Health literacy (HL) is a crucial determinant of individual and public health, significantly affecting the capacity to access, comprehend, and use health-related information [1-3]. Low HL correlates with adverse health outcomes, diminished use of preventive services, and elevated health care costs [1]. In increasingly complex information environments, individuals must not only process health information but also assess its relevance and reliability. The advent of digital media has transformed public engagement with health information through the widespread availability of online resources and tools [4-6]. Repeated engagement with online health resources may foster HL by familiarizing users with medical terminology; supporting navigation of the health system; and providing recurring opportunities to locate, appraise, and apply health information, particularly when trustworthy sources are used [6-8]. At the same time, disparities in HL persist across Europe, often mirroring gradients of age, education, and social status [9,10]. In multilingual regions, such disparities may be further influenced by language-related differences in access to and trust in the information environment [11]. South Tyrol, a predominantly bilingual province in northern Italy where German and Italian are co-official languages, provides a particularly pertinent context for examining these dynamics: its shared health care infrastructure and coexisting language communities allow HL variations to be investigated under stable institutional and policy conditions.

A recent study conducted by Ausserhofer et al [12] represents the inaugural population-based HL assessment in South Tyrol, using the 16-item Health Literacy Survey European Union Questionnaire (HLS-EU-Q16) instrument. The study identified significant disparities between the language groups, with Italian speakers exhibiting higher HL than German speakers. These differences persisted even after controlling for variables such as age, education, chronic illness, and professional background in health and social care. Notably, domain-specific analyses indicated that German speakers scored low in terms of health promotion (HP) literacy. Concurrently, Wiedermann et al [13] investigated the usage of health information and trust in the same regional context. Their findings suggest that Italian speakers use and trust digital health information sources more frequently, whereas German speakers predominantly rely on traditional or interpersonal sources. These variations in engagement with digital health information may elucidate HL disparities that are not solely attributable to sociodemographic factors. Collectively, these studies propose that South Tyrol serves as a natural model for examining the interaction between digital information behavior and linguistic background in shaping HL. The region’s bilingual infrastructure, shared health care system, and controlled regional variation provide a quasi-experimental setting for exploring the mechanisms underlying HL inequalities.

This study investigated whether variations in engagement with digital health information mediate the relationship between language group (Italian vs German) and HL. Furthermore, we examined trust in health information from professional sources and internet-based platforms as potential independent correlates of HL. By examining behavior and perceptual factors associated with HL disparities, this study contributes to the formulation of targeted, culturally, and linguistically responsive public health strategies.

This study used a cross-sectional survey conducted between March and May 2024 among residents of South Tyrol, a multilingual and autonomous province in northern Italy. Eligible participants were individuals aged 18 years or older, proficient in either German or Italian, and randomly selected from the provincial resident registry through stratified sampling based on language groups. The sample size was determined by the established survey methodology and operational expertise of the Provincial Institute of Statistics of the Autonomous Province of Bolzano-South Tyrol (ASTAT). A stratified random sample of 4000 adult residents of South Tyrol was drawn using the SURVEYSELECT procedure in SAS (version 9.2; SAS Institute Inc), with subsequent poststratification weighting performed in ReGenesees (version 2.3; Italian National Institute of Statistics). The stratification criteria included age group (18‐34, 35‐54, and 55+ years), sex, citizenship, and municipality of residence. The questionnaire was administered in a mixed mode design: participants received a bilingual postal invitation containing an individual access code to complete the questionnaire online via the LimeSurvey platform (LimeSurvey GmbH), and respondents who preferred to could instead complete an enclosed paper version and return it by post.

Sociodemographic characteristics included self-reported sex (male or female), age (in years), and age group (18‐34, 35‐54, and 55‐99 years). Residence was classified as urban or rural based on municipal criteria. Educational attainment was categorized into 4 levels: middle school or less, vocational school, high school, and university. Participants also reported whether they were employed in the health or social care sector (yes or no).

