Evolution, hotspots, and future directions of artificial intelligence in asthma research: a Web of Science-based bibliometric analysis [2016–2026]

Introduction

Asthma remains a profound global health burden characterized by extreme heterogeneity in both clinical phenotypes and molecular endotypes (1,2). Because this multidimensional complexity—driven by intricate gene-environment interactions—often exceeds the analytical capacity of traditional, generalized stepwise management frameworks, there is an urgent clinical mandate to transition toward precision prediction and personalized prevention (3,4). In recent years, artificial intelligence (AI)—particularly machine learning (ML) and deep learning (DL)—has emerged as a transformative catalyst in respiratory medicine to address this core clinical challenge (5-7). Unlike linear statistical models, AI algorithms possess inherent advantages in deciphering high-dimensional, non-linear, and multi-modal data (8,9). Currently, AI applications in asthma are proliferating across the management spectrum: in diagnosis, convolutional neural networks are enhancing accuracy by analyzing chest computed tomography imaging and acoustic respiratory signals; in risk prediction, models leverage electronic health records and natural language processing (NLP) to mine unstructured clinical text, precisely stratifying risks for exacerbation and readmission (10-12); furthermore, AI algorithms integrated with wearable sensors and environmental monitoring are making real-time digital phenotyping and personalized trigger alerts a reality (13). Notably, the contemporary frontier epoch marked by large language models promises to further revolutionize clinical decision support and patient interaction (14). In essence, AI is reshaping the entire workflow of asthma management, driving the field toward the era of digital precision medicine (15).

However, this exponential surge in algorithmic innovations has created a severe paradox of information overload for clinicians and researchers (8). Faced with a highly fragmented and interdisciplinary deluge of literature, traditional narrative syntheses or scoping reviews often fall short. They are inherently constrained by subjective selection bias and human reading capacity, making it nearly impossible to objectively depict the panoramic landscape of AI in asthma (16). Consequently, there is an urgent clinical and academic necessity to decode this vast scientific architecture and identify true translational friction between computational development and real-world implementation.

Bibliometric analysis serves as the optimal quantitative vehicle to resolve this bottleneck. Unlike traditional descriptive reviews, bibliometrics utilizes rigorous mathematical and statistical algorithms to objectively map network topologies, trace cross-institutional collaborations, and track evolutionary frontiers via citation burst detection (17,18). This approach provides unique strategic value by offering a data-driven, macroscopic blueprint of the field that cannot be achieved through manual literature synthesis.

To navigate this dense interdisciplinary landscape, this study utilizes the Web of Science Core Collection (WoSCC) database, combined with visualization tools, to conduct a comprehensive bibliometric analysis of publications on AI in asthma from January 1, 2016, to June 4, 2026. This study aims to address the following critical questions: (I) What are the global publication trends and geopolitical distribution characteristics of this domain? (II) Which nations, institutions, and journals constitute the core academic community and how do they collaborate? (III) What is the evolutionary trajectory of AI technologies in asthma research, and where does the current translational friction lie? (IV) What are the potential actionable future frontiers? By answering these questions, we aim to provide a robust, forward-looking strategic reference guide for clinicians, data scientists, and policymakers, fostering a deeper evidence-based convergence of medicine and engineering. We present this article in accordance with the BIBLIO reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0804/rc).

Methods

Data sources and search strategy

Bibliometric data were comprehensively retrieved from the Science Citation Index Expanded and Social Sciences Citation Index horizons within the WoSCC database. The WoSCC was selected as the primary data source due to its global recognition as the most stable, comprehensive, and authoritative database for high-quality peer-reviewed literature (19). Furthermore, it provides granular citation metadata that is natively compatible with mainstream bibliometric visualization platforms, such as CiteSpace and VOSviewer, thereby ensuring the mathematical reproducibility of the subsequent knowledge mapping analysis (16).

The advanced literature retrieval was executed on June 4, 2026. To minimize retrieval omission, the search strategy was meticulously constructed to capture publications at the cross-disciplinary intersection of AI and asthma, utilizing topic search (TS) tags. The standardized search query was formatted as follows: TS = (“artificial intelligence” OR AI OR “machine learning” OR “deep learning” OR “supervised learning” OR “unsupervised learning” OR “supervised clustering” OR “unsupervised clustering” OR “transfer learning” OR “reinforcement learning” OR “neural network*” OR radiomics OR “image segmentation” OR “semantic segmentation” OR “natural language processing” OR “data mining” OR “text mining” OR “ensemble learning” OR “predictive model*”) (20) AND TS = (asthma* OR “bronchial spasm” OR “bronchoconstriction”) (21).

To optimize the contemporary relevance and longitudinal quality of the analysis, the publication timeline was restricted from January 1, 2016, to June 4, 2026. The year 2016 was strategically designated as the chronological baseline, as it marks the global computational breakthrough of DL architectures and the subsequent paradigm shift toward integrating advanced algorithmic intelligence into medical clinical research (8). To ensure data integrity and minimize human extraction bias as mandated by the BIBLIO reporting guidelines, two investigators independently performed the literature screening and metadata verification; any superficial discrepancies were resolved through open consensus or consultation with a third senior clinician. While this rigorous selection strategy ensures high-impact quality, we explicitly acknowledge that the absolute reliance on the English-language WoSCC database introduces inherent selection and linguistic biases. Specifically, excluding localized non-English databases (e.g., major Chinese academic databases) potentially masks regional clinical applications and domestic algorithmic adaptations that possess profound localized utility. Documents such as meeting abstracts, editorial materials, letters, corrections, and proceedings papers were excluded. Following this rigorous screening process, a total of 1,967 valid records were obtained and exported for subsequent analysis (Figure 1).

Figure 1 Literature screening and inclusion flowchart.

Bibliometric tools and visualization analysis

All extracted raw metadata—including titles, abstracts, authors, affiliations, countries, journals, cited references, and keywords—were exported as plain text files for network topology analysis. Following data retrieval and screening, bibliometric analysis and network visualization were performed using CiteSpace 6.4.R1 (64-bit, https://citespace.podia.com/), VOSviewer 1.6.20 (https://www.vosviewer.com/), and the Bibliometrix R-package (based on R 4.5.2, https://cran.r-project.org/web/packages/bibliometrix/). These tools were employed synergistically to comprehensively mine the characteristics of the included literature. Specifically:

• Bibliometrix was utilized to quantify the annual publication output and calculate the annual growth rate of the field (22).
• VOSviewer was applied to construct distance-based network maps, focusing on co-authorship networks and keyword co-occurrence clusters, where node sizing and color modulations visually differentiate to visualize research themes (18).
• CiteSpace was used to conduct advanced network analyses, including dual-map overlays of journals, detection of references/keywords with strong citation bursts, and cluster analysis (17,23). The mathematical outputs are visually rendered as interactive knowledge graphs composed of interconnected nodes and link vectors.

