A Bayesian network analysis of the probabilistic relationships between metabolically healthy obesity and cardiovascular disease risk under new diagnostic criteria

Abstract

BACKGROUND:

The relationship between metabolically healthy obesity (MHO) and cardiovascular disease (CVD) risk remains debated. The critical point may be the lack of consensus on MHO’s definition and diagnostic criteria.

OBJECTIVE:

This study aimed to investigate the association of MHO status with arteriosclerosis-CVD (ASCVD) risk in Chinese under new diagnostic criteria.

METHODS:

Participants aged 35–79 in the 2009 China Health and Nutrition Survey cohort were included. The 10-year ASCVD risk was predicted by the prediction for ASCVD risk in China, and participants with a predicted risk of $\geqslant$ 10% were classified into the high-risk group. The Bayesian network (BN) models were constructed to characterize the multivariable probabilistic connections between metabolically obesity phenotypes and ASCVD risk.

RESULTS:

The 10-year ASCVD risk score and the proportion of individuals at ASCVD high risk were significantly different between metabolically obesity phenotypes ( $P<$ 0.001). BN reasoning results showed that MHO individuals were not significantly associated with a 10-year ASCVD risk. Among metabolically unhealthy individuals, the conditional probability of high ASCVD risk increased with the Body Mass Index (BMI), with the conditional probability of high ASCVD risk was 24.63% (95% CI: 22.81–26.55%), 32.97% (95% CI: 30.75–35.27%) and 40.2% (95% CI: 36.64–43.86%) for metabolically unhealthy normal weight (MUNW), metabolically healthy overweight weight (MHOW), and metabolically unhealthy obesity (MUO) group, respectively. Subgroup analysis showed that MHO individuals were at increased risk of CVD compared with metabolically healthy normal weight (MHNW) individuals only in females.

CONCLUSION:

These results showed that there was no significant increase in ASCVD risk of MHO phenotype based on the new diagnostic criteria, suggesting that MHO is in a relatively healthy state.

Keywords

Metabolically healthy obesity atherosclerosis cardiovascular disease Bayesian networks

1. Introduction

In recent years, overweight and obesity have become a worldwide epidemic. According to The Lancet, from 1975 to 2014, the rate of adult obesity in men increased from 3.2% to 10.8% and from 6.4% to 14.9% in women [1]. Obesity is a significant risk factor for many chronic diseases, such as cardiovascular disease (CVD), and contributes to the increasing morbidity, mortality, disability, and medical costs of these CVDs [2, 3]. Finding a specific subset of the obesity phenotype resilient to obesity-related cardiovascular outcomes allows us to focus interventions on those who may benefit most and develop personalized treatments.

Recently, related studies have found significant heterogeneity in the body fat distribution and metabolic indicators of obese people [4]. Some obese people do not show abnormal cardiometabolic characteristics but have higher insulin sensitivity, healthier blood pressure, blood glucose, blood lipids, and other metabolic conditions, known as “metabolically healthy obesity (MHO)”. There is an ongoing debate about the association between MHO with CVD and other poor clinical prognoses. The previous systematic reviews and meta-analyses have concluded that MHO is associated with an increased risk of developing CVD. However, recently studies have shown that MHO has a significantly lower risk of metabolic diseases, CVD, and other adverse clinical outcome events compared with metabolically unhealthy obesity (MUO) [5, 6]. The high heterogeneity and conflicting results observed in those studies are largely due to the absence of a uniform definition and diagnostic criteria for MHO and the use of different outcome definitions [7, 8].

In previous studies, the definitions related to metabolic health were frequently based on the absence of metabolic syndrome (MetS). Theoretically, CVD and mortality risk of individuals with MHO should be similar to those in metabolically healthy normal-weight (MHNW). However, studies in large-scale cohorts and meta-analyses found significantly increased CVD risk and overall mortality in individuals with MHO despite meeting the most stringent criteria of MetS [9]. Then, some researchers question whether the previous definition and diagnostic criteria of metabolically healthy may not be sufficient to identify risk-free MHO individuals. In 2021, Zembic et al. established a new definition for metabolically healthy based on the mortality follow-up of the third National Health and Nutrition Examination Survey (NHANES-III) and validated it in an independent cohort (UK Biobank) [10]. There is no report on the relationship between the MHO population and CVD risk in Chinese based on their diagnostic criterion.

