Health literacy and the health belief model in type 2 diabetes mellitus

Abstract

Background:

Health literacy (HL) and health beliefs are pivotal but historically separate determinants of self-management in type 2 diabetes mellitus (T2DM), and their combined research landscape has not been mapped. We aimed to characterize the intellectual structure, collaborative networks, and thematic evolution of research on HL and health beliefs in T2DM using a comprehensive bibliometric approach.

Methods:

Bibliographic records were retrieved on February 24, 2026 from the Web of Science core collection (science citation index-expanded, social sciences citation index, and emerging sources citation index) with no year restriction. A 3-concept Topic Search combining HL, health belief (including health belief model [HBM] constructs), and T2DM term sets returned 1750 records. Three sequential exclusion filters were then applied: document type (290), English-only language (23), and Citation Topics Meso domain (592), yielding a final corpus of 845 original articles (1982–2026). Performance indicators were computed in Biblioshiny/bibliometrix (open-source; M. Aria & C. Cuccurullo, University of Naples Federico II) and science mapping in VOSviewer 1.6.20. Analyses were descriptive and structural, with no inferential hypothesis testing.

Results:

The 845 articles were published in 347 journals by 3929 authors from at least 45 countries, yielding 20,426 citations (mean 24.17 per article). Annual output grew from 7 articles in 2005 to 84 in 2025. The United States contributed 39.6% of publications, and international co-authorship reached 17.99%. The HBM was explicitly named in only 7.9% of the corpus. Abstract-text mapping identified 4 thematic clusters: HL measurement and glycemic outcomes; self-management and behavioral constructs including the HBM; qualitative inquiry into patient access and experience; and randomized controlled trial design.

Conclusions:

HL and HBM research streams appear to remain largely separate within the T2DM literature. Integrated theoretical frameworks, culturally diverse samples, and interventions that address both literacy deficits and negative health beliefs are urgently needed. The findings support prioritizing studies that jointly model HL, perceived barriers, and self-efficacy against glycemic endpoints; validating these frameworks in the low- and middle-income regions that carry the greatest diabetes burden; and embedding both constructs within diabetes self-management education and support. Patient-education and counseling journals emerge as the natural venues for this integrated agenda.

1. Introduction

Type 2 diabetes mellitus (T2DM) represents one of the most formidable chronic health challenges of the 21st century, with a global prevalence currently estimated at 589 million adults and projected to escalate to 853 million by 2050.^[1] Accounting for over 90% of all diabetes diagnoses, T2DM is a paradigm of a behaviorally mediated chronic condition: achieving optimal clinical outcomes depends less on pharmacological advances than on the patient’s sustained adherence to multifaceted daily self-management regimens, including nutritional regulation, physical activity, glycemic monitoring, pharmacological compliance, and the strategic use of healthcare services.^[2]

Two psychosocial constructs have emerged as pivotal determinants of self-care engagement: health literacy (HL) and health beliefs. As operationalized by the World Health Organization, HL comprises the cognitive and social competencies that shape an individual’s motivation and capacity to access, process, and apply health-related information.^[3] This construct is further stratified into functional, communicative, and critical levels, each requiring tailored educational and clinical interventions.^[4] Within T2DM populations, HL has been associated with disease-specific knowledge, self-management efficacy, glycemic regulation, and the longitudinal risk of hospitalization.^[5,6] Complementing this, the health belief model (HBM) (which integrates perceived susceptibility, severity, benefits, and barriers, alongside self-efficacy and cues to action) remains a preeminent theoretical framework for explaining the drivers of adherence in T2DM.^[7] Empirical data consistently identify self-efficacy and perceived barriers as the most potent HBM predictors of behavioral compliance across diverse domains, including pharmacological adherence, podiatric care, and glycemic monitoring.^[8]

The relevance of these 2 constructs is not confined to high-income settings. Cross-sectional evidence from Iran links higher HL (particularly its decision-making dimension) to health-promoting behaviors in the general adult population,^[9] while studies of patients with T2DM report that perceived barriers and self-efficacy are among the strongest predictors of self-care behavior and glycemic control, frequently outweighing sociodemographic factors.^[10] These findings, drawn from low- and middle-income contexts that carry a disproportionate share of the global diabetes burden, indicate that literacy and belief determinants are both salient and actionable across heterogeneous health systems. Yet the 2 literatures have developed largely in parallel, and no study has systematically mapped how (or whether) they converge within the T2DM evidence base.

