Soy Consumption and the Risk of Type 2 Diabetes and Cardiovascular Diseases: A Systematic Review and Meta-Analysis

Soy is rich in plant protein, isoflavones, and polyunsaturated fatty acids. To clarify the associations between soy intake and type 2 diabetes (T2D) and cardiovascular diseases (CVDs) events, we performed a meta-analysis and review. A total of 1963 studies met the inclusion criteria, and 29 articles with 16,521 T2D and 54,213 CVDs events were identified by the eligibility criteria. During a follow-up of 2.5–24 years, the risk of T2D, CVDs, coronary heart disease, and stroke in participants with the highest soy consumption decreased by 17% (total relative risk (TRR) = 0.83, 95% CI: 0.74–0.93), 13% (TRR = 0.87, 95% CI: 0.81–0.94), 21% (TRR = 0.79, 95% CI: 0.71–0.88), and 12% (TRR = 0.88, 95% CI: 0.79–0.99), respectively, compared to the lowest sot consumption. A daily intake of 26.7 g of tofu reduced CVDs risk by 18% (TRR = 0.82, 95% CI: 0.74–0.92) and 11.1 g of natto lowered the risk of CVDs by 17% (TRR = 0.83, 95% CI: 0.78–0.89), especially stroke. This meta-analysis demonstrated that soy consumption was negatively associated with the risks of T2D and CVDs and a specific quantity of soy products was the most beneficial for the prevention of T2D and CVDs. This study has been registered on PROSPERO (registration number: CRD42022360504).


Introduction
Soy is a major source of plant protein, dietary fiber, polyunsaturated fatty acids, lecithin, stigmasterol and isoflavones. Soy protein is the only plant-derived complete protein and it accounts for 35-40% of soy, containing all the essential amino acids in animal proteins [1]. Its amino acid composition is similar to human essential amino acid composition, and its content is sufficient [1]. Animal protein is rich in a variety of essential amino acids, but also rich in fat and cholesterol, leading to a high incidence of various chronic diseases [2]. Additionally, soy protein does not contain cholesterol, and the contents of methionine and branched-chain amino acids (BCAAs) are low compared with animal proteins [3,4]. Although leucine or its metabolite β-hydroxy-β-methylbutyric acid can improve muscle function [5], aberrant metabolism of BCAAs and high circulating concentrations of BCAAs are a hallmark of metabolic disorders, including obesity, insulin resistance and type 2 diabetes mellitus, cancer, and heart failure [6,7]. Studies have found that a low BCAA diet can increase the survival rate of premature aging in mice, delay frailty, and promote metabolic health [8]. Restricting dietary BCAAs can increase the health span and life span of mice, which is a potential transformable intervention for the promotion of healthy aging [8]. Polyunsaturated fatty acids, lecithin, and stigmasterol are also the main components of soy. Studies have reported that these ingredients can reduce the total cholesterol and low-density lipoprotein cholesterol levels, improve diabetes and protect cardiovascular health [9,10].

Materials and Methods
This systematic review and meta-analysis was reported based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline [25]. We also adhered to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) reporting guideline [26].

Data Sources and Search Strategy
Three reviewers conducted a systematic search of all articles published up to 14 October 2022, of online databases, including Medline, PubMed, Web of Science core collection, and Embase. The research was restricted to human studies. Detailed search strategies can be found in Table S1. To avoid missing any publications, we also evaluated and checked all studies in the prior reviews and contacted the author of the original study.

Eligibility Criteria and Study Selection
Studies were included in this meta-analysis according to the following criteria: (1) the study design was a case-control or nested case-control or cohort, (2) the exposure factor was soybean, soy food, or soy protein consumption, (3) the outcome of interest was population aged more than 18 years with T2D or CVDs events, including coronary heart disease (CHD) and stroke, and those who reported odds ratio (OR), relative risk (RR), or hazard ratio (HR) associated with 95% confidence interval (95% CI) between the interested exposure and outcome. Exclusion criteria were as follows: (1) Case reports, editorials, letters to editors, comments, conference abstracts, short communications, reviews, cross-sectional, and animal studies were not considered. (2) If study participants were reported more than once, we selected the greatest number of cases or the result with Nutrients 2023, 15, 1358 3 of 18 a longer follow-up time. Three independent reviewers first selected relevant studies by screening the titles and then, reviewed the full texts according to the inclusion/exclusion criteria to decide the final relevant references to be included in the review. Any discrepancy between the three co-authors was resolved with a fourth author.

