The evolution of SSRI research: trajectories of knowledge domains across four decades

Abstract

Background

Since their introduction in 1982, Selective Serotonin Reuptake Inhibitors (SSRIs) have fundamentally transformed psychiatric care, becoming the first-line pharmacological treatment for depression and expanding to treat anxiety disorders, obsessive-compulsive disorder, and other conditions. Over four decades, SSRI research has generated an expansive, multidisciplinary scholarly landscape spanning neuropharmacology, psychiatry, somatic medicine, and environmental sciences. While specialisation creates deep expertise, it also risks intellectual isolation, underscoring a growing need for tools that enhance broad cross-domain awareness and collaboration, which has historically been a driver of scientific innovation.

Methods

We mapped the field’s intellectual structure by analysing 38,961 SSRI-related publications (1982–March 2025) using a hybrid network analysis approach. This method combined citation data with semantic similarity to capture research connections. We applied community detection algorithms to identify 99 distinct research clusters, main path analysis to trace the primary intellectual trajectories of the field, and modularity measures to evaluate the network’s structural organisation.

Results

Publication volume grew linearly over the four decades. Our analysis showed that research became continuously more diversified over time, indicated by rising network modularity and the growing number of clusters required to account for annual publications. The main path trajectory shifted from early pharmacokinetic and comparative efficacy studies toward topics of neuroplasticity and prenatal exposure research. Contemporary research has forked into two prominent branches: exploring fluvoxamine’s therapeutic potential for COVID-19 and environmental pharmacology examining SSRI ecotoxicological effects in aquatic ecosystems. The interactive visualisation further enables the identification of pivotal publications that bridge distinct research domains and initiate new lines of inquiry.

Conclusions

This comprehensive mapping, accompanied by an interactive visualisation, allows researchers and clinicians to trace the co-evolution of knowledge domains. It provides a broad historical perspective to contextualise individual contributions. By using visual modalities, this work expands how diverse audiences can engage with and interpret research, potentially fostering more intuitive and inclusive forms of scientific synthesis.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-026-07867-8.

Introduction

Selective Serotonin Reuptake Inhibitors (SSRIs) represent a landmark class of psychopharmaceuticals that have fundamentally transformed psychiatric treatment since their introduction. Beginning with zimelidine in 1982 (subsequently withdrawn due to rare adverse events), followed by fluoxetine in 1986, SSRIs quickly established themselves as first-line pharmacological treatment for depression, shifting prescribing patterns away from tricyclic antidepressants and monoamine oxidase inhibitors [1–3]. The clinical adoption of SSRIs has been accompanied by a dramatic increase in prescription rates that reflect changing prescribing practices as well as expanding therapeutic applications. Regarding the former, research suggests that since the late 1990s, more individuals are diagnosed with a depressive disorder, a greater proportion of those are treated pharmacologically, and for substantially extended durations [4–7]. This trend has culminated in an almost 50% increase in SSRI prescriptions between 2011 and 2021 across OECD member states [8]. As a secondary factor, the expansion of therapeutic applications contributed to this trend. Over time, SSRIs have received regulatory approval for use in obsessive-compulsive, post-traumatic stress, eating, and anxiety disorders [3, 9]. The rise in SSRI use for anxiety disorders reflects two related developments: more patients are being diagnosed with an anxiety disorder rather than depression, and pharmacological treatment has shifted from benzodiazepines to SSRIs as first-line therapy [10–13]. Furthermore, off-label applications have been documented for conditions including body dysmorphic disorder, fibromyalgia, premature ejaculation, and menopausal vasomotor symptoms [9].

In addition to the expanded applications of SSRIs in psychiatric and somatic diseases, studies over the past four decades have given rise to a remarkably diverse research landscape. This includes investigations into the neurochemical mechanisms underlying mood regulation, preclinical studies using different mouse models of depression to assess effectiveness, and large-scale pharmacovigilance efforts addressing a wide array of safety concerns, including suicidal ideation, withdrawal syndromes, and teratogenic risks [14–17]. Beyond human health, examples of ongoing inquiry range from veterinary medicine and environmental consequences of SSRI use to philosophical reflections on the self [18–20].

