The association between health literacy and out-of-pocket health expenditure: exploring heterogeneity by chronic disease in Zhejiang, China Authors: Xuehai Zhang, Xinyu Wen, Yunfang Zhou, Mingyu Huang, Dingming Yao, Xiujing Hu, Qiaohong Lv, Chun Chen, Yuya Song Published: 2026 ABSTRACT BackgroundHealth literacy has been increasingly emphasized as a critical factor shaping healthcare behaviors and costs. However, empirical evidence on its association with out-of-pocket (OOP) health expenditure remains complex, and little is known about how this relationship varies by chronic disease status. This study aimed to identify distinct health literacy profiles within a large Chinese population, examine their associations with OOP expenditure, and explore how these associations vary by chronic disease status.MethodsA cross-sectional study was conducted using data from a large-scale health literacy survey of 56,863 residents aged 15–69 in Zhejiang, China. Latent profile analysis (LPA) identified health literacy subgroups using the validated Chinese Health Literacy Scale. A Two-Part Model was employed to decompose expenditure into the probability of incurring costs and expenditure intensity. Stratified analyses were conducted to examine variation across chronic disease status.ResultsLPA identified three health literacy profiles with weighted proportions of: Relatively High (61.45%), Moderate (26.54%), and Low (12.01%). The analysis revealed an inverted U-shaped trajectory. The Moderate group exhibited significantly higher total OOP expenditure than the Low group (AME = 554.71 CNY, P < 0.05), driven primarily by elevated expenditure intensity. Conversely, the Relatively High group showed no significant difference (P > 0.05), indicating expenditure levels comparable to the baseline. Interaction analysis suggested potential heterogeneity by chronic disease status (P for interaction = 0.053, in Part 2), which prompted exploratory stratified analyses. Among those with one chronic disease, the Relatively High group showed higher probability of incurring expenditure (Part 1, P < 0.01) but significantly lower expenditure intensity (Part 2, P < 0.05). However, this association was not statistically significant among those with multimorbidity.ConclusionsThe findings indicate that health literacy is associated with expenditure in a non-linear pattern. While higher literacy maintains expenditure at baseline levels, exploratory analyses suggest this protective effect may be attenuated under high disease burden. To enhance both equity and efficiency, policy strategies should focus on cultivating decision-making capabilities for appropriate healthcare utilization among general populations and strengthening financial protection for those with complex needs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12939-026-02913-z. Introduction Health inequity remains a pressing global concern with significant disparities in out-of-pocket (OOP) health expenditure serving as a key indicator of systemic inequalities in healthcare access and affordability [1]. Despite widespread efforts to enhance financial protection, these disparities persist worldwide, reflecting deep-rooted variations in health system structure and performance [2]. In China, although health insurance coverage had reached over 95% of the population by 2024 [3], OOP health expenditures still account for a substantial share of total health spending and disproportionately burden socioeconomically disadvantaged populations [4, 5]. Regional disparities exacerbate this issue, with residents in less developed areas facing weaker healthcare infrastructure and more limited financial protection [6]. While social and structural determinants such as insurance status and geographic access have been extensively examined, growing evidence underscores the importance of individual-level determinants, especially health literacy, in shaping health expenditures [7, 8]. Health literacy is the capacity to access, comprehend, appraise, and apply health information [9]. It has become a key factor influencing healthcare utilization and costs. The prevailing body of empirical evidence suggests an inverse relationship where higher health literacy is associated with lower OOP expenditure, often attributed to more efficient navigation of the healthcare system and better self-management [10–12]. This assumption has shaped much of the global health policy discourse, where enhancing health literacy is seen as a pathway to reducing medical spending. However, this seemingly straightforward relationship may conceal a different dynamic. A separate stream of research suggests that individuals with higher health literacy are more likely to engage in proactive health-seeking behaviors, such as utilizing more preventive services [13, 14]. These findings suggest that, under certain conditions, improved health literacy could actually lead to greater medical spending rather than less. Examining the contexts in which this relationship varies can provide important insights for theory and policy. The relationship between health literacy and OOP health expenditure is shaped not only by individuals’ informational capacity, but also by their health needs. According to Andersen’s behavioral model of health service use [15], health literacy can be conceptualized as a predisposing factor that shapes individuals’ healthcare decision-making capacity. However, actual service utilization and associated OOP health expenditure are also driven by need factors. Among various health-related need factors, chronic diseases deserve particular attention. Unlike short-term illnesses, they involve continuous care and complex management, which increase healthcare use and financial burden [16]. This is especially evident as chronic diseases, once largely confined to older adults, are increasingly prevalent among younger populations globally [17, 18]. Chronic diseases now account for 68% of disability-adjusted life years, with estimated economic costs projected to reach $47 trillion globally by 2030 [19]. Individuals with chronic conditions face greater challenges in disease management and bear heavier healthcare costs [20]. As the population ages, the prevalence of chronic diseases continues to rise, necessitating long-term treatment and contributing to sustained increases in OOP health expenditure [21, 22]. Moreover, the complexity of managing multiple chronic conditions may interact with individual’s health literacy in influencing their healthcare behaviors and financial burden [23–25]. Yet little is known about whether the association between health literacy and health expenditure differs by chronic disease status. Furthermore, much of the existing research on health literacy has treated it as a unidimensional construct [26, 27]; specifically, defining adequate health literacy as achieving ≥ 80% of the total score [28]. The traditional variable-centered approaches (e.g., summed indices) often rely on such arbitrary cut-off points to categorize individuals [29], potentially masking the complexity of health literacy patterns. To address this limitation, our study employs Latent Profile Analysis (LPA). Unlike arbitrary cut-off scores, LPA is a person-centered statistical method that identifies distinct subgroups based on response patterns across multiple indicator variables [30]. Crucially, LPA allowed the data structure itself to reveal distinct intermediate groups without imposing arbitrary thresholds. Consequently, this approach allows us to uncover distinct gradients of health literacy, providing a more nuanced basis for equity-oriented policy targeting. To address these critical gaps in the literature, this study leverages a large, population-based dataset (n = 56,863) from Zhejiang Province, China. As one of the most economically dynamic regions with relatively high levels of health literacy nationwide [31, 32], Zhejiang serves as a representative setting for examining how socioeconomic advancement reshapes the relationship between health literacy and health expenditure. The province’s experience offers a meaningful reference point for other developing regions that are undergoing similar transitions. We aim to identify distinct health literacy profiles among Chinese residents and explore whether the association between these profiles and OOP health expenditure varies across different levels of chronic disease status. This nuanced analysis will provide crucial insights for designing more equitable and effective health policies. Methods Data sources and sample The data for this study were drawn from the 2024 Zhejiang Provincial Health Literacy Surveillance Survey, a large, population-based assessment