Head-to-head comparison of fasting-insulin-based versus non-insulin-based insulin resistance surrogates for predicting all-cause and cardiovascular mortality in hypertensive adults across the glycemic continuum: a prospective cohort study

Abstract

Background

Insulin resistance (IR) is mechanistically linked to hypertension, yet no study has directly compared fasting-insulin-based and non-insulin-based IR surrogates for predicting mortality across the glycemic spectrum in hypertensive adults. We evaluated ten IR indices, three insulin-based [homeostasis model assessment of insulin resistance (HOMA-IR), McAuley index, and quantitative insulin sensitivity check index (QUICKI)] and seven non-insulin-based [triglyceride-glucose index (TyG), stress hyperglycemia ratio (SHR), cardiometabolic index (CMI), atherogenic index of plasma (AIP), estimated glucose disposal rate (eGDR), metabolic score for insulin resistance (METS-IR), and lipid accumulation product (LAP)], for all-cause mortality (ACM) and cardiovascular mortality (CVM) by glycemic status.

Methods

This prospective cohort study included 7,548 hypertensive adults from NHANES 1999–2018, classified as normoglycemia (n = 1,869), prediabetes (n = 3,389), and diabetes (n = 2,290). Mortality data were collected through December 31, 2019. Associations were analyzed using Cox models with three levels of adjustment. Dose-response relationships were modeled with restricted cubic splines.

Results

Over a mean follow-up of 8.7 ± 5.3 years, 1,752 ACM events (23.2%) and 499 CVM events (6.6%) occurred. In fully adjusted models, HOMA-IR independently predicted ACM (per-unit HR 1.017; Q4 HR 1.165) and CVM (per-unit HR 1.012). eGDR showed the strongest overall associations: ACM: per-unit HR 0.807; Q4 HR 0.559; CVM: per-unit HR 0.774; Q4 HR 0.538. TyG predicted ACM (per-unit HR 1.158; Q4 HR 1.233) and CVM per-unit (HR 1.160). In glycemic-stratified analyses, HOMA-IR was the only index with per-unit ACM significance across all three strata (normoglycemia: HR 1.085; prediabetes: HR 1.039; diabetes: HR 1.012). eGDR showed per-unit and Q4 ACM significance in prediabetes and diabetes, and demonstrated significant quartile-level protection for CVM in both prediabetes (Q4 HR 0.465) and diabetes (Q4 HR 0.463), whereas per-unit associations with CVM were observed only in diabetes. TyG was significantly associated with ACM in diabetes (per-unit HR 1.147; Q4 HR 1.337) and with CVM in diabetes (Q4 HR 1.595).

Conclusions

In hypertensive adults, HOMA-IR was the only fasting-insulin-based index independently associated with ACM across all three glycemic strata. eGDR demonstrated the most consistent non-insulin-based associations with ACM and CVM, particularly in prediabetes and diabetes. TyG provided additional prognostic value for ACM and CVM, particularly in patients with diabetes. These findings support selecting IR indices based on glycemic phenotype and data availability.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03244-7.

Research insights

What is currently known about this topic?

• Insulin resistance (IR) is a fundamental factor in hypertension-related cardiovascular morbidity and mortality. Various IR surrogates have been associated with mortality in hypertensive or cardiometabolic populations through separate studies. However, no study has directly compared fasting-insulin-based and non-insulin-based IR surrogates across the entire glycemic spectrum in a prospective mortality study focused on hypertension.

What is the key research question?

• Among ten contemporary IR surrogate indices, which one provides the most reliable and discriminating prediction of all-cause mortality (ACM) and cardiovascular mortality (CVM) in hypertensive adults? Additionally, does the comparative performance of fasting-insulin-based versus non-insulin-based indices vary systematically across normoglycemia, prediabetes, and diabetes?

What is new?

• This is the first study to simultaneously compare 10 IR indices, explicitly grouped as fasting-insulin-based or non-insulin-based, for mortality prediction across the glycemic spectrum in hypertensive adults. HOMA-IR was the only index with independent per-unit associations with ACM across all three glycemic strata. eGDR was the only non-insulin-based index with per-unit ACM significance in prediabetes and diabetes, and the only index with per-unit CVM significance in the diabetic subgroup, with additional CVM protection observed in prediabetes.

How might this study influence clinical practice?

• eGDR, derived from waist circumference, hemoglobin A1c (HbA1c), and hypertension status, appears to be the most informative non-insulin-based surrogate, particularly among individuals with prediabetes or diabetes. HOMA-IR remains the preferred insulin-based index when fasting insulin is available, given its consistent per-unit associations with ACM across all three glycemic strata. TyG provides additional prognostic value for both ACM and CVM risk stratification, specifically in individuals with diabetes. These findings support a precision-based approach to selecting IR indices for hypertensive cardiometabolic risk assessment, guided by glycemic phenotype and resource availability.

Introduction

Hypertension is the leading modifiable risk factor for cardiovascular disease (CVD) and premature death worldwide, affecting over 1.28 billion adults [1]. Hypertension and insulin resistance (IR) are mechanistically interconnected through bidirectional pathophysiological pathways, in which IR elevates blood pressure via sympathetic overactivation, impaired nitric oxide-mediated vasodilation, renal sodium retention, and vascular smooth muscle remodeling, while sustained hypertension, in turn, impairs peripheral glucose disposal [2, 3]. Beyond its role in hypertension development, IR has been linked to all-cause mortality (ACM) and cardiovascular mortality (CVM) through endothelial dysfunction, atherogenic dyslipidemia, and chronic low-grade inflammation, pathways that may accelerate cardiovascular organ damage across the glycemic spectrum [4–6].

Since the gold-standard hyperinsulinemic-euglycemic clamp is impractical for large-scale studies, many surrogate indices have been developed. These fall into two categories based on whether fasting insulin is required: insulin-based indices such as the homeostasis model assessment of insulin resistance (HOMA-IR) [7], McAuley index [8], and the quantitative insulin sensitivity check index (QUICKI) [9], and non-insulin-based indices like triglyceride-glucose index (TyG) [10], atherogenic index of plasma (AIP) [11], estimated glucose disposal rate (eGDR) [12], metabolic score for insulin resistance (METS-IR) [13], lipid accumulation product (LAP) [14], cardiometabolic index (CMI) [15], and stress hyperglycemia ratio (SHR) [16]. This distinction has direct clinical relevance because insulin assays are costly, vary across laboratories, and are often unavailable in epidemiological datasets. Individual IR surrogates have been linked to hypertensive mortality in separate studies. For instance, HOMA-IR showed a U-shaped association with ACM in a coronary heart disease (CHD)-hypertension cohort [4] and predicted cardiovascular events in hypertensive non-diabetic adults [17]. TyG and its adiposity-adjusted variants showed significant non-linear associations with ACM and CVM, as confirmed by machine-learning analyses [18], with the highest joint mortality risk when combined with hypertension [19]. eGDR correlated most strongly with myocardial glucose metabolism [20] and predicted outcomes across cardiometabolic syndrome, chronic kidney disease (CKD), and dyslipidemia cohorts [21–23]. METS-IR predicted the incidence of CVD in hypertensive sleep apnea [24] and incident hypertension across populations [25–27]. AIP was associated with ACM and CVM among hypertension patients [28]. LAP demonstrated age-adjusted predictive performance for hypertension risk [29] and positive associations with arterial stiffness in Chinese hypertensive adults [30]. SHR, a dynamic glycemic marker, independently predicted incident hypertension via lipid-mediated pathways [31], adverse outcomes in pulmonary hypertension [32], and stroke risk, with synergistic interactions with hypertension [33]. Despite this extensive body of evidence, no study has simultaneously compared IR indices within a single hypertensive cohort or systematically evaluated how glycemic phenotype influences their relative prognostic performance.

Notably, IR pathophysiology changes across the glycemic spectrum, from compensatory hyperinsulinemia that maintains euglycemia in normoglycemia, through lipotoxicity-driven peripheral IR with impaired glucose tolerance in prediabetes, to converging glucolipotoxicity with progressive beta-cell failure in overt diabetes [34, 35]. This suggests that insulin-based and non-insulin-based surrogates may exhibit different prognostic performance across glycemic stages. This pathophysiological heterogeneity highlights the need for glycemic stage-specific index evaluation, yet no study has systematically addressed this important gap.

Therefore, this study performed a direct comparison of ten IR surrogates for predicting ACM and CVM in 7,548 hypertensive adults from the National Health and Nutrition Examination Survey (NHANES) 1999–2018, with stratification by glycemic status (normoglycemia, prediabetes, diabetes) and classification of indices as fasting-insulin-based or non-insulin-based, to provide evidence-based guidance for index selection in clinical and epidemiological practice.

Materials and methods

Study design and data source

This prospective cohort study used publicly available NHANES data (https://www.cdc.gov/nchs/nhanes/), a continuous cross-sectional survey of the non-institutionalized US civilian population that employs a stratified, multistage probability cluster sampling design and is conducted by the National Center for Health Statistics (NCHS). Data from ten consecutive two-year cycles (1999–2000 through 2017–2018) were combined. Mortality was determined through probabilistic linkage to the National Death Index (NDI) Plus up to December 31, 2019. All NHANES protocols received approval from the NCHS Research Ethics Review Board, and written informed consent was obtained from all participants. Since this analysis used publicly available, de-identified data, it did not require additional ethics review.

Study population

The initial NHANES 1999–2018 population included 101,316 participants. Exclusions were applied in four structured groups. Target population restrictions were applied first: participants younger than 18 years were excluded (n = 42,112), leaving 59,204 adults. Hypertension was defined as either a self-reported physician diagnosis or a measured mean systolic blood pressure (SBP) of 140 mmHg or higher, or a diastolic blood pressure (DBP) of 90 mmHg or higher, consistent with widely used epidemiological definitions [36]; participants without hypertension were then excluded (n = 34,231), resulting in 24,973 hypertensive adults. Group 2 (missing outcome data): participants with missing mortality or follow-up data were excluded (n = 38). Group 3 (missing key exposure variables, IR index components): participants were then excluded for missing anthropometric data required for IR index calculation, including body mass index (BMI), waist circumference, height, and weight (n = 3,032); missing lipid panel data required for TyG, AIP, LAP, CMI, and METS-IR, including triglycerides, high-density lipoprotein cholesterol (HDL-C), and total cholesterol (n = 1,245); missing hemoglobin A1c (HbA1c) required for glycemic classification and for the calculation of eGDR and SHR (n = 27); missing fasting blood glucose required for glycemic classification and for the calculation of TyG, METS-IR, and SHR (n = 10,490); and missing fasting insulin required for HOMA-IR, the McAuley index, and QUICKI (n = 81). The total excluded for missing IR index components was n = 14,875. Group 4 (missing covariate data): participants were further excluded for missing comorbidity or medical-history data, including CKD, coronary artery disease (CAD), angina, myocardial infarction, heart failure, stroke, cancer, diabetes history, and smoking history (n = 328); missing blood pressure readings required for baseline blood pressure completeness and covariate adjustment (n = 991); missing alcohol consumption history (n = 483); and missing poverty-income ratio (n = 710). The total excluded for missing covariates was n = 2,512. The final analytic cohort consisted of 7,548 adults with hypertension. According to the prespecified glycemic classification (criteria defined in the following subsection), 1,869 participants (24.8%) were classified as normoglycemic, 3,389 (44.9%) as having prediabetes, and 2,290 (30.3%) as having diabetes (Fig. 1).

