The initial dimension differentiates between predictive and explanatory models. Predictive models seek to anticipate future conditions of a health system under diverse scenarios (39). These are frequently predictive and are frequently employed to forecast outcomes such as illness prevalence, hospital admissions, or resource requirements across various scenarios (40). For instance, during the COVID-19 pandemic, prediction models were essential in assessing the effects of measures such as lockdowns and vaccination tactics (41). These models often emphasize accuracy, generalizability, and practical applicability.

In contrast, explanatory models aim to comprehend the mechanisms and reasons behind specific results. These models are predominantly theory-driven and are employed to investigate the fundamental mechanics of complex systems, including health service utilization behaviors, interactions among system components, or the impacts of policy instruments (42). System dynamics models are frequently created to replicate feedback loops and delays in health systems, providing insights into systemic inefficiencies or unexpected effects of treatments (43).

This distinction elucidates whether a model is designed to facilitate quick decision-making through forecasts or to enhance comprehension of system behavior over time.

The second dimension emphasizes the modeling process. At one end are data-driven models that depend primarily on empirical data and statistical or machine learning approaches (44). These models are frequently employed for disease classification, risk prediction, and patient outcomes forecasting (45). The emergence of big data and the widespread use of electronic health records have elevated data-driven approaches, including deep learning, ensemble models, and clustering, particularly in precision medicine and public health surveillance (46).

On the other end are system-oriented models, including system dynamics, agent-based modeling (ABM), and discrete event simulation (DES) (47). These models typically rely on theoretical frameworks and aim to emulate the dynamic interactions among components within health systems. For instance, ABMs can replicate patient trajectories inside sophisticated healthcare systems, whereas DES can analyze patient movement within hospitals or clinics (48). These approaches are particularly practical for capturing nonlinearities, feedback loops, and temporal delays that define real-world health systems (47).

The scholarly literature on health systems modeling can be traced back to 1992, when it began with a modest corpus of merely two articles. However, recent trends indicate a remarkable surge in scholarly output, with 2021 witnessing the highest output on the subject matter, namely, 351 publications. The period under review (1992–2023) saw an annual growth rate of 7.53%, with a high average citation per document index of 22.82. This is a strong growth rate compared with the approximately 3% annual growth rate in the global scientific publication output (62). Substantial growth in health systems modeling research occurred between 2015 and 2022, with 2023 already having 19 publications. There has been a marked surge in the utilisation of health systems modeling in tackling chronic, tobacco-related, and cancer-related illnesses, as well as other health-related concerns. This trend has been attributed to the recognition that systems modeling can contribute to a more comprehensive understanding of the intricate and complex nature underpinning health and disease prevention (63).

The notable upsurge in publications on health systems modeling between 2020 and 2022 can be attributed to the onset of the COVID-19 pandemic. This global crisis inflicted significant harm on human lives and necessitated the development of precise prediction mechanisms to address its impact on public health (1). Many different co-authorship indices are also revealed. The average number of co-authors for each article is the basis for the co-authors per article index (64). In this study, the index coefficient is 9.35.

Table 2 depicts two documents published in 1992, when articles on health systems modeling were first cited. The first article to be published was by Schinnar, Rothbard & Hadley from the University of Pennsylvania titled “A Prospective Management Approach to the Delivery of Public Mental Health Services” in the journal Administration and Policy in Mental Health and Mental Health Services Research in March 1992. This foundational article contributed to the early conceptual framing of health systems modeling by introducing a system-oriented strategy encompassing forecasting service demands, optimizing resource allocation, and employing decision-support mechanisms within public mental health services. Subsequent literature has cited this work in support of broader system-level planning frameworks. For example (65), was built on (66), emphasizing performance evaluation to apply Data Envelopment Analysis (DEA) in funding allocation for behavioral health contractors. Similarly (67), referenced the study about assessing community mental health service needs and forecasting future demand for planning purposes.

The health systems modeling discipline encompasses diverse domains, comprising but not limited to surveillance mechanisms for identifying health threats, interventions aimed at controlling and mitigating the impact of such threats on public health, and predictive technologies designed to address various health-related challenges. The following criteria were used to articulate the most influential publications in health systems modeling: (i) the most cited articles, (ii) publications with the highest number of articles, (iii) publications with the highest number of citations, and (iv) publications with top indexing figures.