conducted by Zhejiang Provincial Center for Disease Control and Prevention. The survey employed the standardized Health Literacy and Tobacco Use Surveillance Questionnaire approved by the Chinese Health Education Center, comprising three components: personal characteristics, health literacy, and tobacco use status. The survey employed a stratified multistage probability-proportional-to-size (PPS) sampling design to recruit a representative sample of permanent residents aged 15–69 years who had resided in the selected survey areas for at least six months. Written informed consent was obtained from all individuals prior to their participation. The initial sample included 56,863 participants. Rigorous on-site quality control procedures were implemented to ensure data completeness, resulting in no missing values for the variables analyzed in this study. To ensure population representativeness, a composite final sampling weight (Wf) was calculated through three sequential adjustments, accounting for the complex survey design. First, a base sampling weight (Ws) was computed as the inverse of the selection probability across all five sampling stages (county/district, township/street, village/community, household, and individual), incorporating local household registry data. Second, a non-response adjustment factor (Wnr) was applied at the county level, calculated as the ratio of the planned sample size to the actual number of respondents in each county. Third, a post-stratification adjustment factor (Wpo) was applied to calibrate the weighted sample to the age-sex structure of the general population using 2024 official population statistics. The final weight was calculated as: Wf = Ws × Wnr × Wpo. For all analyses, primary sampling units were defined as unique combinations of county and township codes, and stratification was based on county. This svyset specification was applied across all models. Health literacy Participants’ health literacy was assessed using the validated Chinese Health Literacy Scale, which evaluates individuals’ ability to understand and apply health-related information. The scale comprises 50 questions totaling 66 points, structured across six core dimensions: Scientific View of Health (8 questions, 11 points), Infectious Disease Literacy (6 questions, 7 points), Chronic Disease Literacy (9 questions, 12 points), Safety and First-Aid Literacy (10 questions, 14 points), Medical Care Literacy (11 questions, 14 points), and Health Information Literacy (6 questions, 8 points). Items were presented in several formats, including true/false questions (1 point per correct response), single-choice questions (1 point per correct answer), and multiple-choice questions (2 points awarded only for selecting all correct options without any incorrect selections). OOP health expenditure A respondent was asked about the OOP health expenditure incurred within the past 12 months, including spending on medicine but excluding any amount that had been or would be reimbursed by health insurance. The primary outcome was defined as the annual OOP health expenditure, derived from respondents’ self-reported total personal healthcare spending over the preceding 12 months. Covariates As covariates, we included a comprehensive set of sociodemographic and health-related factors based on prior research [26, 33]. By controlling for these variables, we aim to estimate the independent association of health literacy on expenditure adjusting for health need. Sociodemographic variables included sex (male, female), age group (15–29, 30–44, 45–59, and 60–69 years), ethnicity (Han, others), household registration (non-local, local), and urbanicity (urban, rural). Marital status was dichotomized into married and unmarried, with the latter comprising participants who were never married, divorced, or widowed. Socioeconomic characteristics covered occupation (public sector, student, agricultural worker, industrial worker, private sector, and others), per capita household income, and education level. Per capita household income was calculated by dividing the total annual household income by the number of household members and categorized into quartiles (Q1: ≤13,333; Q2: 13,334–25,000; Q3: 25,001–50,000; Q4: >50,000 CNY). Education level was categorized into four groups based on the highest level of schooling completed: (1) no formal education (including those who were illiterate or semiliterate); (2) primary school; (3) secondary education (comprising junior high school, senior high school, and vocational/technical secondary school); and (4) higher education (including junior college, bachelor’s degree, and postgraduate degrees). Health-related variables included the number of chronic diseases (0, 1, or ≥ 2), smoking status (current smoker, former smoker, and never smoker), and self-rated health (good, fair, and poor). Statistical analysis All statistical analyses were conducted using Stata, version 17.0. Participant characteristics were summarized using descriptive statistics, with categorical variables presented as frequencies and percentages. Differences in sociodemographic characteristics across health literacy profiles were assessed using the Rao-Scott X2 test to account for the complex survey design. All estimates, including regression coefficients and marginal effects, were derived using survey-weighted estimation procedures to ensure population representativeness. Latent profile analysis LPA was performed to identify distinct health literacy patterns based on the score rates of the six dimensions. A series of Gaussian mixture models with one to five profiles were estimated using standardized weights to prevent entropy inflation during model fitting. Model selection was based on fit indices including Log-likelihood (ll), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Entropy. Classification precision was further evaluated using Average Posterior Probabilities (AvePP) to ensure robust group assignment. Two-part model To investigate the association between health literacy profiles and OOP expenditure while accounting for the complex survey design, we employed a Two-Part Model. Preliminary analysis revealed a substantial mass at zero, with 18.10% (N = 10,290) of participants reporting zero OOP expenditure. Furthermore, the positive expenditure values exhibited a severe right-skewed distribution, ranging from 1 to 1,000,000 CNY. Under these conditions, standard Ordinary Least Squares (OLS) regression may yield biased and less precise estimates [34, 35]. The Two-Part Model addresses these challenges by decomposing the expenditure into two distinct processes. The first part estimated the probability of incurring any expense using a Probit model. The second part, applied to those with positive spending, estimated the expenditure intensity using a Generalized Linear Model (GLM) with a Gamma distribution and a log link. The Gamma specification is appropriate for right-skewed, continuous positive outcomes and was empirically supported by Park’s test (Coef. = 2.00, 95% CI: 1.97, 2.03), which favored it over a Gaussian model. The overall expected OOP expenditure was obtained by combining both parts as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:E(Y\mid\:X)=Pr(Y>0\mid\:X)\times\:E(Y\mid\:Y>0,X)$$\end{document}, and the overall effects were summarized using Average Marginal Effects (AME) on the unconditional expected expenditure in Chinese Yuan (CNY). Interaction and robustness checks Formal statistical tests for interaction were performed by including an interaction term between health literacy profiles and chronic disease status. In the expenditure intensity model (Part 2), the interaction term was marginally significant (P = 0.053). Therefore, exploratory subgroup analyses were performed by stratifying the full Two-Part Model by the number of chronic diseases (0, 1, or ≥ 2). To ensure the robustness of the findings, two sensitivity analyses were performed. First, we assessed variable specification by replacing the categorical profiles with a continuous, standardized health literacy Z-score within the same Two-Part Model framework. Second, to test the assumption of linearity and explore potential non-linear relationships, we extended the continuous specification by adding a quadratic term of health literacy score and re-estimated the Two-Part Model. The significance of the quadratic term was formally tested, and the shape of the association was visualized by plotting the model-predicted marginal means across the range of health literacy scores, derived from the combined Two-Part Model predictions. Statistical significance in this study is indicated by P < 0.05. Results Participant characteristics