Fig. 1.

Participant selection flowchart

Glycemic status classification

Participants were categorized into three mutually exclusive glycemic groups based on the 2024 American Diabetes Association (ADA) criteria [37]. Normoglycemia is defined as HbA1c < 5.7% and fasting glucose < 100 mg/dL. Prediabetes is defined as HbA1c 5.7%-6.4%, or fasting glucose 100–125 mg/dL, in the absence of a diabetes diagnosis. Diabetes is defined as HbA1c ≥ 6.5%, or fasting glucose ≥ 126 mg/dL, or self-reported physician-diagnosed diabetes, or current use of insulin or oral glucose-lowering medications. Both HbA1c and fasting plasma glucose were incorporated to ensure complete ascertainment of dysglycemia across NHANES cycles and to capture individuals with discordant glucose-HbA1c values.

Insulin resistance index calculation

Three indices require fasting insulin for their calculation. HOMA-IR is calculated as (fasting insulin [µU/mL] × fasting glucose [mg/dL]) / 405 [7]. The McAuley index = exp (2.63 − 0.28 × ln [insulin µU/mL] − 0.31 × ln [triglycerides mmol/L]) [8]. QUICKI = 1 / (log10[glucose mg/dL] + log10[insulin µU/mL]) [9].

The remaining seven indices are derived solely from routinely available biochemical and anthropometric parameters, without requiring insulin measurement. TyG = ln (triglycerides [mg/dL] × glucose [mg/dL] / 2) [10]. AIP = log10(triglycerides [mmol/L] / HDL-C [mmol/L]) [11]. eGDR (mg/kg/min) = 21.158 − 0.09 × waist circumference [cm] − 3.407 × hypertension status − 0.551 × HbA1c [%] [12]; hypertension status was set to 1 for all cohort members; higher values indicate greater sensitivity. METS-IR = ln (2 × glucose [mg/dL] + triglycerides [mg/dL]) × BMI [kg/m²] / ln (HDL-C [mg/dL]) [13]. LAP = (waist circumference [cm] − 65) × triglycerides [mmol/L] for males, (waist [cm] − 58) × triglycerides [mmol/L] for females [14]. CMI = (triglycerides [mmol/L] / HDL-C [mmol/L]) × (waist [cm] / height [cm]) [15]. SHR = fasting glucose [mmol/L] / (1.59 × HbA1c [%] − 2.59) [16]. For McAuley, QUICKI, and eGDR, higher values indicate lower IR; for all other indices, higher values indicate greater IR or atherogenic burden. For overall analyses, quartiles were defined based on the full-cohort distribution. For glycemic stratum-stratified analyses, quartiles were defined separately within each glycemic stratum, with Q1 serving as the reference.

In NHANES, fasting plasma glucose and fasting insulin were obtained from venous blood drawn during the morning examination session following an 8–24 h fast. Fasting plasma glucose was measured using the hexokinase enzymatic method. Insulin was quantified using immunoassay platforms that differed across survey cycles: radioimmunoassay in 1999–2002, a chemiluminescent sandwich immunoassay on the Roche Elecsys 2010 in 2011–2012, and immunoenzymometric assays in the other included cycles. HbA1c was measured by high-performance liquid chromatography. Triglycerides, HDL-C, and total cholesterol were determined enzymatically from serum, with triglycerides restricted to the fasting subsample. Waist circumference was recorded at the uppermost lateral border of the right iliac crest. Height and weight were obtained in the mobile examination center; BMI was calculated as weight in kilograms divided by height in meters squared. Trained examiners recorded blood pressure following a standardized auscultatory protocol with a mercury sphygmomanometer through the 2017–2018 cycle, during which NHANES concurrently conducted a methodology comparison study using the Omron HEM-907XL oscillometric device.

Outcome ascertainment

ACM and CVM (ICD-10 codes I00-I09, I11, I13, I20-I51, I60-I69) were the primary outcomes, determined through NDI Plus probabilistic linkage (sensitivity > 97%, specificity > 99%). Survival time was calculated from the mobile examination center visit date to the date of death or December 31, 2019, whichever came first.

Covariate selection

Covariates were prespecified a priori based on established biological and epidemiological associations with both IR and mortality outcomes, and included sociodemographic, anthropometric, lifestyle, and clinical variables. Sociodemographic variables included age (continuous, years), sex (male/female), race/ethnicity (categorized as non-Hispanic White, non-Hispanic Black, Mexican American, other Hispanic, and other/multiracial based on standardized self-report), and poverty-income ratio. Age, sex, and race/ethnicity were included in the primary Cox models, whereas poverty-income ratio was used for cohort completeness and baseline characterization but was not included in the Cox models. Anthropometric measurements included BMI (kg/m², calculated from measured weight and height at the mobile examination center) and waist circumference (cm, measured at the superior border of the iliac crest by trained NCHS examiners). Lifestyle covariates comprised smoking status (categorized as current or former smoker vs. never-smoker, encompassing both combustible and smokeless tobacco use) and alcohol consumption (defined as current use if the participant reported ≥ 12 alcoholic drinks in the preceding year, otherwise classified as non-use). Comorbid conditions were ascertained through self-reported physician diagnosis and included CKD, CAD, heart failure, cerebrovascular accident or transient ischemic attack (stroke), and any malignancy.

Statistical analysis

All analyses were performed using R (version 4.2.2; R Foundation for Statistical Computing) and MSTATA (https://www.mstata.com). Baseline characteristics were summarized as mean ± standard deviation (SD), median (interquartile range, IQR), or n (%), as appropriate, and grouped by glycemic status. Continuous variables were compared using one-way ANOVA or Kruskal-Wallis tests, as appropriate, and categorical variables were compared using chi-square tests. Multivariable Cox proportional hazards regression models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the associations of each IR index with ACM and CVM. Three Cox models were constructed: Model 1 (unadjusted), Model 2 (adjusted for age, sex, race, and BMI), and Model 3 (fully adjusted, including all variables in Model 2 plus smoking, alcohol use, CKD, CAD, heart failure, stroke, and cancer). Both continuous per-unit and quartile-based comparisons are reported for each model. Proportional hazards assumptions were examined using scaled Schoenfeld residuals. Non-linear dose-response relationships were modeled using restricted cubic splines (RCS) within the fully adjusted Cox model (Model 3), with four knots placed at the 5th, 35th, 65th, and 95th percentiles of each index distribution and the median set as the reference value (HR = 1.0). Evidence of non-linearity was evaluated using the P value for non-linearity. Prespecified subgroup analyses, stratified by glycemic status (normoglycemia, prediabetes, diabetes), sex (male, female), and age (< 65 years, ≥ 65 years), were performed under the full Model 3 adjustment to assess potential effect modification. When stratifying by sex or age, the respective variable was excluded from Model 3 adjustments. A sensitivity analysis excluding participants with less than 1 year of follow-up was conducted to assess robustness to potential reverse causation. Statistical significance was defined as P < 0.05 (two-sided).

Results

Baseline characteristics and IR index distributions

Average age increased steadily (54.2 ± 17.2, 60.3 ± 14.9, 63.3 ± 12.3 years; P < 0.001), while the proportion of males was 43.7%, 53.8%, and 53.8% (P < 0.001). BMI, waist circumference, fasting glucose, HbA1c, fasting insulin, triglycerides, and C-reactive protein all increased significantly with worsening glycemic status (all P < 0.001), while HDL-C decreased. The prevalence of CKD (3.6% to 8.5%), CAD (5.4% to 11.3%), heart failure (4.1% to 9.7%), and stroke (5.2% to 8.9%) rose stepwise across the groups (all P < 0.001) (Table 1).

Table 1.

Baseline characteristics by glycemic status

Variables	Normoglycemia (n = 1,869)	Prediabetes (n = 3,389)	Diabetes (n = 2,290)	P value
Age, years	54.2 ± 17.2	60.3 ± 14.9	63.3 ± 12.3	< 0.001
BMI, kg/m²	28.5 ± 6.3	30.3 ± 6.6	32.3 ± 7.1	< 0.001
Waist circumference, cm	97.7 ± 14.7	103.8 ± 15.1	109.6 ± 15.7	< 0.001
Fasting glucose, mg/dL	93.0 (88.9–96.4)	105.0 (101.0-111.0)	137.0 (120.0-172.0)	< 0.001
HbA1c, %	5.3 (5.1–5.5)	5.7 (5.4–5.9)	6.7 (6.1–7.7)	< 0.001
Fasting insulin, µU/mL	7.2 (4.8–11.1)	10.1 (6.7–15.4)	12.6 (7.6–20.8)	< 0.001
Triglycerides, mg/dL	102.0 (73.0-148.0)	118.0 (83.0-168.0)	133.0 (93.0-195.0)	< 0.001
HDL cholesterol, mg/dL	54.0 (45.0–67.0)	51.0 (42.0–62.0)	46.0 (39.0–56.0)	< 0.001
Total cholesterol, mg/dL	200.9 ± 41.8	200.0 ± 42.5	184.0 ± 45.4	< 0.001
LDL cholesterol, mg/dL	118.9 ± 35.5	119.2 ± 36.0	103.6 ± 37.2	< 0.001
C-reactive protein, mg/dL	0.3 (0.1–0.9)	0.5 (0.2–1.4)	0.6 (0.2–1.9)	< 0.001
Poverty-income ratio	2.7 ± 1.6	2.6 ± 1.6	2.3 ± 1.5	< 0.001
Male, n (%)	816 (43.7)	1,824 (53.8)	1,232 (53.8)	< 0.001
Race/ethnicity, n (%)				< 0.001
Non-Hispanic White	871 (46.6)	1,621 (47.8)	877 (38.3)
Non-Hispanic Black	340 (18.2)	699 (20.6)	545 (23.8)
Mexican American	374 (20.0)	544 (16.1)	440 (19.2)
Other Hispanic	173 (9.3)	272 (8.0)	256 (11.2)
Other/Multiracial	111 (5.9)	253 (7.5)	172 (7.5)
Current/former smoker, n (%)	929 (49.7)	1,759 (51.9)	1,191 (52.0)	0.243
Alcohol use, n (%)	1,262 (67.5)	2,298 (67.8)	1,367 (59.7)	< 0.001
Chronic kidney disease, n (%)	67 (3.6)	132 (3.9)	194 (8.5)	< 0.001
Coronary artery disease, n (%)	100 (5.4)	254 (7.5)	258 (11.3)	< 0.001
Heart failure, n (%)	77 (4.1)	161 (4.8)	222 (9.7)	< 0.001
Stroke, n (%)	98 (5.2)	200 (5.9)	203 (8.9)	< 0.001
Cancer, n (%)	204 (10.9)	487 (14.4)	360 (15.7)	< 0.001

Abbreviations: BMI, body mass index; HbA1c, hemoglobin A1c; HDL, high-density lipoprotein; LDL, low-density lipoprotein

All ten IR indices showed notable and consistent differences across glycemic categories (Table 2). HOMA-IR increased from normoglycemia to diabetes, 1.63 (1.08–2.57) to 4.47 (2.52–7.92), while the McAuley index (7.82 to 6.26) and QUICKI (0.35 to 0.31) decreased accordingly (both P < 0.001). eGDR declined by 35.2%, from 6.06 ± 1.34 to 3.93 ± 1.70. TyG, SHR, CMI, AIP, METS-IR, and LAP all increased significantly across glycemic levels (all P < 0.001).