While this diversification has created considerable expertise within specialised domains, it simultaneously presents challenges for researchers seeking to maintain awareness of developments beyond their immediate area of focus [21–23]. This increased specialisation thus potentially reinforces intellectual isolation and can pose challenges to the identification of meaningful connections between different subject matters. Yet, such transitioning of disciplinary boundaries has historically been a driver of scientific innovation [23, 24].

Today, the accumulation of scholarly metadata has created new opportunities for exploring the structure and evolution of scientific disciplines following two major advancements. First, large-scale corpora of scientific publications and their metadata are now accessible from various sources, enabling connections between publications, authors, and institutions to create comprehensive research networks [25, 26]. Second, advances in computing power and natural language processing now make it possible to analyse entire areas of research using a wide variety of data types such as texts, citations, author collaborations or geographical locations. These developments enable researchers to trace the trajectories of disciplines across decades, identify emerging trends, and determine central research questions that define academic discourse, which was not feasible in the past [24]. Thus, rather than focusing on narrow research questions to systematically analyse the literature, as in traditional systematic reviews and meta-analyses, these tools enable a comprehensive mapping of complex research landscapes and offer a panoramic vision of a field that reveals patterns and underlying structures of a knowledge domain.

The primary objective of this work is to map the scientific literature on SSRIs, tracing the knowledge evolution of this entire pharmaceutical class over more than four decades since its first marketing authorisation in 1982. Through network analyses and an interactive visualisation, we aim to reveal not only which research domains were dominant during specific periods but also how dominance shifted, and new domains co-evolved in response to new clinical insights and societal demands. Our goal was to present the breadth and depth of the field in an engaging and accessible way.

Methods

Data collection

We selected the Scopus database due to its extensive multidisciplinary scope and comprehensive coverage, offering greater breadth and completeness compared to alternative databases [26, 27]. We queried the Scopus Search API and extracted all articles related to SSRIs published between January 1982, marking the approval year of the first SSRI, and March 2025. We included all English-language journal articles whose titles or abstracts contained either ‘SSRI’ (abbreviated or spelled out) or the name of any individual SSRI (see eTable 1 in the supplementary materials for the exact API retrieval parameters). We subsequently removed all publications without abstract and used the Abstract Retrieval API to extract all references from the remaining journal articles.

Network construction

We built two complementary networks from the SSRI publication dataset. The first network is a directed citation network that links publications through their Electronic Identifiers (EIDs). The second network is a hybrid undirected network that combines citation relationships with semantic similarity to provide a more comprehensive view of research connections.

To create the hybrid network, we first generated text embeddings from article titles and abstracts using the SPECTER21 model. [28] These embeddings were used to identify k-nearest neighbours for each publication based on cosine similarity. We executed the algorithm with different parameter values (k = 5, 10, 15, and 20) to determine the optimal number of nearest neighbours (see Community Detection below). The citations as well as the cosine similarities of the k nearest neighbours were then integrated into a hybrid undirected network following the approach of Boyack and Klavans [29], who demonstrated that combining direct citations with textual information yields the most accurate clustering of scientific articles.

In the hybrid network, two publications could be connected in two ways: either through direct citation relationships (bidirectional) or semantic similarity (amongst each other’s k most similar articles). We calculated edge weights using the formula: wᵢⱼ = α × cᵢⱼ + (1-α) × sᵢⱼ, where cᵢⱼ represents citation relationships (1 if a citation exists, 0 otherwise) and sᵢⱼ represents semantic similarity (cosine similarity for k-nearest neighbours, 0 otherwise). We explored two weighting schemes by setting α to 0.3 and 0.5. Prior bibliometric studies have demonstrated that hybrid networks combining citation and textual information produce more accurate clustering than citation-only approaches, with textual similarity being a particularly important contributor to clustering quality [29, 30]. The α = 0.5 configuration provides equal weighting between citation and semantic relationships, representing a balanced approach. The α = 0.3 configuration prioritises semantic similarity to assess whether emphasizing textual information would yield more coherent research clusters. This approach creates three distinct connection scenarios: publications that are both cited and semantically similar receive the strongest weights, publications connected only through semantic similarity have weights determined by (1-α) times their cosine similarity, while publications connected solely through citations have weights equal to α.