Statistical analysis

Descriptive statistics were utilized to quantify publication frequencies, absolute counts, and annual growth rates. To ensure the mathematical reproducibility of the CiteSpace analysis, the algorithmic parameters were rigorously calibrated as follows: (I) time slicing: from January 2016 to June 2026, with a time slice of 1 year; (II) node types: selected independently for country, institution, author, journal, keyword, and reference depending on the analysis; (III) pruning: the pathfinder and pruning sliced networks algorithms were applied to eliminate redundant peripheral edges the network structure and highlight critical connections; (IV) clustering: the log-likelihood ratio (LLR) statistical algorithm was employed for extracting cluster labels.

To objectively critique the structural architecture and scientific weight of the identified nodes, the following core bibliometric and statistical indicators were strictly defined and utilized: (I) betweenness centrality: a graph-theoretical metric (normalized from 0 to 1) evaluating a node’s capacity to sit on the shortest path between other node pairs. Nodes exhibiting a centrality score >0.1 are designated as strategic interdisciplinary bridges or critical turning points driving global collaboration (23). (II) citation burst strength: a time-sensitive algorithmic indicator utilizing Kleinberg’s state-transition modeling to detect statistically significant surges in citation frequency or keyword appearance within a compressed timeframe, serving as a highly specific sensor for rapidly evolving research frontiers (24). (III) Modularity score (Q) and weighted mean Silhouette score (S): structural validation metrics assessing the overall reliability of network partitioning. Q >0.3 denotes a robust, well-defined community structure with clear topological boundaries, whereas S >0.7 validates highly convincing, homogeneous clustering with verified internal node similarity (25).

Results

Publication trend analysis

As summarized in Figure 2A, a comprehensive search of the WoSCC database yielded a total of 1,967 qualifying publications originating from 649 distinct sources. These documents were contributed by 11,856 authors, with an international co-authorship rate of 29.44%, highlighting the collaborative nature of this domain. The dataset has accumulated a total of 84,280 references and 28,347 citations, averaging 14.42 citations per document, which indicates a substantial level of academic influence and engagement within the scientific community. As further illustrated in Figure 2B, the annual publication output has demonstrated a robust upward trajectory over the study period, with an overall annual growth rate of 22.24%. The evolution of the field can be categorized into two distinct phases: a steady incubation phase [2016–2018], where publication volume fluctuated between 31 and 57 papers annually, followed by an exponential growth phase [2019–2026] marked by a clear inflection point in 2019. During this latter phase, output accelerated rapidly, nearly doubling from 116 papers in 2020 to 212 in 2021, and eventually peaking at 493 publications in 2025. The apparent reduction in 2026 volume (n=231) is a statistical artifact reflecting the partial-year data retrieval cutoff (June 4, 2026) rather than a plateau in academic interest. Parallel to this growth, the cumulative citation count (Figure 2C) rose consistently from 1,082 in 2016 to 28,347 in 2026, confirming that the expanding volume of literature is being met with increasing scholarly attention and validation.

Figure 2 Publishing and citation trends. (A) Key bibliometric indicators summarizing the dataset. (B) The annual volume of scientific publications. (C) The cumulative number of citations received annually. The publication count for the year 2026 represents a partial year, reflecting data retrieved up to June 4, 2026.

Global geographic distribution and collaborative networks

Spatial and structural mapping of the domain reveals a dual-centric geopolitical landscape, driven predominantly by the massive publication volumes of the United States (US) (n=678) and China (n=476) (Table 1, Figure 3A). However, a significant divergence exists between output quantity and collaboration intensity: while the two high-volume producers rely predominantly on domestic research (single country publications), with international collaboration rates (multiple country publications, MCP) below 23%, countries like the United Kingdom (UK) and Australia demonstrate superior international integration (MCP >47%) (Figure 3B). Topological network analysis mathematically identifies the UK as the ultimate global knowledge broker (highest betweenness centrality =0.14) (Table 1, Figure 3C), bridging the Atlantic and Asia-Pacific clusters, whereas China remains relatively peripheral in terms of global network connectivity despite its high volume (Figure 3C). Additionally, the Middle East is emerging as a new collaborative hotspot (Figure 3C,3D), collectively shaping a global landscape of centralized output but differentiated connectivity.

Figure 3 Global geographic distribution and collaborative landscapes. (A) Global heat map of scientific production. (B) The top 20 most productive countries. (C) Country collaboration networks. (D) Country collaboration density map. MCP, multiple country publications; SCP, single country publications.

Institutional productivity and collaborative networks

The analysis of institutional contributions reveals a distinct hierarchy in research output and collaborative influence. As detailed in Table 2, Harvard Medical School (n=53) and Imperial College London (n=51) lead in publication volume, followed by the Mayo Clinic (n=33) and the University of Washington (n=32). However, a notable discrepancy exists between productivity and network brokerage: the University of Zurich exhibits the highest betweenness centrality (0.09), followed by Johns Hopkins University and Karolinska University Hospital (0.07), acting as critical topological pivots, whereas high-output institutions like Imperial College occupy a less central position. The collaboration network (Figure 4A) is characterized by severe structural fragmentation (density =0.0091), with the largest connected component comprising only 48% of institutions, suggesting a lack of cohesive global integration. Geographically, the density map (Figure 4B) identifies three primary clusters: a dense European network centered on Imperial College and the University of Manchester; a North American clinical hub anchored by Harvard and the Mayo Clinic; and a rapidly emerging East Asian cluster (e.g., Guangzhou Medical University, Seoul National University), which has driven network expansion since 2022 but remains topologically distant from the Euro-American core, starkly reflecting the geopolitical data silos that currently constrain global algorithmic validation.

Figure 4 Institutional productivity and collaborative landscapes. (A) Institution collaboration networks. (B) Institution collaboration density map.

