Table 3. Estimates of etiologic effects of dietary factors and risk of cardiovascular diseases and type 2 diabetes mellitus1.
| Dietary Factor | Outcome | Studies in Each Meta-analysis2 | Source | No. of Subjects | No. of Events | Unit of RR3 | RR (95% CI)3 | Statistical Heterogeneity |
|---|---|---|---|---|---|---|---|---|
| Foods | ||||||||
| Fruits4 | ↓ CHD | 16 cohorts (22 estimates) | Gan, 2015 [71] | 817,977 | 13,786 | per 1 serving/d (100 g/d) | 0.94 (0.91, 0.98) | I2 = 31.7% p = 0.08 |
| ↓ Ischemic stroke | 9 cohorts (10 estimates) | De novo meta-analysis5 [72–79] | 329,204 | 5,517 | per 1 serving/d (100 g/d) | 0.88 (0.83, 0.93) | I2 = 77.1% p<0.001 | |
| ↓ Hemorrhagic stroke | 5 cohorts (7 estimates) | De novo meta-analysis5 [74–77] | 175,035 | 1,535 | per 1 serving/d (100 g/d) | 0.73 (0.62, 0.87) | I2 = 81.4% p<0.001 | |
| Vegetables 6 | ↓ CHD | 9 cohorts | Gan, 2015 [71] | 761,612 | 13,135 | per 1 serving/d (100 g/d) | 0.95 (0.92, 0.98) | I2 = 35.6% p = 0.07 |
| ↓ Ischemic stroke | 9 cohorts (10 estimates) | De novo meta-analysis5 [72–75, 77–79] | 329,204 | 5,515 | per 1 serving/d (100 g/d) | 0.83 (0.75, 0.93) | I2 = 89.9% p<0.001 | |
| ↓ Hemorrhagic stroke | 5 cohorts (7 estimates) | De novo meta-analysis5 [74–77, 79] | 175,035 | 1,535 | per 1 serving/d (100 g/d) | 0.83 (0.72, 0.96) | I2 = 30% p = 0.20 | |
| Beans/legumes | ↓ CHD | 5 cohorts | Afshin, 2014 [80] | 198,904 | 6,514 | per 1 serving/d (100 g/d) | 0.77 (0.65, 0.90) | I2 = 0.2% p = 0.41 |
| Nuts/seeds | ↓ CHD (fatal) | 1 RCT and 5 cohorts | Afshin, 2014 [80] | 206,114 | 6,749 | per 4 servings/wk (4 oz/wk) | 0.76 (0.69, 0.84) | I2 = 27.2% p = 0.23 |
| ↓ CHD (non-fatal) | 1 RCT and 3 cohorts | Afshin, 2014 [80] | 141,390 | 2,101 | per 4 servings/wk (4 oz/wk) | 0.78 (0.67, 0.92) | I2 = 0.0% p = 0.46 | |
| ↓ Diabetes | 1 RCT and 5 cohorts | Afshin, 2014 [80] | 230,216 | 13,308 | per 4 servings/wk (4 oz/wk) | 0.87 (0.81, 0.94) | I2 = 21.6% p = 0.27 | |
| Whole grains | ↓ CVD | 7 cohorts (9 estimates)7 | New GLST8 of Mellen 2008 [54] | 285,217 | 7,005 | per 1 serving/d (50 g/d) | 0.91 (0.86, 0.97) | I2 = 84.0% p<0.001 |
| ↓ CHD | 6 cohorts | New GLST8 of Mellen 2008 [54] | 281,633 | 4,593 | per 1 serving/d (50 g/d) | 0.97 (0.94, 0.99) | I2 = 75.5% p = 0.001 | |
| ↓ Diabetes | 10 cohorts | Aune 2013 [59] | 385,868 | 19,791 | per 1 serving/d (50 g/d) | 0.88 (0.83, 0.93) | I2 = 82% p = <0.0001 | |
| Red meats, unprocessed | ↑ Diabetes | 9 cohorts (10 estimates) | Pan, 2011 [81] | 442,101 | 28,228 | per 1 serving/d (100 g/d) | 1.19 (1.04, 1.37) | I2 = 93% p<0.001 |
| Processed meats | ↑ CHD | 5 cohorts (6 estimates) | Micha, 2010 [82] | 614,062 | 21,308 | per 1 serving/d (50 g/d) | 1.37 (1.11, 1.68) | I2 = 76.2% p = 0.001 |
| ↑ Diabetes | 8 cohorts (9 estimates) | Pan, 2011 [81] | 371,492 | 26,256 | per 1 serving/d (50 g/d) | 1.51 (1.25, 1.83) | I2 = 94.3% p<0.001 | |
| Fish/Seafood9 | ↓ CHD (fatal) | 16 cohorts (17 estimates) | Zheng, 2012 [83] | 315,812 | 4,472 | per 15 g/d (~1–100 g- serving/wk) | 0.94 (0.90–0.98) | I2 = 63.1 p<0.005 |
| Yoghurt | ↓ Diabetes | 9 cohorts | Chen, 2014 [84] | 408,096 | 32,995 | per 1 serving/d (8 oz/d, 244 g/d) | 0.82 (0.70, 0.96) | I2 = 65.3 p = 0.003 |
