Can supplementary private health insurance reduce poverty vulnerability? Evidence from Shandong Province

Abstract

Background

Despite achieving near-universal social health insurance coverage, some Chinese continue to grapple with issues of poverty vulnerability. Supplementary private health insurance (SPHI) serves as a crucial complement to the social health insurance system. This study aimed to investigate its efficacy in mitigating poverty vulnerability.

Methods

Cross-sectional data of 18,426 representative samples were obtained from the Sixth National Health Service Survey (Shandong) conducted in 2018. A three-stage feasible generalised least squares estimation procedure was employed to estimate poverty vulnerability. Additionally, we explored the impact of SPHI on poverty vulnerability using the propensity score matching (PSM) method to balance treatment and control groups along observable dimensions. To address the potential endogeneity issues, we used the instrumental variable (IV) estimation approach to determine how SPHI affects poverty vulnerability.

Results

We found that SPHI reduced the probability of poverty vulnerability for individuals enrolled in social health insurance. Furthermore, SPHI had a more pronounced protective effect on respondents with chronic diseases, those aged over 60 years, and those living in urban areas or western Shandong.

Conclusions

The results imply that supplementary SPHI can be effective in poverty reduction even among populations with basic health insurance. Therefore, we encourage governments in other low- and middle-income countries to consider implementing supplementary SPHI for vulnerable people to reduce medical impoverishment.

Background

The 2030 Agenda for Sustainable Development, adopted by 193 member states in 2015, sets the eradication of poverty in all its forms as its primary goal [1]. One of its sub-objectives is to enhance the resilience of the poor and vulnerable populations and reduce their exposure and vulnerability [1]. Health risk is the most important causative factor of being poor and returning to poverty [2]. Globally, around 100 million people are trapped in health poverty, a type of poverty caused by insufficient affordability and excessive demand for medical services [3]. The worldwide incidence of catastrophic health expenditure (CHE) is approximately 2.14%, and the incidence of poverty is approximately 1.43% [4]. Health-related poverty in China reached approximately 40% of at-risk households face poverty due to medical shocks [5].

Health insurance, as a financial sharing mechanism for medical expenses, plays a crucial role in alleviating illness-related poverty and ensuring timely access to healthcare for individuals who lack the financial means [6]. It has brought a significantly positive impact on individuals and families affected by health poverty. In the short term, health insurance provides financial coverage for medical expenses when an individual falls ill, effectively reducing the occurrence of CHE [7–9]. In the long run, it improves individuals’ health status by enhancing the accessibility of healthcare services. This, in turn, increases labour supply, reduces informal lending for health services, decreases expenditures on precautionary savings, and improves the overall structure of household consumption, ultimately aiming to alleviate poverty [10–12]. Thus, the establishment of a robust insurance system is essential in reducing health poverty, ensuring high-quality medical services and enhancing overall health outcomes [13, 14].

In some low- and middle-income countries, governments have focused on implementing public health insurance programs [15, 16]. However, the limited reimbursement rates and scope have led to health insurance not being very effective in defusing the economic risks of illness or protecting residents’ property [17, 18]. China, as a representative case, has established the world’s largest medical security system, with over 95% of the population covered by social health insurance (SHI) [19, 20]. However, due to the limited funding pool for SHI, the extent of coverage is constrained [21, 22], resulting in a high out-of-pocket ratio for medical expenses and an increasing incidence of CHE over time [11, 23]. In 2022, the personal health expenditure accounted for 27.0% of the total health expenditure in China [24], much higher than the proportion of poverty control expenditure proposed by the World Health Organization of 15–20% [25].

In this context, Supplementary private health insurance (SPHI), as a complement to social insurance, has attracted the attention of researchers [26, 27]. SPHI operates on a voluntary basis and provides coverage for expenses that fall outside the scope of reimbursement by SHI, as well as for high-quality services [28]. SPHI plays a crucial role in bridging the gaps in SHI coverage, mitigating the burden of high medical costs on individuals, improving residents’ ability to invest in health, and meeting their higher healthcare needs [29, 30]. China’s SPHI market has experienced rapid growth, driven by policy support and rising healthcare demands. Key products include critical illness insurance, high-deductible “million-yuan” health insurance, and city-customized insurance (low-cost supplemental coverage). Healthy China 2030 initiative position SPHI as a critical supplement to public health insurance [31]. Aligning private health insurance with ongoing public health insurance reforms could enhance multi-level health care resilience, particularly for high-cost treatments and underserved populations [27, 32]. Studies have confirmed that enrolment in SPHI significantly reduces personal health expenditures [33, 34]. Scholars have also found that supplemental insurance contributes to the efficient utilisation of healthcare services [35, 36]. However, there is limited evidence regarding whether SPHI also contributes to poverty prevention among individuals who are already covered by SHI. Recent years have witnessed growing scholarly attention toward the poverty-mitigating potential of health insurance, particularly in Southeast Asian countries like Vietnam [37] and broader empirical studies [38]. However, these studies primarily focus on specific regions or contexts that offer limited relevance to China’s unique socioeconomic characteristics and the SPHI - a crucial component within the China’s multi-level social health insurance scheme.