Data Extraction
Three investigators used a predefined form to extract information, including first author, publication year, database used, country, study design, sample size, duration of follow-up in cohort studies, type of Food Frequency Questionnaire (FFQ), exposure assessment, ascertainment of outcomes (CHD, stroke, or total CVDs), comparison, adjustment for confounding covariates, and the Newcastle-Ottawa Scale (NOS) [27] score. We extracted HR, RR, or OR those that reflected the maximum extent of adjustment for variables. Any results stratified by sex or endpoints were treated as two separate reports. We have tried to contact the authors to obtain some available research statistics, but we have not yet received a response.

Quality Assessment
The methodological quality of each study was independently evaluated by three reviewers using NOS [27]. This scale consists of three parameters of quality: four stars for the selection of participants, two stars for comparability, and three stars for assessment of exposure (for case-control studies) or outcomes (for cohort studies). A maximum of nine stars can be given to each study. In our systematic review, those that achieved a score of 0-4, 5-7, or 8 or more, were, respectively, considered to be low, moderate, and high-quality publications. Disagreements were resolved by discussing with a fourth author.

Statistical Analysis
Review Manager, version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) and Stata, Version 14.0 (Stata Corp, College Station, TX, USA) were used to perform all statistical analyses. A random-effects model was used to calculate total relative risk (TRR) and 95% CI for the highest category compared to the lowest category analysis of each study in this meta-analysis. Forest plot was created to visually assess the pooled result for each study and overall estimate. Based on the statistical test of I 2 , the Q-test was used to estimate the heterogeneity between studies. A p value less than 0.10 or an I 2 value greater than 50% indicated statistical heterogeneity. Meta-regression and subgroup analyses were used to analyze sources of heterogeneity. Sensitivity analyses were performed by excluding each study to assess the stability of the study. Egger's test and Begg's test were designed to assess publication bias.
Methods described by Greenland [28] and Orsini [29] were carried out to calculate the dose-response analysis between soy foods consumption and T2D/CVDs events. A two-stage random-effects dose-response meta-analysis was applied to examine a probable non-linear association between intakes of soy foods and T2D/CVDs events. A two-stage generalized least squares trend estimation method was employed to explore linear doseresponse relationships. p values were considered significant at the level < 0.05.

Study Characteristics and Quality Assessment
The main characteristics of the included studies were given in Table 1. All the 29 studies selected for quantitative analysis consisted of 1,660,304 participants during the follow-up of 2.5-24 years, including 16,521 T2D events (morbidity) and 54,213 CVDs events (morbidity and mortality), which were composed of 22,112 stroke events, 12,906 CHD events, and 19,195 other CVDs events. Based on the NOS, all the quality scores ranged from 5-9 and 19 studies were of high quality (Table S3).

Study Characteristics and Quality Assessment
The main characteristics of the included studies were given in Table 1. All the 29 studies selected for quantitative analysis consisted of 1,660,304 participants during the follow-up of 2.5-24 years, including 16,521 T2D events (morbidity) and 54,213 CVDs events (morbidity and mortality), which were composed of 22,112 stroke events, 12,906 CHD events, and 19,195 other CVDs events. Based on the NOS, all the quality scores ranged from 5-9 and 19 studies were of high quality (Table S3).

Figure 2.
The pooled risk association of T2D with soy consumption [30][31][32][33][34][35][36][37][38]. In the forest plot, the black point represents the point estimate of the effect size of each study, the gray square means the weight of each study, the line length represents the 95% confidence interval (CI) of the effect size of each study, the diamond represents the meta-analysis synthesizes the summary results of each study, the diamond center represents the point estimate of the effect size of the summary results, and the diamond width represents the 95% CI of the effect size of the summary results. Figure 2. The pooled risk association of T2D with soy consumption [30][31][32][33][34][35][36][37][38]. In the forest plot, the black point represents the point estimate of the effect size of each study, the gray square means the weight of each study, the line length represents the 95% confidence interval (CI) of the effect size of each study, the diamond represents the meta-analysis synthesizes the summary results of each study, the diamond center represents the point estimate of the effect size of the summary results, and the diamond width represents the 95% CI of the effect size of the summary results.

