Correlation of visceral adiposity index and dietary profile with cardiovascular disease based on decision tree modeling: a cross-sectional study of NHANES

Abstract

Background

Visceral adiposity index (VAI) and diets are associated with the risk of cardiovascular disease (CVD). It is unclear how well VAI and diet predict CVD.

Methods

Data were obtained from the National Health and Nutrition Examination Survey (NHANES 2017–2018). Demographic data, diets, biochemical examination, and questionnaire information were collected. VAI was calculated using body mass index, waist circumference, triglycerides, and high-density lipoprotein cholesterol. Binary logistic regression was adopted to examine the correlation of VAI and diets with CVD. A decision tree model was developed to predict CVD risk according to different factors.

Results

2104 participants (mean age: 50.87 ± 17.35 years, 48.38% males) were included. Participants with high levels of VAI (≥ 2.18) had an elevated risk of CVD compared to those with low levels of VAI (≤ 0.76) (OR = 1.654, 95% CI: 1.025–2.669, P = 0.039). Compared with the low protein intake level (≤ 50.34 g), the upper intermediate (72.10–99.92 g) (OR = 0.445, 95% CI: 0.257–0.770, P = 0.004) and high (≥ 99.93 g) levels of protein intake (OR = 0.450, 95% CI: 0.236–0.858, P = 0.015) reduced CVD risk. The decision tree model unveiled that VAI, protein intake, and dietary fiber intake were predictors for CVD.

Conclusion

VAI and protein intake levels are independently associated with CVD risk and have predictive power for CVD. These findings can provide insights into the development of appropriate lifestyle and treatment strategies for patients to reduce the incidence of CVD.

Background

Cardiovascular disease (CVD) is a major contributor to death and disability worldwide. The number of people with CVD has increased from 271 million in 1990 to 532 million in 2019, and CVD remains the primary cause of the global disease burden, posing a serious threat to human life [1]. CVD mainly includes ischemic heart disease, congestive heart failure (CHF), and stroke [2]. The main risk factors for CVD include obesity, insulin metabolism resistance, and metabolic syndrome [3, 4]. Different dietary components may affect CVD development, and healthy dietary patterns help prevent CVD [5,6,7]. According to fat distribution, obesity is classified into homogeneous and visceral types. Visceral adiposity index (VAI) is a reliable indicator for evaluating visceral adiposity, overweight/obesity, and lipid levels [8], [9], and is more convenient and cost-effective than the evaluation using visceral adipose tissue (VAT).

Exploring risk factors for CVD is of great practical significance. Risk factors have previously been analyzed using logistic regression, which quantifies the effect of risk factors on outcome variables but fails to provide better decision-making recommendations [10] and address multicollinearity between variables [11]. In recent years, data mining techniques have been widely used in the medical field, among which decision tree (DT) adopts a tree structure to present the results simply and intuitively, with no specific requirements for data distribution, and can identify the main risk factors for diseases [12]. DT not only solves the possible multicollinearity between variables in the analysis of disease risk factors but also displays the importance of risk factors on outcome variables [13]. Therefore, this study was to explore the correlation of VAI and dietary profile with CVD based on DT models.

Methods

Study populations

The National Health and Nutrition Examination Survey (NHANES) is a population-based cross-sectional study conducted by the Centers for Disease Control and Prevention to assess the health and nutritional status of adults and children in the United States.

NHANES was approved by the Ethical Review Committee of the National Center for Health Statistics. All participants signed informed consent and received a standardized personal interview conducted by a medical professional, as well as laboratory tests and measurements of basic physiological data. The data in NHANES are updated on a biannual cycle.

Data from NHANES (2017–2018) were de-identified for this study, including 9254 participants. Participants (n = 7150) who were < 18 years of age and lacked information on body measurements, lipid testing, or health questionnaires were excluded, and ultimately 2104 participants were included (Fig. 1).

Fig. 1

Participant enrollment

Definitions for VAI and other items

CVD was defined as having a previous diagnosis of CHF, coronary heart disease (CHD), angina pectoris, heart attack, or stroke. Participants who answered “yes” to the question “Ever a doctor told you had CHF” were considered to have CHF. In the same manner, CHD, angina/angina pectoris, heart attack, and stroke were defined.