| Sugar-sweetened beverages | ↑ BMI (when baseline BMI <25 kg/m2)10 | 3 cohorts | Mozaffarian, 2011 [55] | 120,877 | n/a | per 1 serving/d (8 oz/d)111 | 0.10 kg/m2 (0.05, 0.15) | not reported |
| ↑ BMI (when baseline BMI ≥25 kg/m2)10 | 3 cohorts | Mozaffarian, 2011 [55] | 120,877 | n/a | per 1 serving/d (8 oz/d)11 | 0.23 kg/m2 (0.14, 0.32) | not reported | |
| ↑ Diabetes (BMI-adjusted)10 | 17 cohorts | Imamura, 2015 [85] | 464,937 | 38,253 | per 1 serving/d (8 oz/d)11 | 1.27 (1.10, 1.46) | I2 = 73% | |
| ↑ CHD (BMI-adjusted)9 | 4 cohorts | Xi, 2015 [86] | 173,753 | 7,396 | per 1 serving/d (8 oz/d)11 | 1.17 (1.10, 1.24) | I2 = 0.0% p = 0.79 | |
| Nutrients | ||||||||
| PUFA replacing Carbs12 | ↓ CHD | 9 cohorts (12 estimates) | Farvid, 2014 [58] | 262,612 | 12,198 | per 5%E/d | 0.90 (0.85, 0.94) | I2 = 47.3% p = 0.04 |
| PUFA replacing SFA12 | ↓ CHD | 8 cohorts (11 estimates) | Farvid, 2014 [58] | 262,612 | 12,198 | per 5%E/d | 0.91 (0.87, 0.96) | I2 = 55.9% p = 0.01 |
| Seafood omega-3 fats13 | ↓ CHD (fatal) | 4 RCTs and 15 cohorts | Mozaffarian 2006 [87] | 363,003 | 5,951 | per 100 mg/d | 0.85 (0.79, 0.92) | not reported |
| Trans-fats14 | ↑ CHD | 4 cohorts | Mozaffarian, 2006 [88] | 139,836 | 4,965 | per 2% %E/d | 1.23 (1.11, 1.37) | not reported |
| Dietary fiber15 | ↓ CVD | 10 cohorts | Threapleton, 2013 [89] | 1,279,690 | 19,869 | per 20 g/d | 0.76 (0.70, 0.84) | I2 = 45% |
| ↓ CHD | 12 cohorts | Threapleton, 2013 [89] | 1,039,572 | 11,282 | per 20 g/d | 0.76 (0.68, 0.85) | I2 = 33% | |
| ↓ Stroke | 7 cohorts | Threapleton, 2013 [90] | 324,640 | 9,257 | per 20 g/d | 0.81 (0.70, 0.95) | I2 = 59% | |
| ↓ Diabetes | 5 cohorts | Yao, 2014 [91] | 157,336 | 3,029 | per 30 g/d | 0.76 (0.65, 0.88) | not reported | |
| Glycemic load16 | ↑ CHD | 9 cohorts (13 estimates) | Mirrahimi, 2014 [92] | 262,891 | 11,319 | high vs. low | 1.23 (1.06, 1.42) | I2 = 52% p = 0.02 |
| ↑ Stroke | 6 cohorts (9 estimates) | Cai, 2015 [93] | 222,308 | 2,951 | high vs. low | 1.19 (1.05, 1.36) | I2 = 5.0% p = 0.39 | |
| ↑ Diabetes | 17 cohorts (30 estimates) | Bhupathiraju, 2014 [94] | 698,589 | 46,115 | high vs. low | 1.13 (1.08, 1.17) | I2 = 26.4% p = 0.09 | |
| Sodium17, 18 | ↑ CVD (fatal) | 11 cohorts (16 estimates) | Poggio, 2015[95] | 220,249 | 9,628 | high vs. low | 1.12 (1.06, 1.19) | I2 = 57.6% p = 0.002 |
| ↑ SBP, main effect, white, age 50, normotensives18 | 103 RCTs (107 estimates) | Mozaffarian, 2014 [24] | 6,970 | NA | per 2,300 mg/d (100 mmol/d) | 3.74 mm Hg (5.18, 2.29) | not reported | |
| ↑ SBP, additional effect per year of age < or > 5018 | 0.105 mm Hg (0.164, 0.047) | |||||||
| ↑ SBP, additional effect among Blacks18 | 2.49 mm Hg (4.85, 0.13) | |||||||
| ↑ SBP, additional effect among hypertensives18 | 1.87 mm Hg (3.63, 0.12) | |||||||
| Potassium17 | ↓ Stroke | 9 cohorts (11 estimates) | D’Elia, 2011 [96] | 233,606 | 7,066 | per 1,000 mg/d (25.7 mmol/d) | 0.87 (0.79, 0.94) | I2 = 55% p = 0.01 |
1 Dietary factors with probable or convincing evidence, based on the Bradford-Hill criteria for assessing causality [11], for etiologic effects on cardiometabolic outcomes including coronary heart disease (CHD), stroke, cardiovascular disease (CVD), type 2 diabetes, body mass index (BMI), or systolic blood pressure (SBP).
