JMIR Aging is a peer-reviewed, open-access journal dedicated to advancing digital health, technology, and innovation for older adults [1]. It publishes research on eHealth interventions, gerontechnology, aging populations, and health services aimed at improving care and outcomes for aging societies [2]. Despite this clear mission, the underlying thematic structure of its publications remains underexplored, limiting the ability to systematically trace research foci, emerging trends, and evolving priorities [3]. Understanding the journal’s intellectual landscape requires an approach that integrates bibliometric theory, network clustering, and theme identification into a coherent framework.

In bibliometric research, thematic classification plays a central role in mapping intellectual structures and identifying research frontiers [4-6]. The theoretical foundation lies in co-word analysis, which assumes that the co-occurrence of keywords reflects conceptual proximity among research topics [4-9]. Tools such as Bibliometrix, VOSviewer, and CiteSpace operationalize this principle by constructing keyword networks and identifying clusters that represent thematic communities [7-9].

However, existing methods primarily operate at the cluster level, identifying thematic groupings among keywords rather than assigning a single dominant theme to each article [4-6,10,11]. This limits the interpretability of bibliometric maps and constrains microlevel analyses, such as tracing thematic shifts across journals, institutions, or regions.

Although advanced methods such as the following leader clustering algorithm (FLCA) [12] and latent Dirichlet allocation–based topic modeling can uncover latent thematic structures, a key gap remains: Cluster-level theme discovery does not automatically translate into clear, article-level theme representation—for example, mapping text-driven patterns (from item/factor structures) into labels that describe article or person attributes.

To date, few bibliometric models explicitly address this gap with a formal mechanism for assigning cluster-derived themes to individual publications. Without this linkage, thematic labeling becomes less consistent, and cross-sectional or longitudinal analyses are harder to interpret and compare [4,5].

To address this theoretical and methodological gap, an article-level theme assignment algorithm (Theme Assignment Algorithm in Articles [TAAA]) is required. TAAA formalizes the process of identifying each article’s dominant theme from multiple candidate clusters, enhancing both interpretability and reproducibility in bibliometric studies.

Although some prior studies have conceptually mentioned “theme-by-article” analyses [4,5], no standardized nor open-source R-based (or application) implementation currently exists [13]. The proposed framework, TAAA, operationalizes the statistical mode of an article’s keyword cluster assignments to determine its most representative theme (ie, from item/bifactor structures into labels that describe article or person attributes). This method enhances computational efficiency while maintaining theoretical consistency with co-word analysis principles.

Furthermore, the framework integrates visual analytic tools via an application (for the general public) and R-based scripts (for academic researchers), including Kano diagrams [14,15] and Sankey visualizations [11], to illustrate how the newly created theme variable propagates across articles, institutions, and nations. This integration advances existing bibliometric visualization paradigms by enabling article-level interpretability within cluster-based systems.

This study aimed to develop and demonstrate an application and R-based TAAA for assigning dominant themes to publications in JMIR Aging. The contribution of this work is 2-fold:

We hypothesized that TAAA can accurately capture each article’s primary thematic focus and improve interpretability in large-scale bibliometric analyses. We further expected the framework to be adaptable to other journals and research domains, as demonstrated through several practical examples.

We retrieved metadata for 434 articles published in JMIR Aging from 2020 to 2025 using the Web of Science Core Collection (WoSCC) database. The dataset includes bibliographic information such as author names, affiliations, titles, abstracts, and keywords. Study data are provided in Multimedia Appendix 1. Since all records in WoSCC are publicly available and contain no identifiable patient information, institutional review board approval was not required.

To ensure conceptual clarity, the following key terms are defined within the context of this study: keyword source, item testlet, person theme, mapping power (MP), and theme variable.

Keyword source refers to the origin of the keywords used for analysis. This study distinguishes between 3 types: (1) author-provided keywords: terms explicitly selected by the authors; (2) Keywords Plus: terms automatically generated by the WoSCC; and (3) abstract-derived keywords: terms extracted from the article abstracts using a custom R script.