Subgroup Analysis between Soy Foods Consumption and the Risk of T2D and CVDs Events
To investigate the association of specific soy products with T2D and CVDs, we performed a subgroup analysis. All subgroup analyses of soy foods were shown in Table 2. We found that non-fermented and fermented soy food were not associated with T2D events (p > 0.05). However, dried and other soy foods were significantly and negatively associ-     Through a dose-response analysis, the non-linear significant relationships were found between tofu and CVDs (P nonlinearity = 0.003, Figure S3A) and between natto and CVDs (P nonlinearity < 0.001, Figure S3B). Meanwhile, a non-linear significant relationship between the natto intake and stroke events was also observed (P nonlinearity < 0.001, Figure S3C). A daily intake of 26.7 g of tofu reduced the CVDs risk by 18% (TRR = 0.82, 95% CI: 0.74-0.92). When the intake dose exceeded 63 g, the tofu had no protective effect on CVDs ( Figure S3A). A daily intake of 11.1 g of natto lowered the risk of CVDs by 17% (TRR = 0.83, 95% CI: 0.78-0.89), especially stroke. When the daily intake was less than 32 g, natto was beneficial for CVDs (stroke) ( Figure S3B,C). Additionally, no significant linear or non-linear associations were found between miso and CVDs (P linearity = 0.72, P nonlinearity = 0.13; Figure S3D), between natto and CHD (P linearity = 0.91, P nonlinearity = 0.70; Figure S3E), or between miso and stroke (P linearity = 0.62, P nonlinearity = 0.19; Figure S3F).

Subgroup Analysis between Baseline Characteristics and the Risk of T2D and CVDs Events
To further investigate the association between soy consumption and T2D and CVDs under different baseline characteristics, we performed the following subgroup analysis. First, a significant negative correlation between soy consumption and T2D in women (TRR = 0.78, 95% CI: 0.67-0.91) was demonstrated, but neither was found in men (TRR = 0.99, 95% CI: 0.89-1.1) nor in both included together (TRR = 0.56, 95% CI: 0.18-1.70; Table S4) by stratifying analysis. It also negatively correlated between soy consumption and CVDs events in women (TRR = 0.80, 95% CI: 0.68-0.94) but not in men (TRR = 0.96, 95% CI: 0.89-1.04; Table S4). Further analysis found that high soy consumption had a low risk of CHD events in women and men but did not reduce the risk of stroke in women or men (Table S4). Second, we found that soy consumption lowered the risk of T2D and CVDs in Asians and Americans (Table S4) (Table S4). However, it was not associated with the risk of stroke events in cohort studies (TRR = 0.96, 95% CI: 0.87-1.05). Fourth, the follow-up time did not affect the relationship between soy consumption and the risk of T2D, CVDs, CHD, and stroke events (Table S4).

Publication Bias
The visual inspection identified that cohort studies of CVDs had publication bias (P Egger' test = 0.18 and P Begg's test = 0.01; Figure S4A), while case-control studies (P Egger' test = 0.15 and P Begg's test = 0.38; Figure S4B) related to CVDs and studies between soy intake and T2D risk (P Egger' test = 0.17 and P Begg's test = 0.32; Figure S4C) did not. The publication bias of high-quality cohort studies might be attributed to some differences in baseline characteristics, such as subtypes of CVDs and long follow-up times. Sensitivity analysis indicated a stable publication bias between soy and T2D, and between soy and CVDs ( Figure S5A,B).

