization) and reduce the medical expenditures that bring financial risk [33, 34]. Meanwhile, the income effect arises when LTCI relaxes household budget constraints. Through the dual mechanisms of reimbursing direct care costs and releasing informal caregivers into the labor market, LTCI enhances the household’s real purchasing power. This expanded financial capacity enables households to invest in health services that were previously unaffordable.
Accordingly, the effect of LTCI on household financial risk protection may be associated with both the substitution effect from cost savings and the income effect induced by demand. Guided by this theoretical framework, this study aims to systematically evaluate the effect of LTCI on household financial risk protection using a quasi-experimental design, and examine heterogeneity in these effects across different benefit compensation strategies (service-only compensation and mixed compensation). Furthermore, the study aims to employ mediation analysis to test the substitution effect and the income effect. By rigorously evaluating LTCI’s effect, this study contributes to providing valuable insights into the household financial risk protection in China.
A quasi-experimental design was employed to examine the effect of LTCI on financial risk protection using nationally representative survey data, the China Health and Retirement Longitudinal Study (CHARLS). CHE was used to measure financial risk protection among older households in this study. Households covered by UEBMI were selected as the analytical sample. Cities implementing LTCI were included in the pilot group, while those not implementing it were included in the non-pilot group. Given that different cities adopted different benefit compensation strategies during implementation, the differential effects of LTCI on CHE under these varying strategies were examined. Additionally, a difference-in-differences (DID) analysis was employed to compare changes in CHE incidence across LTCI-enrolled and non-enrolled households before and after implementation of the pilot policy.
Based on the theoretical analysis presented earlier, this study developed an analytical framework (Fig. 1) to explore the mechanisms through which LTCI affected household financial risk protection (measured by CHE). The framework identified two parallel mediating the substitution effect and the income effect. In the substitution effect path, LTCI operated by influencing medical service utilization. Specifically, this study employed inpatient service utilization and outpatient service utilization as the mediating variables. This path examined whether LTCI lowered the incidence of household CHE by reducing reliance on high-cost inpatient and outpatient services. In the income effect path, LTCI operated by influencing household economic status. Utilizing per capita household income as the mediating variable, this path analyzed whether LTCI affected the probability of CHE by alleviating household budget constraints and enhancing affordability. Therefore, we would test the mediation hypothesis shown in the figure to clarify the mechanism between LTCI and household financial risk protection.
Fig. 1Analytical framework
CHARLS is a high-quality, nationally representative survey covering residents aged 45 years and older in China, and it provides comprehensive data on demographic information, household characteristics, medical service utilization, household consumption, etc. A multistage proportional probability sampling (PPS) method was employed to select respondents from 150 counties/districts and 450 villages/resident committees across 28 provinces and autonomous regions in China. The national baseline survey was launched in 2011, followed by subsequent surveys in 2013, 2015, 2018, and 2020. Tracking data showed that the sample size of respondents and their spouses gradually increased from 17,708 in the baseline wave (2011) to 19,395 in the fifth-wave survey (2020). More details on the CHARLS sampling procedure and questionnaire are available on the official https://charls.pku.edu.cn/.
Given that this study focused on the sample of households with at least one member aged 60 years and above (the mean age of the older adults was 69.6 years), the following data processing steps were conducted. Firstly, data from the 2015 and 2018 waves of CHARLS were used for the main effect analysis. Information regarding LTCI—including its pilot time, benefit compensation strategies, and coverage scope—was systematically collected and presented in Supplementary Table S1. The year 2015 was identified as the pre-pilot period, whereas 2018 was post-pilot period. Given the assumption that only cities with an LTCI pilot duration of more than 6 months prior to the survey would be affected by the policy, cities piloted after March 2018 were excluded (as the CHARLS 2018 survey began at the end of September). 17 pilot cities were finally included in the pilot group, and 104 cities were included in the non-pilot group. A total of 3,733 households were included, of which 615 households were in the pilot group and 3,118 households were in the non-pilot group. Secondly, a key assumption for DID estimation is that CHE incidence for the two groups (LTCI-enrolled and non-enrolled) must follow parallel trends prior to the implementation of LTCI. Therefore, data from the 2011 and 2013 waves were included to test the parallel trend assumption. The distribution of pilot cities and additional details on raw data processing are shown in Fig. 2.
