Among other factors, susceptibility to health mis/disinformation has been independently linked to individuals’ levels of media literacy (ML)^‡ and health literacy (HL). Health literacy interventions aim to reduce negative health outcomes associated with an inability to obtain, understand, or use essential health information and services. Most HL research focuses on older adults, people with chronic illnesses, and individuals with limited English proficiency. In contrast, ML research typically concentrates on adolescents, emphasizing critical inquiry, empowerment and self-reflexivity. Evidence shows that each construct is independently related to health-related beliefs and behaviors (12, 13). While each play an important role in informed decision-making, neither literacy alone accounts for the set of knowledge, abilities and practices (KAPs) that manifest at their intersection — that is, adult media health literacy.

As a sub-literacy within the overarching domain of ML, adult media health literacy (AMHL) represents a cumulative ability to: (1) access health-related information from various sources; (2) identify key elements of message construction; (3) critically evaluate message credibility, quality and relevance; (4) produce content using a variety of media tools; and (5) engage with a global media culture (76). Each of these is reinforced by specific KAPs, outlined in Supplementary Material 2, that fluctuate throughout the decades-long course of adulthood.

The available body of MHL scholarship reveals four major limitations. The first is conceptual, with little known in regard to how AMHL manifests and whether or how it is linked to health beliefs, behaviors and outcomes. As reported in a preceding scoping review, the vast majority of MHL studies do not reference an explicit theory or model and those that do lack a formal representation or discussion of their framework (76).

Likely stemming from an insufficient theoretical understanding of how ML operates in health contexts, what research exists at the intersection of adult ML and health misinformation tends to focus on more politically charged issues, such as COVID-19 vaccination. Media literacy studies that examine susceptibility to science misinformation, for example, generally account for partisanship or critical reasoning capacities but tend to neglect the unique considerations associated with health behaviors. These can include barriers to high-quality health care and health information; the higher stakes of bodily decision-making; the disproportionate influence of anecdotal evidence; and the role of situational literacy (i.e., how stress can temporarily impair one's cognitive processing) (14). A review of the literature on individual susceptibility to mis/disinformation reveals that most studies focus on a small number of prominent political issues, such as gun control and climate change. As others have pointed out, the findings from these contexts are not necessarily generalizable to health and medicine—hence the need for intersectional constructs such as AMHL (15–19).

The second area in need of greater refinement is methodological. No AMHL study to date has been conducted with an instrument that employs direct measurement (i.e., performance based). Existing instruments for adults measure ML and HL separately, failing to capture a conceptual overlap, and are overly dependent on self-assessment measures. Significant issues of validity associated with self-report questionnaires for knowledge and skills assessments have been well documented (20–25). Yet, they remain prominent due to the burden of analyzing open-ended data, challenges of scale, and the design costs of task- and performance-based measures.

The third issue pertains to sampling. To date, no AMHL studies have been conducted with a representative sample of adults and what adult studies exist largely under-represent working-age adults, generally classified in the U.S. as those between the ages of 25 and 54. This period marks a time when adults are in their critical preventive health care years and most likely to care for children and/or elderly dependents, meaning they not only make consequential decisions for themselves, but also on behalf of their loved ones. The affordances and implications of AMHL among this demographic of health decision-makers are profound and span generations.

Finally, there is a clear need for the inclusion of diverse outcome variables that explore relationships beyond saturated health topics in health misinformation studies, such as vaccination. While there is a great deal of research dedicated to vaccine hesitancy, relatively little is known as it pertains to adult ML and other health topics associated with high levels of digital misinformation. Examples include misinformation on smoking products, diet, sunscreen, reproductive care, and mental health (5, 11, 26–28). Given that chronic non-communicable diseases account for eight of the 10 top causes of death in the U.S., further attention is warranted.

