Study design and setting

A cross-sectional observational study was conducted between 13/07/2025–14/11/2025 across outpatient departments of the Institute of Medicine (IOM), Tribhuvan University Teaching Hospital (TUTH), Kathmandu, Nepal, including the Manmohan Cardiothoracic Vascular and Transplant Centre (MCVTC) and the General Practice (GP) Outpatient Department. MCVTC was selected because of its high patient flow with cardiovascular and metabolic comorbidities, including diabetes, hypertension, dyslipidemia, and non-alcoholic fatty liver disease, making it highly relevant for a study focused on cardiometabolic risk prediction. The GP OPD was additionally included as it provides a standardized general health check-up package, ensuring availability of complete anthropometric and biochemical data required for the study objectives.

Ethical approval

Ethical approval was obtained from the Institutional Review Committee (IRC) of the Institute of Medicine (IOM), Maharajgunj, Kathmandu [Reference number: 717 (6–11) E2 081/082].

Study population and sampling

Patients visiting the outpatient departments of MCVTC and the General Practice OPD of TUTH who had undergone recent clinical evaluations were included. Anthropometric measurements taken during these visits served as the baseline reference point. Corresponding biochemical and radiological investigations were retrospectively extracted from medical records if they were documented within three-month window of clinical evaluation. A non-probability convenience sampling method was used for participant selection.

Sample size was estimated using ROC analysis methodology, aiming to detect a difference in AUC of 0.10 between BMI and WC, with a power of 80% and 5% significance level. Assuming a prevalence of hypertension of 20% [13,14] and adjusting for up to 4 predictors in logistic regression (BMI, WC, age, sex), the minimum required sample size is estimated to be approximately 400–450 participants. This accounts for adequate statistical power to compare predictive strength and control for confounding. To increase precision and statistical power, we have exceeded the sample size and taken 678 participants.

The formula used is:

Where:

- Z₁-α/2 = 1.96 (for 95% confidence)

- Z₁-β = 0.84 (for 80% power)

- AUC₁ - AUC₀ = 0.10 (detectable difference in AUCs)

- p = 0.2 (estimated prevalence of outcome)

Inclusion criteria: Only those patients of age ≥ 18 years, whose complete biochemical and anthropometric data were available and who consent to participate were included.

Exclusion criteria: Pregnant women and patients with known history of malignancy or chronic liver disease were excluded.

Data collection

Written informed consent was taken from all participants. The socio-demographic data and behavioral information (smoking, alcohol intake, diet) were collected through in-person questionnaires.

This study was purely observational and non-interventional. No new diagnoses of hypertension, diabetes mellitus, or thyroid disease were made during the study period. Participants were classified as having these conditions solely based on documented prior diagnoses in their medical records and/or current use of relevant medications. Anthropometric and biochemical measurements obtained during routine clinical evaluation were not used to establish new clinical diagnoses.

The presence of comorbidities was documented based on medical history of diabetes, hypertension and thyroid disease from the patient’s medical records. The biochemical values and radiological findings were taken retrospectively from patient’s medical records which were done within 3 months of the date of the primary clinical encounter where anthropometric data were recorded.

Anthropometric measurements

Height was measured using a measuring tape and recorded to the nearest 0.1 cm. Participants stood barefoot in an upright position with their back against the wall, heels together, and head maintained in the Frankfort horizontal plane.

Weight was measured using a calibrated spring balance placed on a stable, level surface and recorded to the nearest 1 kg. Participants wore light clothing during measurement.

Body mass index (BMI) was calculated as weight (kg) divided by height squared (m²). Waist circumference (WC) was measured using a non-stretchable measuring tape with participants standing upright in a relaxed posture and feet together. Measurements were taken midway between the lower rib margin and the iliac crest at the end of normal expiration.

In addition to BMI and WC, several derived anthropometric indices were evaluated, including waist-to-height ratio (WtHR), waist-to-BMI ratio (WBMIR), Body Roundness Index (BRI), A Body Shape Index (ABSI), and body fat percentage. These indices were included because they have been proposed as alternative measures of central adiposity and cardiometabolic risk and may provide improved risk discrimination compared with traditional anthropometric measures and may be valuable for South Asian population. BRI estimates body roundness and has been associated with visceral adiposity, whereas ABSI standardizes waist circumference relative to height and BMI to quantify abdominal adiposity independent of overall body size.

