Introduction

Metabolic syndrome (MetS) represents a complex cluster of cardiometabolic risk factors, primarily encompassing central obesity, atherogenic dyslipidemia, hypertension, and hyperglycemia.1 Driven by sedentary lifestyles and dietary shifts, the global prevalence of MetS now affects approximately one-quarter of the adult population.2–4 This systemic dysregulation significantly amplifies the risk of developing type 2 diabetes, severe cardiovascular disease (CVD), and premature mortality.1,5 Under widely adopted traditional criteria, clinical diagnosis requires the presence of at least three out of five predefined metabolic thresholds.6 While clinically convenient, this strict dichotomous (yes/no) approach compresses a heterogeneous metabolic spectrum into a single label, often assigning identical diagnoses to patients with divergent risk trajectories. Recognizing these diagnostic limitations, major medical societies are now shifting toward continuous risk staging. Recently, the American Heart Association (AHA) introduced the Cardiovascular-Kidney-Metabolic (CKM) staging framework,7 and the European Atherosclerosis Society (EAS) proposed a pathophysiology-based clinical staging system for systemic metabolic disorders.8 These consensus updates highlight the need to transition from categorical definitions to quantitative severity assessments.

For instance, a patient whose metrics barely cross three thresholds are categorized identically to one with severe abnormalities across all five components.9 Recent methodological evaluations emphasize that this binary framework results in a substantial loss of information regarding the severity of individual risk factors, failing to capture the full spectrum of metabolic abnormality and thereby diminishing predictive precision.10,11 Although clinical practice sometimes employs the number of fulfilled criteria as a crude proxy for severity, this simple counting method fails to capture the nuanced spectrum of MetS severity.12 To overcome this limitation, early methodologies turned to established statistical techniques, paving the way for the first continuous severity scores.13 Techniques such as Confirmatory Factor Analysis (CFA) were instrumental in constructing weighted scores that acknowledge the differential contributions of each MetS component to the underlying pathology. Other approaches employed Z-score standardization or statistical metrics like Mahalanobis distance to generate continuous indices.14 These statistical models facilitated the transition from categorical labeling to continuous severity quantification.

Current continuous severity scores reflect the metabolic dysfunction continuum described in the early phases of the AHA cardiovascular-kidney-metabolic (CKM) framework.7 In these initial phases, statistical scores provide a clear quantification of accumulating adiposity and subclinical risk factors. However, scores derived solely from classic diagnostic components fall short in capturing the advanced multi-organ continuum. They fail to map onto advanced stages that explicitly integrate renal dysfunction and clinical CVD. Furthermore, basic clinical scores lack the mechanistic granularity required by the EAS consensus.8 The EAS emphasizes the necessity of differentiating underlying pathophysiological drivers, such as insulin resistance versus lipid-driven phenotypes. Traditional metrics compress diverse metabolic dysregulations into a unified statistical construct. This obscures the distinct organ-specific damage and molecular etiologies highlighted by both frameworks, underscoring the necessity of transitioning toward machine learning and multi-omics approaches.

The limitations of traditional statistical models in handling high-dimensional data necessitated the application of machine learning (ML). ML algorithms further refine this assessment by identifying complex, non-linear relationships within clinical data, allowing for a more granular definition of metabolic dysfunction than traditional rule-based methods.15 Supervised learning, for instance, can build high-accuracy severity scoring models by integrating multiple variables, while unsupervised learning offers an unbiased approach to discovering novel disease “endotypes” based on intrinsic biological patterns.16 While early real-world applications of these ML models demonstrate improved risk stratification, their routine clinical adoption remains hindered by practical translational challenges. Therefore, in this narrative review, we synthesize current evidence on quantitative MetS assessment, focusing on how data-driven methods improve risk prediction, address clinical validation, and navigate these translational barriers.

Beyond Binary Diagnosis: Statistical Approaches to Continuous Severity Scoring

The quantification of MetS severity has progressed through increasingly complex models to improve predictive accuracy. Figure 1 characterizes these methodologies across eight performance domains, comparing traditional component counts with advanced statistical and digital twin (DT) models. This systematic comparison highlights the trade-offs between clinical interpretability and predictive precision.

Figure 1 Methodological evolution of MetS severity assessment. The figure integrates a timeline of methodological milestones with a multidimensional heatmap comparing performance across eight domains (complexity, data requirement, interpretability, personalization, predictive accuracy, mechanistic insight, cost, and clinical maturity; illustrative scores 1–5 synthesized from literature). For instance, the digital twin maximizes theoretical personalization but currently lacks clinical maturity, reflecting its status as an emerging frontier. This trajectory illustrates the field’s progression from static, binary diagnostic thresholds toward continuous scoring, multi-omics endotyping, and future dynamic models, driving the shift toward precision metabolic medicine.

Counting Abnormal Components as a Proxy for MetS Severity

A simple extension of traditional binary MetS diagnosis involves counting the number of abnormal components as a proxy for severity, stratifying individuals based on fulfilling 0–5 criteria.17,18 This approach provides an intuitive, graded view of risk: a higher tally correlates with a greater burden of complications. Illustratively, each additional MetS component is associated with a 20–40% increase in cardiovascular event risk, and clustering patterns like hypertension with obesity amplify associations with chronic kidney disease and cancers.19–22 Its simplicity aids rapid stratification in resource-limited settings using routine measurements, and it has utility in pediatric populations for early metabolic dysregulation detection.23,24

However, this approach has significant limitations, treating all components equally without accounting for deviation degrees, potentially discarding substantial biological information as noted in information-theoretical analyses.10,11 Such oversimplification overlooks pathophysiological nuances, where mild abnormalities may be equated to severe ones, leading to imprecise risk assessment. This underscores the need for more refined methods, such as the weighted continuous scores discussed in subsequent sections, to better capture MetS’s heterogeneous spectrum.