Therefore, this study aims to investigate the impact of SPHI on poverty prevention among individuals enrolled in SHI. Since China has already eliminated the absolute poverty, we adopted a more forward-looking measure—poverty vulnerability—which captures the dynamic risk of falling into poverty, as opposed to static poverty status [39]. While previous studies have confirmed the role of health insurance in reducing catastrophic health expenditures [40] and preventing illness-induced poverty [41], this study shifts the focus to the likelihood of future poverty from a vulnerability perspective, which remains underexplored. To address potential self-selection bias in SPHI enrolment, we employed the propensity score matching (PSM) method [42]. Since adverse selection by purchasers and cream skimming by insurers may introduce reverse causality between health status and SPHI participation [43, 44], we further adopted an instrumental variable (IV) approach to correct for endogeneity. Additionally, we explored heterogeneity in the protective effects of SPHI across different population subgroups, with the aim of identifying key target groups and providing empirical evidence to inform policy refinement.

Methods

Participants

Shandong, located in East China, is a coastal province with a population of more than 100 million. The urban-rural population ratio of Shandong Province was approximately 1.706:1, and per capita gross regional product was 72,151 Yuan in 2020 [45]. Furthermore, there were significant disparities between different regions in eastern and western and between urban and rural areas. Shandong Province features diverse landforms, including coastal, plain, and mountainous regions. In 2023, its per capita gross domestic product was 90,770 Yuan, slightly surpassing the national average of 89,358 yuan [46]. Given that Shandong’s demographic and economic indicators closely mirror those of China as a whole, the province was considered a highly representative sample [47, 48]. Additionally, both the province’s aging-related metrics and medical resource distribution align with national benchmarks [49]. Therefore, the sample from Shandong Province was chosen to ensure comprehensive representation.

The data for this study was obtained from the Sixth Health Service Survey of Shandong Province in 2018. This is a large survey of health services organised by the National Health Commission and conducted every 5 years. The questionnaire used in this study was developed by National Health Commission. The survey used a multistage stratified cluster sampling method, covering 17 cities in Shandong Province. A total of 35,265 questionnaires were collected through face-to-face interviews, including demographic and socioeconomic characteristics, health service needs, utilisation, and healthcare security.

Our study focuses on the influence of SPHI on poverty vulnerability among residents participating in SHI. Therefore, residents with Shi were included in the study. In addition, people aged 15 years and younger was not involved in some study-related information during the interview process; therefore, the study excluded them and removed the sample with missing information. The final sample included in this study was 18,426 individuals, as shown in Fig. 1.

Fig. 1.

Flow chart of the participants in this study

Measures

Poverty vulnerability

This study focused on the return of individuals and families lifted out of poverty. It was more concerned with the dynamics of poverty than that of the current state. We included the Vulnerability to Expected Poverty (VEP) metric as an explained variable. VEP is a validated measure that predicts long-term poverty [50–52]. By considering different assets owned by the individual or household, such as financial, human, social resources, this measure aims to gauge their overall poverty vulnerability. It takes into account the potential impact of various risks, such as disease, unemployment, natural disasters, or economic downturns [2, 50].

Poverty vulnerability is shaped by various uncertainties, and the Sustainable Livelihoods Approach (SLA) framework is employed to identify both external and internal factors influencing household socioeconomic survival [53]. The core component of the SLA framework is livelihood capital, which is divided into five dimensions: natural capital, physical capital, financial capital, social capital, and human capital. Previous studies have shown that the absence of certain livelihood capital indicators may exacerbate poverty vulnerability [54]. Based on the SLA framework and the survey questionnaire, this study measures livelihood capital across the following dimensions: physical capital, financial capital, social capital, human capital, and health behaviours. Among these, physical capital is assessed through indicators such as the type of drinking water, toilet facilities, distance to the nearest medical institution, type of medical institution, and health records. Social capital is measured by household registration type, place of registration, and number of permanent residents. Financial capital is evaluated through annual per capita household expenditure, registered poor household, and whether the household benefits from minimum living security. Human capital is captured by factors such as gender, age, marital status, education, employment status, the EuroQoL-5 Dimensions (EQ-5D), body mass index (BMI), and presence of chronic disease. Health behaviours are measured through indicators of smoking, alcohol consumption, exercise, and brushing habits. Table 1 presents the index settings and definition of poverty vulnerability prediction.

Table 1.

