Medical insurance scheme and out-of-pocket medical burden among Chinese adults: repeated cross-sectional evidence from CFPS 2016–2022

Abstract

Background

China has achieved broad health insurance coverage, but financial protection may still differ across schemes. We assessed whether insurance scheme was associated with out-of-pocket medical burden among Chinese adults.

Methods

We analyzed repeated cross-sectional adult person-wave data from the China Family Panel Studies for 2016, 2018, 2020, and 2022. Adults were classified as employee/public insured, resident insured, uninsured, or covered by other or multiple schemes. The primary outcome was log-transformed annual out-of-pocket expenditure after reimbursement. Among medical spenders, the second financial-protection outcome was an out-of-pocket ratio of at least 50% of total medical expenditure. Weighted regression models adjusted for age, sex, education, relative income rank, employment, urban residence, chronic disease, and survey year. Exact-scheme and sensitivity analyses tested robustness.

Results

The study included 120,063 adult observations; 107,116 had a known insurance scheme, 49,111 entered the primary out-of-pocket model, and 8,034 medical spenders entered the high-ratio model. In the weighted analytic sample, resident insurance accounted for 69.9% of observations, employee/public insurance for 18.0%, and no insurance for 8.7%. Compared with employee/public insurance, resident insurance was associated with higher log out-of-pocket expenditure (beta 0.275, 95% CI 0.188 to 0.362, p < 0.001), as was no insurance (beta 0.313, 95% CI 0.177 to 0.449, p < 0.001). Among medical spenders, resident insurance and no insurance were associated with higher probability of an out-of-pocket ratio of at least 50% (risk difference 0.174, 95% CI 0.127 to 0.221, p < 0.001; and 0.177, 95% CI 0.098 to 0.256, p < 0.001, respectively). Exact-scheme, province fixed-effect, clustered-standard-error, income-rank, year-stratified, and alternative-threshold analyses were directionally consistent.

Conclusions

In CFPS 2016–2022, resident insurance and no insurance were associated with higher out-of-pocket medical burden than employee/public insurance. Causal interpretation should remain cautious because plan-level benefit design, health-care utilization intensity, and household capacity to pay were not fully observed.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12889-026-28571-y.

Text box 1. Contributions to the literature
- Insurance scheme differences in China are increasingly policy-relevant because nominal coverage is already high.
- Resident insurance, covering approximately 70% of Chinese adults in the analytic sample, was associated with higher out-of-pocket burden than employee/public insurance.
- The protection gap was visible across four CFPS waves (2016-2022), suggesting a persistent pattern rather than a single-year artifact.
- Exact-scheme analyses showed higher burden for NRCMS, URBMI, and URRBMI relative to UEBMI in at least one financial-burden outcome.
- Additional analyses using clustered standard errors, income-rank stratification, alternative thresholds, province fixed effects, and year-stratified models supported the main association pattern.

Background

Financial protection is a central dimension of universal health coverage, but broad insurance coverage does not guarantee equal protection against out-of-pocket expenditure [1, 2]. The World Health Organization and the World Bank emphasize catastrophic and impoverishing health spending as core indicators for monitoring progress toward universal health coverage [2, 3]. China has built a near-universal social health insurance system through several schemes, including employment-linked insurance, resident insurance, and reforms that merged or replaced earlier urban and rural programs [4–7]. By 2020, more than 95% of the Chinese population was nominally covered [6]. Even after coverage expansion, persistent differences in benefit packages, pooling levels, reimbursement ceilings, co-payment rates, outpatient coverage, and access to designated providers may still translate into unequal financial protection across insured groups [7–10].

China’s basic health insurance schemes differ in their financing and benefit design. Urban Employee Basic Medical Insurance (UEBMI) is employment-linked and financed mainly through employer and employee payroll contributions. Resident-oriented schemes, including Urban Resident Basic Medical Insurance (URBMI), the New Rural Cooperative Medical System (NRCMS), and the Basic Medical Insurance for Urban and Rural Residents (URRBMI), have relied more heavily on individual contributions plus government subsidies and historically had lower financing levels and less generous reimbursement than UEBMI [9, 10]. These basic schemes generally cover inpatient services and selected outpatient services and medicines within locally defined benefit packages, but deductibles, co-payment rates, reimbursement ceilings, and covered service lists vary by locality. They should therefore not be interpreted as high-deductible health plans in the US private-insurance sense; rather, they are Chinese public insurance arrangements with heterogeneous benefit generosity.

