Catastrophic health expenditure and commercial health insurance purchase: evidence from China

Abstract

Background

In China, catastrophic health expenditure (CHE) is a predominant factor that drives households into poverty. While commercial health insurance (CHI) can mitigate the adverse consequences of CHE, its uptake remains limited. This study aimed to analyse the association between CHE and the purchase of CHI in China from an ex post perspective.

Methods

This study included 34,534 observations from data from the 2019 China Household Finance Survey. A two-part model was employed to analyse the association between CHE (categorized as none, mild, moderate, or severe) and individual CHI purchase (both purchase behaviour and premium expenditure), guided by an integrated framework. Subgroup analyses were conducted to compare factors influencing CHI purchase between observations with and without household CHE.

Results

The prevalence rates of mild, moderate, and severe CHE were 19.27%, 6.97%, and 6.89%, respectively. The overall CHI purchase rate was 4.48%. Severe CHE was negatively associated with the likelihood of purchasing CHI (OR = 0.475, p < 0.01), whereas moderate CHE was negatively associated with premium expenditure among purchasers (coef.=−1.227, p < 0.05). Ambiguous risk preference was negatively associated with both the likelihood of purchasing CHI and premium expenditure. Factors such as poor self-rated health, older age (≥65 years), male sex, and lack of financial attention were negatively associated with the likelihood of purchasing CHI. Greater educational attainment and a stronger social support network were positively associated with the likelihood of purchasing CHI. Subgroup analyses revealed that the influencing factors of CHI purchase systematically differed between observations with and without household CHE.

Conclusions

CHE is negatively associated with CHI purchase. The government should establish a full-cycle risk protection mechanism for vulnerable groups by deepening health care reform, implementing targeted subsidies, and enhancing public awareness. It is equally important for insurance institutions to improve the accessibility of CHI.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-026-14181-2.

Introduction

Poverty remains a persistent and critical global challenge, with high health care costs being among its diverse drivers [1]. A survey by the World Health Organization (WHO) revealed that in 89 countries, the proportion of people suffering from severe economic hardship due to disease is growing at an annual rate of 11% [2]. In China, catastrophic health expenditure (CHE) has emerged as a predominant risk factor driving households into or back into poverty [3]. The solution to the vicious circle of “disease–poverty–disease” is directly related to the effectiveness of the construction of a “Healthy China”. Universal health insurance coverage is the key to breaking this cycle. China’s social basic medical insurance (SBMI), which covers more than 95% of the population, has provided protection for the majority of Chinese people against the financial shocks caused by illnesses [4].

However, China’s SBMI highlights several flaws. Its sustainability is challenged by factors such as extensive coverage, low security, significant pressure on fund balance, limited fiscal transfer payments, and decentralized management. Moreover, disparities in premium and benefit entitlements undermine the equity and consistency of the programme [5]. Commercial health insurance (CHI) can effectively supplement the SBMI and provide residents with more insurance choices and greater risk protection [6]. Therefore, to bolster the protective measures for the populace and minimize welfare losses, in 2020, China promulgated the “Opinions of the State Council on Deepening the Reform of the Medical Security System”. It clearly stated that by 2030, a comprehensive system will be established. Termed the multilevel medical security system, it will include SBMI as the main body, medical assistance as the support, and CHI as well as charitable donations as supplements. The term CHI refers to insurance under which insurance institutions pay insurance benefits to insured individuals for health-related reasons or the provision of medical services. It covers commercial medical insurance, critical illness insurance, disability income insurance, long-term care insurance, and medical accident insurance.

Despite the current accelerated development momentum of China’s CHI market, it still lags behind the global average level. According to government data1, in 2024, the premium income of CHI accounted for 22.9% of the premium income of personal insurance in China, representing a large gap compared with 30% in mature markets [7]. With respect to the proportion of CHI expenditure to total health care expenditure, China’s total health care expenditure reached 9.06 trillion yuan in 2024, with CHI claim payments amounting to merely 384.8 billion yuan—a proportion of approximately 4.25%, whereas the corresponding proportion in developed countries typically reaches 10%. From the supply-side perspective, the development of China’s CHI market is characterized by regional disparities. Compared with those in the eastern and central regions, insurance institutions in the western regions demonstrate lower operational efficiency and professionalization, which impairs residents’ access to CHI products and the diversity of their choices. Such unbalanced market development undermines the accessibility and equity of CHI products and services, thereby constraining the industry’s growth potential [8]. From the demand-side perspective, the effective demand for CHI among Chinese residents has not yet been fully accessed. In 2024, CHI density was 693.97 yuan per person, with a penetration of only 0.72%. During the same period, CHI density in Japan was 780.90 yuan per capita (calculated on the basis of the 2024 average exchange rate), with a penetration of 3.29%. Characterized by a demand-side-driven nature [9], the mature development of China’s CHI market will benefit from more in-depth research into demand-side purchasing decisions.

CHI plays a pivotal role in controlling medical expenditures and alleviating poverty [10]. When people are affected by critical illnesses, CHI can promptly provide financial compensation, mitigate the financial impact of such shocks, and reduce poverty vulnerability [11]. A U.S. study indicated that CHI benefits reduced the poverty rate among people under the age of 65 by 3.7% points [12]. A growing body of evidence has indicated that CHI complements social medical insurance and promotes the utilization of high-value preventive care, thereby reducing the subsequent use of costly medical services [13, 14]. In addition, studies have emphasized that CHI can safeguard households’ labour supply, increase their livelihood capital, and thereby curb poverty [15].

While existing studies have fully elaborated on the effective control of CHE by CHI, this discussion is based on an ex ante prevention perspective, focusing on CHI as a risk-mitigation tool to block the risks of adverse outcomes. However, these studies focus only on groups that benefit from CHI, with insufficient attention given to vulnerable groups already trapped in distress. When individuals have already experienced CHE, do they possess sufficient ability and motivation to purchase CHI to prevent them from falling into a vicious cycle of poverty? This represents an ex post remediation perspective. Addressing this question will help provide a more comprehensive safety net for vulnerable groups and contribute to the humanized development of the CHI industry.

In summary, this study aims to examine the CHI purchasing decisions of vulnerable groups from an ex post perspective. We seek to address the following two research questions: What is the association between CHE and CHI purchase decisions in China? What factors influence CHI purchase decisions among individuals who have experienced household CHE, and do these factors systematically differ from those affecting individuals without such experience?

The marginal contributions of this paper are as follows. First, on a theoretical level, this study introduces an integrated model that incorporates a social psychology framework into insurance demand analysis, thereby broadening the understanding of CHI purchase, particularly the decision-making mechanisms following negative experiences. Second, this paper contributes a novel perspective by breaking through the existing research paradigm focused on ex ante prevention and, for the first time, systematically analysing the CHI decisions of populations impacted by CHE from an ex post remediation perspective. This approach provides novel policy insights for China and other developing countries to enhance full-cycle risk protection mechanisms for vulnerable groups.

Literature review

Influencing factors of commercial health insurance purchase

The decision to purchase CHI is shaped by a complex interplay of factors across multiple dimensions. From an individual perspective, CHI purchase is linked to factors such as sex [16], age, social status [17], health status [18], social interaction, income [19], education [20], marital status [21], and financial literacy [22]. In addition, household characteristics such as family size [23], population structure [24], and family ties [25] have been regarded as important indicators. From the perspective of the insurance industry, the degree of competition in the insurance market and the product structure [26], product design [27], marketing methods, claim settlement services [28], and the professionalism of insurance agents [29] are likely to be associated with residents’ CHI purchase. From the perspective of the policy environment, the literature has explained the association between CHI purchase and preferential tax policies [30, 31] as well as social insurance coverage [32, 33] from both theoretical and empirical perspectives.

While the aforementioned factors have been extensively studied, the underlying mechanisms driving decision-making warrant further investigation. Imprinting theory offers a valuable perspective in this regard, positing that significant early-life experiences or shocks can create enduring “imprints” on individuals, consequently shaping their subsequent personality traits and decision-making preferences [34, 35]. Many studies have shown that individuals’ economic decisions are strongly associated with the stimuli they have experienced [36, 37]. Therefore, numerous studies have examined the association between experiences or events and CHI purchase. This kind of research can be further divided into two streams: general experiences and adverse experiences. In terms of general experiences, studies have confirmed that life experiences [38, 39] are potentially associated with the demand for long-term care insurance, whereas public health education [40] is positively associated with CHI purchase. In terms of adverse experiences, air pollution has received considerable attention. Studies have shown that as air quality continues to deteriorate, the demand for CHI increases significantly, driven by a mechanism through which adverse experiences enhance individuals’ risk perception [41, 42]. Additionally, the literature has verified the association between public health emergencies and the demand for personal insurance and short-term travel insurance [43, 44]. Some studies have also considered various natural disasters and analysed their effects on the demand for earthquake insurance and flood insurance [45, 46].

While existing studies have comprehensively discussed sociodemographic and external environmental factors, the exploration of adverse experiences from the perspective of imprinting theory remains insufficient. CHE, as a major negative shock characterized by both health trauma and financial devastation, is likely to profoundly influence individuals’ risk-related decision-making. Consequently, this study proposes its first hypothesis:

H1

CHE is negatively associated with individuals’ CHI purchase.