BRI and ABSI were calculated according to the following formulas:

Biochemical measurements

Biochemical parameters including fasting plasma glucose, lipid profile (total cholesterol, triglycerides, HDL-cholesterol, LDL-cholesterol), liver function test (alanine aminotransferase), renal function tests (urea and creatinine), thyroid function tests (triiodothyronine, tetraiodothyronine, and thyroid-stimulating hormone) and serum uric acid were obtained from hospital laboratory records. All investigations were performed as part of routine clinical evaluation using standard automated analyzers in the central laboratory of Tribhuvan University Teaching Hospital. For study inclusion, laboratory values measured within three months of the clinical visit were considered. As this study involved a retrospective review of existing clinical records, the biochemical results analyzed were those previously reported by the central laboratory in accordance with its established standard operating procedures. Fasting venous blood samples (minimum 8–12 h fast) were collected in the morning in specialized vacutainers and processed within three hours of collection. All assays were performed in automated analyzers (Cobas series, Roche Diagnostics) in the ISO-certified central laboratory of TUTH, which participates in an external quality assurance program including External Quality Assessment Services (EQAS) such as Christian Medical College Vellore, India. Fasting plasma glucose was analyzed using the enzymatic hexokinase method. Lipid parameters, including total cholesterol, triglycerides, HDL-cholesterol, and LDL-cholesterol, were assessed using enzymatic colorimetric assays. Liver enzyme alanine aminotransferase (ALT) was measured by the IFCC-recommended kinetic UV method, while thyroid function tests (T3, T4, and TSH) were evaluated using electrochemiluminescence immunoassay (ECLIA). Serum uric acid was determined using the uricase-PAP method.

Radiological assessment

Fatty liver disease was assessed using abdominal ultrasonography as part of routine clinical evaluation. Ultrasonographic findings documented in patients’ medical records were reviewed, and fatty liver was diagnosed based on standard sonographic criteria. Where available, fatty liver was graded as mild (grade 1), moderate (grade 2), or severe (grade 3) according to the radiologist’s report. Consistent with the approach applied to biochemical data, only ultrasonography reports documented within three months of the clinical visit (i.e., the date of anthropometric measurement) were included in hepatic steatosis classification. Sonographic diagnosis of fatty liver was based on standard B-mode ultra-sonographic criteria: increased hepatic echogenicity relative to the renal cortex and posterior attenuation of acoustic beam [15] as interpreted by a qualified radiologist.

Cardiometabolic risk outcomes

The cardiometabolic risk outcomes evaluated in this study included hypertension, diabetes mellitus, dyslipidemia, fatty liver disease (FLD), non-alcoholic fatty liver disease (NAFLD), and thyroid disease.

A condition-specific classification approach was adopted for cardiometabolic outcomes. Dyslipidemia was operationally defined as a binary variable derived from NCEP ATP III lipid thresholds, where the presence of any abnormal lipid parameter (LDL-C ≥ 100 mg/dL, Total Cholesterol ≥ 200 mg/dL, Triglycerides ≥ 150 mg/dL, or HDL-C < 40 mg/dL or males and <50 mg/dL for females) was classified as dyslipidemia [16]. This threshold is widely used in epidemiological and clinical studies for its explicit laboratory cut-offs suitable for population-based classification rather than guidelines designed for treating hyperlipidemia. Diabetes mellitus, hypertension, and thyroid disease were defined based on documented prior physician diagnosis or current pharmacological treatment. This was done because diabetes mellitus, hypertension, and thyroid disease are clinical diagnoses that cannot be established from a single laboratory value alone. The diagnostic criteria for each as per the American Diabetes Association, the American Heart Association, and the American Thyroid Association respectively require either repeat biochemical confirmation, oral glucose tolerance testing, HbA1c correlation [17], multiple blood pressure measurements across separate visits [18,19], or clinical symptom correlation before a diagnosis can be assigned [20,21]. Applying a single retrieved fasting glucose, blood pressure reading, or TSH value from a retrospective record to classify participants as diseased or non-diseased would therefore not satisfy the diagnostic requirements for these conditions and would introduce significant misclassification.