Therefore, this study aimed to explore the association between the MHO population and the risk of arteriosclerosis-CVD (ASCVD) under the new diagnostic criteria based on the Chinese cohort.

2. Materials and methods

2.1 Study population

Participants came from the China Health and Nutrition Survey Cohort study (CHNS). The CHNS cohort study began in 1989 and is an ongoing, long-term follow-up survey that selected populations from 9 provinces through multi-level sampling [11]. A detailed description of the study design, eligibility criteria and sampling has been published previously [12]. The study was approved by the University of North Carolina Review Board and the Ethical Review Committee of the Institute of Nutrition and Health of the Chinese Centers for Disease Control and Prevention. All participants offered informed consent. Participants and data were from the 2009 CHNS database ( $N=$ 11,929). According to the purpose of the study, 6,108 subjects aged 35–79 years were finally included.

2.2 Data collection

The general information of the subjects was obtained by face-to-face questionnaire. Information on demographic data (age, sex, marital status, and education level), behavioral lifestyle (smoking, alcohol consumption, physical activity), personal health status and medical history, family medical history, and medication use were collected.

The physical examinations of subjects were measured by medical professionals following standardized procedures. Body Mass Index (BMI, kg/m²) was calculated as weight (kg)/height (m)², and WHpR was calculated as waist/hip. Blood pressure measurements were repeated 3 times after a 10-minute sedentary rest, and an average of 3 times was taken.

Blood routine and other blood biochemical indexes were measured by the standardized process in the local laboratory at the scene of the investigation. Data on fasting plasma glucose (FPG), lipids, total cholesterol (TC), low-density lipoprotein (LDL-C), high-density lipoprotein cholesterol (HDL-C), triglyceride (TG), and uric acid (UA) were collected.

2.3 Diagnostic criteria of metabolically healthy phenotype

Diagnostic criteria for metabolically healthy refer to the latest diagnostic criteria proposed by Zembic et al.: a) SBP $<$ 130 mm Hg and no use of BP-lowering medication, b) WHpR $<$ 1.03 (men)/0.95 (women), c) no prevalent diabetes; otherwise were metabolically unhealthy [10]. Overweight and obese were defined as BMI $\geqslant$ 24 kg/m² and $\geqslant$ 28 kg/m², respectively, using the diagnostic criteria for overweight and obesity in Chinese adults [13]. The subjects were then divided into three BMI groups: normal weight (18.5 kg/m² $\leqslant$ BMI $<$ 24.0 kg/m²), overweight (24.0 kg/m² $\leqslant$ BMI $<$ 28.0 kg/m²) and obese (BMI $\geqslant$ 28.0 kg/m²).

According to the definition of BMI level and metabolic status, the subjects were divided into six following metabolic obesity phenotypes: MHNW, metabolically healthy overweight weight (MHOW), MHO, metabolically unhealthy normal weight (MUNW), metabolically unhealthy overweight weight (MUOW), andMUO.

2.4 ASCVD risk assessment and grouping

ASCVD risk assessment was performed using a Chinese population-based prediction for ASCVD risk in China (China-PAR) model on an online website: https://www.cvdrisk.com.cn/ASCVD/Eval [14]. Covariables include gender, age, current residence, region, waist circumference, TC, HDL-C, current blood pressure levels, history of antihypertensive medication, diabetes, smoking, and family history of ASCVD. The 10-year ASCVD risk stratification in the China-PAR model was: ASCVD $<$ 10% was a low-intermediate risk, and ASCVD $\geqslant$ 10% was high-risk [15, 16].

2.5 Statistical analysis

Bayesian network (BN) models were used to construct the network structure of the associations between metabolic obesity phenotype and the risk of ASCVD. The BN is a statistical method for inference and data analysis of uncertain information based on the principle of probability propagation [17]. The construction of the BN model mainly includes two parts: structure learning and parameter learning. In the first stage of manual construction, the prior knowledge of clinical experts was effectively embedded into the network structure through Whitelist and Blacklist. In the second stage of data-driven, the Tabu search algorithm and Bayesian information criterion score were used to learn the network structure and parameters [18, 19]. The stabilities of the arcs in the network were examined from 300 bootstrapped networks based on the original database, and the arc strengths were estimated by averaging the probability of the arcs presenting in these bootstrap-resampled network structures. The specific strengths thresholds were selected to construct network structure models of 10-year ASCVD risk and metabolically obesity phenotype.