Research on catastrophic health expenditure

CHE is intricately intertwined with the poverty trap. It not only directly exacerbates the incidence and depth of poverty, particularly among rural and middle- to low-income households [47] but also imposes long-term developmental constraints through mechanisms such as human capital depletion (e.g., reduced labour supply due to illness) and the exhaustion of household financial resources [48]. A study conducted in Brazil emphasized that CHE exacerbates social inequality [49]. Even among households not falling below the poverty line, CHE leads to distorted consumption patterns that crowd out essential developmental expenditures on education and nutrition [50] while potentially triggering mental health issues that reinforce the vicious cycle [51].

Although extensive research has yielded substantial insights into the measurement [52, 53], determinants [54, 55], and direct welfare consequences of CHE [47–51], their behavioural spillover effects remain underexplored. A critical gap persists in the context of CHI, which is widely recognized as a key mechanism for preventing or mitigating CHE. While the scholarly focus has been almost exclusively on its ex ante preventive role, little attention has been given to how vulnerable people, once they have experienced a CHE shock, engage in ex post risk management. The mechanisms underlying their subsequent decisions regarding CHI are particularly overlooked.

The profound shock of a CHE is likely to fundamentally reshape an individual’s risk perception, budget allocation and decision preferences. It can be logically inferred that the decision-making framework for purchasing CHI among CHE survivors is likely driven by a distinct set of factors compared to those without such experiences. Therefore, this study posits its second hypothesis:

H2

The factors influencing CHI purchase differ between individuals from households with CHE and those without CHE.

Methods

Theoretical framework

In this study, the stimulus–organism–response (SOR) model and the motivation–opportunity–ability (MOA) model are integrated within a synthesized analytical framework. This integrated approach is designed to provide a more comprehensive examination of how CHE, in conjunction with other factors, shapes an individual’s CHI purchase decision.

The SOR model provides the foundational logical chain for this study. In this model, a stimulus refers to the occurrence of external factors or events; the organism, as a mediating variable, represents changes in an individual’s physiological and psychological states, such as emotional or cognitive responses; and the response refers to the outcome variable of an individual’s attitudes or behaviours. The model posits that the occurrence of an event alters an individual’s internal emotional states and perceptions of external things, thereby influencing their subsequent behaviours [56]. However, the model defines the organism variable in rather broad terms, failing to clearly delineate its internal composition or the differential roles played by distinct individual traits.

To address this limitation, this study incorporates the MOA model. The MOA model explains individuals’ decision-making from aspects such as attitude, self-perception, and the environment. It first emerged in the study of consumer decision-making in response to commercial advertisements [57]. Specifically, motivation is the psychological drive that prompts individuals to engage in certain behaviours stemming from personal needs, including the need for autonomy, competence, and relatedness [58, 59]. Opportunity refers to external factors facilitating or inhibiting an individual’s behaviour. Although these external factors are closely related to individuals, they are not under their control [60]. Ability is the individual’s potential to affect behaviour efficiency, including cognitive status, skills, and resources [61]. While the MOA model excels in deconstructing the microfoundations of individual decision-making, its conventional application often overlooks a clear external stimulus that acts as the initial trigger for the behaviour.

Accordingly, the integrated framework of this study is depicted in Fig. 1. In this study, household CHE is conceptualized as the pivotal stimulus. The three dimensions of the MOA model—motivation, opportunity, and ability—collectively constitute the organism in the SOR model, detailing the individual’s internal psychological and resource state following the stimulus. The subsequent response is the decision regarding CHI purchase.

Fig. 1.

Integrated theoretical framework. Note: Author’s original

The strength of this integrated framework is twofold. First, it establishes a clear pathway from an external shock to a specific financial behaviour. More importantly, it leverages the MOA model to deconstruct the black box of the organism. This approach allows for a systematic investigation into which specific motivations (e.g., risk preference), abilities (e.g., financial attention), and opportunities (e.g., social support network) play critical roles in this pathway. Consequently, it provides a robust theoretical foundation for understanding the heterogeneity in decision-making mechanisms among individuals who have experienced CHE.

Data sources

The data used in this study were drawn from the China Household Finance Survey (CHFS). This nationally representative longitudinal survey has been conducted biennially since 2011. The CHFS is a nationwide survey organized and conducted by the China Household Finance Survey and Research Centre of Southwestern University of Finance and Economics, Chengdu, China. For the CHFS, data were collected on individuals, households, and communities through tracking to reflect essential demographic characteristics, assets and liabilities, income and consumption, the purchase of insurance, financial literacy, economic expectations, etc. [62]. The CHFS survey adopts a stratified, three-stage sampling design with probability proportional to size, combining onsite visits with quarterly telephone follow-ups to collect and update sample data. Considering the possible impact of the COVID-19 pandemic and the fact that CHFS 2021 and 2023 data are not yet fully publicly available, the data from 2019 were selected for this study. In the 2019 wave of data, the number of observations reached 34,643 households, which covered 29 provinces/cities/autonomous regions, 170 cities, 345 districts and counties and 13,604 village/residential committees in China. Overall, the data are representative at both the national and provincial levels, providing sufficient empirical support for this study.

The initial total sample comprised 34,643 observations. The observations were refined through the following sequential exclusion criteria. First, 10 observations were deleted because they represented individuals under the age of 18 years. Subsequently, 98 observations were excluded because of missing values for key variables concerning CHI purchase. Finally, 1 observation was removed because of a missing value for the variable for CHE. After these criteria were applied, the final domain for analysis included 34,534 observations. The flow chart of the study population selection process is shown in Fig. 2.

Fig. 2.

Flow chart of the study population selection process

Given the limited number of observations with missing values in the key variables—which were excluded—this study employed imputation solely for the control variables. This study addressed missing values through multiple imputation. The analysis included 34,534 observations, with 8.32% having at least one missing value. The missing values pattern, which assumed data to be missing at random, was examined in this study, after which multiple imputation by chained equations was employed to generate 20 imputed datasets. The imputation model specified link functions appropriate to the distribution of the variables with missing values: linear regression for continuous variables, ordered logistic regression for ordered categorical variables, and logistic regression for binary variables. The model was conditioned on a set of complete auxiliary variables to strengthen the plausibility. Parameter estimates and standard errors from the analyses conducted on each imputed dataset were pooled using Rubin’s rules.

Measures

Dependent variable: commercial health insurance purchase

The dependent variable in this study is the household head’s CHI purchase decision, measured along two dimensions: purchasing behaviour and premium expenditure. Purchasing behaviour is a binary variable, coded as 1 if the household head reports holding CHI and 0 otherwise. Premium expenditure is a continuous variable. Within the subsample of household heads who have purchased CHI, this study takes the natural logarithm of their premium expenditure for analysis to mitigate right-skewness in the distribution.

Household mutual assistance is an important mechanism for risk response in Chinese society. Therefore, the occurrence of CHE in a household will have an impact on every family member, especially the core role in the family—the household head. The CHFS questionnaire defines the household head as the person who plays a decisive role in family affairs. Even if the household head does not incur the CHE, the economic and psychological pressure they face is undeniable. Compared with other family members, the head of the household may be more sensitive to this shock. This research aims to explore what CHI purchase decision an individual—not the entire household—makes when exposed to the background risk of CHE and which characteristics lead to differences in these decisions. Considering that the household head is the most distinctive individual in a household, this study selects them as the research subjects. Notably, the theoretical framework of this study is the integrated SOR-MOA model, where the main motivation and ability factors identified need to be measured at the individual level. Thus, defining the CHI purchase at the individual level is consistent with the theoretical framework.

Independent variable: household catastrophic health expenditure

In this study, the occurrence of CHE in observed households was taken as the independent variable. To calculate CHE, this study drew on the approach adopted in a previous study [50]. The CHFS questionnaire provided the total medical expenses of a family during the past year. On this basis, this study excluded the reimbursed expenses and included the medical expenses paid for this family by others to calculate the out-of-pocket expenses. In this study, out-of-pocket expenses were used as the numerator, and total household consumption was used as the denominator to obtain CHE. The WHO defines 40% as the threshold for households to incur CHE—a value commonly used in studies on developing countries [2]. In addition, in studies of the incidence of CHE in middle-income countries such as Brazil [63], Vietnam [64], Turkey [65], and Malaysia [66], 10 and 25% are also regarded as key thresholds alongside 40%. Therefore, the current study drew on the practices of a previous study [67] and defined three levels of CHE as follows: observations with a proportion between 10 and 25% were defined as having mild CHE in the household and coded as “2”; those with a proportion between 25 and 40% were defined as having moderate CHE in the household and coded as “3”; and those with a proportion higher than 40% were defined as having severe CHE in the household and coded as “4”. Observations with no CHE in the household were coded as “1”.

This study defined the occurrence of CHE at the household level, as the CHFS provides only household-level consumption data. Owing to data limitations, this study cannot accurately identify individuals who have experienced CHE, but households that have incurred such expenditures are identifiable.

Control variables

Following the MOA model and drawing on the literature, this study controlled for a comprehensive set of variables.

In the motivation dimension, the variables selected in this study included risk preference and self-rated health status. Many studies have used answers to the question “If you have a sum of money, which of the following projects would you be willing to invest in?” to measure risk preference [68, 69]. Therefore, this study maintained this approach. In terms of self-rated health status, in this study, answers to the question “Compared to your peers, how is your current physical condition?” were used.

In the opportunity dimension, the variables selected for this study included age, family size, permanent place and social support network. Drawing on prior studies [70, 71], this study employed the income and expenditure of red envelopes and gifts (including holiday allowances) during family festivals, as well as those related to weddings and funerals, as proxy variables for social support network.

In the ability dimension, the variables selected in this study included education, total family income, family savings and financial attention. Family savings were measured by the ratio of the sum of household cash and liquid deposits to total financial assets [72]. This study measured financial attention on the basis of answers to the question “How much do you usually pay attention to economic and financial information?”