Main Findings and Interpretation
Besides updating the data by including more studies, there were three strengths compared to the results of the previous meta-analysis related to soy/soy products and CVDs and T2D [23,24,58,59]. First, the subgroup analysis of soy products clarified the specific relationship between soy products and CVDs and T2D. Second, the dose-response analysis based on the subgroup analysis revealed the specific intake dose of bean products and the risk of CVDs. Third, we comprehensively elaborated on the possible mechanisms of soybean components in protecting T2D and CVDs in the discussion, especially soybean protein peptides, isoflavones, unsaturated fatty acids, lecithin, stigmasterol, and low BCAAs. Meanwhile, we also obtained the consistent conclusion that soy intake was negatively associated with the risk of T2D, CVDs, CHD, and stroke events [23,24,58].
It was known that soy protein and flavonoids were important parts of the active ingredients of soy, which could reduce the risk of CVDs and diabetes through antioxidative and anti-inflammatory properties [1,11]. More clinical and basic studies also supported our conclusion [6,23,59]. Animal studies in diabetic rats and mice revealed that consumption of soy protein peptides, such as β-conglycinin, soymorphin-5, and Vglycin, improved muscle glucose uptake, decreased blood glucose and triglyceride levels, increased insulin sensitivity, and restored pancreatic function with higher plasma adiponectin and phosphorylated adenosine monophosphate-activated protein kinase (AMPK) [60,61]. Evidence showed that isoflavone improved the endothelial function, lowered blood pressure and blood glucose, and had beneficial effects on decreasing the cardiovascular risk of patients with T2D by affecting AMPK phosphorylation and glucose transporter 4 (GLUT4) in vitro and in preclinical and clinical studies [62][63][64].
Interestingly, soy does not contain cholesterol, but its plant components, including lecithin, stigmasterol, soy protein peptides, and isoflavones, improve lipid profiles through different mechanisms. For example, phosphatidylcholine in soybean lecithin may regulate lipid metabolism by dissolving cholesterol in the intestine and reducing intestinal cell absorption [9,65]. Secondly, lecithin decreases blood lipids by inhibiting intestinal cholesterol absorption and promoting cholesterol excretion [9]. Lecithin cholesterol acyl transferases (LCAT) are responsible for reverse cholesterol transport from extra-hepatic tissues back to the liver and the activity of LCAT is closely related to cardiovascular health [66]. In recent years, soybean lecithin has been used as a functional food supplement to improve blood lipids [67]. Stigmasterol, as a common phytosterol in soybeans, has no effect on the apolipoprotein E genotype [68], and the composition of intestinal flora [69] may be involved in regulating its cholesterol-lowering effect. Studies have reported that stigmasterol can improve blood pressure and reduce the risk factors of cardiovascular diseases in post-menopausal women [70]. It can also improve blood glucose and induce β-cell regeneration by improving GLUT4 translocation and insulin resistance [10]. Furthermore, soy proteins and isoflavones also reduced blood cholesterol and triglyceride [71,72], which might be beneficial to cardiovascular health and systemic metabolism. Therefore, the lipid-lowering effect of soybeans may be enhanced and mediated by the synergistic effect of intestinal factors and soy components [9]. In the future, soy protein peptides and isoflavones might emerge as complementary or alternative treatments for diabetes and cardiovascular disease [63]. Soy is rich in unsaturated fatty acids and dietary fiber. Additionally, some research has discovered that unsaturated fatty acids, including n-3 long-chain polyunsaturated fatty acids, and linoleic acid, improved fasting glucose levels and insulin resistance, and reduced the risk of stroke and CHD [72][73][74][75]. Similarly, dietary fiber could increase the flora gene richness and activate acetic acid and butyric acid synthesis pathways to promote insulin secretion and intervene in diabetes by modulating gut microbiota [76][77][78].
Additionally, compared to the protein derived from animals, the protein derived from soy had a low concentration of BCAAs of only 4% [79]. However, with the widespread application of metabolomic profiling, increased levels of BCAAs and related metabolites are now widely considered to be a metabolic hallmark of obesity, insulin resistance, and T2D in humans [80]. Meanwhile, some research has reported a direct role of BCAAs in heart failure (HF), vascular disease, hypertension, and arrhythmias [7]. Epidemiologically, increased plasma BCAA concentrations were biomarkers for HF, C, HD, and hypertension and they could predict adverse outcomes in individuals with CHD and HF, and predict stroke, myocardial infarction (MI), coronary revascularization, and death from cardiovascular causes in individuals without CVDs [7]. Thus, this might be one of the key mechanisms through which soy intake could reduce the risk of T2D and CVDs.
Although fried or other soy intakes were negatively associated with T2D, we should be cautious about the result, since only five pooling studies were included. Our result revealed that the consumption of soy foods, especially tofu and natto, was negatively associated with the risk of T2D and total CVDs, including CHD and stroke. This is because fermented soy foods, such as natto, contain higher peptides, amino acids, and other breakdown products and consequently enhanced the antioxidant capacity [11]. Additionally, angiotensin-converting enzyme (ACE) inhibitory peptides, richen in natto, blocked ACE and modulated the rennin-angiotensin system to regulate blood pressure [1]. Natto is a salt-free fermented bean product compared to miso soup. Therefore, natto, but not miso soup was associated with stroke and total CVDs events [1]. The processing of tofu and natto could well reduce the fat content of soy, which might be beneficial in decreasing the incidence of T2D and CVDs [81].
The results of the subgroup analysis by gender showed a significant negative relationship between soy intake with T2D and CVDs in women but not in men. Soy intake was negatively associated with CHD and women had a lower risk than men, whereas it was not associated with stroke in men and women. Such a difference could be attributed to isoflavones. Soybean isoflavones are phytoestrogens that are bound to estrogen receptors in different tissues to exert a two-way regulatory effect on endogenous estrogen and maintain the balance of estrogen levels in women [1]. Isoflavones exerted cardioprotective functions and a protective role of soy against the development of T2D by estrogen-like effect and antioxidation [30,42]. However, soy or isoflavones have no effect on testosterone or estrogen levels in men [72]. In addition, the protective effect of unsaturated fatty acids on CVDs in women was more significant than that in men [75,82]. Through further analysis of the correlation between gender and stroke, we found that men generally had a higher risk of stroke than women, excluding Yu's research. When we excluded the study, the occurrence of stroke in women was significantly negatively correlated, but this was not the case in men. The discrepant differences from these studies may be partly explained by differences in the type and amount of soy products consumed, population characteristics, and variations in bioactive peptides of soy protein and isoflavone metabolism [54].
Results from our subgroup analysis according to the study regions showed a more significant negative relationship between soy intake and CHD and stroke in Asians and Americans. Since only three cohort studies in the United States were summarized, we should be cautious about the protective effect of soy intake on CVDs in Americans. There are more studies from Asians, and the difference could be attributed to the much higher intakes of soy in Asians than in Westerners and different ethnicities, local gastronomy, food habits, cooking style, and processing of soy. Soy consumption was traditionally common and had become mainstream among Asians [72,83].
Results from summarizing case-control and cohort studies showed protective effects of soy foods on the risk of total CVDs and CHD, whereas there were no protective effects between soy foods and stroke in the meta-analysis of cohort studies. Cohort studies included a larger number of population; dietary exposure was first determined and CVDs were detected after a period of follow-up. However, case-control studies may be attributed to selection and recall bias, because the exposure information of dietary intake was reported after the diagnosis of CVDs was determined. Therefore, the overall results of reducing the risk of CVDs, stroke, and CHD should not be overemphasized in case-control studies.