The countries of origin of corresponding authors in health systems modeling research are presented in Table 7 and Figure 2. A corresponding author is the primary contact with the journal's editor and represents all co-authors. Collaboration patterns are captured through two indices: Single-Country Publications (SCP), which reflect intra-country collaboration, and Multiple-Country Publications (MCP), which indicate inter-country collaboration. The MCP counts articles co-authored by individuals from institutions in different countries, while the SCP includes publications with co-authors from the same country.

Table 7 and Figure 2 highlight the top 10 contributing countries (out of 90) to health systems modeling research between 1992 and 2023, based on total citations and SCP/MCP outputs. The United States (733 corresponding authors), the United Kingdom (196), Australia (129), Canada (125), and China (80) lead in terms of combined intra- and inter-country collaboration. A clear correlation exists between high SCP and high MCP scores, suggesting that countries with strong domestic research capacity are well-positioned in international collaborations. Citation frequency follows a similar pattern, reinforcing the influence of these leading nations, with the USA consistently occupying the top slot. Notably, the UK displays a higher MCP than SCP, indicating a strong international orientation in its research collaborations.

The bibliometric analysis identifies the USA and UK as gatekeepers within the global health systems modeling network. The USA, with 562 SCP and 171 MCP publications, and the UK, with an MCP ratio of 0.592, are the most important connectors influencing research agendas, funding, and publication dynamics. Their central roles in the collaboration network emphasize their strategic positioning as global research hubs. Conversely, countries such as Iran (MCP ratio = 0.246), Spain (0.298), and Italy (0.341) are relatively under-connected in this network. Despite their contributions, these lower MCP ratios suggest limited international collaboration, which may stem from barriers such as restricted access to international funding, institutional limitations, language differences, or fewer established global partnerships.

On the other hand, countries like Switzerland (MCP ratio = 0.660), Brazil (0.344), and China (0.388) represent emerging research communities with growing international engagement. Switzerland, in particular, stands out for its high MCP ratio despite a modest publication count (53 articles), signaling its increasing integration and influence within global research networks. Brazil and China, with rising publication outputs and moderate international collaboration, also show strong potential to expand their global research presence further.

Table 8 lists the top 10 authors based on the h-index. Blakely T. and Hay S.I. are the top authors in the field of health systems modeling for the period 1992–2023. Blakey T. is in the top position because of a higher g-index and the number of publications. The author also started to publish in 2009, as opposed to Hay S.I., who was in the second position, which started earlier in 2006. Although Wilson N. has the second most publications (NP = 25), he is in the fourth position because of a lower h-index than Blakely (15), Hay (15), and Kvizhnadze (12). It has been suggested that the h-index predicts future scientific achievements more accurately than other research metrics (79).

A total of 4,069 institutions contributed to health systems modeling research over the 32 years. The 10 most productive institutions are presented in Table 9, ordered according to the number of publications by each institution. The findings are compared to the Academic Ranking of World Universities (ARWU) 2022 (80) and QS World University Rankings 2023 (81), which are internationally recognised rankings of universities in terms of reputation and excellence. Surprisingly, the Tehran University of Medical Sciences in Iran has the highest number of publications (132), although its ARWU ranking is between 401 and 500. The University of Washington in the USA is second with 128 articles, followed by the London School of Hygiene & Tropical Medicine in the UK with 123 articles. Except for one institution in Canada (ranked 9th with 91 articles), all the remaining seven were either in the USA or the UK.

All the institutions and research centres in Table 9 are ranked at least in one of the two world university rankings (ARWU or QS). While the Tehran University of Medical Sciences is perceived to be in the 401–500 position on ARWU 2022 rankings, it has published many articles on health systems modeling. The ARWU's number one institution (Harvard University), number 5 on the QS ranking, has fewer publications (110) in health systems modeling. As noted, the dominance of institutions from the USA, followed by those from the UK, is apparent as these two countries have eight out of the 10 institutions within the top 10 slots. The noticeable inclusion of Iran and Canada in health systems modeling is a positive development.