Despite the substantial growth of the primary literature in both domains, the intersection of HL and health beliefs in T2DM remains considerably under-synthesized as an integrated field of inquiry. Bibliometric analysis (a rigorous quantitative methodology that applies statistical and mapping techniques to large scholarly corpora) offers a systematic framework for delineating field structures, identifying critical knowledge gaps, and tracing intellectual trajectories that conventional narrative reviews frequently obscure.^[11–13] Although recent bibliometric investigations have addressed adjacent areas, including digital HL^[14] and broader applications of the HBM,^[15] a structural mapping of their convergence in T2DM has not yet been undertaken. This gap has practical consequences: in the absence of an integrated evidence map, interventions and educational programs tend to address literacy or beliefs in isolation; researchers cannot readily locate where the 2 literatures connect or diverge; and educators, program designers, and funders lack a basis for prioritizing the integration of these determinants within diabetes self-management education and support (DSMES).

The novelty and necessity of the present study therefore lie in being the first comprehensive bibliometric mapping of HL and health belief research specifically at their intersection in T2DM, analyzing 845 Web of Science (WoS)-indexed publications from 1982 to 2026. We address 4 research questions: How have the volume, citation impact, and growth trajectory of the field evolved over this period?; What is its geographic and collaborative structure, and how concentrated is scientific production?;To what extent, and in what manner, is the HBM integrated into HL research in T2DM?; What are the dominant thematic clusters, and where do the principal gaps for theoretical and experimental integration lie? By answering these questions, the analysis delivers an evidence-based research agenda (identifying underserved regions, the limited penetration of explicit behavioral theory, and the scarcity of experimental designs) of direct utility to researchers, intervention designers, educators, and funders seeking to strengthen DSMES.

2. Materials and methods

2.1. Study design

This bibliometric study was conducted and reported in accordance with the BIBLIO guideline for reporting bibliometric reviews of the biomedical literature.^[16] Because the analysis was confined to publicly available, non-identifiable bibliographic metadata and did not involve human participants, human specimens, or animals, institutional review board approval was not required. All data underlying the findings (the bibliographic dataset, analysis code, and network map files) are provided as Files S1, S2, S3 and S4, Supplemental Digital Content 1.

2.2. Data source and search strategy

Literature was retrieved from the WoS core collection on February 24, 2026 with no year restriction. WoS was selected as the single information source because its core collection provides curated, citation-linked indexing across the 3 databases most relevant to this field (the science citation index-expanded, the social sciences citation index, and the emerging sources citation index) and supports the document type, language, and Citation Topics classifications used for corpus refinement; no other databases, trial registers, gray-literature sources, or author contact were used. Three concept groups were combined with the Boolean AND operator: HL terms, health belief terms (including HBM constructs), and T2DM synonyms. The Topic Search (TS=) operator covered titles, abstracts, author keywords, and KeyWords Plus. The complete Boolean search strings for all 3-concept groups, including synonyms, truncation, and field tags, are provided as File S4, Supplemental Digital Content 4.

At the query stage, the WoS filters restricted results to the Article document type and excluded publisher-invited and early-access review articles, as well as hot-paper and highly-cited-paper designations; no date restriction was imposed. This yielded 1750 records. Three exclusion filters were then applied sequentially: removal of letters, book chapters, editorials, correction notices, conference papers, and meeting abstracts reduced the set to 1460 records (290); exclusion of non-English records (Spanish, Portuguese, German, French, Polish) reduced the set to 1437 records (23); and Citation Topics Meso refinement (restricting results to the diabetes, HL, telemedicine, and nursing citation domains while excluding the psychiatry, psychology, religion, and healthcare-policy clusters, in order to concentrate the corpus on the clinical and health-services literature most directly concerned with diabetes self-management) yielded the final corpus of 845 original English-language articles (592). The WoS export contained no duplicate records. The complete study selection process is summarized in Figure 1.

Figure 1.

Study selection flow diagram. Records were identified in the WoS core collection (SCIE, SSCI, ESCI) using a 3-concept Boolean Topic Search and screened sequentially by document subtype (290), language (23), and Citation Topics Meso domain (592), yielding a final corpus of 845 original English-language articles (1982–2026). ESCI = emerging sources citation index, n = number of records, SCIE = science citation index-expanded, SSCI = social sciences citation index, WoS = Web of Science.

