The new system, called MetsObesity, was developed and validated by Junaid Iqbal, Hou-De Zhou, and colleagues using data from the UK Biobank, a prospective cohort of roughly half a million adults in the United Kingdom. After applying careful exclusion criteria designed to remove conditions that distort body weight or confound cardiovascular outcomes, including pregnancy, cancer, thyroid disease, dementia, and pre-existing cardiovascular disease, the study settled on 346,001 participants with a median age of 57 years at baseline. These individuals were followed for a median of 15 years, during which the researchers tracked the first occurrence of a composite endpoint known as major adverse cardiovascular events, or MACE, comprising acute myocardial infarction, heart failure, stroke, unstable angina, and death from cardio-cerebrovascular causes.

What sets MetsObesity apart from existing frameworks is the breadth of information it folds into a single classification. Rather than relying on body mass index or waist circumference alone, the researchers considered three categories of variables. The first was anthropometric: body mass index, waist circumference, waist-to-hip ratio, and waist-to-height ratio. The second comprised the classic metabolic abnormalities that define metabolic syndrome, including systolic and diastolic blood pressure, fasting blood glucose, high-density lipoprotein cholesterol, triglycerides, along with low-density lipoprotein cholesterol and total cholesterol. The third category, often neglected in conventional risk assessment, captured additional abnormalities: C-reactive protein, a marker of systemic inflammation; serum urate; and the liver enzymes alanine aminotransferase, aspartate aminotransferase, and gamma-glutamyl transferase, which reflect fatty liver disease and hepatic dysfunction.

To decide which of these variables actually mattered, the team turned to machine learning. Recursive feature elimination was used to identify the single best anthropometric predictor, and waist-to-hip ratio emerged at the top of the ranking, outperforming body mass index, waist circumference, and waist-to-height ratio. Random forest models, an ensemble learning method well suited to large, high-dimensional datasets, generated mean decreased accuracy scores to rank the influence of every candidate variable on cardiovascular outcomes. Variables with negative importance scores, including alkaline phosphatase, cystatin C, and serum creatinine, were discarded. Fasting blood glucose beat glycated hemoglobin as the preferred glycemic marker. The final feature set contained twelve variables: waist-to-hip ratio, low-density lipoprotein cholesterol, total cholesterol, systolic blood pressure, diastolic blood pressure, urate, high-density lipoprotein cholesterol, alanine aminotransferase, C-reactive protein, aspartate aminotransferase, gamma-glutamyl transferase, and fasting glucose. SHAP analysis, which quantifies each feature’s contribution to model predictions, confirmed the same hierarchy.

From these building blocks, the researchers constructed eight preliminary classes representing every meaningful combination of anthropometric, classic metabolic, and additional abnormalities, then consolidated them into five final MetsObesity classes. Class 1 describes people with a normal waist-to-hip ratio and no abnormalities at all, the metabolically healthy reference group. Class 2 covers those with normal waist-to-hip ratio but one or more classic metabolic abnormalities such as hypertension or hyperglycemia. Class 3 includes people with normal waist-to-hip ratio who carry one or more additional abnormalities, such as elevated C-reactive protein or liver enzymes, with or without classic metabolic problems. Class 4 captures a high waist-to-hip ratio without additional metabolic abnormalities, while Class 5, the highest-risk tier, combines a high waist-to-hip ratio with additional abnormalities, with or without classic ones. The design deliberately replaces the binary language of metabolically healthy versus unhealthy obesity with a graded spectrum.

The results were striking. In the validation cohort, the proportion of participants who developed major cardiovascular events rose steadily across the classes: 1.4 percent in Class 1, 3.3 percent in Class 2, 5.0 percent in Class 3, 6.0 percent in Class 4, and 8.5 percent in Class 5. After adjusting for medications, age, sex, smoking, alcohol consumption, and socioeconomic deprivation, the hazard ratios for cardiovascular events climbed from Class 1 through Class 5. In the two highest classes, the hazard ratios for major adverse cardiovascular events reached 4.75 and 6.79, several times higher than the hazard ratios produced by body mass index categories, waist circumference, body fat percentage, or the metabolic syndrome definitions recommended by the National Cholesterol Education Program Adult Treatment Panel and the International Diabetes Federation, which yielded hazard ratios between roughly 1.1 and 2.0.

