NHANES utilizes a multi-stage probability-based sampling methodology to yield a nationally representative sample of the non-institutionalized U.S. civilian population. Since 1999, NHANES has operated in two-year cycles, collecting comprehensive data through interviews, physical examinations, and laboratory analyses. Detailed study design information is publicly available^[21]. A total of 41,474 subjects participated in NHANES between 1999 and 2006. After excluding 18,850 participants aged less than 18, 4,498 with missing hypertension, diabetes, and heart rate data, 835 missing AC, TC, and WC data, and 15 lacking mortality data, we ended up with a final of 17,276 individuals [Supplementary Figure 1]. The use of de-identified, publicly available NHANES data waived the requirement for Institutional Review Board review.

Trained healthcare technicians measured right arm circumference (AC), right thigh circumference (TC), and WC following standardized protocols. The left limb was measured if the right was unavailable. For AC, participants stood with arms naturally hanging, elbows bent at 90°, and palms facing upward. AC was measured at the midpoint between the acromion's posterior edge and the olecranon process (tip of the elbow), with the tape positioned perpendicular to the long axis of the upper arm. For TC, participants first sat with knees bent at 90°, and a mark was made at the midpoint between the inguinal crease and the proximal edge of the patella. Participants then stood, shifted their weight to the left leg, slightly flexed the right leg forward, and TC was measured perpendicular to the long axis at the marked point. WC was measured while standing by marking a horizontal line just above the right iliac crest and measuring WC along this line. All measurements were recorded to the nearest millimeter. The sum of arm and thigh circumference (ATC) was defined as the sum of unilateral AC and TC, and ATC/WC was calculated as the ratio of ATC to WC.

Using a unique identifier, each participant was linked to the National Death Index, with follow-up for mortality available through December 31, 2019^[22]. Mortality endpoints included death from any cause (all-cause mortality) and death from cardiovascular diseases, as determined by the following ICD-10 codes: I00-I09, I11, I13, and I20-I51.

Blood pressure (BP), blood samples, resting heart rate measurements, and related questionnaire data were collected at baseline. BP was measured at the Mobile Examination Center (MEC) using a mercury sphygmomanometer. After a 5-min seated rest, three consecutive readings of systolic and diastolic BP were obtained after determining the maximum inflation level; a fourth reading was taken if any measurement was unsuccessful. The final BP value was the average of the valid readings. Hypertension was defined as systolic BP exceeding 140 mmHg, diastolic BP exceeding 90 mmHg, and/or a self-reported physician diagnosis. Blood samples were rapidly centrifuged after collection at the MEC and sent to Johns Hopkins Hospital (Baltimore, MD), where fasting glucose and hemoglobin A1c (HbA1c) levels were determined using a Hitachi automated analyzer. Diabetes was defined as fasting blood glucose over 7 mmol/L, HbA1c above 6.5%, and/or self-reported diagnosis, use of hypoglycemic medications or insulin therapy. Sample processing and assays adhered to NHANES standardized protocols^[23]. Resting heart rate was determined by counting the beats in 30 s and multiplying by two. According to the ESC Tachycardia Management Guidelines, resting tachycardia was defined as a heart rate greater than 100 beats per minute^[24].

Covariates were selected based on established clinical and epidemiological relevance, and were categorized into three domains. (1) Sociodemographic factors: age (categorized as 18-30, 30-42, 42-54, 54-66, > 66 years), sex, race/ethnicity, education level, marital status, and family income-to-poverty ratio (< 1 or ≥ 1); (2) Lifestyle factors: smoking and alcohol consumption status; (3) Physical health indicators: body mass index (BMI), serum creatinine, urine albumin-to-creatinine ratio (uACR), and history of underlying diseases. Participants were considered to have a history of underlying diseases if they reported a prior diagnosis of bronchitis, asthma, coronary heart disease, congestive heart failure, stroke, or malignant tumors.

Analyses accounted for the NHANES complex survey design (weights, clustering, stratification) and were performed in R 4.5.1. Missing covariate data were imputed via random forest. Continuous variables are summarized as weighted mean (standard deviation), and categorical variables as weighted frequency (percentage). Group comparisons utilized the weighted Kruskal-Wallis test for continuous variables and the weighted chi-square test for categorical variables. A two-sided P-value < 0.05 was deemed statistically significant.

To evaluate survival patterns, Kaplan-Meier curves were generated for ATC/WC and ATC across age subgroups, stratified by all-cause and cardiovascular mortality. To assess mortality risk, Cox proportional hazards models were fitted. The associations of ATC/WC and ATC with hypertension, diabetes, and resting tachycardia were examined using generalized linear models. We employed a multi-model adjustment strategy. Model 1 was unadjusted (crude); Model 2 adjusted for basic sociodemographic and lifestyle factors, including smoking and alcohol consumption; Model 3 further adjusted for clinical measures and comorbidities, including serum creatinine, uACR, dyslipidemia, and a history of underlying diseases—and, for mortality outcomes, additionally for baseline hypertension, diabetes, and tachycardia; and Model 4 additionally adjusted for BMI (fully adjusted). For ATC, a sensitivity analysis (Model 5) replaced adjustment for BMI with adjustment for WC. In Model 4, potential nonlinearity was explored using restricted cubic splines with three knots.

