Novel use of structural equation modelling to examine diet and metabolic traits associated with microvascular endothelial dysfunction in middle-aged Chinese males: a cross-sectional study

Abstract

Background

The present study aims to use structural equation modelling (SEM) with multiple regression pathways to examine direct and indirect links from diet and metabolic traits to microvascular endothelial dysfunction (ED) among middle-aged Chinese males.

Methods

The study was conducted in middle-aged Chinese males, who underwent a health check-up between 2018 and 2019. Data on lifestyle behaviour factors (physical activity, diet pattern, sleep quality and diet data underwent factor analysis in advance) and metabolic risk factors referring to metabolic traits were introduced into the SEM to examine inter-relationship among these factors and their association with ED, as evaluated by the reactive hyperaemia index (RHI).

Results

Both exploratory factor analysis and confirmatory factor analysis identified two major dietary patterns: ‘prudent pattern’ and ‘western pattern’. The univariate test suggested that only triglycerides (TGs) and prudent dietary pattern were directly associated with RHI. Furthermore, prudent dietary pattern had an indirect association with RHI via TG (prudent diet→TG: β=−0.15, p<0.05; TG→RHI: β=−0.17, p<0.001). As to confirming the hypothesised association between variables apart, physical activity frequency was correlated to the decrease in TG (β=−0.29, p<0.001), but had no direct correlation to RHI.

Conclusion

The network of direct and indirect associations among diet pattern and cardiometabolic risk factors with RHI measured ED among middle-aged males. The most significant modifiable factors identified were TG and prudent diet pattern, which needs to be targeted as preventive strategies for early microvascular impairment.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Introduced seldomly used diet pattern as risk factors using factor analysis.

• Presented a network consisting of modifiable risk factors using structural equation models methods.

• Not only direct but also indirect effects were also brought into consideration.

Background

Endothelial dysfunction (ED), which is a pathological condition characterised by an imbalance between endogenous vasodilating and vasoconstricting substances, is one of the earliest measurable markers of atherosclerosis,¹ and a powerful predictor of future cardiovascular (CV) events.² ED is usually a systemic disorder, but the small arteries could represent the basis for assessment of the functional capability of the micro-circulatory system due to their easy accessibility. Based on this principle, the reactive hyperaemia-peripheral arterial tonometry (RH-PAT) located in the fingertips is a non-invasive signal technique, which is correlated to the endothelial function of microcirculation via measuring the increase in digital pulse volume amplitude (PVA) during reactive hyperaemia relative to baseline,³ and nitric oxide (NO) plays a central role in the augmentation of PVA.⁴ In light of the predictive value and easy detection, RH-PAT is suitable for health check-up populations for the prognostics of CV disease (CVD), and the risk factors of RH-PAT-measured ED in this population are also worthy of study to prevent ED.

If not all, the metabolic risk factors of CVD were generally associated with ED.⁵ However, as to RH-PAT-measured ED, there are controversies. Some correlations were found in specific populations. For instance, diabetic adolescents with poorly controlled blood glucose had a worse ED reflected by lower reactive hyperaemia index (RHI) score of RH-PAT.⁶ RHI was significantly lower in obese adolescents than in normal-weight adolescents.^{7 8} A weak correlation was found between microvascular ED and body mass, and systolic (SBP) and diastolic blood pressures (DBP), triglyceride (TG), high-density lipoprotein cholesterol (HDL-c) and glucose in middle-aged men.⁹ RHI was associated with BP and HDL-c in chronic kidney disease and rheumatoid arthritis patients.^{10 11} However, a negative profile also existed: RHI measured the microvascular ED, but failed to show the correlations with a risk factor profile (hypercholesterolaemia, hypertension, diabetes and overweight/obesity) in the asymptomatic population.^12–14 In type 2 diabetic patients, the logarithmic-scaled RHI did not correlate with fasting plasma glucose, low-density lipoprotein cholesterol and HDL-c, TG levels, or SBP and DBP.^{15 16}

In the present study, the investigators aimed to explore the association between metabolic risk factors and RH-PAT-measured ED. Furthermore, the behavioural factors were introduced to conceptualise a more advanced structure of predisposing factors of ED as a model. Since cardiometabolic factors were affected by prior level of lifestyle including diet pattern, activity,¹⁷ we speculated that direct and/or indirect relation mediated by cardiometabolic factors might exist between lifestyle and ED. The investigators applied the statistical approach of structural equation modelling (SEM)¹⁸ to simultaneously analyse all relevant regression pathways, in order to better understand the role of these factors in the development of ED. Considering that age and gender are fixed factors and potentially exert significant effect on ED,^{13 19} the investigators narrowed the population scale to middle-aged males to limit their affection on the counteracting effect of the present core factors.

