The impact of demographic characteristics on health literacy in the population: a mixed model analysis based on panel survey data from China Authors: Rui Li, Xinyu Xu, Lingming Gong, Tiance Li, Feidan Deng, Yan Leng, Weiliang Liu, Sheng Gao, Fangyao Chen, Liansen Wang, Jing Han Published: 2025 ABSTRACT AbstractBackgroundHealth literacy is strongly correlated with health outcomes, making it a key indicator of population health. This study aimed to analyze trends in health literacy and assess the impact of key demographic factors on these trends in Shandong Province, China, to inform future health education strategies.MethodsA cross-sectional study analyzed 12,302 samples from 2012, 2016, and 2020, employing mixed-effects models to investigate trends in health literacy and influencing factors in Shandong Province, China.ResultsThis study examined health literacy determinants using data from 12,302 participants in Shandong Province across three waves (2012, 2016, 2020). The analysis revealed substantial age-related disparities, with health literacy prevalence declining from 17.6% among adults aged 25–34 to 9.2% in the 55–64 age group. Urban residents demonstrated significantly higher health literacy rates than rural counterparts (17.5% vs. 8.9%). Mixed-effects models showed each increasing education level was associated with 60% higher odds of health literacy (OR = 1.60, 95% CI: 1.51–1.69), though this effect diminished over time (2012 coefficient: 0.615; 2020 coefficient: 0.365). Urban residence was associated with 39% lower odds of limited health literacy (OR = 0.61, 95% CI: 0.54–0.69). The relationship between self-rated health and health literacy shifted significantly, with coefficients changing from 0.050 (2012) to -0.155 (2020), indicating strengthening protective effects.ConclusionsDespite progress, significant health literacy challenges persist in Shandong Province, characterized by evolving media dynamics, persistent urban-rural disparities, and the changing role of education. These findings call for targeted interventions focusing on high-risk groups such as the older population, bridging the urban-rural gap through improved digital health resource allocation in rural areas, and integrating health literacy into educational systems to ensure sustainable impact. Addressing these priorities is essential for fostering health literacy and achieving a healthier population.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-025-25969-y.Keywords: Health literacy, Demographic characteristics, Panel survey, Mixed model Background Health literacy refers to an individual’s ability to access and understand basic health information and services, and to use this information and services to make informed decisions for maintaining and promoting one’s health [1–3]. Low levels of health literacy can lead to an individual’s inability to make sound health decisions and are associated with many adverse outcomes, such as lower use of preventive services, lower adherence to medication prescriptions, an increased risk of hospitalization, and an increased risk of poor lifestyle habits [4]. Research indicates a strong correlation between residents’ health literacy level and health outcomes. Globally, longitudinal studies such as one conducted in the United States have shown that age, gender, socio-economic status, and baseline health all have a significant impact on health literacy [5].Similarly, within China, a study of health literacy levels across 25 provinces and cities also revealed significant heterogeneity between different regions, urban and rural areas, and different social groups [6, 7]. However, existing research, both globally and in China, largely relies on single-year data. This creates a critical evidence gap regarding the evolution of health literacy and the dynamic influence of its contributing factors over time. Such a gap significantly limits our capacity to determine whether health disparities are widening or narrowing, and it obstructs a meaningful evaluation of the long-term effectiveness of major policy interventions. China has introduced many national policies to improve the health literacy of its population [8]. In 2016, China initiated the “Healthy China 2030” plan [9], aiming to comprehensively improve the health literacy of residents through the active participation of the entire society. In 2019, the State Council of China issued the Healthy China 15 Special Actions. The goal is to ensure that by 2030, the health literacy level of the entire population will be no less than 30%. Research on health interventions suggests that the effectiveness of broad policies like “Healthy China 2030” can be moderated by demographic factors [10]; for instance, their impact on health literacy may vary significantly between urban and rural populations due to differences in resource availability and information access [11]. Furthermore, studies indicate that the uptake of health information disseminated through policies is often influenced by an individual’s age and education level, which can shape health literacy outcomes [12]. Additionally, during the COVID-19 pandemic, health literacy was recognized as a critical component of the public health response, once again drawing intense focus from the public and the government and significantly influencing residents’ health literacy levels [13–15]. This focus was further amplified by the advent of the new media era (e.g., social media, health apps, and short-video platforms), which made accessing health information more convenient than ever. In the Chinese context, the organic integration of these new media with traditional media (e.g., television, radio, and newspapers) has also facilitated the dissemination and sharing of health information, reducing the disparities in health literacy levels among individuals with different demographic characteristics.