2.3. Analytical tools and definitions

Performance indicators (publication trends, citation metrics, journals, and author productivity) were computed using Biblioshiny (bibliometrix R-package 4.x),^[17] with raw data exported to Microsoft Excel (Microsoft Corporation) for verification. Science mapping (co-authorship networks, abstract-text co-occurrence networks, and thematic cluster analysis) was performed using VOSviewer 1.6.20 (developed by the Centre for Science and Technology Studies [CWTS], Leiden University).^[12] Keyword co-occurrence required a minimum of 5 occurrences; abstract-text mapping required a minimum of 10 occurrences per term; and co-authorship networks were restricted to authors with a minimum of 3 documents. These minimum-occurrence thresholds follow standard VOSviewer practice, suppressing infrequently occurring terms to yield interpretable maps, and were specified a priori rather than tuned to the results. All analyses were descriptive and structural in nature comprising: frequency counts, citation metrics (total citations [TC], mean citations per document, and within-dataset h- and g-indices), annual growth rates, and network co-occurrence maps; consistent with the bibliometric design, no inferential statistical hypothesis testing was performed.

Single-country publication and multiple-country publication rates were used to quantify international collaboration. Country of origin was assigned from the corresponding author’s affiliation, as implemented in bibliometrix (the AU_CO field); because this convention attributes each article to a single country, it may underrepresent coauthor countries in multi-country collaborations. An article was classified as applying the HBM when the model was explicitly named (“health belief model” or “HBM”) in its title, abstract, or author keywords. This operational definition is deliberately specific but not sensitive: studies that apply HBM constructs (perceived susceptibility, severity, benefits, barriers, or self-efficacy) or cite foundational HBM references without explicitly naming the model are not captured, so the resulting frequency represents a conservative lower bound on HBM engagement.

3. Results

3.1. Overview of the dataset

The final corpus comprised 845 articles, and its principal bibliometric indicators are summarized in Table 1. Two features stand out. First, the field is strongly collaborative: only 15 articles (1.8%) were single-authored, and documents averaged 5.76 coauthors. Second, citation impact is substantial for a specialized intersection, with a total of 20,426 citations and a mean of 24.17 per document. The 845 articles were published in 347 source journals by 3929 authors from at least 45 countries and cited 23,062 references; the annual growth rate was 1.59%, and the mean document age was 7.47 years.

Table 1.

Main bibliometric indicators for the HL and health beliefs in T2DM literature (n = 845; WoS core collection, 1982–2026).

Indicator	Value	Notes
Total articles	845	WoS core collection, 1982–2026
Source journals	347	All indexed types (SCIE, SSCI, ESCI)
Annual growth rate	1.59%	Linear slope 1982–2025
Mean document age	7.47 years	Calculated from publication year
Average citations per document	24.17	Total TC/ n articles
Total references cited	23,062	Sum of all in-text references
Author keywords (DE)	1566	Unique author-supplied keywords
Keywords Plus (ID)	1057	WoS auto-generated keywords
Total authors	3929	Unique author records
Single-authored documents	15	1.8% of corpus
Coauthors per document	5.76	Mean across all articles
International co-authorship rate	17.99%	Articles with authors from ≥ 2 countries

Data extracted from the Biblioshiny report.

“DE” is the Web of Science / bibliometrix field tag for author-supplied keywords; “ID” is the field tag for Keywords Plus.

ESCI = emerging sources citation index, HL = health literacy, SCIE = science citation index-expanded, SSCI = social sciences citation index, T2DM = type 2 diabetes mellitus, TC = total citations, WoS = Web of Science.

3.2. Annual publication and citation trends

Substantive scholarly output began in 2005 (n = 7) and grew steadily, accelerating after 2016 and peaking in 2025 (n = 84); annual counts for 2014 to 2025 are reported in Table 2. The apparent dip in 2023 (n = 50) most likely reflects incomplete WoS indexing at the time of data extraction rather than a genuine decline, and the 2025 figure is partial (up to February 24, 2026). Citation impact was concentrated in early work: the highest mean citations per document for any single year occurred in 2002 (TC/n = 359.25), driven almost entirely by Schillinger et al.^[5]

Table 2.

Annual publication output, 2014 to 2025.

Year	n
2014	35
2015	39
2016	45
2017	43
2018	52
2019	65
2020	76
2021	65
2022	80
2023	50
2024	79
2025	84

n represents the number of original articles indexed in the WoS core collection per year. 2025 data are partial (up to February 24, 2026).

n = number of articles, WoS = Web of Science.