Fatty liver disease was identified based on ultrasonographic evidence documented in radiology reports. NAFLD was defined as ultrasonographically detected fatty liver in the absence of significant alcohol consumption or secondary causes, based on available clinical documentation. Ultrasound findings of within-three months was used for the diagnosis.

Statistical analysis

Statistical analyses were performed using RStudio (version 2024.12.0 + 467). Descriptive statistics were generated for all continuous and categorical variables. Normality was assessed using the Shapiro–Wilk test. As continuous variables were not normally distributed, data were summarized as median and interquartile range (IQR; Q1–Q3), while categorical variables were reported as frequencies and percentages. Statistical significance was defined as a two-sided p-value < 0.05.

Associations between anthropometric measures, including body mass index (BMI) and waist circumference (WC), and blood parameters were evaluated using Spearman’s rank correlation coefficients. Simple linear regression analyses were subsequently performed to further characterize significant associations and assess the direction and magnitude of linear relationships.

Participants were stratified according to cardiometabolic risk groups, including hypertension (HTN), diabetes mellitus (DM), thyroid disease, fatty liver disease (FLD), non-alcoholic fatty liver disease (NAFLD), and dyslipidemia. Differences in BMI, WC, and blood parameters between two-group comparisons were assessed using the Wilcoxon rank-sum test, whereas comparisons involving more than two groups were evaluated using the Kruskal–Wallis test.

Binary logistic regression analyses were conducted to evaluate the associations of BMI and WC with cardiometabolic outcomes. Results were reported as odds ratios (ORs) with 95% confidence intervals (CIs). Multivariable logistic regression models were adjusted for age, sex, smoking status, alcohol consumption, and dietary factors to account for potential confounding.

The predictive performance of BMI and WC for cardiometabolic outcomes was assessed using receiver operating characteristic (ROC) curve analysis and quantified by the area under the curve (AUC). ROC analyses were performed for anthropometric measures alone and for models incorporating significant covariates. Optimal cut-off values were determined using Youden’s J statistic.

To determine whether alternative obesity-related indices provided superior predictive performance WtHR, WBMIR, BRI, ABSI, and body fat percentage were evaluated using the same logistic regression and ROC analysis framework. The predictive abilities of these indices were compared with those of BMI and WC across all cardiometabolic outcomes.

Patient demographics and descriptives

There were altogether 678 patients comprised on 292 males and 386 females. All continuous variables were non-normal in distribution. The median age was 45 (Q1: 33, Q3: 55). Out of them 97 were smokers, 131 consumed alcohol, and 646 followed a non-vegetarian diet. The prevalence of hypertension, diabetes mellitus, and thyroid disease were 15.7%, 6.2%, and 8.7% respectively. Table 1 summarizes the demographics and descriptive statistics. No significant difference in the waist circumference between males and females (p = 0.136)

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Patient demographics and descriptives statistics of BMI, Waist, Age, and blood parameters.

https://doi.org/10.1371/journal.pone.0353676.t001

Correlation of BMI and waist circumference (WC) with blood parameters

BMI and WC were highly correlated (rho = 0.75). WC had a higher correlation with age than BMI (rho = 0.28 vs. 0.13). With glucose, there was a weak positive correlation of both BMI (rho = 0.13) and WC (rho = 0.12). Both BMI and WC correlated significantly with lipid parameters: triglyceride, LDL, and total cholesterol, but not with HDL. WC had a higher correlation with triglycerides than BMI (rho 0.35 vs. 0.24). Likewise, both BMI and WC correlated significantly with ALT, UA, and urea with similar correlation coefficients. Overall, the highest correlation with any blood parameter observed was between WC and triglyceride. Fig 1 displays the correlation matrix of all variables.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Correlation coefficient heatmap.