We conducted an extensive set of sensitivity analyses. First, we performed subgroup analyses based on age, sex, hypertension, diabetes, smoking status, and history of underlying diseases to assess how ATC/WC and ATC were related to both all-cause and cardiovascular mortality, as well as to CVRFs. Second, we substituted measured BP values for the hypertension diagnosis variable, fasting glucose measurements for the diabetes diagnosis variable, and resting heart rate measurements for the resting tachycardia variable, utilizing linear regression to assess the correlations between ATC/WC, ATC, and CVRFs—to mitigate potential misclassification and validate our primary findings. Finally, to deconstruct the composite metrics, we evaluated the individual components: we examined the associations of unilateral AC and TC separately with mortality and CVRF, and we also constructed and tested arm-to-waist (AC/WC) and thigh-to-waist (TC/WC) ratios.

This study comprised a final cohort of 17,276 individuals. According to weighted quartiles of ATC/WC (cut points: 0.839, 0.902, 0.965), participants were categorized into four groups: Q1 (n = 5,213), Q2 (n = 4,112), Q3 (n = 3,776), and Q4 (n = 4,175). Over a median follow-up of 18.04 years (totaling 358,477 person-years), 3,927 deaths occurred, including 1,047 cardiovascular deaths. The cohort had a mean age of 46.08 (±20.12) years and was 52% female. As detailed in Table 1, when contrasting baseline characteristics, individuals in Q4 were generally younger, had higher educational attainment, and lower smoking prevalence than those in Q1. They also presented with more favorable anthropometric and clinical profiles, including lower values for WC, weight, BMI, serum creatinine, total cholesterol, BP, and resting heart rate. Moreover, the prevalence of multiple comorbidities—including proteinuria, hypertension, diabetes, dyslipidemia, chronic bronchitis, coronary heart disease, heart failure, stroke, and cancer—was lower in Q4 relative to Q1.

Kaplan-Meier analyses stratified by age indicated protective associations for both ATC/WC and ATC with mortality, though the associations emerged in different age groups [Supplementary Figures 2 and 3]. This study evaluated the association between the ATC/WC ratio and mortality using multivariable-adjusted Cox proportional hazards models, confirming significant and independent inverse associations (see Table 2 for details). In Model 3, which did not adjust for BMI or waist circumference, participants in the highest quartile (Q4) of ATC/WC, compared to those in the lowest quartile (Q1), showed a significantly lower risk of all-cause mortality, with a hazard ratio (HR) of 0.49 (95% confidence interval [CI]: 0.40-0.60). The reduction in cardiovascular mortality risk was more pronounced, with an HR of 0.33 (95% confidence interval (95%CI): 0.22-0.49). Notably, the strength of these protective associations remained virtually unchanged after further adjustment for BMI in Model 4 (all-cause mortality: HR = 0.47, 95%CI: 0.38-0.57; cardiovascular mortality: HR = 0.32, 95%CI: 0.21-0.49), demonstrating that ATC/WC effectively accounts for adiposity.

In contrast, the associations for ATC were confounded by overall adiposity. In model 3 (without BMI/WC adjustment), a higher ATC (Q4 vs. Q1) was associated with only a modest reduction in all-cause mortality risk (HR = 0.68, 95%CI: 0.60-0.78), while its association with cardiovascular mortality did not reach statistical significance (HR = 0.80, 95%CI: 0.62-1.04). However, after adjusting for BMI in model 4, the associations strengthened substantially, becoming significantly inverse for both all-cause (HR = 0.47, 95%CI: 0.38-0.57) and cardiovascular mortality (HR = 0.51, 95%CI: 0.34-0.76). This reversal underscores that the true protective effect of limb circumference is masked by its correlation with general obesity and requires control for adiposity.

RCS analyses visually supported these associations. For the ATC/WC, the relationship with mortality was significantly linear. The P-value for non-linearity was 0.255 for all-cause [Figure 1A] and 0.171 for cardiovascular mortality [Figure 1B]. The mortality risk decreased steadily as ATC/WC increased, with a threshold of notable protection above a value of 0.835. In contrast, the association between ATC and mortality was significantly non-linear (P for non-linearity < 0.001), as detailed in Supplementary Figure 4.