(Fig. 1) [16]. However, whether these policies truly affect health literacy has not been thoroughly studied. Timeline for the release of important public health time policies in China The timeline for the release of China's important public health time policies shown in the figure is arranged based on their release dates. The timeline starts from the 12th Five Year Plan in 2011 Therefore, to directly address this longitudinal evidence gap, this study leverages data from Shandong Province, a key region on the eastern coast of China. With a permanent resident population of over 100 million, Shandong is the second-most populous province and features significant economic and demographic diversity, encompassing both developed urban centers and large rural areas. This makes it a highly relevant and representative setting for studying health literacy trends, as it reflects the challenges and opportunities faced by a large, diverse segment of the Chinese population. This study extracted data from the Health Literacy Surveillance Database of Residents in Shandong Province for the years 2012, 2016, and 2020. Guided by the following research questions, this study aimed to: (1) How did the overall level of health literacy and its distribution across key demographic groups (e.g., by age, urban-rural residence, and education) evolve between 2012 and 2020? (2) Did the influence of these demographic and socioeconomic factors on health literacy significantly change during this period of major policy and technological advancement? By analyzing trends across three time points, this study seeks not only to describe changes in health literacy but also to examine how the influence of key factors may have shifted. Through the application of mixed-effects modeling, this research seeks to disentangle the complex interplay between individual characteristics and temporal trends, thereby providing empirical evidence to inform the development of more dynamic and effective health education strategies. Materials and methods Study design and population The data for this study is sourced from Shandong Province, China. The permanent population of Shandong Province is 101.527 million, ranking second in the country in total. The survey questionnaire was conducted in three large-scale surveys in 2012, 2016, and 2020. Based on the geographical distribution and economic development characteristics of Shandong Province, a multi-stage stratified cluster random sampling method was adopted to select 16 counties (cities, districts) as survey points from the total of 136 county-level administrative divisions in the province. The selection was designed to ensure provincial representativeness by combining the socioeconomic stratification criteria from the Shandong Statistical Yearbook with the geographic classification of the province. The primary stratification ensured proportional inclusion of jurisdictions from the eastern, central, and western regions of Shandong, representing high, medium, and low levels of economic development as defined by official per capita GDP rankings. Selected locations included major urban districts, prefecture-level city cores, county-level cities, and counties to reflect the full spectrum of administrative and development types within the province. Within each selected county-level unit, households were systematically sampled from local community registries. Trained investigators then conducted in-person visits to approach potential participants at their residences for face-to-face interviews. The final sample comprised 12,302 completed questionnaires collected across the three survey waves (2012, 2016, and 2020). The selection criteria are 15 years old and above (with compulsory education), of Chinese nationality, and able to read and write. Exclusion criteria include individuals outside the age range and individuals with mental disabilities, assessed through investigator evaluation of comprehension capacity during the informed consent process. This study was conducted based on the Helsinki Declaration. Survey instrument and data collection A standardized protocol for data collection and quality control was implemented under the coordination of the Shandong Provincial Center for Disease Control and Prevention (Shandong CDC). Prior to the main survey, the questionnaire underwent a rigorous development process. Expert review was conducted by a panel comprising 9 specialists in epidemiology, health communication, and health policy from provincial and national CDC institutions. The panel evaluated content validity using a standardized rating form. Subsequently, we conducted on-site pre-testing in two non-sampled communities (one urban, one rural) with 150 participants, employing cognitive interviewing techniques to assess item interpretation and conducting quantitative analysis of response distributions. It is important to note that these pre-test participants were not included in the main study sample. Based on the integrated feedback, we modified ambiguous items, adjusted response categories, and reduced interview duration from 45 to 30 min on average. For the main study, within each selected county-level unit, households were systematically sampled from local community registries using a random start and fixed interval approach. All investigators underwent a mandatory, provincial-level training program. This training, typically conducted over a two-day session before each survey wave, covered the study objectives, a detailed explanation of each questionnaire item, standardized interviewing techniques, ethical guidelines, and protocols for handling difficult situations. To ensure competency, all investigators were required to pass a uniform qualification test. Trained investigators then conducted face-to-face interviews with participants from the selected households for the main study data collection. To guarantee data quality during and after collection, a multi-tiered verification system was employed. Firstly, at the time of interview, field supervisors conducted random spot-checks to observe and correct any deviations from the protocol in real-time. Secondly, upon completion of each questionnaire, investigators were required to perform an immediate on-site check for completeness and logical consistency. Thirdly, all questionnaires were collected by field supervisors who reviewed them again for any obvious errors or omissions before submission. Finally, during the data entry phase, a double-entry verification method was adopted by independent clerks to identify and rectify any data entry errors, ensuring a high level of data integrity. The final sample comprised 12,302 completed questionnaires collected across the three survey waves, with 4,118 in 2012, 4,085 in 2016, and 4,099 in 2020, all of which were deemed valid after quality control checks. In 2012, 5,191 