Fig. 2Distribution of LTCI pilot cities and flow chart of the sample selection process. Notes: The map review number is GS (2024)0650 and drawn by ArcGIS 10.8.1 in Fig. 1(a)
The dependent variable in this study was whether households incurred CHE, a widely recognized measure for assessing household financial risk protection [35]. In this study, CHE was defined as out-of-pocket (OOP) health spending exceeding 40% of a household’s capacity to pay (CTP) [36]. OOP health spending includes both direct and indirect medical costs incurred by households over the past year, excluding any portion reimbursed by health insurance, as reported in the CHARLS questionnaire. CTP refers to a household’s effective income minus basic subsistence needs adjusted for household size. CHE is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ :che=\left{\begin{array}{c}1\text{} ;if;oop {h}/ct{p}{h}\ge:0.4\:0 ; if ; oop_{h}/ct{p}_{h} < 0.4 \end{array}\right.
Where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\:\mathrm{oo}{\mathrm{p}}_{h}\: $$\end{document}refers to a household’s OOP health spending, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\mathrm{ct}{\mathrm{p}}_{h}\: $$\end{document}represents a household’s CTP. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\mathrm{ct}{\mathrm{p}}_{h}\: $$\end{document}is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:ct{p}_{h}=\left\{\begin{array}{c}{{exp}}_{h}-s{e}_{h}\text{}\mathrm{i}\mathrm{f}\text{}\mathrm{s}{\mathrm{e}}_{h}\le\:foo{d}_{h}\\\:{{exp}}_{h}-foo{d}_{h}\text{}\mathrm{i}\mathrm{f}\text{}\mathrm{s}{\mathrm{e}}_{h}>foo{d}_{h}\end{array}\right. $$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{exp}_{h}\: $$\end{document}denotes household total expenditure, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:foo{d}_{h} $$\end{document} refers to household food expenditures. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\mathrm{s}{\mathrm{e}}_{h} $$\end{document} is the subsistence expenditure, which is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\mathrm{s}{\mathrm{e}}_{h}=pl*eqsiz{e}_{h} $$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:pl $$\end{document} denotes the poverty line and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqsiz{e}_{h} $$\end{document} is the equivalent household size. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:pl $$\end{document} is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ pl = \frac{\sum \mathrm{w}_{h} * eqfood_{h} }{\sum \mathrm{w}_{h}}{where\;food\;45< foodexp\;_{h} < food55 } $$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqfoo{d}_{h}\: $$\end{document}denotes equivalent food expenditure, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:\mathrm{foodex}{\mathrm{p}}_{h} $$\end{document} is the proportion of household food expenditure. As indicated by the above equation, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:pl $$\end{document} refers to the weighted average of food expenditure in the 45th to 55th percentile range—this value also represents per capita subsistence expenditure (adjusted for equivalent household size). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqfoo{d}_{h} $$\end{document} can be calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqfoo{d}_{h}=\frac{foo{d}_{h}}{eqsiz{e}_{h}} $$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqsiz{e}_{h} $$\end{document} is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:eqsiz{e}_{h}=hhsiz{e}_{h}^{0.56} $$\end{document} Where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:hhsiz{e}_{h} $$\end{document} denotes household size. #### Control variables Drawing on the conceptual framework of social determinants of health (SDH) and previous studies, socioeconomic characteristics and health-related characteristics were included as control variables [37–39]. Socioeconomic characteristics included region, household size, educational attainment (defined as the highest educational attainment of all household members), households with children aged 5 years or younger, households with older adults aged 60 years or above, and household economic status. Household economic status was grouped into tertiles based on per capita household consumption expenditure, calculated by subtracting OOP health spending from total expenditure and dividing by household size [40]. Health-related characteristics included outpatient service utilization, inpatient service utilization, functional disability, and NCDs among household members. For medical service (1) A household was coded as “Yes” if any member had used inpatient or outpatient services in the past year or month; (2) Otherwise, it was