Much of the existing literature on China compares insured versus uninsured adults or focuses on one scheme at a time [11]. Several studies have examined catastrophic health expenditure at the household level [12–14] or assessed the impact of specific scheme expansions or integration reforms on health-care utilization [15–17]. However, the contemporary policy question is not only whether adults are insured, but whether different insurance arrangements provide meaningfully different protection within a system that already has high nominal coverage. This is especially relevant after integration of earlier urban and rural resident programs into newer urban-rural resident arrangements, because administrative integration does not necessarily imply convergence in effective financial protection [17, 18]. International evidence also suggests that the transition from nominal coverage to effective coverage can stall when benefit design remains heterogeneous [19].

Despite this growing body of work, few studies have directly compared the financial-protection performance of multiple insurance schemes within a single nationally based multi-wave dataset using individual-level reimbursement-adjusted spending data. The China Family Panel Studies (CFPS) provide repeated adult data with information on insurance participation, total medical expenditure, reimbursement-adjusted out-of-pocket expenditure, and self-rated health [4, 5]. This combination allows repeated cross-sectional evaluation of scheme-specific financial burden using directly measured spending outcomes. We therefore used CFPS 2016–2022 to assess whether insurance scheme was associated with out-of-pocket medical burden among Chinese adults. We hypothesized that resident insurance and no insurance would be associated with higher out-of-pocket burden than employee/public insurance, and that more granular exact-scheme analyses would show weaker protection for resident-oriented schemes such as NRCMS and URBMI relative to UEBMI.

Methods

Data source and study design

We used adult person-level data from the 2016, 2018, 2020, and 2022 waves of CFPS, a nationally based longitudinal household survey in China [4, 5]. The present study was structured as a repeated cross-sectional analysis of harmonized adult person-wave observations across waves. Some respondents contributed observations in more than one wave; each observation was classified according to the respondent’s insurance status reported in that wave. Thus, adults who transitioned between schemes were not excluded or treated as a separate exposure-history group. Instead, each person-wave contributed to the scheme group reported at that interview. The main models treated observations as repeated cross-sections with survey-year adjustment, and we added respondent-level clustered standard-error sensitivity models to assess whether within-person correlation changed the inference. The analytic dataset was assembled from wave-specific adult or person files and restricted to respondents aged 18 years or older.

Exposure: insurance scheme

The main exposure was a harmonized four-level insurance variable. Public medical insurance and UEBMI were grouped as employee/public insurance. URBMI, NRCMS, and URRBMI were grouped as resident insurance. Respondents reporting no insurance were coded as uninsured, and respondents indicating multiple schemes or supplementary combinations were coded as other or multiple insurance. Records with unknown insurance status were excluded from the main scheme analysis.

For sensitivity analyses, we also used a more granular exact-scheme variable restricted to observations with one clearly identified scheme plus the uninsured group: public insurance, UEBMI, URBMI, NRCMS, URRBMI, and no insurance. UEBMI was treated as the reference category because it is the main employment-linked benchmark scheme in the recent Chinese insurance system. CFPS does not provide harmonized plan-level information on premiums, deductibles, co-payment rates, reimbursement ceilings, covered drug lists, or province-specific benefit packages, so these benefit-design features could not be directly adjusted for.

Outcomes

The primary outcome was annual individual out-of-pocket medical expenditure after reimbursement, modeled as log(OOP + 1) to reduce right skewness. In CFPS, total medical expenditure was measured using the annual medical-expense item, and OOP expenditure was measured as the amount of medical costs borne by the respondent after excluding amounts already reimbursed or expected to be reimbursed. This OOP measure does not include insurance premiums. It may include unreimbursed patient-paid amounts generated by deductibles, co-payments, co-insurance, uncovered medicines, or services outside reimbursement rules, but CFPS does not separate these components.