Analysis of authors and co-cited authors

The analysis of authorial impact and intellectual heritage delineates a distinct dual-trajectory evolution (Figure 5). As visualized in the production timeline (Figure 5A), Adnan Custovic emerges as the most prolific and enduring contributor, anchoring a major European pediatric asthma research cluster. In contrast, the post-2019 landscape is characterized by the rapid ascent of Chinese scholars (e.g., Wang J, Zhang Y) and a high-impact Medical AI & NLP cohort led by Sohn S, Liu HF, and Wi CI. The co-authorship network (Figure 5B) reveals a polycentric structure that integrates these technical hubs (e.g., the Sohn/Liu/Wi cluster) with established clinical authorities—such as Bousquet J (allergy/immunology) and Wenzel SE (severe asthma)—alongside prominently emerging distinct clusters like that of Mersha TB, representing the field’s multidisciplinary backbone. Crucially, the author co-citation analysis (Figure 5C) uncovers the field’s hybrid epistemic foundations: a clinical domain cluster defined by references to asthma phenotypes and Global Initiative for Asthma Guidelines is intricately interwoven with a methodological domain cluster referencing foundational ML architects like Breiman L (random forest) (26), Chen TQ (XGBoost) (27), and Tibshirani R (Lasso/ridge regression) (28). This structure confirms that the field is built upon the convergence of rigorous respiratory medicine and advanced computational data science.

Figure 5 Author productivity, collaboration, and co-citation networks. (A) Timeline visualization of the top 10 most productive authors. (B) Co-authorship network map. (C) Author co-citation network visualization. TC, total citation.

Analysis of journal publications and collaborations

The publication landscape conforms to Bradford’s Law of Scattering (Figure 6A), identifying a nucleus of 23 core journals that concentrate approximately 30% of the total output. In terms of productivity, Annals of Allergy, Asthma & Immunology (n=137) leads the field, while the notable presence of the engineering journal IEEE Access (n=31) within the core zone underscores the deep integration of computer science into this medical domain. Network analysis (Figure 6B) reveals the structural hierarchy of the field’s knowledge base: the Journal of Allergy and Clinical Immunology absolutely dominates with both the highest citation count [1,045] and the highest betweenness centrality (0.31). Alongside it, critical care authorities like the American Journal of Respiratory and Critical Care Medicine (centrality =0.21) and open-access multidisciplinary journals like PLoS ONE (centrality =0.17) exhibit strong betweenness centrality, functioning as essential knowledge bridges that connect disparate clusters. Furthermore, the dual-map overlay (Figure 6C) elucidates the field’s interdisciplinary architecture, demonstrating how citing journals in the Medicine, Medical, Clinical and Molecular, Biology, Immunology clusters draw heavily from foundational knowledge bases in Molecular, Biology, Genetics and Health, Nursing, Medicine, effectively synthesizing clinical pulmonology with computational and molecular methodologies.

Figure 6 Journal landscape and interdisciplinary knowledge flow. (A) Core sources identified by Bradford’s Law. (B) Cited journal co-citation network visualization. (C) A dual-map overlay of journals illustrating citation trajectories. KS, Kolmogorov-Smirnov.

Keyword co-occurrence, burst, clustering, and evolutionary trends

Keyword co-occurrence and burst analysis

The keyword co-occurrence network (Figure 7A,7B) reveals a topology dominated by the central nodes ML, AI, and asthma, which exhibit the strongest link strength and centrality, surrounded by a secondary semantic layer comprising children, risk, diagnosis, and classification. The burst detection analysis (Figure 7C) further delineates a distinct three-stage evolution: an early foundational phase [2016–2020] characterized by methodological terms like cluster analysis (strength =5.83) and data mining, reflecting an early focus on statistical phenotyping; a transitional phase [2021–2023] focused on data infrastructure such as electronic health records and guidelines; and a current frontier phase [2024–2026] marked by outcome-driven and diagnostic terms such as allergic rhinitis (strength =6.48), efficacy (strength =6.28), computed tomography (strength =5.34), and safety (strength =4.65), signaling a pivotal shift toward clinical validation.

Figure 7 Keyword co-occurrence and burst detection analysis. (A) Keyword co-occurrence network generated by CiteSpace. (B) Keyword co-occurrence network generated by VOSviewer. (C) The top 25 keywords with the strongest citation bursts.

Thematic clustering analysis

The semantic structure of the field was visualized using LLR clustering (Figure 8), yielding a high modularity score (Q=0.5193) and weighted mean silhouette (S=0.7392) that indicate a clearly defined and credible network structure. Eighteen major clusters were identified and categorized into five overarching domains: AI methodologies anchored by the largest clusters #0 artificial intelligence and #4 machine learning serving as the technical engine; clinical applications including #1 efficacy and #6 severe asthma acting as translational bridges; biomarkers & mechanisms represented by #12 cell-free dna and #9 lung disease; environmental epidemiology highlighted by #13 climate change; and algorithmic optimization such as #2 selection and #7 feature selection. The dense connectivity observed between the AI clusters and the clinical application clusters highlights the field’s maturation from theoretical modeling to practical implementation.

Figure 8 Thematic clustering of keywords.

Spatiotemporal evolution and research frontiers

The timeline view (Figure 9) illustrates the generational succession of research hotspots from 2016 to 2026, revealing two distinct evolutionary paths: a methodological path evolving from early cluster analysis and data mining to DL, culminating in advanced algorithms like convolutional neural networks and feature selection; and a clinical focus path transitioning from lung function and childhood asthma to phenotypes, molecular mechanisms and personalized medicine. This trajectory confirms that the field has successfully completed a three-stage evolution from traditional statistics to DL, and ultimately into precision clinical applications. Furthermore, the synchronization of advanced AI terms with clinical keywords like safety, placebo, and adverse reactions in the latest time zones signals a profound paradigm shift, where the research focus is pivoting from pure technology-driven exploration to value-driven clinical translation and rigorous trial validation.

Figure 9 Timeline view of keyword clusters.

Analysis of co-cited references and research frontiers

Citation bursts and shifting frontiers

The analysis of citation bursts (Figure 10A) captures the dynamic shifts in research hotspots, identifying 25 references with surging interest that map the field’s transition through three distinct phases. The inception phase [2017–2020] was led by Finkelstein [2017] (29) (burst strength =12.13), which pioneered AI-driven tele-monitoring systems. This was followed by the methodological expansion phase [2020–2022], characterized by the rise of algorithmic cornerstones like Chen TQ et al. [2016] (27) (XGBoost, strength =8.16), solidifying ML as the standard for predictive modeling. The current clinical translation phase [2023–2026] is characterized by the simultaneous citation bursts of foundational clinical studies, most notably Kaplan A et al. [2021] (9) (burst strength =12.42) and Zein JG et al. [2021] (7) (burst strength =12.14). These are complemented by other high-impact foundational works like Vos T et al. [2020] (30) (strength =11.31) and Exarchos KP et al. [2020] (31) (strength =11.25).The prominence of these works, coupled with recent reviews in The Lancet (Asher MI et al., 2021) (32), indicates that the research frontier has decisively moved towards precision medicine and severe asthma management, focusing on real-world clinical implementation.

Figure 10 Analysis of co-cited references and research frontiers. (A) The top 25 references with the strongest citation bursts. (B) Reference co-citation clustering network. (C) Timeline view of reference co-citation clusters.