2 Number of estimates can be higher than the number of studies if more than one arm in a randomized controlled trial or if estimates were separately reported by sex or age in prospective cohort studies.
3 Based on published or de novo dose-response meta-analyses of prospective cohorts or randomized trials. Meta-analyses were evaluated based on design, number of studies and events, definition of dietary exposure and disease outcomes, statistical methods, evidence of bias, and control for confounders. Relative risks (RRs) were standardized across individual studies per uniform servings of intake. When necessary, original data were extracted from individual studies to perform de novo dose-response meta-analyses using all available data by means of generalized least squares (GLST in STATA) for trend estimation. Effect sizes are relative risks (RRs) (95% confidence intervals (CIs)) except for sugar-sweetened beverage (SSB) effects on BMI (absolute in kg/m2) and sodium effects on SBP (absolute in mm Hg). Effect sizes correspond to the relationship between increased consumption of each dietary target per unit of RR and respective change in cardiometabolic risk (directionality in risk: ↑ increased, ↓ decreased). Proportional effects of major risk factors on cardiometabolic outcomes vary by age, with an inverse log-linear age association [22]. We derived age specific RRs for diet-cardiometabolic disease relationships based on the age patterns of RRs for metabolic risk factors and incident cardiometabolic disease events (see Figure B in S1 File) [22]. Except as indicated (SSBs, sodium), we did not identify sufficient evidence for effect modification by other factors beyond age, e.g. race, obesity, or overall diet quality.
4 Excluding 100% juices.
5 All four of these de novo meta-analyses were performed for consumption of fruits and vegetables and stroke subtypes due to absence of recent published meta-analyses; all details are provided in S1 File.
6 Excluding vegetable juices, starchy vegetables such as potatoes or corn, and salted or pickled vegetables. Because certain beans/legumes (e.g., black beans, lentils) were commonly included as vegetables in many of the identified studies, the etiologic effects identified for vegetables should be considered as representing the effects of vegetables including beans/legumes. We also evaluated etiologic effects of beans/legumes separately.
7 When a trial did not report an effect for total CVD separately (n = 3 cohorts), CHD and stroke estimates from each trial were first pooled using fixed effects (n = 2 cohorts), or the CHD estimate was used in place of CVD when that was the only reported outcome (n = 1 cohort).
8 Data were re-extracted from all original investigations identified in the meta-analysis to assess dose-response using two-step generalized least squares for trend estimation [18, 19].
9 Etiologic effects are limited to fatal CHD only due to absence of probable or convincing evidence for benefits on nonfatal CHD events. Benefits for were identified up to 3.5 servings/week of fish/seafood and 250 mg/d of eicosapentaenoic acid (EPA) + docosahexaenoic acid (DHA), with little evidence for additional benefits at higher intakes.
10 Available evidence suggests that SSBs increase risk through effects on both BMI and additional BMI-independent effects on type 2 diabetes and CHD [22, 23].
11 Depending on study-specific assumptions, use of UK or US conversion factors, and study weighting, the serving size is in this analysis could also be 8.7–9.1 oz.
12 Reported effects are nearly identical for polyunsaturated fats (PUFA) replacing carbohydrates (Carbs) or saturated fats (SFA).
13 Linear reduction in risk observed until 250 mg/day, with little evidence for additional benefits at higher intakes.
14 The overall causal effect was based on 4 cohorts; the final RR (95% CI) used herein was very similar but based on the isocaloric replacement of trans-fats with an equal distribution of SFA, monounsaturated fats (MUFA), and PUFA based on a meta-analysis of 2 cohorts.
15 Possible evidence for larger effects at intakes above 20 g/d.
16 Glycemic load is calculated as the glycemic index of a food multiplied by its carbohydrate content. Higher values reflect both higher glycemic index and higher quantities of refined grains, starches, and sugars. We also identified evidence for causal effects of dietary fiber. Glycemic load and dietary fiber each overlap with foods in this Table including fruits, vegetables, beans/legumes, nuts/seeds, and whole grains.
17 Assessed by 24h dietary recall, food frequency questionnaire, or 24h urine excretion.
18 Available evidence suggests that sodium increases mortality from CHD, stroke, and other BP-related cardiovascular diseases through effects on SBP [22, 24]. For every year above or below age 50, there was 0.105 mm Hg (95% CI: 0.047, 0.164) larger or smaller BP reduction, respectively. Effects on CVD vs. SBP were separately identified and are not independent (i.e., effects on CVD are at least partly mediated by SBP effects).