A second financial-protection outcome was a high out-of-pocket ratio of at least 50%, defined among adults with any medical expenditure as out-of-pocket expenditure accounting for at least half of total annual medical expenditure. This individual-level high out-of-pocket ratio is an analytic marker of severe personal cost exposure; it differs from household catastrophic health expenditure indicators, which are commonly defined as health spending exceeding 10% or 25% of total household consumption or income, or exceeding 40% of household capacity to pay [2, 3, 13]. Because the harmonized adult dataset included only ordinal relative income rank rather than household total consumption, non-food expenditure, or a continuous household capacity-to-pay denominator, we did not construct a formal WHO-style catastrophic health expenditure outcome. Instead, we used income rank as an adjusted covariate and added income-rank stratified and interaction sensitivity analyses. A secondary outcome was poor self-rated health. Additional sensitivity outcomes included high out-of-pocket thresholds of 40% and 60%, continuous out-of-pocket ratio among spenders, and log(OOP + 1) among spenders only.

Covariates

Adjusted models included age, sex, years of education, relative income rank, current economic activity, urban residence, chronic disease, and survey year. Age, years of education, and relative income rank (1 to 5, indicating the respondent’s perceived relative income position in the local area) were modeled as continuous linear covariates. Sex, current economic activity, urban residence, and chronic disease were modeled as binary indicators. Survey year was modeled as a categorical variable. Province code was additionally incorporated in a province fixed-effect sensitivity specification.

Statistical analysis

All descriptive and regression analyses used positive survey weights normalized within year to improve comparability across waves while preserving within-wave weighting structure. Weighted linear regression models with heteroskedasticity-robust standard errors were used for log(OOP + 1), poor self-rated health, and linear probability models of high financial-burden outcomes. Monetary outcomes were analyzed in nominal yuan; we did not apply a separate inflation adjustment because models included survey-year fixed effects to absorb common year-level price changes, and the ratio-based high-burden outcomes are invariant to common scaling of spending within a wave. The main complete-case analytic sample for the primary model required non-missing insurance group, weight, outcome, and all covariates.

We estimated exact-scheme models, unweighted sensitivity models, post-2018 weighted models, province fixed-effect models, respondent-level clustered-standard-error models, income-rank stratified and interaction models for high OOP ratio, year-stratified models, and subgroup analyses by sex, urban-rural residence, age younger than 60 years versus 60 years or older, and chronic disease status. Table 1 reports weighted means or proportions; its p values are descriptive omnibus comparisons calculated using one-way ANOVA for continuous variables and chi-square tests for categorical variables. All analyses were performed in Python using pandas and statsmodels. This study followed the STROBE reporting guidelines for observational studies.

Table 1.

Baseline characteristics by insurance scheme

Characteristic	Employee/public	Resident	Uninsured	Other/multiple	Missing	P value
Age, years	49.5	49.3	43.3	42.4	0.0%	< 0.001
Female	56.7%	49.1%	51.6%	57.0%	0.0%	< 0.001
Married	77.1%	78.4%	57.1%	66.6%	2.4%	< 0.001
Education, years	11.9	7.3	9.0	11.9	12.3%	< 0.001
Economically active	67.8%	76.6%	70.0%	81.8%	8.2%	< 0.001
Relative income rank (1–5)	2.8	2.7	2.6	2.9	8.5%	< 0.001
Urban residence	87.0%	51.2%	63.6%	78.5%	1.6%	< 0.001
Chronic disease	19.2%	16.5%	11.9%	15.7%	3.2%	< 0.001
Poor self-rated health	25.0%	32.2%	27.4%	20.7%	0.0%	< 0.001

Values are weighted means for continuous variables and weighted proportions for binary variables

Results

Sample flow and scheme distribution

The harmonized dataset contained 120,063 adult observations across 2016, 2018, 2020, and 2022. Of these, 107,116 had a known main insurance scheme, 96,956 had both a known main insurance scheme and a positive analysis weight, 49,111 entered the complete-case weighted primary model, and 8,034 medical spenders entered the high out-of-pocket ratio model (Fig. 1). Variable missingness is summarized in Supplementary Table S1.

Fig. 1.