Thematic clustering and intellectual structure

The reference co-citation network (Figure 10B) delineates the field’s static intellectual structure with high credibility (Q=0.8994, S=0.9584). The network is organized into distinctive knowledge domains: the central hub is dominated by cluster #0 (artificial intelligence) and cluster #1 (ige), which represent the fusion of computational algorithms and immunology at the core. Surrounding this core are specialized clinical and translational clusters such as cluster #6 (patient care management) and cluster #7 (decision support systems), which provide the clinical implementation pathways. This topology reveals a dual-core architecture where clinical problem-solving (e.g., wheezing detection, management) is tightly coupled with computational innovation (e.g., informatics, AI), confirming the interdisciplinary nature of the domain.

Evolutionary trajectory and knowledge layers

The timeline view (Figure 10C) visualizes the temporal evolution of these clusters, revealing a clear three-layered knowledge trajectory. The foundational layer [2015–2018], anchored by cluster #9 (wheezing detection) and cluster #7 (decision support systems), established the early diagnostic and computational groundwork for asthma phenotyping. This evolved into the methodological breakthrough layer [2017–2021], where computational tools were deeply integrated into immunological profiling and clinical workflows, represented by the maturation of cluster #4 (informatics) and cluster #2 (ige). Currently, the field has entered the AI-driven clinical integration layer [2021–2026], absolutely dominated by the massive largest cluster #0 (artificial intelligence). This layer signifies a paradigm shift from methodological validation to patient-centric care, where high-impact epidemiological works [e.g., Zein JG, 2021 (7); Kaplan A, 2021 (9)] link modern ML architectures with the active reorganization of knowledge to support global health applications.

Discussion

General overview

The rapid evolution of AI has catalyzed a paradigm shift in asthma research, delivering breakthrough advancements and reshaping our multidimensional understanding of this complex disease from 2016 to mid-2026. The exponential growth of literature, particularly the explosive surge post-2019 (Figure 2), reflects the synergistic drive of widespread electronic health record adoption, the maturation of advanced ML architectures (including DL and foundation models), and the deepening of interdisciplinary convergence between computational sciences and clinical pulmonology (8,15). As the first systematic bibliometric analysis focusing specifically on AI in asthma, this study comprehensively decodes the underlying research landscape and elucidates the evolutionary logic that was previously obscured by the sheer volume of data. Moving beyond traditional descriptive statistics, a deeper synthesis of this mapped landscape reveals three critical dynamics shaping the field: emerging geopolitical and epistemological disparities, the underlying friction in clinical translation, and the urgent necessity for prospective trial validation.

Global geographic landscape: strategic divergence and asymmetry

The country-level analysis unveils a structurally unbalanced global landscape characterized by distinct strategic models. The US and China stand as the first tier but exhibit contrasting developmental archetypes: the US functions as a hegemon, leveraging robust technology export capabilities within a self-sufficient ecosystem, whereas China represents a rising challenger, achieving second place in volume through rapid accumulation but remaining relatively isolated, as evidenced by its low centrality and international collaboration rate. In stark contrast to this bipolarity, the UK, despite a smaller output scale compared to the superpowers, serves as the ultimate global hub. Leveraging the highest betweenness centrality (0.14) and extensive international collaboration (48.6%), it acts as a critical router connecting the American, European, and Asian clusters. Meanwhile, nations like Australia and France have adopted a high-quality niche strategy, exerting significant influence in specific subfields through focused international cooperation (MCP >46%).

This geographic distribution further reflects an emerging data-algorithm-clinical division of labor. North America and Northern Europe dominate the high ground of clinical validation, supported by mature longitudinal cohorts (e.g., National Health Service databases) (33,34). Conversely, East Asian nations demonstrate superior capacity in algorithmic engineering and scale but face the risk of their influence being confined to tool application rather than standard definition due to a lack of globally recognized longitudinal datasets. Furthermore, while the US-UK axis remains the most enduring collaborative channel, the weak direct connectivity between the US and China creates a bipolar standoff. Crucially, a nuanced geopolitical disparity emerges regarding the Global South: while regions like Africa face a severe mismatch between disease burden and research investment (32,35), developing nations such as Argentina (centrality =0.11) and Pakistan (centrality =0.10) exhibit surprisingly high network connectivity. This indicates they are acting as vital, yet under-resourced, regional conduits in an otherwise Euro-American dominated network. Moving forward, leveraging emerging technologies like federated learning to break down geopolitical barriers and construct a multipolar collaborative network will be essential for the field’s sustainable development.

Institutional stratification: structural holes and the value of strategic intermediaries

The institutional analysis echoes the national-level findings, revealing a complex ecosystem where dual-core monopoly coexists with fragmented edges. Harvard Medical School and Imperial College London constitute the global bipolar centers of output; however, the sparse direct connectivity between them indicates the absence of a strong trans-Atlantic innovation axis. Simultaneously, Asian institutions, represented by Chinese medical universities, have formed high-density regional clusters since 2022. Yet, due to data sovereignty constraints and inward-looking evaluation systems, these clusters remain topologically distant from the Euro-American core, risking an island effect (36,37).

Within this fragmented network, the University of Zurich, Johns Hopkins University, and Karolinska University Hospital demonstrate unique strategic value. Despite not leading in volume, they occupy the highest network centrality, functioning as strategic intermediaries that efficiently glue diverse research clusters together. This phenomenon suggests that in a resource-limited context, a connector strategy—prioritizing high-frequency linkages across communities—yields greater network influence than mere volume expansion, a concept strongly aligned with the theory of structural holes (38). Given the currently low network density (0.0091), future strategies should prioritize establishing direct Harvard-Imperial axes, systematically integrating emerging Asian forces via key nodes like Zurich and Johns Hopkins, and engaging engineering powerhouses to bridge the gap between algorithmic sources and clinical scenarios. Ultimately, breaking down these institutional silos is a prerequisite for reducing translational friction and executing the large-scale, cross-continental prospective trials needed to validate AI models.