Table 2.

Distribution of insulin resistance indices by glycemic status

Category	IR Index	Normoglycemia (n = 1,869)	Prediabetes (n = 3,389)	Diabetes (n = 2,290)	P value
Fasting-insulin-based	HOMA-IR	1.63 (1.08–2.57)	2.65 (1.74–4.08)	4.47 (2.52–7.92)	< 0.001
	McAuley	7.82 ± 2.26	6.84 ± 1.96	6.26 ± 2.14	< 0.001
	QUICKI	0.35 (0.33–0.38)	0.33 (0.31–0.35)	0.31 (0.29–0.33)	< 0.001
Non-insulin-based	TyG	8.50 ± 0.55	8.76 ± 0.55	9.21 ± 0.75	< 0.001
	SHR	0.89 ± 0.09	0.93 ± 0.12	0.97 ± 0.23	< 0.001
	CMI	0.47 (0.29–0.83)	0.63 (0.37–1.05)	0.83 (0.49–1.34)	< 0.001
	AIP	-0.07 ± 0.31	0.02 ± 0.32	0.11 ± 0.33	< 0.001
	eGDR	6.06 ± 1.34	5.30 ± 1.39	3.93 ± 1.70	< 0.001
	METS-IR	41.08 ± 10.91	45.44 ± 11.69	51.71 ± 13.41	< 0.001
	LAP	42.21 (24.90-68.33)	54.34 (33.48–87.36)	70.55 (44.78-109.84)	< 0.001

Abbreviations: IR, insulin resistance; HOMA-IR, homeostasis model assessment of insulin resistance; QUICKI, quantitative insulin sensitivity check index; TyG, triglyceride-glucose index; SHR, stress hyperglycemia ratio; CMI, cardiometabolic index; AIP, atherogenic index of plasma; eGDR, estimated glucose disposal rate; METS-IR, metabolic score for insulin resistance; LAP, lipid accumulation product

Mortality outcomes

Over a mean follow-up of 8.7 ± 5.3 years (median 7.9, maximum 20.8 years), 1,752 all-cause deaths (23.2%) and 499 cardiovascular deaths (6.6%) were recorded. ACM was similar in normoglycemia and prediabetes but was highest in diabetes (normoglycemia: 21.1%, prediabetes: 21.4%, diabetes: 27.6%; P < 0.001). CVM rose progressively from 5.7% (normoglycemia) to 6.0% (prediabetes) to 8.2% (diabetes) (P < 0.001).

The association of IR indices with all-cause mortality

Table 3 displays the fully adjusted Cox regression results for ACM. Among insulin-based indices, HOMA-IR demonstrated the strongest independent associations. Each unit increase in HOMA-IR was associated with a 1.7% increase in ACM hazard under Model 3 (HR 1.017, 95% CI 1.012–1.022; P < 0.001), and the quartile dose-response was non-monotonic: Q2 (HR 0.879, 95% CI 0.770–1.003; P = 0.056) and Q3 (HR 0.823, 95% CI 0.712–0.951; P = 0.008) showed neutral-to-protective associations, while Q4 carried a significant excess risk (HR 1.165, 95% CI 1.004–1.351; P = 0.044). The McAuley index was not significant per unit (HR 1.004; P = 0.789) or across quartiles. However, RCS analyses revealed a significant overall association for both ACM (P-overall < 0.001, P-nonlinear < 0.001) and CVM (P-overall = 0.001, P-nonlinear = 0.001), suggesting a non-linear pattern concentrated at extreme McAuley values that is not captured by quartile or per-unit analyses. QUICKI showed a non-linear pattern: the per-unit association was not significant (HR 0.696; P = 0.619), whereas higher quartiles were associated with lower ACM risk, Q2 (HR 0.707, 95% CI 0.615–0.813; P < 0.001), Q3 (HR 0.755, 95% CI 0.657–0.867; P < 0.001), and Q4 (HR 0.859, 95% CI 0.740–0.996; P = 0.044), suggesting a non-linear or threshold-like relationship rather than a simple linear gradient.

Table 3.

Cox regression analysis of IR indices and all-cause mortality

IR Index	Model 1		Model 2		Model 3
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
HOMA-IR
Per-unit	1.012 (1.007, 1.017)	< 0.001	1.019 (1.014, 1.024)	< 0.001	1.017 (1.012, 1.022)	< 0.001
Q1 (n = 1,887; events = 483)	Ref		Ref		Ref
Q2 (n = 1,887; events = 448)	0.842 (0.740, 0.957)	0.009	0.873 (0.766, 0.996)	0.043	0.879 (0.770, 1.003)	0.056
Q3 (n = 1,887; events = 375)	0.706 (0.617, 0.808)	< 0.001	0.809 (0.700, 0.935)	0.004	0.823 (0.712, 0.951)	0.008
Q4 (n = 1,887; events = 446)	0.919 (0.808, 1.045)	0.196	1.223 (1.055, 1.418)	0.008	1.165 (1.004, 1.351)	0.044
McAuley
Per-unit	1.039 (1.016, 1.062)	< 0.001	0.991 (0.965, 1.018)	0.498	1.004 (0.977, 1.031)	0.789
Q1 (n = 1,887; events = 399)	Ref		Ref		Ref
Q2 (n = 1,887; events = 453)	1.161 (1.014, 1.328)	0.030	0.912 (0.796, 1.045)	0.185	0.992 (0.865, 1.139)	0.912
Q3 (n = 1,887; events = 445)	1.162 (1.015, 1.330)	0.030	0.807 (0.701, 0.930)	0.003	0.878 (0.762, 1.012)	0.073
Q4 (n = 1,887; events = 455)	1.277 (1.116, 1.460)	< 0.001	0.880 (0.759, 1.021)	0.092	0.961 (0.827, 1.117)	0.605
QUICKI
Per-unit	4.113 (1.252, 13.510)	0.020	0.447 (0.105, 1.903)	0.276	0.696 (0.166, 2.912)	0.619
Q1 (n = 1,887; events = 446)	Ref		Ref		Ref
Q2 (n = 1,887; events = 375)	0.768 (0.670, 0.881)	< 0.001	0.661 (0.576, 0.760)	< 0.001	0.707 (0.615, 0.813)	< 0.001
Q3 (n = 1,887; events = 448)	0.916 (0.804, 1.045)	0.191	0.714 (0.621, 0.820)	< 0.001	0.755 (0.657, 0.867)	< 0.001
Q4 (n = 1,887; events = 483)	1.089 (0.957, 1.238)	0.196	0.818 (0.705, 0.948)	0.008	0.859 (0.740, 0.996)	0.044
TyG
Per-unit	1.126 (1.052, 1.205)	< 0.001	1.200 (1.114, 1.293)	< 0.001	1.158 (1.074, 1.249)	< 0.001
Q1 (n = 1,889; events = 390)	Ref		Ref		Ref
Q2 (n = 1,885; events = 453)	1.146 (1.001, 1.312)	0.049	1.051 (0.917, 1.205)	0.475	1.044 (0.910, 1.197)	0.539
Q3 (n = 1,887; events = 430)	1.067 (0.930, 1.224)	0.353	1.062 (0.922, 1.223)	0.407	1.030 (0.894, 1.187)	0.681
Q4 (n = 1,887; events = 479)	1.259 (1.102, 1.439)	< 0.001	1.320 (1.148, 1.518)	< 0.001	1.233 (1.071, 1.418)	0.003
SHR
Per-unit	1.848 (1.417, 2.410)	< 0.001	1.764 (1.345, 2.315)	< 0.001	1.751 (1.344, 2.282)	< 0.001
Q1 (n = 1,893; events = 422)	Ref		Ref		Ref
Q2 (n = 1,883; events = 417)	0.923 (0.806, 1.056)	0.243	0.884 (0.772, 1.014)	0.078	0.868 (0.757, 0.995)	0.042
Q3 (n = 1,887; events = 438)	0.969 (0.848, 1.108)	0.646	0.904 (0.788, 1.036)	0.146	0.924 (0.806, 1.060)	0.259
Q4 (n = 1,885; events = 475)	1.169 (1.025, 1.333)	0.020	1.116 (0.976, 1.276)	0.108	1.111 (0.971, 1.270)	0.124
CMI
Per-unit	0.969 (0.929, 1.011)	0.143	1.022 (0.991, 1.055)	0.172	1.016 (0.981, 1.051)	0.376
Q1 (n = 1,887; events = 428)	Ref		Ref		Ref
Q2 (n = 1,887; events = 438)	1.020 (0.893, 1.166)	0.769	0.966 (0.843, 1.106)	0.613	0.945 (0.825, 1.082)	0.413
Q3 (n = 1,887; events = 430)	0.931 (0.814, 1.064)	0.294	1.034 (0.898, 1.192)	0.639	0.978 (0.848, 1.127)	0.757
Q4 (n = 1,887; events = 456)	1.007 (0.882, 1.149)	0.918	1.257 (1.088, 1.452)	0.002	1.151 (0.995, 1.331)	0.058
AIP
Per-unit	0.957 (0.829, 1.105)	0.553	1.271 (1.085, 1.489)	0.003	1.141 (0.971, 1.340)	0.110
Q1 (n = 1,889; events = 422)	Ref		Ref		Ref
Q2 (n = 1,886; events = 446)	1.037 (0.908, 1.185)	0.595	1.016 (0.888, 1.162)	0.815	0.980 (0.856, 1.121)	0.767
Q3 (n = 1,886; events = 420)	0.915 (0.799, 1.047)	0.196	0.988 (0.859, 1.136)	0.863	0.930 (0.809, 1.070)	0.312
Q4 (n = 1,887; events = 464)	1.008 (0.883, 1.150)	0.909	1.243 (1.081, 1.429)	0.002	1.126 (0.979, 1.297)	0.097
eGDR
Per-unit	0.975 (0.947, 1.003)	0.078	0.795 (0.757, 0.835)	< 0.001	0.807 (0.767, 0.849)	< 0.001
Q1 (n = 1,887; events = 425)	Ref		Ref		Ref
Q2 (n = 1,887; events = 418)	0.861 (0.752, 0.985)	0.030	0.594 (0.512, 0.689)	< 0.001	0.602 (0.519, 0.699)	< 0.001
Q3 (n = 1,887; events = 459)	0.968 (0.848, 1.105)	0.630	0.578 (0.489, 0.683)	< 0.001	0.608 (0.514, 0.720)	< 0.001
Q4 (n = 1,887; events = 450)	0.866 (0.758, 0.989)	0.034	0.510 (0.413, 0.629)	< 0.001	0.559 (0.452, 0.691)	< 0.001
METS-IR
Per-unit	0.983 (0.979, 0.987)	< 0.001	1.029 (1.019, 1.039)	< 0.001	1.021 (1.011, 1.032)	< 0.001
Q1 (n = 1,887; events = 545)	Ref		Ref		Ref
Q2 (n = 1,887; events = 464)	0.820 (0.725, 0.928)	0.002	0.946 (0.820, 1.090)	0.442	0.853 (0.739, 0.984)	0.029
Q3 (n = 1,887; events = 409)	0.682 (0.600, 0.776)	< 0.001	0.992 (0.833, 1.181)	0.927	0.884 (0.741, 1.055)	0.172
Q4 (n = 1,887; events = 334)	0.591 (0.516, 0.677)	< 0.001	1.445 (1.130, 1.849)	0.003	1.208 (0.942, 1.550)	0.137
LAP
Per-unit	0.999 (0.998, 1.000)	0.048	1.001 (1.000, 1.001)	0.016	1.001 (1.000, 1.001)	0.048
Q1 (n = 1,887; events = 434)	Ref		Ref		Ref
Q2 (n = 1,887; events = 471)	1.113 (0.977, 1.268)	0.107	1.108 (0.967, 1.269)	0.140	1.104 (0.964, 1.266)	0.153
Q3 (n = 1,887; events = 435)	0.961 (0.841, 1.098)	0.556	1.091 (0.942, 1.262)	0.244	1.049 (0.906, 1.215)	0.520
Q4 (n = 1,887; events = 412)	0.940 (0.821, 1.075)	0.365	1.432 (1.216, 1.686)	< 0.001	1.330 (1.129, 1.567)	< 0.001