Health status was evaluated using a single self-rated health item on a scale from 0 (worst imaginable health) to 100 (best imaginable health). The burden of chronic illness was assessed using a checklist of prevalent conditions, including cardiovascular, respiratory, metabolic, oncological, and mental health disorders. A binary indicator for the presence of any chronic condition was derived (yes or no), along with a variable indicating the number of chronic diseases (range: 0‐9) per respondent. These indicators have been previously documented [13].

Trust in professional health information was measured using a 4-item index reflecting trust in doctors, pharmacists, health institutions, and the information provided by them. The items were rated on a 4-point scale, ranging from “very much (1)” to “not at all (4),” coded inversely, and aggregated to form a trust index [13], with higher scores indicating greater trust. Trust in the internet was assessed separately with a single mistrust item, again ranging from “very much (1)” to “not at all (4),” and it was coded inversely as well.

Digital health information use was operationalized through 4 items assessing the frequency of using search engines, passive browsing, forums, and social media for health-related purposes, on a 4-item scale ranging from 1 (regularly) to 4 (never); item scores were inversely rescaled and summed to create a composite score (range: 4‐16), with higher scores indicating more frequent digital health information use [13].

HL was assessed using the validated HLS-EU-Q16 questionnaire, implemented in German and Italian, with validated translations that demonstrated robust psychometric properties [9,14]. All 16 items were scored on a 4-point difficulty scale, dichotomized according to standard procedures, and summed to yield a raw score of 0 to 16. Respondents with fewer than 14 valid responses were excluded from the HL-score calculation.

Consistent with the procedure reported by Ausserhofer et al [12], domain-specific scores were computed for the 3 HL dimensions represented in the HLS-EU-Q16: (1) health care (HC), (2) disease prevention (DP), and (3) HP. Each domain score was calculated by summing the dichotomized responses for the items assigned to the respective domains, as outlined in the HLS-EU conceptual framework. Only respondents with complete data (ie, all 16 items answered) were included in the domain-level analysis to ensure comparability across dimensions.

Ethical approval for this study was granted by the Institutional Review Board of the Institute of General Practice and Public Health, Bolzano, Italy (Protocol: 20/11/2023, November 20, 2023). The survey was incorporated into the 2024 Statistics Program of South Tyrol, which received approval from the Government of the Autonomous Province of Bolzano, Italy. The data collected for this research are protected under the Italian Statistics Act (Article 9, Legislative Decree 322/1989) and adhere to the regulations for personal data protection (EU Regulation 679/2016 and Legislative Decree 196/2003, as amended by Legislative Decree 101 of August 10, 2018). This study was conducted in accordance with the principles of the Declaration of Helsinki. Sampling and fieldwork were conducted by ASTAT within the official provincial statistics program; the research team received only deidentified records, and no information permitting identification of individual respondents was accessible during analysis. No financial compensation was provided.

Participation in this study was voluntary. Prior to completing the online questionnaire, participants provided informed consent. Completing the paper questionnaire and returning it by post was also regarded as an indication of informed consent. Although the survey adhered to official statistical procedures delineated by Italian law (Legislative Decree 322/1989), participants were not explicitly given the option to withdraw consent after completing the survey. Nevertheless, individuals retained the right to decline participation or request removal from the sample prior to submission.

Of the 4000 individuals invited, 2120 returned the questionnaire, yielding a response rate of 53% (2120/4000). Responses were collected anonymously. Among them, 2090 adults provided complete responses and were included in the analysis, corresponding to a completion rate of 98.6% (2090/2120). Cases with insufficient completion of the HLS-EU-Q16 instrument (ie, fewer than 14 of the 16 items answered) were excluded according to the standard scoring guidelines. After exclusion of cases with insufficient HLS-EU-Q16 completion, 1642 respondents remained. Of these, 142 had missing data on one or more model covariates and were excluded listwise, yielding a final analytic sample of 1500 participants (1080 German-speaking and 420 Italian-speaking) for the mediation analysis (Figure 2).

A post hoc power analysis was conducted using G*Power 3.1.9.6 (Heinrich-Heine-Universität) to estimate the statistical power for detecting a small effect with 7 predictors. Assuming a small effect size (f²=0.02), α=.05, and a total sample size of 1500, the analysis yielded a power exceeding 99%, indicating that the sample was well powered to detect the small effects examined in the mediation model.

