Effect of Plant Protein on Blood Lipids: A Systematic Review and Meta‐Analysis of Randomized Controlled Trials

Abstract

Background

There is a heightened interest in plant‐based diets for cardiovascular disease prevention. Although plant protein is thought to mediate such prevention through modifying blood lipids, the effect of plant protein in specific substitution for animal protein on blood lipids remains unclear. To assess the effect of this substitution on established lipid targets for cardiovascular risk reduction, we conducted a systematic review and meta‐analysis of randomized controlled trials using the Grading of Recommendations Assessment, Development, and Evaluation system.

Methods and Results

MEDLINE, EMBASE, and the Cochrane Registry were searched through September 9, 2017. We included randomized controlled trials of ≥3 weeks comparing the effect of plant protein in substitution for animal protein on low‐density lipoprotein cholesterol, non–high‐density lipoprotein cholesterol, and apolipoprotein B. Two independent reviewers extracted relevant data and assessed risk of bias. Data were pooled by the generic inverse variance method and expressed as mean differences with 95% confidence intervals. Heterogeneity was assessed (Cochran Q statistic) and quantified (I² statistic). The overall quality (certainty) of the evidence was assessed using the Grading of Recommendations Assessment, Development, and Evaluation system. One‐hundred twelve randomized controlled trials met the eligibility criteria. Plant protein in substitution for animal protein decreased low‐density lipoprotein cholesterol by 0.16 mmol/L (95% confidence interval, −0.20 to −0.12 mmol/L; P<0.00001; I²=55%; moderate‐quality evidence), non–high‐density lipoprotein cholesterol by 0.18 mmol/L (95% confidence interval, −0.22 to −0.14 mmol/L; P<0.00001; I²=52%; moderate‐quality evidence), and apolipoprotein B by 0.05 g/L (95% confidence interval, −0.06 to −0.03 g/L; P<0.00001; I²=30%; moderate‐quality evidence).

Conclusions

Substitution of plant protein for animal protein decreases the established lipid targets low‐density lipoprotein cholesterol, non–high‐density lipoprotein cholesterol, and apolipoprotein B. More high‐quality randomized trials are needed to improve our estimates.

Clinical Trial Registration

URL: http://www.clinicaltrials.gov. Unique identifier: NCT02037321.

Clinical Perspective

What Is New?

• Although the cholesterol‐lowering benefit of plant protein sources, such as soy, pulses, and nuts, is well documented, the overall cholesterol‐lowering benefit of plant protein in substitution for animal protein (as meat, dairy, and/or egg alternatives) has not been synthesized.

• The available evidence from randomized controlled trials suggests that 1 to 2 servings of plant protein in substitution for animal protein decreases low‐density lipoprotein cholesterol, non–high‐density lipoprotein cholesterol, and apolipoprotein B by ≈4% in adults with and without hyperlipidemia.

• Because of inconsistency or imprecision in the estimates, the overall quality (certainty) of the evidence is moderate by the Grading of Recommendations Assessment, Development, and Evaluation system, suggesting that more research will refine our estimates.

What Are the Clinical Implications?

• Because the intake of plant protein from soy, nuts, and pulses remains low, there is an opportunity for people to realize the lipid‐lowering benefits of sustainable plant‐based dietary strategies that substitute plant protein for animal protein.

• Plant protein, especially in combination with other cholesterol‐lowering foods (eg, viscous fiber and plant sterols) and/or as an adjunct to lipid‐lowering pharmacotherapy, may have a clinically meaningful benefit in helping people to achieve lipid targets and reduce cardiovascular risk.

Introduction

Cardiovascular disease (CVD) accounts for ≈48% of deaths attributable to noncommunicable disease worldwide and remains the number one cause of mortality.1, 2 Modification by diet and lifestyle of risk factors, particularly dyslipidemia, remains the cornerstone of therapy, according to major cardiovascular guidelines.3, 4

There has been increasing recent interest in plant‐based diets. Vegetarian and vegan dietary patterns and other plant‐based dietary patterns, such as the Mediterranean diet, have been established as dietary patterns that improve lipid profiles and reduce risks of CVD.5, 6, 7 Both the Scientific Report of the 2015 Dietary Guidelines Advisory Committee and 2016 Canadian Cardiovascular Society guidelines recently recommended a vegetarian dietary pattern and a Mediterranean dietary pattern for cardiovascular protection.3, 8 The mechanisms by which these dietary patterns improve cardiovascular risk likely include intrinsic and extrinsic pathways. Plant protein sources, such as soy, dietary pulses, and nuts, have all individually shown lipid‐lowering advantages through their specific components (specific protein fractions [7s‐globulin], viscous fibers, polyunsaturated fatty acids, and plant sterols). Replacement of animal protein with plant protein has also shown advantages through the displacement of saturated fatty acids.9 The combination allows for meaningful reductions in lipids in systematic reviews and meta‐analyses of randomized controlled trials (RCTs).9, 10, 11, 12

Despite the strong biological plausibility supporting their benefit and endorsement of plant‐based diets from recent guidelines, there is still uncertainty as to whether the benefit is attributable to the exchange of plant protein for animal protein or to other aspects of a plant‐based dietary pattern. It remains difficult to isolate specific mechanisms,13, 14, 15 and the strength of the evidence supporting the lipid‐lowering effects of plant protein remains disputed.16, 17, 18, 19 As a result, many authoritative guidelines do not specifically recommend substituting plant protein for animal protein for lipid‐lowering and cardiovascular protection.20, 21, 22, 23 To summarize and evaluate the available evidence, we conducted a systematic review and meta‐analysis using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system of the effect of substituting plant protein for animal protein on the established lipid targets for CVD prevention, low‐density lipoprotein cholesterol (LDL‐C), non– high‐density lipoprotein cholesterol (non–HDL‐C), and apolipoprotein B (Apo‐B), in RCTs.4, 24

Methods

This study was planned and conducted following the Cochrane Handbook for Systematic Review of Interventions.25 Data were reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines.26 The authors declare that all supporting data are available within the article (and its supplementary files).