Community detection and data analysis

We applied the Leiden algorithm using the Constant Potts Model (CPM) as the quality function using the hybrid network to identify thematically coherent research communities within the SSRI literature [31, 32]. In addition to varying the number of nearest neighbours and edge weighting scheme as discussed above, we also adjusted the Leiden algorithm’s resolution parameter (ranging from 0.000001 to 0.005) [30] to identify the most optimal clustering configuration. We subsequently qualitatively evaluated the solutions to best balance cluster granularity with meaningful topic differentiation.

Our qualitative assessment focused on the algorithm’s ability to distinguish between conceptually related but distinct research areas that should remain separate clusters. Specifically, we evaluated whether the clustering result could distinguish between conceptually related but distinct research areas: (1) research on sexual dysfunction versus premature ejaculation, (2) anxiety versus panic disorder, and (3) depression in diabetic patients versus weight management in diabetic patients treated with SSRIs. These pairs were selected pragmatically during parameter optimisation as test cases for determining appropriate cluster granularity. We selected the least granular parameter combination that successfully maintained these topical distinctions. Based on this evaluation, we identified optimal parameters of α = 0.3 (prioritizing semantic similarity over citations), k = 10 nearest neighbours, and a resolution parameter of 0.002. The algorithm was run for 400 iterations, as additional iterations yielded no further improvements in the CPM quality function value (see eFigure 1 in the supplementary materials) [31, 33]. This configuration produced 144 initial clusters from the complete dataset. To ensure robust clusters, we implemented a minimum threshold of 100 publications per cluster, eliminating small, unstable clusters that could introduce analytical noise. This filtering retained 100 clusters, of which one additional cluster was removed as it consisted of publications regarding ‘Gateways to Clinical Trials’, recurring publications that list recent clinical trials from the Clinical Studies Knowledge Area of Prous Science Integrity^® and do not contain specific information on SSRIs. The threshold of 100 publications per cluster serves two purposes: it removes small, unstable clusters that could introduce analytical noise, and it ensures that our analysis focuses on substantive research clusters that remain stable across similar parameter settings. To assess cluster stability across parameter variations, we performed pairwise comparisons between five similar clustering solutions (10 comparisons total). For each comparison, we identified clusters with ≥100 publications in both solutions and matched them using the Hungarian algorithm to optimise total overlap. Jaccard similarity coefficients quantified the overlap between matched cluster pairs.

Our final dataset encompasses 99 clusters and a total of 36,510 publications for subsequent analyses (see eTable 3 for cluster sizes and labels in the supplementary material). Lastly, cluster labels were qualitatively assigned based on examination of the most central papers within each cluster, aided by Term Frequency-Inverse Document Frequency (TF-IDF) analysis to identify distinctive terms within each cluster’s corpus of titles and abstracts. While these automatically extracted keywords provided initial guidance, final label assignment required qualitative engagement with the manuscripts themselves to ensure labels accurately reflected the clinical and research contexts. It is important to note that while these labels reflect the dominant thematic focus, the unsupervised clustering process groups publications based on semantic and citation links that do not adhere to strict temporal or categorical boundaries. Consequently, clusters encompass antecedent and otherwise related research investigating foundational mechanisms, theoretical precursors, or related phenomena of the cluster’s primary subject.

Finally, we conducted main path analysis in Pajek [34] on the directed citation network to identify the primary intellectual trajectories within SSRI research. Main path analysis is a citation-based method that identifies the most critical sequence of papers through which knowledge has flowed over time. The method works by treating the citation network as paths of knowledge transmission: when paper B cites paper A, knowledge flows from A to B. By calculating which paths carry the most citation traffic, the algorithm identifies the dominant route through which foundational ideas have evolved and shaped subsequent research. We used the global key-route search algorithm with search path link count as traversal weights, using seed parameters between 75 and 125 to capture the most influential citation pathways that shaped the field’s development [35, 36].

Data availability and reproducibility

All Python code required to reproduce these analyses is publicly available on GitHub, with the exception of the Pajek implementation [37]. With a valid Scopus subscription, all data can be retrieved using an API key, as demonstrated in the corresponding Jupyter notebook. The visualisation repository can also be accessed on GitHub [38].