By using SEM, the present study is the first to investigate the direct and indirect association of potential behaviour (diet, physical activity and sleep) and cardiometabolic risk factors with RH-PAT-measured ED among middle-aged men.

Methods

Subject and design

A cross-sectional study was conducted. The medical records of 306 participants, who underwent a health check-up at the Health Examination Management Center of Third Xiangya Hospital from 1 November 2018 to 31 August 2019, were included. The inclusion criteria were males between 45 and 59 years old (WHO-recommended middle-aged group) underwent endothelial function examination. The exclusion criteria were symptomatic CVD.

Definitions and measures

SBP and DBP were measured between 08:00 and 10:00, following the guidelines from the American Health Association.²⁰ All measurements were conducted using an automatic digital BP monitor (Omron 9020). The participants were measured after a 10 min rest period, with their feet straight on the ground and their back and arm supported, and with the antecubital fossa at the level of the heart. The maximum cuff inflation was calculated by adding 30 mm Hg to the pulse obliteration pressure, and the cuff was deflated at a constant rate of 2–4 mm Hg per second. Venous blood was collected in the morning after overnight fasting. The serum samples were stored at 4°C, and were subjected to testing (Hitachi 7170s autoanalyser) within 2 days, according to the instruction of the analyser. Fasting blood glucose (FBG), TG and HDL-c were measured using the enzymatic method with the full-automatic biochemical analyser (Hitachi 7170s). The investigators chose to analyse five cardiometabolic components of the metabolic syndrome as categorical risk factors: body mass index (BMI), TG, HDL-c and FBG were categorised into three groups, according to the standards of the Guidelines for prevention and control of overweight and obesity in Chinese adults (2004), and the Guidelines on prevention and treatment of blood lipid abnormality in Chinese adults (2018): BMI (<24, 24–28, ≥28 kg/m²), TG (<1.70, 1.70–2.25, ≥2.26 mmol/L), HDL-c (<1.04, 1.04–1.55, ≥1.55 mmol/L), FBG (<6.1, 6.1–7.0, >7.0 mmol/L or taking antidiabetic medication). BP was binary categorised as <140/90 mm Hg, ≥140/90 mm Hg/or taking antihypertensive medication.

Data about concerning diet, physical activity and sleep quality over the past years were extracted from the ‘self-rated health measurement scale in health check-up’ recommended by the Chinese Health Management Association.²¹ This questionnaire was designed and administered by medical professionals, who collected data on the previous year. Physical activity was defined as moderate-intensity aerobic exercise, including fast walking, running, bicycle riding, rope skipping and swimming. All participants were assigned to categories, according to the activity frequency per week (0 time, 1–2 times, 3–5 times and >5 times) and average duration each time (0 min, <30 min and ≥30 min). Subjective evaluation of sleep quality was categorised into levels of poor, medium and excellent, according to their own state of difficulty of falling asleep, early awakening, dreaminess, easily awakening and shortened sleep duration. The 12 diet items in the scale included common diet behaviour and diet habits (online supplemental table 1): anticipating dinner party per week ((1) ≤1 time, (2) 2–3 times, (3) 4–5 times and (4) >5 times); midnight snack per week ((1) never, (2) ≤1 time and (3) >1 time); three meals on time per week ((1) almost every time, (2) failed 2–3 times and (3) failed >3 times); times of milk and alcoholic drink per week ((1) never, (2) 1–2 times, (3) 3–5 times and (4) ≥6 times); fruit, eggs and legume product per week ((1) ≤2 times, (2) 3–5 times and (3) ≥6 times); vegetable per day ((1) <100 g, (2) 100–200 g, (3) >200 g); meat per day ((1) <50 g, (2) 50–100 g and (3) >100 g); sugary beverage and coffee per week ((1) never, (2) 1–2 times and (3) ≥3 times).