Flow of the analytic sample. Adult CFPS observations from 2016-2022 were screened to the known-scheme weighted sample and then to the complete-case main models

Resident insurance represented 69.9% of the weighted analytic sample, employee/public insurance 18.0%, no insurance 8.7%, and other or multiple insurance 3.5%. Over time, resident insurance remained dominant but declined modestly from 72.6% in 2016 to 66.7% in 2022, while employee/public insurance increased from 16.5% to 20.7%. The uninsured share remained smaller but non-trivial, ranging from 7.4% to 10.0% across waves (Fig. 2A).

Fig. 2.

Scheme distribution and financial burden trends over time. Panel (A) shows weighted scheme shares across 2016-2022. Panel (B) shows weighted prevalence of high out-of-pocket ratio >=50% among medical spenders

Baseline characteristics

Baseline characteristics differed substantially across scheme groups (Table 1). Compared with the resident group, adults in the employee/public group were more urban (87.0% vs. 51.2%), more educated (11.9 vs. 7.3 years), and had a lower prevalence of poor self-rated health (25.0% vs. 32.2%). The uninsured group was younger (43.3 vs. 49.5 years in the employee/public group), less often married (57.1%), and socially intermediate between employee/public and resident groups on several indicators. These differences support the need for multivariable adjustment and conservative causal interpretation.

Main associations with out-of-pocket burden

In the weight-adjusted model, resident insurance was associated with higher log(OOP + 1) than employee/public insurance (beta 0.275, 95% CI 0.188 to 0.362, p < 0.001), corresponding to about 31.7% higher out-of-pocket expenditure. No insurance showed a similar but slightly larger association (beta 0.313, 95% CI 0.177 to 0.449, p < 0.001), corresponding to about 36.8% higher out-of-pocket expenditure. Other or multiple insurance did not differ materially from employee/public insurance (Table 2 and Supplementary Figure S1).

Table 2.

Main weighted regression results

Outcome	Comparison	Estimate	95% CI	P value
Log out-of-pocket expenditure	Uninsured	0.313	0.177 to 0.449	< 0.001
Log out-of-pocket expenditure	Other/multiple	0.045	-0.127 to 0.218	0.608
Log out-of-pocket expenditure	Resident	0.275	0.188 to 0.362	< 0.001
High OOP ratio ≥ 50%	Uninsured	0.177	0.098 to 0.256	< 0.001
High OOP ratio ≥ 50%	Other/multiple	0.008	-0.090 to 0.107	0.867
High OOP ratio ≥ 50%	Resident	0.174	0.127 to 0.221	< 0.001
Poor self-rated health	Uninsured	0.022	0.003 to 0.040	0.023
Poor self-rated health	Other/multiple	0.011	-0.011 to 0.033	0.320
Poor self-rated health	Resident	0.013	-0.000 to 0.026	0.051

Reference group: employee/public insurance. Binary outcomes are modeled using linear probability models

Among medical spenders, resident insurance was associated with a 17.4%-point higher probability of an out-of-pocket ratio of at least 50% than employee/public insurance (95% CI 12.7 to 22.1% points, p < 0.001). No insurance was associated with a 17.7%-point higher probability (95% CI 9.8 to 25.6% points, p < 0.001). In weighted descriptive trends, high out-of-pocket ratio prevalence among spenders remained consistently lower in the employee/public group than in resident or uninsured groups across all four waves (Fig. 2B). In 2022, the weighted prevalence was 46.8% for employee/public insurance, 75.4% for resident insurance, and 78.8% for no insurance.

Associations with poor self-rated health were weaker than those for direct financial burden. Compared with employee/public insurance, the uninsured group had a 2.2%-point higher probability of poor self-rated health (95% CI 0.3 to 4.0% points, p = 0.023), while the resident group showed a borderline association (beta 0.013, 95% CI -0.000 to 0.026, p = 0.051). This pattern suggests that the clearest scheme differences in the present dataset concern financial protection rather than overall self-rated health status.