Model 1: Unadjusted

Model 2: Adjusted for sex, age, race, and BMI

Model 3: Adjusted for sex, age, race, BMI, CKD, heart failure, CAD, stroke, smoking, alcohol use, and cancer

Abbreviations: IR, insulin resistance; HR, hazard ratio; CI, confidence interval; Ref, reference; Q1-Q4, quartiles 1-4; HOMA-IR, homeostasis model assessment of insulin resistance; QUICKI, quantitative insulin sensitivity check index; TyG, triglyceride-glucose index; SHR, stress hyperglycemia ratio; CMI, cardiometabolic index; AIP, atherogenic index of plasma; eGDR, estimated glucose disposal rate; METS-IR, metabolic score for insulin resistance; LAP, lipid accumulation product

Among non-insulin-based indices, eGDR showed the strongest inverse relationship with ACM. The per-unit HR in Model 3 was 0.807 (95% CI 0.767–0.849; P < 0.001), corresponding to a 19.3% decrease in ACM hazard per mg/kg/min increase in eGDR. All three higher quartiles demonstrated consistent significant reductions: Q2 HR 0.602 (95% CI 0.519–0.699; P < 0.001), Q3 HR 0.608 (95% CI 0.514–0.720; P < 0.001), and Q4 HR 0.559 (95% CI 0.452–0.691; P < 0.001). TyG was significantly associated with ACM both per unit (HR 1.158, 95% CI 1.074–1.249; P < 0.001) and in the highest quartile (Q4 HR 1.233, 95% CI 1.071–1.418; P = 0.003). SHR showed the highest per-unit HR for ACM among all ten indices (HR 1.751, 95% CI 1.344–2.282; P < 0.001), although its quartile pattern was non-monotonic: Q2 showed a significant protective association (HR 0.868, 95% CI 0.757–0.995; P = 0.042), whereas Q4 was not significant (HR 1.111; P = 0.124). METS-IR was significant per unit (HR 1.021, 95% CI 1.011–1.032; P < 0.001), with Q2 exhibiting a protective association (HR 0.853, 95% CI 0.739–0.984; P = 0.029), but Q4 was not significant (HR 1.208; P = 0.137). LAP was significant both per unit (HR 1.001, 95% CI 1.000-1.001; P = 0.048) and at Q4 (HR 1.330, 95% CI 1.129–1.567; P < 0.001). AIP (per-unit HR 1.141; P = 0.110) and CMI (per-unit HR 1.016; P = 0.376) did not reach significance per unit or across quartiles. RCS analysis for CMI revealed a significant non-linear overall association with ACM (P-overall = 0.005, P-nonlinear = 0.003), suggesting that the association may be concentrated at the extremes of the distribution.

The association of IR indices with cardiovascular mortality

Table 4 presents the fully adjusted CVM results. HOMA-IR remained significant on a per-unit basis (HR 1.012, 95% CI 1.001–1.022; P = 0.029), with a protective non-monotonic pattern across quartiles: Q2 (HR 0.743, 95% CI 0.578–0.956; P = 0.021) and Q3 (HR 0.680, 95% CI 0.517–0.895; P = 0.006) were protective, while Q4 was not significant (HR 1.094; P = 0.516). The McAuley index was not significant either per unit or across all quartiles. QUICKI again showed no per-unit significance (P = 0.894); however, Q2 (HR 0.622, 95% CI 0.479–0.807; P < 0.001) and Q3 (HR 0.679, 95% CI 0.524–0.880; P = 0.003) demonstrated significant CVM protection, consistent with the threshold pattern observed for ACM.

Table 4.

Cox regression analysis of IR indices and cardiovascular mortality

IR Index	Model 1		Model 2		Model 3
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
HOMA-IR
Per-unit	1.012 (1.002, 1.021)	0.013	1.017 (1.007, 1.026)	< 0.001	1.012 (1.001, 1.022)	0.029
Q1 (n = 1,887; events = 140)	Ref		Ref		Ref
Q2 (n = 1,887; events = 118)	0.761 (0.596, 0.973)	0.029	0.754 (0.587, 0.969)	0.027	0.743 (0.578, 0.956)	0.021
Q3 (n = 1,887; events = 100)	0.647 (0.500, 0.836)	< 0.001	0.682 (0.518, 0.897)	0.006	0.680 (0.517, 0.895)	0.006
Q4 (n = 1,887; events = 141)	1.004 (0.795, 1.268)	0.974	1.187 (0.906, 1.555)	0.215	1.094 (0.835, 1.433)	0.516
McAuley
Per-unit	1.030 (0.988, 1.073)	0.160	0.995 (0.946, 1.047)	0.860	1.015 (0.965, 1.068)	0.556
Q1 (n = 1,887; events = 112)	Ref		Ref		Ref
Q2 (n = 1,887; events = 141)	1.289 (1.006, 1.652)	0.045	1.039 (0.808, 1.337)	0.764	1.159 (0.899, 1.494)	0.256
Q3 (n = 1,887; events = 116)	1.081 (0.834, 1.401)	0.557	0.785 (0.599, 1.029)	0.079	0.869 (0.662, 1.140)	0.311
Q4 (n = 1,887; events = 130)	1.305 (1.014, 1.681)	0.039	0.967 (0.732, 1.278)	0.815	1.090 (0.823, 1.444)	0.547
QUICKI
Per-unit	1.162 (0.119, 11.349)	0.897	0.305 (0.020, 4.754)	0.397	0.832 (0.056, 12.377)	0.894
Q1 (n = 1,887; events = 141)	Ref		Ref		Ref
Q2 (n = 1,887; events = 100)	0.644 (0.499, 0.833)	< 0.001	0.575 (0.443, 0.745)	< 0.001	0.622 (0.479, 0.807)	< 0.001
Q3 (n = 1,887; events = 118)	0.758 (0.594, 0.968)	0.027	0.636 (0.490, 0.824)	< 0.001	0.679 (0.524, 0.880)	0.003
Q4 (n = 1,887; events = 140)	0.996 (0.788, 1.258)	0.974	0.843 (0.643, 1.104)	0.215	0.914 (0.698, 1.198)	0.516
TyG
Per-unit	1.154 (1.018, 1.309)	0.026	1.214 (1.057, 1.395)	0.006	1.160 (1.007, 1.335)	0.039
Q1 (n = 1,889; events = 108)	Ref		Ref		Ref
Q2 (n = 1,885; events = 126)	1.149 (0.888, 1.486)	0.290	1.053 (0.812, 1.365)	0.699	1.035 (0.797, 1.342)	0.798
Q3 (n = 1,887; events = 130)	1.162 (0.901, 1.500)	0.248	1.151 (0.885, 1.497)	0.296	1.110 (0.852, 1.445)	0.440
Q4 (n = 1,887; events = 135)	1.283 (0.996, 1.652)	0.054	1.317 (1.011, 1.715)	0.041	1.216 (0.932, 1.586)	0.149
SHR
Per-unit	1.519 (0.892, 2.585)	0.124	1.354 (0.779, 2.352)	0.283	1.354 (0.780, 2.349)	0.281
Q1 (n = 1,893; events = 133)	Ref		Ref		Ref
Q2 (n = 1,883; events = 113)	0.783 (0.609, 1.008)	0.057	0.747 (0.580, 0.963)	0.024	0.728 (0.564, 0.940)	0.015
Q3 (n = 1,887; events = 126)	0.888 (0.696, 1.133)	0.340	0.810 (0.631, 1.039)	0.097	0.844 (0.657, 1.082)	0.181
Q4 (n = 1,885; events = 127)	0.996 (0.781, 1.270)	0.971	0.925 (0.722, 1.186)	0.540	0.898 (0.701, 1.152)	0.398
CMI
Per-unit	0.964 (0.889, 1.046)	0.379	1.008 (0.945, 1.074)	0.816	0.995 (0.925, 1.071)	0.895
Q1 (n = 1,887; events = 123)	Ref		Ref		Ref
Q2 (n = 1,887; events = 109)	0.883 (0.682, 1.142)	0.343	0.794 (0.610, 1.032)	0.084	0.764 (0.587, 0.994)	0.045
Q3 (n = 1,887; events = 134)	1.005 (0.786, 1.283)	0.971	1.051 (0.810, 1.363)	0.709	0.981 (0.756, 1.272)	0.883
Q4 (n = 1,887; events = 133)	1.018 (0.797, 1.301)	0.887	1.183 (0.905, 1.548)	0.219	1.055 (0.805, 1.383)	0.697
AIP
Per-unit	1.007 (0.770, 1.317)	0.957	1.286 (0.957, 1.729)	0.096	1.115 (0.823, 1.510)	0.483
Q1 (n = 1,889; events = 115)	Ref		Ref		Ref
Q2 (n = 1,886; events = 124)	1.052 (0.816, 1.356)	0.696	1.003 (0.777, 1.296)	0.980	0.955 (0.739, 1.234)	0.724
Q3 (n = 1,886; events = 120)	0.948 (0.734, 1.225)	0.683	0.995 (0.764, 1.296)	0.971	0.928 (0.712, 1.210)	0.582
Q4 (n = 1,887; events = 140)	1.107 (0.865, 1.416)	0.421	1.320 (1.016, 1.714)	0.037	1.165 (0.895, 1.516)	0.257
eGDR
Per-unit	0.921 (0.874, 0.970)	0.002	0.760 (0.696, 0.830)	< 0.001	0.774 (0.706, 0.848)	< 0.001
Q1 (n = 1,887; events = 134)	Ref		Ref		Ref
Q2 (n = 1,887; events = 127)	0.823 (0.646, 1.049)	0.116	0.593 (0.453, 0.776)	< 0.001	0.597 (0.455, 0.783)	< 0.001
Q3 (n = 1,887; events = 121)	0.805 (0.629, 1.029)	0.083	0.523 (0.383, 0.715)	< 0.001	0.553 (0.403, 0.758)	< 0.001
Q4 (n = 1,887; events = 117)	0.704 (0.549, 0.902)	0.006	0.485 (0.328, 0.717)	< 0.001	0.538 (0.362, 0.798)	0.002
METS-IR
Per-unit	0.989 (0.982, 0.997)	0.005	1.030 (1.011, 1.049)	0.001	1.019 (1.000, 1.038)	0.049
Q1 (n = 1,887; events = 136)	Ref		Ref		Ref
Q2 (n = 1,887; events = 135)	0.956 (0.753, 1.213)	0.710	1.068 (0.815, 1.400)	0.632	0.925 (0.704, 1.215)	0.573
Q3 (n = 1,887; events = 119)	0.793 (0.620, 1.014)	0.064	1.122 (0.808, 1.558)	0.493	0.954 (0.686, 1.328)	0.780
Q4 (n = 1,887; events = 109)	0.772 (0.600, 0.993)	0.044	1.849 (1.177, 2.903)	0.008	1.487 (0.942, 2.348)	0.088
LAP
Per-unit	0.999 (0.998, 1.001)	0.230	1.000 (0.999, 1.002)	0.643	1.000 (0.999, 1.001)	0.857
Q1 (n = 1,887; events = 116)	Ref		Ref		Ref
Q2 (n = 1,887; events = 136)	1.206 (0.941, 1.544)	0.139	1.136 (0.878, 1.471)	0.332	1.149 (0.887, 1.489)	0.292
Q3 (n = 1,887; events = 126)	1.039 (0.808, 1.338)	0.764	1.093 (0.829, 1.441)	0.530	1.056 (0.800, 1.393)	0.702
Q4 (n = 1,887; events = 121)	1.033 (0.801, 1.333)	0.800	1.428 (1.049, 1.943)	0.024	1.354 (0.994, 1.845)	0.055