individuals were invited to participate, with 4,692 completing the survey, yielding a response rate of 90.4%; in 2016, 5,243 were invited, with 4,763 completing the survey (90.8%); in 2020, 5,195 were invited, with 4,699 completing the survey (90.5%). This study employed repeated cross-sectional surveys rather than a longitudinal panel design. While the same 16 counties were maintained as survey points across all three waves to ensure geographical consistency, different households were randomly selected during each survey wave using the same systematic sampling methodology from updated community registries. This approach follows standard practice for population health surveillance studies aiming to track population-level trends rather than individual-level changes, thereby reducing the potential biases associated with longitudinal follow-up (such as attrition and practice effects) while maintaining representative sampling at each time point. The investigators also informed the participants that participation was anonymous and voluntary (for a detailed inclusion and exclusion process, refer to Fig. 2). All subjects signed an informed consent form before enrollment. The survey instrument was structured into three main sections: (1) informed consent and survey introduction; (2) health literacy assessment comprising various question formats including scenario-based items; and (3) collection of demographic characteristics and healthcare utilization information. For reference, the specific English and Chinese versions of the questionnaire used in this study are available in Supplementary Material 1. Flowchart of study design and management Ethical considerations This study received ethical approval from the Medical Ethics Review Committee of Shandong Provincial Center for Disease Control and Prevention (SDJK (K)2024-001−01). All eligible participants were informed about the study. A written information sheet and a consent form were provided for participants who wished to participate. Participants were clearly informed that their participation was voluntary and that refusing to participate would not result in adverse consequences. All personal information was kept anonymous throughout the process, and all questionnaire data was stored in a password-protected cabinet. Measurements HLQ(Health literacy Questionnaire) Health literacy was assessed using the official Chinese Health Literacy Questionnaire (56 questions, total score of 73 points) developed by the Chinese Center for Disease Control and Prevention. The instrument was categorized into three aspects: (1) knowledge and attitudes (KAA), (2) health-related behaviour and lifestyle (BAL), and (3) health-related skills (HRS); and six dimensions: (1) scientific views of health (SVH), (2) infectious diseases (ID), (3) chronic diseases (CD), (4) safety and first aid (SAFA), (5) medical care (MC), and (6) health information (HI). The questionnaire employed a combination of true/false, single-choice, and multiple-choice questions with predetermined correct answers. True/false and single-choice questions were scored 1 point for correct answers, while multiple-choice questions were scored 2 points for completely correct answers. The total health literacy score was computed as the sum of all three aspects and six dimensions. Based on the standard established by China’s national health literacy surveillance program [17, 18], participants were classified into two categories: those with inadequate health literacy (total score < 58 points, < 80% of the maximum 73 points) and those with adequate health literacy (total score ≥ 58 points, ≥ 80% of the maximum score). The questionnaire has demonstrated good reliability and validity in previous Chinese population studies [19, 20], and in our sample showed strong internal consistency with a Cronbach’s alpha coefficient of 0.919. Definition of other variables The analysis incorporated key demographic, socioeconomic, and health variables. Sex was recorded as a binary categorical variable with male and female categories. Ethnic group was categorized into Han, Hui, Miao, Uygur, Manchu, Zhuang, and other groups. Age group was classified into six categories: 15–24 years, 25–34 years, 35–44 years, 45–54 years, 55–64 years, and 65–69 years. Educational level was measured as an ordinal variable with the following categories and corresponding typical years of schooling: illiterate or minimally literate (0–2 years); primary school (6 years); junior high school (9 years); high school/vocational high school/technical secondary school (12 years); junior college (14–15 years); undergraduate (16 years); master’s degree and above (≥ 19 years). Career included categories of civil servant, teacher, medical staff, personnel from other public institutions, student, farmer, worker, other business people, and others. Household population was reported as a numerical variable summarized using median and interquartile range. Residence type was classified as urban or rural. For health outcomes, chronic disease status was determined as a binary variable based on self-reported diagnoses. Finally, health status was assessed using a self-rated health question with responses categorized as good, better, fair, worse, or poor [21]. Data analysis There are missing values in the panel data, and applying statistical methods that appropriately deal with missing data can reduce bias and improve the precision of the estimates obtained [22]. The multiple imputation method [23]is a Monte Carlo (or simulation-based) technique that uses a Bayesian inference paradigm to replace each missing value with two or more seemingly reasonable values [24]. The imputation model can be adapted to the type of variable being interpolated [25]. A linear regression model can be chosen if the outcome is a continuous variable, and a logistic regression model can be chosen if the outcome is a binary variable. Variable selection was conducted using backward stepwise regression. This method was employed not only for its conservative property of beginning with the full model containing all pre-specified candidate variables, thereby reducing the risk of omitting potential confounders, but also for its particular utility in exploratory model building with a moderate number of predictors. Our approach was specifically chosen to systematically identify the most