Results

The networks consist of 36,510 nodes (publications) with 338,084 and 551,227 edges (links) for the directed and hybrid networks, respectively. The volume of SSRI-related scientific publications has grown substantially over the past four decades, from approximately 100 articles in the early 1980s to almost 1600 in 2024, while the average number of references per article increased from around 20 to more than 50 in the same time period (supplementary materials, eFigure 2 and eFigure 3). To assess the stability of the research clusters, we compared matched clusters across parameter settings. This yielded Jaccard indices ranging from 0.516 to 0.833 (eTable 2 in the supplementary materials).

Main path analysis and cluster progression

The main path analysis, shown in Fig. 1, reveals distinct trends in the evolution of research on SSRIs (an interactive version is available online2). The trajectory originates with pharmaceutical topics, such as serotonin transporter mechanisms and cytochrome P450 system, involved in the metabolism of SSRIs. Following escitalopram’s 2001 approval, the main path transitions to comparative efficacy studies and pharmacoeconomic analyses across SSRI treatments.

Fig. 1.

Main path analysis of the SSRI literature from 1982 to 2025. This directed citation network highlights the “critical path” of knowledge flow within the field. Nodes represent individual publications, and directed edges represent citation links. The highlighted trajectory follows the path of highest traversal weights, tracing the primary intellectual lineage from early studies at the top to contemporary topics

The late 2000s marked a critical juncture where the main path shifted toward neuroplasticity research, particularly SSRI-induced neurogenesis mechanisms. This foundation enabled the 2010s focus on prenatal exposure studies in rodent models. The contemporary main path has forked into two prominent branches: fluvoxamine’s COVID-19 therapeutic potential and environmental pharmacology examining SSRI ecotoxicological effects in aquatic ecosystems, particularly fish populations.

Figure 2 shows the progression of each cluster within the main path relative to all other main path clusters, highlighting how research areas change significantly in magnitude over time as topics gain and lose prominence in the field (an interactive version is available online3). Note that cluster labels describe the dominant topic within the group. However, clusters include antecedent and related research sharing semantic or citation similarities. For instance, the “Fluvoxamine for COVID-19” cluster includes pre-pandemic research on Sigma-1 receptor mechanisms (e.g., Omi et al., 2014 [39]), which later formed the basis for inquiries into repurposing the drug for COVID-19.

Fig. 2.

Cluster progression of main path clusters. The visualisation tracks the relative prominence of key research areas over time. Cluster labels indicate the dominant theme. Clusters inherently encompass antecedent research on foundational mechanisms, theoretical precursors or otherwise similar phenomena

Research specialisation and diversification

Figure 3a displays the network modularity over time, which measures how strongly a network divides into distinct communities, with higher values indicating more internally cohesive and externally separated clusters [40]. A modularity score near zero suggests random connectivity patterns whereas high modularity scores approaching 1 indicate strong community structures within the network [40]. Figure 3b complements this by showing the minimum number of research clusters required to account for 50% of annual publications. Although this increase partly reflects overall growth in publication volume, it also indicates a proliferation of distinct research topics within the field. In contrast, the modularity measure in Fig. 3a captures the degree of structural separation between these communities. Together, the two indicators point to a pronounced rise in disciplinary diversification over time, followed by a modest decline in the most recent years.

Fig. 3.

Network structure metrics.(a) Modularity: higher values reflect stronger division into distinct sub-disciplines. (b) Cluster concentration: plots the minimum number of clusters required to account for 50% of the total annual publication volume. Higher values indicate that annual publications are distributed across more clusters, rather than being dominated by a few central themes

Interactive visualisation

The visualisation can be accessed online at http://immersive-ssri-evolution.surge.sh/

To examine the diverse foci within the field of SSRI research, the interactive visualisation enables an immersive analysis of the co-evolution and emergence of distinct clusters while identifying critical linkages between them. A demonstration of the interactive features of the visualisation is provided in Video 1. To illustrate one exemplary insight that can be derived from this visualisation, Fig. 4 displays two clusters of studies investigating prenatal SSRI exposure, one examining exposure in humans and the other in rodents. Figure 4a shows these clusters for the period 1982 to 2010, while Fig. 4b presents the same clusters over the entire study period.