Cigarette smoking, for example, remains the leading cause of preventable disease. In 2022, roughly 12% of adults reported smoking cigarettes and 6.5% reported vaping, with rates rising (29, 30). Breast cancer affects one in eight American women and despite the life-saving role of screenings, recent estimates indicate that half of women aged 45–54 have not received a recommend mammogram under current guidelines (31). Colon cancer is the second most common cause of cancer deaths among all adults, yet roughly 30% of those aged 50–75 have not received recommended screenings (32). Lastly, while yearly medical exams for younger healthier adults are generally deemed unnecessary (33), annual visits are recommended for those 40 and older and have been shown to improve health outcomes and reduce mortality through preventive care screenings and interventions. As of 2020, only 5.3% of adults age 35 and older received all high-priority clinical preventive services, including but not limited to screenings for breast, colon, lung and cervical cancer (34). This marks a 5% decrease from 2015.

A wide range of socio-ecological contributors have been documented as predictors of both preventive health care engagement and risky health behaviors (35), though the role of contemporary factors such as ML have remained largely unexplored.

The hypotheses are grounded in a socio-ecological model of health (SEM), which contends that human development is shaped by a series of interconnected variables across five levels: the individual, social, cultural, political, economic and chronological (36).^‖ These variables are often collectively referred to as social determinants of health (SDH), defined by the World Health Organization as the nonmedical factors that shape human health, such as the “conditions in which people are born, grow, work, live, and age, and the wider set of forces […] shaping the conditions of daily life,” including social norms, policies, and economic systems (37).

As illustrated in Figure 1 and Supplementary Material 1, the SDH are organized under five overarching categories and operate as an interdependent system (38). This means that modifying one variable can have potential domino effects. For example, while the SDH of “access to primary care health” is empirically linked with vaccination, other SDH such as higher education, employment, discrimination and ML moderate the relationship. To date, uni- and bi-directional relationships have been documented between ML and eight standing SDH variables: (1) early childhood development and education; (2) enrollment in higher education; (3) high school graduation; (4) health literacy; (5) language and literacy; (6) employment; (7) social cohesion; and (8) civic participation. As shown in the expanded model, ML operates as the 20th SDH within the Social and Community Context block. For a more comprehensive overview of the supporting evidence for this model, see Ashtari (76, pp. 1238–1241).

In this study, AMHL represents a formative construct, meaning causality flows from the Index items to the construct (Figure 2). Hence, the Index comprises a census of items that serve as an explanatory combination of unequally weighted indicators (39). An initial battery of 45 performance items was developed based on the 17 evidence-based KAPs listed in Supplementary Material 2. Each item was scored using a companion rubric that yielded a composite AMHL score. Questions and Index scoring are detailed in Supplementary Material 3.

The following section describes the development and validation of the Index, which reflects best practices established by experts in index methodology (40–48).

The draft instrument and scoring rubric were first reviewed by five subject-matter experts in survey methodology, health communication, ML, and quantitative and qualitative methods. Their feedback validated the Index's theoretical structure (Figure 2). Next, pre-testing was undertaken with a convenience sample of English-proficient adults aged 22–65 (N = 75). Participants submitted anonymous feedback on question wording, as well as the survey's duration, user friendliness, and technical accessibility.

Cognitive interviews were then conducted to reduce measurement error by optimizing question comprehension and response completeness. Following think-aloud protocols, eight hours of one-on-one Zoom interviews were conducted with six working-age adults who were compensated with $20 Amazon gift cards (49). They completed the survey aloud, explaining their thought process, rationale and hesitation around sensitive questions. As a result, select items were removed or reframed to reduce cognitive shortcuts and minimize social desirability effects. The total item count was reduced from 45 to 28.

A second pilot was conducted in March 2023 with 53 target population members who were recruited through Amazon Mechanical Turk and compensated $7. They completed an anonymous survey that included the Index and a slate of health outcome measures. Closed-ended questions were accompanied by required open-ended fields where participants explained how they interpreted questions, analyzed images and videos, and formed their answers. The qualitative data were manually assessed alongside their corresponding closed-ended questions as a cross-validation technique. The item count was reduced from 28 to 21 and the scoring rubric refined and finalized.

The final step involved assessing the AMHL Index's external nomological validity, or the extent to which the measure meaningfully correlates with the theoretically-related outcome variables posited in the hypotheses. Toward this end, the final survey was fielded as part of a larger correlational study. Assuming a 5% alpha, 80% power and one predictor variable (Index score), a minimum sample size of 568 was required as calculated by G-Power statistical software. A non-probability sample of U.S. adults was recruited through Prolific to match national census data for distributions on age, gender, race, and education (50, 51–55). Sociodemographic characteristics are displayed in Table 1.

