Empirically Tested Health Literacy Frameworks

Background: Health literacy is a significant determinant of health behaviors, but the pathways through which health literacy influences health behaviors are not completely clear nor consistent. The purpose of this systematic review is to critically appraise studies that have empirically tested the potential pathways linking health literacy to health behavior. Methods: We performed searches of the electronic databases PubMed, Embase, and CINAHL to identify studies that proposed a conceptual framework and empirically tested the proposed mechanism through which health literacy influences certain health behaviors. Twenty eligible studies were included for analysis. Key Results: The 20 studies addressed various health behaviors: chronic disease self-management (n = 8), medication adherence (n = 2), overall health status (n = 4), oral care (n = 1), cancer screening (n = 1), shared decision-making (n = 1), health information sharing (n = 1), physical activity and eating behaviors (n = 1), and emergency department visits (n = 1). Most studies were conducted in the United States (n = 13) and used a cross-sectional design (n = 15). The Short Test of Functional Health Literacy in Adults was commonly used to assess health literacy levels. Selection of variables and their operationalization were informed by a theoretical model in 12 studies. Age, gender, race/ethnicity, and insurance status were reported antecedents to health literacy. The most commonly tested mediators were self-efficacy (n = 8) and disease knowledge (n = 4). Fit indices reported in the studies ranged from acceptable to excellent. Discussion: Current evidence supports self-efficacy as a mediator between health literacy and health behavior. Further research is needed to identify how health literacy interplays with known psychosocial factors to inform people's use of preventive care services. Future studies should include more disadvantaged populations such as immigrants with high disease burden and those with low health literacy. Theory-based, empirically tested health literacy models can serve as the conceptual basis for developing effective health interventions to improve health behaviors and ultimately decrease the burden of disease in such vulnerable populations. [HLRP: Health Literacy Research and Practice. 2020;4(1):e21–e44.] Plain Language Summary: This review systemically compiles, and critically appraises 20 existing studies that test conceptual frameworks that propose potential pathways through which health literacy affects health behaviors. The findings from this review can help inform the development of health literacy-focused interventions to improve the health behaviors of populations with disease burdens.


Study Selection and Data Extraction
Covidence, an Internet-based software platform that streamlines the production of systematic reviews, was used in the study selection and data extraction process. Our initial database search yielded a total of 900 studies, of which 169 duplicates were removed. To enhance the rigor of the systematic review process, two authors (J.C. and S.D.) independently screened all abstracts and titles for relevance to empirical testing of HL models and frameworks. All conflicts and discrepancies were discussed and resolved through face-to-face group discussions. A total of 676 articles were excluded for nonrelevance to our study's purpose. The full texts of 55 relevant abstracts were then reviewed independently by the study authors (J.C., S.D., M.C., and H.H.) using the study's inclusion and exclusion criteria. We excluded 39 studies for the following reasons: (a) studies did not include or propose an HL framework (n = 27); (b) no empirical data were presented (n = 6); (c) studies did not address the impact of HL on health behavior (n = 3); (d) studies do not include HL as a study variable (n = 1), (e) no full text was available (n = 1); and (f) it was a podium presentation (n = 1). Using the same search terms ( Table A), an additional database search was conducted in March 2019 for studies published since November 2018. After removing duplicates, 90 titles with abstracts were reviewed for relevance. Two study authors (J.C. and S.D.) independently reviewed 17 full texts using the study's inclusion and exclusion criteria. A total of 13 articles were excluded for the following reasons: (a) studies did not propose a HL framework (n = 9); (b); studies did not address the impact of HL on health behavior (n = 2); (c) studies were not written in English (n = 1); and (d) no empirical data were presented (n = 1). Figure 1 provides a detailed description of the selection process. Two study authors (J.C. and S.D.) extracted data from a total of 20 studies for this systematic review. To enhance interrater reliability and the accuracy of information presented, the authors compared key findings and other relevant data, and discrepancies were resolved.

