A Meta-Analysis of 46 Studies Identified by the FDA Demonstrates that Soy Protein Decreases Circulating LDL and Total Cholesterol Concentrations in Adults

ABSTRACT

Background

Certain plant foods (nuts and soy protein) and food components (viscous fibers and plant sterols) have been permitted by the FDA to carry a heart health claim based on their cholesterol-lowering ability. The FDA is currently considering revoking the heart health claim for soy protein due to a perceived lack of consistent LDL cholesterol reduction in randomized controlled trials.

Objective

We performed a meta-analysis of the 46 controlled trials on which the FDA will base its decision to revoke the heart health claim for soy protein.

Methods

We included the 46 trials on adult men and women, with baseline circulating LDL cholesterol concentrations ranging from 110 to 201 mg/dL, as identified by the FDA, that studied the effects of soy protein on LDL cholesterol and total cholesterol (TC) compared with non-soy protein. Two independent reviewers extracted relevant data. Data were pooled by the generic inverse variance method with a random effects model and expressed as mean differences with 95% CI. Heterogeneity was assessed and quantified.

Results

Of the 46 trials identified by the FDA, 43 provided data for meta-analyses. Of these, 41 provided data for LDL cholesterol, and all 43 provided data for TC. Soy protein at a median dose of 25 g/d during a median follow-up of 6 wk decreased LDL cholesterol by 4.76 mg/dL (95% CI: −6.71, −2.80 mg/dL, P < 0.0001; I² = 55%, P < 0.0001) and decreased TC by 6.41 mg/dL (95% CI: −9.30, −3.52 mg/dL, P < 0.0001; I² = 74%, P < 0.0001) compared with non-soy protein controls. There was no dose–response effect or evidence of publication bias for either outcome. Inspection of the individual trial estimates indicated most trials (∼75%) showed a reduction in LDL cholesterol (range: −0.77 to −58.60 mg/dL), although only a minority of these were individually statistically significant.

Conclusions

Soy protein significantly reduced LDL cholesterol by approximately 3–4% in adults. Our data support the advice given to the general public internationally to increase plant protein intake. This trial was registered at clinicaltrials.gov as NCT03468127.

Introduction

The FDA has revisited the data on soy protein and cholesterol reduction and concluded that an unqualified (i.e., significant scientific agreement) heart health claim is no longer justified based on 46 trials (1). In 2006, a scientific advisory from the Nutrition Committee of the AHA reached a similar conclusion (2). However, the AHA never meta-analyzed the 22 trials on which it based its conclusion. Later, when a meta-analysis was carried out, soy protein was found to reduce LDL cholesterol by a significant 4.3% (3). At that time, the displacement (extrinsic) value of soy foods was also highlighted. This reduction occurs when soy foods displace commonly consumed sources of protein in the diet, such as saturated fat-containing meat and dairy. It was estimated that displacement will further reduce LDL cholesterol by approximately 4% (3). At the time of the 2006 AHA soy advisory, the Nutrition Committee acknowledged the heart healthy profile of soy foods as protein-rich foods as had the Nutrition Committee at the time of the 2000 AHA soy advisory (4). Soy foods were recognized as low in saturated fat, lacking in cholesterol, and a source of omega-3 (n–3) fatty acids, but the potentially nonsignificant 3% LDL cholesterol reduction was not considered clinically relevant (2).

However, the small reduction in LDL cholesterol for soy is similar to permitted claims for other plant foods and food components (nuts, viscous fibers, plant sterols, etc.) (5–7). Another advantage of soy protein and soy foods is that soy foods in combination with viscous fiber, nuts, and plant sterols/stanols reduce LDL cholesterol by as much as 29% (8). Soy has been included in numerous clinical practice guidelines for cardiovascular disease (CVD) risk reduction (e.g., those by the National Cholesterol Education Program, Adult Treatment Panel III, Heart UK, European Atherosclerosis Society, and Canadian Cardiovascular Society) (9–12).

To determine whether data under current consideration by the FDA indicate that the previous meta-analysis-based conclusions must be changed, we conducted a meta-analysis of the clinical trials on which the FDA is prepared to make its final decision regarding a soy protein health claim.

Methods

This meta-analysis was conducted according to the Cochrane Handbook for Systematic Reviews and Interventions (13). Results were reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (14).

Data sources

We used the 46 trials compiled by the FDA (1, 4) in its review of the scientific evidence on soy protein intake and lipid markers (LDL cholesterol and TC) as surrogates of CVD.

Study selection

Forty-six (15–60) dietary trials in humans that tested the effects of soy protein reporting effects on surrogate markers of CVD, LDL cholesterol, and/or TC as listed by the FDA were considered for meta-analysis.