coded as “No”. For functional (1) A household was coded as “Yes” if any member experienced functional disability—defined as ADL scores of 2, 3, or 4, or IADL (instrumental activity of daily living) scores of 3 or 4, based on questionnaire responses; (2) Otherwise, it was coded as “No”. More details are available in Supplementary Table S2. ### Statistical analysis #### Difference-in-differences analysis To empirically evaluate the effect of LTCI on household CHE, we employed a DID analysis combined with Coarsened Exact Matching (CEM) to mitigate potential confounding bias [41]. Within the standard DID framework, this study employed the following empirical \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{Y}_{it}={\beta\:}_{0}+{\beta\:}_{1}*{Post}_{t}+{\beta\:}_{2}*{Treat}_{it}+{\beta\:}_{3}*\left({Time}_{t}*{Treat}_{it}\right)+{X}_{it}+{ϵ}_{it} $$\end{document} Where the dependent variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{Y}_{it}\: $$\end{document}denotes CHE for households \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:i $$\end{document} in wave \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:t $$\end{document}. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{Time}_{t}\: $$\end{document}is a time dummy variable, coded as 1 for the post-pilot year (2018) and 0 for the pre-pilot year (2015). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{Treat}_{it} $$\end{document} is a group dummy variable, coded as 1 for the pilot group and 0 for the non-pilot group. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{Time}_{t}*{Treat}_{it} $$\end{document} is the core DID interaction term, capturing the intention-to-treat (ITT) effect of LTCI on CHE in pilot cities—a quantity of particular interest to policymakers [42]. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{X}_{it} $$\end{document} refers to a set of control variables described above, including basic household characteristics and health-related variables. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \:{ϵ}_{it} $$\end{document} is the error term. The validity of the DID analysis relies on two essential assumptions. Firstly, the parallel trends assumption must be satisfied, which requires that no statistically significant difference in CHE incidence existed between the two groups prior to the LTCI pilot (based on counterfactual inference). The study applied an event study method to test this assumption. Secondly, the robustness assumption also needs to be satisfied, requiring that the pilot LTCI has no effect on the non-pilot group. Placebo tests and the CEM method were conducted to evaluate the validity of this assumption. To ensure comparability between the pilot and non-pilot groups, this study performs CEM [43]. Households were matched based on key baseline covariates, including household size, household economic status, and the health status of older adults. The L1 statistic was calculated to assess the improvement in covariate balance post-matching [44, 45]. Unmatched observations were excluded from the subsequent analyses. #### Mediation analysis A bootstrap-based mediation analysis was used to test the substitution-effect and income-effect pathways. Predictors (LTCI), mediators (inpatient utilization, outpatient utilization, and per-capita household income), and the outcome (CHE) were measured in both the 2015 and 2018 CHARLS waves. Therefore, the analysis did not impose a strict temporal ordering and instead evaluated the hypothesized indirect pathways as an associational structure using repeated measurements across the two waves. We treated the two waves as repeated observations, included a wave indicator (2015 vs. 2018) and covariates, and accounted for within-household dependence (e.g., clustering standard errors at the household level) to estimate the overall associational indirect effect. Due to CHE and medical service utilization are binary variables, indirect effects were computed using a counterfactual (predicted-value) decomposition on the predicted-probability scale rather than the product-of-coefficients method for linear models [46, 47]. Statistical significance of the indirect effects was tested using a non-parametric bootstrap method with 1,000 replications to obtain bias-corrected 95% confidence intervals (CI) [48]. Additionally, descriptive statistics were used to summarize households’ basic characteristics. Categorical variables were reported as counts (n) and percentages (%). Continuous variables were summarized using means and standard deviations (SD). All statistical analyses were performed using Stata MP 17.0. A two-sided *P* < 0.05 was considered