Literature Search

We searched MEDLINE, EMBASE, and the Cochrane Register through September 9, 2017, for eligible trials. Table S1 shows our detailed search strategy.

Study Selection

We included randomized, long‐term, dietary intervention trials in human subjects comparing LDL‐C, non–HDL‐C, and/or Apo‐B parameters between plant and animal protein intervention arms. To be included, studies had to be at least 3 weeks in duration and performed in accordance with the minimum trial follow‐up requirement of the US Food and Drug Administration for lipid‐lowering health claims.27 Studies deliberately introducing confounding factors (eg, plant sterols or combined therapeutic interventions) to the plant protein arm were also excluded, including studies applying a broad vegetarian or vegan dietary pattern as opposed to a direct substitution of protein sources. No restrictions were placed on language.

Data Extraction

Study characteristics and results of eligible trials were each extracted by S.S.L. and a coextractor (L.L., S.B.M., S.E.S., E.V., or V.H.). Extracted characteristics include study setting, design, duration, blinding, sample size, participant characteristics, and plant and animal protein diet descriptions. Risk of bias of eligible trials was also assessed by S.S.L. and the same coextractor using the Cochrane risk of bias tool, which categorizes studies as high, low, or unclear risk of bias on the basis of criteria pertaining to selection bias, blinding, incomplete outcome data, and reporting bias.25 PlotDigitizer version 2.5.1 (Free Software Foundation, Boston, MA) was used to extract data from graphs, where applicable. Any discrepancies in data extraction or risk of bias assessment were reconciled by consensus.

Grading of the Evidence

The overall quality (certainty) of evidence was assessed using the GRADE system,28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40 which grades evidence as high, moderate, low, or very low quality. RCTs are graded as high‐quality evidence by default. Scores can then be downgraded on the basis of the following prespecified criteria: risk of bias (weight of studies shows important risk of bias), inconsistency (substantial unexplained interstudy heterogeneity of I² >50%, P<0.10), indirectness (presence of factors that limit the generalizability of the results), imprecision (95% confidence interval [CI] for risk estimates are wide or overlap a minimally important difference of 0.1 mmol/L for LDL‐C and non‐HDL‐C and 0.04 g/L for Apo‐B), and publication bias (evidence of small‐study effects).

Statistical Analysis

We used Review Manager version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) for primary analyses and Stata version 13 (StataCorp, College Station, TX) for meta‐regression and publication bias tests. Data were pooled using the generic inverse variance method with random‐effects models and are expressed as mean differences (MDs) with 95% CIs. All analyses were repeated using fixed‐effects models and parametric bootstrapping as sensitivity analyses. Where there were multiple plant or animal protein arms in a single trial, we pooled intervention arms to obtain a single pairwise comparison, to mitigate unit‐of‐analysis error25; where relevant, these arms were assessed separately for subgroup analyses.

Change‐from‐baseline values were favored, and differences in change‐from‐baseline values were used, where given; otherwise, we used end‐difference values, if reported, or calculated the differences from available data. Non–HDL‐C values were calculated by subtracting HDL‐C from total cholesterol values, where non–HDL‐C values were not directly reported, and the variance sum law was used to derive SDs for non–HDL‐C from total cholesterol and HDL‐C variance data.41 In crossover trials, missing variance data were calculated from t test P values using standard formulas; where P values were unavailable, a correlation coefficient of 0.5 was assumed as a conservative estimate and used to impute SE data.25, 42 Where no variance data were available, the average SD of the MDs across all other included trials was used to derive the SEM difference on the basis of the respective trial's sample size.

Interstudy heterogeneity was evaluated by the Cochran Q statistic and quantified using the I² statistic. P<0.10 was considered significant; an I² value of 50% or higher was considered substantial.25 Potential sources of heterogeneity were investigated by additional sensitivity analyses, in which we recalculated the pooled effect estimate after removing each individual trial, after removing all imputed data, and after imputing alternative correlation coefficients of 0.25 and 0.75. We additionally investigated potential sources of heterogeneity by subgroup analyses. Our a priori subgroups included study design, protein dose, plant and animal protein type, duration of follow‐up, and baseline lipid values. A post hoc analysis was also conducted for protein form (ie, whole food or protein isolate product). Between‐subgroup differences were assessed using meta‐regression with dummy variables.

A post hoc dose‐response analysis was conducted using a piecewise linear meta‐regression via the mkspline function, to assess potential dose thresholds for the continuous subgroup addressing grams of protein substitution.

Publication bias was assessed by inspection of funnel plots and by the use of Egger and Begg tests. Where publication bias was suspected, Duval and Tweedie nonparametric “trim‐and‐fill” analyses were also applied to assess the effect of the imputed “missing” studies.43

Results

Search Results

Figure 1 shows the trial selection process. Our search identified 3917 reports, of which 3689 were excluded on the basis of review of titles and abstracts. The remaining 228 articles were reviewed in full, of which 104 provided data for 112 trial comparisons for inclusion in our analyses.44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144, 145, 146, 147

Figure 1.

Search summary.