ED was measured using the ENDOPAT2000 device (Itamar Medical, Caesarea, Israel), which recorded the digital pulsatile volume changes without involving painful and risky invasive procedures. Micro-arterial tonometry signals were obtained from participants resting in the supine position in a quiet, temperature-controlled room after overnight fasting. Subjects were refrained from smoking and vigorous activity for 12 hours before the examination. Two finger probes were placed on one finger of each hand. The baseline pulse amplitude was recorded during the first 5 min, followed by the 5 min induction of ischaemia induced by inflating upper-arm BP cuff to 60 mm Hg above SBP, with the opposite arm serving as a control, and the occlusion of blood flow was confirmed by the reduction of the ENDOPAT tracing to zero. After 5 min, the cuff was deflated, and the pulsatile tracing was recorded for another 5 min. RHI was automatically calculated using the computer algorithm of the ratio of the hyperaemia and baseline pulse amplitude after control-arm correction. RHI<1.67, determined in previous studies, was defined as ED.^{22 23}

Statistical analysis

Exploratory factor analysis

Exploratory factor analysis (EFA) was first used to explore the latent construct of diet on 50% randomly selected responders. The extraction method of principle axis factoring followed by orthogonal Varimax rotation was conducted to estimate the factor loading, and determine the category and component of latent variables.²⁴ Items of food and dietary habit with absolute factor loading of ≥0.30 entered the corresponding latent variable group. Only latent variables constructed with three or more items could be selected for further analysis to reduce measurement errors.

Confirmatory factor analysis

Another 50% responders were analysed. Confirmatory factor analysis (CFA) was introduced to test the underlying construction of the food groups, and verify the latent construct explored by EFA.

Structural equation modelling

Structural modelling approach were used. First, the conceptual model that specified the correlations among dependent and independent variables is shown in figure 1. The a priori paths emanating from the measured univariate and unmeasured latent variables potentially direct or indirect affecting RHI-measured ED were constructed. The possible two-way correlation that implied two factors that were mutually connected were also given consideration. The SEM was fitted by the maximum likelihood estimation method, the goodness-of-fit of the CFA and SEM. The χ² test, normed χ² test (χ²/df), root mean square error of approximation, goodness-of-fit statistic, the adjusted goodness-of-fit statistic and the comparative fit index were also evaluated to guarantee that the proposed model can be used. All analyses were conducted using IBM SPSS Statistics V.21.0 and Amos V.23.0. P values <0.05 were considered statistically significant.

Figure 1.

The structural model: testing the association of life-style behavioural and cardiometabolic factors with RHI. BMI, body mass index; BP, blood pressure; FBG, fasting blood glucose; HDL-c, high-density lipoprotein cholesterol; RHI, reactive hyperaemia index; TG, triglyceride.

χ² tests were used to compare the frequency distribution of categorical data for metabolic risk factors. The factor score for each dietary pattern were calculated by weighting the consumption of each food item with the corresponding factor loading, and summing the resulting values. Then, the comparison between two groups evaluated by the factor score were made using Student’s t-test.

Patient and public involvement

Since this study is a secondary study based on other studies, there will be no direct patient or public involvement in this study.

Results

As shown in online supplemental table 2, the mean age of the participants was 51.07±3.98 years old, and the decrease in RHI was found in 39.2% of participants. All diet items were randomly incorporated into the EFA on 50% of 306 participants. Four constructs were extracted and the Varimax rotated factor loadings are presented in table 1, and the thresholds of absolute factor loading of each item, corresponding to common factors, were above 0.3. Factor 1 included five items, namely, fruit, legume product, milk, eggs, vegetable and this was labelled as ‘prudent dietary pattern’. Factor 2 included three items, namely, dinner party, meat, alcoholic drink and was labelled as ‘western dietary pattern’. Factors 3 and 4 were eliminated for further CFA, since these contained only two items.

Table 1.