Data extraction

Trial characteristics and outcome data were double extracted by SBM, SSL, EV, or LC. Extracted trial characteristics included trial design, duration, sample size, participant characteristics, soy protein description, outcome data, and funding information. The outcome data included LDL cholesterol and TC. Plot Digitizer version 2.6.8 (61) was used to extract data from graphs, where applicable. Any discrepancies in data extraction were resolved through consensus.

Risk of bias assessment

Two independent reviewers (SBM, SSL, EV, or LC) assessed the methodological quality of the included trials using the Cochrane Collaboration risk-of-bias tool (62). Assessment was done across 5 domains of bias (sequence generation, allocation concealment, blinding, incomplete outcome data, and selective reporting) and assessed. The risk of bias was assessed as low (proper methods taken to reduce bias), high (improper methods creating bias), or unclear (insufficient information provided to determine the bias level). Final assessments were based on consensus among reviewers.

Statistical methods

Data were pooled using the generic inverse variance method with random effects model and expressed as mean differences (MDs) with 95% CIs. We used Review Manager version 5.3 (The Nordic Cochrane Centre, The Cochrane Collaboration) for primary and sensitivity analyses and Stata version 13 (StataCorp) for publication bias and dose–response analyses. When a trial had multiple soy protein or control arms, these were pooled to obtain a single pairwise comparison to avoid a unit-of-analysis error (13). However, when a trial provided multiple comparisons that had independent treatment and control arms, these arms were included as separate comparisons. Pairwise analyses were applied to all crossover trials as described by Elbourne et al. (63).

Change-from-baseline values were favored over end differences. Paired analyses were applied to all trials with the use of a within-individual correlation coefficient between treatments of 0.5 as described by Elbourne et al. (63). Where variance data were missing, the average SD of the baseline and end periods across all other included trials was used to impute the SEM difference, taking into account the imputed trial's sample size (13).

Interstudy heterogeneity was evaluated by the Cochran Q statistic and quantified using the I² statistic. A P value of <0.10 was considered significant, and an I² value of 50% or higher was considered evidence of substantial heterogeneity (13). Sources of heterogeneity were investigated by sensitivity analyses in which each trial was removed separately and the pooled effect re-estimated. A study was considered influential when its removal changed the significance or significantly altered the overall heterogeneity. Sources of heterogeneity were also explored according to 9 a priori subgroup analyses: 1) baseline LDL cholesterol concentrations (<135 mg/dL compared with ≥135 mg/dL); 2) soy protein dose (<25 g/d compared with ≥25 g/d); 3) soy protein food [isolated soy protein (ISP), soy food, and soy milk]; 4) comparator (animal protein, dairy protein, meat protein, mixed, and other); 5) follow-up (median); 6) trial design (parallel compared with crossover); 7) baseline cholesterol level (desirable or borderline; normal, borderline, high; and high); 8) the manner in which soy protein was provided in the studies (feeding or substitution; and added to diet); and 9) the mixture and type of soy, how it was provided, and baseline cholesterol level (desirable or borderline cholesterol level, added ISP; desirable or borderline cholesterol level, feeding or substitution studies with ISP; desirable or borderline cholesterol level, added soy foods; desirable or borderline cholesterol level, feeding or substitution studies with soy foods; normal, borderline, and high cholesterol levels, feeding or substitution studies with ISP; high cholesterol level, added ISP; high cholesterol level, feeding or substitution studies with ISP; high cholesterol level, added soy foods; and high cholesterol level, feeding soy foods]. The last 3 a priorisubgroups were set using the criteria that the FDA used to assess the scientific evidence regarding the relationship between soy protein and CHD for intervention studies (1). We used the criteria for lipids but not for blood pressure and we refered to these as FDA subgroups. Subgroup analysis by risk of bias domains was also undertaken (sequence generation, allocation concealment, blinding of participants, personnel and outcome assessors, incomplete outcome data, and selective reporting). The significance of categorical subgroup analyses was analyzed by meta-regression.

Linear and nonlinear piecewise dose–response analyses were conducted using the metareg and mkspline commands (64).

Publication bias was assessed by visual inspection of funnel plots and by the use of the Egger (65) and Begg (66) tests. If publication bias was suspected, then adjustment for funnel plot asymmetry was done by imputing missing study data using the Duval and Tweedie trim-and-fill method (67).

Results

Forty-six studies were reviewed in full (Figure 1), 3 studies were excluded, 2 (18, 25) did not provide data on LDL cholesterol or TC, and 1 (54) was conducted in a subgroup of a larger study (53) that was already included. The remaining 43 studies provided data for LDL cholesterol (41 studies, 50 study comparisons) and for TC (43 studies, 52 study comparisons). Two studies (46, 49) did not provide data on LDL cholesterol.

FIGURE 1.

Study selection indicating the number of studies identified by the FDA and the number of studies included in the meta-analysis.