Fig. 4.

Temporal research evolution in prenatal SSRI exposure research. Network visualisation showing two research clusters investigating prenatal SSRI exposure: one examining exposure in humans (orange) and another in rodents (blue). Panel a displays the network from 1982 to 2010, while panel B shows the complete dataset through March 2025. Node size represents publication centrality within each cluster, and connecting lines indicate citation relationships or semantic similarity. A bridging publication that connects both research domains is highlighted, demonstrating how some studies link preclinical and clinical research paradigms

Two main observations emerge from this visualisation. First, the studies which are most central to the research cluster on prenatal SSRI exposure in humans had already been conducted before the most central animal focused studies appear. Second, the visualisation reveals how pivotal publications serve as bridges between these two research domains. For example the work by Noorlander et al. [41], which, among other contributions, provides a comparative analysis of placental SSRI transfer between mouse and human models, thereby connecting insights from both clinical and preclinical research paradigms.

This example demonstrates potential uses of network visualisation tools in identifying not only the chronological development of research domains but also the critical studies that facilitate knowledge transfer across different methodological approaches within the field.

Discussion

Four decades have passed since the initial approval of an SSRI, and the corresponding body of research has undergone substantial transformation, both in scale and orientation. In the early years, only a limited number of publications appeared, concentrating on a narrow set of topics, primarily related to pharmacodynamic and pharmacokinetic properties. As the field evolved, however, the growing volume of research has been accompanied by significant diversification across a broader range of disciplines and thematic areas. Beyond this expansion in scope, the network structure itself has transformed, as research increasingly organises into distinct communities with fewer connections between them.

Clinical research continued to examine the benefits and risks of SSRI treatments in a multitude of studies on safety, special populations (such as pregnant individuals or those with specific comorbidities), and pharmacoeconomic analyses. This work often adopted a comparative approach, driven by the increasing approvals of additional SSRIs and related medications. As reflected in many of the clusters identified in our analysis, these applications extend far beyond the primary indication for depression treatment, despite modest efficacy and questioned clinical relevance in many contexts [7, 42, 43].

While the involvement of serotonin in a vast array of physiological processes, such as mood regulation, sleep, digestion, and cognition, provides a biological rationale for this breadth, it simultaneously complicates the concept of therapeutic specificity implied by the term ‘antidepressant’ [44–47]. This nomenclature was solidified to frame these agents as disease-specific treatments rather than non-specific modulators of arousal or emotion. Our network, with its proliferation of clusters across diverse somatic and psychiatric indications—including obsessive–compulsive disorder (Cluster 4), post-traumatic stress disorder (Cluster 12), generalised anxiety disorder (Cluster 34), pain conditions (Cluster 23 and 91), sleep disturbances (Cluster 41), weight management (Cluster 27), menopausal vasomotor symptoms (Cluster 45), premature ejaculation (Cluster 47), irritable bowel syndrome (Cluster 51), autism spectrum disorder (Cluster 57), and eating disorders (Cluster 58)—illustrates the tension between this label of ‘antidepressants’ and the drugs’ broad pharmacological reality.

This tension is equally apparent in the pharmacological domain, as numerous clusters in our network reflect ongoing efforts to identify the precise mechanisms of antidepressant action and to identify discrete biomarkers for depression. These research areas encompass serotonin receptor modulation (Cluster 2), neurogenesis (Cluster 8), inflammatory biomarkers (Cluster 16), the serotonin transporter gene and other pharmacogenetic factors (Clusters 21 and 73), alternative non-SERT pathways (Cluster 75), neurometabolic changes (Cluster 89), and brain-derived neurotrophic factor (BDNF) (Cluster 93). Collectively, these clusters illustrate how the original hypothesis of depression as a serotonin deficiency has shifted to increasingly complex and diverse theories involving neuroplasticity, the BDNF, and inflammatory processes [48–52]. This trajectory may reflect what has been termed the ‘pharmacological bridge’: the assumption that depression can be understood by inferring its pathophysiology from the effects of antidepressants [47]. As simpler explanations stating that depression results from a serotonin deficiency were challenged, the field adopted flexible explanatory frameworks based on the biological action of antidepressants to preserve the underlying assumption that these drugs target specific disease pathologies unique to depression [49, 53]. However, the immense research effort dedicated to identifying discrete biomarkers, visible in numerous network clusters, faces significant conceptual challenges. Not only does depression likely lack a singular pathophysiology to be ‘corrected,’ but the condition itself represents a heterogeneous construct with shifting diagnostic boundaries that have progressively expanded to encompass a wider range of human emotional experiences [13, 54].