Non-significant correlations are crossed.

https://doi.org/10.1371/journal.pone.0353676.g001

Predictability of BMI and WC to levels of blood parameters

The simple linear regression showed no linear relationship and predictability of BMI and WC to glucose levels. WC showed predictive ability to all lipid parameters: triglyceride, LDL, HDL, and cholesterol, whereas BMI was only related with HDL and LDL. Rise in liver enzyme ALT was related only to WC and not BMI. Both BMI and WC could predict uric acid levels. Additionally, renal parameters urea and creatinine was only predictable by WC and not BMI. Table 2 summarizes the linear relationship of all parameters with BMI and WC. For parameters which could be predicted by both BMI and WC, the change in parameters were much higher followed by elevation in 1 unit WC as compared to BMI.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Simple linear regression analysis of BMI and WC with blood parameters.

https://doi.org/10.1371/journal.pone.0353676.t002

Differences of BMI and WC between various cardiometabolic groups

BMI and WC were significantly higher in patients with HTN (p < 0.001) but did not differ based on DM status (p = 0.350 and 0.159). BMI was higher in patients with thyroid disease (p = 0.03), however there was no difference in WC based on thyroid status (p = 0.656). Overall, both BMI and WC were higher in FLD patients than those without FLD. The values significantly differed between FLD grades. While higher FLD grades were observed to have higher WC, they were not statistically significant. Similar variation was observed with BMI with insignificant statistics. Overall, both BMI and WC were associated with all cardiometabolic risk groups except diabetes mellitus (Table 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Differences of BMI and WC between different cardiometabolic risk groups.

https://doi.org/10.1371/journal.pone.0353676.t003

Odds of HTN, DM, Thyroid disease, FLD, NAFLD, and dyslipidemia based on BMI and WC

In the univariate analysis, there was no significant odds of DM and thyroid disease for both BMI and WC. For HTN, only WC had significant odds (OR: 1.04) however, this was modest. Significant odds for FLD, NAFLD, and dyslipidemia was observed for both BMI and WC, however, the OR was higher for BMI based analysis than WC. Similar results were seen when BMI and WC were adjusted for age, sex, smoking status, alcohol use, and diet(vegetarian vs. non-vegetarian). The results have been summarized in table 4. Among the covariates, age was significantly associated with all outcomes with small ORs (1.03–1.08). Sex (male) was significantly associated with HTN, FLD and NAFLD (ORs: 2.07^BMI 1.94^Waist, 1.94^BMI 2.08^Waist, and 2.86 respectively). In contrast male sex was associated with a lower odds for thyroid disease (0.39^{both models}). Interestingly, smoking and alcohol was not associated with any outcomes. Non-vegetarian diet was associated with lower odds of diabetes (OR=0.33^BMI) and thyroid disease (OR=0.31 ^{both models}). Adjusted ORs are displayed in Fig 2. Overall, this indicated that BMI is associated with higher odds of FLD, NAFLD, and dyslipidemia than WC. Male patients are more likely to develop HTN, FLD, and NAFLD while less likely to develop thyroid disease.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Univariate and multivariate logistic regression analysis of BMI and WC with risk groups.

https://doi.org/10.1371/journal.pone.0353676.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Heatmap diagram of covariate-adjusted ORs of multivariate regression analysis.

https://doi.org/10.1371/journal.pone.0353676.g002

Performance of BMI versus WC in classifying cardiometabolic risk outcomes

Hypertension: BMI and WC alone had similar AUCs (0.643 and 0.639; fair) for hypertension. When combined, the increment was mild (0.651). In contrast, there was a sharp increment when age and sex together with BMI and WC (AUC = 0.829; good) (Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. ROC curves of analysis of BMI and WC models to evaluate the accuracy with outcomes.

https://doi.org/10.1371/journal.pone.0353676.g003

Diabetes: Both BMI and WC had a poor AUC (0.543 and 0.579) to classify diabetes. The highest AUC was 0.736 when BMI and WC were combined and supplemented with age and diet (Fig 3E).

Thyroid: Accuracy of both BMI and WC in classifying thyroid disease was low (0.581 and 0.523). Increase in AUC was observed when BMI and WC together were combined with age, sex, and diet (0.687;Fair).