3.3. Journal analysis

The 845 articles appeared in 347 journals, but productivity was concentrated in patient-education outlets (Table 3). Diabetes Educ and Patient Educ Couns ranked first and second by output and, together with JAMA and Diabetes Care, formed the most consistently co-cited sources across the corpus. J Gen Intern Med carried disproportionate citation weight relative to its output (h-index = 11 from 16 articles), reflecting the clinical importance of the work it published.

Table 3.

Top 15 most productive journals and source impact indicators.

Rank	Journal	n	%	h-index	TC	WoS
1	Diabetes Educ	31	3.67	20	1550	SSCI
2	Patient Educ Couns	27	3.20	15	824	SSCI/SCIE
3	BMC Public Health	23	2.72	10	426	SCIE
4	Patient Prefer Adherence	19	2.25	9	327	SCIE
5	BMJ Open	17	2.01	7	212	SCIE
6	J Gen Intern Med	16	1.89	11	1031	SSCI/SCIE
7	PLoS ONE	16	1.89	8	315	SCIE
8	Front Public Health	14	1.66	-	-	ESCI
9	JMIR Res Protoc	13	1.54	-	-	ESCI
10	Prim Care Diabetes	13	1.54	-	-	SCIE
11	Int J Environ Res Public Health	12	1.42	8	272	SCIE
12	BMC Health Serv Res	11	1.30	-	-	SCIE
13	Diabet Med	11	1.30	7	231	SCIE
14	J Health Commun	11	1.30	10	580	SSCI
15	J Diabetes	10	1.18	-	-	SCIE

h = h-index within the present dataset. TC = total citations within the dataset. (-) = not among the top 10 in the source impact analysis. Journal names shown as National Library of Medicine journal-title abbreviations.

ESCI = emerging sources citation index, n = number of articles, SCIE = science citation index-expanded, SSCI = social sciences citation index, TC = total citations, WoS = Web of Science.

3.4. Author analysis and co-authorship networks

The 10 most productive and impactful authors are listed in Table 4. Muscat, D.M. (University of Sydney) led in document count, whereas Schillinger, D. (University of California, San Francisco) accumulated the highest within-dataset citation total and h-index; Rothman (Vanderbilt) and Egede (Medical University of South Carolina) showed high citation impact relative to their output. Notably, 4 of the top 10 authors (Muscat, Ayre, McCaffery, and Nutbeam) are affiliated with the University of Sydney, indicating a coordinated Australian research program in HL measurement and shared decision-making.

Table 4.

Top 10 most productive authors and within-dataset bibliometric impact indicators.

Rank	Author	Docs	h-index	g-index	TC	Affiliation
1	Muscat, D.M.	28	11	21	469	University of Sydney, Australia
2	Schillinger, D.	19	12	19	2439	University of California, San Francisco, USA
3	Wolf, M.S.	16	10	16	452	Northwestern University, USA
4	Ayre, J.	13	7	11	123	University of Sydney, Australia
5	McCaffery, K.J.	13	9	13	321	University of Sydney, Australia
6	Nutbeam, D.	13	8	13	314	University of Sydney, Australia
7	Egede, L.E.	12	11	12	694	Medical Univ. South Carolina, USA
8	Rothman, R.L.	12	10	12	1135	Vanderbilt University, USA
9	Wallston, K.A.	8	8	8	691	Vanderbilt University, USA
10	Shiyanbola, O.O.	10	8	10	338	Univ. of Wisconsin-Madison, USA

h = h-index, g = g-index, TC = total citations within the 845-article dataset. These are not global career indices.

TC = total citations, USA = United States of America.

Figure 2 shows the author co-authorship network generated by VOSviewer. The network comprised 8 distinct collaboration clusters with limited inter-cluster connectivity. The largest cluster (red; upper right) groups the Sydney team (Muscat, Ayre, McCaffery, Cvejic, Nutbeam), reflecting a tightly integrated Australian HL-measurement program. A second major cluster (yellow; lower center) centers on Schillinger and University of California, San Francisco collaborators (Sarkar, Karter, Grant, Herman). A third cluster (dark blue; upper left) links Rothman, Wallston, and Cavanaugh within the Vanderbilt clinical-outcomes network. Cross-cluster links were sparse, indicating that the field’s major research groups operated largely within self-contained networks.

Figure 2.

Author co-authorship network (VOSviewer 1.6.20; minimum threshold: 3 documents per author; n = 845 articles). Node size is proportional to document count; color denotes cluster membership. n = number of articles.