The definition of main variables

Variable	Definition
Physical capital
Types of drinking water	Running water = 0; Others = 1
Types of toilets	Sanitary latrines = 0; Non-sanitary latrines = 1
Distance to the nearest medical institution	Less than 1 km = 0; One kilometer = 1; Two kilometers = 2; Three kilometers = 3; Four kilometers = 4; Five kilometers and above = 5
Types of medical institution	Primary medical institutions and below = 0; Secondary medical institutions = 1; Tertiary medical institutions = 2; Others = 3
Health records	Yes = 0; No = 1
Social capital
Household registration type	Agriculture = 0; Non-agriculture = 1
Place of registration	This county/district = 0; Other counties/districts = 1
Number of permanent residents	The number of permanent residents in the family (person)
Financial capital
Annual per capita household expenditure	Yuan
Registered poor household	Whether it is poor household: Yes = 0; No = 1
Minimum living security	Whether it is Dibao (including Wubao) household: Yes = 0; No = 1
Human capital
The gender of the interviewee	Male = 0; Female = 1
The age of the interviewee	16–59 years old = 0; ≥60 years old = 1
Marital status	Married = 0; Unmarried = 1
Education of the interviewee	No formal education = 0; Primary school = 1; Middle school = 2; High school and above = 3
Employment status	Employment = 0; Unemployment = 1
EQ-5D	The score of EQ-5D (scores)
BMI	The body mass index
Chronic disease	Whether having chronic diseases: Yes = 0; No = 1
Health behaviors
Smoking	Whether you are currently smoke: Yes = 0; No = 1
Alcohol consumption	Whether you consumed alcohol in the past 12 months: Yes = 0; No = 1
Exercise	Whether you exercise every week: Yes = 0; No = 1
Brushing habits	Whether you brush teeth every day: Yes = 0; No = 1

Based on the estimation method of Chaudhuri [55], we predicted future income or assets falling below a predefined poverty line, subject to household capital shocks and the risk of high costs. The VEP was performed using the following equation:

1

Where was the probability of poverty vulnerability, was the vulnerability of family in the period t + j, and poverty line is predefined poverty line.

To estimate the income equation, we introduced a three-stage feasible generalised least squares (FGLS) measure of VEP and assumed that the logarithm of future personal income was normally distributed.

First, the ordinary least squares (OLS) regression was estimated by taking the logarithm of annual per capita household income, using the following equation:

2

3

Where was the calculated total sum of income (), and then divided by the number of family members. referred to physical capital, financial capital, social capital, human capital, and health behaviours. Regression estimation was then performed by substituting the squared residual in Eq. (4) into Eq. (5), and the estimation of the predictor variable () and the residual term σ were obtained after processing.

Second, FGLS estimation was performed using the fitted values obtained in the first step to obtain the estimator and , and substituting them into Eqs. (4) and (5), the expectation and variance of the logarithm of income were obtained.

4

5

Third, select the poverty line (z), calculate the poverty vulnerability of household, and the VEP could be obtained by Eq. 6.

6

In the estimation, the poverty line (z) used the 2018 poverty standard of Shandong Province: an annual per capita income of 3609 yuan. In this study, we drew on Pritchett’s study and considered the highly vulnerable group whose predicted Future annual per capita household income to be below the poverty line with more than 50% probability as a poverty vulnerability status [56, 57]. Poverty vulnerability had two values (yes = 1, no = 0).

Supplementary private health insurance

Of the research objectives, we included data from respondents covered by the SHI, each supplemented by SPHI. The core explanatory variable of our study was the purchase of SPHI, with 1 representing a purchase and 0 representing no purchase.

Empirical strategy

Matching variables

Individual characteristics, household characteristics, enabling resources, health status, health behaviours, and living standard were selected as matching variables [58]. The individual characteristics were gender (male = 0 and female = 1), age (16–59 years old = 0 and ≥ 60 years old = 1), marital status (married = 0 and unmarried = 1; unmarried included separated, divorced, widowed, and never married), and education level (no formal education = 1, primary school = 2, middle school = 3, high school and above = 4). The enabling resources were residence (rural = 0 and urban = 1), residence region (eastern Shandong = 1, central Shandong = 2, and western Shandong = 3), and employment status (employed = 0 and unemployed = 1). Health status included chronic disease status (no = 0, yes = 1) and EQ5D (continuous variable). Health behaviours were smoking (yes = 0, no = 1), alcohol consumption (yes = 0, no = 1), health check-up (yes = 0, no = 1). Living standard included type of drinking water (running water = 0 and others = 1) and sanitary latrines (sanitary latrines = 0 and non-sanitary latrines = 1).

Propensity score matching method

As the SPHI market is free, it is a “self-selection” process for individuals to purchase products. The decision to purchase was influenced by various factors, including individual and family characteristics, social environment, and their relationship with poverty vulnerability. PSM was conducted to solve the endogeneity problem caused by self-selection. This study used a logistic model to estimate the probability of participating in SPHI.

