Visceral adiposity index is positively associated with fasting plasma glucose: a cross-sectional study from National Health and Nutrition Examination Survey 2017–2020

NHANES is a national representative cross-sectional survey on non-institutionalized population in the US with publicly accessible data, conducted by the National Centers for Disease Control and Prevention, which included demographic, dietary, examination, laboratory, questionnaire and limited data. Datasets from NHANES 2017 to 2020 required for the analysis were obtained from the NHANES website (https://​www.​cdc.​gov/​nchs/​nhanes/​). The written informed consents were obtained from all participants, and the program was approved by the Ethics Review Board of National Center for Health Statistics.

Of 15,560 candidates extracted from the NHANES database, 9884 participants aged≥18 without indispensable information for VAI calculation were excluded (missing data of BMI, N = 2423; missing data of waist circumference, N = 603; missing data of HDL-c, N = 2210 and missing data of TG, N = 4648). Individuals without information on fasting plasma glucose (N = 1239) were also excluded. Eventually, a total of 4437 participants were enrolled in this study, and the final selected subjects were divided into 4 groups according to the VAI value. The flow diagram of selection process was provided in Fig. 1.

High-quality anthropometric measurement data was collected by well-trained NHANES staff following standardized examination protocols, using standardized examination procedures and calibrated equipment. A stadiometer with an adjustable head piece and a fixed vertical backboard was used to measure standing height. Participants were weighed in kilograms using a digital weight scale. The formula weight (kilograms)/square of height (meters) was used to determine BMI values. WC was measured to the closest 1 mm at the end of the normal expiration. Draw a horizontal line just above the right ilium’s uppermost lateral border, and wrap the measuring tape around the waist. Blood pressure (BP) in sitting position was measured in the right arm sing Omron HEM-907XL BP monitor. After a rest of 5 min and triplicate BP determinations with at least 1 min interval were taken. The means of the three BP measurements were used in the following analysis.

The demographics questionnaires were asked by accomplished interviewers using Computer-Assisted Personal Interview (CAPI) system. Participants venous blood samples were collected and examined following the NHANES laboratory protocol. All methods of standard biochemistry were measured on the Roche Cobas 6000 (c501 module) analyzer. More detailed information about analyte methodologies, principles, and operating procedures could be found in the Laboratory Method Files at https://​wwwn.​cdc.​gov/​Nchs/​Nhanes/​. The information on treatment of diabetes, hypertension and hyperlipidemia was provided at the questionnaire data.

The collected anthropometric data (BMI and WC) and biochemical data (TG and HDL) were used for VAI calculation by previous reported formula [19], and the units of WC and BMI were cm and Kg/m², respectively. Both TG and HDL were expressed in mmol/L.

FPG was measured using the Roche/Hitachi Cobas C311 UV assay. Diabetes was defined as a self-reported history of diabetes, HbA1c level ≥ 6.5% or FPG level ≥ 7.0 mmol/L, according to the 2019 American Diabetes Association criteria [25].

Smoking was defined as smoking over 100 cigarettes during entire lifetime [26]. Moderate and vigorous activity were defined as ‘activity that causes small, and large increase in breathing or heart rate, respectively [27]. For example. Brisk walking was classified as moderate activity, while carrying heavy loads was categorized as vigorous activity. Stoke was self-reported and physician diagnosis using epidemiological data from NHANES [28]. eGFR was calculated through the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation. Then, CKD were defined according to the KDIGO guidelines [29]: stage 1, urinary albumin-to-creatinine ratio (ACR) ≥ 3 mg/mmol with eGFR ≥90 ml/min/1.73 m²; stage 2, ACR ≥ 3 mg/mmol with 60 ml/min/1.73 m² ≤ eGFR ≤89 ml/min/1.73 m²; stage 3, 30 ml/min/1.73 m² ≤ eGFR ≤59 ml/min/1.73 m² (with or without ACR ≥ 3 mg/mmol); stage 4, 15 ml/min/1.73 m² ≤ eGFR ≤29 ml/min/1.73 m²; and stage 5, eGFR < 15 ml/min/1.73 m².

Covariates in multivariate models might contribute to muddled correlations between VAI and FPG. The following variables were selected as covariates in the current study: age, gender, SBP, DBP, HR, smoking, physical activity, hypertension, hyperlipidemia, stroke, CKD, race, TC, LDL-c, ALT and ALP based on univariate analysis and previous studies.

All statistical analyses were performed using R software and EmpowerStats. Sample weights were adjusted to present nationally representative estimates in all analyses according to the stratified, multistage probability sampling design [30]. Continuous and categorical variables were presented as mean ± standard error and frequencies (percentages), respectively. Participants were equally divided into 4 quartiles according to the calculated VAI value: Q1 (VAI < 0.69), Q2 (0.69 ≤ VAI < 1.18), Q3 (1.18 ≤ VAI < 2.02) and Q4 (VAI ≥ 2.02). One-way analysis of variance and Kruskal-Wallis tests were used for comparison the differences of characteristics between 4 VAI groups for continuous variables and chi-square tests were utilized for categorical variables. The independent relationship between exposure factors and FPG was calculated using univariate analysis and multiple linear regression model. The association between VAI and FPG was expressed using 3 models with different adjustment for confounding factors: a crude model; minimally adjusted model: adjusted for age and gender; fully adjusted model: adjusted for age, gender, SBP, DBP, HR, smoking, physical activity, hypertension, hyperlipidemia, stroke, CKD, race, TC, LDL-c, ALT and ALP. These confounders were selected on the basis of their associations with FPG or an alteration in effect estimate of over 10% [31]. Non-linear relationships between VAI and fasting plasma glucose were explored using generalized additive model (GAM) and smooth curve fitting. Furthermore, we calculated the inflection points using two-piecewise linear regression model. The receiver operating characteristic (ROC) curve was utilized to evaluate the predictive potential of VAI for FPG elevation. The best threshold was determined according to the sum of sensitivity and specificity. A two-sided P < 0.05 was considered statistically significant. Sample weighting was applied for the unequal probability of individual sampling, resulting from complex multistage survey design.