Factor Analysis and Principal Component Analysis: Constructing Weighted Severity Scores

Continuous MetS scores derived from statistical modeling operates on the core hypothesis that an unobservable, latent “MetS factor” underlies the five clinical indicators, which are merely its external manifestations.25 CFA quantifies the contribution of each clinical measure by calculating its “factor loading” on this latent construct, a process that has enabled the development of customized scoring systems for diverse populations stratified by sex, age, and ethnicity.13,14,25–30 In parallel, PCA, a data-reduction technique, consolidates multiple correlated clinical variables into a few uncorrelated “principal components”31 In MetS research, the first principal component (PC1) is frequently adopted as a continuous severity score, as it explains the maximum variance across all original indicators and has been shown to effectively predict the risk of future diabetes and cardiovascular events.32–35Table 1 provides a detailed summary of representative statistical model-based scoring methods.

Table 1 Summary of Statistical Model-Based Continuous MetS Scoring Methods

Standardization, Distance Metrics, and Other Novel Indices

Another class of methods focuses on standardizing clinical indicators with disparate units to facilitate a composite evaluation. Among these, Z-score-based approaches are particularly common, offering an intuitive reflection of an individual’s metabolic health relative to their age- and sex-matched peers.14 A more sophisticated statistical technique utilizes the Mahalanobis distance, which calculates the multidimensional statistical distance between an individual’s risk factor profile and a “healthy” clinical threshold.32,36 Furthermore, researchers have validated novel and easily computable indices, such as the siMS score, which accurately reflects early cardiometabolic burden and tracks continuous risk trajectories even in pre-MetS populations.37,38 More recently, assessment methodologies have begun to integrate modern technology, such as using 3D body scanning and smartphone applications to derive body volume data for predictive models, signaling a progression from purely statistical methods toward more complex data science paradigms.26Table 2 summarizes these innovative clinical threshold-based indices.

Table 2 Summary of Clinical Threshold-Based Continuous MetS Scoring Methods

These pioneering statistical scoring methods, validated in large population-based cohorts,25,29 introduced concepts of “weighting” and a “continuous spectrum” that marked a significant advance over simplistic binary diagnosis. Nevertheless, their reliance on a few core variables and assumptions of linearity limited their utility for the high-dimensional, complex data that characterize modern clinical research. This challenge motivated the use of machine learning, discussed next.

Machine Learning and Multimodal Data Integration for Advanced AssessmentApplication of Machine Learning in MetS Severity Assessment

The application of ML in MetS research has progressively shifted from binary diagnostic classification toward sophisticated prognostic modeling.15,39 Early supervised learning models were primarily designed to differentiate patients with MetS from healthy controls.40,41 While these classifiers often achieve high discriminatory performance, they fundamentally automate traditional categorical thresholds and do not inherently generate a validated, continuous severity metric.42–44 To address this limitation, current methodological advancements utilize regression-based ML approaches to calculate continuous severity scores.39,42 By capturing complex, non-linear feature interactions without assuming equal weights for each component, these models map patients onto a continuous risk continuum.40,41

Crucially, the clinical value of such quantified metrics relies on their prognostic significance.39 As a result, longitudinal ML frameworks increasingly integrate baseline multi-modal data to forecast the exact probability of future adverse outcomes, such as incident cardiovascular events, directly fulfilling the core objective of clinical risk staging.43,44 Parallel to supervised risk scoring, unsupervised cluster analysis provides a data-driven method to explore the intrinsic heterogeneity of MetS without relying on predefined clinical labels.45,46 By stratifying patients into homogeneous subgroups based on underlying biological patterns, these algorithms reveal mechanistically distinct endotypes that often exhibit divergent prognostic trajectories, thereby facilitating targeted precision interventions.16,47

As ML-based phenotyping becomes central to metabolic research, maintaining rigorous methodological transparency is essential.43 Future investigations developing either diagnostic classifiers or longitudinal prognostic models must adopt standardized reporting frameworks, specifically the TRIPOD+AI statement.48 Furthermore, to ensure clinical applicability and mitigate the risk of overfitting, it is imperative to rigorously assess the risk of bias using the PROBAST+AI guidelines before translating these models into routine practice.49

Foundational Research on MetS Subtyping Using Clinical and Anthropometric Data

The potential of unsupervised learning to redefine metabolic diseases has been powerfully demonstrated in foundational research on diabetes subtyping.50 With the increasing availability of large-scale clinical datasets, unsupervised clustering of routine clinical and anthropometric data has become a central method for identifying the intrinsic heterogeneity of MetS.16 The validity of this strategy was first established through breakthrough research in the field of type 2 diabetes. Longitudinal analyses of the Swedish ANDIS cohort have demonstrated that data-driven clustering can effectively stratify adult-onset diabetes into five distinct subtypes, providing a prospectively validated framework for precision risk assessment.51

Following this precedent, research in MetS rapidly advanced, demonstrating the robustness of this approach across different analytical techniques and populations. A landmark study utilizing data from the UK Biobank established a methodological benchmark by identifying five MetS endotypes using the core diagnostic components.16 Its subsequent external validation in the ethnically distinct Taiwan Biobank demonstrated robust cluster stability and reproducibility across algorithms and populations. However, the ultimate utility of unsupervised endotyping relies on clinical actionability. These models must demonstrate that derived endotypes improve risk prediction beyond standard clinical scores to directly guide precision interventions. Complementing this work, another study on a Chinese population employed a different statistical approach called the multi-trait finite mixture regression model (MFMR), using ten functional indicators.45 This analysis similarly identified four major metabolic subtypes, further confirming that the heterogeneity within MetS can be reliably uncovered with varied sets of clinical inputs and clustering algorithms.