statistically significant. ## Results ### Basic characteristics of the sample Table 1 reports the basic characteristics of the sample. Before the implementation of the LTCI pilot (2015), significant differences were observed between the pilot and non-pilot groups in specific socioeconomic and health domains. Compared to the non-pilot group, households in the pilot group exhibited a higher proportion of members with high school education or above (55.75% vs. 51.94%) and a higher prevalence of functional disability (82.58% vs. 77.04%). In terms of healthcare utilization, the pilot group reported higher outpatient service usage (32.75% vs. 28.74%) but slightly lower inpatient service usage (26.48% vs. 27.08%). Additionally, per capita income in the lowest income group was marginally higher in the pilot group. No statistically significant differences were observed in other demographic or household characteristics, supporting the comparability of the groups after controlling for these covariates. Table 1Basic Characteristics of the sampleVariablesNon-pilot GroupPilot Group2015201820152018 *n* % *n* % *n* % *n* %Region Eastern38427.7346827.0120170.0322167.38 Central42630.7653030.583712.894212.80 Western40729.3954931.68155.23216.40 North-eastern16812.1318610.733411.854413.41Household size 1795.7017410.04165.57216.40 298270.9092153.1420471.0816349.70 3 or above32423.3963836.816723.3514443.90Educational attainment Middle school or below66748.1629917.2512744.254614.02 High school42930.9751529.7210436.2410231.10 Undergraduate or above28920.8791953.035619.5118054.88Children aged 5 or below No1,34897.331,63794.4627997.2130592.99 Yes372.67965.5482.79237.01Outpatient service utilization No98771.261,27173.3419367.2522568.60 Yes39828.7446226.669432.7510331.40Inpatient service utilization No1,01072.921,21369.9921173.5223671.95 Yes37527.0852030.017626.489228.05Functional disability No31822.9650128.915017.426921.04 Yes1,06777.041,23271.0923782.5825978.96NCDs No27419.781609.235619.513410.37 Yes1,11180.221,57390.7723180.4929489.63Household economic status, CNY (M, SD) Lowest9,307.442,952.269,133.213,317.358,717.352,793.599,234.333,033.21 Middle19,398.553,470.8519,490.853,368.0319,554.943,453.3019,882.603,206.32 Highest59,073.2759,027.7559,280.3791,122.3758,468.1846,207.7758,238.2967,484.02 Observations1,3851,733287328Unit of household economic Chinese yuan (CNY) ### Univariate analysis of changes in financial risk protection before and after LTCI implementation Table 2 presents the descriptive changes in financial risk protection indicators (CHE, OOP, and CTP) before and after the LTCI implementation. The CHE’s incidence decreased from 18.82% to 12.20% before and after the pilot implementation in the pilot group, whereas it increased from 15.16% to 18.41% in the non-pilot group. Both changes were statistically significant (*P* < 0.05). The DID estimator indicated a reduction of 9.87% in CHE incidence attributed to the policy. Similarly, OOP medical expenses decreased in the pilot group while increasing in the non-pilot group, resulting in a statistically significant reduction (DID = -2,823.59 CNY). Although household CTP increased in both groups, the DID estimator was not statistically significant. Table 2Univariate analysis results of changes in financial risk protection before and after LTCI implementationVariablesGroupPre-pilotPost-pilotDifferenceDIDCHE (%)Pilot Group18.8212.20-6.62**-9.87**Non-pilot Group15.1618.413.25**OOP (CNY)Pilot Group6,782.516,194.81-587.70*-2,823.59**Non-pilot Group6,565.698,801.582,235.89**CTP (CNY)Pilot Group41,611.0149,875.018,264.006.32Non-pilot Group40,703.4848,961.348,257.68This table presents the results of the univariate test based on the pilot and non-pilot groups. The significance levels of 1‰, 1%, and 5% are denoted by ***, **, and *, respectively ### DID analysis for the effect of LTCI on CHE The results of the DID analysis are presented in Table 3. After adjusting for confounding variables, the core interaction term, Time × Treat, yielded an OR of 0.384 (*P* < 0.01), meaning that the implementation of LTCI significantly reduced the incidence of CHE. Robustness checks described in the Supplementary Materials verified these findings. Supplementary Figure S1 confirmed that the parallel trends assumption was satisfied before the pilot and that the results passed the placebo test. Furthermore, the CEM-DID analysis presented in Supplementary Table S3 verified the significant negative effect of LTCI on CHE after balancing covariates. Table 3DID analysis results for the effect of LTCI on CHEVariablesORS.E.z-value95% CILowerUpperTime* Treat0.384**0.124-2.9590.2030.724Time (Ref: 2015) 