7

D indicates the group or intervention factor, with D = 1 indicates the observation group and D = 0 indicates the control group. They are then matched according to propensity scores, pairing those ‘purchasers’ with ‘non-purchasers’ whose main characteristics (matching variables) are relatively similar. Using the matched sample, estimates of the average treatment effect (ATT) of participation on the treated in SPHI can be obtained to assess the poverty prevention effect.

8

represented the probability of individual poverty vulnerability being enrolled in SPHI, while represented the probability of individual poverty vulnerability not being enrolled in SPHI. Because is not directly observable, a counterfactual framework must be constructed. in ATT is the counterfactual effect.

In addition, we divided the samples into five subgroups according to chronic disease conditions, age level, education level, area of residence, and urban and rural environments. We also used four PSM methods to analyse the between-group differences in the effects of SPHI on subsamples of different populations and areas and to explore the heterogeneity of the effects of SPHI on poverty reduction.

Instrumental variable estimates

Although we believed that the richness of our data sets makes the PSM assumptions plausible, we complemented our analysis of the SPHI effect on poverty vulnerability by adopting a two-stage least squares (2SLS). We used SPHI density in the prefecture-level city where the sample population resides as instrumental variable. SPHI density is defined as the proportion of individuals enrolled in SPHI out of the total population. It is primarily associated with enrolment in SPHI and does not have a direct relationship with poverty vulnerability.

The 2SLS model was implemented to further determine the corrective effect of IV on endogeneity. In the first stage, the endogenous variable SPHI was used to regress the IV, namely, , and obtain the fitted values . In the second stage, the was regressed on the explained variable , that was, . The premise of using IV was the existence of endogenous explanatory variable. In addition, the correlation and validity of IV were tested.

Test for the endogeneity of SPHI and IV

The rationale for employing IV was the presence of endogenous explanatory variables. First, we conducted the Hausman test, which rejected the null hypothesis that “all explanatory variables are exogenous” at the 5% significance level. This result confirmed that the density of SPHI should be treated as an endogenous variable, thus justifying the use of the IV method.

Additionally, we assessed the correlation and validity of the IV: (1) Correlation: The IV must be correlated with the endogenous variables and must exert adequate explanatory power over them; (2) Validity: The IV should be uncorrelated with the error terms and should influence the dependent variable solely through the endogenous variables. To establish correlation, we examined the first-stage regression results from the 2SLS regression (p < 0.05). Another condition for a valid instrument is its exogeneity. That is, SPHI density must not be correlated with unobserved determinants of poverty vulnerability. According to the related literature, regional insurance density is independent of other factors that would influence individual poverty [59]. SPHI density reflects the prevalence and accessibility of social health insurance in a specific region, which directly influences individuals’ decisions to participate in SPHI [60, 61]. Furthermore, a weak IV test was conducted with a significance threshold of 1%, confirming that the selected IV possessed sufficient explanatory power for the endogenous variables. Finally, we applied the under-identification test to reject the null hypothesis of under-identification of the IV [62].

Results

Demographics and characteristics of participants

As shown in Table 2, the sample consisted of 18,426 individuals, among which only 2305 (12.51%) were covered by SPHI. Among individuals who were vulnerable to poverty, 95.78% did not participate in SPHI, while only 4.22% did choose to participate. In contrast, among individuals who were not vulnerable to poverty, 85.77% did not participate in SPHI, while 14.23% did participate. These results suggest that participation in SPHI serves as a protective factor against poverty vulnerability.

Table 2.

Description statistics of poverty vulnerability among participants (N = 18426)