Collectively, these large-scale cohort studies demonstrate that unsupervised learning can effectively deconstruct the heterogeneous MetS population into more homogeneous subtypes using only routinely collected clinical data. This data-driven classification approach offers a novel dimension for understanding the intrinsic heterogeneity of MetS. To investigate the molecular mechanisms underlying these clinical subtypes, research is now integrating multi-omics data.

Multi-Omics Data Integration for Advanced Assessment

Figure 2 presents a conceptual framework for this multi-modal data integration, illustrating how patient-centric data streams, from multi-omics to real-time inputs, are processed by a central AI/ML engine. This integrated approach enables the derivation of dynamic severity scores and the discovery of novel disease endotypes.

Figure 2 Conceptual framework mapping multi-modal data integration to clinical translation in metabolic syndrome assessment. The diagram delineates the flow of heterogeneous patient data (genomics, proteomics/metabolomics, wearable metrics, electronic health records, medical imaging, and environmental/behavioral inputs) into a central AI/ML integration engine. This computational processing yields two primary outputs: a dynamic severity score for continuous risk quantification and novel disease endotypes for patient re-stratification. The rightmost panel outlines the essential translational pipeline: prospective validation, data governance and privacy protocols, and clinical decision support integration required to transition these algorithms into routine clinical practice.

Emerging research now integrates multiple data sources, such as multi-omics, imaging, and behavioral data, to construct a comprehensive, dynamic picture of metabolic health. Integrating standard clinical variables with high-dimensional multi-omics, imaging, and longitudinal behavioral data allows for the construction of high-resolution metabolic phenotypes. Such integration enables a more responsive and precise evaluation of MetS severity.52

Non-Invasive Staging with Medical Image-Based Deep Learning

Deep learning, particularly the application of Convolutional Neural Networks (CNNs) to medical image analysis, provides a powerful non-invasive tool for MetS severity assessment. For instance, analyzing abdominal CT scans with CNNs to quantify organ-level changes, such as hepatic steatosis, offers a critical dimension for evaluating MetS severity.53 This method moves beyond circulating biomarkers to directly visualize and quantify the tangible impact of metabolic dysregulation on target organs. This provides a potentially more direct assessment of current disease severity than biomarker levels alone. Artificial intelligence models can extract subtle textures and patterns from medical images that are imperceptible to the human eye, patterns that may correlate with early disease stages or specific pathophysiological states.54

Integrating Multi-Omics Data for Deeper Mechanistic Endotypes

Investigating the molecular mechanisms of metabolic syndrome requires integrating multi-omics data to connect clinical manifestations with underlying biological pathways. Recent deep phenotyping investigations demonstrate the feasibility of this systems-level approach. For instance, integrating metabolomic and microbiome data has successfully defined distinct metabolic endotypes that substantially outperform traditional clinical indices in stratifying insulin resistance and cardiometabolic risk.55

A landmark study, for instance, bypassed traditional clinical diagnoses altogether, clustering patients solely based on their proteomic and metabolomic data. This data-driven reclassification yielded three novel molecular subgroups that transcended conventional disease boundaries, with one group characterized by severe dyslipidemia and another by insulin/glucose dysregulation and immune activation.56 This work shows how multi-omics can reveal shared pathophysiological bases missed by clinical diagnostics. Genomic approaches now use pathway-specific partitioning rather than global polygenic risk scores. Recent large-scale genomic analyses have successfully clustered metabolic disease loci into distinct mechanistic pathways providing a validated genetic framework for mechanism-based risk stratification.57,58 Most recently, investigators integrating plasma metabolomics in cardiovascular-kidney-metabolic (CKM) syndrome identified three metabolic subtypes significantly associated with clinical severity stages, and pinpointed a panel of 20 metabolites that could accurately identify individuals in a high-severity state.59

However, translating these findings into clinical tools faces several challenges. We grapple with the “curse of dimensionality” and the inherent heterogeneity of multi-omics datasets, requiring robust statistical methods to disentangle true biological signals from noise.60,61 Furthermore, ensuring the interpretability of these complex models remains paramount; for clinical utility, we must understand not just what the model predicts, but why.62,63 Yet, the potential rewards are immense. The integration of multi-omics data offers a potential framework for refining MetS classifications through the identification of biologically distinct endotypes, which may inform more targeted prevention and therapeutic interventions.

Future Directions: The Digital Twin Paradigm

The DT framework represents an experimental frontier rather than a mature clinical tool. It envisions utilizing artificial intelligence to integrate longitudinal data into dynamic virtual physiological models.64 While conceptually promising for dynamic assessment, its application in metabolic medicine remains nascent. Realizing this paradigm requires overcoming substantial technical barriers, including continuous data capture, system interoperability, and robust data governance. Furthermore, these computational frameworks necessitate validated mechanistic models and rigorous prospective testing prior to clinical implementation.65–67 Consequently, the DT currently serves as an aspirational research target.

