High blood pressure (BP), or hypertension, with a high prevalence in developed countries, is increasing dramatically in developing countries, in parallel with economic development. It has become a serious burden of global public health, influencing approximately 42 % of the population in England and one billion individuals worldwide(Reference Kearney, Whelton and Reynolds1, Reference Wolf-Maier, Cooper and Banegas2). Furthermore, hypertension has been identified as a large but modifiable risk factor for cardiovascular mortality, which is the leading cause of death in Europe and North America. Considerable attention therefore should be paid to the prevention and control of hypertension.
Soya has received an increasing scientific interest for its beneficial effects on cardiovascular health since the US Food and Drug Administration (Silver Spring, MD, USA) approved a health claim in 1999 that a daily intake of 25 g of soya protein may reduce heart disease. Isoflavones in soya protein, a class of phyto-oestrogens acting as oestrogen mimics, are suggested to have potential hypotensive effects(Reference Rivas, Garay and Escanero3–Reference Welty, Lee and Lew5). However, a recent meta-analysis by Hooper et al. (Reference Hooper, Kroon and Rimm6) has shown that isoflavone extracts did not affect BP, though the result for systolic BP was close to significance. Also, three trials(Reference Llaneza, Gonzalez and Fernandez-Inarrea7–Reference Khaodhiar, Ricciotti and Li9) published subsequently to that meta-analysis did not yield significant results either. Moreover, two trials(Reference Llaneza, Gonzalez and Fernandez-Inarrea7, Reference Chan, Lau and Yiu8) showed that isoflavone extract intervention resulted in an increase in BP compared with control. On the basis of the current evidence, we believe with some confidence that isoflavone extracts have no significant effects on BP. On the other hand, numerous epidemiological studies have been carried out to evaluate the effects of soya protein or soya foods on cardiovascular health as well as BP reductions. Observational studies(Reference Nagata, Shimizu and Takami10–Reference Pan, Franco and Ye12) have indicated that soya protein intake or soya food consumption might be inversely associated with BP. Several randomised controlled trials (RCT) have documented a BP-lowering role of soya(Reference Rivas, Garay and Escanero3, Reference Welty, Lee and Lew5, Reference Teede, Dalais and Kotsopoulos13, Reference He, Gu and Wu14), whereas others have suggested no effect. A previous meta-analysis(Reference Hooper, Kroon and Rimm6) reported that soya protein isolate, but not other soya products or components, significantly reduced diastolic BP (DBP). However, that study was not primarily designed to assess the effects of soya protein or isoflavones on BP, and sources of heterogeneity were not explained.
With newly emerging evidence supporting a BP-lowering role of soya protein rather than isolated isoflavones, we aimed to conduct an updated and comprehensive meta-analysis of RCT to evaluate the effects of soya protein on BP reductions, and to explore the potential sources of heterogeneity across studies for a better understanding of the present literature.
Materials and methods
Search strategy
We searched the PubMed database (National Library of Medicine, Bethesda, MD, USA) through to April 2010 for published studies in the English language using search terms ‘soya OR soya protein OR soybeans OR phytoestrogens OR isoflavones’ and ‘hypertension OR blood pressure’. In addition, we carried out a manual search using reference lists of original articles and recent reviews. Each published paper was independently reviewed and relevant information was extracted by three authors (J.-Y. D., X. T. and Z.-W. W.).
Study selection
To be included in our analysis, a study was restricted to RCT conducted in human subjects and reported a net change in systolic BP (SBP) and DBP during the intervention with their corresponding standard deviations, or appropriate data to calculate these values. In one study, two independent strata (normotension v. hypertension) were included; thus, it was treated as two trials.
Major reasons for exclusion were as follows: (1) using isoflavone extracts as treatment; (2) non-randomised treatment allocation; (3) lack of BP data; (4) a systematic difference between the intervention and control groups other than soya protein; (5) absence of a concurrent control group; (6) recurrent publication of data already represented in the present analysis.
Data extraction
We recorded the study characteristics as follows: (1) first author's name, publication year and country of origin; (2) number of participants; (3) mean age, age range, sex distributions and health status of participants; (4) design details, including whether parallel or cross-over and open, single blind or double blind; (5) study duration; (6) dosage of soya protein and isoflavones, placebo and other treatment interventions; (7) type of soya protein; (8) mean BP at baseline. The Jadad score, a scale that ranges from 0 to 5 according to the descriptions of randomisation, blinding and reporting of participant withdrawals, was used to measure the quality of each trial(Reference Jadad, Moore and Carroll15).