Trial characteristics

In total, the 43 studies included 2607 participants, of whom 37% were men and 63% women (Table 1). Most studies (20/43, 46.5%) included postmenopausal women and/or participants with hypercholesterolemia (21/43, 48.8%). The median age of the participants was 54.9 y (IQR: 44.8, 59.1 y). The study design was almost evenly balanced, with 23 parallel and 20 crossover studies. The median follow-up period was 6 wk (IQR: 4, 12 wk). The mean baseline LDL cholesterol and TC values were 147.6 and 226.3 mg/dL, respectively. All, except 1 study (17) reported being randomized. Most studies reported the use of soy protein, including ISP (34/43, 79.1%). Others included soy foods such as whole soy bean, soy flour, or soy milk products (3/43, 7%); soy milk (1/43, 2.3%); textured soy protein (1/43, 2.3%); or a mixture of ISP, soy foods, and/or soy milk (4/43, 9.3%). The median soy protein dose was 25 g/d (IQR: 23.8, 38.1 g/d). Most studies used dairy protein (31/43, 72.1%) as the comparator. Others used meat protein (1/43, 2.3%); mixed dairy and meat protein (3/43, 7.0%); or other foods, such as wheat protein, cellulose, cereals, biscuits, and other carbohydrates (8/43, 18.6%). The majority of the studies reported funding from both agency and industry (16/43, 37.2%) or industry alone (12/43, 27.9%), whereas 8 studies were funded by an agency alone and 7 studies failed to report funding.

TABLE 1.

Trial characteristics for the studies included in the analyses of the effect of soy protein intake on circulating LDL cholesterol and TC concentrations in adults