Integration of Environmental and Behavioral Data

Recent advances incorporate environmental and behavioral data into ML frameworks to improve MetS severity assessment.68 These factors dynamically modulate MetS progression through gene-environment interactions. Nutrigenomics has become a key pillar of this integration, enabling ML models to predict severity by capturing the non-linear relationships between diet and metabolic pathways.68 For example, a multi-task deep learning model integrating genetic, nutritional, and clinical data demonstrated exceptional performance in MetS prediction (AUC > 0.90), outperforming single-modality approaches.40 Similarly, incorporating lifestyle features such as smoking, alcohol consumption, and sleep duration into computational clustering algorithms has successfully identified “lifemetabotypes” that can predict MetS trajectories with high accuracy.69

Incorporating a Global Epidemiological Dimension

Global epidemiological data add a population-level dimension to the quantification of MetS severity, revealing geographical, socioeconomic, and cultural variations that influence risk stratification.70 As the global prevalence of MetS rises, ML models that integrate these data can achieve cross-cultural generalizability and highlight health inequalities. For instance, a model that fused sociodemographic factors (eg, income, urbanization index) with lifestyle risks achieved an AUC of 0.88, underscoring environmental determinants like vegetable intake and obesity rates can amplify genetic susceptibility in low- and middle-income countries.44 This global perspective facilitates stratified severity scoring, identifies high-risk subgroups, and thereby informs precision public health strategies.

The power to realize this ambitious multi-modal vision is fundamentally rooted in large-scale data infrastructures. Biobank-scale cohorts like the UK Biobank are enabling robust multi-omics discoveries by linking molecular data to long-term health records.71 National initiatives such as the All of Us Research Program are further laying the groundwork for future paradigms like the DT by prospectively integrating genomics, wearables, and clinical data.72 To translate these insights into globally equitable models, the field is leveraging international consortia and privacy-preserving AI like federated learning to train on diverse datasets without centralizing sensitive information.73 The sophistication of these methods is secondary to their real-world impact. Their clinical utility must be validated against hard endpoints, discussed next.

Clinical Validation and Prognostic Significance: Bridging Current Severity with Future Outcomes

The methodologies described above represent a paradigm shift towards high-resolution, data-driven metabolic phenotyping. However, the sophistication of these tools is secondary to their impact on patient outcomes. Whether derived from supervised severity scores, unsupervised endotypes, or multi-modal feature spectra, these novel metrics must be rigorously validated against definitive clinical hard endpoints.16,42,56 This validation process serves as the crucial bridge between data-driven discovery and clinical practice, answering the fundamental question: Does the “current severity” we have quantified truly predict adverse long-term health outcomes? The universal framework for this validation employs survival analysis models to assess the strength of association between different severity strata or subtypes and the future incidence of adverse events.45,46

Consistent validation affirms the prognostic value of these new methodologies. For supervised learning scores, studies confirm that higher model-computed severity scores predict worse outcomes; for instance, individuals identified with MetS by a non-invasive model had a 51% greater risk of future CVD (significant HR).39 Similarly, a large longitudinal cohort of Chinese adults demonstrated a linear dose-response relationship, where individuals in the highest quartile of the continuous MetS severity score faced a 2.8-fold increased risk of incident cardiovascular events.35 Unsupervised learning further demonstrates its value by revealing prognostic heterogeneity. Research has identified subtypes such as a “hyperglycemic” profile and, more critically, validated through long-term follow-up that these subtypes carry markedly different risks. For example, the “hyperglycemic” subtype was most strongly linked to liver and pancreatic cancer, while the “obesity” subtype showed the strongest association with atrial fibrillation.16 These findings refine a generic MetS diagnosis into multiple disease trajectories with specific organ-level predispositions. This stratification directly guides targeted pharmacotherapy. Identifying the obesity-driven phenotype with elevated atrial fibrillation supports incretin-based treatments. Recent evidence confirms that GLP-1 receptor agonists lower incident atrial fibrillation in overweight populations.74 Furthermore, monitoring efficacy requires tracking detailed body composition changes rather than relying solely on body mass index.75

Multi-modal data integration, by providing deeper biological insights, offers even more powerful prognostic information. Image-derived profiles (IDPs) based on deep learning have demonstrated superior capability in predicting future complications compared to traditional clinical definitions.53 Similarly, studies incorporating dynamic data have shown that glycemic variability (GV) is a more potent predictor of major adverse cardiovascular events (MACE) than conventional static glucose measures, with high GV more than doubling the MACE risk (Odds Ratio = 2.21).76 These findings compellingly demonstrate that functional indicators of severity capture prognostic information which traditional, static measures are missing.

In summary, clinical validation is the pivotal step that imbues all novel MetS severity assessment methods with clinical relevance. Demonstrating that greater severity predicts worse outcomes is essential for these approaches to have clinical value.

Clinical Translation, Interpretability, and Future Directions

Machine learning can improve MetS severity quantification, but clinical adoption faces several barriers. However, bridging the gap between research validation and routine clinical adoption presents a distinct set of formidable challenges. Among these, model interpretability, validation rigor, and implementation feasibility are paramount determinants of their ultimate clinical adoption.

The Critical Role of Explainable AI in Deciphering Severity

For any severity score or subtype classification derived from machine learning to gain clinical traction, the underlying decision-making logic must be both transparent and comprehensible. Clinicians and patients alike require trust in a model’s judgment before they can confidently act upon its recommendations.77 However, many high-performance ML models are “black boxes”, a major barrier to clinical translation.78 Explainable AI (XAI) technologies have emerged to open this black box.43 Widely applied XAI techniques, such as SHapley Additive exPlanations (SHAP) and LIME (Local Interpretable Model-agnostic Explanations), have shown early promise in MetS research.43 For instance, SHAP analysis has been used to reveal that high-sensitivity C-reactive protein (hs-CRP) and liver function indicators are key drivers in MetS prediction, while plasma histidine levels are strongly and negatively correlated with hepatic steatosis, identifying it as a key protective factor for the liver.15,79 This interpretability is crucial because it bridges the gap between prediction and mechanistic understanding. By identifying the key factors that drive a “high severity” classification, XAI not only enhances clinician trust in model outputs but also provides direct clues for identifying personalized intervention targets, thereby guiding more precise clinical decisions.