Statistical analysis was performed using R 4.1.2 software. Continuous variables were presented as mean $\pm$ standard deviation; categorical variables were presented as numbers (percentage). The ANOVA test was used to compare continuous variables, and the chi-square test was used to compare categorical variables. The BN construction, structure learning, parameter estimation, and reasoning were performed in the “bnlearn” package in R software [20]. A $P$ value $<$ 0.05 was considered statistically significant.

3. Results

3.1 Characteristics of the study population

Table 1 shows the characteristics of the study population. Among the 6,108 participants, 2,817 (46%) were men. Except for drinking history, there were significant differences in demography, biochemical indexes, hyperuricemia, hypertension and diabetes status among the six groups ( $P<$ 0.05). Within the same BMI levels, the metabolically healthy groups had higher waist, hip circumference, blood pressure, TC, TG, UA, FPG, HOMA-IR and HbA1c levels and lower HDL-C levels than the metabolically unhealthy groups. In the MHO group, 21 (9.91%) had hypertension, 118 (55.14%) had dyslipidemia, and 45 (21.03%) had hyperuricemia, while in the MUO group, the prevalence of the abovementioned disease was 77.99%, 60.89% and 29.74%, respectively.

3.2 ASCVD risk among different metabolic obesity phenotypes

In this study, the 10-year ASCVD risk of this population was 5.09 $\pm$ 5.23%, and 15% of subjects (900/6,108) were at high risk according to the China-PAR model. The 10-year ASCVD risk score and the proportion of participants with a high ASCVD risk were significantly different among metabolically obesity phenotypes ( $P<$ 0.001, Table 2). Weight was positively associated with the 10-year ASCVD risk score ( $P$ -trend $<$ 0.001). The 10-year ASCVD risk score of MHO group was 2.84 $\pm$ 2.49%, while that of MUNW group and MUO group was 7.83 $\pm$ 5.72% and 9.01 $\pm$ 6.66%, respectively. When stratified by ASCVD risk, the high-risk proportion of the metabolically unhealthy groups was significantly higher than that of the metabolically healthy groups.

3.3 Construction of BN model

In this study, the nodes in the BN model include age, sex, metabolic status, BMI level, smoking history, drinking history, hyperuricemia, diabetes, hypertension, dyslipidemia, eGFR, hs-CRP and ASCVD risk classification. As shown in Fig. 1, age and gender are directly connected with ASCVD risk; dyslipidemia indirectly affects the ASCVD risk by affecting metabolic status, gender and other intermediate nodes. In addition, dyslipidemia is indirectly linked to sex through node hyperuricemia. Diabetes and eGFR are indirectly associated with ASCVD through metabolic status and sex, respectively.

3.4 Reasoning of BN model

Figure 2 shows the conditional probability distribution of metabolically obesity phenotype and 10-year