Study, year (reference)	Participants	Age (y)1	Baseline plasma LDL-C (mg/dL)1	Baseline plasma TC (mg/dL)1	Design	Follow-up	Test soy food/product	Soy protein dose (g/d)	Control food/product	Funding sources
Bakhit et al., 1994 (15)	21 (21 M) HC	43 (14)	157.8 (24.0)	189.9	Randomized crossover	4 wk	Muffins with ISP ± cellulose	25	Muffins with casein ± cellulose	Industry
Blum et al., 2003 (16)	24 (24 W) HC, PM	55 (5)	178.5 (28.5)	270.5 (31.9)	Randomized crossover	6 wk	Supplement packets with ISP + isoflavones	25	Supplement packets with milk protein	Industry
Bosello et al., 1988 (17)	24 (12 M, 12 W) OB	25–42	149	217	Parallel	60 d	Commercial textured preparation with soy protein + mineral salts and vitamins	∼28	Commercial textured preparation with casein + mineral salts and vitamins	N/A
Chen et al., 2006 (19)	26 (19 M, 7 W) HC, CKD	58.6 (11.4)	162.4	269.3	Randomized parallel	12 wk	Packet of ISP	30	Packet of milk protein	Agency
Cuevas et al., 2003 (20)	18 (18 W) HC, PM	59 (47–70)	194.8 (25.5)	285.9 (28.8)	Randomized crossover	4 wk	Powder with ISP + isoflavones mixed with water	40	Powder with caseinate mixed with water	AgencyIndustry
Evans et al., 2007 (21)	22 (22 W) PM	61.5 (8.2)	N/A	N/A	Randomized crossover	4 wk	Powder packets and packets of granules with soy protein + isoflavones ± lecithin; powder was mixed with water	25	Powder packets and packets of granules with caseinate ± lecithin; powder was mixed with water	Industry
Gardner et al., 2001 (23)	94 (94 W) HC, PM	59.1 (6.9)	152	230.6	Randomized parallel	12 wk	Powder packets with soy protein ± isoflavones mixed with juice, water, or soup	42	Powder packets with milk protein mixed with juice, water, or soup	AgencyIndustry
Gardner et al., 2007 (22)	28 (6 M, 22 W) HC	52 (9)	183.0 (19.0)	265.2 (51.4)	Randomized crossover	4 wk	Whole bean soy milk or isolated soy protein milk	25	Dairy milk	AgencyIndustry
Goldberg et al., 1982 (24)	4 (3 M, 1 W) NC, 12 (7 M, 5 W) HC	36.8 (18.6)43.6 (12.7)	191.0 (19.0)106.0 (31.0)	260.0 (23.0)176.0 (17.0)	Randomized crossover	6 wk	Soy milk and ISP meat analogues	∼93 ∼98	Nonfat milk and meat protein products (pork, sausage, hamburger, turkey roll, ham, and bologna)	AgencyIndustry
Greany et al., 2004 (26)	12 (12 W) PM25 (25 W) HC, PM	57.5 (2.2)	136.5 (25.9)	202.6 (5.0)	Randomized crossover	6 wk	ISP powder ± probiotic capsules	∼26	Isolated milk protein powder ± probiotic capsules	Agency
Harrison et al., 2004 (27)	213 (112 M, 101 W) HC or HTN	52 (4)	201.1 (42.5)	263.0 (42.5)	Randomized parallel	5 wk	Bread, cereal bars, and cracker biscuits fortified with soy protein ± DHA	25	Bread, cereal bars, and cracker biscuits fortified with placebo ± DHA	AgencyIndustry
Higashi et al., 2001 (28)	14 (14 M)	31 (4)	125.5	196.0	Randomized crossover	4 wk	ISP mixed with milk or yogurt	20	Milk or yogurt	Industry
Høie et al., 2005 (29)	116 (54 M, 62 W) HC	55.2 (9.5)	168.0	268.8	Randomized parallel	8 wk	Powder sachets of soy protein mixed with cold water	25	Powder sachets of milk protein mixed with cold water	N/A
Høie et al., 2005 (30)	117 (63 M, 54 W) HC	53.6 (9.6)	162.3	260.4	Randomized Parallel	8 wk	Powder sachets of soy protein mixed with cold water	15, 25	Powder sachets of milk protein mixed with cold water	N/A
Høie et al., 2006 (31)	80 (29 M, 51 W) HC	53.5	143.3	252.2	Randomized parallel	4 wk	ISP + chocolate-flavored milk + ultra-heat-treated	12.5, 25	Casein + chocolate-flavored milk + ultra-heat-treated	N/A
Høie et al., 2007 (32)	88 (34 M, 54 W) HC	54.6 (9.6)	164.2	253.4	Randomized parallel	8 wk	ISP powder dissolved in 150 mL of cold water	25	Placebo preparation made out of milk protein	Industry
Hori et al., 2001 (33)	21 (21 M) HC	44.9	166.2	265.0	Randomized parallel	3 mo	Powder drink made with soy protein hydrolysate and bound phospholipids	4.5	Powder drink made with casein hydrolyzate	N/A
Jayagopal et al., 2002 (34)	31 (31 W) PM, DM2	62.5 (6.77)2	137.3	220.4	Randomized crossover	12 wk	Sachets of ISP ± isoflavones	30	Sachets with microcrystalline cellulose	N/A
Jenkins et al., 2000 (36)	25 (15 M, 10 W) PM, HLP	60.0 (2)	184.1	268.9	Randomized crossover	3 wk	Cereal made with soy flour + isoflavones	36	Wheat protein + soy oil (7.9 g/d)	AgencyIndustry
Jenkins et al., 2002 (35)	41 (23 M, 18 W) PM, HLP	62 (12.8)	176.5	261.4	Randomized crossover	4 wk	Soy products (low-fat soy milk, soy hot dogs, breakfast links, soy burgers, cold cuts, and tofu burgers) + isoflavones	51	Dairy products (skim milk, yogurt, cottage cheese, very low-fat Hoop cheese, cheese made with skim milk) and egg substitute	AgencyIndustry
Jenkins et al., 1989 (37)	11 (11 W) OB	38 (13.3)	125.7 (43.6)	197.6 (55.1)	Randomized crossover	4 wk	Soy formula with ISP and full-fat soy flour	17.4	Milk formula with milk protein isolate and nonfat dry milk	AgencyIndustry
Kohno et al., 2006 (38)	126 (88 M, 38 W)3 95 (47 M, 48 W)3	44.8	141.4	220.9	Randomized parallel	12 wk20 wk	Candy containing soybean β-conglycinin	5	Candy containing casein	Agency
Lichtenstein et al., 2002 (39)	42 (18 M, 24 W) HC	62.7 (8.8)	160.1 (25.1)	238.6 (18.6)	Randomized crossover	6 wk	ISP ± powder isoflavones	63	Dairy, meat ± powder isoflavones	AgencyIndustry
Liu et al., 2012 (40)	180 (180 W) PM, PD, or DM2	56.2 (4.4)	145.7	217.1	Randomized parallel	6 mo	Powder with soy protein + isoflavones	15	Powder with milk protein ± isoflavones	AgencyIndustry