Performance testing reinforced the picture. MetsObesity achieved a Harrell C-statistic of 0.62, modest in absolute terms but clearly superior to body mass index at 0.57, waist circumference and body fat percentage at 0.55, trunk fat percentage at 0.51, and the IDF and NCEP metabolic syndrome definitions at 0.57 and 0.56. Calibration plots showed good agreement between predicted and observed event probabilities, verified with bootstrap-corrected estimates. Decision curve analysis, which quantifies the clinical net benefit of a prediction model across a range of risk thresholds, demonstrated that MetsObesity delivered superior value across the clinically relevant 2 to 8 percent threshold range, peaking at a standardized net benefit of 0.638 at the 2 percent threshold. Net reclassification improvement analysis showed that MetsObesity meaningfully reclassified risk beyond standard markers, with an NRI of 0.204 at five years and 0.160 at ten years.

Sensitivity analyses tested the system’s robustness from multiple angles. The findings held when missing data were excluded rather than imputed, when the cohort was randomly split in different proportions, and when participants with weight-affecting comorbidities were retained. A subgroup analysis of Asian participants, using population-specific body mass index and body fat thresholds, showed the same pattern of superior prediction compared with conventional measures. Notably, when compared with established cardiovascular risk calculators such as SCORE2 and the Pooled Cohort Equations, MetsObesity showed somewhat lower discrimination, but the authors point out a crucial distinction: those models lean heavily on age and sex, which are non-modifiable, whereas MetsObesity is built exclusively on changeable metabolic and anthropometric factors, making it potentially more actionable for prevention, particularly in younger people in whom age-driven models perform poorly.

The clinical appeal of MetsObesity lies partly in its practicality. Every variable it requires, from waist-to-hip ratio to C-reactive protein and liver enzymes, is already measured in routine health examinations and standard metabolic panels in many healthcare settings. No advanced imaging, genetic testing, or complex continuous risk functions are needed; the categorical thresholds align with established clinical cut-points, allowing bedside application and straightforward communication of risk to patients. The authors suggest that embedding the system into electronic medical records or mobile risk calculators could automate the calculation and ease the workload in busy clinics, and they emphasize that MetsObesity is intended to complement rather than replace established scores like SCORE2 and the Pooled Cohort Equations, especially in overweight and obese populations where traditional tools perform least well.

The study is not without limitations. It was conducted in a predominantly European-ancestry UK population, and external validation in independent, ethnically diverse cohorts remains an essential next step, particularly given that adiposity patterns and anthropometric cut-offs differ across South Asian, East Asian, and Middle Eastern populations. Data on physical activity and diet were unavailable, leaving residual confounding possible, and the system currently addresses cardiovascular outcomes only, not the kidney, cancer, reproductive, and perinatal consequences of obesity. Still, as obesity rates climb worldwide and cardiovascular disease follows in its wake, MetsObesity offers what its authors describe as a paradigm shift from one-size-fits-all assessment to phenotype-specific prevention, a five-step ladder that identifies high-risk individuals years before their first heart attack, stroke, or diagnosis of heart failure, using nothing more exotic than a tape measure and a standard blood panel.

Subject of Research: A novel obesity and metabolic classification system for predicting long-term cardiovascular risk

Article Title: MetsObesity: a novel classification system for predicting 15-year cardiovascular risk in the UK Biobank population

Article References: Iqbal, J., Wu, H.-X., Wu, Y.-X., Jiang, H.-L., Li, L., Zhou, X.-Y., Bu, Y.-H., & Zhou, H.-D. (2026). MetsObesity: a novel classification system for predicting 15-year cardiovascular risk in the UK Biobank population. Journal of Advanced Research, 88, 947-957. https://doi.org/10.1016/j.jare.2026.01.050

Image Credits: AI Generated

DOI: 10.1016/j.jare.2026.01.050

Keywords: obesity, cardiovascular disease, UK Biobank, MetsObesity, risk prediction, metabolic syndrome, waist-to-hip ratio, C-reactive protein, liver enzymes, BMI, precision medicine, cohort study