Quality Assessment
The Joanna Briggs Checklist was the appraisal tool used in the quality assessment of all studies included in this review (Joanna Briggs Institute, 2018). The checklist is a series of questions that authors of observational studies are expected to answer to enhance a study's methodological rigor. Specifically, each study's quality was assessed using seven items addressing selection bias, measurement bias, confounding variables, and appropriate use of statistical analyses (Joanna Briggs Institute, 2018). Studies were assigned a score of 1 for items that were adequately described, and a score of 0 for items that were not addressed by the authors. Total scores for each study ranged from 0 to 7, with a higher total score attributed to higher quality rating. Studies with a total score less than 3 were rated as low quality, studies with total scores ranging from 3 to 4 were rated as medium quality, and studies with total scores of 5 or higher were rated as high quality. Findings from the quality assessments were used to critique the overall methodological strengths and weaknesses of the studies Results of the quality assessment process are shown in Table 1. All of the studies adequately described inclusion criteria and the characteristics of study participants. There was adequate discussion of items addressing selection bias in most studies included in the review: description of inclusion criteria (n = 19), and description of study characteristics (n = 15). Most studies included in the review inadequately addressed measurement bias: identification of confounders (n = 8), use of valid and reliable measurement of outcome (n = 6), and strategy addressing confounders (n = 8). The measurement of outcomes in more than 75% (n = 15) of studies was based on self-reports. Overall, most studies had high (n = 10) to medium (n = 6) quality ratings. Only four studies received a low-quality rating.

Pathways Linking HL and Health Behaviors/Outcomes
All but three studies assessed a number of variables as possible mediators between HL and health behaviors/outcomes (Hou et al., 2018;Intarakamhang & Intarakamhang, 2017;Schillinger et al., 2006). Eight studies examined the mediating effect of self-efficacy on the relationship between HL and diabetes management, heart failure management, and general self-care (Como, 2018;Chen, 2014       Four studies that examined how HL is related to health behavior through disease knowledge found the following: only one study showed a statistically significant mediating effect of knowledge in the context of diabetes management (Brega et al., 2012), and three studies found a direct association between HL and knowledge (Chen, 2015;Cho et al., 2008;Osborn, Paasche-Orlow et al., 2011). All four studies that examined the mediating effect of disease knowledge did not describe how knowledge instruments were scored, however. In addition, all four studies had a large proportion (65%-70%) of study participants with a high school education or less (Chen, 2015;Cho et al., 2008;Osborn, Paasche-Orlow et al., 2011;Zou et al., 2017).
Of the eight studies that examined self-care activities (medication adherence, physical activity, self-monitoring of blood glucose, foot care, healthy diet) as factors linking the pathway between HL and health outcomes (glycemic control, emergency department visits, blood pressure control, and physical and mental health status) (Brega et al., 2012;Cho et al., 2008;Como, 2018;Hickman et al., 2016;E. H. Lee et al., 2016;Y. J. Lee et al., 2016;Osborn, Paasche-Orlow, et al., 2011;Sun et al., 2013), two reported a significant, mediating effect (Brega et al., 2012;E. H. Lee et al., 2016). Both studies controlled for known demographic covariates such as age, gender, education, marital status, treatment regimen (insulin or oral hypoglycemic use), hemoglobin A1c level, as well as duration of disease in the mediation analysis (Brega et al., 2012;E. H. Lee et al., 2016).