Table 1
General characteristics among the study population^a

	MHNW ( $N=$ 2147)	MHOW ( $N=$ 1038)	MHO ( $N=$ 214)	MUNW ( $N=$ 1238)	MUOW ( $N=$ 1044)	MUO ( $N=$ 427)	$P$ -value
Sex (men/women)	994/1153	473/565	90/124	594/644	499/545	167/260	0.024
Age, year	50.39 $\pm$ 10.37	49.89 $\pm$ 9.46	49.14 $\pm$ 9.25	59.58 $\pm$ 10.63	57.53 $\pm$ 10.29	55.63 $\pm$ 10.77	$<$ 0.001
Smoking, $N$ (%)	704 (32.79%)	308 (29.67%)	55 (25.70%)	447 (36.11%)	321 (30.75%)	94 (22.01%)	$<$ 0.001
Drinking, $N$ (%)	727 (33.86%)	378 (36.42%)	64 (29.91%)	387 (31.26%)	351 (33.62%)	127 (29.74%)	0.057
Weight (kg)	55.79 $\pm$ 6.88	66.76 $\pm$ 7.22	76.65 $\pm$ 9.02	56.01 $\pm$ 7.18	67.04 $\pm$ 7.76	77.60 $\pm$ 9.52	$<$ 0.001
Height (cm)	160.84 $\pm$ 8.20	161.38 $\pm$ 8.26	160.54 $\pm$ 8.71	159.88 $\pm$ 8.53	160.99 $\pm$ 8.42	160.05 $\pm$ 8.53	$<$ 0.001
BMI (kg/m²)	21.51 $\pm$ 1.43	25.57 $\pm$ 1.08	29.66 $\pm$ 1.57	21.84 $\pm$ 1.46	25.79 $\pm$ 1.13	30.21 $\pm$ 1.95	$<$ 0.001
Waist circumference (cm)	77.53 $\pm$ 6.78	86.60 $\pm$ 6.41	94.79 $\pm$ 7.76	81.32 $\pm$ 7.56	89.91 $\pm$ 6.95	98.84 $\pm$ 7.67	$<$ 0.001
Hip circumference (cm)	91.41 $\pm$ 5.36	98.40 $\pm$ 5.22	104.85 $\pm$ 5.91	91.63 $\pm$ 6.37	98.50 $\pm$ 5.60	106.23 $\pm$ 6.34	$<$ 0.001
Waist-to-hip ratio	0.85 $\pm$ 0.06	0.88 $\pm$ 0.05	0.90 $\pm$ 0.06	0.89 $\pm$ 0.09	0.91 $\pm$ 0.07	0.93 $\pm$ 0.06	$<$ 0.001
DBP (mm Hg)	75.50 $\pm$ 8.05	78.39 $\pm$ 7.29	79.87 $\pm$ 7.76	86.40 $\pm$ 11.28	88.59 $\pm$ 11.36	92.13 $\pm$ 11.66	$<$ 0.001
SBP (mm Hg)	113.77 $\pm$ 9.59	117.02 $\pm$ 8.52	117.97 $\pm$ 8.13	139.83 $\pm$ 16.87	140.87 $\pm$ 18.04	144.44 $\pm$ 17.72	$<$ 0.001
TC (mmol/L)	4.73 $\pm$ 0.93	4.95 $\pm$ 0.97	5.08 $\pm$ 0.93	5.03 $\pm$ 1.03	5.19 $\pm$ 0.98	5.24 $\pm$ 1.08	$<$ 0.001
HDL-C (mmol/L)	1.51 $\pm$ 0.46	1.35 $\pm$ 0.37	1.23 $\pm$ 0.29	1.50 $\pm$ 0.53	1.34 $\pm$ 0.34	1.31 $\pm$ 0.65	$<$ 0.001
LDL-C (mmol/L)	2.91 $\pm$ 0.94	3.08 $\pm$ 0.92	3.20 $\pm$ 0.89	3.12 $\pm$ 1.03	3.23 $\pm$ 0.95	3.21 $\pm$ 1.26	$<$ 0.001
TG (mmol/L)	1.35 $\pm$ 1.09	1.84 $\pm$ 1.51	2.10 $\pm$ 1.28	1.62 $\pm$ 1.18	2.13 $\pm$ 1.79	2.62 $\pm$ 2.25	$<$ 0.001
FPG (mmol/L)	5.04 $\pm$ 0.62	5.18 $\pm$ 0.64	5.31 $\pm$ 0.62	5.62 $\pm$ 1.55	5.77 $\pm$ 1.80	6.16 $\pm$ 1.83	$<$ 0.001