Exact-scheme analyses

Exact-scheme analyses supported the main interpretation (Table 3). Relative to UEBMI, NRCMS was associated with higher log(OOP + 1) (beta 0.294, 95% CI 0.202 to 0.387, p < 0.001), URBMI was also associated with higher log(OOP + 1) (beta 0.207, 95% CI 0.073 to 0.341, p = 0.002), and no insurance remained strongly associated with higher out-of-pocket burden (beta 0.314, 95% CI 0.175 to 0.452, p < 0.001). For high out-of-pocket ratio among spenders, NRCMS (beta 0.193, 95% CI 0.143 to 0.244, p < 0.001), URRBMI (beta 0.189, 95% CI 0.089 to 0.290, p < 0.001), and no insurance (beta 0.178, 95% CI 0.097 to 0.259, p < 0.001) all showed higher burden than UEBMI.

Table 3.

Exact-scheme sensitivity analyses

Outcome	Comparison vs. UEBMI	Estimate	95% CI	P value
Log out-of-pocket expenditure	Uninsured	0.314	0.175 to 0.452	< 0.001
Log out-of-pocket expenditure	NRCMS	0.294	0.202 to 0.387	< 0.001
Log out-of-pocket expenditure	Public insurance	-0.036	-0.282 to 0.209	0.772
Log out-of-pocket expenditure	URBMI	0.207	0.073 to 0.341	0.002
Log out-of-pocket expenditure	URRBMI	0.121	-0.069 to 0.311	0.212
High OOP ratio ≥ 50%	Uninsured	0.178	0.097 to 0.259	< 0.001
High OOP ratio ≥ 50%	NRCMS	0.193	0.143 to 0.244	< 0.001
High OOP ratio≥ 50%	Public insurance	-0.044	-0.152 to 0.065	0.429
High OOP ratio ≥ 50%	URBMI	0.067	-0.004 to 0.139	0.066
High OOP ratio ≥ 50%	URRBMI	0.189	0.089 to 0.290	< 0.001

Reference group: UEBMI. Exact-scheme analyses are restricted to observations with one clearly identified scheme plus the uninsured group

Additional sensitivity and subgroup analyses

Sensitivity analyses were directionally stable (Supplementary Table S2). Using alternative burden thresholds, resident insurance and no insurance remained associated with higher high-burden risk at both 40% and 60% thresholds. At the 60% threshold, the adjusted risk differences were 0.143 for resident insurance and 0.243 for no insurance relative to employee/public insurance (both p < 0.001). Province fixed-effect models also remained positive for both resident insurance and no insurance in the log(OOP + 1) and high-burden analyses. Continuous out-of-pocket ratio among spenders showed the same direction, with higher ratios for resident insurance (beta 0.102) and no insurance (beta 0.180) relative to employee/public insurance.

Respondent-level clustered standard errors did not materially change the main inference (Supplementary Table S3). In clustered models, resident insurance remained associated with higher log(OOP + 1) (beta 0.275, 95% CI 0.181 to 0.369, p < 0.001) and high OOP ratio (risk difference 0.174, 95% CI 0.126 to 0.221, p < 0.001); the corresponding estimates for no insurance were beta 0.313 (95% CI 0.171 to 0.455, p < 0.001) and risk difference 0.177 (95% CI 0.095 to 0.258, p < 0.001). Income-rank stratified models also supported the main pattern (Supplementary Table S4). Among low-income-rank medical spenders (rank 1–2), resident insurance and no insurance were associated with higher high-OOP-ratio probability than employee/public insurance (risk differences 0.137 and 0.208, respectively). Among income ranks 3–5, the corresponding risk differences were 0.186 and 0.123. Formal interaction terms between low income rank and scheme were not statistically significant for resident insurance or no insurance, suggesting no strong evidence that the scheme association was limited to one income-rank stratum.

Year-stratified models suggested temporal persistence (Supplementary Table S5). Year-by-scheme descriptive outcome estimates are reported in Supplementary Table S6. Resident insurance was positively associated with higher log(OOP + 1) in each wave, with coefficients ranging from 0.178 in 2016 to 0.436 in 2022. For high OOP ratio among spenders, the resident-insurance risk difference relative to employee/public insurance was also positive in each wave, ranging from 0.134 in 2018 to 0.287 in 2022. These year-stratified results support the interpretation that the broad resident-insurance protection gap was not confined to a single survey wave.