Trial Characteristics

The Table summarizes characteristics of the included trials. Detailed characteristics are shown in Table S2. In total, 5774 participants (median age, 54 years) were included in this analysis. There were more women versus men overall (≈5:3 ratio), but this difference is largely attributable to a few large female‐only trials, and the median sex ratio in trials was relatively balanced (44% men). Sixty‐one trials were crossover, and all but 4 were in outpatient settings. Half of the trials were conducted in the United States and Canada (60 of 112), but trials were also distributed across European (24 trials), Asian (10 trials), Middle‐Eastern (9 trials), and South American (3 trials) countries, as well as Australia (6 trials). Of 112 trials, 34 recruited healthy subjects (including healthy postmenopausal women); 51 trials recruited subjects with hyperlipidemia, 4 of which also selected for additional conditions. The remaining 28 trials included participants with various conditions, including renal disease, overweight, obesity, type 2 diabetes mellitus, and hypertension. Average baseline LDL‐C, non–HDL‐C, and Apo‐B measures were 3.81 mmol/L, 4.42 mmol/L, and 1.16 g/L, respectively.

Table 1.

Summary Table of Characteristics

Trial Characteristics	LDL‐C	Non–HDL‐C	Apo‐B
Trial number, N	108	102	37
Total participants	5582	5401	1506
Trial size (participants)a	32 (4–352)	32 (4–352)	32 (4–130)
Male:female ratiob, c	37:63	39:61	51:49
Age, yc, d	54 (44–59)	54 (44–59)	54 (43–60)
Inpatient:outpatient settingb	4:96	3:97	3:97
Baseline serum leveld, e	3.7 (3.0–4.2) mmol/L	4.4 (3.8–5.0) mmol/L	1.2 (1–1.4) g/L
Crossover:parallel study designb	54:46	54:46	57:43
Amount of substitution, gd	29 (23–49)	30 (22–50)	30 (25–50)
Follow‐up duration, wksa	6 (3–208)	6 (3–208)	6 (3–52)
Funding sources (agency:industry:agency‐industry:NR)b	23:19:48:9	23:19:49:10	19:32:43:5
Plant protein source, N	Soy, 91; lupin, 3; legumes, 3; pinto beans, 2; pulses, 2; barley, 1; pea, 1; walnut, 1; various, 4	Soy, 84; legumes, 3; lupin, 3; pinto beans, 2; pulses, 2; barley, 1; pea, 1; walnut, 1; various, 5	Soy, 34; legumes, 1; walnut, 1; various, 1
Animal protein source, N	Dairy, 70; meat, 10; chicken noodle soup, 2; egg, 1; various, 25	Dairy, 64; meat, 10; chicken noodle soup, 2; egg, 1; various, 25	Dairy, 25; meat, 3; egg, 1; various, 8
Protein form, N	Whole food, 38; protein isolate, 72	Whole food, 40; protein isolate, 63	Whole food, 10; protein isolate, 28

Apo‐B indicates apolipoprotein B; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; and NR, not reported.

^aValues are reported as medians (ranges).

^bValues are reported as percentage ratios.

^cIncludes baseline data before dropouts, where final data were not available.

^dValues are reported as medians (interquartile ranges).

^eBaseline serum‐level data correspond to the respective lipid marker for each end point.

Of 112 trials, 94 used soy as the sole plant protein intervention, and 74 used dairy as the sole animal protein intervention. Other plant protein sources included various pulses, nuts, barley, and seeds; other animal protein sources included meat, fatty fish, and eggs. Seventy‐one trials used protein isolate products, 37 used whole foods, and 4 used a combination of the two. The median protein substitution was ≈30 g/d. Trial follow‐up ranged from 3 weeks to 4 years, with a median follow‐up of 6 weeks. Twenty‐five trials obtained funding from publicly funded agencies alone, 22 were supported by industry funding alone, and 55 used a combination of the two.

Most of our included trials were deemed to be “low risk of bias” or “unclear risk of bias” across most domains by the Cochrane Risk of Bias tool. Of the trials rated has high risk of bias, 3 were for allocation concealment, 3 were for blinding, 14 were for incomplete outcome data, and 5 were for selective outcome reporting; 1 trial was considered to have an alternative high‐risk source of bias (substantial macronutrient imbalance in protein interventions for tofu compared with cheese, in the trial by Meredith et al107). Detailed risk of bias assessment data can be found in Figure S1.

Effect on LDL‐C

Figure 2 and Figures S2 and S3 show the effect of plant protein in substitution for animal protein intake on LDL‐C across 108 trials. We found a significant reduction in LDL‐C (MD, −0.16 mmol/L [95% CI, −0.20 to −0.12 mmol/L]; P<0.00001), with evidence of substantial interstudy heterogeneity (I² =55%; P<0.00001). Fixed‐effects model analysis, bootstrap analysis (Table S3), and sensitivity analyses did not alter the direction or significance of the effect estimates. Subgroup analyses were nonsignificant and failed to explain heterogeneity (Figure S4). Post hoc subgroup analyses (Figure S5) failed to identify significant effect modification by protein form on LDL‐C, and post hoc dose‐response analyses (Table S4) did not find a dose threshold for LDL‐C in continuous subgroup analyses.

Figure 2.

Primary analyses. Pooled effect estimates for each end point (squares) shown. Paired analyses were applied to all crossover trials. Data are expressed as mean differences (95% confidence intervals [CIs]), using generic inverse‐variance random‐effects models. Interstudy heterogeneity was tested using the Cochran Q statistic (χ²) at a significance level of P<0.10 and quantified by I²; levels of ≥50% represented substantial heterogeneity. All outcomes had significant pooled effect estimates. Heterogeneity was significant and substantial for low‐density lipoprotein cholesterol (LDL‐C) and non–high‐density lipoprotein cholesterol (HDL‐C), and significant but not substantial for apolipoprotein B (Apo‐B).