Factor loading matrix for diet items using exploratory factor analysis

	Component
Items	1	2	3	4
Fruit	0.460	3.23	0.044	0.044
Legume product	0.372	0.016	0.028	0.010
Milk	0.540	0.111	0.288	0.064
Eggs	0.560	0.108	0.007	0.189
Vegetable	0.334	0.035	0.194	0.087
Alcoholic drink	0.141	0.681	0.275	0.106
Dinner party	0.116	0.622	0.250	0.003
Meat	0.169	0.420	0.090	0.144
Sugared beverages	0.066	0.171	0.663	0.068
Coffee	0.126	0.062	0.472	0.007
Night snack	0.001	0.238	0.022	0.863
Three meals on time	0.214	0.010	0.076	0.306

Factor classification and loading after exploratory factor analysis for each dietary item, with factor loading >0.4 as the inclusion threshold, and bold marked.

Table 2 displayed the measurement model for ‘prudent dietary pattern’ and ‘western dietary pattern’, with observed variables previously recognised in EFA. CFA was used to test the factor structure evaluated by EFA using the data of the remaining 50% of participants. The goodness of fit indices reported in the note of table 2 indicating the fit of two diet measurement models. CFA confirmed the measurement model of the ‘prudent dietary pattern’ and ‘western dietary pattern’ construct.

Table 2.

Factor loading and reliability of identified factors verified by standardised regression weights of confirmatory factor analysis

Prudent diet pattern		Western diet pattern
Items	Factor loading	Items	Factor loading
Milk	0.40	Meat	0.35
Eggs	0.47	Alcoholic beverage	0.53
Fruits	0.59	Dinner party	0.55
Vegetable	0.44
Legume products	0.52

Fit indices of confirmatory factor analysis model: χ²=29.59, df=20, χ²/df=1.48, p=0.077, goodness-of-fit statistic=0.954, adjusted goodness-of-fit statistic=0.918, comparative fit index=0.889, root mean square error of approximation=0.056.

The a priori SEM for evaluating the association of life-style behavioural and cardiometabolic factors with RHI are depicted in figure 1. Based on the theory of CV risk factors impaired artery function and structure, we intended to explore the destructive effect of typical metabolic risk factors including high BP and blood glucose, dyslipidaemia. These risk factors were also affected by prior level of lifestyle mainly including diet pattern, activity and sleeping. Thus, we built the hierarchical SEM. Later, model modifications were made between factors for better fit according to modification indices, and unnecessary paths were eliminated according to p value.

Baseline anthropometric and biochemical data were listed in online supplemental table S1. Table 3 indicates the behavioural and cardiometabolic difference using the univariate analysis according to the presence of impaired RHI. Sleep quality, physical activity frequency and duration, western dietary pattern and most cardiometabolic factors had no statistical difference between the groups, with or without RHI decrease, However, the prevalence of impaired RHI declined in participants with a lower TG and a higher factor score of prudent dietary pattern.

Table 3.

Characteristics of participants according to the presence of endothelial dysfunction

	RHI<1.67	RHI≥1.67	P value
	(n=120)	(n=186)
Activity frequency (times per week)
0	41	65	0.089
1–2	34	32
3–5	29	63
>5	16	26
Activity duration (minutes each time)
0	41	65	0.224
<30	21	20
≥30	58	101
Sleep quality
Excellent	51	61	0.147
Medium	55	92
Poor	14	33
Blood pressure (mm Hg)
<140/90	81	117	0.411
≥140/90+ medication	39	69
Body mass index
Normal	34	66	0.427
Overweight	62	86
Obesity	24	34
Triglyceride (mmol/L)
<1.7	46	112	0.001
1.7–2.26	31	30
≥2.26	43	44
HDL-c (mmol/L)
<1.04	18	22	0.213
1.04–1.55	87	127
≥1.55	15	37
Fasting blood glucose (mmol/L)
<6.1	89	142	0.546
6.1–7.0	16	27
>7.0	15	16
Factor score
Prudent diet pattern	0.12±0.06	0.08±0.06	0.023
Western diet pattern	0.01±0.07	0.008±0.06	0.829

χ² tests were used to compare the frequency distribution of categorical data for metabolic risk factors, and mean±SD was used in continuous data. P<0.05 was considered significant.

HDL-c, high-density lipoprotein cholesterol; RHI, reactive hyperaemia index.