One nuance emerged from the spender-only model for log(OOP + 1): after restricting to those with any medical expenditure, absolute out-of-pocket amount differences largely attenuated. That result suggests the main protection gap may operate more through reimbursement generosity and share of cost borne by the individual than through uniformly larger absolute out-of-pocket payments conditional on receiving care.

Exploratory subgroup analyses showed broadly consistent positive associations across men and women and across urban and rural residents (Fig. 3; Supplementary Table S7). The resident and uninsured effects were especially large in rural and younger-than-60 strata. Effect sizes were smaller and less precise in adults aged 60 years or older and in the chronic-disease subgroup, although most point estimates remained in the same adverse direction.

Fig. 3.

Subgroup associations for log out-of-pocket expenditure. Resident insurance and no insurance were compared with employee/public insurance across sex, residence, age, and chronic-disease strata

Discussion

This study provides a consistent repeated cross-sectional picture from CFPS 2016–2022: relative to employee/public insurance, resident insurance and no insurance were both associated with higher individual out-of-pocket burden, and the difference was substantial when financial protection was measured as the probability that out-of-pocket expenditure constituted at least half of total medical expenditure. The key observation is that the dominant resident-insurance group showed worse financial-protection markers than the employee/public benchmark, even though both categories represent insured populations [9, 10, 18].

The exact-scheme results sharpen that interpretation. Compared with UEBMI, NRCMS and URBMI showed higher out-of-pocket burden, while URRBMI showed elevated high-burden risk among spenders. This pattern is consistent with the broader literature suggesting that expansion of insurance coverage in China did not fully eliminate differences in financing generosity and financial protection across schemes [7, 9, 10]. Our results extend that literature by using a harmonized multi-wave adult dataset with direct reimbursement-adjusted spending variables and by showing that the burden gradient remains visible after adjustment for socioeconomic and health-related covariates.

The 2016–2022 study period overlaps with continued implementation of the URRBMI integration reform, which merged URBMI and NRCMS in many provinces. This policy transition is important for interpretation. The main resident-insurance category should be understood as a harmonized resident-oriented coverage group rather than a single stable benefit package with identical generosity in every year and province. To examine temporal heterogeneity, we used exact-scheme models, province fixed-effect models, post-2018 models, descriptive trend figures, and year-stratified associations. These analyses consistently suggested higher burden among resident-oriented schemes, but CFPS does not provide harmonized province-specific reform timing or plan-level reimbursement rules. Therefore, exposure misclassification and unobserved benefit changes over time remain possible.

The time-trend figure adds an important descriptive layer. Resident insurance remained the dominant coverage arrangement throughout the study period, even as its share gradually declined. More importantly, the high-burden gap between employee/public insurance and resident or uninsured groups persisted across all four waves. This makes the main finding harder to dismiss as a single-year artifact.

The spender-only sensitivity model adds a useful mechanistic nuance. Once the analysis was restricted to adults who had any medical expenditure, scheme differences in absolute log out-of-pocket amount were attenuated, but differences in out-of-pocket ratio remained robust. That pattern suggests the main inequity may lie less in the fact of spending itself and more in reimbursement depth, benefit generosity, or the fraction of costs shifted to the individual. This distinction matters for health-financing policy.

Several limitations should be considered. First, the study is observational and repeated cross-sectional, so causal interpretation should remain conservative. Individuals with different insurance schemes differ systematically in age, residence, employment, and other characteristics, and residual confounding is possible despite multivariable adjustment. Although CFPS has a longitudinal panel structure that could in principle support individual fixed-effect models, exploiting within-person insurance transitions would require additional assumptions about the timing and endogeneity of scheme switches, which we leave for future work.

Second, the study could not fully adjust for health-care utilization. Out-of-pocket expenditure reflects not only insurance generosity but also outpatient visits, inpatient admissions, facility level, disease severity, and possible forgone or delayed care. CFPS provides some health and expenditure information, but the harmonized adult file used here did not contain consistently comparable outpatient and inpatient visit counts, service type, or disease-severity measures across all four waves. The spender-only model partly addressed utilization selection by conditioning on any medical expenditure, but it could not distinguish lower OOP caused by better protection from lower OOP caused by reduced or forgone utilization. This issue is especially relevant if resident-insured adults reduce care because of lower expected reimbursement or higher anticipated costs.