Author networks: paradigm shift and epistemic fusion

The analysis of author collaboration and co-citation clearly delineates the field’s intellectual trajectory from clinical description to technological solution. Early research, dominated by clinical experts like Custovic, focused on traditional statistical modeling (39). In contrast, the recent rise of scholars such as Sohn and Zhang Y marks a shift toward DL and ensemble methods, signaling a significant maturation of technology (40,41). However, the network structure reveals a persistent two-culture divide, where clinical mechanism researchers (e.g., Wenzel, alongside newly emerging clinical-translational clusters led by Mersha TB) (42,43) and AI methodologists (e.g., Liu HF and Wi CI) (44,45) remain in distinct communities with a solidified core-periphery structure. Co-citation analysis further confirms that the field is built upon a hybrid foundation: a methodological cluster providing algorithmic rigor (anchored by Random Forest, XGBoost, Lasso Regression via Tibshirani R) (26-28) and a clinical cluster ensuring medical relevance (anchored by Global Initiative for Asthma Guidelines, phenotyping studies) (46). This dual structure implies that single-discipline researchers are increasingly ill-equipped to lead the frontier. Future breakthroughs will necessitate composite teams possessing both clinical insight and algorithmic literacy, potentially bridging the epistemic gap through dual-PI initiatives.

Journal ecosystem: legitimacy building and niche differentiation

Journal analysis reveals the process of legitimacy building and ecological niche differentiation within this interdisciplinary domain. Currently, a high degree of specialization is evident, with only 23 core journals accounting for approximately 30% of the output. The leadership of Annals of Allergy, Asthma & Immunology indicates that clinicians remain the primary audience, while the entry of IEEE Access signals the formal acceptance of asthma research by the computer science community.

Notably, the structural hierarchy of the co-citation network demonstrates that academic authority and structural influence are highly aligned. Flagship venues like the Journal of Allergy and Clinical Immunology absolutely dominate the field, commanding both the highest citation count [1,045] and the highest betweenness centrality (0.31). Alongside it, critical care authorities like the American Journal of Respiratory and Critical Care Medicine (centrality =0.21) and multidisciplinary hubs like PLoS ONE (centrality =0.17) function as essential knowledge bridges (47). This suggests that researchers should adopt differentiated publication strategies: establishing clinical relevance in flagship journals, validating methodologies in multidisciplinary venues, or pushing algorithmic boundaries in engineering venues. Looking ahead, as multi-omics and generative AI penetrate deeper, we anticipate more top-tier basic science journals entering the core citation network, driving the field to evolve from simply applying AI to defining AI in the context of disease mechanisms (48).

Research hotspots and frontiers

Based on keyword clustering, co-citation analysis, and burst detection, the research landscape of AI in asthma has evolved into four interconnected frontiers: precision phenotyping, predictive monitoring, personalized therapy, and robust methodological infrastructure.

From clinical phenotypes to molecular endotypes: the precision classification revolution

The heterogeneity of asthma remains the primary obstacle to effective management. Traditional phenotyping based on clinical features (e.g., age of onset, atopy) is increasingly insufficient for guiding targeted biologic therapies. Our analysis reveals a paradigm shift where ML is deeply embedded in multi-omics integration (e.g., cluster #12 cell-free dna, cluster #4 machine learning, cluster #9 lung disease).

Early unsupervised clustering (e.g., SARP cohort) has evolved into sophisticated semi-supervised and multi-modal integration. Recent landmark studies, such as the conceptual frameworks proposed by Ray et al. [2022] (49) and the specific application of graph attention networks (GOAT) by Jeong et al. [2023] (50) to identify novel biomarkers like CTNNB1 and JUN for eosinophilic asthma, which were undetectable by gene expression alone. Similarly, Liu et al. [2025] (51) demonstrated that integrating proteomics with ML can accurately discriminate between eosinophilic and neutrophilic endotypes (area under the receiver operating characteristic curve, AUC =0.90), paving the way for molecularly defined diagnosis.

However, the translation of these endotypes into clinical practice is hindered by severe translational friction, primarily stemming from data heterogeneity (batch effects in multi-center omics) and the curse of dimensionality (small sample sizes vs. high-dimensional features), leading to poor external generalization (52). Future research must pivot towards multi-modal fusion, integrating radiomics (echoing the recent burst of computed tomography) and digital biomarkers (wearables) to construct robust classifiers (53). Furthermore, moving from correlation to causal ML—using Mendelian randomization to identify actionable drivers (54)—will be critical for distinguishing causative endotypes from mere associations, a leap that ultimately requires rigorous validation through prospective, biomarker-stratified clinical trials.

Closing the loop: real-time monitoring and exacerbation prediction

Preventing exacerbations is the ultimate goal of asthma management. The burst of keywords like efficacy and safety [2024–2026], coupled with the emergence of trial-specific terms like placebo and adverse reactions in our timeline analysis, signals a definitive transition from retrospective technical validation to rigorous clinical utility assessment.

AI models have demonstrated high predictive accuracy using diverse data sources. Zein et al. [2021] (7) achieved an AUC of 0.85 for hospitalization prediction using EHR data, establishing a benchmark for risk stratification. More recently, Turcatel et al. [2025] (11) leveraged Transformer architectures to predict exacerbations within a 6-month window (AUC =0.964). In the realm of digital health, aligning with the prominently identified cluster #9 (wheezing detection), the DIGIPREDICT study (55) and acoustic analysis models (56) exemplify the shift towards continuous, non-invasive home monitoring using wearables and digital stethoscopes.

However, severe translational friction persists due to data sparsity (exacerbations are rare events requiring imbalanced learning techniques like SMOTE) and the actionability gap (predicting risk 12 months ahead is less actionable than a 7-day warning). Future innovations lie in digital twins (57), which dynamically update risk profiles based on real-time environmental and physiological data, and foundation models, where audio-based pre-training (e.g., Google HeAR) could revolutionize low-cost screening and democratize precision monitoring in resource-limited geopolitical settings (58).

Personalized therapy and optimization: beyond generalized approaches

As asthma treatment advances into the biologics era, AI is becoming indispensable for predicting individual therapeutic responses.

Decision support systems are proving their value. Kaplan et al. [2021] (9) and the subsequent asthma/chronic obstructive pulmonary disease (COPD) differentiation classification tool demonstrated diagnostic accuracy superior to primary care physicians. In pharmacogenomics, Ong et al. [2024] (59) combined genome-wide association study with ML to identify novel loci (TPSAB1) predicting inhaled corticosteroids response. In the realm of targeted therapies, Di Bona et al. [2022] (60) highlighted that ML techniques have significantly enhanced the ability to predict clinical outcomes in severe asthma patients treated with biologics, paving the way for more personalized medication choices.

Despite these algorithmic triumphs, severe translational friction remains a major bottleneck, primarily driven by the lack of standardized definitions for treatment response and the confounding factors inherent in real-world observational data. Heterogeneous treatment effect modeling, using causal forests to identify subgroups benefiting from specific biologics (61), methodological frontier to address these confounders. To further overcome the inherent biases of retrospective data and bridge the epistemological gap between computational modeling and clinical practice, the field is urgently transitioning toward precision-driven adaptive platform trials. These prospective designs utilize algorithm-derived biomarker profiles for dynamic patient allocation, significantly accelerating the rigorous validation of novel severe asthma therapeutics across diverse clinical populations (62).