Statistical analysis
Net changes in SBP and DBP at the end of the intervention were calculated, if not reported. For studies reporting only the mean difference between the treatment and control groups, we set the control group's mean change as zero and the treatment group's mean change as the reported mean difference. Standard errors, CI and P values were converted to standard deviations for the analyses. When the standard deviation for paired differences was not reported, we calculated it from the standard deviation at baseline and at the end of the follow-up on the basis of the method of Follmann et al. (Reference Follmann, Elliott and Suh16), assuming a correlation coefficient of 0·5 between initial and final BP. We assumed equal variances during the trial and between the treatment and control groups.
With each trial weighted by the reciprocal of the variance for BP change, we calculated a weighted mean difference and its corresponding 95 % CI. Both fixed- and random-effects models were used. The homogeneity of effect size across trials was tested by Q statistics (P < 0·10 was considered heterogeneous). If there was significant heterogeneity among the studies, the random-effects model was used; otherwise, the fixed-effects model was acceptable. We also examined the I 2 statistic, which measures the percentage of the total variation across studies that is due to heterogeneity, rather than chance(Reference Higgins, Thompson and Deeks17). A sensitivity analysis was conducted to examine the influence of various exclusion criteria on the overall effect sizes.
To explore the influence of covariates on the net change in BP, we carried out a series of subgroup analyses. Subgroups were selected based on biological plausibility, study design and participant characteristics. Furthermore, we conducted a meta-regression analysis to estimate the effects of various study characteristics on the effect size. Covariates for meta-regression analysis were selected based on the results of the subgroup analysis and biological knowledge. For each trial, the covariates were calculated as average values of the treatment and control groups at baseline. Mean aortic pressure was used as pre-treatment BP. When information on mean values was absent, for example dosage of soya protein, we estimated it as the median of the remaining trials.
To assess the publication bias, we performed both Egger's and Begg's tests(Reference Begg and Mazumdar18, Reference Egger, Davey Smith and Schneider19), and inspected the funnel plots as well. All analyses were performed using STATA version 10.0 (StataCorp, College Station, TX, USA). A P value < 0·05 was considered to be statistically significant, unless otherwise specified.
Results
Characteristics of the studies
After our complete review, twenty-five studies(Reference Rivas, Garay and Escanero3–Reference Welty, Lee and Lew5, Reference Teede, Dalais and Kotsopoulos13, Reference He, Gu and Wu14, Reference Kurowska, Jordan and Spence20–Reference Wong, Kendall and de Souza39) including twenty-seven trials met the inclusion criteria. Characteristics of the trials included in the present meta-analysis are presented in Table 1. The trials published between 1997 and 2010 varied from 12 to 276 participants, with a median of 42 and a total number of 1608. Of the twenty-seven trials, nine were conducted in the USA, eight in Europe and ten in other countries. All trials were conducted in adults, with ages ranging from 18 to 75 years. The subjects' initial mean BP ranged from 110 to 153 mmHg for SBP, and from 69 to 100 mmHg for DBP. Of the twenty-six trials that reported the sex distribution, eight consisted entirely of women, of which six were conducted among post-menopausal women, and two trials consisted entirely of men. There were five trials(Reference Rivas, Garay and Escanero3–Reference Welty, Lee and Lew5, Reference He, Gu and Wu14, Reference Burke, Hodgson and Beilin22) conducted only in hypertensive subjects (210 participants) with a mean baseline BP of 145·2/90·5 mmHg, among which two trials reported antihypertensive therapy use and the other three did not. There were eleven trials that included only normotensive subjects (714 participants) with a mean baseline BP of 126·8/77·9 mmHg. The remaining eleven trials included both hypertensive and normotensive subjects. There were two studies(Reference Hermansen, Sondergaard and Hoie23, Reference Jayagopal, Albertazzi and Kilpatrick24) conducted in patients with type 2 diabetes. Of the twenty-seven trials, fourteen were parallel-designed, and the rest had a cross-over design. The duration of the intervention lasted from 4 to 52 weeks, with a median of 8 weeks. Treatment was double blind in fourteen trials, single blind in one trial and open in twelve trials. There were fifteen trials that used soya protein isolate containing isoflavones and twelve that used other soya foods. Most of the control groups received casein or milk. Dosage of soya protein varied from 18 to 66 g/d, with a median of 30 g/d. Dosage of isoflavones contained in soya protein varied from 23 to 160 mg/d, with a median of 100 mg/d.