Third, the main models use complete-case analysis rather than multiple imputation. Missingness was substantial for some variables, including out-of-pocket expenditure (41.8%), education (15.5%), and income rank (17.5%; see Supplementary Table S1). Some expenditure missingness is likely attributable to skip patterns in which detailed medical expenditure items are asked only of respondents reporting health-care use, but item non-response and non-random missingness cannot be ruled out. If those with missing expenditure differ systematically by scheme, health status, or income, the estimates could be biased. Future analyses using multiple imputation, inverse-probability weighting for missingness, or a two-part utilization-and-spending framework would be useful.

Fourth, the high out-of-pocket ratio thresholds are analytic markers of severe individual burden rather than formal household catastrophic expenditure definitions as used by the WHO/World Bank [2, 3]. They capture the reimbursement-share dimension directly but do not account for household capacity to pay. Income-rank stratified sensitivity analyses were added in response to this limitation, but ordinal income rank cannot replace a household consumption or non-food-expenditure denominator. Fifth, insurance categories were harmonized across waves, which necessarily compresses some scheme-specific detail in the main analysis. We addressed that issue with exact-scheme and province fixed-effect sensitivity models, but some administrative and benefit nuances remain unobserved in CFPS.

Despite these limitations, the study has clear strengths. It uses a large nationally based adult dataset, directly measures reimbursement-adjusted out-of-pocket expenditure, compares multiple insurance arrangements rather than insured versus uninsured only, and shows a coherent pattern across main, exact-scheme, trend, threshold, clustered-standard-error, income-rank, year-stratified, and subgroup analyses. These features support an association-focused contribution to the health services, public health, and health equity literature.

The policy implications of these findings are specific but should be interpreted cautiously. The persistent protection gap under resident-oriented schemes points to several potential levers: improving reimbursement depth, strengthening outpatient coverage, reducing cost sharing that disproportionately affects lower-income enrollees, and narrowing the benefit-design gap between UEBMI and resident-oriented schemes as integration reforms continue. More research is needed to determine how underreporting, forgone care, plan-level reimbursement rules, and benefit design drive the observed OOP differences.

Conclusions

Using CFPS 2016–2022 adult data, we found that resident insurance and no insurance were associated with higher out-of-pocket medical burden than employee/public insurance. Exact-scheme, trend, threshold, province fixed-effect, clustered-standard-error, income-rank, and year-stratified analyses supported this association pattern. The results point to persistent financial-protection differences across Chinese insurance arrangements and support further research into the roles of benefit design, utilization, underreporting, and household capacity to pay.

Abbreviations

Abbreviation

Full term

CFPS

China Family Panel Studies

CHE

Catastrophic health expenditure

CI

Confidence interval

NRCMS

New Rural Cooperative Medical System

OOP

Out-of-pocket

UEBMI

Urban Employee Basic Medical Insurance

URBMI

Urban Resident Basic Medical Insurance

URRBMI

Basic Medical Insurance for Urban and Rural Residents

WHO

World Health Organization

Authors' contributions

XR conceived and designed the study, performed the statistical analyses, and drafted the manuscript. YH contributed to data curation and statistical analysis. SM assisted with data interpretation and manuscript preparation. CJ supervised the study and critically revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The CFPS datasets analyzed during the current study are available from the Institute of Social Science Survey, Peking University, through the official CFPS data platform (https://www.isss.pku.edu.cn/cfps/) subject to a data use agreement. Derived analytic variables and statistical code are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study analyzed de-identified public-use data from the China Family Panel Studies (CFPS). CFPS data collection was approved by the Biomedical Ethics Review Committee of Peking University (IRB00001052-14010). As the present study used pre-existing, fully anonymized secondary data without any intervention on human participants, individual informed consent was not required, and no additional ethics review was necessary under applicable national guidelines. The study was conducted in accordance with the ethical principles outlined in the 2013 Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.