Other contemporary research trends demonstrate a responsiveness of academia to societal concerns, with large emerging clusters focused on two pressing challenges. The first concerns the environmental impact of pharmaceuticals, reflected in Clusters 1 and 64 on aquatic ecotoxicology and environmental degradation of antidepressants. This area has received substantial attention, particularly in relation to water pollution, where SSRI exceedances have been documented in wastewater, surface water, groundwater, and sediments [55, 56]. These findings raise significant concerns about ecological impact, especially within aquatic ecosystems where SSRIs demonstrate measurable effects on organisms [56]. The European Union’s regulatory response through Directive 2008/105/EC, which mandates that the European Commission develop a strategic approach to pharmaceutical water pollution and requires mandatory environmental risk assessments (ERA) for medicines, has both resulted from and accelerated this research trajectory that has expanded dramatically since the mid-2000s [57]. Second, SSRIs, particularly fluvoxamine, have been investigated as potential COVID-19 treatments (Cluster 39, Fluvoxamine for COVID-19) based on their anti-inflammatory properties and potential anti-viral effect [58]. Early trials suggested fluvoxamine might reduce hospitalisation risk in outpatients with mild to moderate COVID-19, but subsequent studies have not been able to confirm these results and regulatory authorities have not recommended fluvoxamine for routine COVID-19 treatment [59]. This research trajectory, however, exemplifies how urgent societal needs can drive rapid drug repurposing efforts, even when evidence of efficacy remains uncertain.

Viewing these trajectories through the lens of scientific field evolution reveals patterns that align with established models of disciplinary maturation while highlighting dynamics specific to regulated biomedical research [60]. The field displays a clear progression from an initial concentration on foundational pharmacokinetics and pharmacodynamics to a broader diversification into applied domains, such as safety monitoring, utilisation patterns, and health economics [60]. However, this internal expansion is distinctively shaped by exogenous drivers, factors beyond scientific inquiry that redirect research focus. For instance, the emergence of research on environmental toxicology followed regulatory mandates like the EU Directive 2008/105/EC, whereas the surge in COVID-19–related publications reflects a crisis-driven diversification prompted by urgent public health needs.

The evolution of new theories on mechanisms of antidepressant action can similarly be interpreted in Kuhnian terms [61]. As discussed earlier, explanatory models have repeatedly shifted, from monoamine and serotonin deficiency hypotheses to more complex accounts centred on neuroplasticity, demonstrating the adaptability of “normal science” when confronted with anomalies. Rather than triggering a paradigm shift, these successive adjustments may suggest efforts to accommodate conflicting evidence by expanding the paradigm’s explanatory scope, thereby preserving the prevailing therapeutic framework.

