Atherogenic, cardiometabolic, and adiposity composite indices for identifying prehypertension and hypertension: conventional regression and explainable machine learning.
Composite indices integrating visceral adiposity, insulin resistance, and lipid burden—particularly CVAI, TyG-WC, TyG-BMI, and METS-IR—were associated with prehypertension and hypertension and may help characterize cardiometabolic risk clustering, though they are not substitutes for direct blood pressure measurement or diagnostic evaluation.
Key Findings
Results
CVAI showed the strongest standardized association with abnormal blood pressure among all composite indices examined.
In fully adjusted analyses, CVAI had an OR of 2.17 (95% CI: 1.80–2.62; P<0.01) for abnormal blood pressure.
CVAI is a Chinese visceral adiposity index integrating anthropometric and biochemical variables.
The association remained significant after multivariable adjustment including anthropometric, biochemical, and questionnaire variables.
The cohort included 983 participants with normal blood pressure (46.02%), 422 with prehypertension (19.76%), and 731 with hypertension (34.22%).
Results
TyG-WC, TyG-BMI, and METS-IR also showed strong standardized associations with abnormal blood pressure.
TyG-WC had an OR of 1.94 (95% CI: 1.64–2.28; P<0.01).
TyG-BMI had an OR of 1.93 (95% CI: 1.66–2.26; P<0.01).
METS-IR had an OR of 1.91 (95% CI: 1.63–2.24; P<0.01).
All four indices (CVAI, TyG-WC, TyG-BMI, METS-IR) represent different combinations of visceral adiposity, insulin resistance, and lipid burden.
Results
Higher quartiles of CVAI, METS-IR, and TyG-WHtR were associated with higher odds of abnormal blood pressure in a dose-response fashion.
The highest quartiles of CVAI, METS-IR, and TyG-WHtR all had higher odds than the lowest quartiles.
P for trend was <0.01 for all three indices.
Quartile analyses were part of a broader analytic strategy including multivariable regression, restricted cubic splines, mixture analysis, and incremental evaluation.
Results
Adding composite index and clinical information to a base model improved cross-validated discriminative performance from AUROC 0.80 to 0.83.
Incremental evaluation showed AUROC improved from 0.80 to 0.83 after adding clinical and index information.
This improvement was assessed using cross-validated AUROC.
The incremental gain suggests composite indices contribute modestly but measurably to identifying abnormal blood pressure beyond baseline variables.
Results
Explainable machine learning using SHAP values identified CVAI, age, fasting glucose, waist circumference, eGFR, and uric acid as top contributors to model predictions.
SHAP (SHapley Additive exPlanations) analysis was used to interpret the machine learning model.
CVAI was the top-ranked feature among composite indices in the SHAP analysis.
Other highly ranked variables included age, fasting glucose, waist circumference, eGFR, and uric acid.
The study used explainable machine learning alongside conventional regression to provide complementary analytical perspectives.
Methods
The study population consisted of 2,136 adults from a single hospital-based health examination center in China.
Participants were recruited from the Health Examination Center of the Second Affiliated Hospital of Shandong First Medical University between 2024 and 2025.
This was a cross-sectional study design.
Blood pressure was classified into three categories: normal, prehypertension, and hypertension.
Analyses included subgroup analyses and sensitivity analyses to assess robustness of findings.
Conclusions
The authors explicitly noted that composite indices are not substitutes for direct blood pressure measurement and that findings support association and screening-value inference, not causal or diagnostic claims.
The abstract states: 'They may help characterize cardiometabolic risk clustering but are not substitutes for direct blood pressure measurement or diagnostic evaluation.'
The authors further specified: 'These findings support association and screening-value inference, not causal or diagnostic claims.'
This distinction was made in the context of a cross-sectional study design, which limits causal inference.
What This Means
This research suggests that certain composite scores calculated from routine blood tests and body measurements—particularly a Chinese visceral adiposity index (CVAI), and indices combining triglycerides with body weight, BMI, or waist-to-height ratio—are meaningfully associated with having elevated blood pressure (prehypertension or hypertension) in Chinese adults undergoing routine health checkups. Among 2,136 adults examined at a hospital health center, about half had normal blood pressure, roughly one in five had prehypertension, and about one in three had hypertension. People with higher values of these composite scores were substantially more likely to fall into the prehypertension or hypertension categories, with the strongest association seen for CVAI, which combines measures of abdominal fat and blood lipids.
The researchers used both traditional statistical methods and machine learning to analyze the data. When machine learning was used to identify which factors most strongly predicted blood pressure status, the CVAI score ranked highest among the composite indices, alongside age, fasting blood sugar, waist circumference, kidney function, and uric acid levels. Adding these composite indices to a predictive model modestly improved its ability to distinguish people with abnormal blood pressure from those with normal blood pressure (accuracy measure improved from 0.80 to 0.83 on a 0–1 scale).
This research suggests that these easy-to-calculate composite scores—derived from information already collected during routine health checkups—could help identify people at higher cardiometabolic risk. However, the authors emphasize that these scores cannot replace actually measuring blood pressure, and because this was a single-center cross-sectional study conducted in China, the findings reflect associations rather than causes. Further research in diverse populations would be needed to determine how broadly applicable these findings are.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
Sun H, Liu Y. (2026). Atherogenic, cardiometabolic, and adiposity composite indices for identifying prehypertension and hypertension: conventional regression and explainable machine learning.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1924726