The prevalence of hypertension [Figure 2A], diabetes [Figure 2B], and resting tachycardia [Figure 2C] decreased significantly with increasing quartiles of ATC/WC (all P < 0.001). In contrast, while ATC quartiles were significantly associated with the prevalence of hypertension [Figure 2D] and diabetes [Figure 2E] (P < 0.001), the relationships appeared nonlinear. No significant association was observed between ATC and the prevalence of resting tachycardia (Figure 2F, P = 0.478).

In logistic regression of model 3, participants in Q4 of ATC/WC had substantially lower odds of hypertension (odds ratio (OR) = 0.36, 95%CI: 0.30-0.42), diabetes (OR = 0.19, 95%CI: 0.13-0.26), and resting tachycardia (OR = 0.25, 95%CI: 0.18-0.37) compared to Q1. Critically, these robust protective associations persisted after further adjustment for BMI in model 4, with ORs of 0.49 (95%CI: 0.41-0.59) for hypertension, 0.26 (95%CI: 0.18-0.37) for diabetes, and 0.28 (95%CI: 0.19-0.39) for resting tachycardia [Table 3], underscoring the ability of ATC/WC to effectively account for adiposity.

In contrast, the associations for ATC alone were confounded by overall adiposity. Before adjustment for BMI or WC (model 3), a larger ATC (Q4 vs. Q1) was paradoxically associated with higher odds of hypertension (OR = 2.70, 95%CI: 2.37-3.07) and diabetes (OR = 2.95, 95%CI: 2.40-3.62), and showed a non-significant trend for resting tachycardia (OR = 1.23, 95%CI: 0.84-1.78). However, after adjusting for BMI (model 4), these adverse associations were not only eliminated but reversed, revealing significant protective effects for hypertension (OR = 0.74, 95%CI: 0.59-0.93) and diabetes (OR = 0.34, 95%CI: 0.24-0.48), and a non-significant protective trend for tachycardia (OR = 0.65, 95%CI: 0.37-1.14). Sensitivity analyses adjusting for WC yielded consistent results. This pattern unequivocally demonstrates that the true protective association of limb circumference is masked by its correlation with overall adiposity.

For ATC/WC, a linear inverse association was observed for hypertension (p non-linear  =  0.89), as shown in Figure 3A, while its relationships with diabetes (p non-linear  =  0.004) and resting tachycardia (p non-linear  =  0.026) exhibited significant non-linear components, as shown in Figure 3B and C. Protective effects appear when ATC/WC exceeds 0.89. ATC showed a linear correlation with hypertension and a nonlinear correlation with diabetes and resting tachycardia [Supplementary Figure 5].

The robustness of the primary findings and potential underlying mechanisms were evaluated through a set of sensitivity analyses.

First, we confirmed the fundamental source of confounding for ATC by demonstrating its significant, linear positive correlations with both BMI and WC [Supplementary Figure 6]. Consistently, higher ATC was linked to greater odds of both general and abdominal obesity [Supplementary Table 1], directly explaining its paradoxical associations before adiposity adjustment.

The protective associations of ATC/WC with mortality and CVRFs remained consistent across all pre-specified subgroups (stratified by sex, age, hypertension, diabetes, smoking status, and history of underlying diseases; Supplementary Figures 7-10), with generally more pronounced effects in males and older adults. In contrast, the association of ATC with outcomes was modified by certain health conditions, further underscoring its instability without considering central adiposity.

The robustness of the findings was further corroborated using continuous measures of cardiovascular risk. Furthermore, in models with full adjustment, ATC/WC showed significant inverse correlations with systolic and diastolic BP, resting heart rate, and fasting blood glucose, and these associations persisted after additional adjustment for BMI [Supplementary Table 2]. Replicating the pattern seen with dichotomous outcomes, ATC was positively associated with these continuous measures in models unadjusted for BMI, but these associations reversed direction after BMI adjustment.

Finally, deconstructing the composite metrics confirmed the generalizability of the central finding. Both the arm-to-waist (AC/WC) and thigh-to-waist (TC/WC) ratios were significantly inversely associated with all mortality and CVRF outcomes [Supplementary Tables 3-7]. Similarly, the individual circumferences (AC and TC) alone were inversely associated with mortality [Supplementary Tables 8 and 9]. Most critically, for CVRFs, the isolated arm (AC) and thigh (TC) circumferences recapitulated the identical paradoxical pattern observed for ATC: they were positively associated with hypertension, diabetes and resting tachycardia before adiposity adjustment, but these associations were attenuated, nullified, or reversed after controlling for BMI or WC [Supplementary Tables 10-12]. This confirms that the confounding effect of overall adiposity is a universal property of limb circumference measures, which is effectively resolved by forming a ratio with WC.

Collectively, these comprehensive sensitivity analyses solidify the premise that ATC/WC is a robust, adiposity-independent indicator of metabolic health, thereby resolving the confounding that inherently plagues the interpretation of limb circumferences alone.