parsimonious model from our set of candidate predictors, which included several multi-category variables (e.g., occupation, education, ethnicity) that would generate a substantial number of parameters in a full model. This methodology aligns with established practices in epidemiological research for balancing model fit with simplicity and interpretability [26, 27]. The elimination process iteratively removed the least contributive and statistically non-significant variables until all remaining predictors were significant. Model selection was guided by the minimization of the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The final six-variable model demonstrated optimal fit with an AIC of 12,847.3 and a BIC of 12,935.6, representing the lowest values among all candidate models evaluated. To ensure the robustness of the final model and address potential concerns regarding multicollinearity, we calculated Variance Inflation Factors (VIF) for all included variables. All VIF values were well below the conservative threshold of 5 (range: 1.08–2.37), indicating that multicollinearity did not substantially influence the parameter estimates in our final model. Linear models are statistical models used to study the linear relationship between explanatory and response variables [28]. The effect of the explanatory variables on the response variable can be divided into two categories, fixed effects and random effects. Random effects are relative to fixed effects and can be used to reflect heterogeneity between different objects and correlation between different observations of the same object [29]. The choice of random effects was based on the hierarchical structure of the data, treating grouping factors (sampling units) as random to allow for broader generalization and more accurate estimation of fixed effects. Fixed effects: An effect is called a fixed effect if the level of the explanatory variable that produces the effect is one of a finite number of levels of that explanatory variable and the effect of that level on the response variable is exactly what we are interested in [25]; Random effect: If the level of the explanatory variable that produces the effect is one of several levels of the explanatory variable, and the level can be seen as randomly selected from an aggregate, then the effect is called a random effect [25]. Generally, our linear models only incorporate fixed effects. However, incorporating random effects allows for better identification of individual variances across different levels or within-group disparities. In our analysis of panel data, we utilized a mixed effects model that integrates both fixed and random effects. Given the binary nature of the outcome, the analyses were conducted using the lme4 package of the R language for the generalized linear mixed effect model. Statistical analysis The study encompassed nine variables: age group, sex, household population, health status, Residence type, educational level, chronic disease, ethnic group, and career. All nine variables were initially included in a backward stepwise regression model. However, after analysis, six variables: age group, sex, household population, health status, Residence type, and educational level were selected for inclusion in the mixed effects model. Random effects and fixed effects were discussed in detail. Statistical analysis was conducted using the R programming language (version 4.1.3) and RStudio software (version 2022.02.1). Results Baseline characteristics Table 1 presents the baseline characteristics of the 12,302 participants, with data expressed as row percentages to show the distribution of health literacy status within each category. Significant disparities were observed between groups for most characteristics (p < 0.001).A strong negative gradient in health literacy prevalence was observed with increasing age. The proportion of individuals with adequate health literacy was highest among younger adults aged 25–34 (17.6%) and 35–44 (17.0%), and declined markedly to 9.2% in the 55–64 age group and 8.6% in the 65–69 age group.Substantial urban-rural disparities were evident. Urban residents exhibited nearly double the health literacy prevalence of their rural counterparts (17.5% vs. 8.9%). Educational attainment demonstrated a dramatic positive association with health literacy, ranging from only 2.9% among the illiterate or minimally literate to 44.3% among those with undergraduate degrees.Occupational distribution revealed striking variations, with medical staff showing the highest health literacy prevalence (44.6%), followed by teachers (31.9%) and civil servants (26.5%). In contrast, farmers had the lowest rate at 8.7%. Better health status was associated with higher health literacy, ranging from 14.9% among those reporting “good” health to 4.0% among those reporting “poor” health. Individuals without chronic diseases had higher health literacy prevalence than those with chronic conditions (13.5% vs. 9.8%).No significant differences in health literacy prevalence were found for sex (p = 0.685) or ethnic distribution (p = 0.493). The Han ethnic group constituted the vast majority of participants in both groups. Table 1The baseline characteristics of population by November 2022CharacteristicTotal NNon-health literacy n (row %)Health literacy n (row %)P-valueAge group< 0.001 15–24 years619527 (85.1%)92 (14.9%) 25–34 years1,5381,268 (82.4%)270 (17.6%) 35–44 years2,2291,850 (83.0%)379 (17.0%) 45–54 years3,3142,907 (87.7%)407 (12.3%) 55–64 years3,2272,929 (90.8%)298 (9.2%) 65–69 years1,3751,257 (91.4%)118 (8.6%)Residence type< 0.001 Urban5,4544,500 (82.5%)954 (17.5%) Rural6,8486,238 (91.1%)610 (8.9%)Sex0.685 Male6,1005,317 (87.2%)783 (12.8%) Female6,2025,421 (87.4%)781 (12.6%)Ethnic group0.493† Han12,22010,667 (87.3%)1,553 (12.7%) Hui5548 (87.3%)7 (12.7%) Miao44 (100.0%)0 (0.0%) Uygur21 (50.0%)1 (50.0%) Manchu32 (66.7%)1 (33.3%) Bourei33 (100.0%)0 (0.0%) Others97 (77.8%)2 (22.2%)Educational level< 0.001† Illiterate or minimally literate1,5781,533 (97.1%)45 (2.9%) Primary school2,7272,529 (92.7%)198 (7.3%) Junior high school4,7524,159 (87.5%)593 (12.5%) High school/vocational/technical2,1371,752 (82.0%)385 (18.0%) Junior college904642 (71.0%)262 (29.0%) Undergraduate17497 (55.7%)77 (44.3%) Master and above63 (50.0%)3 (50.0%)Career< 0.001 Civil servant8361 (73.5%)22 (26.5%) Teacher160109 (68.1%)51 (31.9%) Medical staff12167 (55.4%)54 (44.6%) Other public institution personnel390319 (81.8%)71 (18.2%) Student193166 (86.0%)27 (14.0%) Farmer7,8957,212 (91.3%)683 (8.7%) Worker1,163964 (82.9%)199 (17.1%) Other business people1,012794 (78.5%)218 (21.5%) Other1,2581,021 (81.2%)237 (18.8%) Household population, median (IQR)-3 (2, 4)3 (2, 4)< 0.001*Chronic disease< 0.001 No9,6808,374 (86.5%)1,306 (13.5%) Yes2,6222,364 (90.2%)258 (9.8%)Health status< 0.001† None1110 (90.9%)1 (9.1%) Good4,1963,571 (85.1%)625 (14.9%) Better4,3593,759 (86.2%)600 (13.8%) Fair3,2382,936 (90.7%)302 (9.3%) Worse402370 (92.0%)32 (8.0%) Poor7572 (96.0%)3 (4.0%)P-value from Mann-Whitney U test †P-value from Fisher’s Exact TestAll other P-values from chi-square tests*P-value for Household population from Mann-Whitney U test The baseline characteristics of population by November 2022 P-value from Mann-Whitney U test †P-value from Fisher’s Exact Test All other P-values from chi-square tests *P-value for Household population from Mann-Whitney U test Temporal trends in participant characteristics across the three survey waves are detailed in Supplementary Table 1. Several demographic shifts reached statistical significance (p < 0.001), including notable changes in age distribution, educational attainment, occupational structure, household size, and chronic disease prevalence.The educational profile showed marked improvement, with the proportion of participants with junior college education within this category increasing from 20.1% to 52.9%, and those with undergraduate degrees rising from 12.6% to 62.6%. Concurrently, the proportion of illiterate individuals decreased from 40.6% to 24.4%.Occupational patterns shifted substantially, characterized by a decline in the farming population (from 36.4% to 31.4% of farmers) and an increase in business-related occupations (from 33.5% to 51.3%). The median household size decreased from 4 to 3 persons, while the self-reported prevalence of chronic conditions increased from 29.0% to 37.6% among those with chronic diseases.In contrast, residence type (p = 0.452), sex distribution (p = 0.058), and ethnic composition (p = 0.832) remained relatively stable across the survey waves, showing no statistically significant temporal trends. Mixed effects modeling results By analyzing each of the six meaningful variables as random effect terms, three meaningful models were identified. These were the models with health status, Residence type, and education level as random effect terms, respectively. Below is the interpretation of the results of the three models. Table 2 presents the results from three distinct mixed-effects models, each examining the association of specific predictors with health literacy while controlling for core demographic variables.In Model 1 (M1), which featured educational level as the primary predictor, higher education level was strongly associated with higher health literacy, with each increasing education level corresponding to 60% higher odds of health literacy (OR = 1.60, 95% CI: 1.51–1.69, p < 0.001).Model 2 (M2), focusing on health status, revealed that better self-rated health was unexpectedly associated with lower health literacy (OR = 0.91, 95% CI: 0.85–0.98, p = 0.008).Model 3 (M3), examining residence type, showed that urban residence was associated with substantially higher health literacy compared to rural residence, with rural residents having 39% lower odds of health literacy (OR = 0.61, 95% CI: 0.54–0.69, p < 0.001). Table 2Mixed effects model resultsM1(Modal 1)M2(Modal 2)M3(Modal 3)PredictorsOdds Ratios95% CI P Odds Ratios95% CI P Odds Ratios95% CI P (Intercept)0.04(0.02,0.07)< 0.0010.02(0.00,0.06)< 0.0011.34(0.54,3.36)0.531educational level1.60(1.51,1.69)< 0.0011.60(1.51,1.70)< 0.001Residence type0.62(0.55,0.70)< 0.0010.61(0.54,0.69)< 0.001sex1.10(0.99,1.23)0.0881.10(0.99,1.23)0.0801.10(0.99,1.23)0.085household population1.06(1.02,1.10)0.0041.05(1.02,1.09)0.0051.06(1.02,1.10)0.005age group0.97(0.93,1.02)0.2100.98(0.93,1.02)0.2970.97(0.93,1.02)0.253health status0.91(0.85,0.98)0.0080.91(0.85,0.97)0.005 The table presents the fixed effects estimates from three distinct mixed-effects models (M1, M2, M3). A blank cell indicates that the variable was not specified as a fixed effect in that particular modelOR odds ratio, CI confidence interval Mixed effects model results The table presents the fixed effects estimates from three distinct mixed-effects models (M1, M2, M3). A blank cell indicates that the variable was not specified as a fixed effect in that particular model OR odds ratio, CI confidence interval Table 3 presents the estimates for random effects across three models, each examining how the influence of a key variable on health literacy varied over time (2012, 2016, 2020). These random effects capture the deviation in the slope of each variable’s effect per survey year, providing crucial information about effect heterogeneity across time. In M1, where health status was specified as a random effect, its association with health literacy showed significant temporal variation (p < 0.001). The coefficients became progressively more negative from 2012 (0.050, 95% CI: 0.022–0.080) to 2020 (−0.155, 95% CI: −0.176 to −0.134), indicating that the protective effect of better self-rated health on health literacy strengthened substantially over the study period. In M2, with residence type as the random effect, significant variation was also observed (p < 0.001). The consistently negative coefficients across all years (2012: −0.280, 95% CI: −0.463 to −0.098; 2016: −0.583, 95% CI: −0.751 to −0.414; 2020: −0.493, 95% CI: −0.649 to −0.336) confirm the persistent disadvantage for rural residents. The strongest effect was observed in 2016, suggesting particularly pronounced urban-rural disparities during that survey wave. In M3, where educational level varied randomly over time, the analysis revealed a significant declining trend in its effect magnitude (p < 0.001). While education maintained a positive effect throughout (2012: 0.615, 95% CI: 0.559–0.677; 2016: 0.501, 95% CI: 0.445–0.554; 2020: 0.365, 95% CI: 0.318–0.414), the decreasing coefficients suggest that the unique contribution of education to explaining health literacy variance diminished over time. Table 3Random effects estimatesModelFixed effectRandom effectAIC201220162020 