Item testlet draws from educational assessment terminology, where a “testlet” represents a cluster of keywords grouped by a common thematic thread. In this study, testlets were generated using FLCA [12], where each cluster reflects a cohesive research topic within the broader keyword network.

Person theme is the primary topical label assigned to an individual article. Using TAAA [4,5], the “person theme” is determined by identifying the statistical mode (the most frequent keyword cluster) associated with that specific article (or person).

MP is a metric used to validate the consistency of thematic classification. As illustrated in Figure 1, MP is computed by mapping keyword clusters (item testlets) to article-level classifications (person themes) and comparing them against established true labels.

For the theme variable, cluster leaders were first generated separately for each keyword source and then merged by mapping item-level terms to person-level themes. This design enables rigorous cross-source comparison and validation of mapping performance against external ground-truth labels (TRUE_LABEL), such as gender or cancer versus noncancer status.

A clear and concise reporting format focusing on 10 key metadata elements, including top 10 themes, was used to improve readability and interpretation [11], offering an alternative to the extensive graphs and tables commonly found in traditional bibliometric analyses.

This research consists of 3 parts: (1) introducing TAAA, (2) validating TAAA, and (3) illustrating TAAA with and without MP.

A Venn diagram was used in R [21] to extract the overlaid themes from the 3 sources: Keywords Plus, author-defined keywords, and abstract. The screened themes denote the core elements in the studied journal.

We evaluated TAAA using 4 complementary validation components: interrater and intrarater validation, external validation and robustness across clustering methods, bootstrapping stability, and out-of-sample holdout robustness.

The application of TAAA to textual entities via the application was conducted with 434 articles published in JMIR Aging from 2020 to 2025. The application of TAAA to numerical entities via the application occurred via virtual data as a test on STEM and ARTS for men and women and as a questionnaire on 2 domains. In addition, the TAAA was applied to Gene Expression Omnibus (GEO) data via the application: A gene expression profile, GSE8401 [36], including 31 primary melanoma and 52 metastatic melanoma (MM) clinical samples, was downloaded from the GEO database [37]. The differentially expressed genes between MM and primary melanoma were assessed using MP.

All visualizations and data processing were conducted using the R platform  [16,17,20,23] or GAE [35]. Kano diagrams were generated in the application or R scripts, with all statistical analyses performed in R version 4.1.3  [13]. Visualizations and results were produced within the RStudio integrated development environment  [38]. The way to conduct this study using R scripts is described in Multimedia Appendix 2.

Figure 2 presents a comprehensive summary of 434 publications in JMIR Aging (h-index=28 [39]) analyzed across 10 bibliometric dimensions—country, institution, department, year, article type, author, research area, keyword frequency, and citation count.

The United States, Canada, and the United Kingdom led in publication output, while the University of Toronto (Canada) and Maastricht University (the Netherlands) were among the most productive institutions. Nursing and medical departments contributed prominently, with the majority of papers classified as original research articles.

Notably, Hannah Liane Christie (the Netherlands) was the most prolific author (6 articles [40-45]), whereas Audrey Lebrasseur et al (Canada) produced the most-cited study [46], receiving 252 citations for a rapid review on COVID-19’s impact on older adults. Core keywords such as HEALTH, PEOPLE, CARE, and TECHNOLOGY reflect the journal’s central focus on digital health and aging innovation; see the keyword and theme panels in Figure 2.

Accordingly, this performance overview provides a structural snapshot of JMIR Aging’s research landscape, enabling targeted insight into its geographic reach, institutional strengths, and topical emphasis.

Figure 3 illustrates how the TAAA identified dominant themes across 3 keyword sources: Keywords Plus (external indexing terms): 9 core themes (Figure 3A); author-supplied keywords: 7 core themes representing self-identified research emphases (Figure 3B); and abstract-derived keywords: 9 themes reflecting content-driven topic extraction (Figure 3C).

Across all 3 scenarios, the top themes were HEALTH (108/282, 39.4%), OLDER ADULTS (140/343, 43.1%), and DEMENTIA (38/158, 25.3%).