Table 10 reflects the countries that produced the top 10 highest number of citations within the period under review. In the corpus of selected articles, many authors cited scientific articles originating from the USA, which amassed 24,467 citations. The UK emerged as the second most frequently cited country with 5,139 citations. Australia also received a notable share of citations, with its scientific output accruing 3,361 citations and exhibiting an average article citation of 26.054. Remarkably, South Africa was featured among the top 10 most frequently cited countries, with 744 articles on the subject matter.

The Sankey Plot depicted in Figure 3 illustrates the relationship between the top countries (left field), top keywords (middle field), and top sources (right field). Based on a Sankey diagram, a three-field plot can be used to interpret the proportion of highly used keywords to top countries that use them and the top publishers within the subject. The flows and related quantities are illustrated in proportion to one another on the Sankey diagram. The magnitudes are represented by the width of the lines or arrows; the wider the line, the stronger the flow. The flow symbols can be merged or divided at every process phase along their pathways. Colour can be used to categorise the diagram or to indicate the change from one state of the process to another.

Specifically, the diagram visually depicts how each of the top ten countries connects to the primary keywords they have employed in their publications and the journals in which these articles appeared. Wider lines between countries, keywords, and sources indicate greater frequency or stronger associations. For instance, the USA and UK exhibit robust flows toward the keyword “COVID-19,” signifying extensive research output in this area, primarily disseminated through leading journals such as The Lancet and PLOS One. Conversely, narrower lines highlight more modest associations, such as those observed for countries like South Africa, indicating either lower volume or narrower topical focus in their contributions within these top journals.

A closer look at low- and middle-income countries (LMICs) in the plot, namely Brazil, Iran, India, South Africa, and China, reveals meaningful but less dominant contributions to health systems modeling research. These countries exhibit active engagement with key thematic areas such as “COVID-19,” “public health,” and “modeling,” suggesting that LMICs are contributing to globally relevant and urgent topics. For instance, Brazil and India show visible connections to “cost-effectiveness” and “public health,” which may reflect their ongoing efforts to optimize limited healthcare resources. Iran and South Africa are linked to narrower keyword ranges, while China shows more diverse connections to terms like “health systems” and “epidemiology.” Regarding publication venues, LMICs tend to publish in widely recognized outlets such as PLOS One, BMJ, although their link strength is lower than in high-income countries. These findings emphasize the importance of supporting broader inclusion of LMICs in international health systems modeling research, particularly given their unique perspectives and contextual challenges.

Except for South Africa, all 10 countries (the USA, Brazil, the UK, Canada, Iran, Australia, Spain, India, and China) made noticeable contributions related to COVID-19. The subject was also published in all the top journals in terms of citations identified in Table 5 (The Lancet, PLOS One, New England Journal of Medicine, Medical Care, Journal of the American Medical Association, Journal of Health Sciences, BMJ, PLOS Medicine, Nature, and Social Science & Medicine). The Lancet stands out as the most published journal, covering all the nine most used authors' keywords (DE) identified: COVID-19, cost-effectiveness, epidemiology, SARS-CoV-2 (coronavirus), health systems, public health, modeling, and health policy.

An important approach for determining trending areas and scholarly focus is frequently used keywords (54). The bibliometric analysis also included an analysis of the frequently used keywords. Using another bibliometric tool, VOSviewer, the keywords were ranked according to their total link strength (TLS) and the number of times they occurred. The TLS indicator reflects the strength of all collaborations between a specified researcher and other researchers (82). Table 11 reflects the most frequently used keywords in the research, ranked according to the TLS. As displayed in the table, the keyword “care” has the highest TLS (1,060) and the highest number of occurrences (213). This is followed by “impact” with a TLS of 789 and 161 occurrences. The keyword “health care” has the lowest TLS (474) but the third highest number of occurrences (144). All the keywords are relevant within the health research area. The occurrence of keywords such as “care,” “health,” and “health-care,” individually and in combination, is attributed to the inherent scope and nature of health systems modeling literature. Researchers frequently employ these terms both broadly and specifically; “health” often denotes overarching population outcomes and conditions, “care” emphasizes direct service delivery or interventions, while “health-care” commonly represents the healthcare sector holistically. This lexical variety reflects different research emphases and highlights the interdisciplinary and multifaceted nature of the health systems modeling field, aligning closely with the thematic breadth identified in the bibliometric analysis.