Model 1: Unadjusted

Model 2: Adjusted for sex, age, race, and BMI

Model 3: Adjusted for sex, age, race, BMI, CKD, heart failure, CAD, stroke, smoking, alcohol use, and cancer

Abbreviations: IR, insulin resistance; HR, hazard ratio; CI, confidence interval; Ref, reference; Q1-Q4, quartiles 1-4; HOMA-IR, homeostasis model assessment of insulin resistance; QUICKI, quantitative insulin sensitivity check index; TyG, triglyceride-glucose index; SHR, stress hyperglycemia ratio; CMI, cardiometabolic index; AIP, atherogenic index of plasma; eGDR, estimated glucose disposal rate; METS-IR, metabolic score for insulin resistance; LAP, lipid accumulation product

Among non-insulin-based indices, eGDR again showed the strongest associations with CVM: a per-unit HR of 0.774 (95% CI 0.706–0.848; P < 0.001); Q2 HR of 0.597 (95% CI 0.455–0.783; P < 0.001); Q3 HR of 0.553 (95% CI 0.403–0.758; P < 0.001); and Q4 HR of 0.538 (95% CI 0.362–0.798; P = 0.002), indicating a 46.2% lower CVM hazard when comparing the highest to the lowest eGDR quartile. TyG reached significance for per-unit CVM (HR 1.160, 95% CI 1.007–1.335; P = 0.039), but Q4 was not significant (HR 1.216; P = 0.149). METS-IR was significant per-unit (HR 1.019, 95% CI 1.000-1.038; P = 0.049), though Q4 was not (HR 1.487; P = 0.088). Notably, SHR, despite having the highest per-unit HR among all ten indices for ACM, was completely non-significant for CVM per-unit (HR 1.354, 95% CI 0.780–2.349; P = 0.281), demonstrating a fundamental outcome-specific dissociation; however, SHR Q2 showed a significant protective association with CVM (HR 0.728, 95% CI 0.564–0.940; P = 0.015). LAP was non-significant per-unit (P = 0.857) and at Q4 for CVM (HR 1.354; P = 0.055). CMI Q2 showed a nominally significant protective association with CVM (HR 0.764, 95% CI 0.587–0.994; P = 0.045), although per-unit and overall RCS analyses did not confirm a significant CMI-CVM association (P-overall = 0.255), suggesting this isolated quartile finding should be interpreted with caution.

Non-linear dose-response relationships

Model 3-adjusted RCS analyses (Figs. 2 and 3) primarily revealed non-linear dose-response relationships. For ACM, HOMA-IR showed a J-shaped association, with risk increasing sharply at higher values (P for non-linearity < 0.001). eGDR demonstrated an inverse relationship that weakened at higher levels (P for non-linearity < 0.001). TyG, METS-IR, CMI, LAP, SHR, McAuley index, and QUICKI also showed significant non-linear associations with ACM, although the direction and shape varied according to index orientation and distribution (all P-values for non-linearity < 0.05). AIP showed a mostly linear association with ACM (P for nonlinearity = 0.080) (Fig. 2).

Fig. 2.

Restricted cubic spline analyses of the associations between insulin resistance indices and all-cause mortality in hypertensive adults. Panel A, HOMA-IR; Panel B, McAuley index; Panel C, QUICKI; Panel D, TyG; Panel E, SHR; Panel F, CMI; Panel G, AIP; Panel H, eGDR; Panel I, METS-IR; Panel J, LAP. Red solid lines indicate adjusted hazard ratios from Model 3, shaded areas indicate 95% confidence intervals, blue histograms show the distribution of each index, and the dashed horizontal line indicates HR = 1.0

Fig. 3.

Restricted cubic spline analyses of the associations between insulin resistance indices and cardiovascular mortality in hypertensive adults. Panel A, HOMA-IR; Panel B, McAuley index; Panel C, QUICKI; Panel D, TyG; Panel E, SHR; Panel F, CMI; Panel G, AIP; Panel H, eGDR; Panel I, METS-IR; Panel J, LAP. Red solid lines indicate adjusted hazard ratios from Model 3, shaded areas indicate 95% confidence intervals, blue histograms show the distribution of each index, and the dashed horizontal line indicates HR = 1.0

For CVM, significant non-linearity was observed in HOMA-IR (P for non-linearity < 0.001), McAuley index (P for non-linearity = 0.001), QUICKI (P for non-linearity < 0.001), METS-IR (P for non-linearity = 0.015), SHR (P for non-linearity < 0.001), and eGDR (P for non-linearity = 0.010). TyG, CMI, AIP, and LAP did not show statistically significant non-linearity for CVM, indicating generally linear or non-significantly non-linear patterns in the fully adjusted spline models (Fig. 3).

Subgroup analysis by glycemic status

Tables 5 and 6 present subgroup results by glycemic status. Across all three strata, the pattern of index significance expanded progressively from normoglycemia to diabetes, with both the number of significant indices and the range of outcomes broadening as glycemic status worsened.

Table 5.