Effect on Non–HDL‐C

Figure 2 and Figures S6 and S7 show the effect of plant protein in substitution for animal protein intake on non–HDL‐C across 102 trials. We found a significant reduction in non–HDL‐C (MD, −0.18 mmol/L [95% CI, −0.22 to −0.14 mmol/L]; P<0.00001), with evidence of substantial interstudy heterogeneity (I²=52%; P<0.00001). Fixed‐effects model analysis, bootstrap analysis (Table S3), and sensitivity analyses did not alter the direction or significance of the effect estimates. Subgroup analyses, however, did reveal a greater reduction in non‐HDL‐C in trials with higher baseline non‐HDL‐C levels (between‐subgroup difference, −0.09 mmol/L [95% CI, −0.17 to −0.01 mmol/L]; P=0.03), with a residual I²=43% (Figure S8). Post hoc subgroup analyses (Figure S5) failed to identify significant effect modification by protein form on non–HDL‐C, and post hoc dose‐response analyses (Table S4) did not find a dose threshold in continuous subgroup analyses.

Effect on Apo‐B

Figure 2 and Figures S9 and S10 show the effect of plant protein in substitution for animal protein intake on Apo‐B across 37 trials. We found a significant reduction in Apo‐B by plant protein (MD, −0.05 g/L [95% CI, −0.06 to −0.03 g/L]; P<0.00001), with evidence of moderate interstudy heterogeneity (I² =30%; P=0.05). Fixed‐effects model analysis, bootstrap analysis (Table S3), and sensitivity analyses did not alter the direction or significance of the effect estimates. Subgroup analyses also did not explain the heterogeneity (Figure S11). However, removal of the 2007 study by Azadbakht et al51 modified heterogeneity from significant to nonsignificant (I² =21%; P=0.14). Post hoc subgroup analyses (Figure S5) failed to identify significant effect modification by protein form on non–HDL‐C, and post hoc dose‐response analyses (Table S4) did not find a dose threshold in continuous subgroup analyses.

Publication Bias

Figure S12 shows the funnel plots used to evaluate publication bias; on visual inspection, there was no evidence of asymmetry or small‐study effects for any outcome. The Egger test identified significant publication bias for LDL‐C (P=0.03), but the Begg test was nonsignificant. The Egger and Begg tests were nonsignificant across all other end points. Trim‐and‐fill analyses were conducted for LDL‐C, with data for 8 additional studies imputed to adjust for funnel plot asymmetry (Figure S13). There was no evidence of meaningful small‐study effects. The direction, significance, and size of the pooled effect estimate after inclusion of the imputed studies were not significantly altered (MD, −0.18 mmol/L [95% CI, −0.21 to −0.14 mmol/L]; P<0.001).

GRADE Assessment

Table S5 shows a summary of the GRADE assessments for each end point. The evidence for both LDL‐C and non–HDL‐C was rated moderate quality, on the basis of a downgrade for inconsistency in both analyses. The evidence for Apo‐B was rated moderate quality, on the basis of a downgrade for imprecision.

Discussion

We conducted a systematic review and meta‐analysis of 112 RCTs assessing the effect of plant protein versus animal protein on established lipid targets for CVD prevention in 5774 adult participants with and without hyperlipidemia. Plant protein substitution for animal protein led to modest reductions in LDL‐C (−0.16 mmol/L or ≈4%; 95% CI, ≈3%–5%), non–HDL‐C (−0.18 mmol/L or ≈4%; 95% CI, ≈3%–5%), and Apo‐B (−0.05 g/L or ≈3%; 95% CI, 2%–5%). On the basis of studies finding a one‐to‐one relationship between LDL‐C and cardiovascular risk reductions, these findings would translate to a 4% risk in major cardiovascular events.148, 149

Findings in Relation to the Literature

Our findings are supported by other systematic reviews and meta‐analyses of the effect of individual sources of plant protein in substitution for different macronutrients (not just animal protein) on blood lipids. We showed, in an updated analysis of an American Heart Association analysis, that soy protein produced similar decreases in LDL‐C (≈4%) in RCTs involving participants with and without hyperlipidemia.9 An individual patient‐level pooled analysis of RCTs showed that tree nuts decrease LDL‐C by ≈7%, along with other lipid end points.10 A systematic review and meta‐analysis of the effect of dietary pulses on established lipid targets showed an LDL‐C–lowering effect of ≈5% and a tendency for a non–HDL‐C–lowering effect.12

Our findings are also aligned with previous evidence related to plant protein as part of plant‐based dietary patterns. A systematic review of 13 observational studies and 14 RCTs trials demonstrated the lipid‐lowering benefits of plant‐based diets,6 and a recent systematic review and meta‐analysis of 11 RCTs found significant reductions in LDL‐C and non–HDL‐C following a vegetarian diet.150 We have shown that the Portfolio diet, which combines cholesterol‐lowering foods (including plant protein from soy, pulses, and nuts) along with viscous fibers and plant sterols, produces LDL‐C reductions comparable to lovastatin (−28.6% versus −30.9%) over 4 weeks when all foods were provided. 151 There were more modest reductions of 10% to 15% (with greater reductions seen with greater adherence) when the diet was administered as dietary advice under free living conditions over 6 months.152 Our Eco‐Atkins trial also found greater reductions in LDL‐C with a vegan low‐carbohydrate (“Eco‐Atkins”) diet that emphasizes plant proteins, compared with a high‐carbohydrate, low‐fat, lacto‐ovo vegetarian diet (treatment difference, −0.49 mmol/L).153