Ma et al., 2005 (41)	159 (70 M, 89 W) HC	56.6 (8.4)	181.6	252.6	Randomized parallel	5 wk	Powder packets with soy protein + isoflavones in vanilla or chocolate flavors mixed with water	31.5	Powder packets with milk protein in vanilla or chocolate flavors mixed with water	Industry
Maesta et al., 2007 (42)	46 (46 W) PM	59.4	146.3	233.3	Randomized parallel	16 wk	Chocolate-flavored powder with soy protein ± resistance exercise + 200 mL of skim milk	25	Chocolate-flavored powder with maltodextrin ± resistance exercise + 200 mL of skim milk	AgencyIndustry
Mangano et al., 2013 (43)	97 (97 W) PM	73.14	132.3	207.8	Randomized parallel	1 y	Powder packets with ISP ± isoflavones incorporated into commonly consumed beverages or foods	18	Powder packets with sodium caseinate + whey + egg white protein ± isoflavones incorporated into commonly consumed beverages or foods	Agency
Matthan et al., 2007 (44)	28 (2 M, 26 W) HC	65 (6)	154.3 (20.1)	237.4 (20.9)	Randomized crossover	6 wk	Whole soybean (whole organic soybeans, soynuts, defatted soyflakes, soya granules, and soynut butter), soy flour (soy nutlettes, whole-grain soy flour, and textured soy protein products such as Chicken-Not, Beef-Not, and Turkey-Not), or soy milk (tofu, plain soy yogurt, and plain soy milk) products	37.5	Animal protein (meats, chicken, eggs, and dairy products)	Agency
McVeigh et al., 2006 (45)	35 (35 M)	27.9 (5.7)	106.0 (41.0)	174.0 (48.3)	Randomized crossover	57 d	Powder with ISP + isoflavones	32	Powder with IMP	AgencyIndustry
Mitchell and Collins, 1999 (46)	10 (10 M)	20–50	N/A	186.1	Randomized parallel	4 wk	Soy milk	1 litter = ∼375	Rice milk or semi-skim cow's milk	Agency
Murkies et al., 1995 (47)	47 (47 W) PM	54.9	148.3	232.3	Randomized parallel	12 wk	Raw soy flour, added to a drink or mixed with cereal (or cooked in a muffin)	45	Raw wheat flour, added to a drink or mixed with cereal (or cooked in a muffin)	N/A
Murray et al., 2003 (48)	30 (30 W) PM	54.9	133.8	224.3	Randomized parallel	6 mo	Powder packets with ISP + isoflavones mixed into a morning beverage + estradiol	25	Powder packets with placebo mixed into a morning beverage + estradiol	Industry
Sagara et al., 2004 (49)	50 (50 M) HR for CVD	52.2	N/A	235.5	Randomized parallel	5 wk	Soy powder mixed in cereals, biscuits, and bread rolls + isoflavones	20	Cereals, biscuits, and bread rolls	AgencyIndustry
Santo et al., 2008 (50)	30 (30 M)	24.2 (2.3)	116.5	179.9	Randomized parallel	28 d	Powder packets with ISP ± isoflavones mixed with a beverage of participants’ choice (including milk)	25	Powder packets with IMP mixed with a beverage of participants’ choice (including milk)	Industry
Steinberg et al., 2003 (51)	24 (24 W) PM	54.9 (5.3)	111.8 (18.9)	189 (18.9)	Randomized crossover	6 wk	Powder packets with ISP ± isoflavones mixed with a beverage of participants’ choice	25	Powder packets with total milk protein mixed with a beverage of participants’ choice	Industry
Takatsuka et al., 2000 (52)	52 (52 W) PM	26.1	92.7	176.6	Randomized parallel	2 mo	Usual diet + soy milk (400 mL)	17.2	Usual diet	Agency
Teede et al., 2001 (53)	179 (96 M, 83 W) PM	60.5 (9.6)	148.8	228.2	Randomized parallel	3 mo	Powder sachets with ISP + isoflavones mixed into a beverage	40	Powder sachets with casein mixed into a beverage	Agency
Teixeira et al., 2004 (55)	14 (14 M) DM2, CKD	53–73	98.8	182.7	Randomized crossover	8 wk	Vanilla-flavored powder with ISP incorporated into several dishes or drinks	∼476	Vanilla-flavored powder with casein incorporated into several dishes or drinks	AgencyIndustry
Van Horn et al., 2001 (56)	127 (127 W) PM, HC	66.6 (10.3)	159.3 (29.4)	251.4 (33.3)	Randomized parallel	6 wk	Powder with soy protein + isoflavones + oats (cooked oatmeal or ready-to-eat oat bran cereal) Powder with soy protein + isoflavones + wheat cereal (cream of wheat or non-oat ready-to-eat cereal)	29	Powder with nonfat milk protein + oats (cooked oatmeal or ready-to-eat oat bran cereal) Powder with nonfat milk protein + wheat cereal (cream of wheat or non-oat ready-to-eat cereal)	Industry
van Raaij et al., 1981 (57)	69 (∼41 M, ∼28 W)7	18–28	83.6	152.4	Randomized parallel	4 wk	Specially developed products (milk-like beverages and milk-like beverages fermented to yogurts, brown breads, cookies, and sandwich spreads) with ISP, and gelated product with ISP	54ŧŧ8	Specially developed products (milk beverages and milk beverages fermented to yogurts, brown breads, cookies, and sandwich spreads) with caseinate	AgencyIndustry
Washburn et al., 1999 (58)	42 (42 W) PM	51 (4.8)	126.9 (38.5)	208.0 (40.5)	Randomized crossover	6 wk	Powder packets with soy protein + phytoestrogens mixed with milk, orange juice, yogurt, cereal, or different beverages	20	Powder packets with complex carbohydrate supplement mixed with milk, orange juice, yogurt, cereal, or different beverages	Industry
West et al., 2005 (59)	32 (14 M, 18 W) PM, HC	58 (5.2)	142.6 (27.0)	214.3 (24.4)	Randomized crossover	6 wk	ISP + isoflavones powder, 15 g/d baked into a muffin and 10 g/d integrated into meals provided	25	IMP powder, 15 g/d baked into a muffin and 10 g/d integrated into meals provided	Industry
Wong et al., 1998 (60)	13 (13 M) HC13 (13 M) NC	35.5 (7.2)41.4 (7.8)	193.0 (24.0)110.2 (18.2)	273.0 (20.1)167.1 (18.9)	Randomized crossover	5 wk	ISP incorporated into soy products (beverages, ground meat, and block meat substitutes)	∼759	Animal protein (meat)	AgencyIndustry