Other sociodemographic variables included sex, marriage and employment. Considering that many studies on CHI purchases in China have included SBMI enrolment as a control variable [73, 74], this study followed this practice.

Family size was calculated on the basis of household and individual identification variables. Permanent place was predefined by the survey. All the other variables were derived from the responses of the household head.

Statistical analysis

Given the substantial number of zero values (i.e., nonpurchasers) in the dependent variable—CHI purchase—and the highly right-skewed distribution of premium expenditures, this study employed a two-part model to separately model the purchase behaviour and the premium expenditure, thereby addressing potential estimation bias due to the nonnormal distribution of the dependent variable. The first part (whether an individual purchases CHI) was estimated via a logistic regression model, whereas the second part (the amount of premium expenditure on CHI) was analysed with a multiple linear regression model.

The modelling process of this study is as follows. Let CHI_i = 1 indicate that individual i has purchased CHI. P represents the probability of the purchase behaviour occurring. CHE_i denotes individual i’s household’s occurrence of CHE. Controls_i stands for a set of control variables. α is the intercept. β and γ are coefficients.ε_i represents the error term. The logistic regression model takes the following form:

1

Second, conditional on CHI_i = 1 (i.e., among the group of individuals who purchased CHI), let _i denote the premium expenditure on CHI for the i-th household head. The specification of the second part of the two-part model is as follows:

2

Through the two-part model, this study first verifies the association between household CHE and individual CHI purchase decision. To test the second hypothesis, this study subsequently divides the full observations into two subgroups based on households with CHE and performs separate regressions for each subgroup, thereby enabling a comparison of coefficient differences between them.

To address potential sample selection bias, this study employs the Heckman selection model as a robustness check. First, a selection equation is constructed to identify the factors influencing CHI purchase, which are estimated using a probit model to derive the inverse Mills ratio. In the second stage, the inverse Mills ratio is subsequently incorporated into Eq. (4) for estimation. The Heckman selection model is specified as follows:

3

4

In the Heckman selection model, this study employs the social support network as the exclusion restriction variable. Social support networks have been widely documented as significant factors influencing individuals’ insurance decisions [75, 76]. Robust informal support systems may partially substitute for commercial insurance through risk-sharing mechanisms, thereby significantly reducing an individual’s propensity to purchase CHI. Moreover, social support networks primarily affect decisions by altering insurance purchase intentions, but in theory, they should not directly influence the level of premium expenditure, as the latter is more strongly determined by an individual’s payment capacity and cognitive ability [77].

The results of this study are presented as odds ratios, coefficients and 95% confidence intervals. All analyses were performed in Stata Version 19.0 (StataCorp, College Station, TX). Given the complex sampling design of the CHFS, all analyses employed survey weights to ensure the representativeness of the findings for the national adult population. All analyses were conducted under the svyset command framework in Stata. Consequently, the percentages, means, and model coefficients reported in this study are nationally representative estimates derived through this weighting procedure.

Results

Descriptive statistical analysis

The results of the descriptive statistics for each variable are presented in detail in Table 1. The table shows that 19.27% of households had mild CHE, 6.97% of households had moderate CHE, and 6.89% of households had severe CHE.

Table 1.

Weighted population characteristics and commercial health insurance purchase. (n = 34,534)

Variables	Indicators and Codes	Weighted Mean（S.D.）or %	Weighted Commercial Health Insurance purchase Rate % (95% CI)
Commercial Health Insurance Purchase	No = 0	95.52	__
	Yes = 1	4.48	__
ln Commercial Health Insurance Premium		0.33 (0.16)
Stimuli Dimension
CHE	No in household = 1	66.87	5.62 (5.08~6.16)
	Mild CHE in household = 2	19.27	2.97 (2.37~3.57)
	Moderate CHE in household = 3	6.97	1.40 (0.84~1.96)
	Severe CHE in household = 4	6.89	0.80 (0.41~1.20)
Motivation Dimension
Risk Preference	Tend to invest in high-risk, high-return projects = 1	2.64	5.82 (3.17~8.46)
	Tend to invest in projects with slightly higher risks and slightly higher returns = 2	3.53	13.77 (10.68~16.87)
	Tend to invest in projects with average risk and average return = 3	15.47	8.60 (7.37~9.84)
	Tend to invest in projects with slightly lower risk and slightly lower return = 4	14.69	6.90 (5.80~8.00)
	Unwilling to take any risks = 5	47.70	2.74 (2.35~3.13)
	Ambiguous preference = 6	15.97	1.20 (0.79~1.62)
Self-Rated Health Status	Very good = 1	14.20	6.20 (5.04~7.37)
	Good = 2	30.62	6.34 (5.50~7.17)
	General = 3	38.98	3.78 (3.29~4.27)
	Poor = 4	12.67	1.31 (0.81~1.80)
	Very poor = 5	3.53	0.63 (0.05~1.20)
Opportunity Dimension
Age	18–24 years old = 1	1.11	7.05 (2.79~11.30)
	25–34 years old = 2	9.28	9.12 (7.44~10.80)
	35–54 years old = 3	49.73	6.05 (5.44~6.66)
	55–64 years old = 4	20.97	2.11 (1.70~2.53)
	65 years old or older = 5	18.91	0.06 (0.04~0.07)
Family Size	1–3 person(s)=1	65.48	4.38 (3.89~4.86)
	4–6 persons = 2	32.36	4.78 (4.16~5.40)
	7 persons or more = 3	2.16	3.17 (1.37~4.97)
Permanent Place	Rural = 1	35.61	1.84 (1.41~2.28)
	Urban = 0	64.39	5.94 (5.36~6.53)
Social Support Network	ln gift income and expenditure	5.55 (0.04)	—
Ability Dimension
Total Family Income	10,000 yuan or lower = 1	14.69	2.06 (1.46~2.66)
	10,001– 30,000 yuan = 2	16.07	1.46 (1.07~1.86)
	30,001– 60,000 yuan = 3	21.17	2.65 (2.13~3.18)
	60,001– 90,000 yuan = 4	16.03	4.31 (3.42~5.20)
	Higher than 90,000 yuan	32.04	8.40 (7.50~9.30)
Family Savings		0.51 (0.00)	—
Education	Primary schooling or less = 1	25.91	0.87 (0.61~1.13)
	Junior High School/Senior High School/Technical1Secondary School/Vocational High School = 2	53.91	3.87 (3.39~4.35)
	College or above = 3	20.18	10.75 (9.49~12.01)
Financial Attention	Very concerned = 1	2.16	7.63 (4.67~10.60)
	Relatively concerned = 2	3.66	11.17 (8.41~13.93)
	General = 3	20.49	7.93 (6.92~8.95)
	A little concerned = 4	23.99	5.34 (4.58~6.10)
	Never = 5	49.70	2.02 (1.66~2.37)
Other Dimension
Sex	Male = 1	74.87	3.91 (3.48~4.34)
	Female = 0	25.13	6.20 (5.39~7.00)
Social Basic Medical Insurance	Enroll = 1	93.51	4.53 (4.11~4.94)
	No = 0	6.49	3.86 (2.61~5.11)
Marriage	Married = 1	85.38	4.51 (4.09~4.94)
	Other = 0	14.62	4.30 (3.35~5.25)
Employment	Employment = 1	72.82	5.35 (4.83~5.86)
	Unemployment = 0	27.18	2.17 (1.73~2.61)

Note: All estimates are weighted to represent the target population using complex survey analysis methods

The purchase rate of CHI was still relatively low, at only 4.48%. Individuals with higher purchase rates of CHI included those who had no CHE in the household (Rate = 5.62%, 95% CI = 5.08~6.16); tended to invest in projects with slightly higher risks and slightly higher returns (Rate = 13.77%, 95% CI = 10.68~16.87); had good self-rated health status (Rate = 6.34%, 95% CI = 5.50~7.17); were aged 25–34 years (Rate = 9.12%, 95% CI = 7.44~10.80); had a family size of 4–6 persons (Rate = 4.78%, 95% CI = 4.16~5.40); lived in urban areas (Rate = 5.94%, 95% CI = 5.36~6.53); had a total family income higher than 90,000 yuan (Rate = 8.40%, 95% CI = 7.50~9.30); had a college or above education (Rate = 10.75%, 95% CI = 9.49~12.01); were relatively concerned about financial information (Rate = 11.17%, 95% CI = 8.41~13.93); were female (Rate = 6.20%, 95% CI = 5.39~7.00); were enrolled in SBMI (Rate = 4.53%, 95% CI = 4.11~4.94); were married (Rate = 4.51%, 95% CI = 4.09~4.94); and were employed (Rate = 5.35%, 95% CI = 4.83~5.86).

The proportion of female household heads was 25.13%, which is reasonable in China. Studies using data from the China Health and Nutrition Survey, self-conducted household surveys, and the fifth National Health Service Survey organized by the Chinese government have reported female household head proportions very close to those of this study, at 19.00% [78], 24.80% [79], and 25.40% [80]^, respectively.