Limitations

Several methodological considerations warrant attention. First, our network construction involved multiple parameter choices, including the number of nearest neighbours, edge weighting scheme, and resolution parameters. Each of these could yield alternative clustering configurations and potentially different narratives about the field’s evolution. While we employed systematic criteria to optimise these parameters toward achieving the most homogeneous clusters, alternative parameter combinations might have highlighted different structural features of the research landscape. However, similarity measures indicated that clustering solutions with similar granularity produced largely overlapping cluster compositions, particularly for larger clusters. Our implementation of a 100-publication minimum threshold further addresses this concern by focusing analysis on more stable clusters across parameter variations. Second, the assignment of cluster labels inevitably involves subjective elements of interpretation and prioritisation. In our approach, we relied on the most central publications within each cluster, complemented by TF-IDF analysis to identify the highest-scoring, distinctive terms. This combination provides a systematic basis for highlighting each cluster’s core themes. However, it cannot fully capture all thematic nuances contained within the full set of publications. Alternative interpretations are possible, and researchers with different domain expertise might reasonably characterise certain clusters differently. Third, our clustering approach identifies similarities across multiple dimensions, and most papers do not clearly fit within a single cluster. For example, the study by Noorlander et al. [41] of prenatal SSRI exposure incorporated diverse methodological components (high performance liquid chromatography analysis, magnetic resonance imaging, and elevated plus maze), different pharmaceutical exposures (fluoxetine and fluvoxamine), multiple settings (rodent and human), and various safety outcomes (cardiovascular and behavioural). Nonetheless, the algorithm assigns each study to a single cluster according to its strongest semantic and citation associations. Fourth, heterogeneous citation practices across disciplines may distort the network structure [62]. For instance, medical sciences contain fewer citations compared to genetics. However, our hybrid network approach mitigates some of these biases by identifying and connecting conceptually related publications that may not be directly connected through citations alone. Finally, we did not examine institutional, geographical, or funding information, as our primary objective was to map knowledge domains and their evolution. However, future studies should analyse the role of funding sources and institutional influence to provide insights into the social and economic forces shaping SSRI research trajectories.

Conclusion

This analysis highlights the remarkable breadth of SSRI research, which reflects the complex nature of the serotonin system itself. Due to the increased diversification and heterogeneity of research clusters and knowledge trajectories, this field is particularly well suited for an interactive network visualisation that enables users to chronologically trace the co-evolution of research areas. Finally, incorporating interactive visual modalities expands how we engage with and interpret research. Traditional keyword-based retrieval can exclude non-specialists who lack the domain-specific vocabulary to construct effective search queries [63, 64]. By contrast, interactive visualisations facilitate exploratory search, allowing users to identify relevant concepts through spatial proximity and structural connections rather than precise terminology. This shift from querying to navigating may lower the barrier to entry, enabling students, interdisciplinary scholars, and the public to intuitively map the field’s evolution and identify pivotal lines of research without prerequisite expertise.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviation

API

Application Programming Interface

BDNF

Brain-Derived Neurotrophic Factor

COVID-19

Coronavirus Disease 2019

EID

Electronic Identifier

ERA

Environmental Risk Assessment

OECD

Organisation for Economic Co-operation and Development

SSRI/SSRIs

Selective Serotonin Reuptake Inhibitor(s)

Author contributions

Concept and design: L.W., D.D.G., and CEH. Acquisition, analysis, or interpretation of data: L.W., D.D.G. Drafting of the manuscript: L.W. Critical review of the manuscript for important intellectual content: D.D.G. and C.E.H. Analysis: L.W. Administrative, technical, or material support: C.E.H. Supervision: C.E.H.

Funding

Open access funding provided by Copenhagen University. This study was not funded.

Data availability

The data used in this study are available from Scopus. Researchers with appropriate Scopus access can retrieve the same dataset and replicate the analysis. The repositories containing the code for data retrieval and analysis are available at: https://github.com/jarolim14/SSRI-Evolution.

Declarations

Ethics approval and consent to participate

Not applicable

Consent for publication

Not applicable

Competing interests

This study was performed under the umbrella of the Copenhagen Centre for Regulatory Science (CORS), a cross-faculty university anchored institution involving various public and private stakeholders, as well as patient organisations. The centre is purely devoted to the scientific aspects of the regulatory field and has a patient-oriented focus. The research is not a company-specific product, nor directly company related. C.E.H. is employed by CORS, and L.W. and D.D.G. are PhD candidates at CORS. In the last three years, CORS and C.E.H. have received funding from Novo Nordisk for projects not related to this study. L.W.‘s PhD fellowship is funded by a grant to CORS from Novo Nordisk A/S; however, the company was not involved in any aspect of this work. The funders were not involved in the conception of ideas, study design, data collection methodology, statistical analysis, interpretation of results, manuscript preparation, or the decision to publish. The researchers maintained complete independence throughout all phases of this work, from initial conceptualisation through final dissemination of findings.