¹Values are means, means ± SDs, or ranges. CKD, chronic kidney disease; DM2, diabetes mellitus type 2; HC, hypercholesterolemic; HLP, hyperlipidemic; HR for CVD, high risk of cardiovascular disease; HTN, hypertension; IMP, isolated milk protein; ISP, isolated soy protein; M, men; N/A, not available; NC, normocholesterolemic; OB, obese; PD, prediabetes; PM, post-menopausal/perimenopausal; W, women.

²Mean age ± SD from 33 eligible participants.

³126 subjects were included in test 1 and 95 in test 2; however for data analysis, 12 and 7 subjects from test 1 and test 2, respectively, were excluded.

⁴Mean age for 131 randomized participants.

⁵Soy dose taken from Soya Unsweetened Drink (Provamel) as 3.7 g/100 mL.

⁶ISP = 0.5 g/d.

⁷Men and women taken from the percentage of selected subjects (60% men and 40% women).

⁸65% of protein, 8.4% of energy (∼54 g protein based on 2565 kcal).

⁹Based on 2000 kcal/d, ∼75% soy protein from 20% total protein intake.

Risk of bias

No serious risk of bias was detected for either outcome (Supplemental Figure 1).

Effect on LDL cholesterol

Soy protein intake in 50 trial comparisons demonstrated a significant reduction in LDL cholesterol (MD: −4.76 mg/dL; 95% CI: −6.71, −2.80 mg/dL), equivalent to −3.2% (95% CI: −4.5, −1.9%; P < 0.0001). Although there was evidence of substantial interstudy heterogeneity (I² = 55%, P < 0.0001) (Figure 2), the direction of effect for the majority of trials favored soy protein (37/50, 74%). The systematic removal of each trial did not alter the direction or significance of the effect estimates or the evidence of heterogeneity (Supplemental Table 1). Neither our a priori subgroup analyses nor the a priori FDA subgroup analyses showed any effect modification (Supplemental Figures 2 and 3). Dose–response analyses (Supplemental Figures 4A and 5A) did not identify a dose–response effect or a threshold for LDL cholesterol.

FIGURE 2.

Forest plot for the effect of soy protein intake on circulating LDL cholesterol concentration in adults. Overall effect is represented by the black diamond. ¹Total n = 37, but reported n = 71 for soy protein diets and n = 72 for milk protein diets. ²Twelve subjects were excluded in data analysis in test 1 and 7 in test 2, but it was not specified from which arm; 50% dropout rate was taken from each. Data are expressed as mean differences with 95% CIs, using the generic inverse variance method with random effects models. Paired analyses were applied to all crossover studies. Interstudy heterogeneity was tested by the Cochran Q statistic at a significance level of P < 0.10 and quantified by I²; level of ≥50% represented substantial heterogeneity. n, number of participants.

Effect on TC

Soy protein intake in 52 trial comparisons demonstrated a significant reduction in TC (MD: −6.41 mg/dL; 95% CI: −9.30, −3.52 mg/dL), equivalent to −2.8% (95% CI: −4.1, −1.5%; P < 0.0001). Again, although there was evidence of substantial interstudy heterogeneity (I² = 74%, P < 0.0001) (Figure 3), the direction of the effect for the majority of trials favored soy protein (38/52, 73%). The systematic removal of each trial did not alter the direction or significance of the effect estimates or the evidence of heterogeneity (Supplemental Table 1). Similarly, neither our a priori subgroup analyses nor the a priori FDA subgroup analyses showed any effect modification (Supplemental Figures 6 and 7). Again, the dose–response analyses did not identify a dose–response effect or threshold for TC (Supplemental Figures 4B and 5B).

FIGURE 3.