To illustrate theme overlap, Figure 4 (Venn diagram) shows that DEMENTIA was the only theme consistently identified in all 3 TAAA runs, underscoring its centrality in the journal’s research portfolio.

Hence, the convergence of DEMENTIA across independent keyword sources validates TAAA’s reliability in identifying dominant thematic foci and highlights the journal’s sustained emphasis on cognitive health in aging research.

Figure 5 demonstrates how assigned themes distribute across countries, highlighting geographic research orientations. The United States remained a research leader, followed by Canada and the United Kingdom.

DEMENTIA and CARE are closely associated with US authors, whereas ASSESSMENT is linked to Canadian research. SCALE and PEOPLE themes are common in UK-based publications. The Sankey diagram in Figure 5A visualizes these relationships—flow width represents the proportional connection between countries and their dominant themes.

In Figure 5B, a network chart of country-theme was uploaded by the input form of rectangular data via an application created by our research team [25,34].

Together, these analyses reveal cross-level variation in thematic prominence—DEMENTIA dominates within specific institutions but shares space with broader aging-related themes at the national level, suggesting specialized research clusters driving global impact.

The findings of this study confirmed that that the TAAA can reveal meaningful thematic alignments at the journal, country, and institute levels, supporting deeper insights into scholarly trends and research spots, as shown in Table 3.

Table 4 summarizes key conceptual and methodological differences between TAAA and traditional co-word or cluster-based approaches, underscoring TAAA’s capacity for article-level theme identification and reproducible bibliometric analysis.

The results indicate that the 3 analytical scenarios identified 9, 7, and 9 core themes from their respective keyword sources, with HEALTH (39.4%), OLDER ADULTS (43.1%), and DEMENTIA (25.3%) emerging as the most prevalent themes. Among these, only DEMENTIA was consistently highlighted across all analyses. The application of dual heatmaps demonstrated the algorithm’s capacity for comparative bibliometric mapping in JMIR Aging, with a mapping precision of 0.33 providing quantitative evidence of a 2-stage publication pattern.

These findings support the hypothesis that TAAA effectively assigns a dominant theme to each article using a mode-based approach that aligns with established thematic structures.

This study advances bibliometric methodology by introducing the TAAA [17], which systematically assigns dominant themes at the article level using mode-based clustering, improving on previous techniques  [4,5] with R-based implementation  [16,17]. Unlike traditional keyword co-occurrence analyses that remain at the cluster level (eg, a study  [10] exploring trends and hotspots in research on exercise and Alzheimer’s disease), TAAA enables article-level thematic assignment by adding a dedicated theme column to the dataset. This supports more granular analyses of journal focus, research trends, and associations between article entities.

The application and R-based TAAA framework  [17,32-34] enhances reproducibility and computational efficiency while enabling seamless integration with visualization tools such as Kano and Sankey diagrams  [11,14,15] for clearer thematic mapping with dual heatmaps. Applying TAAA [17,32-34] to JMIR Aging publications demonstrates its practical utility, consistently identifying core themes (eg, DEMENTIA) across multiple keyword sources. Notably, this approach can be applied to other journals to highlight their core themes for readers in future research.

TAAA thus offers a replicable, scalable method for bibliometric studies, supporting country-, institution-, and journal-level analyses that were previously limited by manual theme assignment.

In addition to analytical use, the extraction of keywords directly from article abstracts provides a potential editorial application. By comparing abstract-derived keywords with author-supplied keywords and Keywords Plus terms, editors can evaluate the thematic coherence of submitted or published manuscripts. Discrepancies between the automatically derived and author-assigned keywords may indicate whether an article aligns with the journal’s primary thematic domains or falls outside its intended scope. Such comparison offers a data-driven mechanism for theme validation, assisting editorial boards with maintaining topical consistency, refining journal aims, and identifying emerging research areas within the publication’s portfolio.