A total of 56,863 participants were included in the study (unweighted demographic details are provided in Supplementary Table S1). To ensure the findings were representative of the population aged 15–69 years in Zhejiang, China, a composite weight was applied. Based on the weighted analysis (Table 1), the study population represented a population of approximately 51.37 million. The weighted population structure was characterized by a slight male predominance (52.63%), a predominance of Han ethnicity (98.70%), and a middle-aged demographic, with the largest proportion aged 45–59 years (33.11%), followed by those aged 30–44 years (32.15%). Geographically, more than half of the population (53.26%) resided in rural areas, and the vast majority (90.58%) held local household registration; regarding marital status, the majority of the population was married (76.09%). In terms of socioeconomic status, the population exhibited relatively high educational attainment, with 35.06% holding a higher education degree and 48.37% completing secondary education. The occupational distribution was diverse, largest proportion being private sector employees (29.85%) and agricultural workers (18.98%). Regarding per capita household income, the distribution was somewhat skewed towards the middle-to-upper range, with 32.41% falling into the third quartile (25,001–50,000 CNY). Health-related indicators showed that the majority of the population was in good health. Specifically, 81.20% reported no chronic diseases, and 74.29% were never smokers. Self-rated health was generally positive, with 71.78% rating their health as “good” and only 2.63% as “poor”. Latent profile analysis LPA was conducted to identify distinct subgroups of respondents based on their health literacy score rates across six domains. Models specifying one to five latent profiles were estimated, and model fit was assessed using the Log-likelihood (ll), Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Entropy and Average Posterior Probabilities (AvePP). Fit statistics for all candidate models are presented in Supplementary Table S2. As the number of latent profiles increased, both AIC and BIC values decreased monotonically, indicating incremental improvements in model fit. However, entropy values declined with the inclusion of additional profiles, suggesting reduced classification precision. Although the five-profile solution yielded the lowest AIC and BIC values, the decrease in these information criteria became less pronounced after the three-profile solution. Furthermore, the entropy for the five-profile solution (0.83) was lower than that observed in the two-profile (0.92) and three-profile (0.88) solutions. Considering the balance between model fit (AIC/BIC), classification accuracy (Entropy > 0.80), parsimony, and substantive interpretability, the three-profile solution was selected as the optimal model. This solution demonstrated high classification quality with an entropy value of 0.88 and average posterior probabilities ranging from 0.913 to 0.967, indicating clear and reliable separation among the latent classes. Based on these fit statistics and the interpretability of the profiles, the three-profile model was selected as the optimal solution. The three latent profiles were named based on their score rates across the six health literacy domains (see Fig. 1). Profile 1 demonstrated generally high scores in every domain and was labeled the “Relatively high health literacy” group. Profile 2 exhibited moderate scores across all domains and was therefore named the “Moderate health literacy” group. Finally, Profile 3, with significantly lower average scores across all domains, was designated as the “ Low health literacy” group. In the unweighted analytical sample, the Low, Moderate, and Relatively High Health Literacy profiles accounted for 17.38%, 31.71%, and 50.92% of participants, respectively. After applying survey weights to represent the population aged 15–69 in Zhejiang, the distribution shifted, with the Relatively High Health Literacy group comprising the majority (61.45%), followed by the Moderate (26.54%) and Low (12.01%) groups (Table 1). Fig. 1Profile plot of mean score rates across six health literacy domains by latent profile Profile plot of mean score rates across six health literacy domains by latent profile Characteristics of participants across different latent profile memberships Table 1 presents the demographic, socioeconomic, and health-related characteristics of the study population, weighted to represent the adult residents of Zhejiang Province (estimated N = 51,367,126). Statistically significant differences were observed across the three profiles for all examined variables, except ethnicity (P = 0.133). Participants in the Relatively High Health Literacy profile were predominantly younger, with the largest proportion aged 30–44 years (40.90%). This group exhibited the highest educational attainment, with half of the individuals having completed higher education (50.80%), and was more likely to be employed in the private (36.25%) and public sectors (19.44%). They were also disproportionately represented in the highest income quartile (25.68%), resided in urban areas (49.03%), and held non-local household registration (9.91%). In terms of health status, this profile demonstrated the most favorable outcomes, including the highest proportion of individuals without chronic diseases (87.77%), the highest prevalence of good self-rated health (76.26%), and the lowest rate of current smoking (16.13%). In contrast, participants classified into the Low Health Literacy profile were substantially older, predominantly comprised of middle-aged and older adults (45–69 years), with 45.97% aged 45–59 and 39.66% aged 60–69. This group had markedly lower educational attainment, with the majority having completed primary school or less (combined 52.33%), and was disproportionately engaged in agricultural work (43.20%). They were more likely to belong to the lowest per capita household income quartile (34.38%) and to reside in rural areas (58.18%) with local household registration. Health disadvantages were pronounced in this profile, including the highest prevalence of multimorbidity (7.70%), relatively higher rates of current smoking (21.76%), and the largest proportion of individuals reporting poor or fair self-rated health (combined 39.53%). The Moderate Health Literacy profile generally displayed characteristics intermediate between the Low Health Literacy and Relatively High Health Literacy profiles across most demographic, socioeconomic, and health indicators, except for current smoking rates, where the prevalence was slightly higher than in the Low Health Literacy group. Table 1Weighted descriptive statistics of the study populationVariableTotal(Weighted N = 51,367,126)Low HL(Weighted N = 6,168,456)Moderate HL(Weighted N = 13,631,418)Relatively High HL(Weighted N = 31,567,251)P-value Sex < 0.001 Male27,036,652 (52.63)2,931,937 (47.53)7,184,809 (52.71)16,919,906 (53.60) Female24,330,473 (47.37)3,236,520 (52.47)6,446,609 (47.29)14,647,345 (46.40) Age group < 0.001 15–29 years10,221,238 (19.90)276,109 (4.48)1,459,638 (10.71)8,485,491 (26.88) 30–44 years16,517,066 (32.15)610,089 (9.89)2,995,148 (21.97)12,911,829 (40.90) 45–59 years17,008,351 (33.11)2,835,688 (45.97)5,875,795 (43.10)8,296,869 (26.28) 60–69 years7,620,470 (14.84)2,446,570 (39.66)3,300,837 (24.21)1,873,063 (5.93) Ethnicity 0.133 Han50,698,041 (98.70)6,074,831 (98.48)13,477,091 (98.87)31,146,120 (98.67) Ethnic minorities669,085 (1.30)93,625 (1.52)154,328 (1.13)421,132 (1.33) Household registration 0.042 Non-local4,841,217 (9.42)504,394 (8.18)1,209,399 (8.87)3,127,424 (9.91) Local46,525,908 (90.58)5,664,062 (91.82)12,422,019 (91.13)28,439,827 (90.09) Urbanicity < 0.001 Urban area24,008,064 (46.74)2,579,460 (41.82)5,951,991 (43.66)15,476,613 (49.03) Rural area27,359,062 (53.26)3,588,996 (58.18)7,679,427 (56.34)16,090,638 (50.97) Occupation < 0.001 Public sector7,357,998 (14.32)177,126 (2.87)1,043,379 (7.65)6,137,493 (19.44) Student3,653,997 (7.11)93,633 (1.52)588,622 (4.32)2,971,743 (9.41) Agricultural worker9,750,794 (18.98)2,664,965 (43.20)3,877,470 (28.45)3,208,359 (10.16) Industrial worker8,491,215 (16.53)1,427,659 (23.14)2,926,470 (21.47)4,137,086 (13.11) Private sector15,332,528 (29.85)870,328 (14.11)3,018,817 (22.15)11,443,384 (36.25) Others6,780,593 (13.20)934,746 (15.15)2,176,660 (15.97)3,669,186 (11.62) Marital status < 0.001 Unmarried12,282,254 (23.91)831,143 (13.47)2,322,208 (17.04)9,128,904 (28.92) Married39,084,871 (76.09)5,337,314 (86.53)11,309,211 (82.96)22,438,347 (71.08) Education level < 0.001 No formal education1,875,164 (3.65)950,583 (15.41)781,606 (5.73)142,976 (0.45) Primary school6,636,378 (12.92)2,277,274 (36.92)2,989,657 (21.93)1,369,447 (4.34) Secondary education24,844,028 (48.37)2,737,597 (44.38)8,088,369 (59.34)14,018,062 (44.41) Higher education18,011,556 (35.06)203,002 (3.29)1,771,787 (13.00)16,036,766 (50.80) Income quartiles (CNY) < 0.001 Q1 (≤ 13,333)10,408,780 (20.26)2,120,685 (34.38)3,394,764 (24.90)4,893,330 (15.50) Q2 (13,334 − 25,000)13,967,358 (27.19)1,941,729 (31.48)4,257,788 (31.24)7,767,841 (24.61) Q3 (25,001–50,000)16,647,816 (32.41)1,611,412 (26.12)4,236,836 (31.08)10,799,568 (34.21) Q4 (> 50,000)10,343,172 (20.14)494,630 (8.02)1,742,031 (12.78)8,106,512 (25.68) Chronic diseases < 0.001 041,708,270 (81.20)3,964,086 (64.26)10,038,460 (73.64)27,705,724 (87.77) 17,981,476 (15.54)1,729,482 (28.04)2,919,244 (21.42)3,332,750 (10.56) ≥21,677,380 (3.27)474,888 (7.70)673,715 (4.94)528,777 (1.68) Smoking status < 0.001 Current smoker9,441,823 (18.38)1,342,388 (21.76)3,008,548 (22.07)5,090,887 (16.13) Former smoker3,763,355 (7.33)554,035 (8.98)1,306,510 (9.58)1,902,810 (6.03) Never smoker38,161,948 (74.29)4,272,034 (69.26)9,316,360 (68.34)24,573,554 (77.85) Self-rated health < 0.001 Poor1,352,119 (2.63)342,686 (5.56)491,978 (3.61)517,455 (1.64) Fair13,145,797 (25.59)2,095,618 (33.97)4,073,159 (29.88)6,977,020 (22.10) Good36,869,210 (71.78)3,730,152 (60.47)9,066,282 (66.51)24,072,777 (76.26)Notes: Data are presented as weighted population estimates (N) and weighted column percentages (%). N represents the estimated total population aged 15–69 of Zhejiang Province (N = 51,367,126). P-values were calculated using the second-order Rao-Scott adjusted chi-square test to account for the complex survey design (stratification, clustering, and weights) Table 1Weighted descriptive statistics of the study populationVariableTotal(Weighted N = 51,367,126)Low HL(Weighted N = 6,168,456)Moderate HL(Weighted N = 13,631,418)Relatively High HL(Weighted N = 31,567,251)P-value Sex < 0.001 Male27,036,652 (52.63)2,931,937 (47.53)7,184,809 (52.71)16,919,906 (53.60) Female24,330,473 (47.37)3,236,520 (52.47)6,446,609 (47.29)14,647,345 (46.40) Age group < 0.001 15–29 years10,221,238 (19.90)276,109 (4.48)1,459,638 (10.71)8,485,491 (26.88) 30–44 years16,517,066 (32.15)610,089 (9.89)2,995,148 (21.97)12,911,829 (40.90) 45–59 years17,008,351 (33.11)2,835,688 (45.97)5,875,795 (43.10)8,296,869 (26.28) 60–69 years7,620,470 (14.84)2,446,570 (39.66)3,300,837 (24.21)1,873,063 (5.93) Ethnicity 0.133 Han50,698,041 (98.70)6,074,831 (98.48)13,477,091 (98.87)31,146,120 (98.67) Ethnic minorities669,085 (1.30)93,625 (1.52)154,328 (1.13)421,132 (1.33) Household registration 0.042 Non-local4,841,217 (9.42)504,394 (8.18)1,209,399 (8.87)3,127,424 (9.91) Local46,525,908 (90.58)5,664,062 (91.82)12,422,019 (91.13)28,439,827 (90.09) Urbanicity < 0.001 Urban area24,008,064 (46.74)2,579,460 (41.82)5,951,991 (43.66)15,476,613 (49.03) Rural area27,359,062 (53.26)3,588,996 (58.18)7,679,427 (56.34)16,090,638 (50.97) Occupation < 0.001 Public sector7,357,998 (14.32)177,126 (2.87)1,043,379 (7.65)6,137,493 (19.44) Student3,653,997 (7.11)93,633 (1.52)588,622 (4.32)2,971,743 (9.41) Agricultural worker9,750,794 (18.98)2,664,965 (43.20)3,877,470 (28.45)3,208,359 (10.16) Industrial worker8,491,215 (16.53)1,427,659 (23.14)2,926,470 (21.47)4,137,086 (13.11) Private sector15,332,528 (29.85)870,328 (14.11)3,018,817 (22.15)11,443,384 (36.25) Others6,780,593 (13.20)934,746 (15.15)2,176,660 (15.97)3,669,186 (11.62) Marital status < 0.001 Unmarried12,282,254 (23.91)831,143 (13.47)2,322,208 (17.04)9,128,904 (28.92) Married39,084,871 (76.09)5,337,314 (86.53)11,309,211 (82.96)22,438,347 (71.08) Education level < 0.001 No formal education1,875,164 (3.65)950,583 (15.41)781,606 (5.73)142,976 (0.45) Primary school6,636,378 (12.92)2,277,274 (36.92)2,989,657 (21.93)1,369,447 (4.34) Secondary education24,844,028 (48.37)2,737,597 (44.38)8,088,369 (59.34)14,018,062 (44.41) Higher education18,011,556 (35.06)203,002 (3.29)1,771,787 (13.00)16,036,766 (50.80) Income quartiles (CNY) < 0.001 Q1 (≤ 13,333)10,408,780 (20.26)2,120,685 (34.38)3,394,764 (24.90)4,893,330 (15.50) Q2 (13,334 − 25,000)13,967,358 (27.19)1,941,729 (31.48)4,257,788 (31.24)7,767,841 (24.61) Q3 (25,001–50,000)16,647,816 (32.41)1,611,412 (26.12)4,236,836 (31.08)10,799,568 (34.21) Q4 (> 50,000)10,343,172 (20.14)494,630 (8.02)1,742,031 (12.78)8,106,512 (25.68) Chronic diseases < 0.001 041,708,270 (81.20)3,964,086 (64.26)10,038,460 (73.64)27,705,724 (87.77) 17,981,476 (15.54)1,729,482 (28.04)2,919,244 (21.42)3,332,750 (10.56) ≥21,677,380 (3.27)474,888 (7.70)673,715 (4.94)528,777 (1.68) Smoking status < 0.001 Current smoker9,441,823 (18.38)1,342,388 (21.76)3,008,548 (22.07)5,090,887 (16.13) Former smoker3,763,355 (7.33)554,035 (8.98)1,306,510 (9.58)1,902,810 (6.03) Never smoker38,161,948 (74.29)4,272,034 (69.26)9,316,360 (68.34)24,573,554 (77.85) Self-rated health < 0.001 Poor1,352,119 (2.63)342,686 (5.56)491,978 (3.61)517,455 (1.64) Fair13,145,797 (25.59)2,095,618 (33.97)4,073,159 (29.88)6,977,020 (22.10) Good36,869,210 (71.78)3,730,152 (60.47)9,066,282 (66.51)24,072,777 (76.26)Notes: Data are presented as weighted population estimates (N) and weighted column percentages (%). N represents the estimated total population aged 15–69 of Zhejiang Province (N = 51,367,126). P-values were calculated using the second-order Rao-Scott adjusted chi-square test to account for the complex survey design (stratification, clustering, and weights) Weighted descriptive statistics of the study population Notes: Data are presented as weighted population estimates (N) and weighted column percentages (%). N represents the estimated total population aged 15–69 of Zhejiang Province (N = 51,367,126). P-values were calculated using the second-order Rao-Scott adjusted chi-square test to account for the complex survey design (stratification, clustering, and weights) Differences in OOP health expenditure by health literacy profiles Table 2 presents the estimates from the Two-Part Model assessing the association between health literacy profiles and annual OOP health expenditure. The model controls for complex survey design and a comprehensive set of sociodemographic and health-related covariates. Regarding the primary predictor, Moderate Health Literacy was associated with significantly higher OOP expenditure compared to the Low Health Literacy reference group. While the probability of incurring any expenditure (Part 1: Probit) did not differ significantly (Coef. = 0.034, 95% CI: -0.025, 0.093), the expenditure intensity (Part 2: GLM) was significantly higher (Coef. = 0.186, 95% CI: 0.026, 0.347). Consequently, the Total Average Marginal Effect (AME) indicated that individuals in the Moderate Health Literacy group spent, on average, 554.71 CNY more (95% CI: 48.83, 1060.60) than those in the Low Health Literacy group. However, the Relatively High Health Literacy profile showed no statistically significant difference in total OOP expenditure compared to the Low Health Literacy group (AME = -254.62 CNY, 95% CI: -620.14, 110.90), indicating that expenditure levels for this group returned to the baseline. Among covariates, health status indicators were the substantial drivers of expenditure, following expected clinical patterns. Multimorbidity (≥ 2 chronic diseases) was associated with a significant surge in annual spending (AME = 3,582.42 CNY, 95% CI: 2614.28, 4550.56) relative to those with no chronic conditions. Conversely, favorable self-rated health served as a strong protective factor, with individuals reporting “good” health spending approximately 8,597.55 CNY less than those reporting “poor” health. Regarding sociodemographic factors, rural residents showed a higher probability of incurring expenditure (Part 1: Coef. = 0.156, 95% CI: 0.119, 0.193) but lower expenditure intensity (Part 2: Coef. = -0.173, 95% CI: -0.316, -0.030), which resulted in no significant difference in total annual expenditure compared to urban residents. Based on the predictive margins from the fully adjusted Two-Part Model, the adjusted average annual OOP expenditure followed an inverted U-shaped pattern across health literacy profiles. The predicted expenditure was highest in the Moderate Health Literacy group (3,143 CNY, 95% CI: 2,580, 3,706). Subsequently, the Relatively High Health Literacy group had a predicted expenditure of 2,334 CNY (95% CI: 2,112, 2,555). While this level was not statistically different from the Low Health Literacy baseline (2,588 CNY, 95% CI: 2,358, 2,818), it represented a substantial decline from the peak observed in the Moderate group. Post-estimation pairwise testing confirmed that the expenditure in the Moderate group was significantly