Variables	Total sample	Vulnerable to poverty (n = 3173)	Non-Vulnerable to poverty (n = 15253)
SPHI
Yes	2305 (12.51%)	134 (4.22%)	2171 (14.23%)
No	16,121 (87.49%)	3039 (95.78%)	13,082 (85.77%)
Age
16–59 years old	12,072 (65.52%)	936 (29.50%)	11,136 (73.01%)
≥ 60 years old	6354 (34.48%)	2237 (70.50%)	4117 (26.99%)
Gender
Male	8477 (46.01%)	1433 (45.16%)	7044 (46.18%)
Female	9949 (53.99%)	1740 (54.84%)	8209 (53.82%)
Education level
No formal education	3138 (17.03%)	1250 (39.39%)	1888 (12.38%)
Primary school	4426 (24.02%)	941 (29.66%)	3485 (22.85%)
Middle school	7128 (38.68%)	714 (22.50%)	6414 (42.05%)
High school and above	3734 (20.27%)	268 (7.45%)	3466 (22.72%)
Marital status
Married	15,641 (84.89%)	2501 (78.82%)	13,140 (86.15%)
Unmarried	2785 (15.11%)	672 (21.18%)	2113 (13.85%)
Employment status
Employment	10,993 (59.66%)	1167 (36.78%)	9826 (64.42%)
Unemployment	7433 (40.34%)	2006 (63.22%)	5427 (35.58%)
Chronic disease
Yes	5728 (31.09%)	1414 (44.56%)	4314 (28.28%)
No	12,698 (68.91%)	1759 (55.44%)	10,939 (71.72%)
EQ5D	0.96 (0.08)	0.93 (0.11)	0.97 (0.07)
Smoking
Yes	4560 (24.75%)	782 (24.65%)	3778 (24.77%)
No	13,866 (75.25%)	2391 (75.35%)	11,475 (75.23%)
Alcohol consumption
Yes	5243 (28.45%)	784 (24.71%)	4459 (29.23%)
No	13,183 (71.55%)	2389 (75.29%)	10,794 (70.77%)
Health check-up
Yes	9073 (49.24%)	1996 (62.91%)	7077 (46.40%)
No	9353 (50.76%)	1177 (37.09%)	8176 (53.60%)
Residence
Rural	14,537 (78.90%)	2548 (80.30%)	11,989 (78.60%)
Urban	3889 (21.10%)	625 (19.70%)	3264 (21.40%)
Residence region
Eastern Shandong	7656 (41.55%)	1457 (45.92%)	6199 (40.64%)
Central Shandong	7069 (38.36%)	1018 (32.08%)	6051 (39.67%)
Western Shandong	3701 (20.09%)	698 (22.00%)	3003 (19.69%)
Type of drinking water
Running water	14,299 (77.60%)	2272 (71.60%)	12,027 (78.85%)
Others	4127 (22.40%)	901 (28.40%)	3226 (21.15%)
Sanitary latrines
Sanitary latrines	14,986 (81.33%)	2165 (68.23%)	12,821 (84.06%)
Non-sanitary latrines	3440 (18.67%)	1008 (31.77%)	2432 (15.94%)
Household size	2.91 (1.30)	2.71 (1.46)	2.96 (1.26)

We use mean and SD for continuous variables and frequency (n) and percentage for discrete variables; EQ5D: EuroQol-5 dimension; SPHI Supplementary private health insurance

In addition, the average age was 50.82 years, and the average score of EQ5D was 0.96. People who are vulnerable to poverty are typically older, have lower levels of education, reside in rural areas, face higher unemployment rates, experience chronic illnesses, have lower EQ5D scores without running water, without sanitary latrines, compared to those who are not vulnerable to poverty.

Matching and balanced test

To ensure the robustness of the PSM results, the nearest neighbour matching method (K = 1 and K = 4), calliper matching (calliper = 0.01), and kernel Matching were used. Although different Matching methods were used, the bias of the overall covariates was lower than 5% and no longer significant between the two groups. In addition, pseudo R² decreased from 0.079 before Matching to 0.000–0.003 after matching, and LR CHI² decreased from 1100.68 to 1.34–19.81, as shown in Table 3. The results indicated that comparability between the intervention and control groups was improved using PSM.

Table 3.

Results of balance tests of explanatory variables before and after PSM

Matching method	P-value	Pseudo R2	LR Chi2	Mean-bias	Med-bias
Unmatched	0.000	0.079	1100.68	15.1	12.6
K-nearest matching (K = 1)	0.940	0.003	19.81	2.1	1.4
K-nearest matching (K = 4)	1.000	0.001	7.25	1.1	0.9
Caliper matching (caliper = 0.01)	1.000	0.000	1.34	0.5	0.4
Kernel matching	1.000	0.001	7.33	1.6	1.2

Impact of supplementary private health insurance on poverty vulnerability

Table 4 shows the PSM results of the effect estimates for the total sample and subgroups using the four PSM methods. The ATT values obtained for the four matching methods were − 0.028 (K = 1), −0.024 (K = 4), −0.026, and − 0.035 (all at the 1% significance level), indicating that participation in SPHI reduces poverty vulnerability by 2.4–3.5%.

Table 4.