NAFLD: BMI and WC alone had similar AUCs (0.763 and 0.753; acceptable) for NAFLD. When combined, the increment was mild (0.774). The highest AUC was 0.781 when combined together with age and sex (Fig 3C).

FLD: BMI and WC alone had similar AUCs (0.774 and 0777; acceptable) for FLD. When combined, the increment was mild (0.792). In contrast, there was a yield of AUC 0.820 when age and sex were combined with BMI and WC for FLD, indicating good accuracy (Fig 3B).

Dyslipidemia: BMI and WC alone had similar AUCs (0.749 and 0.750; acceptable) for dyslipidemia. When combined, the increment was mild (0.795). The highest AUC was 0.806 when combined together with age and sex (Fig 3D).

Overall, BMI and WC standalone had good accuracy only for FLD, NAFLD, and dyslipidemia and accuracy improved when combining associated covariates (Fig 3).

Optimal cut-offs of BMI and WC for risk outcomes

Optimal BMI and waist circumference (WC) cut-offs for cardiometabolic risk were derived from ROC analysis using Youden’s J statistic (S1 Table). For hypertension and diabetes, BMI thresholds of 24.1–24.4 kg/m² showed high sensitivity (88.7%–90.5%) but low specificity (<40%). Thyroid disease required a higher BMI (25.95 kg/m²) with moderate sensitivity and specificity (64.4% and 55.4%). Fatty liver disease and NAFLD showed the strongest discrimination, with BMI 25.1–26.0 kg/m² and WC 88.5–93.5 cm. Dyslipidemia was best predicted by BMI 24.8 kg/m². Overall, BMI ~ 25 kg/m² and WC ~ 88–93 cm indicate substantially increased cardiometabolic risk, especially for fatty liver disease.

Association of additional obesity metrics with cardiometabolic risk outcomes

Table 5 shows that alternative obesity indices displayed varying predictive power for cardiometabolic risks. BRI was the strongest predictors, showing significant associations and higher AUCs for fatty liver (0.746) and NAFLD (0.739), outperforming BMI and WC. WtHR was linked to dyslipidemia (OR = 2.82, AUC = 0.702). It showed a higher odds of FLD and NAFLD than BMI or WC, however, the AUCs were lower. WBMIR showed inverse associations with fatty liver and NAFLD, while ABSI had weak and non-significant results (AUC ~ 0.5). Overall, BRI demonstrated superior discriminatory accuracy for HTN, FLD, and NAFLD while WtHR stood superior for its association with dyslipidemia. These findings indicated that these obesity metrics might have varying implications for different cardiometabolic risks.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Logistic regression and ROC analysis of additional obesity metrics with risk outcomes.

https://doi.org/10.1371/journal.pone.0353676.t005

Central adiposity indices (WC, WtHR, BRI) vs. BMI – quantitative comparisons

Our findings support that central adiposity measures outperform BMI for metabolic outcomes, with WtHR and WC better predicting diabetes, CVD, and overall risk; a WtHR of 0.5 is proposed as a simple global threshold [5,22]. In a Colombian cohort (n = 29,236), Montoya Castillo et al. found WtHR was the strongest predictor of cardiometabolic risk (OR 3.04 unadjusted; OR 1.99 adjusted for WC, OR 2.48 for BMI), while BMI risk plateaued near 30 kg/m² and WtHR showed a linear relationship [23]. These patterns mirror our findings, with central adiposity indices showing more linear, sensitive associations with risk, while BMI plateaus at higher values; accordingly, WC and derived measures (BRI, WtHR) often provide incremental discrimination beyond BMI alone.

At the discrimination level, our AUCs align with prior reports. We observed acceptable-to-good AUCs for BMI and WC in FLD and NAFLD (FLD: BMI 0.774, WC 0.777; NAFLD: BMI 0.763, WC 0.753), with combined models plus age/sex improving performance into the good range (FLD AUC 0.820). Similarly, Suwała et al. reported comparable sex-specific differences (men: BMI 0.816 vs WC 0.804; women: WC 0.762 vs BMI 0.739) and proposed intermediate BMI and WC cut-offs to optimize cardiovascular risk discrimination [24]. Our optimal cut-offs (BMI ≈ 25–26 kg/m²; WC ≈ 88–93.5 cm) align with cross-population risk thresholds for mid-20s BMI and high-80s/low-90s WC.