Someone who had smoked at least 100 cigarettes in their lifetime was defined as a smoker [14].

Participants who answered “yes” to “Ever a doctor told you had hypertension” and claimed the doctor’s diagnosis of hypertension and use of antihypertensive drugs were considered to have hypertension. Participants who answered “yes” to “Ever a doctor told you had diabetes” and claimed the doctor’s diagnosis of diabetes and use of hypoglycemic agents were considered to have diabetes.

Depending on sex, VAI was calculated by the respective formulae:

Males: VAI = waist circumference (WC) (cm)/(39.68 + 1.88 × BMI kg/m²) × (TG[mmol/l]/1.03) × (1.31/HDL-C [mmol/l]).

Females: VAI = WC (cm)/(39.58 + 1.89 × BMI kg/m²) × (TG[mmol/l]/0.81) × (1.52/HDL-C [mmol/l]) [9, 15].

High-density lipoprotein cholesterol (HDL-C) was measured using the magnesium/dextran sulfate solution. TG was measured using the method of Wahlefeld (Roche, 2014), and low-density lipoprotein cholesterol was calculated based on the directly measured values of total cholesterol, triglyceride (TG), and HDL-C using the Friedewald Wald equation. Body mass index (BMI) was calculated as weight (kg) divided by the square of height (m) (kg/m²) [16].

Statistical methods

Data analyses were implemented using SPSS24.0 (IBM, Armonk, NY). Continuous variables were displayed as mean ± standard deviation ($\overline{x }\pm s$) or median and quartiles M [P25, P75], and categorical variables were depicted as a frequency and percentage (%). Continuous variables were analyzed using independent samples t-tests and non-parametric tests, and categorical variables were analyzed using chi-square tests. Participants were grouped according to quartiles of VAI, and dietary profiles were analyzed using one-way ANOVA. GraphPad Prism was utilized for drawing violin plots. Possible risk factors for CVD were explored using binary logistic regression analysis. The CVD risk under different conditions was predicted using DT models. P < 0.05 was defined as significant.

Results

Basic information

This study enrolled 2104 participants from NHANES (2017–2018) (mean age: 50.87 years) with 48.38% males and 51.62% females; 34.17% of participants aged 61 years and above; and the largest proportion (56.56%) received high school education and above; 37.79% were hypertensive, 19.58% were diabetic, and 43.49% were smokers.

Based on the presence or absence of CVD, the study populations were assigned into a CVD group (n = 253) and a N-CVD group (n = 1851). The univariate analysis manifested statistical differences between the two groups in demographic information (sex, age, race, smoking) and the prevalence of hypertension and diabetes (all P < 0.05) (Table 1).

Table 1 Basic information of study populations

Comparisons of VAI and dietary profiles

The univariate analysis noted marked differences in VAI, intake levels of protein, carbohydrate, dietary fiber, and total energy between the two groups (all P < 0.05). VAI was greatly lower in the N-CVD group [1.24 (0.73, 2.14)] than in the CVD group [1.54 (0.99, 2.44)], and the intake levels of protein, carbohydrates, dietary fiber, and total energy were notably lower in the CVD group (all P < 0.05). Fat intake was lower in the CVD group than in the N-CVD group, without significant differences (P = 0.08) (Table 2, Fig. 2).

Table 2 VAI and dietary profiles in study populations

Fig. 2

VAI and dietary profiles in study populations;* P < 0.05

Dietary profiles at different VAI levels

All study subjects were grouped by quartiles of VAI and named Q1, Q2, Q3, and Q4 groups from the lowest to the highest. Kruskal–Wallis test revealed marked differences (P < 0.05) in the protein and fat intake levels of different VAI groups. Fat intake gradually decreased with increased VAI levels (Table 3).