3.5. Most-cited articles

The 14 most-cited articles are listed in Table 5.^[5,18–30] The foundational study by Schillinger et al^[5] (JAMA, 2002) established the link between functional HL and glycemic control and remains, by a wide margin, the most-cited work in the field. The second-ranked study, Choi and DiNitto^[18] (2013), marked the emergence of eHealth literacy as a distinct subfield. Six of the 14 appeared in major clinical journals (JAMA, Diabetes Care, Ann Intern Med) between 2001 and 2013, whereas the most recent entry, Walker et al^[29] (2020), signals a digital turn toward eHealth literacy in underserved populations.

Table 5.

Top 14 most-cited articles in the corpus.

Rank	Reference	Journal (Year)	TC	Title (abbreviated)
1	Schillinger et al[5]	JAMA (2002)	1287	Association of health literacy with diabetes outcomes
2	Choi & DiNitto[18]	J Med Internet Res (2013)	594	Digital divide among low-income homebound older adults
3	Sarkar et al[19]	Diabetes Care (2006)	404	Self-efficacy and diabetes self-management across race/ethnicity
4	Ishikawa et al[20]	Diabetes Care (2008)	379	Functional, communicative, and critical health literacy in T2DM
5	Cavanaugh et al[21]	Ann Intern Med (2008)	274	Association of numeracy and diabetes control
6	Rothman et al[22]	JAMA (2004)	272	Influence of patient literacy on a T2DM disease management program
7	Sarkar et al[23]	J Gen Intern Med (2011)	232	Validation of self-reported health literacy across diverse populations
8	Rubin[24]	Am J Med (2005)	209	Adherence to pharmacologic therapy in T2DM patients
9	Bains & Egede[25]	Diabetes Technol Ther (2011)	205	HL, diabetes knowledge, self-care behaviors, and glycemic control
10	Karter et al[26]	Diabetes Care (2010)	198	Barriers to insulin initiation: the TRANSLATE study
11	Heisler et al[27]	Diabetes Care (2005)	196	Relationships among HL, diabetes knowledge, and self-care behaviors
12	Aljasem et al[28]	Diabetes Educ (2001)	193	Barriers and necessary conditions for self-care behaviors in T2DM
13	Walker et al[29]	Telemed eHealth (2020)	175	eHealth literacy in underserved populations
14	Lee et al[30]	Patient Educ Couns (2016)	159	Pathways of empowerment, HL, self-efficacy, and glycemic control

TC = total citations in the WoS Core Collection as of 24 February 24, 2026. Superscript numbers correspond to the reference list.

T2DM = type 2 diabetes mellitus, TC = total citations.

3.6. Keyword and abstract-text analysis

The 20 most frequent abstract terms are reported in Table 6. “Health literacy” dominated the vocabulary, followed by clinical and self-management terms such as “care,” “type 2 diabetes,” “glycemic control,” and “self-management.” Terms associated with the HBM (“health belief,” “health belief model,” and “efficacy”) occurred far less frequently (52–63 occurrences), indicating a moderate but clearly subordinate presence in the abstract vocabulary.

Table 6.

Top 20 most frequent words in article abstracts (minimum frequency: 10).

Rank	Word	Freq	Rank	Word	Freq
1	health literacy	413	11	management	112
2	care	187	12	association	110
3	type 2 diabetes	163	13	mellitus	108
4	glycemic control	144	14	prevalence	93
5	adults	141	15	diabetes mellitus	88
6	outcomes	139	16	type 2 diabetes mellitus	87
7	self-management	136	17	adherence	74
8	diabetes	123	18	impact	64
9	knowledge	120	19	risk	63
10	education	119	20	medication adherence	62

Frequencies based on abstract-text analysis in Biblioshiny/bibliometrix. Common stop-words were removed prior to analysis.

Freq = Frequency.

Figure 3 shows the abstract-text co-occurrence network generated by VOSviewer, in which 4 thematic clusters were visible. The large red cluster (left and center) groups measurement and clinical-association terms: “scale,” “literacy level,” “cross-sectional study,” “hba1c,” “self care,” “efficacy,” “health belief,” and “health belief model.” The yellow cluster (upper center) captures HL-measurement methodology: “instrument,” “item,” “reliability,” “subscale,” “cronbach,” “content validity,” and “ehealth literacy scale.” The green cluster (right) groups qualitative-inquiry terms: “interview,” “theme,” “focus group,” “access,” “challenge,” “provider,” and “health information.” The blue cluster (lower center) encompasses intervention and trial terms: “baseline,” “control group,” “effectiveness,” “trial,” “randomized trial,” and “blood pressure.” The co-location of “health belief model” within the red measurement cluster, alongside “efficacy” and “hba1c,” visually confirms the HBM’s role as a behavioral determinant studied primarily in association with glycemic and adherence endpoints rather than in dedicated theoretical investigations.