Two-part model analysis results for commercial health insurance purchase

Table 2 presents the results of the two-part model analysis. The results from the first part indicated that compared with individuals from households without CHE, those from households with severe CHE were less likely to purchase CHI (OR = 0.475, p < 0.01). Compared with individuals who tended to invest in high-risk, high-return projects, those with ambiguous risk preferences were less likely to purchase CHI (OR=0.513, p < 0.05). Compared with people who self-rated their health status as very good, those who self-rated their health status as poor were less likely to purchase CHI (OR = 0.587, p < 0.05). Compared with the young group aged 18–24 years, the elderly group aged 65 years and older were less likely to purchase CHI (OR = 0.286; p < 0.01). The likelihood of individuals purchasing CHI increased with increasing social support network strength (OR = 1.042, p < 0.01). Compared with individuals with primary schooling or less, those with better education were more likely to purchase CHI (OR = 2.094, p < 0.001; OR = 2.703, p < 0.001). Compared with individuals who were very concerned about financial information, individuals who were a little concerned or never concerned about financial information were less likely to purchase CHI (OR = 0.636, p < 0.05; OR = 0.448, p < 0.001). Compared with females, males were less likely to purchase CHI (OR = 0.607, p < 0.001).

Table 2.

Estimated two-part model of commercial health insurance purchase. (survey-Weighted estimates)

Variables	Purchase(n = 34,534)		ln Commercial Health Insurance Premium (n = 1,142)
	OR(95%CI)	P value	Coefficient(95%CI)	P value
CHE
No in household*	—	—	—	—
Mild CHE in household	0.942(0.744~1.194)	0.622	−0.216(−0.823~0.390)	0.484
Moderate CHE in household	0.660(0.427~1.019)	0.061	−1.227(−2.427~0.026)	0.045
Severe CHE in household	0.475(0.285~0.790)	0.004	−0.387(−1.362~0.588)	0.435
Risk preference
Tend to invest in high-risk, high-return projects*	—	—	—	—
Tend to invest in projects with slightly higher risks and slightly higher returns	1.390(0.806~2.395)	0.236	−0.568(−1.626~0.490)	0.292
Tend to invest in projects with average risk and average return	1.169(0.691~1.976)	0.561	−0.801(−1.649~0.046)	0.064
Tend to invest in projects with slightly lower risk and slightly lower return	1.107(0.660~1.857)	0.701	−0.515(−1.358~0.329)	0.231
Unwilling to take any risks	0.829(0.493~1.394)	0.480	−0.845(−1.689~0.000)	0.050
Ambiguous preference	0.513(0.283~0.929)	0.028	−1.843(−3.329~-0.357)	0.015
Self-Rated Health Status
Very good*	—	—	—	—
Good	0.973(0.760~1.246)	0.827	0.242(−0.383~0.867)	0.447
General	0.833(0.657~1.057)	0.133	0.034(−0.623~0.691)	0.919
Poor	0.587(0.380~0.908)	0.017	0.000(−1.373~-1.373)	1.000
Very poor	0.435(0.170~1.116)	0.083	−0.369(−2.787~2.049)	0.764
Age
18–24 years old*	—	—	—	—
25–34 years old	1.378(0.668~2.841)	0.385	2.097(−0.289~4.484)	0.085
35–54 years old	1.433(0.714~2.878)	0.311	2.560(0.172~4.947)	0.036
55–64 years old	0.781(0.382~1.595)	0.496	2.125(−0.404~4.654)	0.099
65 years old or older	0.286(0.131~0.626)	0.002	−0.684(−3.516~2.149)	0.635
Family Size
1–3 person(s)*	—	—	—	—
4–6 persons	1.018(0.850~1.220)	0.844	0.221(−0.244~0.686)	0.350
7 persons or more	1.115(0.609~2.044)	0.724	0.882(−0.001~1.766)	0.050
Permanent Place
Urban*	—	—	—	—
Rural	0.770(0.572~1.038)	0.087	−0.651(−1.306~0.004)	0.051
Social Support Network	1.042(1.013~1.071)	0.004	−0.040(−1.113~0.034)	0.290
Total Family Income
10,000 yuan or lower*	—	—	—	—
10,001– 30,000 yuan	0.608(0.404~0.917)	0.018	0.259(−0.628~1.146)	0.566
30,001– 60,000 yuan	0.753(0.530~1.070)	0.113	−0.098(−1.024~0.829)	0.835
60,001– 90,000 yuan	1.016(0.726~1.423)	0.925	−0.080(−0.957~0.796)	0.857
Higher than 90,000 yuan	1.273(0.901~1.798)	0.171	−0.313(−1.158~0.530)	0.465
Family Savings	0.809(0.632~1.036)	0.093	−0.038(−0.826~0.750)	0.924
Education
Primary schooling or less*	—	—	—	—
Junior High School/Senior High School/Technical Secondary School/Vocational High School	2.094(1.488~2.947)	0.000	0.749(−0.401~1.900)	0.201
College or above	2.703(1.843~3.965)	0.000	0.976(−0.253~2.205)	0.119
Finance Attention
Very concerned*	—	—	—	—
Relatively concerned	1.214(0.718~2.054)	0.468	0.453(−0.311~1.218)	0.244
General	0.787(0.505~1.229)	0.292	−0.444(−1.253~0.364)	0.280
A little concerned	0.636(0.418~0.967)	0.034	−0.038(−0.829~0.754)	0.926
Never	0.448(0.289~0.693)	0.000	−0.004(−0.871~0.864)	0.994
Sex
Female*	—	—	—	—
Male	0.607(0.504~0.733)	0.000	0.100(−0.434~0.636)	0.711
Social Basic Medical Insurance
No*	—	—	—	—
Enroll	0.894(0.620~1.290)	0.549	−0.709(−1.339~-0.079)	0.027
Marriage
Other*	—	—	—	—
Married	0.930(0.694~1.246)	0.626	−0.116(−0.836~0.604)	0.751
Employment
Unemployment*	—	—	—	—
Employment	1.199(0.905~1.564)	0.213	0.246(−0.429~0.922)	0.474

Note: * stands for reference group. Multicollinearity was assessed using the unimputed sample (n = 31,660) as VIFs were not available for the multiply imputed model. All VIFs were < 10 (mean = 1.28), indicating no severe multicollinearity. See Supplementary Material for full results

The results from the second part indicated that compared with individuals from households without CHE, those from households with moderate CHE had lower premium expenditures on CHI (coef.=−1.227, p < 0.05). Compared with individuals who tended to invest in high-risk, high-return projects, those unwilling to take any risks and those with ambiguous risk preferences reported lower premium expenditures (coef.= −0.845, p = 0.050; coef.=−1.843, p < 0.05). Compared with the young group aged 18–24 years, the group aged 35–54 years had higher premium expenditures (coef.=2.560, p < 0.05). Compared with households with 1–3 members, households with 7 or more members had higher premium expenditures (coef.=0.882, p = 0.050). Compared with their urban counterparts, rural residents had lower premium expenditures (coef.=−0.651, p = 0.051). Individuals enrolled in SBMI had lower premium expenditures than those without SBMI did (coef.=−0.709, p < 0.05).

Heckman selection model analysis results

Table 3 presents the estimation results of the Heckman selection model. The primary purpose of this model is to diagnose and correct for potential selection bias arising from unobserved factors. The coefficient of the inverse Mills ratio was statistically insignificant (IMR = 0.819; p = 0.508), indicating the absence of severe sample selection bias in this study and confirming the reliability of the estimates based on the two-part model. Furthermore, a comparison between the results in Tables 3 and 2 reveals no substantial changes in the estimated coefficients or significance levels of the key independent variables. Taken together, this evidence suggests that selection bias does not pose a serious threat to the estimation results of this study, thereby ensuring the unbiasedness and robustness of the main findings.

Table 3.

Estimated Heckman selection model of commercial health insurance purchase. (survey-Weighted estimates)