Deploying these models in clinical workflows requires overcoming several challenges. First are the data challenges: building robust and equitable ML models demands large-scale, diverse, and standardized datasets. Yet, most existing models are trained on specific populations, and their generalizability to different geographical regions and ethnic groups remains a significant challenge.39 Second is the need for rigorous model validation: all models must undergo stringent external validation before clinical deployment to prevent overfitting, a scenario where a model performs perfectly on training data but poorly on new, unseen data, which could lead to erroneous clinical judgments.43 Finally, there are the issues of cost and accessibility: models reliant on multi-omics analyses or advanced medical imaging are limited in their application in resource-constrained settings due to high costs and technical requirements.52 Consequently, a vital direction for both academia and industry is the development of predictive models based on easily accessible, low-cost, non-invasive metrics to enhance technological accessibility and equity.80,81

Conclusion and Future Directions

The quantification of MetS severity is undergoing a paradigm shift, transitioning from rigid binary diagnostic thresholds toward high-resolution, data-driven phenotypic staging. The primary contribution of recent advancements is the capacity of machine learning and multi-omics integration to map patients onto a continuous risk continuum and uncover mechanistically distinct disease endotypes. However, bridging the gap between computational discovery and clinical application requires overcoming significant translational barriers. Future efforts must strictly prioritize prospective validation against hard clinical endpoints, harmonize endotype definitions across diverse populations, ensure algorithmic transparency through explainable AI, and adhere to rigorous reporting standards. Embedding these validated, dynamic severity models into clinical decision support systems will be the definitive step in realizing precision metabolic healthcare.

AI Disclosure

The authors utilized the AI language model DeepSeek for assistance in translating and polishing the English text of this paper. The authors have meticulously verified all factual content, data, and scientific interpretations, and bear complete responsibility for the final work.

Data Sharing Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

Author Contributions

JM: Conceptualization, Investigation, Writing – original draft; JC: Investigation, Writing – original draft, Writing – review & editing; HZ: Validation, Writing – review & editing; XL: Conceptualization, Writing – review & editing; BZ: Conceptualization, Supervision, Writing – review & editing. All authors have given final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Funding

This work was supported by Natural Science Foundation of Wuhan under Grant No. 2025020701020284 (PI:JM). The authors declare that this funding source was not involved in the study design, data collection, analysis, preparation, or decision to submit the article for publication.

Disclosure

The authors declare that they have no competing interests.

References

1. Alberti KG, Eckel RH, Grundy SM, et al. Harmonizing the metabolic syndrome: a joint interim statement of the International Diabetes Federation Task Force on Epidemiology and Prevention; National Heart, Lung, and Blood Institute; American Heart Association; World Heart Federation; International Atherosclerosis Society; and International Association for the Study of Obesity. Circulation. 2009;120(16):1640–12. doi:10.1161/circulationaha.109.192644

2. Noubiap JJ, Nansseu JR, Lontchi-Yimagou E, et al. Geographic distribution of metabolic syndrome and its components in the general adult population: a meta-analysis of global data from 28 million individuals. Diabet Res Clin Pract. 2022:109924. doi:10.1016/j.diabres.2022.109924

3. Noubiap J, Nansseu J, Nyaga U, et al. Worldwide trends in metabolic syndrome from 2000 to 2023: a systematic review and modelling analysis. Nat Commun. 2025;17. doi:10.1038/s41467-025-67268-5

4. Maimani BA, Farhin F, Saha S, et al. The global prevalence of metabolic syndrome in adults and its association with lifestyle factors: a systematic review and meta-analysis. Cureus. 2026;18. doi:10.7759/cureus.105453

5. Van Walree E, Jansen I, Bell N, et al. Disentangling genetic risks for metabolic syndrome. Diabetes. 2022;71:2447–2457. doi:10.2337/db22-0478

6. Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Executive Summary of The Third Report of The National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, And Treatment of High Blood Cholesterol In Adults (Adult Treatment Panel III). JAMA. 2001;285(19):2486–2497. doi:10.1001/jama.285.19.2486

7. Ndumele CE, Rangaswami J, Chow SL, et al. Cardiovascular-kidney-metabolic health: a presidential advisory from the American heart association. Circulation. 2023;148(20):1606–1635. doi:10.1161/CIR.0000000000001184

8. Romeo S, Vidal-Puig A, Husain M, et al. Clinical staging to guide management of metabolic disorders and their sequelae: a European atherosclerosis society consensus statement. Eur Heart J. 2025;46(38):3685–3713. doi:10.1093/eurheartj/ehaf314

9. Wan JY, Goodman D, Makhnoon S, Norden-Krichmar TM, Wu B, Edwards KL. Heterogeneity in familial clustering of metabolic syndrome components in the multiethnic GENNID study. Obesity. 2024;32(1):176–186. doi:10.1002/oby.23914

10. Amouzegar A, Honarvar M, Masoumi S, Agahi S, Azizi F, Mehran L. Independent association of metabolic syndrome severity score and risk of diabetes: findings from 18 years of follow-up in the Tehran lipid and glucose study. BMJ Open. 2024;14(9):e078701. doi:10.1136/bmjopen-2023-078701