Impact of SPHI on poverty vulnerability using PSM method

		K-nearest matching (K = 1)			K-nearest matching (K = 4)			Caliper matching (caliper = 0.01)			Kernel matching
		ATT	SE	T-stat	ATT	SE	T-stat	ATT	SE	T-stat	ATT	SE	T-stat
Full-sample		−0.028***	0.009	−3.19	−0.024***	0.007	−3.63	−0.026***	0.006	−4.08	−0.035***	0.006	−5.64
Chronic disease
Having chronic disease		−0.046**	0.021	−2.15	−0.047***	0.016	−2.87	−0.051***	0.015	−3.40	−0.069***	0.015	−4.65
Non-chronic disease		−0.019**	0.009	−2.14	−0.022***	0.007	−3.10	−0.018***	0.007	−2.75	−0.023***	0.007	−3.56
Age
16–59 years old		−0.029***	0.007	−3.81	−0.021***	0.006	−3.60	−0.021***	0.005	−4.13	−0.023***	0.005	−4.58
≥ 60 years old		−0.072**	0.035	−2.03	−0.065**	0.027	−2.40	−0.077***	0.024	−3.15	−0.087***	0.024	−3.62
Education level
Middle school and below		−0.025**	0.010	−2.36	−0.031***	0.009	−3.57	−0.027***	0.008	−3.43	−0.040***	0.008	−5.24
High school and above		−0.020*	0.018	−1.73	−0.015*	0.009	−1.72	−0.019**	0.009	−2.25	−0.020**	0.008	−2.38
Residence
Urban		−0.031***	0.009	−3.14	−0.023***	0.007	−3.17	−0.028***	0.007	−4.00	−0.039***	0.007	−5.62
Rural		−0.031*	0.017	−1.78	−0.035**	0.014	−2.50	−0.026**	0.013	−2.00	−0.029**	0.013	−2.27
Residence region
Eastern		−0.022	0.015	−1.47	−0.017	0.011	−1.48	−0.018	0.011	−1.63	−0.026**	0.010	−2.42
Central		−0.030***	0.011	−2.71	−0.023***	0.008	−2.76	−0.020**	0.008	−2.33	−0.027***	0.008	−3.21
Western		−0.067***	0.021	−3.13	−0.049***	0.016	−3.06	−0.061***	0.014	−4.17	−0.068***	0.014	−4.77

* P < 0.05, ** P < 0.01, *** P < 0.001; SPHI Supplementary Private health insurance

Table 4 summarizes the impact of SPHI on poverty vulnerability of residents with different individual characteristics and regions. The results using different PSM showed that SPHI reduced poverty vulnerability by 4.6–6.9% among those with chronic diseases, which was significantly higher than that of those without chronic diseases. It reduced by 6.5–8.7% for older adults, significantly higher than that of the young and middle-aged group. It reduced by 2.5–4.0% for the group middle school and below, significantly higher than the group of high school and above. In terms of residence region, SPHI reduced poverty vulnerability of residents in the western region by 4.9–6.8%, which was significantly higher than reductions in the eastern and central regions. It also reduced poverty vulnerability of urban residents by 2.3–3.9%, which was significantly higher than reductions of rural residents.

Regression results using instrumental variable approach

We identified possible IV based on theory and existing literature. First, the endogeneity test was carried out to examine whether all explanatory variables are exogenous. The test results rejected the null hypothesis, indicating that endogenous variables exist and IV estimate is appropriate. Second, the weak IV test was carried out. According to Staiger and Stock, the F statistic of this study is 15.170, which is higher than 11.59 [63, 64]. The null hypothesis of the weak IV can be rejected. Third, the null hypothesis of under-identification of IV is rejected by the under-identification test [62]. Summary of tests for choice of the model, endogeneity of the SPHI, as shown in Table 5.

Table 5.

Summary of tests for choice of the model, endogeneity of the SPHI

Test	Explanation	Results	Comment
Endogeneity test: Hausman test	H0: All explanatory variables are exogenous. Rejection of H0 indicates that endogenous variables exist and IV estimate is appropriate.	=5.20, P-value = 0.023	SPHI is endogenous, IV estimate is appropriate.
Weak IV test	H0: weakly identified. Rejection of H0 & t > 11.59 indicates that the model is not weak.	F statistic = 15.170	IV is not weak.
Under-identification test	H0: IV is under-identification	Statistic value = 15.174 P-value = 0.000	IV is not under-identification.

After kernel Matching and accounting for possible endogeneity, the 2SLS analysis confirmed that SPHI had a statistically significant negative impact on poverty vulnerability (coefficient: −0.663, P < 0.05). This finding is consistent with the results obtained from the PSM analysis, as shown in Table 6.

Table 6.

Results of instrumental variable Estimation after kernel matching (N = 18304)