Forest plot for the effect of soy protein intake on circulating TC concentration in adults. Overall effect is represented by the black diamond. ¹Total n = 37, but reported n = 71 for soy protein diets and n = 72 for milk protein diets. ²Twelve subjects were excluded in data analysis in test 1 and 7 in test 2, but it was not specified from which arm; 50% dropout rate was taken from each. Data are expressed as mean differences with 95% CIs, using the generic inverse variance method with random effects models. Paired analyses were applied to all crossover studies. Interstudy heterogeneity was tested by the Cochran Q statistic at a significance level of P < 0.10 and quantified by I²; level of ≥50% represented substantial heterogeneity. n, number of participants; TC, total cholesterol.

Publication bias

Visual inspection of funnel plots for publication bias showed no evidence of asymmetry or small-study effects for LDL cholesterol or TC (Supplemental Figure 8). Both Egger and Begg tests were nonsignificant for both LDL cholesterol and TC.

Discussion

The meta-analysis of the 46 trials on which the FDA is basing its decision to revoke the heart health claim for soy protein confirms a highly significant but small effect of soy protein intake on both TC and LDL cholesterol concentrations. No dose–response effect was observed, possibly due to a limited range of soy protein dose in the available studies. We regard these effects as intrinsic effects of soy due to the equivalent nature of the selected control trial foods.

The 3.2% reduction in LDL cholesterol is somewhat lower than the reduction found in previously published meta-analyses (68, 69) and much lower than that initially estimated by Anderson et al. (70) in 1995. The extrinsic or substitution value of soy would likely further lower LDL cholesterol if soy foods displace their animal product equivalents that have higher saturated fat and cholesterol content (3). The overall effect in real life could be potentially higher than that seen in these trials. In a previous analysis of NHANES III data, the LDL cholesterol advantage of substituting 13 g/d soy protein for comparable animal protein foods was estimated to lower LDL cholesterol by approximately 3.6%, whereas 25 g/d soy protein was estimated to lower LDL cholesterol by 4.3% (3). These potential substitution advantages are important because the general public of developed countries, including those in North America (71, 72), Europe (73, 74), and Asia (75), is currently being advised to eat more plant-based foods and especially sources of vegetable protein, including nuts, seeds, and legumes. In this regard, the high quality of soy protein and its safety are notable, as indicated by the FDA in the evaluation of the original soy health claim and also by AHA in its soy advisory (2).

One mechanism of action of the intrinsic cholesterol-lowering effect of soy protein has been proposed to relate to the 7S globulin fraction of soy protein, which is common to other legume proteins and appears to inhibit hepatic Apo B synthesis (76). The soy isoflavones have also been proposed to have lipid-lowering potential, but this mechanism of action is less likely (35), although they may contribute to antioxidative effects (35).

It is recognized that soy foods have been a component of the group of foods recommended by the FDA as cholesterol lowering (nuts, plant sterols, and viscous fibers) (5–7). Soy protein as part of a dietary portfolio of cholesterol-lowering foods has been recommended by agencies concerned with CVD health (the National Cholesterol Education Program, Adult Treatment Panel III 2004 update, Heart UK, Canadian Cardiovascular Society, and European Atherosclerosis Society) (9, 11, 68, 77, 78).

Strengths and limitations

The weakness of this analysis is that it was based on data already gathered by the FDA rather than involving a formal search assessment of the literature with the outcomes assessed by two independent reviewers. Nevertheless, these data were extracted by the FDA as representing those trials on which a final decision would be made concerning the soy protein health claim. Because we are addressing the question raised by the FDA, our inclusion criteria included only those trials selected by the FDA.

We also noted considerable inconsistency (heterogeneity) in the treatment effects. However, as was shown by inspection of the forest plots, the individual effect estimates for most trials showed a direction of effect that favored a reduction in LDL cholesterol (Figure 2) and TC (Figure 3). Furthermore, it is not at all unusual for there to be significant heterogeneity even for effects that are well established such as the cholesterol-lowering effect of phytosterols/stanols (79) and oat β-glucan (80).

The strengths of this analysis are the large number of trials used; the median follow-up period was 6 wk, which allows for the assessment of a moderate duration of intervention; none of the trials were rated as having a serious risk of bias; and there was no evidence of publication bias. Another strength is that this analysis is timely, in a similar manner as the analysis performed in 2010 in response to the AHA advisory on soy protein in 2006 (3). In addition, although the number of significant individual trials was relatively small, the overall result of combining multiple small clinical trials provides the highest form of evidence on which clinical decisions are made in other situations (81).

Implications

Overall, soy decreased LDL cholesterol by ∼3.2% and TC by 2.8% as a result of consuming ∼25 g soy protein on a daily basis. The lipid reduction was modest but highly significant even if heterogeneous. Nevertheless, as part of a dietary strategy to reduce serum cholesterol, soy use in a dietary portfolio has proven effective to the extent that it is included in guidelines of many jurisdictions (11, 12, 82, 83).