HOMA-IR	2.62 $\pm$ 4.12	3.02 $\pm$ 2.61	4.55 $\pm$ 4.43	3.96 $\pm$ 9.93	5.06 $\pm$ 9.22	6.17 $\pm$ 7.09	$<$ 0.001
HbA1c	5.40 $\pm$ 0.47	5.56 $\pm$ 0.54	5.68 $\pm$ 0.52	5.70 $\pm$ 0.92	5.84 $\pm$ 0.93	6.15 $\pm$ 1.11	$<$ 0.001
Insulin	11.35 $\pm$ 18.31	12.86 $\pm$ 9.81	18.74 $\pm$ 16.80	14.30 $\pm$ 30.10	17.55 $\pm$ 22.50	21.14 $\pm$ 19.13	$<$ 0.001
Hemoglobin	138.75 $\pm$ 20.48	143.09 $\pm$ 19.65	143.50 $\pm$ 16.66	140.33 $\pm$ 21.04	144.65 $\pm$ 19.71	145.39 $\pm$ 19.70	$<$ 0.001
Uric acid ( $\mu$ mol/L)	283.80 $\pm$ 91.43	306.79 $\pm$ 109.99	324.71 $\pm$ 83.62	307.86 $\pm$ 92.08	338.14 $\pm$ 110.50	350.00 $\pm$ 120.92	$<$ 0.001
hs-CRP (mg/L)	1.91 $\pm$ 5.78	2.41 $\pm$ 5.95	3.04 $\pm$ 3.84	2.90 $\pm$ 9.63	2.87 $\pm$ 5.04	3.67 $\pm$ 5.57	$<$ 0.001
eGFR (ml/min $\times$ 1.73 m²)	79.66 $\pm$ 14.04	80.32 $\pm$ 13.18	80.12 $\pm$ 13.34	72.60 $\pm$ 15.28	73.80 $\pm$ 14.67	75.93 $\pm$ 15.11	$<$ 0.001
Residence, $N$ (%)							0.027
Urban	686 (31.95%)	301 (29.00%)	81 (37.85%)	372 (30.05%)	360 (34.48%)	134 (31.38%)
Rural	1461 (68.05%)	737 (71.00%)	133 (62.15%)	866 (69.95%)	684 (65.52%)	293 (68.62%)
Region, $N$ (%)							$<$ 0.001
North	634 (29.53%)	361 (34.78%)	80 (37.38%)	359 (29.00%)	372 (35.63%)	186 (43.56%)
South	1513 (70.47%)	677 (65.22%)	134 (62.62%)	879 (71.00%)	672 (64.37%)	241 (56.44%)
Age group, years, $N$ (%)							$<$ 0.001
30–39	338 (15.74%)	152 (14.64%)	38 (17.76%)	40 (3.23%)	40 (3.83%)	26 (6.09%)
40–49	762 (35.49%)	405 (39.02%)	86 (40.19%)	196 (15.83%)	205 (19.64%)	111 (26.00%)
50–59	635 (29.58%)	318 (30.64%)	59 (27.57%)	365 (29.48%)	349 (33.43%)	146 (34.19%)
60–69	288 (13.41%)	125 (12.04%)	27 (12.62%)	372 (30.05%)	293 (28.07%)	87 (20.37%)
$\geqslant$ 70	124 (5.78%)	38 (3.66%)	4 (1.87%)	265 (21.41%)	157 (15.04%)	57 (13.35%)
Hypertension, $N$ (%)	96 (4.47%)	72 (6.94%)	21 (9.81%)	774 (62.52%)	725 (69.44%)	333 (77.99%)	0
Diabetes, $N$ (%)	0 (0.00%)	0 (0.00%)	0 (0.00%)	142 (11.47%)	122 (11.69%)	83 (19.44%)	$<$ 0.001
Dyslipidemia, $N$ (%)	497 (23.15%)	399 (38.44%)	118 (55.14%)	426 (34.41%)	518 (49.62%)	260 (60.89%)	$<$ 0.001
Hyperuricemia, $N$ (%)	199 (9.27%)	159 (15.32%)	45 (21.03%)	176 (14.22%)	261 (25.00%)	127 (29.74%)	$<$ 0.001