Figure 3.

Abstract-text co-occurrence network (VOSviewer 1.6.20; minimum: 10 occurrences per word; n = 845 articles). Node size is proportional to occurrence frequency; color denotes thematic cluster. Red = HL measurement and HBM constructs; yellow = psychometric validation; green = qualitative/patient experience; blue = trial/intervention design. HBM = health belief model, HL = health literacy.

4. Discussion

4.1. Field growth and citation dynamics

This bibliometric study documents the progressive institutionalization of HL and health beliefs as central constructs in T2DM research, with annual output growing roughly 12-fold from 2005 to 2025. The 1.59% annual growth rate, while modest compared with rapidly expanding fields, reflects a field that has passed its initial growth phase and is consolidating toward methodological diversification and applied intervention design. The concentration of high citation counts in early foundational studies is consistent with the citation lifecycle of clinical epidemiology: foundational observational and measurement studies have accumulated citations for decades, whereas recent intervention studies are still accruing them.^[11] The high mean citation rate of 2002 publications (TC/n = 359.25) is almost entirely attributable to Schillinger et al,^[5] which continues to function as the field’s primary reference anchor.

4.2. Diabetes educator and patient education and counseling as field anchors

The ranking of Diabetes Educ as the most productive journal and Patient Educ Couns as second, together accounting for 6.9% of the entire corpus, confirms that patient-education-focused journals constitute the primary intellectual hubs for this intersection of HL, patient communication, and T2DM self-management. The co-citation analysis further positioned both journals alongside JAMA and Diabetes Care as the 4 most consistently co-cited sources. This concentration is noteworthy given the breadth of the 347-journal field and underscores the central role of diabetes-education and patient-communication outlets in advancing this research domain.

4.3. Geographic concentration and the collaboration deficit

The United States (US) contributed 39.6% of publications, a share that is disproportionately high relative to the global T2DM burden, which falls predominantly in South Asia, Sub-Saharan Africa, and Latin America.^[1] Although the international co-authorship rate of 17.99% is higher than that reported in some bibliometric studies of chronic-disease management, it still indicates that more than 80% of papers are domestically coauthored. The co-authorship network (Fig. 2) reveals that the 2 dominant collaborative hubs (the US clinical-outcomes network [Vanderbilt–University of California, San Francisco–Northwestern] and the Australian HL-measurement programme centered on the University of Sydney) operate largely in isolation from each other and from research groups in the regions that bear the greatest diabetes burden. This matters because the determinants of self-care are not uniform across settings. Evidence from Iran, for example, shows that HL (especially its decision-making dimension) predicts health-promoting behavior in the general adult population,^[9] and that perceived barriers and self-efficacy are dominant predictors of self-care behavior and glycemic control among patients with T2DM.^[10] The near-absence of such settings from the collaborative core suggests that instruments and belief-based counseling approaches validated in high-income, high-literacy populations may not transfer directly to the contexts where they are most needed, and that current levels of international collaboration remain insufficient to close this gap.

4.4. The HL–HBM theoretical fragmentation

The central finding of this study is the persistent theoretical fragmentation between HL and health belief constructs in the T2DM literature. The HBM appears as a dominant abstract term in only 7.9% of the 845 articles and is co-located in the abstract network (Fig. 3) within the same cluster as “efficacy” and glycemic-outcome measures, suggesting that when the HBM is applied, it functions primarily as a predictor of adherence outcomes rather than as an organizing framework integrated with HL constructs. Systematic reviews confirm this asymmetry: while the association between HL and diabetes knowledge is well established, the evidence on the effect of HL on self-care behaviors is inconsistent and measurement-dependent,^[31] whereas HBM-based studies consistently identify self-efficacy and perceived barriers as the dominant determinants of self-care.^[8,32]