Variables	LnCommercial Health Insurance Premium
	Selection model (Commercial Health Insurance purchase, n = 34,534)		Outcome model (n = 1,142)
	Coefficient(95%CI)	P value	Coefficient(95%CI)	P value
CHE
No in household*	—	—	—	—
Mild CHE in household	−0.051(−0.132~0.030)	0.215	−0.168(−0.633~0.296)	0.477
Moderate CHE in household	−0.092(−0.233~0.049)	0.200	−0.949(−1.813~-0.084)	0.032
Severe CHE in household	−0.212(−0.388~-0.035)	0.019	−0.985(−2.194~0.225)	0.111
Risk preference
Tend to invest in high-risk, high-return projects*	—	—	—	—
Tend to invest in projects with slightly higher risks and slightly higher returns	0.243(0.044~0.442)	0.017	0.228(−0.0897~1.353)	0.691
Tend to invest in projects with average risk and average return	0.177(0.002~0.352)	0.048	−0.193(−1.187~0.801)	0.704
Tend to invest in projects with slightly lower risk and slightly lower return	0.134(−0.043~0.311)	0.138	−0.417(−1.397~0.562)	0.404
Unwilling to take any risks	0.014(−0.159~0.187)	0.871	−0.545(−1.477~0.387)	0.252
Ambiguous preference	−0.175(−0.372~0.022)	0.081	−1.481(−2.701~-0.261)	0.017
Self-Rated Health Status
Very good*	—	—	—	—
Good	−0.030(−0.115~0.056)	0.498	−0.140(−0.580~0.301)	0.535
General	−0.070(−0.156~0.017)	0.115	−0.374(−0.842~0.094)	0.117
Poor	−0.196(−0.331~-0.062)	0.004	−0.424(−1.354~0.505)	0.371
Very poor	−0.322(−0.598~-0.047)	0.022	−1.679(−3.692~0.334)	0.102
Age
18–24 years old*	—	—	—	—
25–34 years old	0.284(−0.005~0.573)	0.054	1.380(−0.233~2.993)	0.094
35–54 years old	0.249(−0.036~0.534)	0.087	1.712(0.128~3.296)	0.034
55–64 years old	0.008(−0.284~0.300)	0.957	1.053(−0.476~2.581)	0.177
65 years old or older	−0.383(−0.686~-0.081)	0.013	−1.799(−3.646~0.048)	0.056
Family Size
1–3 person(s)*	—	—	—	—
4–6 persons	0.015(−0.049~0.080)	0.646	0.079(−0.261~0.420)	0.647
7 persons or more	−0.023(−0.230~0.185)	0.830	0.382(−0.824~1.589)	0.534
Permanent Place
Urban*	—	—	—	—
Rural	−0.213(−0.296~-0.130)	0.000	−0.703(−1.379~-0.026)	0.042
Social Support Network	0.022(0.013~0.030)	0.000	—	—
Total Family Income
10,000 yuan or lower*	—	—	—	—
10,001– 30,000 yuan	−0.102(−0.239~0.035)	0.143	−0.131(−0.995~0.733)	0.766
30,001– 60,000 yuan	−0.029(−0.149~0.092)	0.641	−0.239(−0.954~0.476)	0.513
60,001– 90,000 yuan	0.044(−0.079~0.167)	0.480	−0.027(−0.752~0.698)	0.942
Higher than 90,000 yuan	0.164(0.050~0.278)	0.005	−0.083(−0.840~0.674)	0.829
Family Savings	−0.134(−0.223~-0.045)	0.003	−0.092(−0.676~0.493)	0.759
Education
Primary schooling or less*	—	—	—	—
Junior High School/Senior High School/Technical Secondary School/Vocational High School	0.220(0.124~0.317)	0.000	0.354(−0.448~1.157)	0.388
College or above	0.383(0.267~0.498)	0.000	0.679(−0.419~1.776)	0.226
Finance Attention
Very concerned*	—	—	—	—
Relatively concerned	0.073(−0.124~0.270)	0.469	0.500(−0.453~1.454)	0.303
General	−0.102(−0.274~0.071)	0.247	−0.021(−0.893~0.851)	0.962
A little concerned	−0.167(−0.340~0.006)	0.059	0.142(−0.789~1.072)	0.765
Never	−0.342(−0.516~-0.168)	0.000	0.170(−0.978~1.318)	0.771
Sex
Female*	—	—	—	—
Male	−0.258(−0.326~-0.191)	0.000	−0.210(−0.842~0.423)	0.516
Social Basic Medical Insurance
No*	—	—	—	—
Enroll	−0.032(−0.159~0.096)	0.627	−0.385(−1.094~0.324)	0.288
Marriage
Other*	—	—	—	—
Married	−0.039(−2.222~-1.433)	0.407	0.069(−0.440~0.577)	0.792
Employment
Unemployment*	—	—	—	—
Employment	0.128(0.041~0.214)	0.004	0.090(−0.467~0.646)	0.752

Note: * stand for reference group

Subgroup analysis results

Table 4 presents the subgroup analysis results. In the motivation dimension, this study incorporated risk preference and self-rated health status. With respect to risk preference, among individuals from households with CHE, those with ambiguous risk preferences were less likely to purchase CHI than those who tended to invest in high-risk, high-return projects were (OR = 0.228, p<0.05). Among those from households without CHE, individuals with ambiguous risk preferences had lower premium expenditures than those who tended to invest in high-risk, high-return projects did (coef.=−1.801, p < 0.05). With respect to self-rated health status, in households without CHE, individuals who rated their health status as very poor were less likely to purchase CHI than those who reported very good health status were (OR = 0.182, p < 0.05). However, among those who did purchase CHI, those with very poor health status had higher premium expenditures (coef.=1.522, p < 0.05). In households with CHE, compared with those with very poor health status, individuals with very poor health status had lower premium expenditures (coef.=-2.749, p<0.05).

Table 4.

Estimated two-part model of commercial health insurance purchase in different groups. (survey-Weighted estimates)

Variables	Purchase(n = 34,534)				LnCommercial Health Insurance Premium (n = 1,142)
	No CHE in household (n = 21,226)		Have CHE in household (n = 13,418)		No CHE in household (n = 923)		Have CHE in household (n = 219)
	OR (95%CI)	P value	OR (95%CI)	P value	Coefficient (95%CI)	P value	Coefficient (95%CI)	P value
Risk Preference
Tend to invest in high-risk, high-return projects*	—	—	—	—	—	—	—	—
Tend to invest in projects with slightly higher risks and slightly higher returns	1.722 (0.928~3.196)	0.085	0.376 (0.109~1.293)	0.120	−0.657 (−1.834~0.520)	0.272	0.297 (−1.927~2.520)	0.790
Tend to invest in projects with average risk and average return	1.420 (0.783~2.575)	0.248	0.439 (0.154~1.246)	0.122	−0.837 (−1.823~0.148)	0.095	−0.997 (−3.244~1.251)	0.378
Tend to invest in projects with slightly lower risk and slightly lower return	1.246 (0.696~2.231)	0.459	0.641 (0.227~1.811)	0.401	−0.527 (−1.522~0.468)	0.298	−0.678 (−2.707~1.351)	0.506
Unwilling to take any risks	0.950 (0.529~1.706)	0.864	0.403 (0.149~1.089)	0.073	−0.801 (−1.819~0.216)	0.122	−0.744 (−2.677~1.188)	0.444
Ambiguous preference	0.617 (0.318~1.200)	0.154	0.228 (0.071~0.735)	0.013	−1.801 (−3.551~-0.052)	0.044	−2.357 (−5.379~0.665)	0.124
Self-rated health status
Very good*	—	—	—	—	—	—	—	—
Good	0.934 (0.723~1.207)	0.600	1.537 (0.763~3.097)	0.229	0.320 (−0.333~0.973)	0.335	−1.279 (−2.562~0.003)	0.051
General	0.806 (0.624~1.040)	0.098	1.183 (0.583~2.398)	0.641	0.082 (−0.648~0.812)	0.825	−0.701 (−1.824~0.422)	0.216
Poor	0.633 (0.348~1.150)	0.133	0.663 (0.296~1.485)	0.317	−0.180 (−1.970~1.610)	0.843	−0.852 (−2.685~0.981)	0.356
Very poor	0.182 (0.043~0.775)	0.021	0.767 (0.213~2.763)	0.685	1.522 (0.322~2.723)	0.013	−2.749 (−5.089~-0.409)	0.022
Age
18–24 years old	—	—	—	—	—	—	—	—
25–34 years old	1.188 (0.557~2.536)	0.655	5.821 (0.664~51.064)	0.112	2.262 (−0.085~4.608)	0.059	−3.100 (−5.331~-0.868)	0.007
35–54 years old	1.281 (0.612~2.678)	0.511	4.294 (0.518~35.613)	0.177	2.767 (0.426~5.107)	0.021	−2.273 (−4.570~0.025)	0.052
55–64 years old	0.712 (0.328~1.547)	0.391	2.063 (0.234~18.220)	0.514	2.548 (0.050~5.046)	0.046	−4.584 (−7.421~-1.747)	0.002
65 years old or older*	0.273 (0.114~0.653)	0.004	0.640 (0.069~5.896)	0.693	0.108 (−2.781~2.997)	0.941	−6.967 (−9.891~-4.044)	0.000
Family Size
1–3 person(s)*	—	—	—	—	—	—	—	—
4–6 persons	1.071 (0.881~1.303)	0.490	0.758 (0.509~1.130)	1.174	0.472 (−0.029~0.972)	0.065	−0.771 (−1.718~0.177)	0.109
7 persons or more	1.230 (0.578~2.619)	0.591	0.811 (0.303~2.173)	0.677	1.402 (0.488~2.315)	0.003	−0.262 (−2.398~1.873)	0.806
Permanent Place
Urban*	—	—	—	—	—	—	—	—
Rural	0.882 (0.642~1.213)	0.440	0.422 (0.249~0.714)	0.001	−0.802 (−1.455~-0.149)	0.016	0.005 (−1.920~1.930)	0.996
Social Support Network	1.038 (1.007~1.070)	0.017	1.067 (1.000~1.138)	0.050	−0.057 (−0.140~0.026)	0.179	−0.005 (−0.143~0.133)	0.941
Total Family Income
10,000 yuan or lower*	—	—	—	—	—	—	—	—
10,001– 30,000 yuan	0.682 (0.417~1.116)	0.128	0.406 (0.202~0.819)	0.012	1.127 (−0.034~2.287)	0.057	−1.085 (−2.805~0.636)	0.212
30,001– 60,000 yuan	0.854 (0.551~1.322)	0.480	0.502 (0.263~0.960)	0.037	0.400 (−0.784~1.584)	0.506	−1.308 (−2.902~0.286)	0.106
60,001– 90,000 yuan	1.252 (0.824~1.903)	0.291	0.446 (0.237~0.842)	0.013	0.141 (−0.992~1.274)	0.807	0.447 (−1.242~2.137	0.598
Higher than 90,000 yuan	1.537 (1.019~2.320)	0.041	0.643 (0.330~1.251)	0.193	−0.058 (−1.095~-0.980)	0.913	−0.323 (−2.298~1.653)	0.745
Family Savings	0.757 (0.566~1.012)	0.060	1.007 (0.622~1.629)	0.978	0.038 (−0.874~0.949)	0.935	−0.484 (−2.008~1.040)	0.527
Education
Primary schooling or less*	—	—	—	—	—	—	—	—
Junior High School/Senior High School/Technical Secondary School/Vocational High School	2.033 (1.389~2.977)	0.000	2.284 (1.311~3.979)	0.004	1.459 (0.026~2.892)	0.046	−1.145 (−2.538~0.257)	0.105
College or above	2.663 (1.739~4.077)	0.000	2.919 (1.470~5.797)	0.002	1.714 (0.198~3.231)	0.027	−0.709 (−2.519~1.102)	0.436
Finance Attention
Very concerned*	—	—	—	—	—	—	—	—
Relatively concerned	1.062 (0.619~1.823)	0.826	5.339 (1.114~25.596)	0.036	0.339 (−0.488~1.166)	0.420	2.893 (−1.772~7.558)	0.219
General	0.688 (0.436~1.087)	0.109	3.291 (0.757~14.305)	0.112	−0.584 (−1.437~0.268)	0.178	1.382 (−3.023~5.787)	0.532
A little concerned	0.565 (0.364~0.879)	0.011	2.465 (0.557~10.917)	0.235	−0.056 (−0.880~0.768)	0.893	1.244 (−3.332~5.819)	0.588
Never	0.435 (0.275~0.688)	0.000	1.133 (0.70~4.763)	0.864	−0.321 (−1.254~0.612)	0.498	2.901 (−1.593~7.395)	0.201
Sex
Female*	—	—	—	—	—	—	—	—
Male	0.609 (0.496~0.749)	0.000	0.582 (0.382~0.886)	0.012	0.169 (−0.446~0.785)	0.589	−0.155 (−0.933~0.622)	0.690
Social Basic Medical Insurance
No*	—	—	—	—	—	—	—	—
Enroll	0.975 (0.663~1.432)	0.896	0.655 (0.287~1.497)	0.316	−0.815 (−1.603~-0.026)	0.043	0.091 (−0.931~0.113)	0.859
Marriage
Other*	—	—	—	—	—	—	—	—
Married	0.886 (0.644~1.220)	0.458	1.120 (0.590~2.125)	0.730	−0.234 (−1.010~0.541)	0.553	0.052 (−1.571~1.674)	0.949
Employment
Unemployment*	—	—	—	—	—	—	—	—
Employment	1.199 (0.870~1.652)	0.267	1.268 (0.792~2.031)	0.322	0.332 (−0.452~1.115)	0.405	−0.480 (−1.664~0.704)	0.420