higher than in the Relatively High Health Literacy group (Difference = 809 CNY, P = 0.026), effectively completing the inverted U-shaped pattern (see Fig. 2). Table 2Two-Part Model estimates of the association between health literacy profiles and OOP expenditureVariablesPart 1: ProbitCoef. (95% CI)Part 2: GLMCoef. (95% CI)Total AMEdy/dx (95% CI) [CNY]Health Literacy Profiles (Ref: Low HL)Moderate HL0.034 (-0.025, 0.093)0.186* (0.026, 0.347)554.71* (48.83, 1060.60)Relatively High HL0.053 (-0.010, 0.116)-0.116 (-0.264, 0.033)-254.62 (-620.14, 110.90)Sex (Ref: Male)Female0.080*** (0.034, 0.126)-0.031 (-0.255, 0.193)-34.37 (-636.25, 567.51)Age group (Ref: 15–29)30–44 years-0.018 (-0.096, 0.060)0.096 (-0.152, 0.343)251.77 (-414.35, 917.89)45–59 years-0.106* (-0.191, -0.022)-0.047 (-0.303, 0.210)-175.33 (-843.00, 492.34)60–69 years-0.150** (-0.243, -0.057)0.102 (-0.165, 0.370)186.59 (-520.25, 893.43)Ethnicity (Ref: Han)Ethnic minorities0.017 (-0.138, 0.173)0.038 (-0.253, 0.330)115.52 (-705.25, 936.29)Household registration (Ref: Non-local)Local-0.024 (-0.088, 0.039)0.151* (0.010, 0.292)364.98* (20.37, 709.59)Urbanicity (Ref: Urban area)Rural area0.156*** (0.119, 0.193)-0.173* (-0.316, -0.030)-374.68 (-781.99, 32.62)Occupation (Ref: Public sector)Student-0.002 (-0.118, 0.114)-0.531* (-0.944, -0.118)-1177.47* (-2181.65, -173.28)Agricultural worker0.174*** (0.101, 0.248)-0.023 (-0.375, 0.329)48.38 (-956.99, 1053.74)Industrial worker-0.070 (-0.140, 0.001)-0.252 (-0.609, 0.104)-676.48 (-1692.27, 339.32)Private sector0.054 (-0.007, 0.115)-0.200 (-0.563, 0.162)-487.22 (-1514.92, 540.48)Others0.080* (0.007, 0.153)0.056 (-0.338, 0.451)223.72 (-921.19, 1368.64)Marital status (Ref: Unmarried)Married0.076** (0.019, 0.134)0.172* (0.023, 0.321)478.35** (117.29, 839.40)Education level (Ref: No formal education)Primary school0.025 (-0.056, 0.107)-0.074 (-0.262, 0.114)-182.43 (-705.88, 341.02)Secondary education0.094* (0.013, 0.175)-0.093 (-0.281, 0.095)-190.08 (-712.80, 332.64)Higher education0.144** (0.045, 0.244)0.008 (-0.264, 0.279)115.36 (-662.27, 892.99)Income quartiles (Ref: Q1(≤ 13,333))Q2 (13,334 − 25,000)0.002 (-0.052, 0.056)0.103 (-0.065, 0.272)294.88 (-207.96, 797.72)Q3 (25,001–50,000)-0.068* (-0.121, -0.016)-0.111* (-0.219, -0.003)-319.98* (-600.05, -39.91)Q4 (> 50,000)-0.137*** (-0.200, -0.075)0.016 (-0.185, 0.218)-42.98 (-581.80, 495.84)Chronic disease (Ref: 0)10.400*** (0.348, 0.452)0.392*** (0.276, 0.507)1266.52*** (938.81, 1594.24)≥20.662*** (0.558, 0.766)0.886*** (0.700, 1.072)3582.42*** (2614.28, 4550.56)Smoking status (Ref: Current smoker)Former smoker0.203*** (0.125, 0.280)0.377*** (0.218, 0.537)1149.33*** (676.88, 1621.78)Never smoker0.178*** (0.123, 0.234)0.185 (-0.032, 0.401)554.58* (33.84, 1075.32)Self-rated health (Ref: Poor)Fair-0.597*** (-0.755, -0.439)-1.028*** (-1.164, -0.892)-6866.13*** (-8139.18, -5593.09)Good-0.956*** (-1.111, -0.801)-1.655*** (-1.787, -1.524)-8597.55*** (-9857.78, -7337.31) Observations 56,86346,57356,863Notes:Results were estimated using a Two-Part Model adjusted for the complex survey design with a composite final weightPart 1 (Probit): Estimates the probability of an individual incurring any OOP healthcare expenditure (OOP > 0)Part 2 (GLM): Estimates the expenditure intensity among those with positive spending, utilizing a Generalized Linear Model with a Gamma distribution and log linkTotal Average Marginal Effect (AME): Represents the average change in annual OOP expenditure (in CNY) associated with each variable, derived from the combined predictions of both model parts* p < 0.05, ** p < 0.01, *** p < 0.001. CI = 95% Confidence Interval. Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places Table 2Two-Part Model estimates of the association between health literacy profiles and OOP expenditureVariablesPart 1: ProbitCoef. (95% CI)Part 2: GLMCoef. (95% CI)Total AMEdy/dx (95% CI) [CNY]Health Literacy Profiles (Ref: Low HL)Moderate HL0.034 (-0.025, 0.093)0.186* (0.026, 0.347)554.71* (48.83, 1060.60)Relatively High HL0.053 (-0.010, 0.116)-0.116 (-0.264, 0.033)-254.62 (-620.14, 110.90)Sex (Ref: Male)Female0.080*** (0.034, 0.126)-0.031 (-0.255, 0.193)-34.37 (-636.25, 567.51)Age group (Ref: 15–29)30–44 years-0.018 (-0.096, 0.060)0.096 (-0.152, 0.343)251.77 (-414.35, 917.89)45–59 years-0.106* (-0.191, -0.022)-0.047 (-0.303, 0.210)-175.33 (-843.00, 492.34)60–69 years-0.150** (-0.243, -0.057)0.102 (-0.165, 0.370)186.59 (-520.25, 893.43)Ethnicity (Ref: Han)Ethnic minorities0.017 (-0.138, 0.173)0.038 (-0.253, 0.330)115.52 (-705.25, 936.29)Household registration (Ref: Non-local)Local-0.024 (-0.088, 0.039)0.151* (0.010, 0.292)364.98* (20.37, 709.59)Urbanicity (Ref: Urban area)Rural area0.156*** (0.119, 0.193)-0.173* (-0.316, -0.030)-374.68 (-781.99, 32.62)Occupation (Ref: Public sector)Student-0.002 (-0.118, 0.114)-0.531* (-0.944, -0.118)-1177.47* (-2181.65, -173.28)Agricultural worker0.174*** (0.101, 0.248)-0.023 (-0.375, 0.329)48.38 (-956.99, 1053.74)Industrial worker-0.070 (-0.140, 0.001)-0.252 (-0.609, 0.104)-676.48 (-1692.27, 339.32)Private sector0.054 (-0.007, 0.115)-0.200 (-0.563, 0.162)-487.22 (-1514.92, 540.48)Others0.080* (0.007, 0.153)0.056 (-0.338, 0.451)223.72 (-921.19, 1368.64)Marital status (Ref: Unmarried)Married0.076** (0.019, 0.134)0.172* (0.023, 0.321)478.35** (117.29, 839.40)Education level (Ref: No formal education)Primary school0.025 (-0.056, 0.107)-0.074 (-0.262, 0.114)-182.43 (-705.88, 341.02)Secondary education0.094* (0.013, 0.175)-0.093 (-0.281, 0.095)-190.08 (-712.80, 332.64)Higher education0.144** (0.045, 0.244)0.008 (-0.264, 0.279)115.36 (-662.27, 892.99)Income quartiles (Ref: Q1(≤ 13,333))Q2 (13,334 − 25,000)0.002 (-0.052, 0.056)0.103 (-0.065, 0.272)294.88 (-207.96, 797.72)Q3 (25,001–50,000)-0.068* (-0.121, -0.016)-0.111* (-0.219, -0.003)-319.98* (-600.05, -39.91)Q4 (> 50,000)-0.137*** (-0.200, -0.075)0.016 (-0.185, 0.218)-42.98 (-581.80, 495.84)Chronic disease (Ref: 0)10.400*** (0.348, 0.452)0.392*** (0.276, 0.507)1266.52*** (938.81, 1594.24)≥20.662*** (0.558, 0.766)0.886*** (0.700, 1.072)3582.42*** (2614.28, 4550.56)Smoking status (Ref: Current smoker)Former smoker0.203*** (0.125, 0.280)0.377*** (0.218, 0.537)1149.33*** (676.88, 1621.78)Never smoker0.178*** (0.123, 0.234)0.185 (-0.032, 0.401)554.58* (33.84, 1075.32)Self-rated health (Ref: Poor)Fair-0.597*** (-0.755, -0.439)-1.028*** (-1.164, -0.892)-6866.13*** (-8139.18, -5593.09)Good-0.956*** (-1.111, -0.801)-1.655*** (-1.787, -1.524)-8597.55*** (-9857.78, -7337.31) Observations 56,86346,57356,863Notes:Results were estimated using a Two-Part Model adjusted for the complex survey design with a composite final weightPart 1 (Probit): Estimates the probability of an individual incurring any OOP healthcare expenditure (OOP > 0)Part 2 (GLM): Estimates the expenditure intensity among those with positive spending, utilizing a Generalized Linear Model with a Gamma distribution and log linkTotal Average Marginal Effect (AME): Represents the average change in annual OOP expenditure (in CNY) associated with each variable, derived from the combined predictions of both model parts* p < 0.05, ** p < 0.01, *** p < 0.001. CI = 95% Confidence Interval. Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places Two-Part Model estimates of the association between health literacy profiles and OOP expenditure Notes: Results were estimated using a Two-Part Model adjusted for the complex survey design with a composite final weight Part 1 (Probit): Estimates the probability of an individual incurring any OOP healthcare expenditure (OOP > 0) Part 2 (GLM): Estimates the expenditure intensity among those with positive spending, utilizing a Generalized Linear Model with a Gamma distribution and log link Total Average Marginal Effect (AME): Represents the average change in annual OOP expenditure (in CNY) associated with each variable, derived from the combined predictions of both model parts * p < 0.05, ** p < 0.01, *** p < 0.001. CI = 95% Confidence Interval. Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places Fig. 2Predicted annual OOP health expenditure by health literacy profiles. notes Bars represent the predictive margins (adjusted predicted means) of annual OOP expenditure (in CNY) derived from the Two-Part Model. Error bars indicate 95% confidence intervals. These estimates are based on the fully adjusted regression model presented in Table 2 Predicted annual OOP health expenditure by health literacy profiles. notes Bars represent the predictive margins (adjusted predicted means) of annual OOP expenditure (in CNY) derived from the Two-Part Model. Error bars indicate 95% confidence intervals. These estimates are based on the fully adjusted regression model presented in Table 2 Health literacy and OOP health expenditure across chronic disease status Formal interaction tests were conducted to examine potential heterogeneity. The interaction between health literacy and chronic disease status was not statistically significant in the probability model (Part 1, P for interaction = 0.519) but approached the significance threshold in the expenditure intensity model (Part 2, P for interaction = 0.053). Based on these results, this marginal signal prompted an exploratory stratified examination to better understand potential variations in expenditure intensity. Table 3 presents the stratified estimates from the Two-Part Model. Among participants without chronic diseases, the Moderate Health Literacy group exhibited a significantly higher total annual OOP expenditure compared to the Low Health Literacy group (AME = 503.76 CNY, P < 0.05). This was accompanied by a significantly higher expenditure intensity (Part 2 Coef. = 0.252, P < 0.05), while the probability of incurring expenditure did not differ significantly (Part 1 Coef. = 0.028, P > 0.05). No significant difference was observed for the Relatively High Health Literacy group in this subgroup (AME = -101.70 CNY, P > 0.05). Among participants with one chronic disease, the associations differed between the two model components. Compared to the Low Health Literacy group, individuals with Relatively High Health Literacy showed a significantly higher probability of incurring any OOP expenditure (Part 1 Coef. = 0.186, P < 0.01). However, among those who incurred expenditure, the Relatively High Health Literacy group had a significantly lower expenditure intensity (Part 2 Coef. = −0.224, P < 0.05). The combined total average marginal effect indicated a marginally lower annual OOP expenditure for the Relatively High Health Literacy group compared to the reference group (AME = − 765.64 CNY, P < 0.10). Among participants with multimorbidity (two or more chronic diseases), the point estimate for the total average marginal effect in the Relatively High Health Literacy group was 772.94 CNY compared to the Low Health Literacy group. This difference was not statistically significant (P = 0.563). Similarly, no statistically significant associations were found for this subgroup in either the probability model (Part 1) or the expenditure intensity model (Part 2). Figure 3 displays the predicted marginal annual expenditures. For participants without chronic diseases, the predicted expenditure was highest in the Moderate Health Literacy group. In the one chronic disease subgroup, the Relatively High Health Literacy group had a predicted expenditure of 3,562 CNY, which was numerically lower than the Low Health Literacy group (4,327 CNY). In the multimorbidity subgroup, the Relatively High Health Literacy profile showed a numerically higher mean predicted expenditure (8,973 CNY) compared to the Low Health Literacy profile (8,200 CNY), although the confidence intervals overlapped substantially. Table 3Exploratory stratified analysis for Two-Part Model estimates of the association between health literacy and OOP expenditure by chronic diseaseVariablesNo Chronic Disease(n = 41,340)Coef. (95% CI)One Chronic Disease(n = 12,371)Coef. (95% CI)Multimorbidity (≥ 2)(n = 3138)Coef. (95% CI)Part 1: Probability of Any OOP aLow HLRef.Ref.Ref.Moderate HL0.028 (-0.042, 0.098)0.080 (-0.037, 0.197)-0.104 (-0.362, 0.153)Relatively High HL0.033 (-0.042, 0.107)0.186 ** (0.058, 0.315)-0.073 (-0.336, 0.190)Part 2: Expenditure Intensity bLow HLRef.Ref.Ref.Moderate HL0.252* (0.041, 0.464)0.041 (-0.152, 0.234)-0.171 (-0.419, 0.076)Relatively High HL-0.073 (-0.238, 0.092)-0.224* (-0.419, -0.029)0.097 (-0.207, 0.400)Total AME (CNY) cLow HLRef.Ref.Ref.Moderate HL503.76* (28.82, 978.70)243.64 (-624.62, 1111.90)-1356.28 (-3254.47, 541.92)Relatively High HL-101.70 (-376.60, 173.20)-765.64† (-1560.00, 28.72)772.94 (-1847.12, 3393.00)Notes:* P < 0.05, ** P < 0.01, *** P < 0.001, † P < 0.10 Coef. represents the regression coefficient.Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places.a Part 1 (Probit): Modeling the likelihood of incurring any OOP expenditure (i.e., OOP > 0).b Part 2 (GLM): Modeling the intensity of expenditure among those with positive spending (Gamma family, log link).c Total AME: Estimates represent the Total Average Marginal Effects (in CNY) derived from a Two-Part Model.Interaction Analysis: Formal tests revealed no significant interaction between health literacy and chronic disease status in Part 1 (P = 0.519), whereas the interaction in Part 2 was marginally significant (P = 0.053), supporting the stratified presentation of expenditure intensityThe regression models automatically excluded 14 observations (11 in the one chronic disease group and 3 in the multimorbidity group) due to perfect prediction Table 3Exploratory stratified analysis for Two-Part Model estimates of the association between health literacy and OOP expenditure by chronic diseaseVariablesNo Chronic Disease(n = 41,340)Coef. (95% CI)One Chronic Disease(n = 12,371)Coef. (95% CI)Multimorbidity (≥ 2)(n = 3138)Coef. (95% CI)Part 1: Probability of Any OOP aLow HLRef.Ref.Ref.Moderate HL0.028 (-0.042, 0.098)0.080 (-0.037, 0.197)-0.104 (-0.362, 0.153)Relatively High HL0.033 (-0.042, 0.107)0.186 ** (0.058, 0.315)-0.073 (-0.336, 0.190)Part 2: Expenditure Intensity bLow HLRef.Ref.Ref.Moderate HL0.252* (0.041, 0.464)0.041 (-0.152, 0.234)-0.171 (-0.419, 0.076)Relatively High HL-0.073 (-0.238, 0.092)-0.224* (-0.419, -0.029)0.097 (-0.207, 0.400)Total AME (CNY) cLow HLRef.Ref.Ref.Moderate HL503.76* (28.82, 978.70)243.64 (-624.62, 1111.90)-1356.28 (-3254.47, 541.92)Relatively High HL-101.70 (-376.60, 173.20)-765.64† (-1560.00, 28.72)772.94 (-1847.12, 3393.00)Notes:* P < 0.05, ** P < 0.01, *** P < 0.001, † P < 0.10 Coef. represents the regression coefficient.Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places.a Part 1 (Probit): Modeling the likelihood of incurring any OOP expenditure (i.e., OOP > 0).b Part 2 (GLM): Modeling the intensity of expenditure among those with positive spending (Gamma family, log link).c Total AME: Estimates represent the Total Average Marginal Effects (in CNY) derived from a Two-Part Model.Interaction Analysis: Formal tests revealed no significant interaction between health literacy and chronic disease status in Part 1 (P = 0.519), whereas the interaction in Part 2 was marginally significant (P = 0.053), supporting the stratified presentation of expenditure intensityThe regression models automatically excluded 14 observations (11 in the one chronic disease group and 3 in the multimorbidity group) due to perfect prediction Exploratory stratified analysis for Two-Part Model estimates of the association between health literacy and OOP expenditure by chronic disease Notes: * P < 0.05, ** P < 0.01, *** P < 0.001, † P < 0.10 Coef. represents the regression coefficient.Coefficients for Part 1 and Part 2 are rounded to three decimal places. AME values are rounded to two decimal places. a Part 1 (Probit): Modeling the likelihood of incurring any OOP expenditure (i.e., OOP > 0). b Part 2 (GLM): Modeling the intensity of expenditure among those with positive spending (Gamma family, log link). c Total AME: Estimates represent the Total Average Marginal Effects (in CNY) derived from a Two-Part Model. Interaction Analysis: Formal tests revealed no significant interaction between health literacy and chronic disease status in Part 1 (P = 0.519), whereas the interaction in Part 2 was marginally significant (P = 0.053), supporting the stratified presentation of expenditure intensity The regression models automatically excluded 14 observations (11 in the one chronic disease group and 3 in the multimorbidity group) due to perfect prediction Fig. 3Predicted annual OOP expenditure by health literacy profiles and chronic disease status. nOTE Estimates represent the predicted marginal annual OOP expenditure (in CNY) derived from stratified Two-Part Models. Vertical bars indicate 95% confidence intervals (CIs). While the formal interaction test in the probability model (Part 1) was non-significant, the interaction in the expenditure intensity model (Part 2) approached the significance threshold (P for interaction = 0.053). Consequently, this stratified visualization is presented as an exploratory analysis to suggest differential expenditure patterns across chronic disease burden levels Predicted annual OOP expenditure by health literacy profiles and chronic disease status. nOTE Estimates represent the predicted marginal annual OOP expenditure (in CNY) derived from stratified Two-Part Models. Vertical bars indicate 95% confidence intervals (CIs). While the formal interaction