Primary studies that model literacy and belief constructs jointly, rather than in isolation, reinforce this interpretation. In a cross-sectional study of Iranian patients with T2DM, Babazadeh et al found that self-efficacy and perceived barriers (both belief constructs) were stronger predictors of self-care behavior and glycated hemoglobin (HbA1c) than sociodemographic factors, illustrating that literacy alone is insufficient to explain glycemic outcomes.^[10] The same research group showed, in the context of cervical-cancer screening, that HL and cognitive-belief constructs (perceived barriers and self-efficacy) act together as determinants of preventive behavior, indicating that the value of integrated literacy–belief models extends beyond diabetes.^[33] At the interventional level, a randomized controlled trial based on the Health Action Process Approach improved diet and medication adherence and reduced HbA1c in patients with T2DM, demonstrating that belief- and self-regulatory constructs are both modifiable and clinically consequential when explicitly targeted.^[34] Taken together, these studies support the central inference of the present map: the field’s separation of literacy and belief research is a missed opportunity, because the 2 sets of determinants interact to shape self-management. The theoretical implication is important, patients with adequate HL may still fail to adopt self-care behaviors if they perceive high barriers or low self-efficacy,^[6,35] whereas patients with strong self-efficacy may self-manage successfully despite HL limitations through practical compensatory strategies.^[3] These interaction effects have rarely been formally modeled within a single study, representing a tractable and high-yield research priority for the diabetes-education and counseling community.

4.5. Thematic evolution and emerging directions

The 4-cluster abstract-text network (Fig. 3) reveals not only the current thematic structure but also the field’s trajectory. The qualitative cluster (green; interview, theme, focus group, access, culture) has grown substantially since 2015, reflecting increasing recognition that HL limitations and health beliefs are embedded in social and cultural systems that are not reducible to individual cognitive deficits, a pattern also emphasized by scoping-review evidence on HL and self-care among immigrant populations with T2DM.^[36] The intervention–randomized controlled trial cluster (blue; trial, effectiveness, baseline, control group) indicates a maturing field beginning to shift from descriptive associations toward experimental tests of HL-sensitive interventions, consistent with systematic evidence that tailored, community-delivered HL interventions improve glycemic outcomes and diabetes knowledge.^[6] The prominence of “ehealth literacy scale” in the measurement cluster (yellow) points to digital HL as a rapidly growing sub-domain, consistent with the appearance of Walker et al^[29] in the most-cited list. Future research is likely to focus increasingly on integrated digital-HL models, culturally adapted HBM–HL frameworks for low- and middle-income settings, and patient-communication interventions that simultaneously address literacy and belief barriers.

4.6. Implications for diabetes research and clinical care

For the diabetes-research community, the most actionable implication of this analysis is the need to move beyond the parallel investigation of HL and health beliefs toward integrated models that link both constructs to glycemic and broader metabolic outcomes. Because the HBM appeared in only 7.9% of the corpus and co-occurred with efficacy and HbA1c terms rather than with HL measurement, the field has effectively treated beliefs as a downstream predictor rather than as a mechanism that interacts with literacy. Studies that jointly model HL, perceived barriers, and self-efficacy, and test their combined association with HbA1c, treatment adherence, and metabolic risk factors, would address a clear and tractable gap.

The geographic concentration of the evidence base has direct consequences for clinical diabetes care. With 39.6% of the output from the US and the dominant collaboration hubs located in high-income, high-literacy settings, the external validity of HL- and HBM-informed approaches in the regions carrying the greatest T2DM burden (South Asia, Sub-Saharan Africa, and Latin America) remains largely untested. Validation of HL instruments and belief-based counseling approaches in these populations should be a precondition for incorporating them into international diabetes-management guidelines.

The thematic structure also signals where experimental work is required. The predominance of measurement and cross-sectional studies, and the comparatively small intervention–trial cluster, indicate that the field is rich in description but limited in causal evidence. Pragmatic, culturally adapted trials that simultaneously target literacy deficits and negative health beliefs, and that report metabolic endpoints, would convert the descriptive consensus mapped here into actionable clinical evidence. Finally, the strong qualitative and patient-experience signals in the network underscore that literacy and beliefs are embedded in social and cultural contexts rather than reducible to individual cognitive deficits; for clinicians, this supports complementing routine literacy screening with structured assessment of perceived barriers and self-efficacy, so that educational efforts are matched to the beliefs that ultimately govern self-management behavior.