Table 3 Diets at different VAI levels

Risk factors for CVD

The logistic regression model showed that VAI and protein intake level were related to CVD (Supplementary file 1). Covariates such as demographic information and previous comorbidities were further included in the logistic regression model to explore risk factors for CVD. It was found that age, education level, race, hypertension, history of diabetes, smoking, VAI, and protein intake level were associated with CVD (all P < 0.05) (Table 4). There was no multicollinearity among all covariates. CVD risk in middle-aged individuals was 4.364 times higher than young individuals (OR = 4.364, P < 0.001); whereas CVD risk in older individuals was 9.097 times higher than young individuals (OR = 9.097, P < 0.001). A high level of education might be a protective factor against CVD. Compared to those with below high school education, individuals with high school education had a 60.5% risk of developing CVD (OR = 0.605, P = 0.024), and those with above high school education had a 65.8% risk of CVD (OR = 0.632, P = 0.017). CVD risk in other races was 50.4% of that in non-Hispanic blacks (OR = 0.504, P = 0.002). Patients with hypertension had 2.747 times the risk of other CVD compared to those without hypertension (OR = 2.747, P < 0.001). CVD risk in diabetic individuals was 2.011 times higher than non-diabetic ones (OR = 2.011, P < 0.001). Smoking individuals had 1.576 times higher CVD risk than those who did not smoke (OR = 1.576, P < 0.001). Individuals with high levels of VAI (≥ 2.18) had 1.654 times higher CVD risk than those with low levels of VAI (≤ 0.76) (OR = 1.654, P = 0.035). Individuals with upper intermediate (72.1–99.92 g) and higher (≥ 99.93 g) levels of protein intake had 44.5% (OR = 0.445, P = 0.004) and 45.0% (OR = 0.450, P = 0.015) of CVD risk compared to those with low levels of protein intake (≤ 50.34 g). The effects of VAI, diet, and other factors on CVD were also analyzed in sex subgroups (Supplementary file 2).

Table 4 Binary logistic regression analysis of factors influencing CVD

Risk factors for CVD based on the DT model

Thirteen factors including sex, age, race, education level, hypertension, diabetes, smoking, VAI, protein, fat, carbohydrate, energy, and dietary fiber were included in the DT model. Eight influencing factors (age, race, hypertension, diabetes, smoking, VAI, protein, and dietary fiber) were screened out in the model.

The DT model had 6 layers and 26 nodes (Fig. 3). There were 14 paths in the DT model, namely:

• 1.  ≤ 60 years old, no hypertension, smoking, other races, with an approximately 1.3% risk of CVD.

• 2.  ≤ 60 years old, no hypertension, smoking, non-Hispanic whites or blacks, with an approximately 5.8% risk of CVD.

• 3.  ≤ 60 years old, no hypertension, no smoking, with 0.6% risk of CVD.

• 4.  ≤ 60 years old, hypertension, ≤ 40 years old, with an approximately 5.4% risk of CVD.

• 5.  ≤ 60 years old, hypertension, 41 ~ 60 years old, VAI ≤ 1.28, with about 11.2% risk of CVD.

• 6.  ≤ 60 years old, hypertension, 41 ~ 60 years old, VAI > 1.28, non-Hispanic whites or blacks, with about 33.3% risk of CVD.

• 7.  ≤ 60 years old, hypertension, 41 ~ 60 years old, VAI > 1.28, other races, with about 14.9% risk of CVD.

• 8.  ≥ 61 years old, no diabetes, non-Hispanic whites, no hypertension, with about 19.2% risk of CVD.

• 9.  ≥ 61 years old, no diabetes, non-Hispanic whites, hypertension, with about 29.9% risk of CVD.

• 10.  ≥ 61 years old, no diabetes, non-Hispanic blacks or other races, dietary fiber intake ≤ 10.60 g, with about 25.3% risk of CVD.

• 11.  ≥ 61 years old, no diabetes, non-Hispanic blacks or other races, dietary fiber intake > 10.60 g, with about 8.1% risk of CVD.

• 12.  ≥ 61 years old, diabetes, smoking, protein intake ≤ 72.10 g, with about 48.8% risk of CVD.

• 13.  ≥ 61 years old, diabetes, smoking, protein intake > 72.10 g, with about 31.7% risk of CVD.

• 14.  ≥ 61 years old, diabetes, no smoking, with about 27.5% risk of CVD.

Fig. 3

Risk factors for CVD based on the DT model

Discussion

This cross-sectional study of American adults incorporated data from NHANES 2017–2018 and discovered that VAI and protein intake levels were associated with CVD development, with high levels of VAI increasing CVD risk and high protein intake reducing CVD risk. Meanwhile, VAI, protein intake, and dietary fiber intake levels were identified as effective predictors of CVD in the DT model.