Characteristics of all trials included in the present meta-analysis
BP, blood pressure; X, cross-over; NR, not reported; DB, double blind; SPI, soya protein isolate; P, parallel; SB, single blind; DASH, Dietary Approaches to Stop Hypertension.
Net changes in blood pressure
Tests for heterogeneity indicated that the treatment effect was significantly different across studies (P < 0·001 for SBP and P < 0·001 for DBP). I 2 values were 65·7 % for SBP and 61·5 % for DBP. Compared with control, intervention was associated with an average net change in BP ranging from − 17·0 to 5·0 mmHg for SBP and − 12·2 to 4·0 mmHg for DBP. For SBP, a trend towards intervention-related reduction in BP was observed in seventeen trials, with five trials showing a significant reduction. For DBP, a trend towards intervention-related reduction in BP was observed in nineteen trials, with a significant reduction in five trials. Overall pooled estimates of the effect of soya protein were − 2·21 mmHg (95 % CI − 4·10, − 0·33; P = 0·021) for SBP (Fig. 1) and − 1·44 mmHg (95 % CI − 2·56, − 0·31; P = 0·012) for DBP (Fig. 2).
Pooled effect size of soya protein with isoflavones on systolic blood pressure. WMD, weighted mean difference.
Pooled effect size of soya protein with isoflavones on diastolic blood pressure. WMD, weighted mean difference.
Sensitivity analysis
Table 2 presents the results of sensitivity analysis that excluded trials based on different criteria. Overall pooled estimates for treatment effects in BP changed little in the sensitivity analysis.
(Number of trials, I 2 and 95 % confidence intervals)
Stratified analysis
Table 3 shows the results from stratified analyses. SBP and DBP were significantly reduced in both the hypertensive and normotensive groups. Furthermore, the mean decrease in the hypertensive group was 8·58 mmHg (95 % CI − 15·10, − 2·06; P = 0·010) for SBP and 5·24 mmHg (95 % CI − 9·40, − 1·08; P = 0·014) for DBP, markedly greater than that in the normotensive group. Significant BP reductions in response to the intake of soya protein were also observed in trials using carbohydrate as the control diet, with a longer duration, using soya protein isolate as treatment, and in trials employing a parallel design. Moreover, the reductions tended to be greater among trials using carbohydrate than those using milk or casein as the control diet, among trials with a parallel design than those with a cross-over design and among trials with the intervention duration of at least 12 weeks than those with less than 12 weeks.
(Number of trials, I 2 and 95 % confidence intervals)
Meta-regression analysis
We first conducted univariate meta-regression analyses for each of the following variables: pre-treatment BP; duration of intervention; sample size; BMI; dosage of soya protein; dosage of soya isoflavones. There was a significant inverse association of effect size with pre-treatment BP for both SBP (Fig. 3) and DBP (Fig. 4), but not with other covariates (all P>0·3; e.g. P values are 0·791 and 0·492 for isoflavone dosage v. net changes in SBP and DBP, respectively). In our multivariate meta-regression, pre-treatment BP remained significant after adjustment for the other covariates (P = 0·011 for SBP and P = 0·039 for DBP). Pre-treatment BP was therefore proved to be a significant and independent predictor for heterogeneity, strengthening the results of the present subgroup meta-analyses.
Net change in systolic blood pressure according to pre-treatment blood pressure (r − 0·54, P = 0·006). Diameters of circles depend on the weights in the random-effects model.
Net change in diastolic blood pressure according to pre-treatment blood pressure (r − 0·45, P = 0·024). Diameters of circles depend on the weights in the random-effects model.
Publication bias
There was no sign of publication bias when examining the funnel plots (plot not shown). Results from Begg's and Egger's tests also did not indicate the evidence of publication bias (SBP: Begg P = 0·297, Egger P = 0·988; DBP: Begg P = 0·381, Egger P = 0·243).