Based on the theory of CV risk factors impaired artery function and structure, we intended to explore the destructive effect of typical metabolic risk factors including high BP and blood glucose, dyslipidaemia. These factors were also affected by prior level of lifestyle mainly including diet pattern, activity and sleeping. Thus, we built this SEM. Besides, model modification were made between metabolic risk factors for better fit according to modification indices.

The best-fit modified SEM was somewhat different from the hypothesised model (figure 2). Changes were made from the present hypothesised model, according to the crisis ratio (CR) value and p value. The investigators stepwise eliminated the paths with the unstandardised CR lower than 1.4. The path coefficients on a one-way arrow among these multiple regression pathways supported the results of that univariate analysis, in which only TG and prudent dietary pattern were directly associated with RHI (β=−0.17 and 0.16, p<0.05). Furthermore, RHI had an indirect association with prudent dietary pattern via TG (prudent diet→TG: β=−0.15, p<0.05; TG→RHI: β=−0.17, p<0.001). As to confirming the hypothesised association between variables apart, physical activity frequency was negatively correlated to increased TG (β=−0.29, p<0.001), but there was no direct correlation to RHI. It turned out that sleep quality and diet-related BP and BMI had no association with ED. Meanwhile, a two-way correlation was rebuilt in cardiometabolic or behaviour factors, according to the modification indices in covariance. The western diet pattern was not negatively correlated with prudent diet, but has an activity frequency. The model fit indices below figure 2 for SEM revealed that the tested model is generally qualified for fitting the data.

Figure 2.

The final structural model indicating standardised regression weights after testing the association of life-style behavioural and cardiometabolic factors with RHI. The fit indices were generally acceptable: χ²=232.416, df=110, χ²/df=2.113, p=0.000, goodness-of-fit statistic=0.918, adjusted goodness-of-fit statistic=0.886, comparative fit index=0.845, root mean square error of approximation=0.060. The solid-line pathway represented statistically important links with p<0.05, and the dashed lines means that the path has a crisis ratio of >1.4, but failed to pass the hypothesis tests with p≥0.05. *p<0.05, **p<0.01. BMI, body mass index; BP, blood pressure; FBG, fasting blood glucose; HDL-c, high-density lipoprotein cholesterol; RHI, reactive hyperaemia index; TG, triglyceride.

Discussion

SEM allows for multiple linear equations, including direct and indirect effects, and latent variables, features not allowed by traditional regression methods.^{25 26} Our data derived common factors contained cluster of dietary observed variables, in this case, SEM was a proper way to incorporate integrated variables.

To our knowledge, the present study presented the first examination of the direct and indirect effects of modifiable risk factors on ED using SEM. The appropriate fit required the original hypothesised model to undergo changes. After the elimination of factors that made no contribution, only TG and prudent dietary pattern were directly associated with RHI. Furthermore, RHI had an indirect association with prudent dietary pattern via TG. In the field of health management, risk prevention and management of CVD are vital important study topics, especially learning risk factors when diseases were detected at the earliest stage. Another more widely applied macrovascular ED measurement is flow-mediated dilation (FMD), which is estimated as the percent change of brachial artery diameter at maximal dilation during hyperaemia, when compared with the baseline value. ED in the conduit arteries might be more important in subjects with existing atherosclerosis, while that in resistance small vessels might be an early indicator of arteriosclerosis risk.^{27 28} However, there was no correlation found between FMD and RHI when the two parameters were simultaneously measured.²⁹ Thus, the risk factors of RHI-measured ED, as an early marker of vascular impairment, are still worth digging in relatively healthy populations when undergoing health checkups.