BMI is globally recognized as a simple indicator of obesity, but there may be some limitations in defining an obese individual based on BMI [17]. Two individuals with similar BMI values may differ in health risks because BMI alone cannot be used as a biomarker for overall adiposity. Several studies have pointed out that VAT quantified using CT or MRI is associated with atherosclerosis and cardiac and metabolic risk [18, 19]. Visceral adipocytes increase the production of cytokines, chemokines, and other molecules such as tumor necrosis factor and leptin [20], leading to endothelial injury, vascular smooth muscle cell proliferation, and inflammatory responses, ultimately inducing atherosclerotic plaques [20, 21]. WC is the most readily available index and is a clinical parameter for indirectly assessing abdominal fat, but it fails to differentiate between subcutaneous and visceral fat. VAI, a composite index of WC, BMI, triglycerides, and HDL-C, can reflect visceral fat content with higher sensitivity [9].

Our study illustrated that people with higher levels of VAI had a significantly enhanced risk of CVD compared to those with lower levels of VAI. Data from the ATTICA study by G-M Kouli et al. demonstrated that VAI was independently correlated with an elevated 10-year risk of CVD, particularly in males [22]. Liying Zheng et al. also showed a correlation between VAI and CVD risk [23]. However, some studies reach different conclusions. Aysegul Gulbahar et al. showed that VAI did not predict the 10-year CVD risk in postmenopausal females [24]. This phenomenon may be due to hormonal changes in postmenopausal women, which lead to changes in body weight and body composition. Differences in body fat distribution between males and females may also explain the discrepancy. Our study did not find the effect of sex on the results.

The 2021 American Heart Association dietary guidelines stated that poor diets were strongly associated with increased morbidity and mortality of CVD [25]. The present study found that the upper intermediate level (> 72.10 g) of protein intake might be associated with a lower CVD risk. High-protein diets are widely accepted to lose weight and improve muscle mass for fitness, but their effects on CVD are controversial. Thomas L Halton et al. followed 82,802 women for 20 years and claimed that higher protein intake was not associated with an increased CHD risk in women [26]. A prospective study by Pagona Lagiou followed 43,396 Swedish women for 15.7 years and unveiled through dietary questionnaires that high-protein diets increased the risk of CVD without consideration of the source of carbohydrates and proteins [27]. Some studies, however, had opposite conclusions. A prospective cohort study of 70,696 participants noted that high plant protein intake was associated with low overall mortality and CVD-related mortality [28]. Mingyang Song et al. showed similar conclusions [29]. The discrepancy may be owing to the different sources of protein (animal or plant) and definitions of high-protein diets. The present study did not separate the sources of protein, and the total intake of proteins from plant and animal sources was included.

Past studies have suggested that higher dietary fiber intake is associated with a lower CVD risk, and total dietary fiber intake is negatively associated with all-cause mortality and CVD-related mortality [30]. Dietary fiber can reduce postprandial absorption of glucose and lipids, facilitate the excretion rate of bile acids, diminish cholesterol and LDL-C levels, and impair the effects of pro-inflammatory cytokines on plaque stability [31,32,33]. Our DT models also highlighted that low levels of dietary fiber intake might be associated with a higher prevalence of CVD.

Overall, this is the first study to examine the correlation of VAI and dietary profiles with CVD and to provide a more accurate prediction of CVD risk through DT models. It provides a healthy lifestyle against CVD for people and a relatively simple assessment tool for clinicians. However, this study also has some limitations. First, this was a cross-sectional study and therefore only correlations could be explored rather than causality. Second, the diagnosis of CVD was derived from patients’ claims of physician reports and lacked corresponding diagnostic imaging findings. Finally, our study did not include follow-up information and failed to judge the impact of VAI and diet changes on the results. Therefore, generalization of the results to all regions would require more comprehensive and detailed data on the various ethnic groups around the world for further analysis.

Conclusion

VAI and protein intake levels are independently associated with CVD risk, and CVD risk can be predicted by VAI, protein intake, and dietary fiber intake. These results provide recommendations on appropriate lifestyle and treatment strategies for patients to reduce the incidence of CVD.