Table 1 described the baseline characteristics of the study population. Four thousand four hundred thirty-seven individuals were divided into 4 groups according to the VAI value. 2175 (49.02%) were males, and the average age of included subjects was 44.41 ± 18.94. Subjects in the 4th VAI quartile group had significantly higher levels of age, anthropometric indexes (including BMI, SBP, DBP, HR, WC and hip circumference), WBC, neutrophil, hemoglobin, platelet, FPG, HbA1c, TG, TC, HLD-c, LDL-c, globulin, ALT, γGT, and uric acid. It is worth noting that FPG increased rapidly from 5.55 to 6.70 mmol/L with the increasing VAI quartiles (p < 0.0001). The percentage of smoking, hypertension, hyperlipidemia, diabetes, stroke, and CKD history, and the proportion of taking glucose-lowering, antihypertensive and lipid-lowering drugs also rose with the increasing of VAI quartiles. However, subjects with higher VAI value presented lower levels of albumin and total bilirubin, percentage of male and vigorous physical activity were less frequent. Additionally, different race distribution was found, as the VAI quartile increased, the proportion of Mexican American, Other Hispanic and Non-Hispanic White increased, while Non-Hispanic Black decreased. No statistical significances were observed in other variables among the VAI quartile groups.

Univariate linear analysis was performed to evaluate the relationship between the variables and FPG. As shown in Table 2, age, male, SBP, DBP, HR, WC, hip circumference, smoking, hypertension, hyperlipidemia, diabetes, stroke, CKD, TG, TC, HDL-c, LDL-c, ALT and ALP were all positively associated with FPG level (p < 0.05). Vigorous physical activity and HDL-c were negatively related to FPG level (p < 0.05). Compared with Mexican American, we observed a significant negative correlation between Non-Hispanic White and FPG, whereas no significant association was found between moderate physical activity, AST, total bilirubin, other races and FPG. Of note, VAI was significantly positively correlated with FPG (β = 0.24, 95% CI: 0.21–0.27, p < 0.0001).

Multiple linear regression model was used to evaluate the independent relationship between VAI and FPG. Table 3 displayed the β (95% CI) of VAI for FPG in different models. Higher VAI levels were remarkably associated with increased FPG levels. As a continuous variable, an increase of one unit in VAI was associated with 0.24 (95% CI: 0.21–0.27, p < 0.0001), 0.21 (95% CI: 0.19–0.24, p < 0.0001), and 0.52 mmol/L (95% CI: 0.41–0.63, p < 0.0001) higher FPG level, respectively, in crude model, minimally adjusted and fully adjusted model. As a categorical variable, 4th VAI quartile group was associated with 0.71 mmol/L (95% CI: 0.47–0.95, p < 0.001) higher FPG level after controlling all the potential confounding factors in fully adjusted model compared with participants with lowest FPG level in Q1.The trend test remained significant (p < 0.001).

After the correction of multiple potential confounding factors, the association between VAI and FPG levels remained significant in all subgroups stratified by age, gender, hypertension and hyperlipidemia. No significant association between VAI and FPG levels was found in diabetes subgroup in the fully adjusted model (β = 0.59, 95% CI: − 0.41-1.19, p = 0.0535). To further assess the impact of glucose-lowering drugs on FPG, the association between VAI and FPG was explored in diabetic patients receiving and not receiving antidiabetic treatment, respectively. The results showed that VAI was not associated with FPG in participants with diabetes no matter they received antidiabetic drugs or not (Table 4).

Generalized additive model (GAM) and smooth curve fitting analysis were performed to explore the nonlinear relationship between VAI and FPG. As presented in Fig. 2, a nonlinear association between VAI and FPG levels was observed after adjusting for age, gender, BMI, WC, SBP, DBP, HR, smoking, physical activity, hypertension, hyperlipidemia, stroke, CKD, race, TG, TC, HDL-c, LDL-c, ALT and ALP. FPG displayed an increasing trend with the increasement of VAI. Furthermore, threshold effect analysis was conducted using two-piecewise linear regression, and 4.02 was identified as the inflection point. We observed a dramatically positive correlation between VAI and FPG when VAI was below 4.02 (β = 0.73, 95% CI: 0.59–0.87, p < 0.0001), and the positive association between VAI and FPG also existed when VAI was higher than 4.02 (β = 0.23, 95% CI: 0.07–0.40, p = 0.0063) (Table 5). The smooth curve fittings in different subgroups stratified by age, gender, diabetes and hypertension were shown in Fig. 3. Similarly, non-linear relationship between VAI and FPG in subgroups were observed except for non-diabetic patients.

To further explore the predictive value of VAI for FPG elevation, participants were divided into elevated FPG group (FPG ≥ 7.0 mmol/L) and control FPG group (FPG < 7.0 mmol/L) according to the definition of diabetes. ROC curve analysis was performed to explore the predictive value of VAI for identifying FPG ≥ 7.0 mmol/L (Fig. 4) after adjusting for age, gender, SBP, DBP, HR, smoking, physical activity, hypertension, hyperlipidemia, stroke, CKD, race, TC, LDL-c, ALT and ALP. The area under the curve (AUC) was 0.7169 (95% CI: 0.6948–0.7389), the cut-off value of VAI was 1.4315, and the corresponding sensitivity and specificity were 64.03 and 69.90%, respectively. Additionally, accuracy, positive and negative likelihood ratio were presented in Table 6.