Strengths and Limitations
There are several strengths to this study. First, we conducted a comprehensive literature search for all the available evidence from existing cohorts and case-control studies and performed a dose-response analysis to investigate the associations of soy foods intake with T2D and CVDs events. Second, an extensive subgroup analysis and meta-regression were conducted to explore the potential sources of heterogeneity and their influence on pooled risk estimates.
Several limitations should be considered. First, although we conducted a systematic literature search and included several eligible studies, the dose-analysis was incomplete due to a lack of essential data. Additionally, the exposure medians for individual studies in the dose-response analysis were approximately estimated and some errors might exist. Second, heterogeneity could be due to the diet habits of different populations, such as varieties and components of soy products, and different intake levels and frequencies of soy foods. Although we carried out stratified analyses using the random effects model, the evident heterogeneity still existed, which was caused by sociodemographic risk factors for T2D and CVDs, and other risk factors. Third, assessing the quality of evidence using the NOS scale might lead to subjectivity.

Conclusions
This systematic review and meta-analysis found that soy intake was negatively associated with T2D and CVDs (including CHD and stroke). The daily intake of 26.7 g of tofu reduced CVDs risk by 18%. A daily intake of 11.1 g of natto lowered the risk of CVDs by 17%, especially stroke. Our findings may support a specific intake of tofu and natto for preventing CVDs and providing greater longevity. More prospective and multi-center studies should be performed to find the protective association between dose-analysis of soy foods and T2D and CVDs.  Table S4: Stratified analyses between baseline characteristics and the risk of T2D and CVDs events, Figure S1: The pooled risk association of coronary heart disease with the consumption of soy foods, Figure S2: The pooled risk association of type 2 diabetes and cardiovascular diseases with the consumption of soy protein, Figure S3: Dose-response analysis for the potential curvilinear or linear associations of soy intake with cardiovascular diseases, coronary heart disease, and stroke, Figure S4: Begg's funnel plot of risk association of (A) cohort studies of cardiovascular diseases, (B) case-control studies of