20181.558**0.3701.8480.9792.482Treat (Ref: Non-pilot Group) Pilot Group1.5940.2353.1151.1942.128Region (Ref: Eastern) Central1.0580.1720.1020.7701.455 Western1.0260.1680.1570.7451.412 North-eastern0.494**0.115-3.1790.3120.781Household size (Ref: 1) 20.620*0.138-2.2190.4000.961 3 or above0.457**0.115-3.4880.2790.747Educational attainment (Ref: Middle school or below) High school1.1290.2160.5890.7771.642 Undergraduate or above0.9740.189-0.9120.6661.426Children aged 5 or below (Ref: No) Yes0.446*0.180-1.8150.2020.983Outpatient service utilization (Ref: No) Yes1.471**0.1903.1771.1411.895Inpatient service utilization (Ref: No) Yes6.288***0.97012.2174.6478.508Functional disability (Ref: No) Yes0.8730.121-0.5920.6661.145NCDs (Ref: No) Yes2.994***0.7303.9681.8574.828Household economic status (Ref: Lowest) Middle0.289***0.053-7.4930.2020.413 Highest0.025***0.008-13.3410.0130.047 Constant0.122***0.046-6.2700.0580.256 Log likelihood-1,293.593 Wald chi2263.280 Prob > chi2< 0.001The significance levels of 1‰, 1%, and 5% are denoted by ***, **, and *, respectively ### Heterogeneity analysis of the benefit compensation strategy of LTCI Table 4 shows the different effects of benefit compensation strategies. In Model 1, the interaction term was not statistically significant (*P* > 0.05), suggesting that the service-only benefit compensation did not independently lead to a significant reduction in CHE incidence in this sample. In Model 2, the interaction term was 0.402 (*P* < 0.01), indicating that the mixed benefit compensation model (combining service and cash benefits) statistically significantly reduced the CHE incidence compared to the non-pilot group. Table 4Effect of different benefit compensation strategies on CHEVariablesModel 1Model 2ORS.E.ORS.E.Time*Treat0.2900.2280.402**0.130Time (Ref: 2015) 20181.522**0.2171.538**0.223Treat (Ref: Non-pilot Group) Pilot Group1.5230.8311.5610.392Control Variables *YES* *YES* Constant0.090***0.0350.110***0.039Log likelihood-1,118.950-1,271.871Wald chi2232.660259.240Prob > chi2< 0.001< 0.001The significance levels of 1‰, 1%, and 5% are denoted by ***, **, and *, respectively. Model 1 and Model 2 represent the DID analysis results for service-only benefit compensation and the mixed benefit compensation, respectively ### Mechanism analysis To verify the pathways driving financial protection, we tested the substitution effect and the income effect using bootstrap mediation analysis (1,000 resamples). As illustrated in Fig. 3, LTCI implementation was statistically associated with the mediators. Specifically, the policy was associated with significant reductions in both outpatient service utilization (Coef. = -0.910, *P* < 0.01) and inpatient service utilization (Coef. = -0.105, *P* < 0.05). Simultaneously, the policy was associated with a significant increase in per capita household income (Coef. = 1.708, *P* < 0.05), supporting the income effect hypothesis. The decomposition of effects revealed statistically significant indirect pathways. The indirect effect via outpatient service utilization was − 0.026 (*P* < 0.001), and via inpatient service utilization was 0.031 (*P* < 0.05). The indirect effect via per capita household income was − 0.011 (*P* < 0.05). Fig. 3Mechanism analysis ## Discussion In China, LTCI was implemented to reduce the financial burdens and protect families from financial risk related to LTC [49]. To rigorously evaluate LTCI’s financial risk protection, this study systematically assessed the effects of LTCI and its different benefit compensation strategies on reducing CHE using a DID analysis. The analysis employed data from the 2015 and 2018 waves of CHARLS, focusing on households with at least one member aged 60 years and above. Several findings of this study are highlighted below. ### LTCI’s role in financial risk protection and its interpretation within a global context This study provides quasi-experimental evidence that China’s LTCI pilot significantly reduced the incidence of CHE among urban households with older adults. This finding aligns with the global trend of “de-familialization” in long-term care, where policy interventions are replacing traditional family caregiving to reduce financial risk. Consistent with experiences in Japan and South Korea, where the introduction of LTCI significantly curbed the depletion of household assets for care [50, 51], our results confirm that China’s LTCI acts as an important “financial firewall,” effectively protecting families from the financial risk of functional disability. This effect was primarily manifested as a significant decrease in OOP medical expenses and a simultaneous increase in household