P M1educational level Residence type household populationage groupsexhealth_status8614.80.050(0.022, 0.080)−0.087(−0.110, −0.063)−0.155(−0.176, −0.134)< 0.001M2educational level household population health statusage groupsexmonitoring_point8624.2−0.280(−0.463, −0.098)−0.583(−0.751, −0.414)−0.493(−0.649, −0.336)< 0.001M3Residence type household population health statusage groupsexeducational_level8614.70.615(0.559, 0.677)0.501(0.445, 0.554)0.365(0.318, 0.414)< 0.001 Each model tests a different random effect while controlling for specified fixed effects. Coefficients represent the deviation in each variable’s effect on health literacy for the specified year. M1 tests variation in the effect of health status; M2 tests variation in the effect of residence type; M3 tests variation in the effect of educational levelOR odds ratio, CI confidence interval, AIC Akaike Information Criterion Random effects estimates educational level Residence type household population age group sex 0.050 (0.022, 0.080) −0.087 (−0.110, −0.063) −0.155 (−0.176, −0.134) educational level household population health status age group sex −0.280 (−0.463, −0.098) −0.583 (−0.751, −0.414) −0.493 (−0.649, −0.336) Residence type household population health status age group sex 0.615 (0.559, 0.677) 0.501 (0.445, 0.554) 0.365 (0.318, 0.414) Each model tests a different random effect while controlling for specified fixed effects. Coefficients represent the deviation in each variable’s effect on health literacy for the specified year. M1 tests variation in the effect of health status; M2 tests variation in the effect of residence type; M3 tests variation in the effect of educational level OR odds ratio, CI confidence interval, AIC Akaike Information Criterion Sensitivity analysis A sensitivity analysis was conducted using complete-case analysis (without multiple imputation) to assess the robustness of our primary findings to missing data. The results from the analysis of the raw dataset (Supplementary Tables S2 and S3) were substantively consistent with those from the multiple imputation analysis presented in the main text. All key associations maintained their direction and statistical significance, with only minimal variations in the magnitude of the effect estimates. Similarly, the temporal patterns of the random effects were preserved. This consistency suggests that our primary conclusions are robust and not substantially influenced by the missing data handling method. Discussion and conclusion This study examining health literacy determinants across three time points (2012, 2016, 2020) reveals significant fluctuations in their relative importance over time. Educational attainment consistently predicted health literacy, though its effect size diminished in later years. Conversely, the association between self-rated health and health literacy strengthened progressively, with better health becoming an increasingly important correlate. Residential disparities demonstrated notable variation, with the urban-rural gap peaking in 2016 before moderating in 2020. These temporal dynamics highlight that health literacy influences are not static, suggesting the need for periodically updated intervention strategies that account for these evolving patterns. Our findings reveal a clear age-related gradient in health literacy, with prevalence declining substantially from 17.6% among adults aged 25–34 to only 8.6% in those aged 65–69. This pattern of decreasing health literacy with advancing age aligns with observations from diverse populations. Studies in other Asian countries like Japan and South Korea have reported similar age-related disparities, while research in Western populations, such as the European Health Literacy Survey, also confirms this persistent generational gap [30–32]. Several interconnected mechanisms may explain these persistent generational gaps. Older adults face multiple barriers in accessing and processing health information, including limited digital literacy, decreased familiarity with navigating complex online health platforms, and potential age-related cognitive declines that affect information processing speed and memory [33]. Furthermore, vision and hearing impairments more prevalent in older populations, may hinder both access to and comprehension of health materials [34]. The digital transformation of health information presents particular challenges for older adults, who often struggle with the rapid transition to e-health services and mobile health applications [35]. This consistent cross-cultural pattern suggests that the digital transformation of health information presents a universal challenge for older adults, who may face greater difficulties in accessing and processing health information through new media platforms [36–38]. However, the magnitude of this age-related disparity in the Shandong Province of China cohort appears more pronounced than in some Western populations, possibly reflecting China’s particularly rapid digital transition and unique socioeconomic development trajectory. Therefore, while community-based health education and improving environmental health literacy friendliness remain crucial [39], our findings specifically highlight the urgent need for age-friendly digital health strategies that account for China’s unique context and pace of change. The persistent urban-rural disparities in health literacy identified in our mixed-effects models reflect profound structural inequities in healthcare resources and access. These disparities manifest concretely in several ways: rural areas in Shandong Province face significant shortages of healthcare professionals, with physician density per thousand population substantially lower than in urban areas [40]. Additionally, rural residents often encounter geographical barriers in accessing tertiary hospitals and specialized care, leading to delayed diagnoses and treatment. The limited healthcare infrastructure in rural communities further restricts opportunities for routine health education and counseling that rural residents typically receive during regular medical visits. This resource disparity creates a self-perpetuating cycle where lower health literacy leads to poorer health outcomes, which in turn further limits capacity for health information processing and healthcare navigation. The particularly pronounced disparity observed