DISCUSSION
To our knowledge, this is the first systematic review to critically appraise studies that have empirically tested the potential pathways linking HL to health behaviors and health outcomes. We found evidence to support that theoretically selected mediators (i.e., self-efficacy, disease knowledge, selfcare activities, and patient-provider communication) mediate the identified relationship between HL and chronic disease management, with self-efficacy as the commonly tested mediator (E. H. Y. J. Lee et al., 2016). Our findings show that unless people possess adequate HL, they may perceive low confidence in their abilities to manage their chronic diseases. In addition, improving people's HL is an essential first step to increasing their knowledge about their disease, improving their ability to adequately perform selfcare activities, and effectively communicate and collaborate with health care providers in their chronic disease management (Charlot et al., 2017;Chisholm-Burns, Spivey, & Pickett, 2018). We also found evidence to support that intervention outcomes (glycemic control, medication adherence) differ by the HL levels of study participants, suggesting HL as a moderator (Schillinger et al., 2006;Soones et al., 2017). This finding highlights an important implication for future research, particularly in relation to intervention research as it relates to the role of HL beyond mediation.
We identified several factors that may have contributed to the mixed findings we reported: study design, selection bias, small sample sizes, measurement errors, and non-theoryguided operationalization of study variables. Although all studies in this review aimed to examine the pathways linking HL to health behaviors and outcomes, these studies exclusively used cross-sectional and a mixed-methods designs, which preclude causality and temporality. Secondly, only 7 of 20 studies conducted sample size calculations and power analyses a priori (Chen, 2015;Como, 2018;Hou et al., 2018;Intarakamhang & Intarakamhang, 2017;E. H. Lee et al., 2016;Y. J. Lee et al., 2016;Photharos et al., 2018). The lack of statistical power in most of the studies could account for the mixed findings reported. Thirdly, although all U.S.-based studies used well-validated HL measures, the remaining studies either lacked psychometric testing results or had only been tested in a single population; therefore, the validity and reliability of those measures could not be established (Intarakamhang & Intarakamhang, 2017;E. H. Lee et al., 2016;Y. J. Lee et al., 2016;Sun et al., 2013;Zou et al., 2017). Also important is that the studies were predominantly across a convenience sample of female, urban-dwelling adults with less than a high school education who were recruited from health care facilities. Therefore, findings cannot be generalized to other populations that do not use the health care system due to language barriers or a lack of health insurance. Finally, theory provides a systematic foundation and a logical pathway for illustrating the relationship among various study concepts and variables. However, only a limited number of studies (n = 12) included in the review explained how theory informed the selection and operationalization of study variables, delimiting the generalizability of findings.
Findings from this review call for the need to use theoretically grounded, methodologically rigorous research with statistically powered sample sizes to adequately examine the interplay between HL and health behaviors or outcomes in diverse study populations. For example, the studies included in this review exclusively used a cross-sectional design to test the indirect pathways linking HL to health behaviors. Hence, there is still a need for establishing temporality and causality using more rigorous study designs such as longitudinal cohort design. Several studies have used longitudinal data to examine the role of HL on health behaviors and outcomes; however, they did not meet the inclusion criteria for this review because the authors did not specify a HL conceptual framework to be tested (Kobayashi, Wardle, & Wagner, 2015;Washington, Curtis, Waite, Wolf, & Paasche-Orlow, 2018). In addition, although a recent systematic review showed that HL has gained importance on the European health agenda, none of the studies identified from our extensive search of various database were conducted in Europe (Sørensen et al., 2015). Further, among U.S.-based studies, all were conducted on female, English-speaking adults (Brega et al., 2012;Chen, 2014;Cho et al., 2008;Como, 2018;Crook et al., 2016;Guo et al., 2014;Hickman et al., 2016;Jin et al., 2019;Osborn, Cavanaugh, et al., 2011;Osborn et al., 2010;Osborn, Paasche-Orlow, et al., 2011;Schillinger et al., 2006;Soones et al., 2017). Although people who belong to ethnic/racial minority groups and those with low English proficiency, particularly immigrants, are known to be disproportionately burdened by low HL, they were excluded from the U.S.-based studies (Alper, 2018;Wang et al., 2013). In particular, African immigrants, an exponentially increasing immigrant group in the U.S. with worse health outcomes in comparison to other immigrant groups, were excluded in all the U.S.-based studies (Anderson, 2015). Although there is a possibility that African immigrants were categorized as Black Americans in some of these studies, it has been established that people of African descent (Black, African immigrant, and Afro-Caribbean) in the U.S. have different cultural and linguistic characteristics that affect their health outcomes differently. Therefore, there is a need to disaggregate these subgroups in health research (Commodore-Mensah et al., 2017;Forney-Gorman & Kozhimannil, 2016).

STUDY STREGNTHS
The Cochrane Collaboration and the U.S. Institute of Medicine have endorsed that review teams must have content and methodological expertise (Bigendako & Syriani, 2018;Gøtzsche & Ioannidis, 2012;Institute of Medicine, 2011). A major strength of this study is that our contributors have undergone training in systematic review methodology and have published prior reviews (Cajita, Cajita, & Han, 2016;Han, Floyd, et al., 2018;. Additionally, most of the authors are clinicians with expertise in health promotion among populations with poor health literacy. These skillsets helped us capture a heterogeneity of opinions and allowed for high interrater reliability when reviewing articles for inclusion in the review. These strengths add to the degree of confidence when reporting our study findings, which also speaks to the thoroughness of this systematic review.

STUDY LIMITATIONS
This systematic review is limited in that despite our extensive database searches, there may be other relevant and unpublished studies that may not have been identified. Therefore, the theories we identified as guiding the development of HL conceptual frameworks may not be exhaustive. The majority of studies included in this review assessed HL using REALM and TOFHLA, which assess reading ability and comprehension, respectively, but do not comprehensively address the multidimensionality of HL (i.e., ability to understand written text, speak and listen effectively, and use quantitative data to make appropriate health decisions) (Sørensen et al., 2012). Most studies used a cross-sectional design that precludes causality and temporality. In addition, we only included studies published in English. This may have also resulted in the small number of studies included in this review as well as the number of studies that included non-English-speaking populations.

CONCLUSION
Our review adds to the existing body of knowledge on the impact of HL on health behavior by providing a comprehensive understanding of how theory informs the development of HL conceptual frameworks, and the systematic selection and evaluation of variables that inform HL-focused studies. We found evidence to support that HL is related to health behaviors, particularly chronic disease management, through mediators such as self-efficacy and disease knowledge.