CTP [9, 52]. This specific reduction in OOP expenses can be attributed to the policy’s comprehensive design, which substitutes high-cost medical interventions with cost-effective care services [53]. Evidence from China further supports this phenomenon that access to professional care significantly could improvs insured individuals’ self-rated health and reduce depressive symptoms, thereby accumulating health capital that lowered the demand for medical services and ultimately curbed OOP expenditures [54]. These benefits are particularly pronounced among vulnerable subgroups, including patients with NCDs and rural residents [55, 56]. ### The mixed benefit compensation a comparative perspective A key finding of this study is the heterogeneity of benefit compensation the mixed benefit compensation (combining service and cash benefits) demonstrated a significant protective effect on CHE, whereas the service-only benefit compensation did not. This finding offered a unique dialogue with international models. Internationally, Japan’s LTCI (implemented in 2000) strictly adopted a service-only benefit compensation to ensure care quality and prevent family caregivers from becoming hidden labor merely for cash subsidies [57, 58]. Conversely, Germany’s LTCI (implemented in 1995) offers a choice, where nearly 50% of beneficiaries opt for cash allowance to compensate family caregivers, reflecting a cultural preference for home care [59, 60]. Our finding that the mixed benefit compensation outperformed the service-only benefit compensation in China, differing from the mature markets of Japan and Germany. On the one hand, from the supply-side perspective, unlike Japan, which possessed a robust community care infrastructure before 2000, China’s formal care market is still nascent. In service-only benefit compensation pilot cities, the shortage of qualified long-term care facilities may prevent eligible families from utilizing their benefits effectively, rendering the insurance nominal. However, it is important to point that this is a theoretical deduction rather than empirical evidence derived from our data, and future studies are needed to empirically test the role of care supply shortages. On the other hand, from the demand-side perspective, the mixed benefit compensation, partly similar to the German model, provides cash that directly supplements household disposable income. In lower-income Chinese households, this cash liquidity is crucial for covering indirect medical costs (e.g., transportation, nutrition) that induce CHE [61]. Therefore, the mixed benefit compensation has been proven to perform effectively in China. It bridges the gap between rigid market supply and flexible household needs, successfully strengthening financial risk protection. However, this comes at the potential risk of underperforming in health maintenance if the cash is not strictly tied to care provision. As Lei and Zhang highlighted that cash benefits in China’s LTCI pilots may trigger a “cash-out puzzle” [62]. Their study found that while the service-only benefit compensation significantly reduced the mortality, the cash compensation model showed no such significant effects. This issue is related to the principal-agent relationship within the household. In this context, family members (agents) may focus more on the overall household benefit or their own leisure, rather than meeting the specific care needs of the older adults (principals). As a result, they may receive benefits without providing necessary care in return. That is saying, China’s current or future challenge is not simply choosing between models, but solving the trade-off between household financial risk protection and health quality as it transitions toward a more mature care system. ### Mechanisms: substitution and income effects The mediation analysis confirmed two protective the substitution effect and the income effect. First, regarding the substitution effect, LTCI reduced CHE by decreasing medical service utilization, especially inpatient service utilization. This mirrors the historical experience of Japan, where LTCI was explicitly introduced to solve the crisis of social hospitalization (bed-blocking by the patients with no medical needs) [63]. Our results indicate that China is successfully replicating this efficiency by lowering the shadow price of community care, LTCI corrects the distortion where families used hospitals as “nursing homes” covered by medical insurance. This structural adjustment is consistent with findings from Canada, the UK, and South Korea [33, 64, 65]. Second, regarding the income effect derived from labor force participation, LTCI