Note: * stand for reference group

In terms of the opportunity dimension, the factors included were age, family size, permanent place and social support network. With respect to age, among individuals from households without CHE, those aged 65 and older were less likely to purchase CHI than those in the 18–24 reference group were (OR = 0.274, p < 0.05). Moreover, those aged 35–54 and 55–64 who purchased CHI had higher premium expenditures (coef.=2.767, p < 0.05; coef.=3.548, p < 0.05). Among those from households with CHE, all older age groups had lower premium expenditures than did the 18–24 age group did (coef.=-3.100, p < 0.05 for the 25–34 age group; coef.=-2.273, p < 0.05 for the 35–54 age group; coef.=−4.584, p < 0.05 for the 55–64 age group; coef.=-6.967, p < 0.05 for the 65 and older age group). With respect to family size, in households without CHE, individuals from large households (≥7 members) reported higher premium expenditures than those from small households (1–3 members) did (coef.=1.402, p < 0.05). With respect to permanent place, among individuals from households with CHE, compared with urban residents, rural residents were less likely to purchase CHI (OR = 0.422, p = 0.001). Among those without CHE, rural residents who purchased CHI had lower premium expenditures (coef.=−0.802, p < 0.05). The presence of a social support network was positively associated with the purchase behaviour of CHI in both groups—those without CHE (OR = 1.038, p < 0.05) and those with CHE (OR = 1.067, p = 0.05).

In the ability dimension, the incorporated factors include total family income, family savings, education, and financial attention. With respect to total family income, among individuals from households without CHE, those with a total family income of 90,000 yuan or above were more likely to purchase CHI than those with a total family income of 10,000 yuan or lower were (OR = 1.537, p < 0.05). Among those from households with CHE, individuals in all middle-income groups—10,001– 30,000 yuan (OR = 0.406, p < 0.05), 30,001– 60,000 yuan (OR = 0.502, p < 0.05), and 60,001– 90,000 yuan (OR=0.446, p < 0.05)—were less likely to purchase CHI than the lowest-income reference group was. With respect to education, among individuals from households without CHE, better education was positively associated with both the likelihood of purchasing CHI and the amount of premium expenditure. Among those from households with CHE, better education was positively associated with CHI purchase but was not associated with premium expenditure. With respect to financial attention, among individuals from households without CHE, those who reported paying attention to financial information a little (OR = 0.565; p < 0.05) or never (OR = 0.435, p < 0.001) were less likely to purchase CHI than those who were very concerned. Among those from households with CHE, individuals who were relatively concerned about financial information were more likely to purchase CHI than those who were very concerned (OR = 5.339, p < 0.05).

Additionally, the study incorporated variables such as sex, SBMI enrolment, marriage, and employment. Among individuals both from households without CHE and those with CHE, males were less likely to purchase CHI than females were (OR = 0.609, p < 0.001; OR = 0.582, p < 0.05). With respect to SBMI enrolment, among individuals from households without CHE, those enrolled in SBMI had lower premium expenditures for CHI than those not enrolled did (coef.=−0.815, p < 0.05).

Discussion

The incidence of CHE within Chinese households remains a cause for concern. Without effective control, on the one hand, individuals’ health may be jeopardized, and on the other hand, the human capital of society may decrease, and income inequality is likely to increase [81]. Although CHI can transfer the risk of CHE well, most Chinese people are still resistant to it, and there is a considerable gap in effective demand. In addition, the CHI purchase decisions of vulnerable groups who have experienced CHE remain unclear. On the basis of these practical contradictions, this study aimed to explore the potential association between CHE and individual CHI purchase.

This study revealed that the severity of CHE has differential effects on individuals’ CHI decisions: severe CHE was significantly negatively associated with the likelihood of individuals purchasing CHI, whereas moderate CHE was negatively associated with the amount of premium expenditure; in contrast, mild CHE had no statistically significant association with either purchasing behaviour or premium expenditure. From an economic perspective, Keynes’s absolute income hypothesis is relevant. The squeeze that CHE places on disposable income exhibits an intensity gradient. Mild CHE only compresses the space for discretionary consumption. Individuals can retain preventive expenditures such as CHI by cutting back on experiential or entertainment consumption, resulting in no significant impact on their CHI purchase decision. Moderate CHE, however, has a substantial effect on disposable income, forcing individuals to seek a balance in CHI consumption. To reconcile protection needs with affordability, they tend to opt for low-premium, basic protection-oriented products, which in turn leads to a significant reduction in premium expenditure. In contrast, severe CHE directly squeezes nonessential consumption categories, with CHI—regarded as a nonessential good—bearing the brunt of cuts, ultimately inhibiting purchase behaviour. The life cycle hypothesis further explains this heterogeneity: Precautionary savings accumulated during individuals’ peak income years can smooth consumption fluctuations caused by mild to moderate CHE, enabling them to maintain CHI purchases or adjust premium spending. However, when the intensity of medical expenditure reaches the threshold of severe impact, the protective capacity of savings is breached, households lose their ability to smooth consumption, and CHI purchases inevitably suffer [82].

Liquidity constraint theory supplements this explanation from a cash flow perspective. Individual consumption is not only influenced by income but also constrained by the accessibility of short-term cash flow [83]. Mild and moderate CHE can be alleviated through precautionary savings, short-term borrowing, or financial assistance from relatives and friends, ensuring that individuals’ cash liquidity remains unbroken. Consequently, mild or moderate CHE results only in adjustments to premium expenditure, rather than the complete cessation of CHI purchase behaviour. In contrast, severe CHE often plunges households into deep debt, with cash flow fully allocated to repay debt and cover basic living expenses, leading to irreversible liquidity exhaustion. Even if a CHI demand exists, individuals are forced to abandon purchase because of their inability to afford it.

The above three theories support the results of this study, but they overlook psychological factors. Therefore, this study attempted to further explain the results with the prospect theory of Daniel Kahneman and Amos Tversky. Most people are nonrational economic agents and exhibit psychological biases such as the gambler’s fallacy2, overconfidence3, and loss aversion4. These biases affect their judgement of future uncertain events and subsequently alter their decisions and behaviours. Specifically, even after experiencing household CHE, individuals may underestimate the probability of such events recurring in the future because of overconfidence and a luck mentality. Investing part of their income to avoid such low-probability events may be considered risky behaviour by nonrational economic agents. Therefore, influenced by loss aversion, the demand for CHI may be suppressed. Furthermore, even if individuals’ preventive motives strengthen after CHE, they may prefer to save to cope with future uncertainties rather than purchase “nonessential” CHI [84]. Even if they do purchase CHI, they tend to opt for products with lower premiums.

Furthermore, the conservation of resources theory offers insights from a psychological perspective. It emphasizes that individuals’ decision-making capacity relies on the dual support of economic resources and psychological resources. Mild and moderate CHE only partially deplete individuals’ economic resources, whereas psychological resources (such as decision-making energy and risk perception ability) remain without undue strain. Consequently, individuals can still rationally assess the value of CHI, resulting in no changes to purchase behaviour under mild CHE and only adjustments to premium expenditure under moderate CHE. In contrast, severe CHE may trigger the depletion of both types of resources. Consequently, in addition to the exhaustion of economic resources, long-term illness care and debt pressure can lead to decision fatigue and mental exhaustion among individuals, undermining their attention to and judgement of long-term risk planning. As a result, decisions requiring delayed gratification—such as purchasing CHI—are naturally shelved. A study conducted in China discussed the loss and recovery of psychological resources among older adults during public health emergencies, providing potential empirical support for this research [85].

