depicts the flow of study selection for the analysis. Of the 140 potentially relevant references identified, 117 were excluded following review of abstract and title. A total of 23 full-text articles were retrieved and reviewed for inclusion. We further excluded 8 articles due to multiple publications from an individual trial, 1 study was shorter than 3 weeks in duration, 1 was excluded due to lack of control diet, 2 articles were excluded because they reported insufficient information to calculate an effect size and/or its variance and 1 was eliminated because the control diet was an active cholesterol lowering intervention. A total of 10 randomized controlled trials were included in this meta-analysis, representing data collected from 268 participants.(21
Flow diagram of articles identified and evaluated during the study selection process.
The baseline characteristics of the study participants and designs of the randomized controlled trials are presented in . Out of the 10 trials, 4 were conducted in the United States, 2 in Australia, 2 in Spain, and 1 each in Chile, and New Zealand. A total of 268 participants were included in the analysis. There were 188 men in the trials, representing 70.1% of all participants and 5 trials had exclusively male participants. Study participants ranged in age from 18 to 78. Participants with high, borderline high, and normal cholesterol levels were included and no studies included participants who were taking cholesterol-lowering drugs. Total cholesterol at baseline was reported for 8 studies and study means ranged from 199 to 294.6 mg/dL(22
). Mean baseline LDL- and HDL-cholesterol were reported in 6 studies and mean LDL cholesterol ranged from 138 to 200.4 mg/dL while mean HDL cholesterol ranged from 39.0 to 58.0 mg/dL.(22
) Trials primarily employed the crossover design, however 2 were parallel and 1 was factorial in design. Most trials matched macronutrient and energy content between the legume diet and control diet groups, including amounts of saturated and total fat in the diets (). Various non-soy legumes were represented; intervention diets included the addition of mixed legume dishes, whole chickpeas, field beans ground into flour, whole pinto beans, canned baked beans, whole peas, and whole navy beans, among others (). Comparison groups consisted of calorie and macronutrient-matched control diets, often with a wheat-based or canned vegetable substitution. Intervention durations ranged from 3 to 8 weeks. Most of the studies were conducted in free-living adults, though 1 trial was conducted in a metabolic lab (26
) and in one trial all meals were eaten at the study kitchen.(21
Baseline characteristics of participants and design characteristics of randomized controlled trials of non-soy legume consumption.
Dietary composition of non-soy legume diets as compared to control diets.
Mean net changes and corresponding 95% CIs for total serum cholesterol were reported in 10 studies (21
), and HDL, LDL cholesterol and triglycerides were reported in 9 studies (21
). For total cholesterol, the mean net changes in each study ranged from 1.2 to −31.4 mg/dL. The pooled mean net change from a legume diet was −11.76 mg/dL (95%CI: −16.06, −7.47, p<0.001; Χ2
for heterogeneity p=0.26), , Panel A. The mean net changes for HDL cholesterol ranged from −4.03 to 5.91 mg/dL, and the pooled mean net change was 0.85 mg/dL (95%CI: −1.62, 3.32, p=0.05; Χ2
for heterogeneity p=0.005), , Panel B. For serum LDL cholesterol, mean net changes ranged from −18.91 to 0.0 mg/dL and pooled mean net change was −7.98 mg/dL (95%CI: −11.41, −4.54, p<0.001; Χ2
for heterogeneity p=0.65), , Panel C. For triglycerides, mean net changes ranged from −43.36 to 0.80 mg/dL. Pooled mean net change for serum triglycerides was −18.94 mg/dL (95%CI: −38.04, 0.17, p=0.05; Χ2
for heterogeneity p<0.001), , Panel D. Only 3 studies reported information on VLDL as an outcome measure (not shown), and mean net changes ranged from −8.24 mg/dL to 1.00 mg/dL and the pooled mean net change was −3.34 mg/dL (95%CI: −9.13, 2.45, p=0.26; Χ2
for heterogeneity p<0.001).
Mean net change in total (Panel A), HDL (Panel B), and LDL cholesterol (Panel C) and triglycerides (Panel D) and corresponding 95% confidence intervals by trial and pooled.
We examined the potential for publication bias by plotting sample sizes versus mean net change for total cholesterol, HDL, LDL, VLDL, and triglycerides among the trials included in this meta-analysis using Begg's rank correlation test (p=0.09, p=0.75, p=0.30, p=0.12 and p=0.99 for total, HDL, LDL and VLDL cholesterol, and triglycerides, respectively) and Egger's linear regression tests (p=0.19, p=0.53, p=0.41, p=0.01 and p=0.55 for total, HDL, LDL and VLDL cholesterol, and triglycerides, respectively).
We also examined heterogeneity between studies. Heterogeneity among the effect sizes of individual trials for total, LDL, HDL and VLDL cholesterol and triglycerides had I-square values of 19.7%, 63.8%, 0%, 89.0% and 97.9% respectively. Significant heterogeneity remained between studies for HDL and VLDL cholesterol and triglycerides, therefore, we performed a meta-regression analysis to examine characteristics of the trials and/or their study populations which may affect the heterogeneity in mean net change for lipid parameters. Significant predictors (p<0.05) of the mean net change in total cholesterol among studies included the number of male participants and the length of the intervention phase. For LDL cholesterol, significant predictors included mean age, number of male participants, study design, number of participants and duration of the study. For triglycerides only mean age was a significant predictor and for HDL cholesterol none of the study characteristics were significant predictors.
We then performed sensitivity analyses on net change of lipid concentration using gender distribution (trials with 100% vs. <100% male participants), median duration of the intervention (< 5 vs. ≥5 weeks), study design (crossover vs. parallel or factorial design), and type of control diet (matched vs. other), shown in . In the influence analysis, exclusion of any single study did not change the significance of the pooled estimates for total cholesterol, LDL, HDL, VLDL, or triglycerides.
Sensitivity analysis of mean net change in serum lipid concentrations using different exclusion criteria.