11. Amouzegar A, Honarvar M, Masoumi S, Khalili D, Azizi F, Mehran L. Trajectory patterns of metabolic syndrome severity score and risk of type 2 diabetes. J Transl Med. 2023;21(1):750. doi:10.1186/s12967-023-04639-w

12. Wu X, Cheung CKY, Ye D, et al. Serum thrombospondin-2 levels are closely associated with the severity of metabolic syndrome and metabolic associated fatty liver disease. J Clin Endocrinol Metab. 2022;107(8):e3230–e3240. doi:10.1210/clinem/dgac292

13. Perez-Castro E, Godinez-Jaimes F, Vazquez-Medina MU, Ocharan-Hernandez ME, Vargas-De-Leon C. Derivation and validation of sex-specific continuous metabolic syndrome scores for the Mexican adult population. Sci Rep. 2022;12(1):9659. doi:10.1038/s41598-022-10963-w

14. Fernandez-Aparicio A, Perona JS, Schmidt-RioValle J, Montero-Alonso MA, Navarro-Perez CF, Gonzalez-Jimenez E. cMetS based on Z-scores as an accurate and efficient scoring system to determine metabolic syndrome in spanish adolescents. J Pers Med. 2022;13(1):10. doi:10.3390/jpm13010010

15. Behkamal B, Asgharian Rezae F, Mansoori A, et al. A machine learning-based framework for predicting metabolic syndrome using serum liver function tests and high-sensitivity C-reactive protein. Sci Rep. 2025;15(1):21725. doi:10.1038/s41598-025-06723-1

16. Lim AMW, Lim EU, Chen PL, Fann CSJ. Unsupervised clustering identified clinically relevant metabolic syndrome endotypes in UK and Taiwan Biobanks. iScience. 2024;27(7):109815. doi:10.1016/j.isci.2024.109815

17. Zhao X, Lu C, Song B, et al. The prevalence and clustering of metabolic syndrome risk components in Chinese population: a cross-sectional study. Front Endocrinol. 2023;14:1290855. doi:10.3389/fendo.2023.1290855

18. Fernández-Verdejo R, Galgani JE. Exploring the sequential accumulation of metabolic syndrome components in adults. Sci Rep. 2022;12(1):15925. doi:10.1038/s41598-022-19510-z

19. Guembe MJ, Fernandez-Lazaro CI, Sayon-Orea C, Toledo E, Moreno-Iribas C, Investigators RS. Risk for cardiovascular disease associated with metabolic syndrome and its components: a 13-year prospective study in the RIVANA cohort. Cardiovasc Diabetol. 2020;19(1):195. doi:10.1186/s12933-020-01166-6

20. Gustavo de Sousa Barbalho Y, Morato Stival M, Ramos de lima L, et al. Impact of metabolic syndrome components in high-risk cardiovascular disease development in older adults. Clin Interv Aging. 2020;15:1691–1700. doi:10.2147/CIA.S252589

21. Huang Y, Fu R, Zhang J, et al. Dynamic changes in metabolic syndrome components and chronic kidney disease risk: a population-based prospective cohort study. BMC Endocr Disord. 2025;25(1):137. doi:10.1186/s12902-025-01958-5

22. Zhan ZQ, Chen YZ, Huang ZM, et al. Metabolic syndrome, its components, and gastrointestinal cancer risk: a meta-analysis of 31 prospective cohorts and Mendelian randomization study. J Gastroenterol Hepatol. 2024;39(4):630–641. doi:10.1111/jgh.16477

23. Obeidat AA, Ahmad MN, Ghabashi MA, et al. Developmental trends of metabolic syndrome in the past two decades: a narrative review. J Clin Med. 2025;14(7):2402. doi:10.3390/jcm14072402

24. Molina-Luque R, Ulloa N, Gleisner A, Zilic M, Romero-Saldana M, Molina-Recio G. An approach to early detection of metabolic syndrome through non-invasive methods in obese children. Children. 2020;7(12):304. doi:10.3390/children7120304

25. Huh JH, Lee JH, Moon JS, Sung KC, Kim JY, Kang DR. Metabolic syndrome severity score in Korean adults: analysis of the 2010-2015 Korea national health and nutrition examination survey. J Korean Med Sci. 2019;34(6):e48. doi:10.3346/jkms.2019.34.e48

26. Medina Inojosa BJ, Somers VK, Lara-Breitinger K, Johnson LA, Medina-Inojosa JR, Lopez-Jimenez F. Prediction of presence and severity of metabolic syndrome using regional body volumes measured by a multisensor white-light 3D scanner and validation using a mobile technology. Eur Heart J Digit Health. 2024;5(5):582–590. doi:10.1093/ehjdh/ztae059

27. Yang S, Yu B, Yu W, et al. Development and validation of an age-sex-ethnicity-specific metabolic syndrome score in the Chinese adults. Nat Commun. 2023;14(1):6988. doi:10.1038/s41467-023-42423-y

28. Dai H, Fu Q, Xu K, et al. Constructing a metabolic integral score model for the quantification of metabolic dysfunction and tendency. Nutr Metab Cardiovasc Dis. 2022;32(3):658–665. doi:10.1016/j.numecd.2021.12.014

29. Honarvar M, Masoumi S, Mehran L, Khalili D, Amouzegar A, Azizi F. Development and validation of a continuous metabolic syndrome severity score in the Tehran Lipid and Glucose Study. Sci Rep. 2023;13(1):7529. doi:10.1038/s41598-023-33294-w