Methodological infrastructure: trust, interpretability, and privacy

Underpinning all applications is the robustness of the AI models themselves. The co-citation network highlights a reliance on foundational tools like XGBoost and Lasso regression, but DL is rapidly gaining ground.

Advanced architectures like graph neural networks (50) and bidirectional encoder representations from transformers for clinical NLP (63) are pushing performance boundaries. The black box nature of DL remains a primary driver of translational friction and a barrier to clinical trust. Zheng et al. [2024] (64) emphasize the need for inherently interpretable models rather than post-hoc explanations, which can be unstable. Furthermore, data silos and privacy regulations necessitate the adoption of federated learning to enable collaborative training without data sharing (65), offering a vital technical workaround for the geopolitical data sovereignty constraints discussed previously. Finally, the generation of synthetic patient data offers a transformative solution to the dual challenges of data scarcity and patient privacy, as recently demonstrated in asthma research (66,67).

These four frontiers are not isolated but synergistic. Phenotyping defines the target; monitoring provides the feedback; personalized therapy executes the intervention; and methodological rigor ensures safety and trust. The convergence of these fields towards clinical efficacy—ultimately overcoming historical translational friction and driving towards prospective global validation—marks the maturation of AI in asthma from a technological novelty to a cornerstone of precision medicine.

Strengths and limitations

The primary strengths of this study lie in its forward-looking timeliness and methodological rigor. First, to the best of our knowledge, this represents the most comprehensive bibliometric analysis of AI in asthma to date, covering the critical exponential growth period from January 1, 2016, to June 4, 2026. Unlike previous reviews, this study acutely captures the latest paradigm shifts triggered by advanced DL architectures and multi-modal integration, filling a gap in the literature regarding these disruptive technologies. Second, we employed a novel methodological triangulation strategy by synergizing CiteSpace, VOSviewer, and the Bibliometrix R-package. This multi-tool approach overcame the algorithmic limitations of single-software analyses, allowing for the cross-validation of macro-level cluster structures with micro-level temporal bursts, thereby ensuring the robustness of our findings. Furthermore, moving beyond traditional descriptive statistics, we provided strategic intelligence by elucidating the North-South geopolitical divide, the clinical-algorithmic epistemic gap, and the value of strategic intermediary institutions, offering actionable insights for policymakers to optimize resource allocation and foster interdisciplinary collaboration.

Despite these merits, several limitations warrant acknowledgement. First, data retrieval was restricted to the WoSCC. While WoSCC is widely recognized as the gold standard for high-quality academic literature and bibliometric analysis (19), this exclusivity may underrepresent technical innovations published in top-tier computer science conferences or preprint servers, potentially underestimating early-stage algorithmic breakthroughs. Additionally, the restriction to English-language publications introduces a severe linguistic and selection bias. Although our results highlight China as the second-largest producer of research in this domain, excluding major Chinese academic databases fundamentally skews the epistemological mapping of the global landscape. This limitation effectively erases localized clinical applications and domestic algorithmic adaptations from the global map, inadvertently overrepresenting Western research paradigms. Second, given that approximately 40% of the included literature was published after 2023, an intrinsic citation lag exists. Consequently, high-quality recent studies—particularly those concerning digital twins and foundation models—may exhibit artificially low citation counts despite their scientific significance; however, this temporal limitation was mitigated through the application of burst detection analysis. Finally, the metadata-driven nature of bibliometrics precludes a granular assessment of the intrinsic risk of bias in clinical evidence or the reproducibility of AI code in every study, necessitating future systematic reviews or meta-analyses to complement these macroscopic findings.

Conclusions

In conclusion, this bibliometric analysis illuminates the transformative trajectory of AI in asthma research over the past decade, revealing a paradigm shift from descriptive symptomatology to predictive precision medicine. Driven by the dual engines of big data availability and algorithmic maturation, the field has experienced exponential growth, particularly post-2019. Our findings highlight a global landscape characterized by asymmetric multipolarity: while the US and China dominate in output volume, European nations like the UK and Germany play indispensable roles as connectivity hubs, bridging the gap between algorithmic powerhouses and clinical leadership. Thematically, the field has successfully executed a three-stage evolution—advancing from early statistical phenotyping to ML-based risk stratification, and currently pivoting towards DL and large language models—thereby birthing critical frontiers in molecular endotyping, digital twin monitoring, and personalized biologic therapy.

However, the transition from algorithmic accuracy to real-world efficacy remains the ultimate challenge. Future breakthroughs will depend on dismantling three structural barriers: (I) data silos, which must be overcome via federated learning and multi-modal integration; (II) the North-South divide, which requires equitable collaboration to prevent algorithmic bias against underrepresented populations; and (III) the black box problem, which demands inherently interpretable models to foster clinical trust. Ultimately, the next decade of AI in asthma will not be defined merely by model complexity, but by the successful translation of these digital tools into tangible improvements in patient outcomes and global health equity.

Acknowledgments

We sincerely thank all those who have offered their help and support for this work.

Footnote

Reporting Checklist: The authors have completed the BIBLIO reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0804/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0804/prf