Subgroup analysis by glycemic status: IR indices and all-cause mortality

IR Index	Normoglycemia n = 1,869, ACM = 394		Prediabetes n = 3,389, ACM = 725		Diabetes n = 2,290, ACM = 633
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
HOMA-IR
Per-unit	1.085 (1.005, 1.172)	0.037	1.039 (1.004, 1.076)	0.028	1.012 (1.006, 1.018)	< 0.001
Q1	Ref		Ref		Ref
Q2	0.879 (0.670, 1.154)	0.354	0.833 (0.681, 1.020)	0.077	0.849 (0.681, 1.058)	0.145
Q3	0.796 (0.587, 1.080)	0.143	0.734 (0.587, 0.920)	0.007	0.843 (0.671, 1.057)	0.139
Q4	1.067 (0.751, 1.516)	0.719	1.001 (0.788, 1.270)	0.997	1.041 (0.826, 1.312)	0.731
McAuley
Per-unit	0.989 (0.933, 1.049)	0.714	1.027 (0.981, 1.075)	0.249	1.021 (0.979, 1.066)	0.332
Q1	Ref		Ref		Ref
Q2	0.921 (0.681, 1.244)	0.590	0.982 (0.792, 1.218)	0.869	0.848 (0.676, 1.065)	0.156
Q3	0.686 (0.499, 0.941)	0.020	0.927 (0.744, 1.155)	0.500	0.950 (0.758, 1.190)	0.653
Q4	0.923 (0.660, 1.291)	0.641	1.008 (0.797, 1.276)	0.945	0.928 (0.731, 1.177)	0.537
QUICKI
Per-unit	0.847 (0.030, 24.287)	0.923	5.803 (0.357, 94.381)	0.217	3.413 (0.392, 29.732)	0.266
Q1	Ref		Ref		Ref
Q2	0.746 (0.538, 1.034)	0.079	0.733 (0.586, 0.916)	0.006	0.809 (0.644, 1.017)	0.069
Q3	0.818 (0.593, 1.128)	0.221	0.833 (0.668, 1.037)	0.103	0.815 (0.647, 1.027)	0.083
Q4	0.933 (0.656, 1.326)	0.698	0.997 (0.786, 1.265)	0.981	0.960 (0.762, 1.210)	0.731
TyG
Per-unit	1.022 (0.822, 1.271)	0.843	0.970 (0.836, 1.125)	0.684	1.147 (1.026, 1.283)	0.016
Q1	Ref		Ref		Ref
Q2	0.857 (0.644, 1.140)	0.289	0.950 (0.767, 1.176)	0.638	1.048 (0.834, 1.316)	0.689
Q3	0.843 (0.630, 1.127)	0.249	1.020 (0.823, 1.264)	0.859	0.907 (0.718, 1.145)	0.410
Q4	1.002 (0.740, 1.356)	0.991	0.967 (0.773, 1.211)	0.771	1.337 (1.067, 1.675)	0.012
SHR
Per-unit	1.469 (0.403, 5.363)	0.560	1.605 (0.816, 3.155)	0.170	1.438 (1.054, 1.962)	0.022
Q1	Ref		Ref		Ref
Q2	0.955 (0.715, 1.276)	0.755	0.961 (0.778, 1.187)	0.711	0.862 (0.685, 1.084)	0.204
Q3	0.910 (0.682, 1.213)	0.519	0.985 (0.795, 1.221)	0.892	0.800 (0.634, 1.011)	0.062
Q4	1.075 (0.805, 1.437)	0.624	1.075 (0.866, 1.334)	0.513	1.193 (0.958, 1.485)	0.114
CMI
Per-unit	1.098 (0.936, 1.286)	0.251	0.970 (0.904, 1.042)	0.406	1.007 (0.963, 1.053)	0.754
Q1	Ref		Ref		Ref
Q2	0.840 (0.634, 1.112)	0.223	0.984 (0.795, 1.217)	0.880	1.061 (0.849, 1.327)	0.602
Q3	0.812 (0.604, 1.091)	0.166	0.976 (0.779, 1.224)	0.834	0.976 (0.774, 1.232)	0.841
Q4	1.033 (0.750, 1.423)	0.844	1.112 (0.885, 1.396)	0.362	1.171 (0.924, 1.485)	0.191
AIP
Per-unit	0.985 (0.667, 1.455)	0.941	1.015 (0.783, 1.317)	0.910	1.123 (0.871, 1.448)	0.371
Q1	Ref		Ref		Ref
Q2	0.874 (0.662, 1.152)	0.339	0.950 (0.770, 1.173)	0.635	1.107 (0.885, 1.386)	0.373
Q3	0.774 (0.578, 1.036)	0.086	0.897 (0.721, 1.117)	0.332	0.946 (0.747, 1.198)	0.645
Q4	1.009 (0.743, 1.372)	0.953	1.050 (0.843, 1.307)	0.665	1.130 (0.894, 1.428)	0.306
eGDR
Per-unit	0.830 (0.679, 1.014)	0.068	0.843 (0.731, 0.971)	0.018	0.842 (0.783, 0.905)	< 0.001
Q1	Ref		Ref		Ref
Q2	0.794 (0.557, 1.133)	0.204	0.670 (0.520, 0.864)	0.002	0.644 (0.509, 0.816)	< 0.001
Q3	0.639 (0.418, 0.977)	0.038	0.726 (0.539, 0.977)	0.035	0.515 (0.396, 0.668)	< 0.001
Q4	0.791 (0.465, 1.346)	0.387	0.657 (0.451, 0.958)	0.029	0.555 (0.414, 0.744)	< 0.001
METS-IR
Per-unit	1.001 (0.970, 1.032)	0.952	1.012 (0.993, 1.031)	0.218	1.012 (0.998, 1.027)	0.100
Q1	Ref		Ref		Ref
Q2	0.944 (0.691, 1.288)	0.714	0.912 (0.732, 1.136)	0.411	0.785 (0.624, 0.988)	0.039
Q3	1.148 (0.759, 1.737)	0.512	0.868 (0.658, 1.143)	0.313	0.960 (0.729, 1.264)	0.771
Q4	1.698 (0.926, 3.112)	0.087	1.080 (0.724, 1.610)	0.707	1.096 (0.735, 1.634)	0.654
LAP
Per-unit	1.003 (1.000, 1.005)	0.065	0.999 (0.998, 1.001)	0.318	1.001 (1.000, 1.002)	0.087
Q1	Ref		Ref		Ref
Q2	0.810 (0.605, 1.085)	0.158	1.104 (0.891, 1.368)	0.365	0.945 (0.755, 1.182)	0.618
Q3	1.034 (0.762, 1.403)	0.831	0.956 (0.761, 1.202)	0.702	0.984 (0.776, 1.249)	0.896
Q4	1.055 (0.725, 1.536)	0.780	1.133 (0.878, 1.462)	0.336	1.172 (0.909, 1.511)	0.221

Fully adjusted model: sex, age, race, BMI, CKD, heart failure, CAD, stroke, smoking, alcohol use, and cancer. Quartiles for each IR index were defined separately within each glycemic stratum, with Q1 serving as the reference

Abbreviations: IR, insulin resistance; ACM, all-cause mortality; HR, hazard ratio; CI, confidence interval; Ref, reference; Q1-Q4, quartiles 1-4; HOMA-IR, homeostasis model assessment of insulin resistance; QUICKI, quantitative insulin sensitivity check index; TyG, triglyceride-glucose index; SHR, stress hyperglycemia ratio; CMI, cardiometabolic index; AIP, atherogenic index of plasma; eGDR, estimated glucose disposal rate; METS-IR, metabolic score for insulin resistance; LAP, lipid accumulation product

Table 6.

Subgroup analysis by glycemic status: IR indices and cardiovascular mortality

IR Index	Normoglycemia n = 1,869, CVM = 106		Prediabetes n = 3,389, CVM = 205		Diabetes n = 2,290, CVM = 188
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
HOMA-IR
Per-unit	0.998 (0.847, 1.176)	0.983	1.048 (0.987, 1.114)	0.123	1.006 (0.993, 1.019)	0.356
Q1	Ref		Ref		Ref
Q2	0.969 (0.570, 1.646)	0.906	0.625 (0.423, 0.923)	0.018	0.625 (0.411, 0.951)	0.028
Q3	0.716 (0.390, 1.316)	0.282	0.545 (0.354, 0.841)	0.006	0.828 (0.554, 1.235)	0.355
Q4	0.867 (0.433, 1.737)	0.687	0.938 (0.611, 1.438)	0.769	0.886 (0.584, 1.345)	0.571
McAuley
Per-unit	1.061 (0.947, 1.188)	0.311	1.031 (0.945, 1.124)	0.493	1.021 (0.944, 1.104)	0.605
Q1	Ref		Ref		Ref
Q2	1.137 (0.643, 2.011)	0.658	1.047 (0.705, 1.556)	0.819	1.008 (0.660, 1.542)	0.969
Q3	0.616 (0.328, 1.159)	0.133	0.837 (0.549, 1.277)	0.410	1.146 (0.754, 1.740)	0.524
Q4	1.113 (0.576, 2.150)	0.750	1.180 (0.763, 1.825)	0.456	1.022 (0.656, 1.593)	0.923
QUICKI
Per-unit	19.014 (0.037, 9,805.912)	0.355	6.937 (0.032, 1,498.381)	0.480	2.280 (0.039, 131.889)	0.691
Q1	Ref		Ref		Ref
Q2	0.827 (0.433, 1.579)	0.565	0.580 (0.383, 0.880)	0.010	0.934 (0.617, 1.412)	0.745
Q3	1.108 (0.596, 2.060)	0.746	0.666 (0.443, 1.000)	0.050	0.705 (0.451, 1.104)	0.127
Q4	1.149 (0.573, 2.301)	0.696	1.063 (0.693, 1.630)	0.780	1.128 (0.743, 1.712)	0.571
TyG
Per-unit	0.965 (0.628, 1.482)	0.869	0.924 (0.699, 1.222)	0.579	1.190 (0.971, 1.457)	0.093
Q1	Ref		Ref		Ref
Q2	0.750 (0.425, 1.325)	0.322	0.787 (0.523, 1.182)	0.248	1.135 (0.748, 1.721)	0.552
Q3	0.898 (0.513, 1.571)	0.706	0.955 (0.643, 1.418)	0.820	0.789 (0.504, 1.237)	0.302
Q4	0.959 (0.530, 1.735)	0.890	0.877 (0.576, 1.335)	0.539	1.595 (1.064, 2.392)	0.024
SHR
Per-unit	1.208 (0.090, 16.230)	0.886	0.804 (0.220, 2.933)	0.741	1.265 (0.685, 2.337)	0.453
Q1	Ref		Ref		Ref
Q2	0.834 (0.474, 1.469)	0.530	0.773 (0.525, 1.136)	0.190	0.938 (0.619, 1.422)	0.764
Q3	0.753 (0.425, 1.335)	0.332	0.831 (0.563, 1.225)	0.350	0.750 (0.485, 1.159)	0.195
Q4	1.198 (0.697, 2.060)	0.514	0.764 (0.509, 1.149)	0.196	1.180 (0.787, 1.769)	0.423
CMI
Per-unit	1.069 (0.763, 1.499)	0.697	0.959 (0.840, 1.093)	0.529	0.991 (0.893, 1.100)	0.862
Q1	Ref		Ref		Ref
Q2	0.843 (0.484, 1.467)	0.545	0.875 (0.577, 1.328)	0.530	1.529 (1.025, 2.280)	0.037
Q3	0.735 (0.409, 1.319)	0.302	1.011 (0.666, 1.536)	0.958	0.820 (0.513, 1.311)	0.408
Q4	1.022 (0.545, 1.919)	0.946	1.090 (0.708, 1.678)	0.696	1.459 (0.940, 2.265)	0.092
AIP
Per-unit	1.055 (0.488, 2.284)	0.891	0.985 (0.604, 1.609)	0.953	1.081 (0.677, 1.726)	0.745
Q1	Ref		Ref		Ref
Q2	0.872 (0.508, 1.498)	0.621	0.975 (0.650, 1.461)	0.902	1.604 (1.069, 2.406)	0.022
Q3	0.695 (0.391, 1.236)	0.215	0.874 (0.575, 1.329)	0.530	0.904 (0.566, 1.446)	0.675
Q4	0.964 (0.526, 1.767)	0.905	1.128 (0.746, 1.705)	0.569	1.351 (0.870, 2.098)	0.180
eGDR
Per-unit	0.692 (0.470, 1.017)	0.061	0.825 (0.633, 1.075)	0.155	0.809 (0.710, 0.920)	0.001
Q1	Ref		Ref		Ref
Q2	0.673 (0.346, 1.309)	0.243	0.549 (0.351, 0.858)	0.009	0.492 (0.319, 0.757)	0.001
Q3	0.644 (0.297, 1.398)	0.266	0.440 (0.254, 0.765)	0.004	0.457 (0.288, 0.724)	< 0.001
Q4	0.454 (0.165, 1.249)	0.126	0.465 (0.233, 0.929)	0.030	0.463 (0.276, 0.776)	0.003
METS-IR
Per-unit	1.018 (0.959, 1.080)	0.557	1.012 (0.977, 1.047)	0.510	1.008 (0.981, 1.037)	0.546
Q1	Ref		Ref		Ref
Q2	1.033 (0.560, 1.905)	0.917	0.837 (0.546, 1.283)	0.415	0.769 (0.508, 1.166)	0.216
Q3	1.256 (0.559, 2.824)	0.581	0.924 (0.556, 1.536)	0.761	0.846 (0.510, 1.404)	0.518
Q4	1.576 (0.480, 5.171)	0.453	1.236 (0.600, 2.547)	0.566	1.003 (0.480, 2.092)	0.995
LAP
Per-unit	1.002 (0.996, 1.008)	0.521	0.999 (0.996, 1.002)	0.421	1.000 (0.998, 1.002)	0.738
Q1	Ref		Ref		Ref
Q2	0.767 (0.429, 1.372)	0.371	1.025 (0.677, 1.550)	0.908	0.967 (0.649, 1.442)	0.871
Q3	0.933 (0.509, 1.710)	0.822	0.880 (0.571, 1.355)	0.561	0.935 (0.605, 1.443)	0.761
Q4	1.283 (0.629, 2.615)	0.493	1.133 (0.704, 1.823)	0.608	1.070 (0.669, 1.711)	0.779