The sample characteristics are presented in Table 1. Statistically significant differences were observed between German- and Italian-speaking participants across several sociodemographic and health-related variables.

The educational attainment, age distribution, and residency status differed significantly between the groups. Italian speakers were more frequently urban residents, older, and more likely to have completed a university education. German speakers had higher rates of vocational school attendance than Italian speakers. Italian-speaking participants reported lower self-rated health, a higher prevalence of chronic diseases, and a greater number of comorbidities. While trust in health information from health professionals and on the internet was comparable between groups, digital use scores were higher among Italian speakers. HL scores, both overall and in the domains of HC, DP, and HP, were also higher among Italian speakers. These differences were statistically significant, although the effect sizes were small. No significant group differences were found in sex distribution or employment in the health or social sectors.

Pearson correlations between sociodemographic, health, trust, and information-related variables are shown in Table 2. HL showed a small positive correlation with education, subjective health status, and trust in internet health information, as well as with trust in health information from health care professionals. It was also positively correlated with the language group (Italian speakers) while showing a small negative correlation with the number of chronic diseases.

Language group membership (Italian vs German) was positively correlated with HL and was weakly but significantly associated with higher digital health information use. Digital use of health information was strongly associated with trust in internet health information and was moderately associated with younger age and a lower number of chronic conditions.

Age was broadly associated with other predictors, showing moderate-to-strong correlations with education, health status, the number of chronic diseases, and digital use. However, age showed no significant correlation with HL. Sex was minimally correlated with digital use but was unrelated to HL.

These findings informed the selection of covariates for the mediation analysis. Trust in internet health information emerged as a central correlate of digital health use.

To investigate whether the use of digital health information mediates the relationship between language group (Italian vs German) and HL, a mediation analysis was conducted using PROCESS Model 4. The final model included the language group (X) as the predictor, the use of digital health information (M) as the mediator, and the HL score (Y) as the outcome. Covariates entered in both equations were age, sex, educational level, self-rated health status, trust in professional health information, and trust in internet health information. The results are presented in Table 3.

The total effect of language group on HL was significant (B=0.98, 95% bootstrap CI 0.61 to 1.35), with Italian speakers showing higher HL scores than German speakers. Path a (effect of language group on digital health information use) was positive and significant (B=0.72; P<.001), indicating greater digital health information use among Italian speakers. However, path b (effect of digital health information use on HL) was not significant in the fully adjusted model (B=0.02; P=.49). Consequently, the indirect effect was small and nonsignificant, with a percentile bootstrap CI that included zero (B=0.018, 95% CI –0.035 to 0.074). Digital health information use therefore did not mediate the relationship between language group and HL. The direct effect of language group remained substantial and significant (c′: B=0.96; P<.001), indicating that the language-based HL difference is not accounted for by digital health information use. Among the covariates, trust in professional health information (B=0.39; P<.001), trust in internet health information (B=0.61; P<.001), self-rated health status (B=0.04; P<.001), and age (B=0.02; P<.001) were independent positive predictors of HL.

In summary, digital health information use did not mediate the association between language group and HL. Trust in professional and internet health information emerged as the more substantial correlates of HL, underscoring the role of informational trust over mere digital engagement.

The mediation model (Model 4) demonstrated an overall satisfactory explanatory power. The model predicting digital health information use (mediator M) accounted for 39% of the variance (R²=0.39; P<.001), indicating that the combination of language group, trust in internet health information, age, sex, educational level, self-rated health status, and trust in professional health information explains a substantial portion of the variance in digital information behavior.

For the outcome model predicting HL, the model explained 15% of the variance (R²=0.15; P<.001). Within this model, trust in professional health information, trust in internet health information, and self-rated health status were the strongest positive predictors, while digital health information use was not significantly associated with HL once these factors were accounted for.