4.7. Strengths

This study has several strengths. To our knowledge, it is the first bibliometric synthesis to map HL and HBM research specifically at their intersection within the T2DM literature, rather than treating either construct in isolation. The analysis draws on the full indexed record of the WoS Core Collection (1982–2026) without date restriction, capturing the field from its earliest publications. Reporting followed the BIBLIO guideline for bibliometric reviews, and the search strategy, dataset, analysis code, and network files are provided as Supplementary Files to allow full reproduction. Two complementary tools were used (Biblioshiny/bibliometrix for performance indicators and VOSviewer for science mapping) with cross-verification in Microsoft Excel, adding analytical robustness. Finally, the operational definitions (for example, the explicit-naming criterion for HBM classification) were specified a priori and their consequences quantified, so that the reported figures can be interpreted transparently as conservative bounds.

4.8. Limitations

This study had several limitations. First, data were retrieved exclusively from the WoS Core Collection; Scopus, PubMed, and EMBASE were not included, potentially omitting relevant literature indexed elsewhere. Second, the English-only language restriction may underrepresent non-Anglophone research output. Third, the restriction to the “Article” document type excludes systematic reviews and meta-analyses, which are major carriers of cumulative evidence in this field. Fourth, citation counts represent a single snapshot (February 24, 2026) and will continue to evolve, particularly for articles published after 2022. Fifth, bibliometric mapping characterizes the structure of scientific communication but does not evaluate the methodological quality or clinical validity of individual studies. Sixth, the Citation Topics Meso refinement excluded the psychiatry, psychology, religion, and healthcare-policy clusters. Because health belief research is also published in these domains, this filter, together with the requirement that the HBM be explicitly named, likely undercounts HBM-informed studies, so the observed 7.9% should be read as a conservative lower bound on HBM engagement, and the inference of theoretical separation interpreted accordingly. Seventh, country of origin was assigned from the corresponding author’s affiliation (the bibliometrix AU_CO field); in multi-country collaborations this may over-attribute output to high-income countries, so geographic figures, including the 39.6% US share, warrant cautious interpretation. Eighth, the minimum-occurrence thresholds used for the co-occurrence and co-authorship networks, although conventional, were not subjected to sensitivity analysis, and alternative thresholds could alter the granularity of the resulting structures. Ninth, a small number of foundational references cluster around a single, highly-cited HL research network; although central to the field, future syntheses should remain attentive to representing a broad range of research groups. Accordingly, the findings require interpretation alongside systematic reviews of the primary evidence.

5. Conclusions

This bibliometric analysis provides the first comprehensive map of the HL and health belief research landscape in T2DM, covering 845 articles from 347 journals across more than 4 decades. The field has achieved quantitative maturity, but 3 structural deficiencies constrain its translational impact: geographic concentration in high-income Anglophone settings, persistent theoretical fragmentation between HL and HBM research streams, and an underdeveloped experimental evidence base. Patient-education-focused journals occupy central positions as both productive and highly-cited venues and are uniquely placed to advance integrated HL–HBM intervention frameworks.

Building on the strengths of this analysis (its comprehensive coverage, reproducible methods, and explicit, quantified definitions) future studies should pursue 4 priorities. First, they should develop and test theoretically unified models that link HL levels to HBM constructs, particularly perceived barriers and self-efficacy, and estimate their combined association with HbA1c and metabolic outcomes. Second, they should validate HL instruments and belief-based counseling approaches in culturally diverse and resource-limited settings, especially the South Asian, Sub-Saharan African, and Latin American regions that carry the greatest diabetes burden. Third, they should expand pragmatic, culturally adapted randomized trials that simultaneously target literacy deficits and negative health beliefs while reporting metabolic endpoints. Fourth, they should strengthen cross-national and cross-group collaboration to broaden the geographic and institutional base of the field. Pursuing this agenda (ideally through the patient-education and counseling journals that anchor the field) would substantially strengthen the evidence base for DSMES programs worldwide.

Acknowledgments

The author thanks the academic library and information services that provided access to the Web of Science Core Collection. No source of external funding supported this work, and the author declares no relationships or activities that could be perceived as a conflict of interest. No individuals other than the sole author are named in this section.

Author contributions

Conceptualization: Hatice Ağrali.

Data curation: Hatice Ağrali.

Formal analysis: Hatice Ağrali.

Investigation: Hatice Ağrali.

Methodology: Hatice Ağrali.

Project administration: Hatice Ağrali.

Resources: Hatice Ağrali.

Software: Hatice Ağrali.

Supervision: Hatice Ağrali.

Validation: Hatice Ağrali.

Visualization: Hatice Ağrali.

Writing – original draft: Hatice Ağrali.

Writing – review & editing: Hatice Ağrali.