30. Kotarsky CJ, Lang JM, Shostak ES, et al. The simplified method for quantifying metabolic syndrome (siMS) score reflects an increased cardiometabolic burden: adiposity, hemodynamics, and HRV findings in young adults. J Diabetes Metab Disord. 2026;25(1):66. doi:10.1007/s40200-026-01870-2

31. Greenacre M, Groenen PJF, Hastie T, D’Enza AI, Markos A, Tuzhilina E. Principal component analysis. Nat Rev Method Primers. 2022;2(1):100. doi:10.1038/s43586-022-00184-w

32. Yi M, Wang X, Li Y, et al. Association between metabolic syndrome score and subclinical atherosclerosis. Rev Cardiovasc Med. 2025;26(3):26811. doi:10.31083/RCM26811

33. Tang Z, Zhen Y, Zhang L, Liu X, Ma J. Prevalence and factors associated with metabolic syndrome in first hospitalization for major depression disorder patients. Sci Rep. 2023;13(1):15496. doi:10.1038/s41598-023-42720-y

34. Hsu PW, Chen YR, Sheu WH. MetS-Z: a gender- and age-specific scoring system for predicting type 2 diabetes. J Diabetes Investig. 2025;16(5):907–916. doi:10.1111/jdi.70004

35. Lin -J-J, Dai P-Y, Zhang J, et al. Association between metabolic syndrome severity score and cardiovascular disease: results from a longitudinal cohort study on Chinese adults. Front Endocrinol. 2024;15:1341546. doi:10.3389/fendo.2024.1341546

36. Ghosh A, Ghosh AK, SahaRay R, Sarkar S. Classification using global and local Mahalanobis distances. J Multivar Anal. 2025;207:105417. doi:10.1016/j.jmva.2025.105417

37. Shin H, Shim S, Oh S. Robust metabolic syndrome risk score based on triangular areal similarity. PeerJ Comput Sci. 2024;10:e2015. doi:10.7717/peerj-cs.2015

38. Dimitrijevic-Sreckovic V, Petrovic H, Dobrosavljevic D, et al. siMS score- method for quantification of metabolic syndrome, confirms co-founding factors of metabolic syndrome. Front Genet. 2022;13:1041383. doi:10.3389/fgene.2022.1041383

39. Park JH, Jeong I, Ko GJ, Jeong S, Lee H. Development of a predictive model for metabolic syndrome using noninvasive data and its cardiovascular disease risk assessments: multicohort validation study. J Med Internet Res. 2025;27:e67525. doi:10.2196/67525

40. Lee M, Park T, Shin JY, Park M. A comprehensive multi-task deep learning approach for predicting metabolic syndrome with genetic, nutritional, and clinical data. Sci Rep. 2024;14(1):17851. doi:10.1038/s41598-024-68541-1

41. Akbarzadeh M, Alipour N, Moheimani H, et al. Evaluating machine learning-powered classification algorithms which utilize variants in the GCKR gene to predict metabolic syndrome: tehran cardio-metabolic genetics study. J Transl Med. 2022;20(1):164. doi:10.1186/s12967-022-03349-z

42. Li S, Zhang Y, Fu K, et al. Machine learning ensemble meets clinical practice: developing a real-world risk prediction model for metabolic syndrome using super learner and scorecard approaches. J Adv Res. 2026;82:701–711. doi:10.1016/j.jare.2025.06.072

43. Tsai ML, Chen KF, Chen PC. Harnessing electronic health records and artificial intelligence for enhanced cardiovascular risk prediction: a comprehensive review. J Am Heart Assoc. 2025;14(6):e036946. doi:10.1161/JAHA.124.036946

44. Li Z, Wu W, Kang H. Machine learning-driven metabolic syndrome prediction: an international cohort validation study. Healthcare. 2024;12(24):2527. doi:10.3390/healthcare12242527

45. Ding K, Zhou Z, Ma Y, et al. Identification of novel metabolic subtypes using multi-trait limited mixed regression in the chinese population. Biomedicines. 2022;10(12):3093. doi:10.3390/biomedicines10123093

46. Gonzalez-Martos R, Galeano J, Ramirez-Castillejo C, et al. Unsupervised clustering of biochemical markers reveals health profiles associated with function and survival in active aging. Sci Rep. 2025;15(1):30546. doi:10.1038/s41598-025-14580-1

47. Taurbekova B, Sarsenov R, Yaqoob MM, et al. Cluster analysis in diabetes research: a systematic review enhanced by a cross-sectional study. J Clin Med. 2025;14(10):3588. doi:10.3390/jcm14103588

48. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. Br Med J. 2024;385:e078378. doi:10.1136/bmj-2023-078378

49. Moons KGM, Damen JAA, Kaul T, et al. PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods. BMJ. 2025;388:e082505. doi:10.1136/bmj-2024-082505

50. Mizani M, Dashtban A, Pasea L, et al. Identifying subtypes of type 2 diabetes mellitus with machine learning: development, internal validation, prognostic validation and medication burden in linked electronic health records in 420 448 individuals. BMJ Open Diabetes Res Care. 2024;12:e004191. doi:10.1136/bmjdrc-2024-004191

51. Asplund O, Thangam M, Prasad RB, et al. Comorbidities and mortality in subgroups of adults with diabetes with up to 14 years follow-up: a prospective cohort study in Sweden. Lancet Diabetes Endocrinol. 2026;14(1):29–40. doi:10.1016/s2213-8587(25)00283-9

52. Hu C, Jia W. Multi-omics profiling: the way towards precision medicine in metabolic diseases. J Mol Cell Biol. 2021;13(8):576–593. doi:10.1093/jmcb/mjab051