Hypertension – joint contribution of BMI and WC and comparison with other cohorts

Hypertension was the most common outcome (15.7%), with higher BMI and WC in affected participants (BMI 26.8 vs 25.3 kg/m²; WC 93 vs 89 cm; p < 0.001), though only WC showed a significant unadjusted association (OR 1.04 per cm). Adjusting for age and sex reduced both effects but improved discrimination (AUC ~ 0.64 → 0.829). This pattern, WC adding value while anthropometry requires demographic adjustment, aligns with studies highlighting abdominal obesity as a key driver of elevated blood pressure and mirrors Montoya Castillo’s findings linking WC and WtHR to cardiometabolic risk [23,25,26]. Feng et al. found BMI linked more to hypertension and WC to dysglycemia/dyslipidemia in northern China, with optimal cut-offs BMI 24 and WC 86/78 cm (men/women), showing both measures capture distinct but meaningful biological signals [27].

Dyslipidemia – visceral fat signal and comparison of effect sizes

Dyslipidemia was common (536/678). WC showed the strongest correlation with triglycerides (rho 0.35), and WC/WtHR/BRI better predicted lipid abnormalities, while BMI predicted LDL and HDL less consistently. In ROC analyses, BMI and WC had acceptable discrimination (AUC BMI 0.749; WC 0.750), improving modestly with covariates (AUC 0.806). These patterns mirror meta-analyses showing WtHR outperforms BMI for diabetes, metabolic syndrome, and cardiovascular outcomes, and a Canadian cohort where WC, not BMI, strongly predicted serum lipids [28,29]. These quantitative patterns are consistent with meta-analytic evidence showing WtHR and WC frequently outperform BMI for lipid endpoints [5,22] and with longitudinal analyses where BMI and WC trajectories predict incident dyslipidemia, Zhang et al. found BMI/WC trajectories associated with later dyslipidemia with gut microbiota as a mediator [30]. Feng et al. similarly reported stronger WC-dyslipidemia associations than BMI [28]. WC’s strong association with triglycerides supports visceral fat’s role in delivering hepatic free fatty acids, driving an atherogenic, hypertriglyceridemic profile.

FLD/NAFLD – where central indices and BRI shine

Hepatic outcomes showed the strongest anthropometric gradients: BMI increased from 24.0 to 30.0 kg/m² and WC from 86 to 104 cm (p < 0.001). Both BMI and WC strongly predicted FLD/NAFLD (AUC 0.753–0.777). BRI and WtHR are superior visceral fat indicators, showing higher NAFLD odds than ABSI. A meta-analysis (~60,000 participants) reported pooled BRI AUC 0.803, OR 2.87, supporting BRI’s value for liver-disease risk assessment beyond BMI or WC [31,32]. Still, WC alone captured much of the clinical signal, which argues for its pragmatic use as a first-line screening tool in clinical practice.

Diabetes – limited discrimination by single anthropometrics

Anthropometry poorly discriminated diabetes (AUC BMI 0.543; WC 0.579), but adding demographics and diet improved AUC to ~0.736, highlighting the influence of age, sex, and lifestyle. Attenuation after adjustment has been noted by Seo et al. and others [26,33]. Prior cohort studies show that anthropometric indices’ predictive value for diabetes, hypertension, and lipid abnormalities drops after adjusting for age, sex, or other confounders – a pattern reflected in our data, where BMI and WC alone poorly discriminated diabetes (AUC 0.543–0.579) and associations with hypertension weakened after adjustment [34]. Mediation analyses show that 25–37% of the link between central adiposity (WC or WtHR) and diabetes is mediated by lipid abnormalities, highlighting that anthropometry alone may miss metabolic risk reflected in lipid measures [35]. Thus, anthropometry alone is insufficient for diabetes screening in many clinical populations – biochemical screening remains essential.