Word clouds can also be used to graphically illustrate frequently used keywords. A word cloud refers to a graphic that depicts word frequency as a quick method to determine the main idea of written content (14). The keywords that appear larger in the graphic represent the most frequently used keywords in the analysed text. The analysis of the content of the 2,023 documents produced 4,163 keywords, which encompassed, among others, terminologies within the healthcare industry, the countries involved, the various techniques used, the participants within the various articles, and the inputs and outputs from the articles. These are depicted in Figures 4, 5. The figures graphically depict the most frequently occurring keywords such as those appearing in Table 11 as well as others used in health systems modeling including “healthcare”, “disease”, “prevalence”, “model”, “transmission”, “interventions”, “health systems”, “population”, “transmission” and “quality”.

A thematic map is used to generate insight into the status quo of current research in a particular field and the prospects of sustaining future research (54). Thematic analysis is helpful in enlightening scholars and stakeholders about the possibility of creating novel thematic research topics within a particular discipline (52). A thematic map illustrates how the study themes have changed over time in bibliometric analysis. It is based on bibliometric data that details the frequency and co-occurrence of keywords, authors, journals, and other bibliographic variables in scientific publications (52). Each point on a thematic map represents a research topic, and the interval between the points indicates how related the topics are to one another (54). The similarities between topics increase as they become more contiguous. The size of the points reflects how frequently the topic appears in the literature. The main research themes in a discipline may be found, and their history can be understood by looking at a thematic map. Additionally, one can spot developing topics and follow their evolution over time (54). This data can be employed to direct research efforts, pinpoint research gaps, and drive policy choices. A thematic map with four quadrants is depicted in Figure 6.

Quadrant 1 (the upper right quadrant) represents the motor themes, which are highly developed and mature (driving themes) (55). Quadrant 1 reflects “cost-effectiveness,” “modeling,” “modeling,” “HIV,” and “tuberculosis.” Quadrant 2 (the bottom right quadrant) resembles the basic themes that are generally growing but have not reached maturity, as in Quadrant 1 (54). This quadrant contains themes with high density but low centrality. These themes might be growing, but still require more investment to mature. The themes “economics” and “models” are growing themes that still demand considerable investment to mature. They focus on economic modeling techniques within health systems. In Quadrant 3 (the top left quadrant), the niche themes are more developed, although they are not yet well connected to other themes within the field (64). They have low centrality and density, suggesting they are either at an early exploration stage or losing momentum. “Primary care,” “simulation,” “machine learning,” “electronic health records,” and “mental health” are depicted as the developed themes that were yet to be well-correlated to other themes within the field of systems health modeling. Emerging trends like machine learning and electronic health records show the potential to grow as they address modern healthcare needs. Quadrant 4 (the bottom left quadrant) represents themes that are either emerging or declining (disappearing) (61). Basic themes exhibit high centrality but low density, making them foundational yet not deeply explored. They underpin the field and show promise for future research. “COVID-19,” “health systems,” “epidemiology,” “public health,” and “SARS-CoV-2” have either declined in terms of scientific research or scholars are now gaining interest in them.

Table 12 enumerates the terms for each cluster according to their frequency. The table used author keywords with the Leiden clustering algorithm to identify research clusters. The analysis used a minimum cluster frequency of 5 per thousand documents, with a complete counting method and a minimum of 5 labels per cluster. This approach revealed six distinct clusters comprising 63 items, interconnected through 403 links with a cumulative total link strength of 727. This comprehensive clustering approach emphasises the complex relationships and thematic congruence within health systems modeling.

Table 12 reveals the following clusters and overlapping clusters:

Cluster 1: Focuses on public health challenges and their solutions, including breast cancer, tuberculosis, malaria, and HIV. It also emphasises cost-effectiveness analysis, screening, and improving quality of life through prevention, treatment, and healthcare policy, particularly in regions like South Africa.

To further identify and elucidate the intellectual structures, this section discusses the co-authorship, co-citation, and bibliographic coupling networks conducted using VOSviewer.