Variable	OLS	2SLS (First stage)	2SLS (Second stage)
SPHI (ref: no)		SPHI density
Yes	−0.048*** (0.006)	6.206*** (1.593)	−0.663** (0.312)
Age (ref: 16–59 years old)
≥ 60 years old	0.173*** (0.008)	−0.076*** (0.007)	0.125*** (0.026)
Gender (ref: male)
Female	−0.055*** (0.007)	0.011 (0.007)	−0.049*** (0.009)
Education level (ref: no formal education)
Primary school	−0.120*** (0.011)	0.005 (0.008)	−0.118*** (0.010)
Middle school	−0.156*** (0.011)	0.036*** (0.008)	−0.135*** (0.015)
High school and above	−0.192*** (0.011)	0.076*** (0.010)	−0.146*** (0.026)
Marital status (ref: others)
Married	0.041*** (0.009)	−0.007 (0.007)	0.037*** (0.009)
Employment status (ref: employment)
Unemployment	0.071*** (0.006)	−0.037*** (0.006)	0.048*** (0.013)
Chronic disease (ref: no)
Yes	0.025*** (0.007)	0.004 (0.006)	0.028*** (0.007)
EQ5D	−0.281*** (0.049)	−0.006 (0.035)	−0.287*** (0.043)
Smoking (ref: yes)
No	0.005 (0.008)	0.018** (0.007)	0.016 (0.010)
Alcohol consumption (ref: yes)
No	0.014* (0.008)	−0.045*** (0.005)	−0.013 (0.016)
Health check-up (ref: yes)
No	−0.010** (0.005)	−0.024*** (0.005)	−0.025** (0.010)
Residence (ref: rural)
Urban	0.018*** (0.006)	0.020*** (0.006)	0.033*** (0.011)
Residence region (ref: eastern)
Central	−0.038*** (0.006)	0.028*** (0.005)	−0.022** (0.008)
Western	−0.012 (0.007)	0.050*** (0.007)	0.020 (0.018)
Type of drinking water (ref: running water)
Others	0.029*** (0.007)	−0.007 (0.006)	0.024*** (0.008)
Sanitary latrines (ref: sanitary latrines)
Non-sanitary latrines	0.095*** (0.008)	−0.024*** (0.007)	0.079*** (0.012)
Household size	0.019*** (0.002)	−0.002 (0.002)	0.018*** (0.002)

* P < 0.05, ** P < 0.01, *** P < 0.001; SPHI Supplementary Private health insurance; Standard errors in parentheses

Discussion

Using quasi-experimental methods, we were able to eliminate the endogenous interference of insurance adverse selection [65]. Our findings indicate that SPHI can significantly reduce poverty vulnerability by 2.5–4.0%. The policy significance of reducing extends beyond income increases, generating notable spillover effects [66]. For instance, children from families lifted out of poverty are more likely to access education, while family members are more inclined to invest in health (e.g., regular health check-ups and chronic disease management) [67]. These improvements contribute to the enhancement of human capital, establishing a long-term positive cycle of “poverty reduction - livelihood capital growth - income increase” [66, 68]. With nearly universal coverage under SHI, SPHI is expected to enhance access to health services and play a crucial role in protecting family assets and livelihood capital. However, due to health insurance selection bias and affordability constraints [43, 44], the 10.01% higher SPHI uptake among non-vulnerable groups compared to poverty-vulnerable populations. In high-income countries like the United States, South Korea, and Chile, studies have confirmed SPHI can reduce the probability of CHE and debt risk for both general population and those suffering from diseases [69, 70]. In addition, in low- and middle-income countries like China and Georgia, it has been also confirmed that the SPHI has a positive effect on income growth [71, 72]. Some existing studies have proved that the relationship between SPHI and SHI is not a complete competition and substitution. In contrast, coordinated development can improve the consumers’ utility, and the complementary effect is more obvious [73, 74].

We found that the ownership rate of SPHI among the non-vulnerable population was 10.01% higher than that among the poverty-vulnerable population. Income, education, and health status are primary contributors to the inequality in SPHI coverage [75]. As demonstrated by the results of this study, higher levels of education are associated with a lower likelihood of experiencing poverty vulnerability. A considerable body of literature confirmed that individuals with average or poor health status were less likely to purchase insurance [76, 77]. This can be attributed to the fact that health status reflected individuals’ risk preferences to some extent. That was, individuals who are motivated by the need to mitigate the health and economic risks associated with illness are more likely to adopt a healthy lifestyle, thereby maintaining better physical health [78]. There are several potential mechanisms through which SPHI reduces poverty vulnerability. First, SPHI complements SHI benefits by offering lower treatment thresholds, higher cost-sharing, and a broader range of services, which helps mitigate the occurrence of CHE [79, 80]. Second, SPHI enhances access to high-quality healthcare, enabling timely interventions that restore and preserve human capital, thereby reducing the risk of future poverty [81]. Third, SPHI increases access to risk-reducing services and promotes the utilisation of preventive care, improving health outcomes and reducing overall healthcare costs [35, 82]. For instance, as the study results demonstrated, SPHI is associated with an increased likelihood of residents engaging in health check-ups and a decreased likelihood of smoking and alcohol consumption. Additionally, this paper posits an underexplored mechanism. SPHI’s financial protection against disease risks has enhanced the living conditions of insured households (e.g., access to running water and sanitary latrines), thereby improving their standard of living and reducing their vulnerability to poverty.