53. Leiby JS, Lee ME, Shivakumar M, Choe EK, Kim D. Deep learning imaging phenotype can classify metabolic syndrome and is predictive of cardiometabolic disorders. J Transl Med. 2024;22(1):434. doi:10.1186/s12967-024-05163-1

54. Sah AK, Afzal M, Elshaikh RH, et al. Innovative strategies in the diagnosis and treatment of liver cirrhosis and associated syndromes. Life. 2025;15(5):779. doi:10.3390/life15050779

55. Chakaroun RM, Pradhan M, Björnson E, et al. Multi-omic definition of metabolic obesity through adipose tissue–microbiome interactions. Nature Med. 2026;32(1):113–125. doi:10.1038/s41591-025-04009-7

56. Chen D, Zhao X, Sui Z, et al. A multi-omics investigation of the molecular characteristics and classification of six metabolic syndrome relevant diseases. Theranostics. 2020;10(5):2029–2046. doi:10.7150/thno.41106

57. Ghatan S, van Rooij J, van Hoek M, et al. Defining type 2 diabetes polygenic risk scores through colocalization and network-based clustering of metabolic trait genetic associations. Genome Med. 2024;16(1):10. doi:10.1186/s13073-023-01255-7

58. Suzuki K, Hatzikotoulas K, Southam L, et al. Genetic drivers of heterogeneity in type 2 diabetes pathophysiology. Nature. 2024;627(8003):347–357. doi:10.1038/s41586-024-07019-6

59. Zhou M, Sun W, Gao Y, et al. Metabolomic profiling reveals interindividual metabolic variability and its association with cardiovascular-kidney-metabolic syndrome risk. Cardiovasc Diabetol. 2025;24(1):315. doi:10.1186/s12933-025-02881-8

60. Vidanagamachchi SM, Waidyarathna K. Opportunities, challenges and future perspectives of using bioinformatics and artificial intelligence techniques on tropical disease identification using omics data. Front Digit Health. 2024;6:1471200. doi:10.3389/fdgth.2024.1471200

61. Ouhmouk M, Baichoo S, Abik M. Challenges in AI-driven multi-omics data analysis for Oncology: addressing dimensionality, sparsity, transparency and ethical considerations. Inf Med Unlocked. 2025;57:101679. doi:10.1016/j.imu.2025.101679

62. Bergomi L, Nicora G, Orlowska MA, et al. Which explanations do clinicians prefer? A comparative evaluation of XAI understandability and actionability in predicting the need for hospitalization. BMC Med Inform Decis Mak. 2025;25(1):269. doi:10.1186/s12911-025-03045-0

63. Hou J, Wang LL. Explainable AI for clinical outcome prediction: a survey of clinician perceptions and preferences. AMIA Jt Summits Transl Sci Proc. 2025;2025:215–224.

64. Joshi S, Shamanna P, Dharmalingam M, et al. Digital twin-enabled personalized nutrition improves metabolic dysfunction-associated fatty liver disease in type 2 diabetes: results of a 1-year randomized controlled study. Endocr Pract. 2023;29(12):960–970. doi:10.1016/j.eprac.2023.08.016

65. Reed J, Dong T, Eaton E, et al. Continuous glucose monitoring for glycaemic control and cardiovascular risk reduction in patients with type 2 diabetes not on insulin therapy: a clinical trial. Diabetes Obes Metab. 2024;26(7):2881–2889. doi:10.1111/dom.15608

66. Domenico MD, Allegri L, Caldarelli G, et al. Challenges and opportunities for digital twins in precision medicine from a complex systems perspective. Npj Digital Med. 2025;8(1). doi:10.1038/s41746-024-01402-3

67. Zou H, Zheng X, Wu L, Zhang S, Chang P, Chen Y. Digital twins in cardiovascular disease: a scoping review. Int J Med Inform. 2025;206:106138. doi:10.1016/j.ijmedinf.2025.106138

68. Islam MS, Wei P, Suzauddula M, et al. The interplay of factors in metabolic syndrome: understanding its roots and complexity. Mol Med. 2024;30(1):279. doi:10.1186/s10020-024-01019-y

69. Higuera-Gómez A, de la OV, San-Cristobal R, et al. Computational algorithm based on health and lifestyle traits to categorize lifemetabotypes in the NUTRiMDEA cohort. Sci Rep. 2024;14(1):24835. doi:10.1038/s41598-024-75110-z

70. Pigeot I, Ahrens W. Epidemiology of metabolic syndrome. Pflugers Arch. 2025;477(5):669–680. doi:10.1007/s00424-024-03051-7

71. Park J, Peña-Tauber A, Talozzi L, Greicius MD, Le Guen Y. Rare genetic associations with human lifespan in UK Biobank are enriched for oncogenic genes. Nat Commun. 2025;16(1):2064. doi:10.1038/s41467-025-57315-6

72. Denny JC, Rutter JL, Goldstein DB, et al. The “All of Us” Research Program. N Engl J Med. 2019;381(7):668–676. doi:10.1056/NEJMsr1809937

73. Sheller MJ, Edwards B, Reina GA, et al. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data. Sci Rep. 2020;10(1):12598. doi:10.1038/s41598-020-69250-1

74. Karakasis P, Vlachos K, Antoniadis AP, et al. Effect of GLP-1 receptor agonists and co-agonists on atrial fibrillation risk in overweight or obesity: a meta-analysis of randomized controlled trials. Metabolism. 2025;175:156463. doi:10.1016/j.metabol.2025.156463