The Qualtrics survey, conducted between July and December 2023, is provided in Supplementary Material 2. Eligible participants were aged 25–64, English proficient and residing in the U.S. They received $4 per submission ($13.33 per hour). The full survey included two attention-checks; 21 AMHL Index items; seven sociodemographic questions; and eight health-related outcome measures: mammography, colon cancer screening, COVID-19 and flu vaccination, annual exam attendance, smoking, vaping, and chronic condition diagnosis. A total of 603 participants completed the anonymous survey. Fourteen were removed for failing attention checks or straightlining, resulting in a final sample of 589, exceeding the required minimum.

First, regression analyses were conducted in R 4.4.2. to inspect variance inflation factors (VIF), a standard metric of collinearity. High multicollinearity—which is when items (“indicators”) are redundant and too strongly correlated with one another—can undermine the stability and interpretability of indicator weights in formative models, which are intended to comprise a range of items with distinct contributions. Unlike reflective measures,¶ formative indicators collectively form the construct rather than reflecting an underlying latent factor. Items are not intended to be interchangeable and removing one would theoretically alter the meaning and integrity of the whole.

Next, Partial Least Squares Structural Equation Modeling (PLS-SEM) was employed to establish nomological validity—i.e., whether scores on the Index significantly correlate with the outcomes of interest. PLS-SEM is the recommended statistical approach as it specifies indicators as direct causes of a composite score without requiring internal consistency or covariance (39, 44, 56–58).

Finally, covariate-adjusted linear and logistic regressions were conducted to test whether the Index would serve as an independent predictor after controlling for known co-variates. To account for interaction effects, multiplicative interaction terms were added to the logistic regression models to test whether associations between AMHL and the outcome variables were moderated by key demographic variables (education, race, age, gender, insurance).

The highest score possible on the AMHL Index is 43 and the lowest is 0. Participants’ scores were calculated using the pre-established rubric (Supplementary Material 3). Scores were normally distributed as confirmed by a histogram and QQ plot. The sample mean was 30.3 (SD = 6.4) and scores ranged from 13 to 43.

As a general guideline, VIF values below 10 are acceptable and values below 3.3 are considered excellent (59, 60). All 30 Index indicators** had VIF values considerably below 10, with 27 items below 3.3 (Table 2). There are no serious multicollinearity issues and the conceptual framework effectively captured each variable and the overarching AMHL construct. All indicators were retained for subsequent analysis.

Using PLS-SEM, the model was further validated through an examination of strength and significance of path coefficients, and explained variance of outcome variables (R²). As shown in Table 3, results reflect the stability and relevance of indicator contributions. The structural model demonstrated the Index's ability to explain variance across all outcomes. While bootstrapping procedures were not feasible due to the binary character of the outcome variables, the theoretical consistency and stability of the model paths provide strong confidence in the external validity of the Index.

Logistic regression models were run for each outcome, controlling for known covariates. As shown in Table 4, the overall findings show that one's composite score on the Index significantly predicted positive preventive behaviors in six of the seven health-behavior models, as well as in the final health-outcome model. The results validate the importance of adjusting for SDH and confirm AMHL's measurable impact on health decision-making.

This study employed a non-probability sampling method, which may limit the generalizability of the findings and potentially introduce selection bias and affect the external validity of the results. However, the use of quota-based sampling helped minimize these limitations. Given that the sample was closely matched to national demographic data for age, gender, race, and education—and in light of minimal demographic deviations—we elected not to weight the data. This decision simplified interpretability and preserved analytical clarity, though it is acknowledged as a potential limitation.

For explanatory purposes, future studies should include additional health measures—such as sunscreen usage, reproductive health, diet and physical activity—as well as theoretically-guided mediation analyses. Given the dynamic nature of MHL across the lifespan, longitudinal validation would also illuminate causal relationships and fluctuation by age. In addition, adapting the Index for use in countries with universal health care or robust public media systems may reveal how AMHL manifests in less commercialized contexts and provide insights into the adaptability of the tool across diverse populations and public health systems. Research on pragmatic short- and long-term interventions is also urgently needed.