In addition, this study also found that SPHI is more effective for patients with chronic diseases and older adults. In a report published by the World Bank, Disease Control Priorities: Improving Health and Reducing Poverty, scholars found that the cost of spending on treatment and care for chronic diseases present long-term persistent characteristics [83]. Additionally, some scholars have found that SHI covers inpatient medical expenses in China. But it is not effective enough to reduce financial risks and ensure equity in health services for patients with chronic diseases who take medication for a long time [84, 85]. Furthermore, participation in SPHI reduced poverty vulnerability for those aged over 60 years, significantly higher than that for those under 60 years. This result is consistent with previous studies showing that health insurance has a more significant effect on property protection for older adults [86]. They tend to face a greater possibility of illness and a heavier financial burden owing to their physiological characteristics [86–88]. Therefore, the comprehensive use of SPHI and SHI can be a new benefit package for older adults. This can be a way to prevent poverty among the older adult population due to illness.

The levels of economic development and medical resources in the eastern and central regions of Shandong Province were better than those in the west [85]. Participation in SPHI reduced the poverty vulnerability of residents in western Shandong, significantly higher than that of those in the eastern and central regions. This shows that SPHI is more effective in supplementing poverty prevention in less developed regions with low levels of protection. Interestingly, the urban-rural disparities in the effects of SPHI contradict the findings of previous studies. While other research has indicated that, despite near-universal SHI coverage, rural residents often face a higher proportion of out-of-pocket health costs and a greater likelihood of CHE [89, 90], the present study contradicts these findings, revealing that participation in SPHI reduced the poverty vulnerability of urban residents, significantly higher than that of rural residents. One key difference between urban and rural populations lies in the stability of household income. Urban residents generally experience more income instability, a result of differences in employment structures and income sources between urban and rural areas. Urban income is largely derived from non-agricultural sources, such as wages and business income [91]. In contrast, agricultural income remains a substantial portion of rural households’ earnings, providing a buffer against income fluctuations [92]. The impact of disease risk shocks may, therefore, be more detrimental to urban households, as they often depend on the health and productivity of the primary income earner. Moreover, urban residents tend to have higher health service demands and expenditures [93]. SPHI offers more comprehensive financial protection, particularly when seeking medical care, which explains its stronger impact on urban populations. However, while the urban population benefits more from SPHI, rural populations exhibit higher baseline vulnerability. This suggests that future research should explore how SPHI performs under different scenarios of social health insurance reimbursement.

Several important limitations must be foregrounded. First, while we employed PSM and IV methods, residual endogeneity concerns persist - our instrument (insurance density) may correlate with regional economic factors that independently affect poverty vulnerability. Second, the exclusive focus on Shandong Province limits generalizability to other Chinese regions or countries with different economic contexts. Third, the analysis insufficiently explores how SPHI’s effectiveness interacts with existing social health insurance reimbursement policies, a critical mediator of its potential impact.

Conclusion

Our findings indicate SPHI may help reduce poverty vulnerability, though longitudinal studies are needed to confirm long-term impacts. While these results offer potential insights, we recommend that policymakers consider pilot programs to evaluate the feasibility and effectiveness of SPHI in different contexts. In low- and middle-income countries, efforts could start with achieving basic and widespread SHI coverage, with SPHI positioned as a supplementary option tailored to local economic conditions and health system structures. Particular attention should be paid to high-risk groups, such as older adults, rural residents, individuals in the western region, and those with chronic diseases, for whom SPHI could play a crucial role as a component of health financing by relevant authorities. Additionally, populations with relatively disadvantaged socioeconomic statuses often experience poorer health outcomes [94]. Thus, strengthening SPHI infrastructure—such as improving product pricing accuracy and addressing adverse—may help expand coverage and enhance targeted poverty alleviation efforts, especially in rural areas where uptake is lower.

Abbreviations

SPHI

Supplementary private health insurance

IV

Instrumental variable

PSM

Propensity score matching

CHE

Catastrophic health expenditure

SHI

Social health insurance

VEP

Vulnerability to expected poverty

SLA

Sustainable Livelihoods Approach

BMI

Body Mass Index

EQ5D

EuroQoL-5 dimension

2SLS

Two-stage least squares

Author contributions

J.M. and S.D. drafted the manuscript. J.S., J.M., S.D., J.L., J.H. and P.L. contributed to data collection and analysis. J.S. and J.L. conceptualized and designed the study, and critically reviewed and revised the manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (Grant Number: 72274108) and Natural Science Foundation of Shandong Province (Grant Number: ZR2022MG003).

Data availability

The data that support the findings of this study are available from medical management service center of Shandong health commission but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of medical management service center of Shandong health commission.

Declarations

Ethics approval and consent to participate

This study protocol was approved and organized by Health Commission of Shandong Province. This study was reviewed and approved by the Institutional Review Board (Academic Research Ethics Committee) of Shandong University School of Public Health. All procedures were in accordance with the ethical standards of the Helsinki Declaration. Written informed consents clarifying the study purposes were obtained from each participant.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.