Figure 7 visualization reveals the international collaboration landscape in health systems modeling. Unsurprisingly, the United States emerges as a dominant node with extensive global linkages, followed by the United Kingdom and Australia. The thickness of connecting lines indicates strong bilateral partnerships, while the presence of clusters reveals regionally cohesive research communities. Equally, to depict the interconnectedness among institutions involved in health systems modeling, Figure 8 showcases the institutional alliances. Key academic and research hubs such as the University of Washington and Tehran University of Medical Sciences are prominently positioned, reflecting prolific output and collaboration. While institutional partnerships within countries are robust, the figure highlights significant cross-border affiliations. Importantly, this co-authorship network accentuates how institutional alliances fuel interdisciplinary research, with the structure indicating both central leadership and peripheral but growing contributors.

The study also mapped the intellectual influence of journals based on how frequently they are cited together, as shown in Figure 9. Leading journals like The Lancet and PLOS One occupy central positions, indicating foundational status. The dense saturation observed suggests high cohesion around certain core outlets. These visualizations highlight epistemic dependencies and shared knowledge bases in health systems research.

Country-level co-citation patterns are effectively highlighted in Figure 10, revealing how national research outputs are interlinked through shared citations. The USA, UK, and Australia form the triadic core, indicating their significant academic influence. The map also reveals emerging scholarly alignment among countries such as Iran and China, reinforcing global diversification. While centrality demonstrates intellectual leadership, peripheral countries, though less connected, show promising citation visibility.

The bibliographic coupling analysis in Figure 11 identifies journals that cite similar sources, implying thematic or methodological alignment. Notable journals like BMJ, PLOS One, and BMC Health Services Research are tightly linked, demonstrating a shared intellectual foundation. This coupling provides insight into the publication ecosystem and can guide authors seeking suitable publication venues. Additionally, the clustering of sources reveals disciplinary convergence zones and may indicate emerging niches within health systems modeling literature.

Figure 12 maps institutions that exhibit similar citation behaviors, signaling aligned research interests. Institutions like Harvard, Oxford, and Tehran University of Medical Sciences form tight bibliographic clusters, reflecting thematic or methodological synergy. The network shows which institutions are thematically aligned and where collaborative potential may exist. Thus, this figure reflects research alignment and is a strategic tool for identifying potential institutional partnerships for future collaborative studies.

Figure 13 reveals how countries converge around similar bibliographic sources, suggesting thematic affinity and shared intellectual priorities.

Figure 13 reveals how countries converge around similar bibliographic sources, suggesting thematic affinity and shared intellectual priorities. As expected, the United States, the United Kingdom, and Australia dominate. However, within tightly coupled nodes, countries like Iran and Brazil reflect their growing alignment with global research themes. Importantly, this analysis complements co-authorship and citation maps by showing how national research agendas are intellectually positioned, offering insight into global knowledge integration.

The dense nodes in Figure 14 represent authors such as Wang, Eckelman, White, and Walker, indicating their strong influence through frequent co-citation. This reflects their foundational roles in advancing interdisciplinary research and methodological innovations in health systems modeling and has made significant contributions to health systems modeling.

Drawing insights from the themes that emerged from this bibliometric analysis, we propose three priority areas for future research. These priorities are designed to leverage emerging trends and contribute to developing more efficient and effective health systems.

From a practical point of view, this study offers valuable insights for researchers, policymakers, and practitioners in the health sector. In outlining the most prominent scholars, journals, and institutions in health systems modeling, it guides individuals seeking collaboration, publication venues, and authoritative sources. The study's focus on the application of modeling in addressing various health issues, including pandemic response and chronic disease management, underlines the practical utility of modeling techniques in real-world health decision-making. Furthermore, identifying emerging themes and the thematic map of the field can inform funding agencies and research institutions about potential areas for investment and development, thus driving innovation in healthcare solutions and policies.

The study relied on data from the WoS database only, which, although extensive, may not capture all relevant publications in the field of health systems modeling. There is an opportunity to expand on the findings from this bibliometric analysis to include other databases to capture a broader spectrum of research outputs in health systems modeling. Another limitation of this study is the absence of consultation with a medical librarian or information retrieval expert to optimize the search strategy. Consequently, there is a possibility of missing relevant literature, especially from non-indexed sources or databases not included in our search. Future studies should aim to employ a more exhaustive search strategy, integrating multiple databases and the expertise of information retrieval specialists to ensure a more comprehensive inclusion of literature.