test in the probability model (Part 1) was non-significant, the interaction in the expenditure intensity model (Part 2) approached the significance threshold (P for interaction = 0.053). Consequently, this stratified visualization is presented as an exploratory analysis to suggest differential expenditure patterns across chronic disease burden levels Robustness checks To verify the stability of our findings, we conducted sensitivity analyses using a continuous standardized health literacy Z-score in place of the latent profiles. First, under a linear specification of the Two-Part Model (Supplementary Table S3), health literacy exhibited divergent associations across the two decision stages. While higher health literacy was associated with a significantly increased probability of incurring expenditure (Part 1 Coef. = 0.033, P < 0.05), it was associated with a significant reduction in expenditure intensity (Part 2 Coef. = -0.100, P < 0.05). The net result indicated a reduction in total annual OOP expenditure (AME = -246.69 CNY; 95% CI: -496.33, 2.95)for every one-standard-deviation increase in health literacy. Second, to capture the non-linear dynamics observed in the main analysis, we introduced a quadratic term for health literacy. The analysis confirmed distinct patterns. In the probability model (Part 1), the quadratic term was not statistically significant (Coef. = 0.006, P = 0.574), indicating a linear relationship for access. In contrast, the expenditure intensity model (Part 2) exhibited a significant negative quadratic term (Coef. = -0.073, P = 0.036). This finding corroborates that the inverted U-shaped association is primarily driven by the intensity of healthcare utilization. Specifically, predicted expenditure peaked at an intermediate level (Z ≈ -1), mirroring the characteristics of the Moderate Health Literacy group identified in the latent profile analysis. The predicted pattern is visualized in Supplementary Figure S1. Discussion This study identified distinct health literacy profiles among residents in Zhejiang, China and examined the association between these profiles and OOP health expenditure. Our fully adjusted Two-Part Models suggested a non-linear pattern resembling an inverted U-shaped association, where total expenditure peaked at the Moderate Health Literacy level and declined relative to this peak in the Relatively High Health Literacy level. In addition, this non-linear pattern was further corroborated by sensitivity analyses using continuous health literacy scores, where a significant quadratic term confirmed the inverted U-shaped trajectory. The Two-Part Model analysis indicated that health literacy was not a significant predictor of the probability of incurring expenditure but was significantly associated with expenditure intensity among those who utilized healthcare. Furthermore, exploratory stratified analyses revealed that this association is context-dependent. The efficiency advantage of high health literacy appeared most pronounced among individuals with a single chronic disease, whereas the association was attenuated among those with multimorbidity. These findings suggest that health literacy functions as a complex modulator of healthcare utilization efficiency and individual financial expenditure. Prior research has widely positioned limited health literacy as a driver of inefficient healthcare spending, leading to higher costs through delayed care or reliance on emergency services [8, 11]. In the Chinese context, studies have similarly argued that higher health literacy contributes to a lower financial burden [26]. These studies tend to frame health literacy as a mechanism for cost reduction and have predominantly utilized binary classifications (e.g., adequate vs. limited) or linear assumptions to describe the overall trend. Our study seeks to extend these findings by applying a Two-Part Model to distinguish between the probability of incurring expenditure and the intensity of utilization. Regarding the probability of incurring expenditure, our analysis showed no significant variation across health literacy profiles. This finding stands in contrast to contexts where limited health literacy acts as a significant barrier to entry [36]. Instead, it suggests that the threshold for initiating care is relatively consistent across all literacy levels, implying that health literacy is not a primary gatekeeper for initial access in this setting. This pattern of broad accessibility is further contextualized by our sociodemographic findings. The rural residents were found to be significantly more likely to incur expenditure than their urban counterparts. Rather than implying disparate health needs alone, this pattern indicates that physical and financial barriers do not disproportionately hinder rural populations from entering the healthcare system. This phenomenon is likely related to the region’s robust healthcare infrastructure. Zhejiang, a province at the forefront of China’s economic development [37], has made substantial progress in strengthening rural primary healthcare infrastructure, achieving high coverage of village clinics and township health centers [38, 39]. Consequently, the high geographic accessibility of basic services appears to create an enabling environment where the decision to initiate care is less constrained by literacy or location. However, equity in access does not necessarily translate to equity in efficiency. Once consumption is initiated, health literacy becomes a key factor associated with the intensive margin of expenditure represented by Part 2. We propose that the observed inverted U-shaped pattern may reflect a dynamic evolution in how individuals interact with the healthcare system, a process where capability potentially influences economic behavior. The significantly higher expenditure intensity observed in the Moderate Health Literacy group compared to the Low Health Literacy reference group aligns with characteristics described in Nutbeam’s “Interactive Health Literacy” framework [40]; Individuals at this stage possess the cognitive and social skills to extract information and act independently within the healthcare environment. From the perspective of Grossman’s Health Capital Model, this increased engagement may reflect a heightened demand for health investment [41]. These individuals recognize the value of health capital and actively seek to preserve it. Drawing on prior evidence that links improved health literacy to advanced stages of behavioral change and increased utilization of preventive services [42, 43], we infer that the Moderate Health Literacy group has progressed to a more active state of health engagement compared to the Low Health Literacy baseline. Consistent with this activation hypothesis, previous studies have demonstrated that literacy promotes the uptake of proactive health behaviors, such as cancer screening and health engagement [44, 45]. However, unlike the High Health Literacy group, the Moderate Health Literacy group may lack the allocative efficiency to utilize these services cost-effectively. Possessing partial knowledge, they may be sufficiently active to initiate care yet lack the discernment to maximize value. Consequently, their heightened engagement may be characterized as “action without mastery”. In contrast, the trajectory for the Relatively High Health Literacy group suggests a potential shift toward optimized utilization. Aligning with the “Critical Health Literacy” level of Nutbeam’s framework [40], these individuals are likely better equipped to critically analyze health information and engage in shared decision-making. Instead of simply consuming more, they appear to engage in what may represent optimized utilization. Although the expenditure intensity for this group did not differ statistically from the low health literacy baseline, the numerical decline from the Moderate peak is noteworthy. We acknowledge that the advantageous socioeconomic profile of this group likely confers a degree of financial protection. However, these socioeconomic advantages cannot fully account for the findings. Even after adjusting for occupation and income, the expenditure intensity numerically recedes from the significant peak observed in the Moderate group and statistically returns to baseline levels. This pattern suggests that allocative efficiency plays a plausible complementary role in preventing the specific type of inefficiently elevated expenditure observed in the Moderate group. This aggregate pattern may reflect gains in allocative efficiency, a mechanism that was suggested in our exploratory stratified analysis. Specifically, in the “One Chronic Disease” subgroup (Table 