However, certain studies have also demonstrated that there may be a positive association between adverse events and CHI purchases, since these events may modify the public’s insufficient perception of risk, lack of insurance literacy, and probabilistic estimation bias towards uncertain events [44, 46]. This discrepancy in findings may stem from differences in the adverse events examined between this study and previous research or potentially from previous studies adopting a long-term perspective in their analyses.

By incorporating factors across dimensions such as motivation, opportunity, and ability, this study yields several additional findings. Individuals with greater risk preferences exhibited higher premium expenditures on CHI. Regarded as “adventurers”, these individuals tend to adopt high-risk lifestyles and decision-making patterns while demonstrating greater proactivity in weighing risks against returns. They perceive CHI as a risk-hedging tool and acknowledge its potential protective benefits; thus, they are more inclined to purchase mid- to high-end CHI products featuring high coverage limits and broad coverage scopes [86, 87]. Furthermore, these adventurers are more likely to participate in financial markets, hold a larger volume of financial assets, and prefer higher-risk transactions and instruments with potentially greater returns. Consequently, compared with risk-averse individuals, they are more likely to achieve wealth appreciation, which in turn stimulates demand for CHI.

Prospect theory also helps explain this counterintuitive result. Compared with risk-averse individuals, adventurers may be less sensitive to losses and hold different reference points. Specifically, the reference point for risk-averse individuals is their current wealth level without purchasing insurance contracts; thus, CHI premiums are perceived as losses and are not accepted. In contrast, the reference point for adventurers is their future wealth level after they purchase an insurance policy, where premium investments are likely to yield returns for their high-risk behaviours in the long term. Given that higher premium inputs typically correspond to more comprehensive coverage and greater potential benefits—providing these individuals with an enhanced sense of security when engaging in high-risk activities—they are willing to bear higher premium expenditures [88]. Notably, this characteristic is detrimental to the CHI market, as it reflects potential adverse selection.

Individuals with poor self-rated health status were less likely to purchase CHI. Many studies have confirmed that people with average or poor health status are less likely to purchase CHI [89, 90]. This phenomenon can be attributed to health status reflecting individuals’ risk preferences to a certain extent. That is, individuals, motivated by the need to avoid the health and economic risks brought by diseases, actively choose a healthy lifestyle, thus maintaining good physical condition [91]. This motivation may subsequently also drive them to purchase CHI. In addition, the effect heuristic theory provides a reasonable explanation. This theory posits that when individuals have a strong internal affect towards something, the value assigned to it increases [92], after which the demand for individuals to purchase insurance for this item with high subjective value increases accordingly [93]. Specifically, people in good health status may emotionally invest in and attach great importance to their own health. Therefore, under the influence of the effect of heuristic thinking, they actively purchase CHI. The opposite is true for people with poor health status. Furthermore, previous studies have revealed that self-affirmation can promote healthy behaviours by providing a buffer against potential self-threats [94]. Obviously, in this study, better self-rated health status was a form of self-affirmation. Therefore, people demonstrating this form of self-affirmation were likely to engage in more active health promotion behaviours or to make health investments, such as purchasing CHI. Furthermore, among individuals from households with CHE, those with better health status had higher premium expenditures. This may be attributed to their heightened awareness of the uncertainty of health risks following the CHE shock—despite their current good physical condition, they still tended to purchase high-value CHI products.

Notably, among individuals from households without CHE, if those with poor health status did purchase CHI, they often opted for high-premium products. This could be because their perceived health risks drive them to pursue more comprehensive coverage, which points to a practical challenge in the CHI market—adverse selection. However, this could also be attributed to more lenient health restrictions for policyholders in high-premium products. Adverse selection remains a severe issue in China’s CHI market. The loss ratio of CHI has consistently remained high, with some CHI businesses even facing insolvency because of excessive claims, which has seriously undermined the profitability of insurance institutions and the sustainable development of the industry. The high loss ratio is attributed to two main factors: the rapid increase in health expenditure and the persistent prevalence of false health declarations that are difficult to eradicate. These findings indicate that the Chinese government should promptly promote the interconnection of SBMI data and CHI data to mitigate information asymmetry. Moreover, insurance institutions should leverage big data and artificial intelligence to increase the accuracy of underwriting and claims settlement while accelerating the implementation of risk-based tiered pricing [95].

Compared with younger adults, older adults were less likely to purchase CHI, while middle-aged adults had higher premium expenditures on CHI. These findings are similar to the conclusions drawn by Wan et al. [96], thereby providing mutual corroboration. This age-based differentiation may be related to the insurance institutions’ underwriting threshold settings. Currently, most CHI products in China impose age restrictions and stringent health declaration requirements, making them inaccessible to elderly people [97]. Moreover, the intensifying population ageing confronting Chinese society is another contributing factor. The growing elderly population has increased the support burden on younger adults, posing challenges to affordability and further constraining their premium expenditures. Additionally, against the backdrop of ageing, the prevalence of chronic diseases among older adults has increased. Even if older adults have a strong demand for CHI, the existence of entry barriers results in their persistently low purchase rate. Middle-aged adults fall within the peak range of CHI purchases, primarily because of two factors: First, their physical conditions have not yet entered the high-risk phase of old age, and second, they are supported by stable and sound economic circumstances.

Against the backdrop of population ageing, the solidification of this outcome is detrimental to universal medical security coverage. As a group with high medical needs, elderly people exhibit a rigid and diversified growth trend in their medical expenditures. If the accessibility of CHI for this group remains insufficient in the long term, their high medical expenditures will be borne mainly by the SBMI fund and individuals directly. For the SBMI, this is likely to lead to fund depletion, narrowed reimbursement scope, and severely restricted security functions [98]. Consequently, residents’ sense of gain will diminish, which may ultimately result in insufficient financing. The construction of China’s multilevel medical security system may thus face severe challenges.

Individuals from extended families and those without CHE had higher premium expenditures, which may stem from two underlying mechanisms. The first is increased risk exposure. A larger family size implies a higher potential incidence of health risk events, intensifying individuals’ demand for the protective coverage of CHI and thus motivating them to pay higher premiums. Second, extended families may either opt for high-value CHI products through group coverage arrangements or possess the economic capacity to bear higher premiums because of their higher total household income.

For both individuals with CHE and those without CHE, a social support network promoted CHI purchase, yet its impact on premium expenditure was not statistically significant. This heterogeneous effect can be explained through three interconnected mechanisms. First, the information-sharing mechanism plays a pivotal role in social support networks. Frequent gift exchange typically signifies broader and closer social ties, through which individuals may more frequently access CHI product information, obtain enrolment experience sharing, and even receive recommendations from trusted agents. This support effectively reduces the information search costs and cognitive barriers associated with evaluating CHI contracts—key deterrents to CHI purchase, particularly for individuals with limited financial literacy or risk awareness. By streamlining the decision-making process, the social support network lowers the threshold for initiating CHI purchase. Second, gift exchange reflects informal risk-sharing arrangements and social capital accumulation. In the Chinese context, gift exchanges inherently operate as long-term reciprocal social contracts. When confronting health shocks, this network can provide timely financial assistance and emotional support, acting as an implicit safety net. This informal risk mitigation mechanism enhances individuals’ resilience to future uncertainties, fostering greater confidence in investing in CHI—designed to address long-term, high-value risks—rather than relying solely on short-term precautionary savings. The accumulated social capital thus incentivizes proactive risk management through formal insurance channels. Third, group effects within social support networks shape individual decision-making. When friends, relatives, or community members widely hold CHI, individuals may be motivated to purchase CHI to conform to group norms and maintain social integration. This bandwagon effect strengthens the intention to purchase, as CHI ownership becomes normalized within the social circle.

In contrast, the choice of premium expenditures was driven more by individual-specific factors, diluting the marginal impact of the social support network. In other words, social support networks effectively push individuals above the threshold of CHI adoption by addressing information asymmetries, enhancing risk resilience, and leveraging group norms. However, the specific premium amount—which reflects coverage depth and product tier—depended primarily on personal resources (e.g., income, savings) and risk attitudes (e.g., risk preference, risk perception). These individual characteristics exerted a more direct and dominant influence on premium decisions than the social support network did, leading to the insignificant effect observed on premium expenditure.

With respect to sex, this study revealed that compared with male sex, female sex was associated with a greater likelihood of purchasing CHI. This phenomenon can be explained by the social roles played by women. In traditional culture, women are expected to take on more household care responsibilities, which may prompt them to pay more attention to the potential risk and well-being of themselves and their families, thereby encouraging them to purchase CHI [99]. This tendency can also be attributed to the increased economic independence or authority in households headed by women in recent years, providing them with a stronger material base and autonomy to purchase CHI [100].

Regardless of whether CHE occurs, individuals with higher educational attainment were more likely to purchase CHI. To a large extent, this association may be mediated by health literacy and financial literacy. Evidence has indicated that education is significantly positively correlated with both health literacy [101] and financial literacy [102], while higher levels of these two literacies positively affect CHI purchases. Such effects occur because, under the influence of these two literacies, individuals have a clearer perception of disease risks and burdens, demonstrate greater initiative in risk management, and possess a deeper understanding of the risk transfer function of CHI, thereby reducing resistance and distrust towards CHI products [40, 73]. However, for individuals from households with CHE, education no longer had a significant positive effect on their premium expenditure. From the perspective of economic resources, the shock induced by CHE has severely depleted wealth. Although individuals with higher levels of education possess health literacy and financial literacy, enabling them to clearly recognize the value of CHI, the exhaustion of economic resources prevents them from translating their cognitive advantages into higher premium investments.