Discussion
The present study identified significant reductions in the overall estimates in both SBP and DBP during soya protein interventions. Soya protein consumption significantly reduced SBP and DBP in both hypertensive and normotensive subjects, and the reductions were markedly greater in hypertensive subjects than those in normotensive subjects. Meta-regression analyses further revealed a significantly inverse association between pre-treatment BP and the level of BP reductions. In addition to pre-treatment BP, the type of control diet used as a comparison also has an impact on the BP reductions.
Our findings have potential public health implications. Although the BP reductions reported here are moderate at the individual level, a reduction of 4–5 mmHg in SBP and 2–3 mmHg in DBP would be expected to substantially reduce the risk of stroke (by about 20 %), CHD (by 10 %) and all-cause mortality (by 8 %) at the population level(Reference McInnes40). Given the high prevalence of hypertension(Reference Kearney, Whelton and Reynolds1, Reference Wolf-Maier, Cooper and Banegas2), even a slight reduction in BP may contribute to a considerable public health benefit from soya protein consumption. Moreover, when used for other purposes, such as improving lipid profiles(Reference Anderson, Johnstone and Cook-Newell41, Reference Zhan and Ho42), the moderate hypotensive effects of soya protein doubtlessly provide extra benefits.
The observed substantial heterogeneity is most probably explained by pre-treatment BP and the type of control diet. It is probable that subjects with higher BP have more room for improvement. In fact, the study by Rivas et al. (Reference Rivas, Garay and Escanero3) that enrolled mild-to-moderate hypertensive patients and reported substantial BP reductions has been partly responsible for the observed heterogeneity, as exclusion of this outlier resulted in I 2 values < 50 % for both SBP and DBP. Furthermore, trials(Reference Welty, Lee and Lew5, Reference Teede, Dalais and Kotsopoulos13, Reference He, Gu and Wu14, Reference Kurowska, Jordan and Spence20, Reference Allison, Gadbury and Schwartz26, Reference Anderson and Hoie31, Reference Lukaczer, Liska and Lerman34, Reference Anderson, Fuller and Patterson35, Reference Matthan, Jalbert and Ausman37–Reference Wong, Kendall and de Souza39) conducted in normotensive subjects yielded consistent and significant, although small, BP reductions, with little evidence of heterogeneity (P = 0·175 for SBP and P = 0·339 for DBP). This is probably because normotensive subjects are evidently healthy and are unlikely to take antihypertensive drugs that may potentially mask the effects of soya protein on BP.
Of note, the type of control diet appeared to play an important role in the effects of soya protein on BP. Growing evidence shows that milk product consumption has potential benefits on BP regulations(Reference Kris-Etherton, Grieger and Hilpert43). When compared with milk or casein, soya protein showed small but not significant reductions in BP, which indicated soya protein may not be superior to milk products in improving BP. When compared with carbohydrate, soya protein consumption produced greater and significant BP reductions. These findings are consistent with the current evidence that partial substitution of refined carbohydrate with dietary protein can lower BP(Reference Appel, Sacks and Carey44, Reference Hodgson, Burke and Beilin45). Thus, further studies of soya protein on BP should take the type of control diet into consideration.
In addition to pre-treatment BP and type of control diet, other characteristics of trials are worth considering, including duration of intervention and study design. Although the association of intervention duration with BP reductions was not found in the meta-regression analyses, results from subgroup analyses indicated that duration of at least 12 weeks would be necessary for intervention. Parallel-designed trials showed a greater reduction in BP than cross-over-designed trials. This is possibly because parallel-designed trials usually have longer duration that was associated with significant BP reductions as observed earlier. It is also possible that carry-over effects have residual impacts on BP reductions, as washout periods were short in most cross-over trials.