Furthermore, studies have found an association between plant‐based diets and cardiovascular disease. The PREDIMED (Prevención con Dieta Mediterránea) trial showed that a predominantly plant‐based Mediterranean diet supplemented with nuts as a source of plant protein decreases major cardiovascular events.154 Prospective cohort studies offer further support showing that dietary patterns high in plant proteins, such as Mediterranean and vegetarian dietary patterns, are associated with reduced cardiovascular events.155, 156, 157, 158 An analysis of the Harvard cohorts found that low‐carbohydrate and high‐protein diets were associated with increased mortality, but inversely correlated with mortality and particularly CVD mortality when based on plant protein.159 Other prospective cohort studies have also shown that plant‐based diets are associated with a mortality benefit.160 On the other hand, increased intake of animal protein sources has been associated with negative health outcomes. A pooled analysis of the Harvard cohorts found that red meat consumption was associated with increased risks of total, cardiovascular, and cancer mortality.161 Other large, prospective, cohort studies have found an association between animal protein sources and disease or mortality.162, 163, 164

There are several mechanisms by which plant protein may exert a lipid‐lowering effect. One explanation is that the plant protein source acts as a vehicle for other established antiatherogenic agents, such as plant sterols or soluble fiber; similarly, the displaced animal protein source could also act as a vehicle for hypercholesterolemic agents, such as saturated fat and cholesterol.13, 14, 15, 24 Interestingly, our post hoc subgroup analyses did not find a significant difference between protein isolate products and whole food sources for any given end point, suggesting that the cholesterol‐lowering effects are at least, in part, attributable to the plant protein itself rather than just the associated nutrients.

An alternative explanation relates to the amino acid breakdown encountered in plant proteins versus animal proteins; in particular, lysine, which is more prevalent in animal proteins, has been shown to increase cholesterol levels in animal models, whereas arginine, which is found more in plant proteins, has been found to have the opposite effect.165, 166, 167 The cholesterol‐lowering effect of arginine has also been demonstrated in a 5‐week arginine feeding trial in humans,168 but otherwise there are limited human studies investigating this subject. Proposed mechanisms for these effects involve bile acid production and binding of hepatic LDL receptors.166, 169

A Priori Subgroup Analyses

Our results appear to be robust to different trial conditions. Similar to a previous meta‐analysis by Anderson et al,170 we did find that increased baseline values amplified the effects seen in non–HDL‐C reduction. However, our overall analyses indicate that the lipid‐lowering effects of plant protein apply to both hypercholesterolemic and normal subjects, because the normocholesterolemic subgroup also showed a significant improvement in non–HDL‐C, and similar subgroup analyses in LDL‐C and Apo‐B were nonsignificant. The beneficial effects otherwise held across a range of ages and health statuses, and all other subgroup analyses were nonsignificant.

Strengths and Limitations

Our systematic review and meta‐analysis has several strengths and limitations. The strengths include the identification of all available evidence through a systematic search strategy, the inclusion of RCTs that provide the greatest protection against bias, quantitative syntheses of the data, and assessment of the overall quality of the evidence using the GRADE system.

The limitations of our systematic review and meta‐analysis relate to inconsistency in the treatment effects and imprecision. Evidence of unexplained inconsistency in treatment effects was seen for 2 of the established therapeutic lipid end points. There was substantial interstudy heterogeneity in our LDL‐C and non–HDL‐C analyses, which was not fully explained by sensitivity or subgroup analyses. Evidence of imprecision was seen in Apo‐B, because the 95% CI for effect estimates for Apo‐B overlapped the prespecified minimally important difference of 0.04 g/L. Apo‐B also showed evidence of moderate interstudy heterogeneity; however, the statistical significance of heterogeneity was eliminated by the removal of the 2007 study by Azadbakht et al.51 We also considered downgrading for indirectness of the evidence. A relatively large proportion of the available trials evaluated soy as the sole plant protein source (94 of 112 trials) and/or dairy as the sole animal protein source (74 of 112 trials). Subgroup analyses, however, did not reveal evidence of significant effect modification by protein sources across any of the 3 end points, which suggests that the effects seen apply across varying plant and animal protein sources. Several plant protein sources, however, were not evaluated, including wheat (gluten), rice, and other grains. In addition, there were limited studies with extended follow‐up duration, which would help assess issues of long‐term adherence.

Taking into account these strengths and limitations, the evidence was assessed by the GRADE system as moderate quality for a cholesterol‐lowering effect of plant protein in substitution for animal protein across LDL‐C, non–HDL‐C, and Apo‐B markers.

Implications

Current adult protein intakes average ≈80 to 100 g/d in the United States and Europe. Of this intake, ≈30% is from plant protein sources.171, 172 The median intervention of 30 g protein substitution per day across trials included in our analyses reflects the substitution of 1 to 2 servings of meat for plant protein substitutes or 3 250‐mL cups of dairy milk for soy milk. This additional substitution would mean a shift to diets with >50% plant protein, which can be attained by following healthy dietary patterns, such as vegetarian, Mediterranean, and Portfolio dietary patterns.173, 174, 175 Given the low current consumption of plant protein‐rich foods, such as soy and pulses, in Canada and the United States, there remains a significant opportunity to realize the benefits of making such dietary changes.176, 177, 178