It turned out that only TG in the metabolic syndrome cluster was associated with microvascular ED in the present health-checkup population, which was inconsistent with part of the previous reports listed in the introduction section. The observed strength of each metabolic risk factor significantly varied in several populations reported in previous studies and the present study. According to the heterogeneous strength of RHI-related risk factors, the investigators analysed the potential causality as follows: (1) the estimated ED burden originated from cardiometabolic risk factors should be affected by intrinsic characteristics of specific population in the magnitudes of their effects. Indeed, the sensitivity of endothelial function to risk injury varied owing to the differentiation of population age, gender, race, region and disease category.^30–32 The first four stratifications that referred to genetic background or physiological properties were complex and non-modifiable. Concerning disease background, most previous studies revealed subjects with cardiometabolic disease or autoimmune disease as mentioned in the introduction. When comparing to subjects with diseases, the health-checkup population possessed relatively narrow range of abnormal risk factor levels, which were supposed to be below their counterpart under the pathological state. Thus, the SEM revealed such correlation eliminating the strength of BP, blood glucose and other lipid index. Furthermore, unlike the present SEM with multiple regression pathways, most previous evaluations were univariate analyses without adequate adjustment, which might lead to inflated effect sizes. In addition, a casual evidence supported the exaggerated power: lowering BP and blood glucose using antihypertensive and antidiabetic medications for weeks does not improve the RHI-measured ED.^33–36 As to the negative effect of TG, the evidences were far more than robust. TG level was the only factor that significant associated with RHI with exclusion simultaneously for other cardiometabolic risk factors in healthy adults³⁷; acute hypertriglyceridemia in rats was accompanied with the impairment of NO-dependent relaxation in artery,³⁸ and a short-term use of fenofibrate lowering TG could achieve the efficacy of RHI improvement.³⁹

Regarding the life-style behaviour concerning diet, physical activity and sleep quality, this is the first study that explored the effect of diet pattern on microvascular ED. It was observed that prudent diet pattern consisted of recognised healthy food indirectly ameliorated the impairment of ED through lowering TG, meanwhile, a direct beneficial effect on ED was also found. The underlying mechanism for this advantage of prudent diet might be the high levels of contained vitamins, since it has been proven that vitamins C, D and E restored the macrovascular or microvascular ED.^40–42

With regard to specific food, the intake of blueberry, artichoke improved ED, the fruit and vegetable contained polyphenol metabolites that might account for this advantage.^{43 44} It remains unknown whether protein-contained milk, egg and legume are involved in ED improvement, and more exploration are needed. Indeed, in south of China, although a traditional dinner party generally tends to consume all kinds of alcoholic beverage and plenty of meat, no previous data revealed the effect of alcohol and meat on RHI-measured microvascular ED. However, several studies that evaluate the alcohol effect on the FMD index were controversial: chronic moderate to heavy alcohol consumption caused macrovascular ED in Asian men,^{45 46} even more, FMD was significantly impaired in light alcohol drinkers.⁴⁷ Conversely, the consumption of moderate and high alcoholic beverages was independently associated with better FMD, when compared with no alcohol consumption,^{48 49} especially when consuming red wine.⁵⁰ In general, the high dietary intake of food rich in antioxidants, such as specific fruits, vegetables and red wine, has a positive effect in improving FMD and endothelial function.⁵¹ Thus, the diet pattern should be taken into account, and the food type should be further subdivided if possible. Aerobic exercise provides CV benefits without doubt. A previous study reported that physical activity improves microvascular ED.^{7 52} Furthermore, the present study suggested that this advantage might be mediated by TG reduction. However, other mediated factors should be further explored.

The limitations of the present study are, as follows: (1) the data was cross-sectional. Therefore, the present hypothesised pathway did not reflect the timing dimension and causality. (2) The self-reported life-style behaviour data may also be subjectively affected by social desirability bias. That is, the participants are likely to over-report the ‘healthy’ behaviours. (3) When conducting the factor analysis, the accepted value of factor loading, the reliability and validity of CFA in latent construct were not as desirable as the standard threshold should be. This was due to the wide variations of food major items and the diet was divided into broad categories, thus, the close inter-correlation or—aggregation between these measured variables was hard to achieve. A similarly low variance could be found in all of the factor analysis concerning food category.^53–55

Conclusions

In conclusion, these present results revealed the network of direct and indirect associations among life-style behavioural (diet, physical activity and sleep) and cardiometabolic risk factors with RH-PAT-measured ED among middle-aged men underwent health-checkup. The most significant modifiable factors identified were TG and prudent diet pattern, which needs to be targeted as a preventive strategy for early microvascular impairment.

Data availability statement

Data obtained from the third Xiangya hospital, on reasonable request and with the approval of the third Xiangya hospital, data may be obtained from the corresponding author.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

This study involves human participants. The present study was conducted according to the principles expressed in the Declaration of Helsinki, and was approved by the Ethics Committee of the Third Xiangya Hospital (2020-S609). Participants gave informed consent to participate in the study before taking part.