The greater prominence of DEMENTIA observed at the institutional level compared with the country level reflects the localized concentration of specialized research centers and collaborative networks focusing on dementia and cognitive aging. Institutions such as the University of Toronto and Harvard Medical School host long-standing dementia research programs that generate high keyword density and thematic coherence within institutional clusters.

At the national level, however, the thematic landscape becomes more diffuse as research activities span multiple subfields of aging and health technology. This pattern suggests that DEMENTIA functions as a core institutional specialization within certain centers of excellence rather than a uniform national research focus, highlighting TAAA’s capacity to reveal hierarchical differences in thematic intensity across analytical levels.

Although Hannah Liane Christie from the Netherlands published the most articles, with 6 publications [24-29] in JMIR Aging, the most highly cited article [46] was authored by Audrey Lebrasseur and colleagues from Canada, with 252 citations. This study with a rapid review synthesizes research on COVID-19’s impact on older adults, revealing increased psychological symptoms, ageism, and reduced social interactions. The findings highlight the need for tailored strategies to maintain social ties and address at-risk older adults’ specific needs.

The scalable framework of a reproducible application or R-based TAAA maps co-word clusters to one dominant theme per article using a deterministic mode rule.

Validated mapping via dual heatmaps and MP quantifies alignment with TRUE_LABEL, supporting a stage-like pattern while showing the boundaries are not strictly year-split.

Practical visuals integrating Kano, Sankey/SankeyMATIC, volcano, network, and receiver under the operating curve outputs provide interpretable, actionable bibliometric insights.

Beyond bibliometrics, TAAA can be adapted to clinical and other data-intensive settings where subject-level labels must be inferred from multi-item structures rather than directly observed. In clinical research, bifactor or multidomain measures are often reduced to composite scores and evaluated using logistic regression analysis to predict the patient status. TAAA offers a complementary structure-based alternative: It converts item-level patterns into a discrete subject style by assigning the dominant cluster (mode) across a subject’s activated features, yielding an interpretable 2-style classification suitable for downstream prediction and stratification.

For example, TAAA can be applied to medication records to derive a patient’s dominant treatment pattern (eg, drug class or regimen cluster), supporting analyses of treatment pathways and utilization without requiring explicit brand-company linkage. More broadly, the same framework can harmonize heterogeneous administrative datasets by inferring prevailing categories from fragmented records. These extensions can be implemented using the web-based application [32] provided in this study

First, the analysis relied primarily on Keywords Plus from the WoSCC, which may omit domain-specific concepts or vary in granularity compared with controlled vocabularies such as MeSH used in PubMed. Future studies could integrate multiple bibliographic databases and standardized ontologies to improve thematic coverage and semantic precision.

Second, the mode-based theme assignment in TAAA simplifies articles to a single dominant theme, which may underrepresent genuinely multitopic publications. Extensions incorporating weighted frequencies or probabilistic theme assignments could better capture thematic complexity and are available for further development, as outlined in our GitHub repository.

Third, although abstract- and title-based keyword extraction was automated using R [16] and the application [32], certain preprocessing steps (eg, keyword normalization and cluster interpretation) may still require expert oversight. Greater automation of preprocessing and quality control would further enhance scalability and reproducibility.

Fourth, interrater validation uses two study authors, acknowledges potential expectation bias, and describes blinding procedures more explicitly. Future studies could consider adding more experts to assign themes to articles in comparison withd TAAA.

Finally, validation was conducted using a single journal (JMIR Aging), limiting generalizability. Future work will apply TAAA across multiple journals and domains to more fully assess its robustness, scalability, and cross-disciplinary applicability.

This study introduces the application or R-based TAAA as a systematic approach for assigning dominant themes at the article level. Application to JMIR Aging publications demonstrates that TAAA enhances interpretability and reproducibility while enabling nuanced bibliometric analyses across journals, countries, and institutions. By bridging cluster-level co-word structures with article-level thematic representation, TAAA provides a transparent framework for large-scale thematic mapping. Future work may extend this approach by integrating semantic embeddings or applying the framework to multidisciplinary datasets.