Although the reductions in LDL‐C, non–HDL‐C, and Apo‐B on their own were modest (<5%), plant protein can still contribute to meaningful reductions in lipids. On the basis of the evidence from the Portfolio diet, the lipid‐lowering effects of individual food components, which include plant protein from soy, pulses, and nuts, are additive, such that the LDL‐C–lowering effect (≈5%–10%) of each of the 4 components of the Portfolio diet food can be summed to achieve meaningful reductions.3, 147, 148 Several large trials and cohort studies have shown that such reductions are associated with improved cardiovascular outcomes.179, 180, 181, 182, 183, 184, 185 The 2016 Canadian Cardiovascular Guidelines further highlighted the superior predictive value for CVD of non–HDL‐C and Apo‐B, both of which were reduced by plant protein.3 The implication is that plant protein as part of a comprehensive lipid‐lowering dietary pattern alone or as an add‐on to other lipid‐lowering therapy can help people achieve their lipid targets and reduce CVD risk.

Despite the existing evidence for benefit, current dietary guidelines do not wholly reflect the demonstrated benefits of plant protein versus animal protein and tend to place animal sources of protein on the same level as plant sources.20, 21, 22 In particular, the 2015 to 2020 Dietary Guidelines for Americans recommend seafood, meats, poultry, eggs, nuts, seeds, and soy products indiscriminately as options for protein sources and suggest that the vegetarian dietary patterns described are only for those already following a vegetarian diet (which is incongruent with the Scientific Report of the 2015 Dietary Guidelines Advisory Committee on which the the 2015 to 2020 Dietary Guidelines for Americans is based).8, 22, 23

Conclusions

In conclusion, our aggregate analyses demonstrate a benefit of plant protein in substitution for animal protein on established lipid targets for CVD prevention in adults with and without hyperlipidemia. To our knowledge, this is the first systematic review and meta‐analysis to directly evaluate the effects of plant protein as well as plant for animal protein replacement. These findings presents an opportunity for patients, clinicians, and guidelines to exploit the lipid‐lowering benefits of a sustainable plant‐based dietary strategy that is associated with improved overall health outcomes. Our confidence in the evidence for the LDL‐C–, non–HDL‐C–, and Apo‐B–lowering effects of plant protein, however, is limited by inconsistency for LDL‐C and non–HDL‐C and imprecision for Apo‐B. Further large, high‐quality, randomized controlled trials investigating plant protein sources beyond soy, particularly in young and healthy participants, would be useful to help better understand the role of plant protein in cardiovascular risk reduction.

Author Contributions

All authors had full access to all of the data (including statistical reports and tables) in this study and take full responsibility for the integrity of the data and the accuracy of the data analysis. Conception and design: Li and Sievenpiper. Analysis and interpretation of the data: Li, Blanco Mejia, de Souza, Leiter, Kendall, Jenkins, and Sievenpiper. Drafting of the article: Li. Critical revision of the article for important intellectual content: Li, Lytvyn, Blanco Mejia, Stewart, Viguiliouk, Ha, de Souza, Leiter, Kendall, Jenkins, and Sievenpiper. Final approval of the article: Li, Lytvyn, Blanco Mejia, Stewart, Viguiliouk, Ha, de Souza, Leiter, Kendall, Jenkins, and Sievenpiper. Statistical expertise: de Souza. Attainment of funding: Kendall, Jenkins, and Sievenpiper. Administrative, technical, or logistic support: Blanco Mejia. Collection and assembly of data: Li, Lytvyn, Blanco Mejia, Stewart, Viguiliouk, and Ha. Guarantor: Sievenpiper.

Sources of Funding

This work was funded by the Canadian Institutes of Health Research (CIHR; funding reference number 129920) through the Canada‐Wide Human Nutrition Trialists' Network. The Diet, Digestive Tract, and Disease Centre, funded through the Canada Foundation for Innovation and the Ministry of Research and Innovation's Ontario Research Fund, provided the infrastructure for the conduct of this project. Jenkins was funded by the Government of Canada through the Canada Research Chair Endowment. Sievenpiper was funded by a PSI Graham Farquharson Knowledge Translation Fellowship, Diabetes Canada Clinician Scientist award, CIHR INMD/Canadian Nutrition Society New Investigator Partnership Prize, and Banting & Best Diabetes Centre Sun Life Financial New Investigator Award. Viguiliouk was supported by a Toronto 3D Knowledge Synthesis and Clinical Trials Foundation Internship Award. None of the sponsors had a role in any aspect of the present study, including design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, and approval of the article or decision to publish.