in 2016 may reflect accelerated urbanization during this period, where urban health resources expanded more rapidly than rural ones, widening the existing gap. Addressing these fundamental inequities requires not only health education interventions but also substantial investment in rural healthcare infrastructure and workforce development. In the study, the initial effect of health status on health literacy was the smallest in 2012, suggesting that earlier health literacy efforts may not have been more widely applied. Interestingly, there was an inverse relationship between self-perceived health and health literacy in the same year. This may indicate a compensatory effect, whereby individuals with poorer health are more active in seeking health information to manage their condition. This finding aligns with some previous research suggesting that individuals with poorer health status may become more actively engaged in health information seeking as a coping mechanism to manage their conditions [41]. However, this pattern could also reflect the complex, bidirectional relationship between health literacy and health outcomes documented in other studies [42, 43], where initial health challenges may prompt information engagement that doesn’t immediately translate into improved health literacy scores due to measurement limitations or time lags. However, the trend in 2016 and 2020 that better self-perceived health corresponds to higher health literacy likely reflects the maturation of health literacy interventions following the launch of the Healthy China 2030 program in 2016. This national strategy has specifically promoted health literacy through several mechanisms: extensive public health campaigns through traditional and new media, the integration of health literacy education into primary care services, and the development of standardized health communication materials. The most effective aspects appear to be the widespread dissemination of core health knowledge through multiple channels and the training of healthcare providers in patient communication. However, significant gaps remain in reaching vulnerable populations, particularly in rural areas where access to both digital and traditional health resources remains limited. These policy impacts are supported by multiple studies demonstrating improved health literacy following targeted interventions [44, 45]. The evolving relationship between self-rated health and health literacy suggests that while the initiative has successfully raised general awareness, future efforts need to address the social determinants of health literacy more comprehensively. Therefore, we strongly advocate for enhanced implementation of comprehensive, governmental, and multisectoral strategies to improve rural health literacy. Successful models from other regions offer valuable insights. A community-based intervention in rural Japanese towns demonstrated remarkable success by combining public awareness campaigns with empowerment of residents and civic participation, through welfare programs that promoted social engagement among senior citizens [46], while Thailand’s “Village Health Volunteer” program demonstrated how trained local laypersons can bridge health literacy gaps in rural communities [47]. Building on such evidence, we recommend: increased public investment in rural digital health infrastructure, enactment of supportive regulations for standardized health communication, and establishment of community health worker programs adapted to China’s rural context. These approaches should prioritize meaningful public participation by rural residents throughout implementation [48]. Education plays a crucial role in improving health literacy, though its relative influence has diminished from 2012 to 2020. This trend may reflect several factors: while formal education establishes fundamental knowledge and cognitive skills, its standardized curriculum may not adequately address the rapidly evolving health information landscape [49]. Concurrently, informal learning channels - including digital media, community health programs, and peer networks - have become increasingly significant in disseminating timely health information. The observed pattern suggests a “catch-up” effect where basic health knowledge has become more widespread, reducing education’s marginal contribution. This transition indicates that health literacy development is shifting from relying primarily on formal education to requiring complementary informal learning mechanisms that can better address contemporary health challenges and information dynamics. On the other hand, it is possible that the presence of new media has diversified the accessibility of education, narrowing the gap between education and making health literacy more accessible. Chinese citizens are increasingly utilizing online sources of information to take care of their own and their families’ needs. By the end of 2023, the number of Internet users in China reached 1.092 billion, with an Internet penetration rate of 77.5% [50]. Evidence suggests different digital platforms show varying effectiveness across demographics - social media like WeChat and Douyin reach younger urban populations effectively, while specialized health forums better serve older adults seeking detailed information [51]. To maximize the synergy between formal education and new media, schools and universities should integrate digital health literacy into curricula, equipping students with skills to critically evaluate online health information and navigate digital health resources effectively [52].New media outreach and education are critical to address health literacy disparities between urban and rural areas, education levels, and age groups. Public health initiatives should leverage these platform-specific usage patterns to develop tailored interventions, such as simplified mobile applications for rural residents and interactive content for younger generations. It is recommended “Internet+” interventions because Internet + healthcare services have become an emerging healthcare service model in China, aiming to improve accessibility and quality of healthcare services [53]. At the same time, more personalized and interactive health education content can be designed for different age groups to improve the health literacy level of the population. Through