Fully adjusted model: sex, age, race, BMI, CKD, heart failure, CAD, stroke, smoking, alcohol use, and cancer. Quartiles for each IR index were defined separately within each glycemic stratum, with Q1 serving as the reference

Abbreviations: IR, insulin resistance; CVM, cardiovascular mortality; HR, hazard ratio; CI, confidence interval; Ref, reference; Q1-Q4, quartiles 1-4; HOMA-IR, homeostasis model assessment of insulin resistance; QUICKI, quantitative insulin sensitivity check index; TyG, triglyceride-glucose index; SHR, stress hyperglycemia ratio; CMI, cardiometabolic index; AIP, atherogenic index of plasma; eGDR, estimated glucose disposal rate; METS-IR, metabolic score for insulin resistance; LAP, lipid accumulation product

For ACM, HOMA-IR was the only index maintaining per-unit significance across the full glycemic spectrum, normoglycemia (HR 1.085; P = 0.037), prediabetes (HR 1.039; P = 0.028), and diabetes (HR 1.012; P < 0.001), demonstrating that insulin-mediated hepatic IR carries independent prognostic information at every stage of glucose dysregulation. eGDR showed per-unit ACM significance in prediabetes (HR 0.843; P = 0.018) and diabetes (HR 0.842; P < 0.001), with a consistent inverse quartile pattern in both groups. In the diabetic subgroup, additional ACM signals emerged: TyG per-unit and the highest quartile (reflecting glucolipotoxicity), SHR per-unit, and a protective METS-IR quartile association. In normoglycemia, only McAuley Q3 and eGDR Q3 contributed quartile-level ACM protection beyond HOMA-IR (Table 5).

For CVM, no index reached per-unit significance in normoglycemia, consistent with the limited event count (n = 106). In prediabetes, a protective pattern emerged across both index categories, with HOMA-IR, QUICKI, and eGDR showing significant quartile-level protection for CVM, suggesting that selected IR surrogates may help identify lower CVM risk at this transitional glycemic stage. In the diabetic subgroup, eGDR was the only index showing both a significant per-unit association (HR 0.809; P = 0.001) and consistent quartile-level associations with CVM, supporting its role as the most informative non-insulin-based predictor in this group. TyG Q4 (HR 1.595; P = 0.024), CMI Q2, and AIP Q2 showed nominally significant adverse associations with CVM, suggesting an additional lipid-related cardiovascular risk pattern in established diabetes (Table 6).

Sex- and age-stratified analyses

Sex-stratified analyses (Supplementary Table S1) revealed consistent associations between ACM and measures such as HOMA-IR, TyG, and eGDR in both males and females under Model 3. The eGDR per-unit HRs were 0.776 (95% CI 0.721–0.835; P < 0.001) in males, and 0.826 (95% CI 0.769–0.887; P < 0.001) in females; HOMA-IR per-unit HRs were 1.016 (P < 0.001), and 1.018 (P < 0.001); TyG per-unit HRs were 1.124 (P = 0.021) and 1.153 (P = 0.017), respectively. The SHR was significant in females (HR 2.013, 95% CI 1.486–2.727; P < 0.001) but not in males (HR 1.297; P = 0.240).

Age-stratified analyses (Supplementary Table S2) revealed significantly stronger associations in participants under 65 years. The eGDR per-unit HR was 0.655 (95% CI 0.607–0.706; P < 0.001) for those under 65 compared to 0.872 (95% CI 0.818–0.930; P < 0.001) for those 65 and older. TyG was significant only in the younger group (HR 1.334, 95% CI 1.180–1.508; P < 0.001), but not in older adults (HR 1.015, P = 0.749). HOMA-IR remained significant in both groups (under 65: HR 1.017, P < 0.001; 65 and older: HR 1.012, P = 0.002).

Sensitivity analysis

To assess robustness against reverse causation from undiagnosed terminal illness at baseline, a sensitivity analysis excluding participants with less than one year of follow-up (n = 7,461; deaths = 1,665) was conducted; results closely matched the primary analysis (Supplementary Table S3): HOMA-IR per-unit HR 1.017 (P < 0.001), TyG HR 1.149 (P < 0.001), SHR HR 1.667 (P < 0.001), eGDR HR 0.809 (P < 0.001), and METS-IR HR 1.019 (P < 0.001), indicating that the primary findings are not materially influenced by reverse causation.

Discussion

In this direct comparison of ten IR surrogates for ACM and CVM across the glycemic continuum in 7,548 hypertensive adults, several key findings emerge. HOMA-IR was the only index with per-unit significance for ACM across all three glycemic categories, establishing it as the most widely applicable insulin-based predictor. eGDR demonstrated the most consistent non-insulin-based associations, with per-unit and Q4 ACM significance in prediabetes and diabetes, per-unit and quartile-level CVM associations in diabetes, and Q4-level CVM protection in prediabetes. HOMA-IR showed a J-shaped dose-response curve, while the McAuley index did not achieve consistent independent associations after full adjustment. Glycemic status significantly influenced the comparative performance, and IR-mortality associations were stronger in adults under 65 years, especially for eGDR and TyG. These findings define a comparative prognostic hierarchy of IR surrogates across glycemic phenotypes in hypertension.

The consistent performance of eGDR constitutes the central finding. By combining waist circumference, HbA1c, and hypertension status, eGDR captures three major risk areas without needing fasting insulin. Mechanistically, eGDR correlates more strongly with myocardial glucose metabolic rate measured by positron emission tomography with 18 F-fluorodeoxyglucose (PET-FDG) than with HOMA-IR, and it independently predicts myocardial mechano-energetic efficiency [20]. Previous analyses using NHANES data have established eGDR as an independent predictor of ACM and CVM in cardiometabolic syndrome [38] and metabolic dysfunction-associated steatotic liver disease [39]. At the same time, lower eGDR has been associated with higher mortality among older non-diabetic adults [40]. In the Multi-Ethnic Study of Atherosclerosis (MESA) cohort, each 1-SD increase in eGDR was linked to a 30% reduction in atherosclerotic CVD (ASCVD) risk [41]. Longitudinal data from the China Health and Retirement Longitudinal Study (CHARLS) showed that sustained low eGDR increased the risk of new-onset CVD by 2.5 times [42], with each 1-SD increase leading to a 17–19% decrease in CVD risk [22, 43]. Our findings build on this evidence by showing that eGDR was the only non-insulin-based index that demonstrated both per-unit and Q4 associations with ACM in prediabetes and diabetes. It was also the only index that demonstrated both per-unit and consistent quartile-level associations with CVM in the diabetic subgroup. In prediabetes, eGDR showed significant quartile-level protection for CVM, although a per-unit association was not observed. In the diabetic subgroup, although AIP Q2 achieved nominal significance, no other index demonstrated consistent CVM associations comparable to eGDR.

The J-shaped HOMA-IR dose-response, with neutral-to-protective intermediate quartiles and increased risk at the extremes, mirrors the previously observed U-shaped relationship in CHD-hypertension cohorts [4]. HOMA-IR was confirmed as an independent predictor of cardiovascular events in hypertensive non-diabetic adults [17] and has also been associated with hypertension across diabetic and non-diabetic populations [44]. Beyond mortality outcomes, prior evidence also suggests that the coexistence of hypertension and HOMA-IR-defined IR markedly increases future diabetes risk [45]. The McAuley index showed no independent mortality prediction after full adjustment, aligning with clamp-based validation data [46], despite population-specific associations including reduced CHD risk in Black Americans [47] and heightened cancer mortality in a prospective cohort of men and women [48]. Mechanistically, the J-shaped pattern suggests that extreme HOMA-IR values reflect hepatic lipid overproduction and systemic inflammation that are not fully captured by covariate adjustment, whereas moderate hyperinsulinemia indicates adaptive hepatic physiology; very low HOMA-IR in hypertensive adults may reflect impaired insulin secretion rather than true sensitivity. QUICKI’s quartile-specific significance, without a per-unit gradient, indicates a threshold effect, suggesting that it detects clinically relevant IR thresholds not captured by linear analysis, with important implications for categorical risk classification.