Disclosures

Lytvyn is part of the Grading of Recommendations Assessment, Development, and Evaluation Working Group. Ha received support from a Canadian Institutes of Health Research (CIHR) doctoral award, David Sackett scholarship, and Ashbaugh Graduate scholarship. She has received payment from the World Health Organization (WHO) for work on a systematic review and meta‐analysis commissioned by the WHO for work on the relation of saturated fatty acids and polyunsaturated fatty acids with health outcomes. She and her peers received a cash prize for placing second in the regional “Mission Impulsible” Competition hosted by Pulse Canada, where they conceived and developed a marketable food product that contained dietary pulses. She received a travel award to attend the “Journey Through Science Day,” hosted by PepsiCo and the New York Academy of Sciences, and the Nutrica Travel Award from the Diabetes and Nutrition Study Group of the European Association for the Study of Diabetes (EASD). de Souza has served as an external resource person to the World Health Organization's Nutrition Guidelines Advisory Group on trans fats, saturated fats, and polyunsaturated fats. The WHO paid for his travel and accommodation to attend meetings from 2012‐2017 to present and discuss this work. He has also done contract research for the Canadian Institutes of Health Research's Institute of Nutrition, Metabolism, and Diabetes, Health Canada, and the World Health Organization for which he received remuneration. He has held a grant from the Canadian Foundation for Dietetic Research as a principal investigator, and is a co‐investigator on several funded team grants from Canadian Institutes of Health Research. He received compensation for a lecture on dietary fat given at McMaster Pediatric Nutrition Days in 2016. Kendall has received research support from the Advanced Foods and Materials Network, Agricultural Bioproducts Innovation Program through the Pulse Research Network, Agriculture and Agri‐Food Canada, Almond Board of California, Barilla, Calorie Control Council, CIHR, Canola Council of Canada, INC International Nut and Dried Fruit Council Foundation, National Dried Fruit Trade Association, Kellogg, Loblaw Companies Ltd., Pulse Canada, Saskatchewan Pulse Growers and Unilever. He has received consultant fees from American Pistachio Growers; speaker fees from American Peanut Council, Tate & Lyle and The WhiteWave Foods Company; and travel funding from Sabra Dipping Company, Tate & Lyle, International Tree Nut Council Research & Education Foundation, California Walnut Commission, Sun‐Maid, The Peanut Institute, General Mills, Oldways Foundation and International Nut and Dried Fruit Council Foundation. He is a member of the Clinical Practice Guidelines Expert Committee for Nutrition Therapy of the European Association for the Study of Diabetes (EASD), the Diabetes and Nutrition Study Group of the EASD and the International Carbohydrate Quality Consortium, and is the Director for the Toronto 3D Knowledge Synthesis and Clinical Trials Foundation. Jenkins has received research grants from Saskatchewan Pulse Growers, the Agricultural Bioproducts Innovation Program through the Pulse Research Network, the Advanced Foods and Material Network, Loblaw Companies Ltd., Unilever, Barilla, the Almond Board of California, Agriculture and Agri‐food Canada, Pulse Canada, Kellogg's Company, Canada, Quaker Oats, Canada, Procter & Gamble Technical Centre Ltd., Bayer Consumer Care, Springfield, NJ, Pepsi/Quaker, International Nut & Dried Fruit (INC), Soy Foods Association of North America, the Coca‐Cola Company (investigator initiated, unrestricted grant), Solae, Haine Celestial, the Sanitarium Company, Orafti, the International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, the Canola and Flax Councils of Canada, the Calorie Control Council, the CIHR, the Canada Foundation for Innovation and the Ontario Research Fund. He has received in‐kind supplies for trial as a research support from the Almond board of California, Walnut Council of California, American Peanut Council, Barilla, Unilever, Unico, Primo, Loblaw Companies, Quaker (Pepsico), Pristine Gourmet, Bunge Limited, Kellogg Canada, WhiteWave Foods. He has been on the speaker's panel, served on the scientific advisory board and/or received travel support and/or honoraria from the Almond Board of California, Canadian Agriculture Policy Institute, Loblaw Companies Ltd, the Griffin Hospital (for the development of the NuVal scoring system, the Coca‐Cola Company, EPICURE, Danone, Diet Quality Photo Navigation (DQPN), FareWell, Verywell, True Health Initiative, Saskatchewan Pulse Growers, Sanitarium Company, Orafti, the Almond Board of California, the American Peanut Council, the International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, Herbalife International, Pacific Health Laboratories, Nutritional Fundamental for Health, Barilla, Metagenics, Bayer Consumer Care, Unilever Canada and Netherlands, Solae, Kellogg, Quaker Oats, Procter & Gamble, the Coca‐Cola Company, the Griffin Hospital, Abbott Laboratories, the Canola Council of Canada, Dean Foods, the California Strawberry Commission, Haine Celestial, PepsiCo, the Alpro Foundation, Pioneer Hi‐Bred International, DuPont Nutrition and Health, Spherix Consulting and WhiteWave Foods, the Advanced Foods and Material Network, the Canola and Flax Councils of Canada, the Nutritional Fundamentals for Health, Agri‐Culture and Agri‐Food Canada, the Canadian Agri‐Food Policy Institute, Pulse Canada, the Saskatchewan Pulse Growers, the Soy Foods Association of North America, the Nutrition Foundation of Italy (NFI), Nutra‐Source Diagnostics, the McDougall Program, the Toronto Knowledge Translation Group (St. Michael's Hospital), the Canadian College of Naturopathic Medicine, The Hospital for Sick Children, the Canadian Nutrition Society (CNS), the American Society of Nutrition (ASN), Arizona State University, Paolo Sorbini Foundation and the Institute of Nutrition, Metabolism and Diabetes. He received an honorarium from the United States Department of Agriculture to present the 2013 W.O. Atwater Memorial Lecture. He received the 2013 Award for Excellence in Research from the International Nut and Dried Fruit Council. He received funding and travel support from the Canadian Society of Endocrinology and Metabolism to produce mini cases for the Canadian Diabetes Association (CDA). He is a member of the International Carbohydrate Quality Consortium (ICQC). His wife is a director and partner of Glycemic Index Laboratories, Inc., and his sister received funding through a grant from the St. Michael's Hospital Foundation to develop a cookbook for one of his studies.Sievenpiper has received research support from the Canadian Institutes of health Research (CIHR), Diabetes Canada, PSI Foundation, Banting and Best Diabetes Centre (BBDC), Canadian Nutrition Society (CNS), American Society for Nutrition (ASN), Calorie Control Council, INC International Nut and Dried Fruit Council Foundation, National Dried Fruit Trade Association, The Tate and Lyle Nutritional Research Fund at the University of Toronto, and The Glycemic Control and Cardiovascular Disease in Type 2 Diabetes Fund at the University of Toronto (a fund established by the Alberta Pulse Growers). He has received speaker fees and/or honoraria from Diabetes Canada, Canadian Nutrition Society (CNS), Dr. Pepper Snapple Group, Dairy Farmers of Canada, Nutrition Foundation of Italy (NFI), C3 Collaborating for Health, Sprim Brasil, WhiteWave Foods, Rippe Lifestyle, mdBriefcase, Alberta Milk, FoodMinds LLC, Memac Ogilvy & Mather LLC, PepsiCo, The Ginger Network LLC, International Sweeteners Association, and Pulse Canada. He has ad hoc consulting arrangements with Winston & Strawn LLP, Perkins Coie LLP, and Tate & Lyle. He is a member of the European Fruit Juice Association Scientific Expert Panel. He is on the Clinical Practice Guidelines Expert Committees of Diabetes Canada, European Association for the study of Diabetes (EASD), and Canadian Cardiovascular Society (CCS), as well as an expert writing panel of the American Society for Nutrition (ASN). He serves as an unpaid scientific advisor for the Food, Nutrition, and Safety Program (FNSP) and the Technical Committee on Carbohydrates of the International Life Science Institute (ILSI) North America. He is a member of the International Carbohydrate Quality Consortium (ICQC), Executive Board Member of the Diabetes and Nutrition Study Group (DNSG) of the EASD, and Director of the Toronto 3D Knowledge Synthesis and Clinical Trials foundation. His wife is an employee of Unilever Canada. No competing interests were declared by Li, Blanco Mejia, Stewart, Viguiliouk, and Leiter. There are no patents, products in development, or marketed products to declare.