Funding: This study was supported by Jiangxi Provincial Science and Technology Program of Traditional Chinese Medicine (No. 2024A0053), Talent Team Program—Ganpo Talented Scholars Support Plan: Key Discipline Academic and Technical Leader Training Project (No. 20243BCE51125), Science and Technology Program of Jiangxi Provincial Health Commission (No. 202510018), and Clinical Research Cultivation Program, The First Affiliated Hospital of Nanchang University (No. YFYLCYJPY202523).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-0804/coif). H.Y. reports funding support from Jiangxi Provincial Science and Technology Program of Traditional Chinese Medicine (No. 2024A0053). W.S. reports funding support from Talent Team Program—Ganpo Talented Scholars Support Plan: Key Discipline Academic and Technical Leader Training Project (No. 20243BCE51125); Science and Technology Program of Jiangxi Provincial Health Commission (No. 202510018); and Clinical Research Cultivation Program, The First Affiliated Hospital of Nanchang University (No. YFYLCYJPY202523). The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Yuan L, Tao J, Wang J, et al. Global, regional, national burden of asthma from 1990 to 2021, with projections of incidence to 2050: a systematic analysis of the global burden of disease study 2021. EClinicalMedicine 2025;80:103051. [Crossref] [PubMed]
2. Ramonell RP, Gauthier MC, Ray A, et al. Biologic Medications for Severe Asthma: Implications for Understanding Pathogenic Heterogeneity and Endotypes. Annu Rev Med 2025;76:339-55. [Crossref] [PubMed]
3. Kanabar SS, Pavord ID, Hinks TSC. Treatable traits and treatment options in asthma. Respir Med 2026;254:108736. [Crossref] [PubMed]
4. Perez-Garcia J, Cardenas A, Lorenzo-Diaz F, et al. Precision medicine for asthma treatment: Unlocking the potential of the epigenome and microbiome. J Allergy Clin Immunol 2025;155:298-315. [Crossref] [PubMed]
5. Gogali A, Kyriakopoulos C, Potonos D, et al. Artificial Intelligence in Asthma: Current Status, Opportunities, and Pitfalls. J Allergy Clin Immunol Pract 2025;13:2891-901. [Crossref] [PubMed]
6. Amaral JLM, Sancho AG, Faria ACD, et al. Differential diagnosis of asthma and restrictive respiratory diseases by combining forced oscillation measurements, machine learning and neuro-fuzzy classifiers. Med Biol Eng Comput 2020;58:2455-73. [Crossref] [PubMed]
7. Zein JG, Wu CP, Attaway AH, et al. Novel Machine Learning Can Predict Acute Asthma Exacerbation. Chest 2021;159:1747-57. [Crossref] [PubMed]
8. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019;25:44-56. [Crossref] [PubMed]
9. Kaplan A, Cao H, FitzGerald JM, et al. Artificial Intelligence/Machine Learning in Respiratory Medicine and Potential Role in Asthma and COPD Diagnosis. J Allergy Clin Immunol Pract 2021;9:2255-61. [Crossref] [PubMed]
10. Wi CI, Sohn S, Rolfes MC, et al. Application of a Natural Language Processing Algorithm to Asthma Ascertainment. An Automated Chart Review. Am J Respir Crit Care Med 2017;196:430-7.
11. Turcatel G, Xiao Y, Caveney S, et al. Predicting Asthma Exacerbations Using Machine Learning Models. Adv Ther 2025;42:362-74. [Crossref] [PubMed]
12. Votto M, De Silvestri A, Postiglione L, et al. Predicting paediatric asthma exacerbations with machine learning: a systematic review with meta-analysis. Eur Respir Rev 2024;33:240118. [Crossref] [PubMed]
13. Bousquet J, Schünemann HJ, Sousa-Pinto B, et al. Concepts for the Development of Person-Centered, Digitally Enabled, Artificial Intelligence-Assisted ARIA Care Pathways (ARIA 2024). J Allergy Clin Immunol Pract 2024;12:2648-2668.e2. [Crossref] [PubMed]
14. Thirunavukarasu AJ, Ting DSJ, Elangovan K, et al. Large language models in medicine. Nat Med 2023;29:1930-40. [Crossref] [PubMed]
15. Tan LD, Nguyen N, Lopez E, et al. Artificial Intelligence in the Management of Asthma: A Review of a New Frontier in Patient Care. J Asthma Allergy 2025;18:1179-91. [Crossref] [PubMed]
16. Donthu N, Kumar S, Mukherjee D, et al. How to conduct a bibliometric analysis: An overview and guidelines. Journal of Business Research 2021;133:285-96.
17. Chen C. CiteSpace: A Practical Guide for Mapping Scientific Literature. Hauppauge, NY: Nova Science, 2016.
18. van Eck NJ, Waltman L. Software survey: VOSviewer, a computer program for bibliometric mapping. Scientometrics 2010;84:523-38. [Crossref] [PubMed]
19. Birkle C, Pendlebury DA, Schnell J, et al. Web of Science as a data source for research on scientific and scholarly activity. Quantitative Science Studies 2020;1:363-76.
20. Li X, Zhang W, Fang J, et al. Advance in the use of artificial intelligence of pulmonary nodule: evolution, trends, and future directions. Int J Surg 2025;111:7135-54. [Crossref] [PubMed]
21. Lista-Paz A, Bouza Cousillas L, Jácome C, et al. Effect of respiratory muscle training in asthma: A systematic review and meta-analysis. Ann Phys Rehabil Med 2023;66:101691. [Crossref] [PubMed]
22. Aria M, Cuccurullo C. bibliometrix: An R-tool for comprehensive science mapping analysis. Journal of Informetrics 2017;11:959-75.
23. Chen CM. CiteSpace II: Detecting and visualizing emerging trends and transient patterns in scientific literature. Journal of the American Society for Information Science and Technology 2006;57.
24. Kleinberg J. Bursty and Hierarchical Structure in Streams. Data Mining and Knowledge Discovery 2003;7:373-97.
25. Chen C, Ibekwe-SanJuan F, Hou J. The structure and dynamics of cocitation clusters: A multiple-perspective cocitation analysis. Journal of the American Society for Information Science and Technology 2010;61:1386-409.
26. Breiman L. Random Forests. Machine Learning 2001;45:5-32.
27. Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; San Francisco, California, USA: Association for Computing Machinery; 2016:785-94.
28. Tibshirani R. Regression Shrinkage and Selection Via the Lasso. Journal of the Royal Statistical Society: Series B (Methodological) 1996;58:267-88.
29. Finkelstein J, Jeong IC. Machine learning approaches to personalize early prediction of asthma exacerbations. Ann N Y Acad Sci 2017;1387:153-65. [Crossref] [PubMed]
30. Global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 2020;396:1204-22.
31. Exarchos KP, Beltsiou M, Votti CA, et al. Artificial intelligence techniques in asthma: a systematic review and critical appraisal of the existing literature. Eur Respir J 2020;56:2000521. [Crossref] [PubMed]
32. Asher MI, Rutter CE, Bissell K, et al. Worldwide trends in the burden of asthma symptoms in school-aged children: Global Asthma Network Phase I cross-sectional study. Lancet 2021;398:1569-80. [Crossref] [PubMed]