This study examined the association between language group membership and HL in a linguistically diverse region, focusing on whether the use of digital health information mediates this relationship. Consistent with previous research conducted in South Tyrol, Italian-speaking participants demonstrated higher HL scores than their German-speaking counterparts. Expanding on prior findings, the current analysis identified that the use of digital health information did not mediate language-related differences in HL: the indirect pathway was small and its CI included zero, whereas the direct language group effect remained robust. The comprehensive mediation model further revealed that lower trust in professional health information was independently associated with lower HL. These findings highlight the complex interplay between language group, digital engagement, and informational trust in shaping HL within a universal health care system.

Several limitations constrain the interpretation of the findings of this study. First, the cross-sectional design precludes any causal inference regarding the relationships between language group, digital health information use, trust, and HL. Although mediation models can suggest pathways of association, the temporal ordering of these factors remains unclear [41]. Reverse pathways are equally plausible: individuals with higher HL may be more selective in their digital engagement and more discerning in whom they trust, thus the associations modeled here as predictors of HL may partly reflect its consequences. Second, although validated Italian- and German-language versions of the HLS-EU-Q16 were used, unmeasured cultural factors may have influenced both HL and digital engagement. Despite linguistic validation, the conceptual equivalence of HL assessments across cultural and social contexts cannot be fully assured [42].

Furthermore, the measurement of digital usage captured frequency but did not account for the depth, quality, or the critical evaluation of information. Similarly, trust was assessed using single items or brief indices, which may oversimplify a multidimensional construct that evolves through experience and context. Additionally, approximately 21% (448/2090) of respondents had incomplete responses to the HLS-EU-Q16 and were excluded, potentially introducing selection bias and affecting the generalizability of the HL estimates [13].

Although age was statistically controlled for in all multivariate models, the higher average age of the Italian-speaking cohort may have introduced residual confounding. Age is intricately associated with digital engagement and trust dynamics, potentially attenuating or obscuring the stronger associations between language group, digital use, and HL.

The sample was limited to the 2 predominant language groups in South Tyrol, thereby excluding Ladin speakers, migrants, and other minority populations in South Tyrol. This restriction constrains the generalizability of the findings to broader multilingual and multicultural contexts.

Despite these limitations, this study has several strengths. It is based on a large, population-based sample with a high response rate, enhancing the generalizability of findings within the South Tyrolean context. The use of validated HL instruments in both German and Italian ensures linguistic comparability. Additionally, the comprehensive analysis, including mediation modeling and consideration of trust and digital engagement, provides novel insights into the complex interplay between language, digital behaviors, and HL in a multilingual region.

Future research should encompass a broader range of linguistic and cultural groups and integrate variables such as migration background, acculturation, and digital competencies. Longitudinal or mixed methods studies could elucidate the temporal dynamics between digital engagement, trust, and HL development. Using more nuanced measurements of trust, including multidimensional scales, alongside assessments of digital information quality and processing skills would yield a deeper understanding of these relationships. Addressing item nonresponse through multiple imputation and examining patterns of missingness may also enhance the robustness and representativeness of HL assessment.

This population-based study from South Tyrol confirms that language group membership is a critical social determinant of HL in multilingual regions. Italian-speaking participants exhibited higher HL levels and more frequent engagement with digital health information than their German-speaking counterparts. However, the use of digital health information did not significantly mediate the observed differences in HL between the language groups. Moreover, trust in internet health information, while strongly predictive of digital use, was positively associated with HL but did not contribute to explaining the language-based disparities.

These findings suggest that while digital engagement and trust in online sources are positively linked to HL, they are insufficient on their own to overcome the structural and cultural barriers that contribute to language-based disparities. A more comprehensive strategy that integrates digital engagement with strengthened critical appraisal skills and targeted support for minority language groups is necessary to address these inequalities.

Therefore, addressing HL inequalities requires public health strategies that go beyond enhancing digital access. Interventions must focus on strengthening critical appraisal skills, promoting digital HL, and fostering trust in professional health information sources. In multilingual regions such as South Tyrol, tailored approaches that account for linguistic, cultural, and informational contexts are essential for equitable health care communication.

Future research should use longitudinal and mixed methods designs to better understand the temporal dynamics among digital engagement, trust, and the development of HL. Including underrepresented linguistic minorities and incorporating multidimensional measures of trust, digital literacy, and cultural orientation will be vital to informing effective, equity-oriented health policies and education in diverse societies.