Thyroid disease – a BMI-specific association

Our finding that BMI but not WC was higher in thyroid disease aligns biologically, as hypothyroidism raises weight via lower metabolic rate and fluid retention rather than visceral fat. A meta-analysis of 38 studies (49,427 individuals) reported large BMI increases (ES 1.092; 95% CI 0.755–1.429) but only moderate WC effects (ES 0.759; 95% CI 0.419–1.099) [36]. In a Chinese study of 193 with subclinical hypothyroidism, some thyroid-homeostasis indices were linked to BMI, but associations with WC and WtHR disappeared after adjusting for age, sex, metabolic status, and other confounders [37]. These sources support that thyroid dysfunction, particularly hypothyroidism, is more strongly linked to overall adiposity (BMI) than central obesity (WC), consistent with our finding that BMI but not WC was higher in our thyroid disease subgroup.

A study of Mexican Americans using SAFHS and NHANES 2007–2010 data found WC independently associated with a composite Thyroid Function Index, even after adjusting for BMI, age, sex, and other covariates; each SD increase in WC raised the odds of hypothyroidism [38]. Collectively, these findings reinforce that thyroid-adiposity associations differ across populations and disease phenotypes, but they provide empirical support for our observation that BMI retains unique clinical information for thyroid disease that is not captured by WC, and that BMI should not be abandoned solely in favor of central adiposity indices when evaluating endocrine risk.

Prospective evidence and hard outcomes – WC matters for mortality when diabetes coexists

High WC (>94 cm) with diabetes strongly raised cardiovascular mortality risk (HR 3.78, 95% CI 3.35–3.98) in 1,521 adults over 9.2 years, adding prognostic value beyond traditional risk factors [28]. These data highlight that WC identifies a high-risk central-adiposity/metabolic phenotype. Similarly, Montoya Castillo showed WtHR predicts combined cardiometabolic burden (ORs ~ 2–3), capturing risk beyond single outcomes [23].

Cut-offs – international convergence but population specificity

Across our and others’ studies optimal cut-offs cluster in the mid-20s for BMI and high-80s to low-90s cm for WC. In our cohort, Youden-derived thresholds concentrated near BMI ~ 25–26 kg/m² and WC ~ 88–93.5 cm (e.g., FLD: BMI 25.08, WC 88.5 cm; NAFLD: BMI 26.03, WC 93.5 cm). Suwała et al. proposed BMI 27.6 and WC 93/99 cm for SCORE2-based ICVR in a European cohort [24]; Montoya Castillo’s Colombian study found WtHR > 0.5 more strongly associated with risk than BMI. Adiposity measures included WtHR > 0.5, WC ≥ 80 cm (women)/ ≥ 90 cm (men), and BMI with overweight ≥25 kg/m² and obesity ≥30 kg/m² [23]. Asian studies (e.g., Taiwanese cohorts) show lower BMI/WC cut-offs due to ethnic differences in body composition and cardiometabolic risk. Optimal BMI ranged 24.5–25.7 kg/m² (men) and 22.6–24.0 kg/m² (women); WC 83.7–89.4 cm (men) and 73.5–80.4 cm (women), confirming Asians develop risk at lower thresholds than Caucasians [39]. These patterns support two practical points: (1) central indices are robust across settings, and (2) optimal numeric thresholds need local calibration for ethnic and age structure before wholesale adoption.

Strengths and limitations

Strengths of our work include direct head-to-head comparison of BMI and WC (plus several derived indices) across multiple clinically relevant endpoints in a real-world hospital setting, and reporting of discrimination statistics (AUCs), odds ratios, correlation coefficients, and empiric cut-offs for each outcome. However, we were limited by the cross-sectional design (no incident events), possible under-ascertainment/misclassification for diabetes and early NAFLD, single-site hospital sampling limiting population generalizability, and residual confounding from lifestyle/socioeconomic variables imperfectly measured. Because hypertension, diabetes mellitus, and thyroid disease were classified using existing medical records and medication history without active diagnostic confirmation, some degree of underdiagnosis or misclassification, particularly for undiagnosed or early disease cannot be excluded.