33. Al Sallakh MA, Vasileiou E, Rodgers SE, et al. Defining asthma and assessing asthma outcomes using electronic health record data: a systematic scoping review. Eur Respir J 2017;49:1700204. [Crossref] [PubMed]
34. Wang E, Wechsler ME, Tran TN, et al. Characterization of Severe Asthma Worldwide: Data From the International Severe Asthma Registry. Chest 2020;157:790-804. [Crossref] [PubMed]
35. The Global Asthma Report 2022. Int J Tuberc Lung Dis 2022;26:1-104.
36. Rieke N, Hancox J, Li W, et al. The future of digital health with federated learning. NPJ Digit Med 2020;3:119. [Crossref] [PubMed]
37. Rajpurkar P, Chen E, Banerjee O, et al. AI in health and medicine. Nat Med 2022;28:31-8. [Crossref] [PubMed]
38. Burt RS. Structural Holes The Social Structure of Competition. Harvard University Press; 1992.
39. Dharmage SC, Perret JL, Custovic A. Epidemiology of Asthma in Children and Adults. Front Pediatr 2019;7:246. [Crossref] [PubMed]
40. Sagheb E, Wi CI, King KS, et al. AI model for predicting asthma prognosis in children. J Allergy Clin Immunol Glob 2025;4:100429. [Crossref] [PubMed]
41. Zhang Y, Hai Y, Song B, et al. Screening and Validation of Potential Biomarkers of Immune Cells in Childhood Asthma Patients via Mendelian Randomization and Machine Learning. J Inflamm Res 2025;18:2583-600. [Crossref] [PubMed]
42. Wenzel SE. Severe Adult Asthmas: Integrating Clinical Features, Biology, and Therapeutics to Improve Outcomes. Am J Respir Crit Care Med 2021;203:809-21. [Crossref] [PubMed]
43. Kang M, Ko E, Mersha TB. A roadmap for multi-omics data integration using deep learning. Brief Bioinform 2022;23:bbab454. [Crossref] [PubMed]
44. Sohn S, Wang Y, Wi CI, et al. Clinical documentation variations and NLP system portability: a case study in asthma birth cohorts across institutions. J Am Med Inform Assoc 2018;25:353-9. [Crossref] [PubMed]
45. Wi CI, Pongdee T, Seol HY, et al. Application of a natural language processing algorithm to early asthma ascertainment for adults in the era of electronic health records. J Allergy Clin Immunol Glob 2026;5:100618. [Crossref] [PubMed]
46. Rajvanshi N, Kumar P, Goyal JP. Global Initiative for Asthma Guidelines 2024: An Update. Indian Pediatr 2024;61:781-6.
47. Leydesdorff L. Betweenness centrality as an indicator of the interdisciplinarity of scientific journals. J Am Soc Inf Sci Technol 2007;58:1303-19.
48. Reel PS, Reel S, Pearson E, et al. Using machine learning approaches for multi-omics data analysis: A review. Biotechnol Adv 2021;49:107739. [Crossref] [PubMed]
49. Ray A, Das J, Wenzel SE. Determining asthma endotypes and outcomes: Complementing existing clinical practice with modern machine learning. Cell Rep Med 2022;3:100857. [Crossref] [PubMed]
50. Jeong D, Koo B, Oh M, et al. GOAT: Gene-level biomarker discovery from multi-Omics data using graph ATtention neural network for eosinophilic asthma subtype. Bioinformatics 2023;39:btad582. [Crossref] [PubMed]
51. Liu D, Zhao L, Liu Y, et al. Predictive model for inhaled corticosteroid response in hospitalized asthma: implications for precision medicine. J Thorac Dis 2025;17:7112-23. [Crossref] [PubMed]
52. Dessie EY, Gautam Y, Ding L, et al. Development and validation of asthma risk prediction models using co-expression gene modules and machine learning methods. Sci Rep 2023;13:11279. [Crossref] [PubMed]
53. Qian Y, Wang F, Gao J, et al. Integrating Extracellular Vesicles Proteomics and Clinical Parameters to Develop a High-Precision Predictive Model for Severe Asthma. J Inflamm Res 2025;18:17945-60. [Crossref] [PubMed]
54. Wang H, Tu J, Zhang J, et al. Integrated multi-omics profiling reveals shared mechanistic pathways in asthma-allergic rhinitis comorbidity: A hybrid machine learning framework leveraging Mendelian randomization for precision diagnostics. World Allergy Organ J 2025;18:101120. [Crossref] [PubMed]
55. Chan AHY, Te Ao B, Baggott C, et al. DIGIPREDICT: physiological, behavioural and environmental predictors of asthma attacks-a prospective observational study using digital markers and artificial intelligence-study protocol. BMJ Open Respir Res 2024;11:e002275. [Crossref] [PubMed]
56. Emeryk A, Derom E, Janeczek K, et al. Home Monitoring of Asthma Exacerbations in Children and Adults With Use of an AI-Aided Stethoscope. Ann Fam Med 2023;21:517-25. [Crossref] [PubMed]
57. Khosh-Amadi N, Talebi S, Cheung F, et al. Development of a spatially contextualised and AI-enabled digital twin for asthma-specific indoor air quality management. Journal of Building Engineering 2025;118:115027.
58. Ehtesham A, Kumar S, Singh A, et al. Pediatric Asthma Detection with Googles HeAR Model: An AI-Driven Respiratory Sound Classifier. CoRR 2025;abs/2504.20124.
59. Ong MS, Sordillo JE, Dahlin A, et al. Machine Learning Prediction of Treatment Response to Inhaled Corticosteroids in Asthma. J Pers Med 2024;14:246. [Crossref] [PubMed]
60. Di Bona D, Spataro F, Carlucci P, et al. Severe asthma and personalized approach in the choice of biologic. Curr Opin Allergy Clin Immunol 2022;22:268-75. [Crossref] [PubMed]
61. Wang T, Keil AP, Buse JB, et al. Glucagon-like Peptide 1 Receptor Agonists and Asthma Exacerbations: Which Patients Benefit Most? Ann Am Thorac Soc 2024;21:1496-506. [Crossref] [PubMed]
62. Israel E, Denlinger LC, Bacharier LB, et al. PrecISE: Precision Medicine in Severe Asthma: An adaptive platform trial with biomarker ascertainment. J Allergy Clin Immunol 2021;147:1594-601. [Crossref] [PubMed]
63. Agnikula Kshatriya BS, Sagheb E, Wi CI, et al. Identification of asthma control factor in clinical notes using a hybrid deep learning model. BMC Med Inform Decis Mak 2021;21:272. [Crossref] [PubMed]
64. Zheng L, Ohde JW, Overgaard SM, et al. Clinical Needs Assessment of a Machine Learning-Based Asthma Management Tool: User-Centered Design Approach. JMIR Form Res 2024;8:e45391. [Crossref] [PubMed]
65. Xu J, Talankar S, Pan J, et al. Combining Federated Machine Learning and Qualitative Methods to Investigate Novel Pediatric Asthma Subtypes: Protocol for a Mixed Methods Study. JMIR Res Protoc 2024;13:e57981. [Crossref] [PubMed]
66. Pammi M, Shah PS, Yang LK, et al. Digital twins, synthetic patient data, and in-silico trials: can they empower paediatric clinical trials? Lancet Digit Health 2025;7:100851. [Crossref] [PubMed]
67. Hodges R, Tokunaga K, LeGrand J. A novel method to create realistic synthetic medication data. JAMIA Open 2023;6:ooad052. [Crossref] [PubMed]