Supporting information

Table S1. Search Strategy

Table S2. Full Table of Characteristics

Table S3. Bootstrap Analyses

Table S4. Post‐Hoc Dose Response

Table S5. GRADE Assessment

Figure S1. Cochrane risk of bias. Risk of bias assessment using Cochrane Risk of Bias Tool.

Figure S2. LDL‐C forest plot, random‐effects model.

Figure S3. LDL‐C forest plot, fixed‐effects model.

Figure S4. LDL‐C visual subgroup. Point estimates for each subgroup level (squares) are the pooled effect estimates. The dashed line represents the pooled effect estimate for the overall (total) analysis. The residual I² value indicates the interstudy heterogeneity unexplained by the subgroup. Statistically significant pairwise subgroup effect modification by meta‐regression analyses at P<0.05.

Figure S5. Post‐hoc subgroups. Point estimates for each subgroup level (squares) are the pooled effect estimates. The dashed line represents the pooled effect estimate for the overall (total) analysis. The residual I² value indicates the interstudy heterogeneity unexplained by the subgroup. Statistically significant pairwise subgroup effect modification by meta‐regression analyses at P<0.05.

Figure S6. Non‐HDL‐C forest plot, random‐effects model.

Figure S7. Non‐HDL‐C forest plot, fixed‐effects model.

Figure S8. Non‐HDL‐C visual subgroup. Point estimates for each subgroup level (squares) are the pooled effect estimates. The dashed line represents the pooled effect estimate for the overall (total) analysis. The residual I² value indicates the interstudy heterogeneity unexplained by the subgroup. Statistically significant pairwise subgroup effect modification by meta‐regression analyses at P<0.05.

Figure S9. Apo‐B forest plot, random‐effects model. HC indicates hypercholesterolemic; IF, isoflavones; LF, low‐fat; N, normal; NIF, no isoflavones. The pooled effect estimate (diamond) is shown. Paired analyses were applied to all crossover trials. Data are expressed as MDs with 95% CIs, using generic inverse‐variance random‐effects models. Inter‐study heterogeneity was tested using the Cochran Q statistic (chi‐square) at a significance level of P<0.10 and quantified by I², levels of ≥50% represented substantial heterogeneity.

Figure S10. Apo‐B forest plot, fixed‐effects model. HC indicates hypercholesterolemic; IF, isoflavones; LF, low‐fat; N, normal; NIF, no isoflavones. The pooled effect estimate (diamond) is shown. Paired analyses were applied to all crossover trials. Data are expressed as MDs with 95% CIs, using generic inverse‐variance fixed‐effects models. Inter‐study heterogeneity was tested using the Cochran Q statistic (chi‐square) at a significance level of P<0.10 and quantified by I², levels of ≥50% represented substantial heterogeneity.

Figure S11. Apo‐B visual subgroup. Point estimates for each subgroup level (squares) are the pooled effect estimates. The dashed line represents the pooled effect estimate for the overall (total) analysis. The residual I² value indicates the interstudy heterogeneity unexplained by the subgroup. Statistically significant pairwise subgroup effect modification by meta‐regression analyses at P<0.05.

Figure S12. Funnel plots. Publication bias funnel plots for LDL (A), non‐HDL (B), and apolipoprotein B (C). The solid line represents the pooled effect estimate expressed as the weighted mean difference (MD) of each analysis, and dashed lines represent pseudo‐95% confidence limits. Circles represent effect estimates of included trials. P‐values of Egger and Begg tests for publication bias are shown at top right for each analysis. *Statistically significant (P<0.05).

Figure S13. LDL‐C trim‐and‐fill funnel plot. The horizontal line represents the pooled effect estimate expressed as a mean difference. The diagonal lines represent the pseudo 95% CIs of the mean difference. The clear circles represent effect estimates for each included study.

(J Am Heart Assoc. 2017;6:e006659 DOI: 10.1161/JAHA.117.006659.)29263032