Urban residents were more likely to purchase CHI, and this result may be related to the urban–rural disparity in CHI product accessibility. In fact, insurance institutions in China offer different CHI products in urban and rural areas. Urban residents have access to a richer range of product options (e.g., critical illness insurance, outpatient insurance, special drug insurance and million-dollar medical insurance). In contrast, products launched in rural areas offer mostly short-term medical insurance with narrow coverage and low insured amounts, making it difficult for such products to meet the actual risk needs of rural residents. Some products even directly list remote rural areas as restricted regions. In addition, the setup of insurance purchase channels varies significantly between urban and rural areas. A diversified insurance network has been established in urban regions, which greatly enhances the convenience of CHI purchases for urban residents. In rural areas, however, offline branches are extremely scarce, with a lack of offline personnel to assist in interpreting clauses, handling insurance applications and claims settlements. Moreover, online insurance purchases are constrained by poor network conditions, resulting in the formation of barriers to purchasing. Insurance institutions differ in their product marketing efforts between urban and rural areas, characterized by intensive marketing in urban areas and a lack of marketing in rural areas. Studies have shown that when the development of the insurance industry bridges the urban–rural gap in product accessibility, the demand for CHI among rural residents increases significantly [8].

Although no association was identified between individuals’ CHI purchase behaviour and their enrolment in SBMI, those enrolled in SBMI had significantly lower premium expenditures. This finding partly reflects the competitive relationship between these forms of insurance [103, 104]. Through continuous improvement, China’s multilevel medical security system has not only essentially achieved universal medical security coverage but has also made remarkable progress in improving residents’ health conditions and alleviating the burden of medical expenses. However, the synergistic effect between CHI and SBMI has not been fully exerted, largely because of the lack of differentiated development between them—CHI has not provided supplementary coverage for the areas not covered by SBMI. Owing to its positioning as ensuring coverage for basic medical needs, SBMI inherently has limitations, including insufficient coverage for expenses outside the reimbursement catalogue, inadequate maximum reimbursement limits to cope with high medical expenses of serious illnesses, relatively low coverage content, and poor user experience with traditional reimbursement processes. In response, CHI needs to address each of these limitations to enhance residents’ sense of gain. Specifically, CHI products should continuously expand coverage boundaries and incorporate high-value special drugs, advanced medical technologies, and imported medical consumables outside the SBMI catalogue to fill coverage gaps. Currently, the Chinese government has initiated the development of an innovative drug catalogue for CHI, which will become a key driver for promoting the differentiated development of the two systems. Moreover, insurance institutions need to set deductibles and claim limits on a scientific basis, focusing on covering the remaining high medical expenses after SBMI reimbursement to form a joint guarantee force. In addition, insurance institutions should strengthen product attractiveness by increasing the supply of value-added services, such as free physical examinations, cancer screenings, chronic disease management, and traditional Chinese medicine conditioning, to cater to the public’s demand for health protection. Finally, the construction of an integrated platform for SBMI and CHI is crucial. It plays a key role in realizing one-stop settlement, improving claim efficiency, and optimizing residents’ medical treatment and claim experience.

This study also revealed that regardless of CHE occurrence, financial attention may be positively associated with CHI purchase. In the insurance transaction market, information asymmetry exists between policyholders and insurers. Policyholders often fail to understand the relevant information in insurance contracts, which seriously affects their willingness to purchase insurance. Alternatively, the insurance products they purchase may be seriously inconsistent with their own needs. When residents increase their attention to financial and economic information, they will learn more about insurance products and be exposed to more real insurance cases [105]. As a result, they can obtain a better understanding of the role of CHI.

Finally, through subgroup regression analyses, this study revealed that the CHI purchase behaviour of individuals from households with CHE was influenced primarily by factors in the ability dimension. Despite the CHE shock weakening the positive association between economic capacity and CHI demand, education and financial attention still had significant stimulating effects. This finding implies that enhancing cognitive capabilities is crucial for fostering CHI demand and expanding coverage for vulnerable groups. With respect to premium expenditure, only self-rated health status and age remained influential factors, with both having a significantly negative impact. This finding indicates that the most vulnerable groups tend to purchase CHI products with the lowest coverage, as reflected by their minimal premium expenditures. However, it remains unclear whether such product configurations are sufficient to help these CHE-impacted groups escape the current or future poverty risks induced by health shocks.

For individuals without CHE, their CHI purchase behaviour and premium expenditure were shaped by the combined effects of multidimensional factors, with two key findings worthy of attention. First, older adults and individuals with poor health status are disadvantaged because of market constraints, highlighting the need to develop specialized CHI products tailored to these populations. Second, although family size and SBMI enrolment did not stimulate purchase behaviour, they significantly influenced premium expenditure. These findings suggest that family group CHI products hold substantial promotion value and that promoting the differentiated development of SBMI and CHI constitutes a critical measure for advancing the CHI market.

Some limitations of the present study should be acknowledged. First, there is still room for expansion of the factors. For example, factors such as the herd effect, overconfidence and loss aversion can be added to the motivation dimension; factors such as insurance company marketing and incentive policies can be added to the opportunity dimension; and factors such as insurance literacy can be added to the ability dimension. Furthermore, although this study controlled for a wide range of observed factors, unobserved confounding variables may still exist. Second, owing to the cross-sectional nature of the study, causal relationships between variables cannot be inferred. On the basis of previous research exploring the relationship between CHI and CHE, reverse causality is plausible—specifically, the failure to purchase CHI may lead to the occurrence of CHE. However, this study defined CHI purchase at the household head level and CHE at the household level, a design that slightly mitigates the interference of reverse causality. Nevertheless, future research should construct panel data to further identify the dynamic causal relationship between CHE and CHI purchase decisions. In addition, it would be valuable to use longitudinal studies to analyse the long-term effects of CHE on CHI purchase and the heterogeneity of such effects. For instance, studies could examine differences in post-CHE purchase decisions between groups that had purchased CHI before the occurrence of CHE and those that had not. Third, the results of this study all relied on CHFS data, and these data may have deficiencies such as respondents’ recall bias and incorrect reporting.

Conclusions

On the basis of the research findings, the following are the policy implications of this paper. Importance should be attached to the potential adverse spillover effect of CHE on individuals’ CHI purchase. Controlling CHE must serve as a long-term focus for the government. The government should further deepen the reform of the health care system and improve cost control measures in the three systems of pharmaceuticals, health care services, and medical insurance. This is a fundamental strategy to alleviate CHE. Furthermore, the government should monitor at-risk populations. Leveraging multidimensional data—including medical insurance claims records, health check-up results, and household economic status—the government should implement hierarchical management for groups vulnerable to CHE, including individuals with poor health status, low income, and elderly and fragile individuals whose households have already experienced CHE. It is essential to update risk levels and CHI needs in real time and to provide targeted CHI subsidies to these groups. Such support not only prevents them from falling into poverty but also sustains their ability and demand to purchase CHI.

Equally critical is that the government and insurance institutions should be dedicated to the joint mobilization of residents’ motivation to purchase CHI. The government and insurance institutions should correct the overly conservative risk tendencies of the public. They should promote typical cases through channels such as the internet and communities and moderately convey risk information such as the probability of occurrence and serious harm of CHE to the public to help them establish an accurate view of risks and encourage them to actively seek external protection to diversify risk. The government should also take advantage of the potential promoting effect of financial attention on individuals’ CHI purchase. It is necessary to strengthen financial education, regularly hold financial literacy training for the public and vigorously popularize financial knowledge to nudge residents into a correct understanding of risk and CHI.

Furthermore, the government should urge and assist insurance institutions in overcoming adverse selection, ensuring that vulnerable groups excluded from CHI contracts can also obtain adequate protection. The elderly population and rural residents benefit from the improved accessibility of CHI products. For elderly people, insurance institutions should refine product design and develop inclusive bundled products for seniors, such as long-term care insurance combined with medical insurance, while optimizing insurance purchase channels to be age friendly. For rural residents, insurance institutions should continue to build a three-pronged sinking network covering marketing, products, and services; accelerate the deployment of professional teams; and assist residents in purchasing and utilizing CHI. For individuals with poor health status, insurance institutions should innovate CHI products that cover preexisting conditions while standardizing health declaration procedures to ensure the sustainable operation of such products. For other populations, insurance institutions should provide CHI products tailored to different scenarios and needs—such as customized products or health management-oriented products—to fully tap into CHI demand.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CHE

Catastrophic health expenditure

WHO

World Health Organization

SBMI

Social Basic Medical Insurance

CHI

Commercial health insurance

SOR Model

Stimulus–organism–response model

MOAModel

Motivation–opportunity–ability model

CHFS

China Household Finance Survey

Author contributions

YL and CL conceived and designed this study and critically reviewed the article. SS performed the statistical analysis and drafted the manuscript. All authors interpreted data, contributed to critical revisions, and approved the final version of the article.

Funding

Not applicable.

Data availability

The datasets generated and analysed during the current study are available in the CHFS repository, [https://chfs.swufe.edu.cn/]. All data are available upon application.

Declarations

Ethics approval and consent to participate

For this research, we used a publicly available secondary data set with all individual identifiers removed prior to making the data set available publicly. No ethical approval was required due to the type and nature of data set used.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.