Table 2

The 10-year ASCVD risk scores and grades of different metabolic obesity phenotypes^a

	MHNW	MHOW	MHO	MUNW	MUOW	MUO	$P$ -value
The 10-year ASCVD	2.55 $\pm$ 2.71	2.82 $\pm$ 2.71	2.84 $\pm$ 2.49	7.83 $\pm$ 5.72	8.19 $\pm$ 5.92	9.01 $\pm$ 6.66	$<$ 0.001
risk scores
ASCVD risk, $N$ (%)							$<$ 0.001
Low-medium risk	2088 (97.25%)	1008 (97.11%)	210 (98.13%)	892 (72.05%)	727 (69.64%)	283 (66.28%)
High risk	59 (2.75%)	30 (2.89%)	4 (1.87%)	346 (27.95%)	317 (30.36%)	144 (33.72%)

^a. The 10-year ASCVD risk scores and grades of different metabolic obesity phenotypes were evaluated based on China-PAR.

Figure 1.

Bayesian network model DAG plot of metabolic obesity phenotype and 10-year risk of ASCVD.

Figure 2.

Inference and gender subgroup analysis of the 10-year risk of ASCVD in different metabolic obesity phenotypes.

ASCVD risk. There was no significant increase in the risk of ASCVD in the MHO group compared with the MHNW group (3.04%, 95% CI: 1.64–5.42; vs. 2.37%, 95% CI: 1.92–2.94). However, in the metabolically unhealthy groups, the high-risk risk of ASCVD increased with elevated BMI levels; in the MUNW group (normal weight), the ASCVD high risk was 24.63% (95% CI: 22.81%–26.55%), while in the MUO group (obese) the ASCVD high risk was 40.2% (95% CI: 36.64%–43.86%).

After stratified by sex, there was no significant increase in the risk of ASCVD in the MHO group compared with the MHNW group in men. In women, the 10-year risk of ASCVD in the MHO group was 4.14%, which was significantly higher than that in the MHNW group (1.10%, Fig. 2).

4. Discussion

In this study, we observed that the conditional probability of 10-year ASCVD risk in the Chinese MHO population was 3.04%. There was no significant increase in the risk of ASCVD in the MHO group compared with the MHNW group, while the 10-year ASCVD risk increased significantly among metabolically unhealthy groups. Besides, the significant associations between MHO phenotypes and 10-year ASCVD risk were only found in females. The new diagnostic criteria can accurately distinguish “healthy” obese people, but there may be some gender heterogeneity.

At present, MHO is an important research direction in the field of obesity, and the focus is mainly on whether MHO can be defined as a benign category of obesity. Based on the previous theoretical definition, there is no significant difference in the risk of CVD or other adverse clinical outcome events between the MHO and normal populations [7, 8, 21, 22]. For example, Lin et al. found that stable MHO status did not increase ASCVD risk based on a Chinese cohort study with an average follow-up of 4.4 years [8]. Similarly, Mongraw-Chaffin et al. found no significant increase in the risk of CVD in MHO groups compared with MHNW groups based on the USA MESA prospective cohort study [7]. However, recently, some large prospective cohort studies have shown that the MHO group is significantly associated with poor prognosis of CVD and the risk of all-cause mortality [5, 23, 24]. Two prospective cohort studies from the US Nurses’ Health Study (5) and the UK Whitehall II [24] have confirmed that MHO individuals without CVD history at baseline were at increased risk for CVD after long-term follow-up. Similarly, in 2020, a national cohort study based on The China Kadoorie Biobank also showed that MHO individuals had an 8% higher risk of developing 10-year CVD than MHNW individuals [6]. The association of MHO phenotype with CVD risk varies from non-significant to statistically significantly positive.

Results on CVD and mortality risks of MHO remain inconsistent. Using varying definitions for metabolically healthy might be responsible for the observed heterogeneity. It is reported that there are more than 30 different diagnostic criteria for MHO, according to which the prevalence rate of MHO ranges from 1.4% to 40.2% [25]. Therefore, diagnosing MHO scientifically and reasonably is an essential prerequisite for studying the relationship between MHO and other metabolic obesity phenotypes and disease prognosis. Zembic et al. found that the phenotype of MHO based on conventional MetS or HOMA-IR diagnosis has different results on CVD risk in the NHANES-III and UK Biobank cohorts. Thus, they proposed new diagnostic criteria for MHO and confirmed that MHO was not associated with increased CVD and all-cause mortality, suggesting that the new diagnostic criteria are superior in distinguishing between at-risk and not-at-risk obese individuals [10]. To our knowledge, this study was the first to explore the association of metabolically obesity phenotypes with a 10-year risk of ASCVD in a Chinese population using a BN model. According to the new criteria, about 1/3 of obese individuals are classified as MHO phenotype; compared with MHNW individuals, there was no significant increase in ASCVD risk in the MHO population. On the contrary, the risk of ASCVD in metabolically unhealthy was significantly elevated and independent of BMI levels. The results of the latest Korean community population cohort with new diagnostic criteria were consistent with this study [26], which found no significant association between MHO and CVD risk (subclinical atherosclerosis). Furthermore, we further explored the relationship between MHO and 10-year ASCVD in different genders, and found no significant association of MHO and ASCVD risk in males. We speculate that the new diagnostic criteria may be gender-specific. However, since this study is only a cross-sectional study, it is necessary to carry out relevant cohort studies in the future to verify this conclusion.