online platforms, health knowledge can be effectively disseminated to a wider audience(Fig. 3). Intervention strategies and policy recommendations for improving health literacy in China The interplay between these patterns and overall health policy and the emergence of new media presents a complex picture. The COVID-19 pandemic has accelerated changes in health information consumption, highlighting the need for agile policy frameworks. To effectively integrate new media into health literacy campaigns, we recommend strategies: (1) developing official health authority accounts on popular platforms (WeChat, Douyin) to disseminate verified information through short videos and interactive content; (2) establishing rapid response mechanisms to counter health misinformation during public health emergencies; and (3) creating digital health literacy training modules for community health workers to enhance their capacity for social media-based health education. The “Healthy China” initiative should incorporate these approaches, particularly by allocating dedicated resources for new media operations and developing evaluation metrics to assess the reach and impact of digital health communication (Fig. 4). Schematic framework of China's constantly changing environment and challenges reflected in the improvement of health literacy in China In conclusion, this study identifies key determinants of health literacy in Shandong Province, revealing a pronounced age gradient, persistent urban-rural disparities, and the evolving role of education. To address these findings, we recommend: (1) establishing digital health literacy training for older adults through community centers; (2) integrating critical evaluation of online health information into school curricula; and (3) developing simplified health communication protocols for rural patients. Future research should systematically address several critical questions emerging from this study, particularly focusing on vulnerable populations including rural older adults with limited digital access and individuals with lower educational attainment. Key research priorities should include: longitudinal studies to track health literacy development trajectories, evaluation of tailored intervention models for specific population subgroups, and comparative studies of different delivery approaches to determine optimal resource allocation strategies. These research directions would substantially advance both theoretical understanding and practical interventions for health literacy improvement in China. Limitations Although this study provides valuable insights into health literacy in Shandong Province, several limitations should be acknowledged. First, the generalizability of our findings to other regions of China may be limited, as the sample was drawn exclusively from Shandong Province, which may differ socioeconomically from other areas such as less-developed western provinces or major metropolitan centers. Second, the assessment of chronic disease status was based on self-reported diagnoses collected through household interviews. While trained interviewers provided standardized explanations of chronic conditions and specifically asked about physician-diagnosed diseases, we were unable to verify these self-reports against medical records due to privacy protections and field study constraints. This may introduce potential reporting bias. Future research would benefit from incorporating medical record verification where feasible, expanding geographical coverage, and employing longitudinal designs to track health literacy dynamics. Conclusion In conclusion, our analysis of the health literacy of the Shandong population in China shows that although progress has been made in health literacy, significant challenges and disparities remain. The changing role of new media, the urban-rural divide, and the fundamental importance of education are all key factors that must be taken into account in future health literacy initiatives. As we move beyond pandemics, it is clear that multifaceted approaches to addressing the changing dynamics of health information and its consumption are critical to promoting health literacy and ultimately achieving healthier populations. Supplementary Information Supplementary Material 1. Supplementary Material 2. Acknowledgements AcknowledgementsThe authors thank the staff of the Shandong CDC for technical support. Norm This study was conducted based on the Helsinki Declaration. Abbreviations Health Literacy Questionnaire Odds Ratio Model 1 Model 2 Model 3 Shandong Provincial Center for Disease Control and Prevention Medical Ethics Review Committee Coronavirus Disease 2019 Linear Mixed-Effects Models using Eigen and S4 RStudio Authors’ contributions Conceptualization: L.R.; Methodology: X.X., R.L.; Investigation: L.G., T.L., J.H., F.C., F.D.; Formal Analysis: R.L., L.G., T.L., J.H., F.C., Y.L., W.L., S.G., F.D.; Writing – Original Draft: X.X., R.L., J.H., F.C., L.W.; Writing – Review & Editing: X.X., R.L., J.H., F.C., L.W., S.G.; Supervision: R.L. All authors have read and agreed to the final manuscript. Funding This work was funded by research grants from the Project of Shaanxi Academy of Environmental Sciences (202208146), Shandong Province Medical and Health Science and Technology Development Plan Project(202312071011)and Shandong Province Humanities and Social Sciences Research Projects(2023-JKZX-11). The funders had no involvement in data collection and analysis or the preparation of this article. The analysis and interpretation of the evidence was done by the authors themselves, not by the funders. Data availability The datasets obtained during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate All respondents have informed consent forms. Ethics Approval Opinion of the Medical Ethics Review Committee of Shandong Provincial Center for Disease Control and Prevention (SDJK (K) 2024-001-01) Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Rui Li and Xinyu Xu contributed equally to this work. Contributor Information Fangyao Chen, Email: chenfy@xjtu.edu.cn. Liansen Wang, Email: wlsen@163.com. Jing Han, Email: bbbishop@126.com. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Data Availability Statement The datasets obtained during the current study are available from the corresponding author on reasonable request.