The present study has several strengths. First, it provides a comprehensive report of the effects of soya protein consumption on BP, based on the pooled evidence from twenty-seven RCT conducted in a wide range of geographical locations. The participant characteristics such as patient background, baseline BP status and ethnicity varied across trials, indicating that our findings might have a certain degree of external validity to be generalised to a broader population. Second, compared with the previous meta-analysis(Reference Hooper, Kroon and Rimm6), the present analysis identified and included thirteen more RCT through an updated search. The enlarged sample size enhanced the power to detect a significant difference and provide more precise estimates of effects, and allowed us to perform a meta-regression analysis to explore sources of heterogeneity. We also excluded studies in which other dietary factors were substantially different other than soya protein; therefore, soya protein with isoflavones was more likely to be the major contributor to the BP-lowering effects. Third, we investigated the potential sources of substantial heterogeneity across studies by a series of subgroup and meta-regression analyses, thereby increasing the clinical relevance of our conclusions and the scientific understanding of the relevant studies(Reference Thompson46). In addition, sensitivity analyses supported the robustness of our findings, and there was little evidence of publication bias in the present study.
Limitations in the present study should also be acknowledged. Substantial heterogeneity across trials that makes our findings complicated to interpret is the primary one. However, we have identified the major sources of inconsistency through a series of subgroup and meta-regression analyses. Since some trials were not primarily designed to test the hypotensive effects of soya protein, a rigorous methodology for measuring BP and the use of antihypertensive medications among participants were seldom reported. Likewise, unmeasured dietary factors that potentially influenced BP outcomes, such as dietary Na and K intake during the intervention, may not be controlled and balanced. In addition, there were only five trials(Reference Rivas, Garay and Escanero3–Reference Welty, Lee and Lew5, Reference He, Gu and Wu14, Reference Burke, Hodgson and Beilin22) conducted in hypertensive participants, and the pooled results in this target group showed an even higher degree of heterogeneity. For all of these reasons, the findings of the present meta-analysis should be considered with caution, and further high-quality large-scale RCT are warranted, especially in the hypertensive population.
The underlying mechanisms by which soya protein with isoflavones may influence BP remain unclear. Isoflavones in soya are widely considered as the major contributors to the hypotensive effects of soya(Reference Rivas, Garay and Escanero3–Reference Welty, Lee and Lew5). It is plausible that isoflavones, as oestrogen mimics, may exhibit antihypertensive activities by stimulating the production of the potent vasodilator NO(Reference Squadrito, Altavilla and Squadrito47). Isolated isoflavones, however, appear to have little effect on BP as observed previously, suggesting the necessity for isoflavones to be ingested in combination with soya protein. Alternatively, hypotensive effects of soya may be mediated not only by oestrogenic mechanisms, but also by other soya components. Arginine, for example, which is a precursor of NO and with a high content in soya protein, might also exert BP-lowering effects(Reference Dudasova and Grancicova48). Additionally, pinitol isolated from soyabeans has been documented to significantly decrease BP in patients with type 2 diabetes(Reference Kim, Kim and Kang49).
Equol, a metabolite of the soya isoflavone daidzein produced by gut bacteria, has recently received considerable interest due to its higher bioactivity compared with other isoflavones(Reference Setchell, Brown and Lydeking-Olsen50). Because only approximately 30–50 % of the population is able to metabolise daidzein to equol, conflicting results from previous studies have led to the ‘equol hypothesis’ that equol producers are more likely to benefit from consuming soya foods than non-equol producers(Reference Setchell, Brown and Lydeking-Olsen50). However, several clinical studies(Reference Meyer, Larkin and Owen28, Reference Kreijkamp-Kaspers, Kok and Bots33, Reference Tormala, Appt and Clarkson51) investigating the effects of soya isoflavones on BP have observed no significant difference between equol producers and non-equol producers. Limitations of these studies including small sample sizes may result in insufficient statistical power to detect the difference. Thus, the equol hypothesis is not supported by the current evidence and deserves further study.
Conclusions
In summary, current evidence demonstrates that soya protein intake, compared with a control diet, significantly reduces both SBP and DBP, but the BP reductions are related to pre-treatment BP levels of subjects and the type of control diet used as a comparison.
Acknowledgements
The present study was supported in part by grants from the National Natural Science Foundation of China (30771808) and from the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry (K5126829) to L.-Q. Q. All authors have no conflicts of interest. J.-Y. D. and L.-Q. Q. were responsible for the study design, data analyses, interpreting the results, drafting of the manuscript and critical revision of the manuscript for important intellectual contents; J.-Y. D., X. T. and Z.-W. W. were responsible for the literature search and data extraction; P.-C. X. and K. H. assisted in the data analyses and critical revision of the manuscript for important intellectual contents. All authors read and approved the final manuscript.