The use of BN model in this study helps reveal the complex network risk mechanism between the pathogenesis and factors of ASCVD. In addition to the direct relationship between metabolic status and BMI level and ASCVD, dyslipidemia, diabetes, eGFR, and other variables can indirectly affect the risk of ASCVD by mediating the relationship between metabolic status and sex. In terms of data interaction, compared with conventional statistical models, the BN model can further describe the relationship between factors and the impact of interaction on the risk of ASCVD through the network diagram between variables, and does not affect the interpretability of the network. In addition, another important advantage of the BN model is that the complex model is decomposed into multiple sets of refined simple models through Markov blanket theory, making these simple models easier to calculate and analyze.

This study has some limitations. First, although CHNS is a long-term follow-up cohort study, it only uses its 2009 data set, which belongs to cross-sectional data, so the directed arc direction between nodes or variables only represents probability dependence, not causality. Further studies with follow-up cohort data are needed. Second, this study only covers the Chinese population, so the extrapolation and applicability of the conclusions in other ethnic populations must be treated cautiously.

5. Conclusions

In summary, this study performed a BN to construct a 10-year ASCVD risk model of metabolic obesity phenotype. Meanwhile, the inference results showed that people with MHO classified by the new diagnostic criteria are not at increased risk for ASCVD. In the future, follow-up cohort data will be used to verify the new definition. This study helps identify factors dynamically stabilizing MHO and implement targeted interventions for these factors as early as possible.

Funding

The research was supported by the National Natural Science Foundation of China.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Footnotes

Acknowledgments

We thank all the participants in this study and all the collaborators of the research hospital.

Conflict of interest

The authors declare that they have no conflict of interest.

References

Collaboration

NCDRF

. Trends in adult body-mass index in 200 countries from 1975 to 2014: A pooled analysis of 1698 population-based measurement studies with 19.2 million participants. Lancet. 2016; 387(10026): 1377-1396.

Dai

Alsalhe

Chalghaf

Riccò

Bragazzi

. The global burden of disease attributable to high body mass index in 195 countries and territories, 1990–2017: An analysis of the Global Burden of Disease Study. PLoS Med. 2020; 17(7): e1003198.

Williams

Mesidor

Winters

Dubbert

Wyatt

. Overweight and obesity: Prevalence, consequences, and causes of a growing public health problem. Curr Obes Rep. 2015; 4(3): 363-70.

Karelis

Brochu

Rabasa-Lhoret

. Can we identify metabolically healthy but obese individuals (MHO). Diabetes Metab. 2004; 30(6): 569-72.

Eckel

Kuxhaus

Stefan

Schulze

. Transition from metabolic healthy to unhealthy phenotypes and association with cardiovascular disease risk across BMI categories in 90 257 women (the Nurses’ Health Study): 30 year follow-up from a prospective cohort study. Lancet Diabetes Endocrinol. 2018; 6(9): 714-724.

Gao

Guo

Bian

Yang

, et al. Metabolically healthy obesity, transition to unhealthy metabolic status, and vascular disease in Chinese adults: A cohort study. PLoS Med. 2020; 17(10): e1003351.

Mongraw-Chaffin

Foster

Anderson

CAM

Burke

Haq

Kalyani

, et al. Metabolically healthy obesity, transition to metabolic syndrome, and cardiovascular risk. J Am Coll Cardiol. 2018; 71(17): 1857-1865.

Lin

Zhang

Jiang

, et al. Transition of metabolic phenotypes and risk of subclinical atherosclerosis according to BMI: A prospective study. Diabetologia. 2020; 63(7): 1312-1323.

Opio

Croker

Odongo

Attia

Wynne

McEvoy

. Metabolically healthy overweight/obesity are associated with increased risk of cardiovascular disease in adults, even in the absence of metabolic risk factors: A systematic review and meta-analysis of prospective cohort studies. Obes Rev. 2020; 21(12): e13127.

10.

Zembic

Eckel

Stefan

Baudry

Schulze

. An empirically derived definition of metabolically healthy obesity based on risk of cardiovascular and total mortality. JAMA Netw Open. 2021; 4(5): e218505.

11.

Yan

Zhang

Gordon-Larsen

, et al. The expanding burden of cardiometabolic risk in China: The China Health and Nutrition Survey. Obes Rev. 2012; 13(9): 810-21.