In conclusion, soy protein lowers LDL cholesterol by a small but significant amount. Our data fit with the advice given to the general public internationally to increase plant protein intake (71, 84).

Notes

This work was funded by the Canadian Institutes of Health Research (CIHR; 129920) through the Canada-wide Human Nutrition Trialists' Network. The Diet, Digestive tract, and Disease (3-D) Centre, funded through the Canada Foundation for Innovation (CFI) and the Ministry of Research and Innovation's Ontario Research Fund (ORF), provided the infrastructure for the conduct of this project. JLS was funded by a PSI Graham Farquharson Knowledge Translation Fellowship, Canadian Diabetes Association (CDA) Clinician Scientist Award, CIHR Institute of Nutrition, Metabolism and Diabetes (INMD)/Canadian Nutrition Society (CNS) New Investigator Partnership Prize, and Banting & Best Diabetes Centre (BBDC) Sun Life Financial New Investigator Award. DJAJ was funded by the Government of Canada through the Canada Research Chair Endowment. JLS has received research support from the CIHR, Diabetes Canada, PSI Foundation, BBDC, CNS, American Society for Nutrition (ASN), Calorie Control Council, International Nut and Dried Fruit Council Foundation (INC), National Dried Fruit Trade Association, The Tate and Lyle Nutritional Research Fund at the University of Toronto, The Nutrition Trialists Fund at the University of Toronto (a fund established by the Calorie Control Council), 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 in-kind food donations to support a randomized controlled trial from the Almond Board of California, California Walnut Commission, American Peanut Council, Barilla, Unilever, Unico, Primo, Loblaw Companies, Quaker (Pepsico), Kellogg Canada, and WhiteWave Foods. He has received travel support, speaker fees, and/or honoraria from Diabetes Canada, CNS, Mott's LLC, Dairy Farmers of Canada, FoodMinds LLC, PepsiCo, The Ginger Network LLC, International Sweeteners Association, Nestlé, Pulse Canada, Canadian Society for Endocrinology and Metabolism, GI Foundation, Abbott, Biofortis, American Society for Nutrition (ASN), Health Sciences North, and Physicians Committee for Responsible Medicine. He has ad hoc consulting arrangements with Perkins Coie LLP, Tate & Lyle, and Wirtschaftliche Vereinigung Zucker e.V. 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), Canadian Cardiovascular Society, and Obesity Canada. He serves as an unpaid scientific advisor for the Food, Nutrition, and Safety Program and the Technical Committee on Carbohydrates of the International Life Science Institute North America. He is a member of the International Carbohydrate Quality Consortium (ICQC), Executive Board Member of the Diabetes and Nutrition Study Group of the EASD, and Director of the Toronto 3D Knowledge Synthesis and Clinical Trials Foundation. PK-E serves on the Avocado Nutrition Science Advisory and HUMAN Scientific Advisory Board and has grants from the National Cattlemen's Beef Association, Peanut Institute, INC, California Strawberry Commission, Ocean Spray Cranberries, Inc., McCormick Science Institute, and Hass Avocado Board. DJAJ 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 Canada and Netherlands, 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, Pepsi/Quaker, INC, Soy Foods Association of North America, Coca-Cola Company (investigator initiated, unrestricted grant), Solae, Haine Celestial, Sanitarium Company, Orafti, International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, Soy Nutrition Institute (SNI), the Canola and Flax Councils of Canada, the Calorie Control Council, CIHR, CFI, and ORF. He has received in-kind supplies for trials as 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, and 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), Coca-Cola Company, EPICURE, Danone, Diet Quality Photo Navigation, Better Therapeutics (FareWell), Verywell, True Health Initiative, Institute of Food Technologists, SNI, Herbalife Nutrition Institute, Saskatchewan Pulse Growers, Sanitarium Company, Orafti, the American Peanut Council, the International Tree Nut Council Nutrition Research and Education Foundation, the Peanut Institute, Herbalife International, Pacific Health Laboratories, Nutritional Fundamentals for Health, Barilla, Metagenics, Bayer Consumer Care, Unilever Canada and Netherlands, Solae, Kellogg, Quaker Oats, Procter & Gamble, Abbott Laboratories, 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, Agri-Culture and Agri-Food Canada, the Canadian Agri-Food Policy Institute, Pulse Canada, the Soy Foods Association of North America, the Nutrition Foundation of Italy, 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, CNS, ASN, Arizona State University, Paolo Sorbini Foundation, and INMD. He received an honorarium from the US 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 CDA. He is a member of the 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.

Author disclosures: SBM, MM, SSL, EV, LC, TAK, KS, and AM, no conflicts of interest.

Supplemental Table 1 and Supplemental Figures 1–8 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/.

Abbreviations used: CVD, cardiovascular disease; ISP, isolated soy protein; MD, mean difference; TC, total cholesterol.