The dissociation of SHR, which showed the highest per-unit ACM estimate among all indices, yet no significant per-unit association with CVM, appears to be less well characterized in prior literature. SHR has shown broader prognostic relevance: predicting incident hypertension via lipid-mediated pathways [31], adverse outcomes in chronic thromboembolic pulmonary hypertension [32], and stroke risk with synergistic hypertension interactions nearing a threefold increase [33]. Our finding that SHR associates with ACM but not CVM suggests that SHR encodes non-atherosclerotic mortality pathways, possibly infection-related, neoplastic, or respiratory, rather than atherogenic IR itself, requiring outcome-specific clinical interpretation. It should be noted that the RCS analysis did reveal a significant overall CVM association for SHR, attributable to the significant protective signal at Q2; when protective and null effects across different SHR value ranges are averaged into a single per-unit estimate, the opposing directions cancel out and statistical significance is lost, underscoring the importance of interpreting nonlinear indices beyond their per-unit hazard ratios. The female-specific SHR association suggests possible sex differences in the prognostic relevance of relative hyperglycemia, although the underlying mechanisms require further investigation. The amplified associations of TyG with diabetic subgroups align with the glucotoxicity-lipotoxicity synergy that drives accelerated atherogenesis, as confirmed by machine-learning analyses [18], with the highest combined CVM risk observed when elevated TyG coincides with hypertension [19]. TyG mediated 38–42% of BMI-ACM associations in diabetes-hypertension cohorts [49], and our stratified analyses showed that TyG was significantly associated with ACM only in the diabetes subgroup, consistent with a stronger prognostic role of glucolipotoxicity in established diabetes. METS-IR’s significant per-unit associations with ACM and CVM extend prior evidence linking METS-IR to cardiometabolic and cardiovascular risk into a hypertension-specific mortality setting [24–27]. The non-monotonic quartile pattern, with Q2 exhibiting a protective association yet Q4 failing to reach significance despite a harmful per-unit trend, is consistent with the significant nonlinear dose-response confirmed by RCS analysis, and likely reflects residual age- and BMI-related confounding across the METS-IR distribution that persists even after covariate adjustment. LAP demonstrated age-adjusted predictive performance for hypertension [29], associations with arterial stiffness in hypertensive adults [30], non-linear relationships between hypertension and hyperuricemia [50], and sex-specific performance [51]; however, its signal diminished after full adjustment, likely captured by included covariates. AIP was shown to predict metabolic outcomes over 9 years in a Taiwanese study [52], similar to CMI [53]; its predictive signal weakened after full adjustment, likely reflecting collinearity with the lipid covariates included in Model 3.

The glycemic stratum-specific findings reflect the evolving tissue-specific pathophysiology of IR across the cardiometabolic continuum. In normoglycemia, the exclusive per-unit significance of HOMA-IR aligns with the primary role of hepatic insulin resistance as the earliest detectable IR defect, which promotes endothelial dysfunction, dyslipidemia, and low-grade inflammation through hepatic glucose overproduction and compensatory hyperinsulinemia [5, 6]. Indeed, elevated HOMA-IR is associated with subclinical multiterritorial atherosclerosis in individuals with entirely normal HbA1c, independently of conventional cardiovascular risk factors [54], which mechanistically explains why hepatic IR confers a measurable mortality risk even before overt glucose dysregulation. At the glycemic transition to prediabetes, the emergence of both HOMA-IR and eGDR as independent ACM predictors, along with broad convergence of CVM-protective signals across multiple indices, reflects a pathophysiological inflection point in which skeletal muscle IR becomes increasingly dominant as a metabolic defect. Skeletal muscle accounts for approximately 80% of insulin-stimulated glucose disposal, and insulin resistance in this tissue is established as the primary defect driving progression of prediabetes and associated cardiovascular risk [55, 56]. The wide-ranging CVM protection observed across both index categories in prediabetes suggests that any meaningful improvement in systemic insulin sensitivity at this stage translates into cardiovascular benefit, underscoring the potential for early intervention. In established diabetes, the coexistence of hepatic (HOMA-IR), glucolipotoxicity (TyG), and skeletal muscle (eGDR) IR signals as independent ACM predictors reflects the multidimensional, tissue-redundant nature of insulin resistance at this stage. The unique CVM signals of TyG, CMI, and AIP exclusively in the diabetic subgroup support the idea that atherogenic dyslipidemia becomes a leading cardiovascular driver only after chronic hyperglycemia is established, intensifying lipid peroxidation, platelet activation, endothelial dysfunction, and plaque vulnerability through the glucolipotoxicity mechanism that underpins TyG’s diabetes-specific prognostic signal [57].

The stronger associations observed in participants under 65 years old, with eGDR showing approximately a 35% reduction in mortality risk per unit in younger versus only 13% in older adults, and TyG being significant only in the younger group, suggest that competing non-IR mortality pathways dilute these signals at older ages [40, 58]. This emphasizes the clinical importance of quantifying IR in younger hypertensive patients. This age-related attenuation is similar to that observed in the Taiwanese longitudinal study, where AIP predicted metabolic outcomes in 40-64-year-olds but lost significance above age 65 [52], aligning with LAP’s age-adjusted cutoff values for hypertension prediction [29]. The sex-stratified results show consistent associations of eGDR, HOMA-IR, and TyG in both sexes, supporting sex-independent pathophysiological processes. The female-specific signal for SHR has precedent: the HOMA-IR-CHD association was significant in men but not women in the Jackson Heart Study [47]; METS-IR showed sex-specific effects in Chinese adults [59], and LAP demonstrated sex-specific predictive performance in the elderly [51]. These findings highlight the importance of sex-stratified analysis and position eGDR as the preferred sex-agnostic clinical surrogate, with comparable protective associations in both sexes.

Clinically, eGDR, which only requires waist circumference, HbA1c, and hypertension status, appears to be the preferred non-insulin-based surrogate for IR-related mortality risk stratification, particularly among hypertensive adults with prediabetes or diabetes. It is especially valuable in resource-limited settings where fasting insulin measurement is unavailable. HOMA-IR remains informative when fasting insulin data are available because of its unique per-unit ACM significance across all three glycemic strata; its J-shaped dose-response curve may provide additional insight into risk stratification. TyG provides additional prognostic value, particularly in diabetes, by predicting both ACM and CVM, supporting its use as a lipid-metabolic risk marker when glucolipotoxicity is suspected. The finding that SHR predicts ACM but not CVM indicates it should not be used solely as a cardiovascular risk marker; its female-specific nature also warrants sex-stratified clinical validation. The age-dependent decline highlights the importance of early metabolic assessment in younger hypertensive patients, where IR-related risk is most modifiable through targeted lifestyle and pharmacological strategies.

Several limitations deserve consideration. The observational design prevents causal inference, and residual confounding from unmeasured variables such as physical activity, dietary patterns, glucose- and lipid-lowering medications, antihypertensive adherence, and socioeconomic factors cannot be ruled out. Several covariate choices merit specific acknowledgment. Blood pressure and antihypertensive medication use were not included in the primary Model 3 because they may partly function as downstream intermediates linking IR to mortality, and their inclusion could therefore introduce potential over-adjustment when estimating the total IR-mortality association. Nevertheless, these variables may also contribute to residual confounding, and future sensitivity analyses incorporating blood pressure levels and antihypertensive medication use are warranted. Glycemic status was used as a stratification variable rather than a model covariate in the overall analysis, which may introduce confounding if glycemic categories differ in both IR levels and baseline mortality risk. The poverty-income ratio was used as an exclusion criterion for covariate completeness but was not included as a model covariate, potentially leaving residual socioeconomic confounding. Physical activity, dietary patterns, educational attainment, and glucose- and lipid-lowering medications are available in NHANES but were not incorporated into Model 3; this reflects an analytical choice rather than data unavailability and may introduce additional residual confounding. Blood pressure thresholds differ across guidelines; our cohort used the ≥ 140/90 mmHg threshold consistent with the 2024 European Society of Cardiology (ESC) guideline [36]. NHANES sampling weights were not applied because our primary goal was to examine exposure-outcome relationships rather than produce nationally representative prevalence estimates. IR indices derived from a single fasting sample may underestimate associations, as longitudinal eGDR monitoring has shown additional predictive value [42, 60]. The analytic cohort was restricted to participants in the NHANES fasting sub-sample, which may have selectively included healthier individuals relative to the full hypertensive population; however, the glycemic gradient in mortality risk was preserved across all exclusion stages, suggesting that this selection did not materially distort the primary exposure-outcome relationships of interest. Fasting-morning restriction may introduce selection bias, and variability in insulin assays between laboratories could impact the reproducibility of HOMA-IR, McAuley, and QUICKI. Formal competing risks analyses using Fine-Gray subdistribution hazard models were not performed because non-cardiovascular deaths accounted for 71.5% of all deaths, standard Cox estimates for cardiovascular mortality may overestimate the subdistribution hazard in the presence of competing events, and future analyses incorporating Fine-Gray models are warranted. NDI-linked mortality data through December 31, 2019, might underestimate event rates, and ICD-10-based classification could misclassify cardiovascular deaths. The prediabetes classification in this study followed the ADA criterion of fasting blood glucose ≥ 100 mg/dL (5.6 mmol/L), consistent with standard NHANES analytical conventions. It should be noted that the World Health Organization (WHO) defines impaired fasting glucose using a higher threshold of ≥ 110 mg/dL (6.1 mmol/L), which would classify a smaller proportion of participants as prediabetic; findings may not be fully generalizable to studies that use the WHO definition. Lastly, race/ethnicity-specific eGDR calibration requires further validation, and findings from U.S. hypertensive adults may not generalize to other populations or healthcare settings.

Conclusions

In hypertensive adults, fasting-insulin-based and non-insulin-based IR surrogates provide complementary rather than interchangeable prognostic information. HOMA-IR was the only fasting-insulin-based index independently associated with ACM across all three glycemic strata. eGDR demonstrated the most consistent non-insulin-based associations with ACM and CVM, particularly in prediabetes and diabetes. TyG provided additional prognostic value for ACM and CVM, particularly in patients with diabetes. These findings support a phenotype-informed approach to IR index selection guided by glycemic status and data availability.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

F.A.A. was responsible for the study design, provided statistical support, analyzed the data, drafted the manuscript, and was primarily responsible for the final content. A.Q.M., W.Z., J.A., L.L., and G.Y. were involved in preparing and analyzing the data and in visualizing the results. F.A.A., Y.L., A.Q.M., W.Z., J.A., L.L., and G.Y. independently replicated and validated the statistical analyses. F.A.A., Y.L., G.Z., and W.C. supervised the study, acquired funding, and contributed to manuscript revisions. All authors actively participated in the research process, made substantial contributions to manuscript revisions, and carefully reviewed and approved the final version.

Funding

This work was supported in part by National Natural Science Foundation of China, Research Fund for International Scientists (W2433190), National Natural Science Foundation of China (82570595,82500403), Shanghai Hospital Development Center, Clinical Technologies Promotion and Optimization Management Project (SHDC12024120), Tibet Natural Science Foundation of China (XZ2022ZR-ZY27(Z), XZ202301ZR0032G), Shanghai Municipal Health Commission, 2025 Integrated Traditional Chinese and Western Medicine Collaborative Pilot Program for General Hospitals (ZHYYZXYYD-202505), Tongji University, “Medicine + X” Interdisciplinary Research Program (2025-0553-ZD-02).

Data availability

NHANES data are publicly available at https://www.cdc.gov/nchs/nhanes/. Analytical code is available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The NCHS Research Ethics Review Board approved all NHANES protocols. Written informed consent was obtained from all participants. This analysis of de-identified, publicly available data required no additional institutional review.

Competing interests

The authors declare no competing interests.