12.

Fang

Shen

. Dietary Nonheme, Heme, and Total Iron Intake and the Risk of Diabetes in Adults: Results From the China Health and Nutrition Survey. Diabetes Care. 2020; 43(4): 776-784.

13.

Zeng

Dong

Zhao

Chen

Song

, et al. Optimal cut-off values of BMI, waist circumference and waist: Height ratio for defining obesity in Chinese adults. Br J Nutr. 2014; 112(10): 1735-44.

14.

Yang

Chen

Huang

, et al. Predicting the 10-Year Risks of Atherosclerotic Cardiovascular Disease in Chinese Population: The China-PAR Project (Prediction for ASCVD Risk in China). Circulation. 2016; 134(19): 1430-1440.

15.

Yang

Chen

Cao

Liu

, et al. Risk stratification of atherosclerotic cardiovascular disease in Chinese adults. Chronic Dis Transl Med. 2016; 2(2): 102-109.

16.

Chinese Society of Cardiology of Chinese Medical Association. Chinese Guideline on the Primary Prevention of Cardiovascular Diseases. Cardiol Discov. 2021; 1.02: 70-104.

17.

Pearl

. Probabilistic Reasoning in Intelligent Systems. San Francisco: Morgan Kaufmann Publishers, Inc. 1988.

18.

Tian

Che

Liu

. A Bayesian Network Analysis of the Probabilistic Relationships Between Various Obesity Phenotypes and Cardiovascular Disease Risk in Chinese Adults: Chinese Population-Based Observational Study. JMIR Med Inform. 2022; 10(3): e33026.

19.

Rezaeiahari

Brown

Ali

Datta

Tilford

. Understanding racial disparities in severe maternal morbidity using Bayesian network analysis. PLoS One. 2021; 16(10): e0259258.

20.

Scutari

. Learning Bayesian Networks with the bnlearn R Package. Journal of Statistical Software. 2010; 35(3): 1-22.

21.

Appleton

Seaborn

Visvanathan

Hill

Gill

Taylor

, et al. Diabetes and cardiovascular disease outcomes in the metabolically healthy obese phenotype: A cohort study. Diabetes Care. 2013; 36(8): 2388-94.

22.

Soriguer

Gutierrez-Repiso

Rubio-Martin

García-Fuentes

Almaraz

Colomo

, et al. Metabolically healthy but obese, a matter of time? Findings from the prospective Pizarra study. J Clin Endocrinol Metab. 2013; 98(6): 2318-25.

23.

Hansen

Netterstrom

Johansen

Rønn

Vistisen

Husemoen

, et al. Metabolically healthy obesity and ischemic heart disease: A 10-Year Follow-Up of the Inter99 Study. J Clin Endocrinol Metab. 2017; 102(6): 1934-1942.

24.

Hinnouho

Czernichow

Dugravot

Nabi

Brunner

Kivimaki

, et al. Metabolically healthy obesity and the risk of cardiovascular disease and type 2 diabetes: The Whitehall II cohort study. Eur Heart J. 2015; 36(9): 551-9.

25.

Franco

Morais

Cominetti

. Normal-weight obesity syndrome: Diagnosis, prevalence, and clinical implications. Nutr Rev. 2016; 74(9): 558-70.

26.

Seo

Cho

Seo

Ahn

Hong

, et al. Association between Metabolically Healthy Obesity and Subclinical Atherosclerosis in the Cardiovascular and Metabolic Diseases Etiology Research Center (CMERC) Cohort. J Clin Med. 2022; 11(9): 2440.

A Bayesian network analysis of the probabilistic relationships between metabolically healthy obesity and cardiovascular disease risk under new diagnostic criteria

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSION:

Keywords

1. Introduction

2. Materials and methods

2.1 Study population

2.2 Data collection

2.3 Diagnostic criteria of metabolically healthy phenotype

2.4 ASCVD risk assessment and grouping

2.5 Statistical analysis

3. Results

3.1 Characteristics of the study population

3.2 ASCVD risk among different metabolic obesity phenotypes

3.3 Construction of BN model

3.4 Reasoning of BN model

Table 1 General characteristics among the study populationa

5. Conclusions

Funding

Data availability

Footnotes

Acknowledgments

Conflict of interest

References

Table 1
General characteristics among the study population^a