helped adult children and spouses remain in the labor market by replacing informal caregiving time. This effect is particularly potent in East Asian societies (China, Japan, Korea) characterized by strong filial piety norms, where the baseline opportunity cost of caregiving is higher than in Western welfare countries [30, 66–68]. The increased household income from sustained employment acts as a secondary buffer against medical shocks, reducing household financial risk. ### Policy implications According to these findings and international comparisons, three policy implications are proposed. First, our findings indicate that the mixed benefit compensation strategy currently provided a better financial protection. Therefore, an immediate shift toward a unified service-only benefit compensation, as seen in Japan, should be approached with caution. Retaining a mixed benefit compensation may act as a necessary buffer, potentially ensuring that families receive tangible support (cash) in areas where professional services might still be inaccessible. Second, greater focuses should be placed on the formalization and capacity building of family caregivers. Insights from Germany suggest that cash benefits should not be viewed merely as passive subsidies, but rather as compensation for informal care services. Linking cash benefit compensation to training for family caregivers would help safeguard care quality. This approach effectively combines the flexibility of cash benefits with the quality assurance of service-based provision. Third, policy design should give higher priority to preventive and early-intervention care to reinforce the substitution effect identified in our analysis. To maximize substitution effects, To maximize the substitution effect, more efforts should be prioritized to establish a tiered system of targeted preventive benefits for individuals with mild functional impairments, as evidenced by Japan’s 2000 reforms. By promoting early interventions before severe disability develops, such as rehabilitation and functional training, the system can reduce long-term dependence on costly hospital-based care and address financial risk at its source. Overall, expanding LTCI coverage and continuously enhancing benefit packages should be prioritized in China’s LTCI reform. ### Limitations This study has several limitations. First, recall bias may be unavoidable in CHARLS, particularly regarding specific expenditure items relying on retrospective self-reports. While this is a common challenge in large-scale surveys, measurement error may affect the precision of our estimates. Second, due to data constraints, this study was unable to estimate the heterogeneity of CHE incidence across different long-term care settings (e.g., home-based vs. institutional care). Given that cost structures vary significantly between care settings, future studies are warranted to identify which setting offers the highest cost-effectiveness using more granular data. Third, to ensure the comparability of the treatment and control groups, our analysis was restricted to participants enrolled in the UEBMI. Consequently, caution should be exercised when generalizing these findings to rural residents or those covered by the Urban and Rural Residents Basic Medical Insurance. Finally, a key limitation is that predictors, mediators, and outcomes were measured in both the 2015 and 2018 waves, which precludes establishing a strict temporal ordering required for causal mediation. Therefore, the estimated indirect effects should be interpreted as associational rather than causal. Future studies using methods such as Cross-Lagged Panel Model are needed to better identify causal mediating effects. ## Conclusions This study provides strong empirical evidence that China’s LTCI pilot policy significantly reduced the CHE incidence among urban households. Notably, we identified a transitional advantage of the mixed benefit compensation strategy (combining service and cash benefits), which demonstrated superior financial protection compared to the service-only benefit compensation in the current market environment. Mechanism analysis confirms that these protective effects were mediated through a dual the substitution effect, driven by reduced reliance on high-cost medical care, and the income effect, driven by enhanced household purchasing power. These findings suggest that expanding LTCI coverage should be a policy priority. Furthermore, a phased optimization strategy is recommended to refine the LTCI system, as opposed to a premature unification of benefit compensation models. ## Supplementary Information Supplementary Material 1.