3), Relatively High Health Literacy was associated with significantly lower expenditure intensity compared to the baseline. Such behavior aligns with the pursuit of allocative efficiency, where individuals aim to achieve health outcomes through more discerning use of resources. In this context, advanced health literacy is theorized to function as a form of human capital that enhances the marginal product of health inputs [41]. By demonstrating the competence to navigate medical choices rationally [8, 46], these individuals appear capable of managing the demands of proactive care without driving total expenditure above baseline levels, effectively maintaining total expenditure at levels comparable to the baseline. Regarding potential heterogeneity, we examined whether the associations varied by chronic disease burden. The interaction for expenditure intensity approached the significance threshold. Although not strictly significant, this signal prompted an exploratory stratified examination to better understand the nuances of utilization behavior. In these exploratory analyses, the association between health literacy and expenditure intensity appeared more discernible among individuals with no or one chronic disease and attenuated among those with multimorbidity. This pattern aligns with the “Depreciation Rate” concept in Grossman’s model [47]. From this perspective, medical expenditure for healthier populations functions as a discretionary investment in preserving health capital, where health literacy effectively modulates efficiency. In contrast, for individuals with multiple chronic diseases, spending often becomes non-discretionary “maintenance” directed toward essential treatments [22, 48] to offset the rapid depreciation of their health capital. In this context of high disease burden, the biological imperative of treating multimorbidity constrain individual agency in health management, suggesting that the protective effect of health literacy may have a ceiling when facing high disease burden. Furthermore, an exploratory examination of the “No Chronic Disease” and “One Chronic Disease” subgroups suggest how the function of health literacy may vary with disease progression. Among individuals with no chronic conditions, the cost burden is primarily observed in the Moderate Health Literacy group. In contrast, the efficiency advantage of the Relatively High Health Literacy group appears more pronounced in the “One Chronic Disease” subgroup. For this group, while the probability of incurring expenditure was higher (Part 1), the expenditure intensity was significantly lower (Part 2). This suggests that for early-stage chronic disease management, high health literacy may facilitate improved allocative efficiency. Limitations Several limitations should be considered when interpreting our findings. First, the cross-sectional design precludes causal inference. While we adjusted for a comprehensive set of confounders, reverse causality cannot be ruled out. Longitudinal data are needed to establish the temporal precedence of these associations. Second, regarding generalizability, our sample was drawn from Zhejiang Province, an economically developed region in China. While this setting offers insights into the trajectory of health expenditure in rapidly developing contexts, the findings may not fully generalize to less developed regions with different healthcare infrastructures and literacy baselines. Future studies using nationally representative samples are warranted to examine whether these patterns hold across diverse socioeconomic settings. Third, data on OOP expenditure and chronic disease status were self-reported, which may be subject to recall bias. Additionally, although we controlled for key socioeconomic variables including occupation, household income, and education, a limitation of this study is the lack of differentiation between specific health insurance schemes (e.g., Urban Employee vs. Urban and Rural Resident Basic Medical Insurance). Since reimbursement rates vary across schemes, some residual confounding may remain in our estimates of OOP expenditure. Future research utilizing granular administrative data is warranted to explore the specific impacts of different insurance benefit designs. Implications The observed inverted U-shaped relationship between health literacy profiles and expenditure intensity suggests a potential need for health policy to move beyond a sole focus on improving healthcare access toward greater attention to allocative efficiency. The contrast between the cost-intensive engagement of Moderate Health Literacy and the rationalized utilization of Relatively High Health Literacy suggests that mere competency without mastery may inadvertently result in cost-intensive activation rather than optimized utilization. Consequently, health education should evolve from promoting basic knowledge to cultivating critical decision-making capabilities. This involves empowering residents to navigate tiered healthcare systems effectively, for instance, by encouraging the use of primary care for common conditions rather than bypassing gatekeepers for specialist services. Importantly, this educational shift should be supported by a strengthened primary care system. By enhancing the service capacity of family doctors and community health centers, the system can effectively act as a navigator and gatekeeper, channeling the heightened patient activation into cost-effective management pathways. This combined approach helps facilitate the transformation from the cost-intensive engagement of the Moderate Health Literacy group to the rationalized utilization exemplified by the Relatively High Health Literacy group. Furthermore, the exploratory stratified analysis points to the potential value of a differentiated management strategy that considers the potential ceiling effect of individual empowerment. While higher health literacy appears to promote efficient expenditure among individuals with limited chronic conditions, our findings suggest that this mechanism may attenuate in the context of multimorbidity, where complex care needs may limit the influence of behavioral competencies. This observed pattern supports the consideration of a tiered health management strategy. For generally healthy populations or those with early-stage chronic conditions, policies should leverage the efficiency-enhancing potential of high health literacy to promote rational care. However, for high-need populations managing multimorbidity, enhancing health literacy alone might be insufficient. For these vulnerable groups, strengthening systemic financial protection, such as targeted medical assistance and catastrophic coverage, remains the fundamental priority. Future interventions should recognize that while health literacy is a powerful tool for individual empowerment, it is likely most effective when complemented by robust protections as disease complexity increases. Finally, the Zhejiang experience offers a forward-looking perspective for health equity in developing regions. As healthcare systems mature and access barriers diminish, disparities in health outcomes may increasingly stem from differences in individual capability. Consequently, future health equity policies would benefit from expanding the definition of equity to incorporate competency empowerment alongside financial protection. This approach shifts the focus from simple access to meaningful utilization, ensuring that citizens possess the essential capabilities to navigate the healthcare system efficiently. This ensures that citizens possess the necessary support regardless of their health status. Conclusion In summary, our study uncovers a nuanced, inverted U-shaped association between health literacy and OOP expenditure, revealing that in high-access settings like Zhejiang, the challenge of health equity is increasingly characterized by efficiency gaps rather than simple access barriers. Specifically, expenditure intensity peaks significantly at the Moderate Health Literacy level but follows a downward trajectory at the Relatively High Health Literacy level, returning to the low health literacy baseline. Regarding disease burden, exploratory stratified analyses suggest that this inverted U-shaped pattern may be more pronounced in early-stage disease management but seems attenuated among individuals with multimorbidity. These findings point to a potential context-dependent association, implying that the relationship between health literacy and expenditure intensity may vary according to the complexity of chronic conditions. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 all supplementary material all supplementary material